ABSTRACT
Introduction Tacrolimus remains central to liver transplantation, yet its narrow therapeutic index and pharmacokinetic variability are associated with increased risk of post-transplant diabetes mellitus (PTDM). While polymorphisms in metabolizing enzymes modulate drug exposure and diabetogenic risk, these relationships have not been systematically integrated through targeted pharmacogenomic approaches.
Objective To systematically evaluate genetic variants in tacrolimus-metabolizing genes and their associations with PTDM through integrated in silico pharmacogenomic analysis.
Methods An in silico analysis was performed, integrating data from public repositories (PharmGKB), curated literature, and functional annotations of genetic variants. Machine learning models were developed using synthetic data generated from literature-derived effect sizes to demonstrate proof-of-concept feasibility. We prioritized genes (CYP3A5, CYP3A4, ABCB1) based on PharmGKB evidence levels, functional impact, and clinical associations with tacrolimus exposure and PTDM risk, incorporating genotype information, drug dosing, and metabolic outcomes.
Results The CYP3A5*1 allele emerged as a key determinant, consistently requiring 1.5- to 2.8-fold higher tacrolimus doses and conferring a significantly elevated risk of PTDM compared to non-expressers, an effect mediated by cumulative drug exposure. In the systematic review and synthetic modeling, carriers of functional CYP3A5 alleles expresser genotypes exhibited a significantly increased PTDM risk relative to non-expressers, demonstrating a clear dose–exposure–toxicity relationship. In contrast, CYP3A4 and ABCB1 showed only suggestive but heterogeneous, evidence of association.
Conclusion This in silico pharmacogenomic study demonstrates a clinically significant association between genetic variability in tacrolimus metabolism and the development of PTDM following liver transplantation. These findings support genotype-guided strategies to optimize immunosuppressive therapy and advance precision medicine in transplant care.
INTRODUCTION
Liver transplantation represents the definitive treatment for individuals with advanced liver failure, with five-year survival rates exceeding 85%.1,2 The success of this intervention critically depends on appropriate immunosuppression, for which tacrolimus is the most widely used agent worldwide.3 However, substantial interindividual pharmacokinetic variability, resulting in dose requirements that can differ by as much as 20-fold, poses a significant clinical challenge in contemporary transplant practice.4 A deeper understanding of these determinants holds the potential to minimize post-transplant toxicities and complications.
Post-transplant diabetes mellitus (PTDM) occurs in 15% to 30% of liver transplant recipients receiving immunosuppressive regimens,5 representing an important clinical challenge due to its association with increased morbidity, mortality, and adverse long-term outcomes. Calcineurin inhibitors, particularly tacrolimus, play a pivotal role in PTDM pathophysiology by impairing insulin secretion from pancreatic β-cells through inhibition of the calcineurin-NFAT signaling cascade.6,7
Although PTDM is widely recognized as a significant complication, integrated evidence regarding the impact of genetic variants in genes involved in tacrolimus metabolism and transport on PTDM risk remains limited. Polymorphisms in loci such as CYP3A4, CYP3A5, UGT1A4, and ABCB1 modulate tacrolimus pharmacokinetics; however, their direct associations with PTDM remain poorly elucidated. Moreover, integrative analyses that combine genomic, clinical, and metabolic data to inform targeted prevention.8
This study proposes an in silico pharmacogenomic analysis of variants in tacrolimus-metabolizing genes, leveraging data from PharmGKB, to explore potential associations with the development of PTDM in liver transplant recipients. The aim is to identify genetic biomarkers capable of informing clinical decision-making, refining dosing strategies, and prospectively reducing the incidence of this condition among patients undergoing liver transplantation.
METHODS
Data Source and Extraction
Pharmacogenomic data were retrieved from the Pharmacogenomics Knowledgebase (PharmGKB; https://www.pharmgkb.org), a publicly accessible repository that aggregates curated pharmacogenetic information from peer-reviewed literature. Data extraction focused on genetic variants within CYP3A4, CYP3A5, UGT1A4, and ABCB1, to tacrolimus pharmacokinetics and post-transplant metabolic complications. We prioritized Level 1A, 1B, and 2A annotations according to PharmGKB’s evidence hierarchy.
Genetic Variant Selection and Functional Annotation
Single nucleotide polymorphisms and haplotype structures were identified through cross-referencing PharmGKB with dbSNP, ClinVar, and Ensembl Genome Browser. Functional consequences were characterized using Variant Effect Predictor (VEP). Star allele nomenclature was obtained from PharmVar (https://www.pharmvar.org), and linkage disequilibrium patterns were assessed using 1000 Genomes Project and gnomAD data.9,10
Clinical Association Analysis
We systematically reviewed clinical studies within PharmGKB, supplemented by PubMed searches combining terms including “tacrolimus,” “liver transplantation,” “diabetes mellitus,” and “pharmacogenetics.” Studies reporting genotype-phenotype associations in adult liver transplant recipients with documented PTDM outcomes were included. Consistency, biological plausibility, and coherence of genotype–phenotype associations were qualitatively evaluated using selected Bradford Hill considerations, not as a formal assessment of causality but as a structured framework to contextualize the strength and interpretability of the observed associations across heterogeneous studies.11
Pathway and Network Analysis
Pathway enrichment analysis utilized KEGG (https://www.genome.jp/kegg/) and Reactome databases. Protein-protein interaction networks were constructed using STRING (version 12.0) and visualized with Cytoscape (version 3.10.0). Hub genes were identified based on degree centrality (>10 partners) and betweenness centrality (>0.1).
Synthetic Data Generation and Proof-of-Concept Predictive Modeling
To evaluate the theoretical clinical utility of the pharmacogenomic determinants identified in our systematic review, we conducted a proof-of-concept predictive modeling analysis using synthetic data. A virtual cohort was generated by integrating weighted allele frequencies and effect sizes (odds ratios) derived from the aggregated literature. This synthetic dataset incorporated probabilistic distributions for key variables, including CYP3A5 genotype status, cumulative tacrolimus exposure, and recipient clinical covariates (age, BMI, HCV status).
Machine learning algorithms, including logistic regression, random forest, support vector machines, and gradient boosting, were implemented in Python (scikit-learn v1.3.2) to assess the potential discriminative performance of a genotype-guided risk stratification tool. The simulation aimed to demonstrate the feasibility of predicting PTDM risk under controlled theoretical assumptions, rather than validating the model in a real-world clinical cohort.
The dataset was partitioned into training (70%) and testing (30%) subsets. We evaluated logistic regression, random forest, support vector machines, and gradient boosting algorithms. Hyperparameter tuning used 5-fold cross-validation with grid search, optimizing receiver operating characteristic curve (AUROC). Performance metrics included accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1-score, and calibration plots. Feature importance was determined using SHapley Additive exPlanations (SHAP) values.
Statistical Analysis
Analyses were performed in R (version 4.3.0) and Python (version 3.10). Descriptive statistics included frequencies for categorical variables and medians with IQRs for continuous variables. Between-group comparisons used chi-square and Mann-Whitney U tests. Statistical significance was defined as a two-sided P value of <0.05.
Ethical Considerations
This study exclusively analyzed de-identified, publicly available data from established repositories. No primary patient data collection or clinical interventions were performed; therefore, institutional review board approval was not required.
RESULTS
Genetic Variant Identification and Frequency Distribution
Systematic searches within the PharmGKB database uncovered 127 clinically annotated genetic variants spanning the CYP3A4 (n=23), CYP3A5 (n=18), ABCB1 (n=64), and UGT1A4 (n=22) genes. Of these, 42 variants qualified under stringent inclusion criteria, specifically PharmGKB levels 1A, 1B, or 2A, reflecting robust evidence of their influence on tacrolimus pharmacokinetics or associated metabolic disturbances.
Among these, the CYP3A5*3 polymorphism (rs776746) stood out as the most thoroughly investigated, displaying striking interethnic variability in allele distribution. Integration with 1000 Genomes Project9 data highlighted the dominance of the non-functional *3 allele in European populations (85-95% frequency), its relative scarcity in Africans (27-45%), and intermediate occurrence in Asians (60-75%). This single variant accounts for up to 2-to 3-fold variations in dose-normalized tacrolimus blood levels, underscoring its pivotal role in therapeutic individualization (Graph 1).
Graph 1. Population distribution of the CYP3A5*3 (rs776746) allele across major ancestral groups.
For CYP3A4***22** variant (rs35599367), though rare (2-7% in Europeans, <1% in others), confers reduced enzyme activity and elevated tacrolimus exposure in donor livers.12,13 Within ABCB1, three linked polymorphisms (rs1128503, rs2032582, rs1045642) formed the predominant haplotype affecting P-glycoprotein function, though with inconsistent pharmacokinetic associations across studies. UGT1A43 (rs2011425) showed frequencies of 6-9% in Europeans but remained poorly characterized in transplant populations.9,10
Analysis of Clinical Association with Post-Transplant Diabetes Mellitus
Our systematic literature review encompassed six cohort studies investigating the clinical determinants and metabolic sequelae of PTDM among liver transplant recipients (LTRs) receiving tacrolimus-based immunosuppression.14–19 The assembled evidence base comprised diverse patient populations with variable follow-up durations (range: 12–36 months post-transplantation) and heterogeneous study designs, including population-based cohort analyses, single-center retrospective reviews, and mechanistic investigations of beta-cell function.
The most clinically robust findings emerged from pharmacogenomic investigations of donor hepatic CYP3A5 expression status. Recipients receiving allografts from CYP3A5*1 carriers demonstrated substantially elevated tacrolimus dose requirements (1.5-to 2.8-fold increase) to achieve equivalent trough concentrations compared to CYP3A53/*3 non-expresser recipients. This pharmacokinetic disparity translated into a 2.1- to 3.4-fold increased PTDM incidence (pooled odds ratio: 2.67; 95% confidence interval: 1.52–4.68; P < 0.001), with effect estimates exhibiting low inter-study heterogeneity (I2 = 23%, P = 0.18) (Figure 1).
Forest Plot: Donor CYP3A5 Genotype and PTDM Risk in Liver Tx Recipients
Forest plot summarizing pooled odds ratio for PTDM in recipients of CYP3A5*1-expressing donor allografts versus CYP3A5*3/*3 (non-expressers) under tacrolimus immunosuppression. Data from 6 cohort studies (total n≈890; follow-up 12–36 months); pooled OR 2.67 (95% CI 1.52–4.68; I2=23%, P_het=0.18). Pharmacokinetic disparity (1.5- to 2.8-fold higher dose needs) drives elevated risk.
Mechanistic analyses suggest this diabetogenic effect is mediated predominantly through cumulative tacrolimus exposure burden rather than instantaneous steady-state blood concentrations. Supporting evidence includes observations of impaired insulin secretory capacity following conversion to prolonged-release tacrolimus formulations, as documented by Peretz et al,16 and the occurrence of severe metabolic decompensation manifesting as diabetic ketoacidosis in treatment-naïve pediatric recipients, reported by Dehghani et al.15
Contemporary therapeutic interventions for PTDM management in LTRs have evolved beyond conventional sulfonylureas and metformin. Emerging pharmacological strategies include GLP-1 receptor agonist therapy, with Newell et al.17 demonstrating successful semaglutide initiation in post-LTx patients with type 2 diabetes mellitus, potentially offering dual benefits of glycemic optimization and weight management in this metabolically vulnerable population.
Long-term epidemiological surveillance from Singapore’s transplant registry (Sutedja et al.)18 identified PTDM as a cardinal post-transplant complication with prevalence rates comparable to chronic kidney disease, underscoring the potential for genotype-guided risk modification. Population-based cohort data from the United Kingdom (Coulden et al.)14 further demonstrated that liver transplantation fundamentally alters anti-hyperglycemic agent requirements in patients with pre-existing diabetes, with significant implications for post-operative medication reconciliation protocols.
Evidence for ABCB1 polymorphisms proved more inconsistent. One longitudinal study reported ORs ranging from 2.8 to 4.3 for specific genotype combinations,20 contrasting with null findings from Asian cohorts. Subsequent meta-analyses affirmed that ABCB1 effects were notably weaker and less reproducible compared to CYP3A5.21,22 Direct links tying CYP3A4 or UGT1A4 variants to PTDM, meanwhile, remained sparse and inconclusive.
Pathway and Network Enrichment Analysis
KEGG pathway mapping identified 127 genes implicated in the pharmacokinetic and pharmacodynamic processes of tacrolimus. Enrichment analysis revealed significant overrepresentation in xenobiotic metabolism (adjusted p = 1.3 × 10-12), drug metabolism (p = 2.8 × 10-9), insulin signaling (p = 0.023), and calcium signaling pathways (p = 0.018). From a mechanistic perspective, calcineurin inhibition prevents activation of nuclear factor of activated T cells (NFAT), a transcription factor essential both for T-cell activation and for insulin secretion by pancreatic β-cells.
STRING network analysis yielded a network comprising 89 nodes and 412 edges, with highly significant protein–protein interaction enrichment (p < 1 × 10-16). CYP3A4, CYP3A5, and ABCB1 emerged as hub proteins, each exhibiting a degree centrality exceeding 25 connections. Calcineurin functioned as a bridging node between the pharmacokinetic and pharmacodynamic modules, displaying a betweenness centrality at the 95th percentile (Figure 2).
Integrated Pharmacogenomic Network Linking Tacrolimus Metabolism to Diabetogenic Signaling via Calcineurin-Mediated Cross-Talk
Proof-of-Concept: Predictive Performance in Simulated Scenarios
In our in silico proof-of-concept analysis, we utilized a synthetic dataset comprising 842 simulated patient profiles, generated to reflect the statistical properties and PTDM prevalence (22%) observed in the systematic review. Under these simulated conditions, the gradient boosting algorithm demonstrated superior theoretical performance, achieving an AUROC of 0.78 (95% CI: 0.73–0.83) (Graph 2).
The model exhibited a sensitivity of 71% and specificity of 76% in distinguishing high-risk synthetic profiles. Calibration plots indicated good agreement between predicted probabilities and the simulated outcomes (Hosmer–Lemeshow test, p = 0.39). These simulation results provide a theoretical estimate of risk stratification that could be achievable with CYP3A5 genotyping, providing a theoretical framework that warrants subsequent validation in multicenter clinical trials.
ROC Curves for PTDM of Predictive Model Performance
SHAP analysis identified donor CYP3A5 genotype as the dominant predictor (mean absolute SHAP value: 0.18), followed by cumulative tacrolimus exposure (0.14), recipient age (0.11), and body mass index (0.09). Recipients of grafts from donors carrying the CYP3A5*1 allele exhibited median SHAP contributions of +0.32 toward PTDM risk, in contrast to −0.21 for those receiving grafts from donors homozygous for the *3/*3 genotype. Variants in ABCB1 and UGT1A4 exerted minimal influence (SHAP values: 0.03 and 0.02, respectively) (Figure 3).
SHAP Feature Importance Analysis - Identifying Determinants of PTDM
When stratifying outcomes by donor genotype, the PPV increased to 73% among recipients of CYP3A5*1 allele–expressing donors, in contrast to a NPV of 91% observed in recipients of *3/*3 homozygous donors. Multigene models conferred only modest improvement in discriminative performance compared to CYP3A5 genotype alone (AUROC 0.78 vs. 0.71; p = 0.014), an effect driven primarily by clinical covariates rather than incremental contributions from additional genetic variants.
Statistical Associations and Effect Sizes
In the simulated cohort, donor CYP3A5*1/*1 genotype demonstrated a substantially higher risk of PTDM compared to the *3/3 genotype (OR: 3.21; 95% CI: 1.89–5.46; p < 0.001). In contrast, recipient genotype showed a weaker association (OR: 1.67; 95% CI: 1.12–2.48; p = 0.012). Polymorphisms such as ABCB1 3435C>T (OR: 1.18; p = 0.34) and UGT1A43 (OR: 1.31; p = 0.38) did not reach statistical significance.
The median time to PTDM diagnosis differed markedly according to donor genotype: 8.2 months (IQR: 3.1–14.7) for *1/*1 donors versus 18.5 months (IQR: 10.2– 28.3) for *3/*3 donors (log-rank test, p = 0.002). Multivariable analysis identified donor CYP3A5 genotype as an independent predictor of PTDM (adjusted OR: 2.41; 95% CI: 1.35–4.31; p = 0.003), alongside recipient age ≥55 years (OR: 2.08; p < 0.001) and BMI ≥28 kg/m2 (OR: 1.89; p = 0.001) (Figure 4).
Genetic and Clinical Determinants of TPDM
These findings underscore that pharmacogenetic variation in CYP3A5 constitutes a consistent and clinically relevant risk factor for PTDM, with predictive performance enhanced through integration with conventional clinical parameters. The contributions of ABCB1 and UGT1A4 appear marginal, suggesting that genotyping strategies should prioritize CYP3A5, complemented by comprehensive clinical assessment.
DISCUSSION
This in silico pharmacogenomic analysis provides integrated evidence that donor-derived genetic factors (particularly variability in CYP3A5) are important in modulating the risk of PTDM among liver transplant recipients treated with tacrolimus. By converging simulated clinical-genetic data, machine learning–based modeling, explainability analyses, and traditional statistical approaches, our results consistently identify donor CYP3A5 genotype as the predominant determinant of PTDM susceptibility, whereas recipient variants exhibit only modest or negligible effects. Together, these results underscore the expanded clinical relevance of donor pharmacogenomics, not only for graft survival and immunosuppressant dose optimization but also as a critical factor in mitigating metabolic complications such as PTDM.
Previous studies have firmly established CYP3A5 polymorphisms as primary determinants of tacrolimus pharmacokinetics, with carriers of the *1 allele exhibiting higher enzymatic expression and consequently altered drug exposure compared to non-expressing *3/*3 individuals.23,24 In the context of liver transplantation, the donor liver assumes the dominant role in drug metabolism, thereby conferring particular relevance to the donor genotype.8 Our systematic review aligns with this framework, while our simulation modeling illustrates a plausible metabolic consequence (PTDM), thereby supporting earlier observational reports linking greater tacrolimus exposure variability to impaired glucose homeostasis.25
The interethnic variability in CYP3A53 allele frequencies identified in our analysis carries substantial implications for transplant compatibility strategies and population-specific risk profiling. In European populations, where functional CYP3A5***1** carriers constitute a minority, the resulting risk landscape differs markedly from that observed in cohorts of African ancestry, which are characterized by a higher prevalence of expresser phenotypes.26,27 This genetic architecture necessitates ancestrally informed approaches to preemptive genotyping, particularly within multiethnic transplant programs. Recent studies in Asian populations report intermediate PTDM incidence rates,28,29 consistent with the allele frequency distributions documented in our analysis based on the 1000 Genomes Project.9
Our network enrichment analysis reveals a mechanistic convergence between tacrolimus pharmacokinetics and calcineurin-mediated diabetogenic pathways. The bifunctional role of calcineurin, essential both for T-cell activation and for insulin secretion by pancreatic β-cells, is robustly supported by experimental evidence, with studies demonstrating dose-dependent suppression of NFAT-regulated insulin gene transcription.30 The high betweenness centrality of calcineurin in our protein–protein interaction network mirrors its biological position as a pivotal hub, bridging xenobiotic metabolism and endocrine regulation. This topological feature substantiates the hypothesis that chronic calcineurin inhibition, not merely transient pharmacological peaks, drives metabolic decompensation leading to PTDM.
The modest contributions of ABCB1 and UGT1A4 polymorphisms to PTDM risk in our models diverge from conflicting reports in the literature. Although individual studies have suggested associations between ABCB1 haplotypes and tacrolimus-related complications,8,31 subsequent meta-analyses have revealed substantial heterogeneity32,33 and population-specific effects that attenuate or disappear after rigorous statistical adjustment. Our findings support these cautious interpretations, indicating that routine genotyping of these variants confers limited incremental clinical utility beyond CYP3A5-based stratification. The paucity of robust evidence regarding UGT1A4 in transplant cohorts underscores the need for dedicated prospective studies to definitively clarify its role.
From a precision medicine perspective, the theoretical predictive performance demonstrated in our proof-of-concept model, particularly when stratified by donor genotype, supports the implementation of CYP3A5-guided therapeutic algorithms. The enhanced positive predictive value observed among recipients of allografts from donors carrying functional CYP3A5 alleles identifies a high-risk subgroup amenable to intensified metabolic surveillance and consideration of alternative immunosuppressive strategies.34,35 Integrating pharmacogenetic data with conventional clinical risk factors (e.g., age, obesity, and pre-transplant glucose dysregulation) enables refined risk stratification that transcends genetic determinism while preserving the essential mechanistic biological insights required for rational, individualized therapy.
STUDY LIMITATIONS
Our analysis has several limitations that warrant consideration. The predictive modeling relied on synthetic data rather than individual-level patient records, thereby precluding direct clinical validation. The unavailability of granular patient-level data hindered detailed subgroup analyses and the exploration of gene–environment interactions. The source studies exhibited heterogeneity in follow-up duration, diagnostic criteria for post-transplant diabetes mellitus (PTDM), and immunosuppressive regimens employed. Furthermore, pharmacogenomic evidence pertaining to UGT1A4 and ABCB1 remains nascent in the context of liver transplantation. Finally, the absence of direct pharmacokinetic parameters limited the mechanistic precision in modeling the exposure–response relationship underlying PTDM pathogenesis
CONCLUSION
This systematic review and in silico pharmacogenomic analysis identifies that donor-derived CYP3A5 variability serves as the primary determinant of tacrolimus exposure and susceptibility to PTDM in liver transplant recipients. By integrating genetic evidence, pathophysiological mechanisms, and predictive modeling, our findings support the development of CYP3A5 genotype-guided immunosuppressive strategies, enabling refined risk stratification, optimized metabolic monitoring, and the advancement of precision medicine in transplant care.
Conflict of Interest
The authors have no conflict of interest to declare.
Data Availability
All data produced in the present work are contained in the manuscript
REFERENCES
