Analysis of ADME gene array in Chronic Myeloid Leukemia identifies association between ABCC4, POR, SPG7 and UGT2B7 gene variants and imatinib response

MARIA DEL ROSARIO ANADON; MARIA BELEN FONTECHA; MARTIN LEDESMA; FERNANDA NORIEGA; BEATRIZ MOIRAGHI; RAQUEL BENGIÓ; IRENE LARRIPA; ARIELA FUNDIA

Conferencia; 25th Annual John Goldman Conference on Chronic Myeloid Leukemia: Biology and Therapy; 2023

BackgroundTargeted therapy of chronic myeloid leukemia (CML) with imatinib achieves excellent anddurable responses. However, many cases undergo treatment failure, relapse or developresistance, supporting the need for new prognostic biomarkers. Imatinib resistance is mainlydue to BCR-ABL1 point mutations, but other mechanisms include clonal chromosomalevolution, BCR-ABL1 amplification, DNA repair defects, signaling pathways activation orpharmacogenomic variations. DNA-array technologies promoted massive detection ofgermline genetic variants influencing response. However, scarce DNA-array studies ofvariations in genes involved in absorption, distribution, metabolism and excretion (ADME)processes were undertaken in CML.AimsTo perform a massive analysis of germline variants in ADME genes to assess their involvementin treatment failure and resistance development in imatinib-treated CML patients.MethodsA total of 66 CML patients (22 responders and 42 no responders) treated with imatinib (400mg) as first-line treatment were evaluated using the Infinium Global Screening Array-24 v2.0(GSA-array). This platform contains 743,848 variants, of which 14,609 are knownpharmacogenomic markers. Filtering VCF data was performed by excluding variants occurringin X/Y chromosomes, intergenic regions and pseudogenes. Only variants with 100% genotypingcall rates were included in the analysis, discarding all missing values. Furthermore, data werefiltered to consider only genes previously associated with therapeutic failure in CML. We applya feature selection supervised machine learning algorithm (Binary discriminant analysis) torank SNVs according to their power to discriminate between the two patient groups with thebind R package. The algorithm outputs a t.score, and values greater than 2.5 or less than -2.5were considered significant with a 95% confidence. Finally, statistically significant variantswere tested for their relationship with therapy outcome by analyzing standard genetic models(additive, recessive and dominant) for disease penetrance considering responders as thecontrol group and wild-type homozygotes as reference genotypes using the SNPStats softwarewith p