Pharmacogenetics—Not Just Low‐Hanging Fruit

The current issue of Clinical Pharmacology on the other hand, it may lead to some confusion when comparing new data with literature that used partly other definitions over decades. Overall, however, this comprehensive work is a milestone that will be used as a reference for future research and clinical applications. A further interesting article in this issue “Prognostic implication of CYP2C19 genotype according to clinical risk stratification after drug-eluting stent implantation” by Park et al. comes from a Korean consortium dealing with an allegedly old topic, the relation of CYP2C19 genotype, clopidogrel and clinical outcome after drug-eluting stent implantation.10 First observations on outcome worsening of clopidogrel-treated patients with acute coronary syndromes and planned percutaneous coronary interventions (PCI)11 or acute myocardial infarction12 with respect to CYP2C19 poor metabolizer status, go back to 2009, followed by numerous further studies. The Clinical Pharmacogenetic Implementation Consortium published guidelines already in 2011, with the latest update in 2022.13 In essence, CYP2C19 poor metabolizers receiving the ADP-receptor antagonist prodrug clopidogrel are at risk because of significantly reduced platelet inhibition. In the prognostic study of Park et al.,10 lasting from 2003 to 2018, over 8,000 Korean coronary artery disease (CAD) patients were included who had been genotyped before for CYP2C19 and received aspirin and clopidogrel as P2Y12-receptor antagonist. Importantly, the CYP2C19 status was not considered in the clinical procedures. The outcome was stratified to the predicted phenotype considering the CHADS-P2A2RC and the TRS 2*P CAD risk score. As expected, the risk of outcome worsening increased with increasing clinical score. However, the impact of the genotypically predicted CYP2C19 phenotype was much stronger in the group of intermediate and poor metabolizers compared to rapid metabolizers in patients with high CHADS-P2A2RC score (difference 3.6%) than in patients with low score (difference 0.6%). Bleeding was not affected by the CYP2C19 genotype. Similar results were obtained when applying the TRS 2*P (difference 2.8% vs. 0.5%), indicating that sufficient platelet inhibition is of much higher importance for high risk patients than for those with a better cardiovascular score. The question remains if all patients with acute coronary syndrome (having a high score) should always be treated with alternative ADP-receptor antagonists—as recommended by the international guidelines14—or if the CYP2C19 genotype result should be considered in prasugrel treatment only for CYP2C10 IM/PMs. As the authors explain, East Asian patients treated with prasugrel or ticagrelor suffer from bleeding more frequently, while there is no clear anti-ischemic benefit. This original study could therefore serve as a starting point for further studies that take into account risk scores in order to better exploit the advantages of stratified antithrombotic therapy. Ethnic diversity also has a major impact on the implementation of pharmacogenetics in clinical practice. This has been particularly evident in the treatment with cytostatic drugs such as fluoropyrimidines, taking into account the dihydropyrimidine dehydrogenase (DPYD) genotype,15 or thiopurines, taking into account thiopurine methyltransferase (TMPT) and nudix hydrolase 15 (NUDT15) genotypes, as recently updated in this journal.16 For the immunosuppressant tacrolimus, current guidelines consider the cytochrome P450 3A5 genotype (CYP3A5) to be the most important factor influencing the concentration-dose ratio.17 In 2022, the PharmVar Consortium redefined the star allele nomenclature for CYP3A5.18 In the European population, the common loss-of-function allele CYP3A5*3 leads to a deficiency in CYP3A5 activity in 80–85% of the population, while another loss-of-function allele (CYP3A5*6) is common in the sub-Saharan population but rare in Europeans. In this issue, a comprehensive study of nearly 1,500 kidney transplant recipients from diverse ethnic backgrounds showed that carriers of the CYP3A5*6 allele in the sub-Saharan African and Caribbean populations had higher tacrolimus metabolism rates compared to carriers of CYP3A5*3, resulting in lower dose-normalized area-under-the-concentration-time-curve (AUC) values at steady state.19 The differences could not be explained by variants in the CYP3A4 locus such as *22. This observation clearly shows that other factors in the regulation and/or activity of CYP3A and related enzymes must be taken into account in order to better predict dosing in patients of different ethnic origins. To date, monitoring trough concentrations or AUC of tacrolimus is still indispensable for optimal dosing to prevent rejection of kidney transplants. While tacrolimus dosing guided by CYP3A5 genotype represents a well-known and partly established example of individualized therapy, genetic testing for cisplatin-induced ototoxicity (CIO)—in contrast to aminoglycoside-induced ototoxicity—still remains far from clinical implementation. An estimated 48% of adults and 67% of children experience permanent hearing loss after cisplatin chemotherapy.20 While research on cisplatin ototoxicity has traditionally focused on systemic exposure and damage to sensory hair cells, a comprehensive review in this CPT issue shifts the spotlight to the stria vascularis, a tissue lining the cochlea, as an underexplored though key site of cisplatin-induced ototoxicity.21 Several studies have attempted to identify genetic variants contributing to CIO, including variants in the polyspecific organic cation transporter 2 (OCT2, SLC22A2). However, the predictive value of previously identified genetic markers has been unsatisfactory, and clinical risk factors still remain central to risk assessment. CIO is driven by a complex genetic profile, with risk arising from the combined effects of many genetic variants with individually small effects. In their review, Lazetic et al. provide a comprehensive overview of genetic variants highly expressed in the stria vascularis and significantly associated with CIO, including genes involved in cellular cisplatin uptake and accumulation in the stria vascularis (e.g., SLC22A2) or in DNA damage repair processes (e.g., ERCC2). Furthermore, a variety of additional variants that were not significantly associated with CIO in previous genome-wide association studies (GWAS) are discussed that may influence susceptibility through effects on the antioxidant system, endoplasmic reticulum stress response or inflammation. To generate more conclusive evidence and significant outcomes, larger studies with less heterogeneous cohorts are warranted. Besides, multi-omics technologies can enhance our understanding of cisplatin-induced transcriptomic and epigenomic changes, bearing the potential to improve prediction of CIO risk prior to treatment, and to identify novel otoprotective targets, moving pharmacogenomics from descriptive association toward actionable mechanistic insights. The FDA approval of sodium thiosulfate to reduce the risk of CIO in children with non-metastatic solid tumors has marked a clinical breakthrough, though it has limitations such as potential interference with cisplatin efficacy and restriction to pediatric patients.22, 23 This underscores the need for more specific genetic and molecular targets that could lead to the development of new therapies or the repurposing of approved drugs to mitigate COI and consequently improve quality of life. Precision medicine can only be as reliable as the data that form its foundation. Missing data are common in pharmacogenomic research and can complicate the reliability of study conclusions and ultimately dosing recommendations. Imputation for high-dimensional genetic datasets, covering millions of SNPs, has largely relied on single imputation and the use of one best-guess genotype per SNP, ignoring uncertainty. In this CPT issue, Asiimwe et al. now extend multiple imputation to high-dimensional SNP data, considering genotype probabilities (for each possible genotype at each SNP), imputation uncertainty and pre-imputation missingness based on genotype imputation server outputs.24 In contrast to single-imputation methods, only multiple imputation appeared accurate and reached adequate coverage, that is, a high proportion of 95% confidence intervals containing the true value of pharmacokinetic (PK) parameters. The comprehensive analysis investigates a wealth of imputation strategies, bias and precision metrics, missingness levels, and PGs-related missing data mechanisms (missing completely at random: same probability of missingness for all individuals; missing at random: missing SNPs more likely for males; missing not at random: mutant-allele carriers 3x more likely than wild-types). As a further strength, the imputation framework was not only applied to simulated data, but also validated based on real-world clinical datasets and the UK Biobank, showing that multiple imputation could recover known pharmacogenomic associations, reduce false positives, and detect signals missed by single imputation. Three additional insights from the study are particularly noteworthy. Contrary to common practice excluding SNPs with low genotyping rates or missing data,25 properly handled incomplete data appeared preferable to omission. Second, the multiple imputation approach came at only modest additional computational costs, enabling a workflow comparable to single imputation. Third, large language models (Open AI, GPT-4o) performed poorly on the imputation task (producing similar results to mode imputation), serving as a reminder to carefully scrutinize and not overrate their outputs for critical data-imputation tasks. Imputation of missing data is commonly followed by covariate selection, another key component investigated by the authors. Here, machine learning and dimensionality reduction offered support for covariate selection, but their benefits for SNP selection were less consistent compared to traditional GWAS. The multiple imputation study was exemplified using a PK dataset on tuberculosis treatment together with SNP data from the Luhya population in Kenya (1,000 Genomes Project),26 and focused on the assessment of fixed-effects PK parameters. Although the promising validation results based on clinical data suggest a high potential for the approach, further evaluations using more complex models and including random effects are desirable to confirm its generalizability. Validated multiple imputation frameworks could then offer the opportunity to reanalyze clinical trial data with incomplete genetic samples, potentially uncovering previously overlooked dose–response biomarkers. The articles published in this issue of CPT highlight continued progress in pharmacogenetic research, including efforts to refine nomenclatures, consider genetic variants related to ethnic diversity, and use novel imputation tools in genome-wide analyses. In addition, there is still a need for a better understanding of the contribution of genetic variation to adverse drug reactions, such as cisplatin-induced ototoxicity. The biggest challenge remains answering the question of which patients will benefit significantly from drug selection and dosage adjustment. Prominent examples include the CPIC and Dutch Pharmacogenetic Working Group (DPWG) guidelines, which have become established in clinical practice and are increasingly recognized by regulatory authorities. Some refer to these examples as “low hanging fruit,” but there remains a need to evaluate the benefits in other diseases, particularly with regard to severity and stage, as nicely illustrated by one of the articles published in this issue. Thus, it is more than just “low hanging fruit.” No funding was received for this work. The authors declared no competing interests for this work.