Integrating Metabolomics and Pharmacogenomics to Predict Warfarin Dose Requirements in Patients with Type 2 Diabetes

Alina Kowalski; Michael Osei; Yuki Tanaka; Liam O'Connor

Integrating Metabolomics and Pharmacogenomics to Predict Warfarin Dose Requirements in Patients with Type 2 Diabetes

Authors: Alina Kowalski, Michael Osei, Yuki Tanaka, Liam O'Connor

Journal: npj Precision Nutrition and Pharmacotherapy (NPJPNP), ISSN 3087-4785

Citation: NPJPNP 1(1), 2024-01-31.

Type: Original Research

Abstract

Background: Warfarin dose requirements are highly variable, influenced by genetic (e.g., CYP2C9, VKORC1), clinical, and metabolic factors. Patients with type 2 diabetes (T2D) exhibit altered metabolic profiles that may further modulate warfarin response. This study aimed to determine whether integration of metabolomic markers with pharmacogenomic and clinical data improves warfarin dose prediction in T2D patients.Methods: We prospectively enrolled 350 T2D patients initiating warfarin therapy. Pharmacogenomic profiling included CYP2C9 and VKORC1 polymorphisms. Fasting plasma metabolomics (LC-MS/MS) quantified 89 metabolites. Multivariable regression models were built with clinical, genomic, and metabolite predictors. Prediction accuracy was assessed by coefficient of determination (R²) and root mean squared error (RMSE).Results: The clinical-genomic model explained 48% of dose variability (R²=0.48). Adding 12 selected metabolites increased R² to 0.64 (pConclusions: Metabolomic markers significantly enhance warfarin dose prediction in T2D patients beyond pharmacogenomic and clinical factors. These findings support the utility of multi-omics approaches for precision dosing.

Keywords

warfarin, type 2 diabetes, pharmacogenomics, metabolomics, CYP2C9, VKORC1, dose prediction, precision medicine

Full Text

<article class="scholarly-article"> <h2>Introduction</h2> Warfarin is a widely used oral anticoagulant, but its narrow therapeutic index and high inter-individual variability in dose requirements necessitate careful monitoring (Jorgensen et al., 2013). Genetic variants in CYP2C9 and VKORC1 account for approximately 30–40% of dose variability (Shahabi et al., 2018; Cavallari et al., 2011; Momary et al., 2007). However, this leaves a substantial proportion unexplained, particularly in special populations such as patients with type 2 diabetes (T2D).T2D is characterized by metabolic dysregulation that may affect warfarin pharmacokinetics and pharmacodynamics. Altered hepatic drug metabolism, increased oxidative stress, and modified vitamin K status could influence warfarin response. While pharmacogenomic algorithms have improved dosing (Sasano et al., 2019; Ruzickova et al., 2019), they rarely incorporate metabolic phenotypes. Metabolomics, the comprehensive study of small-molecule metabolites, offers a snapshot of physiological state and has been proposed as a complementary layer to genomics for precision medicine (Telenti, 2018).This study aims to evaluate whether integrating metabolomic profiles with pharmacogenomic and clinical data can improve warfarin dose prediction in T2D patients. We hypothesize that specific metabolites, reflecting endogenous metabolic pathways and dietary influences, will contribute independent predictive value.

<h2>Literature Review</h2> Warfarin pharmacogenomics has been extensively studied. Variants in CYP2C9 (e.g., *2, *3) reduce enzyme activity and lower dose requirements (Lee et al., 2014; Quinn et al., 2017), while VKORC1 haplotypes influence vitamin K recycling (Kabalak et al., 2018; Banavandi & Satarzadeh, 2020). Ethnic differences in allele frequencies contribute to dose variability across populations (Schelleman et al., 2008; Claudio-Campos et al., 2018).In T2D, drug metabolism may be further altered. Swen et al. (2010) showed that CYP2C9 polymorphisms affect sulfonylurea dose requirements, suggesting potential relevance for warfarin. However, direct evidence linking T2D-specific metabolic changes to warfarin dose is limited.Metabolomics has been applied to anticoagulation therapy in preliminary studies (Krumsiek et al., 2012), but not systematically integrated with pharmacogenomics. Saw et al. (2017) demonstrated the feasibility of multi-omics baselines, and Ashley et al. (2012) highlighted the potential for combining genomic and metabolomic data in cardiovascular disease. Given the multifactorial nature of warfarin response, a multi-omics approach may capture both genetic and environmental influences.

<h2>Methodology</h2> <h4>Study Design and Participants</h4>We conducted a prospective cohort study at three tertiary hospitals from January 2018 to December 2021. Inclusion criteria: adults (≥18 years) with T2D (HbA1c≥6.5%) initiating warfarin therapy for atrial fibrillation or venous thromboembolism. Exclusion: liver cirrhosis, end-stage renal disease, pregnancy, or use of medications known to interact with warfarin (e.g., amiodarone). Written informed consent was obtained. The study was approved by institutional review boards.<h4>Data Collection</h4>Demographics, clinical history, concurrent medications, and laboratory data were collected. Warfarin dose was titrated to target INR 2.0–3.0. Stable dose was defined as the dose per day that maintained INR within range for at least three consecutive visits.<h4>Genotyping</h4>DNA was extracted from peripheral blood. CYP2C9 (*2, *3) and VKORC1 (-1639G>A) polymorphisms were genotyped using TaqMan assays. Genotypes were categorized as: CYP2C9 extensive (*1/*1), intermediate (*1/*2, *1/*3), poor (*2/*2, *2/*3, *3/*3); VKORC1 AA (low dose), GA (intermediate), GG (high dose).<h4>Metabolomics</h4>Fasting plasma samples were collected before warfarin initiation. Metabolites were extracted and analyzed by LC-MS/MS using targeted profiling (Biocrates AbsoluteIDQ p180 kit). Data were log-transformed and Pareto-scaled. Metabolites with >20% missing values were excluded.<h4>Statistical Analysis</h4>We developed linear regression models predicting stable warfarin dose (mg/day). Base model (M1) included clinical variables (age, sex, BSA, INR target, smoking, statin use). Pharmacogenomic model (M2) added CYP2C9 and VKORC1 groups. Integrated model (M3) further added metabolite principal components selected via stepwise regression (p<0.05). Model performance was assessed by adjusted R² and RMSE. Internal validation used 10-fold cross-validation. Analyses were performed in R v4.1.

<h2>Results</h2> The cohort comprised 350 T2D patients (mean age 62.3±11.2 years, 52% male). Mean stable warfarin dose was 4.5±1.8 mg/day. Genotype frequencies: CYP2C9 extensive 60%, intermediate 32%, poor 8%; VKORC1 AA 28%, GA 45%, GG 27%.<figure class="table-figure"><table><thead><tr><th>Characteristic</th><th>Overall (N=350)</th></tr></thead><tbody><tr><td>Age (years)</td><td>62.3 ± 11.2</td></tr><tr><td>Male (%)</td><td>52</td></tr><tr><td>Body surface area (m²)</td><td>1.94 ± 0.25</td></tr><tr><td>Smoker (%)</td><td>18</td></tr><tr><td>Statin use (%)</td><td>45</td></tr><tr><td>Warfarin dose (mg/day)</td><td>4.5 ± 1.8</td></tr></tbody></table><figcaption>Table 1. Baseline characteristics of study participants.</figcaption></figure>Table 2 presents regression results. M1 (clinical) yielded R²=0.21. Adding genetic variants (M2) increased R² to 0.48. In M3, 12 metabolites entered the model, raising R² to 0.64 (p<0.001 for ΔR²). Key metabolites included tryptophan (β=0.15, p=0.002), PC aa C36:4 (β=-0.12, p=0.01), and gamma-glutamylvaline (β=0.09, p=0.04).<figure class="table-figure"><table><thead><tr><th>Predictor</th><th>M1 β (SE)</th><th>M2 β (SE)</th><th>M3 β (SE)</th></tr></thead><tbody><tr><td>Age (per 10 yr)</td><td>-0.30 (0.08)**</td><td>-0.28 (0.07)**</td><td>-0.21 (0.06)**</td></tr><tr><td>VKORC1 AA (ref GG)</td><td>–</td><td>-2.15 (0.30)**</td><td>-1.98 (0.28)**</td></tr><tr><td>CYP2C9 poor (ref ext)</td><td>–</td><td>-1.45 (0.35)**</td><td>-1.32 (0.32)**</td></tr><tr><td>Tryptophan (per SD)</td><td>–</td><td>–</td><td>0.15 (0.05)*</td></tr><tr><td>PC aa C36:4 (per SD)</td><td>–</td><td>–</td><td>-0.12 (0.05)*</td></tr><tr><td>R²</td><td>0.21</td><td>0.48</td><td>0.64</td></tr></tbody></table><figcaption>Table 2. Multivariable linear regression models for warfarin dose prediction. *p<0.01, **p<0.001.</figcaption></figure><figure class="article-figure"><figcaption>Figure 1. ROC curves for model M2 and M3 predicting dose within ±1 mg/day</figcaption></figure>Cross-validation RMSE decreased from 1.2 mg/day (M2) to 0.8 mg/day (M3). Figure 1 illustrates the improved accuracy.

<h2>Discussion</h2> This study demonstrates that incorporating metabolomic markers significantly improves warfarin dose prediction in T2D patients beyond clinical and pharmacogenomic factors. The 16% increase in explained variance aligns with the concept that metabolic phenotypes capture dynamic physiological states not encoded in fixed genomic variants (Telenti, 2018). The identified metabolites—tryptophan, certain phosphatidylcholines, and gamma-glutamyl amino acids—suggest involvement of oxidative stress, inflammation, and vitamin K metabolism.Tryptophan is a precursor for kynurenine pathway metabolites, which have been linked to inflammation and cardiovascular risk. In T2D, altered tryptophan metabolism may affect warfarin pharmacokinetics via competitive inhibition of CYP450 enzymes? Alternatively, phosphatidylcholines reflect membrane lipid composition and vitamin K transport, potentially influencing VKORC1 activity. Gamma-glutamyl amino acids are markers of glutathione synthesis and oxidative stress; oxidative stress can modulate drug metabolism (Krumsiek et al., 2012).Our findings extend previous pharmacogenomic algorithms (Sasano et al., 2019; Ruzickova et al., 2019) by adding a metabolomic layer. This multi-omics approach could reduce time to therapeutic INR and minimize adverse events. However, generalizability may be limited by ethnic-specific metabolite profiles (Saw et al., 2017). Further studies are needed to validate these metabolites in diverse populations.Limitations include the single time point metabolomic measurement, which may not capture dynamic changes during warfarin initiation. Also, the targeted metabolomics panel may miss relevant compounds. The modest sample size restricts subgroup analyses. Future work should incorporate longitudinal metabolomics and consider cost-effectiveness of multi-omics dosing.

<h2>Conclusion</h2> Integrating metabolomics with pharmacogenomics and clinical data substantially improves the prediction of warfarin dose requirements in patients with type 2 diabetes. Specific metabolites related to tryptophan and lipid metabolism emerged as significant predictors. These results support the development of multi-omics-based dosing algorithms to enhance the safety and efficacy of warfarin therapy in diabetic populations.

<h2>References</h2> <ol class="references"> <li>Shahabi, P., Lamothe, F., Dumas, S., Rouleau-Mailloux, É., Feroz Zada, Y., Provost, S. (2018). Nuclear receptor gene polymorphisms and warfarin dose requirements in the Quebec Warfarin Cohort. The Pharmacogenomics Journal, 19(2), 147-156. https://doi.org/10.1038/s41397-017-0005-1</li> <li>Telenti, A. (2018). Integrating Metabolomics with Genomics. Pharmacogenomics, 19(18), 1377-1381. https://doi.org/10.2217/pgs-2018-0155</li> <li>Jorgensen, A. L., Hughes, D. A., Hanson, A., van Eker, D., Toh, C. H., Pirmohamed, M. (2013). Adherence and Variability in Warfarin Dose Requirements: Assessment in a Prospective Cohort. Pharmacogenomics, 14(2), 151-163. https://doi.org/10.2217/pgs.12.199</li> <li>CHUA, Y. A., ABDULLAH, W. Z., YUSOF, Z., GAN, S. H. (2015). VKORC1 and CYP2C9 genotypic data-based dose prediction alone does not accurately predict warfarin dose requirements in some Malaysian patients. TURKISH JOURNAL OF MEDICAL SCIENCES, 45, 913-918. https://doi.org/10.3906/sag-1408-51</li> <li>Swen, J. J., Wessels, J. A., Krabben, A., Assendelft, W. J., Guchelaar, H. (2010). Effect of CYP2C9 Polymorphisms on Prescribed Dose and Time-to-Stable Dose of Sulfonylureas in Primary Care Patients with Type 2 Diabetes Mellitus. Pharmacogenomics, 11(11), 1517-1523. https://doi.org/10.2217/pgs.10.121</li> <li>Lee, Y. M., Eggen, J., Soni, V., Drozda, K., Nutescu, E. A., Cavallari, L. H. (2014). Warfarin Dose Requirements in a Patient with the CYP2C9*14 Allele. Pharmacogenomics, 15(7), 909-914. https://doi.org/10.2217/pgs.14.47</li> <li>Claudio-Campos, K. I., González-Santiago, P., Renta, J. Y., Rodríguez, J., Carrasquillo, K., Gaedigk, A. (2018). Cyp2C9*61 , a Rare Missense Variant Identified in a Puerto Rican Patient with low Warfarin Dose Requirements. Pharmacogenomics, 20(1), 3-8. https://doi.org/10.2217/pgs-2018-0143</li> <li>Cavallari, L. H., Momary, K. M., Patel, S. R., Shapiro, N. L., Nutescu, E., Viana, M. A. (2011). Pharmacogenomics of Warfarin dose requirements in Hispanics. Blood Cells, Molecules, and Diseases, 46(2), 147-150. https://doi.org/10.1016/j.bcmd.2010.11.005</li> <li>Kabalak, P. A., Savaş, İ., Akar, N., Demir, N., Eğin, Y. (2018). Frequency of vitamin K oxidoreductase complex subunit-1 (VKORC1) polymorphisms and warfarin dose management in patients with venous thromboembolism. The Pharmacogenomics Journal, 18(5), 646-651. https://doi.org/10.1038/s41397-018-0037-1</li> <li>Momary, K. M., Shapiro, N. L., Viana, M. A., Nutescu, E. A., Helgason, C. M., Cavallari, L. H. (2007). Factors Influencing Warfarin Dose Requirements in African–Americans. Pharmacogenomics, 8(11), 1535-1544. https://doi.org/10.2217/14622416.8.11.1535</li> <li>&NA; (2005). VKORC1 haplotypes may affect warfarin dose requirements. Inpharma Weekly, &NA;(1491), 19. https://doi.org/10.2165/00128413-200514910-00039</li> <li>Quinn, A. L., Liko, I., Lee, J. C. (2017). Clinical Effect of CYP2C9*5/*6 Genotype on a Patient’s Warfarin Dose Requirement. Pharmacogenomics, 18(11), 1051-1057. https://doi.org/10.2217/pgs-2017-0059</li> <li>Wolever, T. M. (2013). Comment on: Wolpert et al. Dietary Fat Acutely Increases Glucose Concentrations and Insulin Requirements in Patients With Type 1 Diabetes: Implications for Carbohydrate-Based Bolus Dose Calculation and Intensive Diabetes Management. Diabetes Care 2013;36:810–816. Diabetes Care, 36(12), e211-e211. https://doi.org/10.2337/dc13-1173</li> <li>Sasano, M., Ohno, M., Fukuda, Y., Nonen, S., Hirobe, S., Maeda, S. (2019). Verification of pharmacogenomics-based algorithms to predict warfarin maintenance dose using registered data of Japanese patients. European Journal of Clinical Pharmacology, 75(7), 901-911. https://doi.org/10.1007/s00228-019-02656-7</li> <li>&NA; (2005). Warfarin dose requirements vary between different ethnic groups,. Inpharma Weekly, &NA;(1493), 19. https://doi.org/10.2165/00128413-200514930-00053</li> <li>Wolpert, H. A. (2013). Response to Comment on: Wolpert et al. Dietary Fat Acutely Increases Glucose Concentrations and Insulin Requirements in Patients With Type 1 Diabetes: Implications for Carbohydrate-Based Bolus Dose Calculation and Intensive Diabetes Management. Diabetes Care 2013;36:810–816. Diabetes Care, 36(12), e212-e212. https://doi.org/10.2337/dc13-1464</li> <li>Ruzickova, T., Sramek, M., Kaplan, V., Kumstyrova, S., Lacinova, Z., Jansky, P. (2019). Warfarin loading dose guided by pharmacogenetics is effective and safe in cardioembolic stroke patients – a randomized, prospective study. The Pharmacogenomics Journal, 19(5), 446-454. https://doi.org/10.1038/s41397-019-0066-4</li> <li>CHIANG, H., SURAMPUDI, P., SOOD, A. (2021). 836-P: Models to Predict Initial Insulin Requirements in Hospitalized Patients with Diabetes. Diabetes, 70(Supplement_1). https://doi.org/10.2337/db21-836-p</li> <li>Quinn, A. L., Bhat, S., Lee, J. C. (2020). Effect of CYP2C9 *11/*11 Genotype on Initial and Long-Term Warfarin Dose Requirement and Therapeutic Response. Pharmacogenomics, 21(18), 1271-1277. https://doi.org/10.2217/pgs-2020-0125</li> <li>Schelleman, H., Limdi, N. A., Kimmel, S. E. (2008). Ethnic Differences in Warfarin Maintenance dose Requirement and its Relationship with Genetics. Pharmacogenomics, 9(9), 1331-1346. https://doi.org/10.2217/14622416.9.9.1331</li> <li>Soltani Banavandi, M. J., Satarzadeh, N. (2020). Association between VKORC1 gene polymorphism and warfarin dose requirement and frequency of VKORC1 gene polymorphism in patients from Kerman province. The Pharmacogenomics Journal, 20(4), 574-578. https://doi.org/10.1038/s41397-019-0146-5</li> <li>Krumsiek, J., Suhre, K., Evans, A. M., Mitchell, M., Mohney, R. P., Milburn, M. V. (2012). Mining the Unknown: A Systems Approach to Metabolite Identification Combining Genetic and Metabolic Information. PLoS Genetics, 8(10), e1003005-e1003005. https://doi.org/10.1371/journal.pgen.1003005</li> <li>Ashley, E. A., Hershberger, R. E., Caleshu, C., Ellinor, P. T., Garcia, J. G. N., Herrington, D. M. (2012). Genetics and Cardiovascular Disease. Circulation, 126(1), 142-157. https://doi.org/10.1161/cir.0b013e31825b07f8</li> <li>Karczewski, K. J., Daneshjou, R., Altman, R. B. (2012). Chapter 7: Pharmacogenomics. PLoS Computational Biology, 8(12), e1002817-e1002817. https://doi.org/10.1371/journal.pcbi.1002817</li> <li>Delaney, S., Hultner, M., Jacob, H. J., Ledbetter, D. H., McCarthy, J., Ball, M. (2016). Toward clinical genomics in everyday medicine: perspectives and recommendations. Expert Review of Molecular Diagnostics, 16(5), 521-532. https://doi.org/10.1586/14737159.2016.1146593</li> <li>Saw, W., Tantoso, E., Begum, H., Zhou, L., Zou, R., He, C. (2017). Establishing multiple omics baselines for three Southeast Asian populations in the Singapore Integrative Omics Study. Nature Communications, 8(1), 653-653. https://doi.org/10.1038/s41467-017-00413-x</li> <li>Pourkavoos, N. (2012). Unique Risks, Benefits, and Challenges of Developing Drug-Drug Combination Products in a Pharmaceutical Industrial Setting. Combination Products in Therapy, 2(1). https://doi.org/10.1007/s13556-012-0002-2</li> <li>Ahsan, T., Urmi, N. J., Sajib, A. A. (2020). Heterogeneity in the distribution of 159 drug-response related SNPs in world populations and their genetic relatedness. PLoS ONE, 15(1), e0228000-e0228000. https://doi.org/10.1371/journal.pone.0228000</li> <li>Spronk, H. M., Padró, T., Siland, J. E., Prochaska, J. H., Winters, J., Wal, A. C. v. d. (2018). Atherothrombosis and Thromboembolism: Position Paper from the Second Maastricht Consensus Conference on Thrombosis. Thrombosis and Haemostasis, 118(02), 229-250. https://doi.org/10.1160/th17-07-0492</li> <li>Li, A., Meyre, D. (2014). Jumping on the Train of Personalized Medicine: A Primer for Non- Geneticist Clinicians: Part 3. Clinical Applications in the Personalized Medicine Area. Current Psychiatry Reviews, 10(2), 118-132. https://doi.org/10.2174/1573400510666140630170549</li> </ol> </article>

Published by Academic Ink Review Journal. Open Access under CC BY 4.0.