CLINICAL PHARMACOGENOMICS FOR ANTI-EPILEPTIC DRUG RESPONSE IN EPILEPSY PATIENT MANAGEMENT IN INDIA

Abstract

Inter-individual variability in drug response, broadly including drug efficacy and its safety, is an increasing problem globally. Such variability has detrimental effect on patients leading to increased financial burden to reduced quality of life. Genomic factors being one of the reasons for this variability, and is explored in the field of study called ‘pharmacogenomics’ (PGx). A broader definition of PGx, ‘is the study of genomic technologies to enable the discovery and development of novel drugs and the optimization of drug dose and choice in individual patients to maximize efficacy and minimise toxicity. The efficacy of different drugs have been reported to vary from 25% to 80%. It is famously stated that more than 90% of the drugs only work in 30 to 50% of the people. And in terms of drug toxicity, adverse drug reactions (ADRs) affect about 15% of people in hospital. The aim of PGx is to define the underlying genetic mechanisms and ultimately to implement pharmacogenetic testing to improve treatment outcome. Another advantage of understanding the genetic basis of variable drug response can be used as a tool to expand the use of existing drugs for new indications as well as for identification of new drug targets or drug development. After three decades of research in PGx, there are vital gaps in achieving translational efficiency when advancing towards clinical implementation. With the enormous amount of articles published every year (approximately 6 lakh articles in PubMed so far), less than 1% reaches to clinical trials (5491 randomised controlled trials). And among these only 133 genes are known with PGx outcome and are included for drug labelling warnings approved by the United Stated food and drug administration (FDA) for 363 drugs. This large translational gap needs to reduce. Secondly, most of the PGx markers established (163 markers for 118 drugs) are used to predict drug toxicity because clear and discrete phenotypic end-points are available for assessment like skin rashes, liver toxicity. But only a handful of markers (73 markers for 20 drugs) are known to predict drug response due to heterogeneous end-points like drug clearance rate, drug/metabolite ratio, and metabolizer status. More homogenous patient cohorts studying drug response outcomes are required to identify genome-based markers for poor response. And lastly, so far after several candidate gene studies, genome-wide association (GWAS) are shaping the future of genetic association models and it should attempt to explore the genome of patients across different ethnic groups but unfortunately, most of the GWAS are confined to western population like American (AMR) and European (EUR). Only a few in East Asian (EAS) encompassing Chinese and Japanese ethnicity, and no GWAS in African (AFR) 3 or South Asian (SAS) group so far. It is inevitable to screen the genome of all global populations to identify conclusive markers predicting PGx outcomes. Thus, our study was aimed to elucidate PGx markers to prognose therapeutic phenotypes in patients with epilepsy (PWE) & develop a platform for evidence-based testing for clinical implications. In this study, we first developed a semi-automated text mining approach, using R package, pubmed.mineR, to retrieve published articles with PGx information in the form of disease–drug–gene- genetic polymorphism relationships to obtain PGx related data for epilepsy treatment for easier therapeutic guidelines. Further we evaluated our approach by comparing its performance (precision and accuracy) with the other available benchmark datasets like PharmGKB and compared the results retrieved with the FDA approved PGx markers used for drug labelling to weigh its clinical ability and accuracy. We identified 2304 PGx relationships pertaining to 1753 disease types and 666 drugs. Our approach showed performance precision of 80.6% with benchmark datasets like Pharmacogenomic Knowledgebase (PharmGKB) (90.4%), Online Mendelian Inheritance in Man (OMIM) (60.0%), and comparative toxicogenomics database (CTD) (72.9%). From a total of 2,304 PGx relationships identified, a total of 127 (68%) are coinciding with the 362 US-FDA approved 362 pharmacogenomic markers used in drug labelling, indicating that our approach has a greater precision in data extraction with PGx information for drug response prediction. Subsequently, we performed genome-wide genotyping on 789 PWE of North Indian origin to identify genetic variants significantly associated with poor response to commonly prescribed anti-epileptic drugs (AEDs) like phenytoin (PHT), carbamazepine (CBZ) or valproic acid (VPA). This GWAS was performed using commercially available Illumina Infinium Global screening array (GSA)-24 v2.0 with psych customization. On performing quality control (QC) based on different parameters to exclude out the poor quality single nucleotide polymorphisms (SNPs) and samples, we conducted logistic regression using age, sex and 2 principal components (PC1, PC2) as covariates in PLINK 1.9 assuming an additive model and evaluated the association of each imputed SNP. Our GWAS of AED response revealed suggestive evidence for association at 29 genomic loci (p <5.0 x 10-5 ) but no significant association reflecting its limited power. The suggestive associations highlight candidate genes that are implicated in metabolism of AEDs are known targets to these drugs. The top SNP rs60633642 associated with overall poor response to AED [OR (95% CI) = 1.98(1.50-2.60), p< 1.185 x10-6 ] is an intergenic variant (SPTLC3; ISM1) down-regulating the expression of SPTLC3 gene in different tissues, most significant at brain caudate basal ganglia 4 tissue (p<0.0003). Functional annotation of these genomic loci based on enrichment analysis revealed functions like regulation of K + transmembrane transport, neuron development, Ca2+ transport to be enriched with maximum fold enrichment (12 fold, p< 5.0 x10-6 ). Likewise, genetic variants associated with poor response to phenytoin, carbamazepine, valproic acid within the suggestive cut-off were four, thirteen and eleven, respectively. Our study is the first of its kind, investigating genetic association of AED response specific to Indian population using a genome-wide approach. The findings from this study upon replication and diagnostic predictability assessment can be used for upcoming pharmacogenetic studies. Lastly, we estimated the diagnostic accuracy of these identified PGx markers for which we overlapped our GWAS results with already published markers known with PGx response to AEDs as well as our findings from text mining in our study. In conclusion, 88 commercial PGx marker are known related to AED response. Among these 19 SNPs overlapped with our GWAS findings for overall poor response. Assessing the diagnostic predictability of these 19 markers showed moderate accuracy (50-60%). These markers are promising candidates for PGx application after appropriate validation and replication. Eight out of these 19 genetic variants are already in use for drug labelling approved by the FDA. Strengthening the fact that genome-based markers can be exploited for application in precision medicine in epilepsy treatment. In conclusion, our study provides a robust text mining semi-automated R-package for retrieving promising PGx variants. These variants were screened in our population specific cohort highlighting significant loci associated with poor response to AEDs like phenytoin, carbamazepine and valproic acid. Although, further replication in independent sample cohort can strengthen statistical power, or functional validation of the associated loci could help provide mechanistic insights for biological relevance to pharmaco-response. Eleven out of the 19 SNPs identified from our GWAS data have moderate diagnostic accuracy are promising candidates which upon validation can be used for PGx application in prognosis of poor response in epilepsy specific for our population.