Lower Dose‐Normalized Tacrolimus Exposure in CYP3A5*6 vs. *3 Loss‐of‐Function Allele Carriers: A Longitudinal Retrospective Real‐World Study in Kidney Transplant Recipients

Abstract

Pharmacogenomic research has historically focused on individuals of European ancestry, leading to the underrepresentation of genetic variants common in non‐European populations. This bias is exemplified by CYP3A5*6, a functionally consequential variant common in individuals of African ancestry (MAF: 11–19%) but virtually absent in Europeans (MAF: 0.15%). We conducted a retrospective, longitudinal cohort study using real‐world data from 1,461 adult kidney transplant recipients across 67 countries, analyzing 4,293 dose‐normalized 24‐hour area‐under‐the‐curve (AUC_0‐24) measurements of tacrolimus. Patients with CYP3A5*1/*1 were excluded. Linear mixed‐effects models (LME) were used to assess the association between CYP3A5*6 carriage and tacrolimus exposure, adjusting for clinical factors and ancestry using both HLA‐based principal components and country of birth. CYP3A5*6 carriers had a 17% lower dose‐normalized AUC_0‐24 than CYP3A5*3 carriers (P = 0.015). Sensitivity analyses using dose‐normalized trough concentrations (C₀) confirmed these findings, with a 20% lower exposure in CYP3A5*6 carriers (P = 0.011). An interval‐based analysis demonstrated persistently lower tacrolimus exposure across the first post‐transplant year. All CYP3A5*6‐containing genotypes showed significantly lower dose‐normalized AUC_0‐24 compared to CYP3A5*3/*3, the most common genotype in European populations, with the largest reductions observed in CYP3A5*1/*6 (−39%; P < 0.001) and CYP3A5*3/*6 (−18%; P = 0.006). African origin, defined by country of birth, was independently associated with a 23% higher AUC_0‐24 (P < 0.001). This is the first study to demonstrate a differential effect on tacrolimus exposure between the CYP3A5*6 and CYP3A5*3 loss‐of‐function alleles. Our results may help bridge the ethnicity gap, advance the applicability of pharmacogenomic findings, and promote health equity.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Tacrolimus pharmacokinetics vary widely among kidney transplant recipients, partly due to genetic polymorphisms in CYP3A5. Most pharmacogenomic studies have focused on the CYP3A5*3 loss‐of‐function allele, which is common in individuals of European ancestry, while the CYP3A5*6 loss‐of‐function allele, prevalent in African populations, remains under‐investigated.

• WHAT QUESTION DID THIS STUDY ADDRESS?

This study evaluated whether the CYP3A5*6 allele influences dose‐normalized tacrolimus exposure differently from the well‐characterized CYP3A5*3 allele in a diverse, real‐world kidney transplant cohort.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

Using 4,293 dose‐normalized tacrolimus AUC_0‐24 measurements from 1,461 kidney transplant recipients across 67 countries, we demonstrate that CYP3A5*6 carriers have significantly lower dose‐normalized tacrolimus exposure compared to CYP3A5*3 carriers, independent of ancestry and clinical covariates. This is the first study to demonstrate a differential effect of the CYP3A5*6 allele compared to CYP3A5*3 on tacrolimus exposure.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

By demonstrating that a variant nearly absent in European datasets meaningfully influences tacrolimus exposure, our findings underscore the importance of incorporating African genome data in dosing algorithms and prioritizing inclusive pharmacogenomic research. This could improve health equity, expand the applicability of pharmacogenomic findings, and advance personalized medicine for all.

Pharmacogenomics optimizes patient outcomes by leveraging genetic variation to explain interindividual variability in drug response. ¹ However, most studies have historically focused on individuals of European ancestry, leading to underrepresentation, or complete omission, of genetic variants prevalent in non‐European populations. This Eurocentric bias compromises the accuracy of pharmacogenomic testing and dosing guidelines for diverse populations, perpetuating health disparities. ² , ³ Currently, there is a growing emphasis on expanding genomic research to underrepresented populations to enhance the applicability of pharmacogenomic findings and advance health equity. ² , ⁴ , ⁵ , ⁶

Tacrolimus, a cornerstone of maintenance immunosuppressive therapy in kidney transplantation, exemplifies the clinical importance of incorporating genetic diversity into pharmacogenomic research. Genetic variation in the cytochrome P450 enzymes CYP3A4 and CYP3A5 significantly influences tacrolimus pharmacokinetics (PK) and dose requirements. ⁷ In particular, the CYP3A5 genotype strongly influences tacrolimus dosing. Carriers of at least one functional *1 allele express the enzyme and metabolize tacrolimus more rapidly, requiring higher doses, whereas homozygous carriers of loss‐of‐function (LOF) alleles (e.g., *3/*3) have reduced metabolism and attain higher blood concentrations at standard doses. ⁸ , ⁹ , ¹⁰

The CYP3A5*3 allele is the most well‐characterized LOF variant, with a high frequency in patients of European descent (MAF: 92%). ¹¹ , ¹² In contrast, CYP3A5*6, with a high prevalence in Sub‐Saharan African (MAF: 19%) and Afro‐Caribbean/Afro‐American populations (MAF: 11%), but is nearly absent in Europeans (MAF: 0.15%), ¹¹ is much less well‐characterized. As a result, only ~15–20% of Europeans are CYP3A5 expressors, compared with ~50–80% of individuals of Sub‐Saharan African ancestry. ¹⁰ , ¹¹ Like CYP3A5*3, CYP3A5*6 ablates protein function. ¹⁰ Yet, its impact on tacrolimus metabolism remains poorly characterized due to underrepresentation in pharmacogenomic studies. In addition to CYP3A5, other pharmacogenes, including CYP3A4, ABCB1, and PPARA, have been associated with tacrolimus PK. However, findings have been inconsistent with sometimes conflicting results, and their clinical impact appears substantially smaller compared to CYP3A5. ⁷ , ¹⁰ , ¹³ , ¹⁴ The limited representation of population‐specific variants in pharmacogenomic studies is evident in the field of kidney transplantation, ¹⁵ where the lack of genetic diversity has restricted the transferability of findings across populations and contributed to disparities in outcomes. ³ Notably, individuals of Sub‐Saharan African and African American ancestry have a higher prevalence of chronic kidney disease, progress more frequently to end‐stage kidney disease, ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ and experience worse outcomes following kidney transplantation, including a higher risk of graft failure compared to individuals of European or other ancestries. ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵

Here, we analyzed a cohort of kidney transplant recipients with unique and extensively sampled PK profiles up to 13 months post‐transplantation to evaluate the impact of CYP3A5*6 on the PK of tacrolimus. Unlike previous studies that assessed tacrolimus exposure based solely on trough concentrations (C₀), ¹³ we also determined tacrolimus exposure using area under the concentration‐time curve (AUC) measurements to provide a more comprehensive understanding of how CYP3A5*6 may affect tacrolimus PK across genetically diverse populations, particularly in comparison to the well‐characterized CYP3A5*3 allele.

MATERIALS AND METHODS

Study design

This retrospective, longitudinal cohort study utilized real‐world clinical data obtained from the transplant center database at Leiden University Medical Center (LUMC). We included adult patients (≥ 18 years) who underwent kidney transplantation between 2011 and 2024 and in whom immunosuppressive treatment following transplantation consisted of tacrolimus, mycophenolic acid, and prednisolone. Inclusion criteria were:

1. ≥ 1 tacrolimus AUC_0‐24 with paired C₀ (as defined in the Therapeutic Drug Monitoring and Pharmacokinetic Analysis section).

2. Documented CYP3A5 genotype.

3. Follow‐up until 13 months post‐transplant or censoring (death/graft loss/tacrolimus discontinuation/lost to follow‐up).

Available therapeutic drug monitoring data (AUC_0‐24 with paired C₀) during follow‐up was analyzed. Exclusion criteria were:

1. Patients homozygous for CYP3A5*1/*1 (to enable direct comparison between CYP3A5*3 and CYP3A5*6 allele carriers).

2. Simultaneous multiorgan transplant recipients (e.g., kidney‐liver or kidney‐pancreas).

3. Prior liver transplant (due to undefined donor CYP3A contribution to tacrolimus metabolism). ²⁶

4. Patients initiated on steroid‐free immunosuppressive regimens (avoiding steroid‐induced CYP3A activity). ²⁷

5.  MeltDose® extended‐release tacrolimus (LCPT) (Envarsus®) (due to fundamentally different PK [distal gut absorption] ²⁸ and consequently potentially reduced CYP3A5‐mediated metabolism). ²⁹

Study population and ancestry classification

For pharmacogenomic analysis, patients were stratified by the CYP3A5*3 or *6 allele. Two groups were defined:

• CYP3A5*6 carriers (genotypes *1/*6, *3/*6 or *6/*6).

• CYP3A5*3 carriers (genotypes *1/*3 or *3/*3).

In addition to this primary grouping, all CYP3A5 diplotypes were modeled to enable comparison of each group against CYP3A5*3 homozygotes, the most common genotype in European populations. ¹¹ To account for ancestry‐related confounding, given the higher prevalence of *6 in individuals of African descent, we employed both genetic (HLA‐based principal component analysis (PCA)) and demographic (country of birth) data to infer ancestral background. In the absence of genome‐wide data, primary ancestry inference was based on HLA genotyping. This was performed at first‐field resolution prior to 2015 and at high resolution (second‐field, via next‐generation sequencing) thereafter (see Genotyping). HLA genes are among the most polymorphic in the human genome, with allele frequencies that vary significantly across populations, making them particularly informative for ancestry inference. ³⁰ , ³¹ As a sensitivity analysis, we also modeled origin, defined by country of birth, an established proxy for ancestral background in epidemiological research, ³² which in the Dutch healthcare system is routinely derived from official documentation (e.g., passport or residency permit) at the time of patient registration. A binary variable was created to distinguish individuals of African origin from those of non‐African origin. The full ancestry inference methodology is described in the Supplementary Information .

Genotyping

Genotyping for CYP3A5*3 (rs776746) and CYP3A5*6 (rs10264272) is routinely performed for all kidney transplant patients from 2011 onward in the ISO15189‐accredited laboratory of the department of Clinical Pharmacy and Toxicology of the LUMC. DNA was isolated from EDTA‐anticoagulated whole blood samples. CYP3A5*3 (rs776746) and CYP3A5*6 (rs10264272) alleles were genotyped using TaqMan SNP assays on the TaqMan 7500 Real‐Time PCR System (Applied Biosystems, Nieuwerkerk a.d. IJssel, The Netherlands), following manufacturer protocols. All variant allele frequencies were in Hardy–Weinberg equilibrium (P > 0.05 by χ² test). HLA genotyping for HLA‐A, ‐B, ‐C, ‐DRB1, and ‐DQB1 was performed in the ISO15189‐accredited HLA laboratory of LUMC. Full HLA genotyping details are provided in the Supplementary Information .

Therapeutic drug monitoring and pharmacokinetic analysis

Tacrolimus concentrations were measured at the ISO15189‐certified Clinical Pharmacy and Toxicology laboratory of LUMC using validated LC–MS/MS techniques. Tacrolimus exposure was monitored according to institutional protocol using both C₀ and AUC, with AUC as the primary PK parameter. AUC values were transformed and analyzed as 24‐hour AUC (AUC_0‐24) to facilitate comparison between patients receiving once‐ and twice‐daily formulations. Monitoring was mainly performed by abbreviated 3‐point sampling (0, 2, 3 h) with Bayesian estimation, while full sampling (0, 1, 2, 3, 4, 5, 6 h) was typically performed during inpatient admission in the first post‐transplant weeks (see Supplementary Information for details).

Data collection

Tacrolimus AUC and C₀ measurements were retrieved from our institutional transplant center database, which prospectively captures clinical and PK data from all adult kidney transplant recipients. Demographic and physiological data (age, sex, country of birth, time since transplantation, hematocrit, body weight) were extracted from electronic health records using Ctcue (IQVIA Patient Finder Solution‐CTcue B.V., Amsterdam, The Netherlands). Covariate values were aligned to AUC measurements by selecting the value closest in time. These parameters were selected based on their well‐recognized influence on tacrolimus PK ²⁰ to enable accurate comparison between CYP3A5*3 and *6 carriers. Full details of manual data verification procedures can be found in Supplementary Information  S1.

Statistical analysis

To assess the impact of CYP3A5*6 on tacrolimus PK over time, we employed linear mixed‐effects models (LME) on log‐transformed, dose‐normalized AUC_0‐24, to account for repeated measurements per individual. Patients contributed multiple AUC measurements over time, with follow‐up extending up to 13 months post‐transplant or until censoring (see Study Design). Each patient was modeled with a random intercept to account for individual differences in baseline tacrolimus exposure. Time since transplantation (in days) was included as a continuous fixed‐effect variable, with both linear and quadratic terms to capture potential nonlinear PK trends.

The primary genotype effect was modeled as a binary variable, comparing CYP3A5*6 carriers to noncarriers, as previously defined. A genotype × time interaction term was included in the primary model to evaluate whether the effect of CYP3A5*6 varies over the post‐transplant course. In secondary analyses, the number of *6 alleles per individual (0, 1, or 2) was modeled as a continuous variable to assess its additive genetic effects on tacrolimus PK. To compare CYP3A5 diplotype groups, we additionally modeled CYP3A5 genotype as a categorical variable including all observed combinations (*1/*3, *1/*6, *1/*1, *3/*6, *6/*6, and *3/*3) using CYP3A5*3/*3 as a reference. To reduce sparsity and assess model stability, we additionally modeled the number of *1 alleles per individual alongside the previously defined number of *6 alleles. This approach increases robustness and allows for more stable effect estimates by assuming additive genetic effects and avoiding the formation of diplotype groups with low frequency in the cohort. All models were adjusted for prespecified covariates known to influence tacrolimus PK ¹³ : age, hematocrit, body weight, sex, and ancestry. In all primary models, genetic ancestry was adjusted for using the first five principal components as extracted from HLA‐based PCA.

We conducted sensitivity analyses to assess the robustness of our findings. First, we repeated the primary model using log‐transformed, dose‐normalized C₀ as the outcome instead of AUC_0‐24. Second, we refitted the model without the genotype × time interaction term to estimate the average effect of CYP3A5*6 carrier status across the entire follow‐up period.

To visualize genotype‐dependent differences in tacrolimus exposure over time, we performed both interval‐based and continuous descriptive analyses. For the interval‐based analysis, we calculated the mean and standard error of log‐transformed, dose‐normalized AUC_0‐24 at standard therapeutic drug monitoring timepoints (2 weeks, 6 weeks, 6 months, and 1 year), comparing CYP3A5*6 carriers vs. noncarriers. To capture the full range of available data, including carriers with AUC_0‐24 outside predefined windows, we additionally applied LOESS smoothing to all log‐transformed AUC_0‐24 measurements over continuous time with standard settings.

Ethics statement

This study was approved by the Institutional Review Board of Leiden University Medical Center (nWMO‐D4‐2025‐005) and conducted in accordance with the Declaration of Helsinki.

RESULTS

Overview of study cohort

We analyzed 1,461 adult kidney transplant recipients with 4,293 paired AUC_0‐24 and C₀ samples, of which ~25% were obtained by full sampling during inpatient admission in the first two post‐transplant weeks and the remainder by abbreviated 3‐point sampling. Out of these, 35 (2.5%) carried at least one CYP3A5*6 allele (genotypes *1/*6, *3/*6, or *6/6), and 1,380 (97.5%) did not (genotypes *1/*3 or *3/*3). Clinical characteristics revealed that CYP3A5*6 carriers were younger at transplantation (median 49 vs. 54 years, P = 0.020) and had higher baseline hematocrit (median 0.35 vs. 0.32, P < 0.001), but comparable body weights (P = 0.381) and sex distribution (P = 0.978). AUC_0‐24 – C₀ sampling density was balanced between genotypes (median 3 samples/patient), with measurements tightly aligned to clinical parameters (median 0‐day delays). Detailed patient characteristics per CYP3A5 *6 carrier status are shown in Table 1 , and baseline characteristics by CYP3A5 genotype are presented in Table S1 .

Table 1.

Patient characteristics stratified by CYP3A5 carrier status

Characteristic	CYP3A5*3 carrier (n = 1,380)	CYP3A5*6 carrier (n = 35)	Total (n = 1,415)	P‐value
Demographics
Male sex—n (%)	844 (61%)	22 (63%)	866 (61%)	0.978
Age at transplantation—median [IQR], years	54 [45–65]	49 [41–58]	54 [45–65]	0.020
Born in the Netherlands—n (%)	1,012 (73%)	3 (8.6%)	1,015 (72%)	< 0.001
European origina— n (%)	1,056 (77%)	3 (8.6%)	1,059 (75%)	< 0.001
African originb—n (%)	122 (8.8%)	23 (66%)	145 (10%)	< 0.001
Donor type and source
Living donor—n (%)	807 (58%)	8 (23%)	815 (58%)	‐
Postmortal donor—n (%)	573 (42%)	27 (77%)	600 (42%)	< 0.001
Underlying kidney disease				0.262
Glomerular diseases—n (%)	330 (24%)	7 (20%)	337 (24%)	—
Cystic kidney diseases—n (%)	220 (16%)	2 (5.7%)	222 (16%)	—
Hypertensive/vascular nephropathy—n (%)	170 (12%)	6 (17%)	176 (12%)	—
Diabetic kidney disease—n (%)	146 (10%)	6 (17%)	152 (11%)	—
Urological disorders—n (%)	58 (4.2%)	0 (0%)	58 (4.1%)	—
Multifactorial/eci/other—n (%)	456 (31%)	14 (37%)	470 (31%)	—
Pharmacokinetic and clinical sampling
AUC0‐24 – C0 samples—n	4,068	109	4,177	—
AUC0‐24 – C0 samples per patient—median [IQR]	3 [2–4]	3 [2–4]	3 [2–4]	—
Hematocrit value at time of transplantation, median [IQR]	0.32 [0.29–0.35]	0.35 [0.32–0.39]	0.32 [0.29–0.35]	< 0.001
Body weight at time of transplantation, median [IQR], kg	76.5 [66.4–87.0]	73.4 [66.1–84.3]	76.3 [66.4–87.0]	0.381
Timepoint pharmacokinetic sampling
AUC0‐24 – C0 samples 0–2 weeks—n (%)	1,316	34	1,350	—
AUC0‐24 – C0 samples 2–6 weeks—n (%)	449	12	461	—
AUC0‐24 – C0 samples 6 weeks–6 months—n (%)	1,248	33	1,281	—
AUC0‐24 – C0 samples 6–13 months—n (%)	1,055	30	1,085	—

Data presented as n (%), median [interquartile range (IQR)] as indicated. CYP3A5*6 carriers: ≥ 1 CYP3A5*6 allele (genotypes *1/*6, *3/*6 or *6/*6); CYP3A5*3 carriers: no CYP3A5*6 alleles (genotypes *1/*3 or *3/*3). Abbreviations: AUC_0‐24‐C₀, 24‐hour area under the concentration–time curve with paired trough level.

These country groupings were determined a priori and based on relative frequencies within our study population; countries not observed in the dataset were not included.

Hematocrit, body weight, and temporal alignment of these clinical parameters with AUC data are available in Table S2 .

^a European origin as defined by country of birth: the Netherlands, Belgium, Germany, Denmark, Greece, Hungary, Italy, Poland, Portugal, Romania, Russia, Scotland, Spain, the United Kingdom, the United States, Australia, New Zealand, or Andorra (see Study Population and Ancestry Classification).

^b African origin as defined by country of birth: Afro‐Caribbean regions (Suriname, Curaçao, the Netherlands Antilles, Dominican Republic, or Guadeloupe) or sub‐Saharan Africa (Ghana, Nigeria, Eritrea, Ethiopia, Angola, Somalia, Sierra Leone, Sudan, Zimbabwe, South Africa, Guinea, Democratic Republic of Congo, Senegal, or Cape Verde) (see Study Population and Ancestry Classification).

Global distribution of analyzed patients

Analyzed kidney transplant recipients were born in 67 unique countries, spanning 15 global regions as defined by the United Nations Statistics Division (M49 classification). Most patients were born in Europe (n = 3,241 samples), followed by Asia (n = 464), Africa (n = 404), Latin America and the Caribbean (n = 422), Oceania (n = 10), and North America (n = 8). The most represented countries of birth included the Netherlands (n = 2,978), Suriname (n = 331), Morocco (n = 159), and Indonesia (n = 61). A full list of countries and corresponding UN region classifications for the analyzed study population is provided in Table S3 . Country‐of‐birth classification revealed marked differences between groups (Table 1 ). Using our predefined classifications, 77% of *3 carriers were born in European‐designated countries (e.g., the Netherlands, other European countries, United States, Australia) vs. 8.6% of *6 carriers (P < 0.001). Conversely, 66% of *6 carriers were born in African‐designated countries (sub‐Saharan Africa/Afro‐Caribbean regions) compared to 8.8% of *3 carriers (P < 0.001). Among CYP3A5*6 carriers, country‐of‐birth analysis indicated that 40% were born in Sub‐Saharan African countries, most frequently Somalia, Ethiopia, Ghana, and Nigeria. Another 29% were born in Afro‐Caribbean regions, including Suriname (17%), the Dutch Antilles, and the Dominican Republic. Morocco accounted for 20% of *6 carriers. In contrast, CYP3A5*3 carriers were predominantly of European origin, with 73% born in the Netherlands and most others in European countries. A smaller proportion were born in countries such as Suriname (6.3%), Morocco (3.2%), and Somalia (0.9%). A global origin map, based on country of birth, visually illustrates the geographical spread of transplanted patients included in the study (Figure 1 ).

Figure 1.

Global distribution of analyzed kidney transplant recipients by country‐of‐birth‐based origin. World map showing the country of birth for all included kidney transplant recipients, used as a proxy for ancestry classification. Countries are shaded by origin group: European (salmon), Sub‐Saharan African (green), Afro‐Caribbean (blue), and Other (black). Small island nations classified as Afro‐Caribbean (Curaçao, Dominican Republic, Guadeloupe, Dominica, and the Netherlands Antilles) are represented with colored points at approximate locations. Countries shown in white had no patients represented in the study dataset. Classification was based on predefined groupings using United Nations regional codes and study‐specific origin definitions.

Ancestry‐adjusted effect of CYP3A5 *6 compared to *3 on tacrolimus exposure

The LME analysis of 4,177 dose‐normalized AUC_0‐24 measurements showed that CYP3A5*6 carrier status was independently associated with a 17% lower dose‐normalized AUC_0‐24 compared to *3 carriers (95% CI: −28% to −3%; estimate: 0.835; P = 0.015) (Table 2 ). This effect remained statistically significant after adjusting for age, sex, hematocrit, body weight, and time since transplantation, as well as for genetic ancestry using HLA‐based PCA. Although one ancestry component (PC4) was independently associated with AUC_0‐24 (95% CI: −2% to −0.1%; estimate: 0.989; P = 0.032), this did not affect the statistical significance of the difference in AUC_0‐24 between CYP3A5*6 and *3 carriers. Details of the PCA procedure and ancestry clustering are provided in the Supplementary Results (Figure  S3 ).

Table 2.

Fixed‐effect estimates from the linear mixed‐effects model evaluating the association between CYP3A5*6 carrier status and dose‐normalized tacrolimus AUC_0‐24 (area under the concentration–time curve from 0 to 24 hours)

Term	% Change in AUC0‐24	95% CI (%)	Estimate	P‐value
CYP3A5*6 carrier (vs *3 carrier)	−17%	−28% to −3%	0.835	0.015
Time since transplant (days)	+0.2%	+0.2% to +0.3%	1.002	< 0.001
(Time since transplant)2 (days2)	−0.0005%	−0.0006% to −0.0004%	1.000	< 0.001
Age at AUC0‐24 (years)	+0.7%	+0.5% to +0.8%	1.007	< 0.001
HLA PC1	+0.5%	−0.6% to +1.2%	1.003	0.52
HLA PC2	+0.2%	−0.8% to +1.1%	1.002	0.709
HLA PC3	+0.1%	−0.9% to +1.1%	1.001	0.793
HLA PC4	−1.1%	−2.1% to −0.1%	0.989	0.032
HLA PC5	−0.7%	−1.7% to +0.4%	0.993	0.228
Hematocrit (per 1%)	+2.4%	+2.1% to +2.6%	1.024	< 0.001
Body weight (kg)	−0.3%	−0.4% to −0.1%	0.997	< 0.001
Time from hematocrit to AUC0‐24 (days)	−0.1%	−0.3% to +0.1%	0.999	0.429
Female (vs. male)	+10%	+6% to +14%	1.115	< 0.001
CYP3A5*6 × Time since transplant	+0.0007%	−0.05% to +0.05%	1.000	0.976

Results are reported as percent change in AUC_0‐24, estimate (i.e., exponentiated model estimates), 95% confidence intervals (CI), and P‐values. The model included time since transplant (linear and quadratic terms), age, sex, hematocrit, body weight, and five HLA‐derived principal components (PC) to adjust for genetic ancestry. A random intercept per patient was included to account for repeated measures. The model intercept is not shown.

Sensitivity and robustness analyses

Sensitivity analyses confirmed the robustness of our findings. Using dose‐normalized C₀ as the outcome, CYP3A5*6 carriers showed 20% lower C₀ compared with CYP3A5*3 carriers (95% CI: −32% to −5%; estimate: 0.804; P = 0.011) (Table S4 ). Covariates, including hematocrit, body weight, sex, and time since transplantation, were significantly associated with tacrolimus exposure. As an additional model, we used African origin, defined based on country of birth, as a proxy for ancestry. This analysis showed that CYP3A5*6 carriers had a 20% lower dose‐normalized AUC_0‐24 compared to CYP3A5*3 carriers (95% CI: −31% to −8%; estimate: 0.798; P = 0.002), consistent with our main analysis. The ancestry proxy itself was not significantly associated with tacrolimus AUC_0‐24 (95% CI: −7% to +7%; estimate: 0.994; P = 0.877), and its inclusion did not alter the direction or magnitude of the association between CYP3A5 genotype and tacrolimus exposure (Table S7). Additional sensitivity analyses, including the number of CYP3A5*6 alleles per individual and country of birth subgroup analyses, are detailed in the Supplementary Information.

Interval‐based analysis of tacrolimus exposure

To visualize genotype‐dependent differences in tacrolimus exposure over time, we performed an interval‐based descriptive analysis of dose‐normalized AUC_0‐24 at standard therapeutic drug monitoring intervals (i.e., 2 weeks, 6 weeks, 6 months, and 12 months). As illustrated in Figure 2 a , CYP3A5*6 carriers consistently showed lower tacrolimus dose‐normalized exposure compared to CYP3A5*3 throughout the first post‐transplant year. At 2 weeks, the mean dose‐normalized AUC_0‐24 was 26 ± 1.3 ng·h/mg/mL in *6 carriers vs. 33 ± 0.4 ng·h/mg/mL in noncarriers. This difference widened at 6 weeks (32 ± 3.8 vs. 46 ± 0.9 ng·h/mg/mL) and at 6 months (39 ± 5.0 vs. 53 ± 1.0 ng·h/mg/mL). Although the gap narrowed by 12 months (49 ± 7.8 vs. 55 ± 1.4 ng·h/mg/mL), *6 carriers still exhibited lower mean exposure. These data are visualized in Figure 2 a and numerically summarized in Table S11 . Statistical comparison of non‐dose‐normalized AUC_0‐24 and C₀ confirmed significantly lower exposure in *6 carriers only at 0–2 weeks post‐transplant (P = 0.0058), but no significant differences were observed at later time intervals (Tables S12 and S13 ), showing the value of therapeutic drug monitoring guided dosing to achieve target attainment. To incorporate all available data and reduce potential bias from time window selection, we additionally applied LOESS smoothing to dose‐normalized AUC_0‐24 values over continuous follow‐up. As shown in Figure 2 b , this visualization supports the interval‐based findings by confirming a persistent trend toward lower exposure in CYP3A5*6 carriers throughout the first year post‐transplant.

Figure 2.

Dose‐normalized tacrolimus exposure (AUC_0‐24) over time stratified by CYP3A5*6 carrier status. Tacrolimus AUC_0‐24 values are stratified by CYP3A5*6 genotype status: green for *6 carriers and orange for noncarriers. (a) Mean dose‐normalized tacrolimus AUC_0‐24 (ng·h/mL per mg) ± standard error at standard therapeutic drug monitoring timepoints (2 weeks, 6 weeks, 6 months, and 1 year post‐transplant), stratified by CYP3A5*6 genotype. CYP3A5*6 carriers consistently exhibited lower tacrolimus exposure compared to noncarriers, with the largest differences observed at 6 weeks and 6 months. Data underlying this panel are provided in Table S11 . (b) LOESS‐smoothed longitudinal trend of dose‐normalized AUC_0‐24 across all available measurements during the first post‐transplant year. Individual values (dots) are stratified by CYP3A5*6 carrier status (green: *6 carriers; orange: noncarriers). Solid lines represent LOESS‐smoothed trends with shaded 95% confidence intervals. The y‐axis is log₁₀‐transformed to account for the skewed distribution of AUC_0‐24 values and to reduce the impact of extreme outliers. Vertical dotted lines mark standard therapeutic drug monitoring timepoints.

Tacrolimus exposure stratified by CYP3A5 diplotype

We compared tacrolimus exposure across CYP3A5 diplotypes using LME models with CYP3A5*3/*3 as a reference, adjusting for genetic ancestry via HLA‐based PCA. All diplotype groups showed significantly lower dose‐normalized AUC_0‐24 compared to CYP3A5*3/*3. Among the CYP3A5*6‐containing diplotypes, individuals with the CYP3A5*1/*6 diplotype had a 39% reduction in dose‐normalized tacrolimus AUC_0‐24 compared to the reference group (P < 0.001), while CYP3A5*3/*6 carriers demonstrated an 18% reduction (P = 0.006). Based on three PK samples from a single patient, CYP3A5*6/*6 homozygotes showed a directionally similar effect (P = 0.183). The largest reduction overall was seen in CYP3A5*1/*1 homozygotes (P < 0.001), followed by *1/*3 (P < 0.001). Results were consistent when using the country of birth as an ancestry proxy. A side‐by‐side comparison of both ancestry‐adjusted models is summarized in Table 3 .

Table 3.

Comparative fixed‐effect estimates from two linear mixed‐effects models evaluating the association between CYP3A5 diplotype and dose‐normalized tacrolimus AUC_0‐24, using either HLA‐based principal components (PCA) or country of birth as proxy for ancestry

CYP3A5 diplotype	HLA PCA‐based ancestry adjustment			Country‐of‐birth ancestry adjustment
	% Change	95% CI	P	% Change	95% CI	P
CYP3A5*1/*1 (vs *3/*3)	−47%	−53% to −41%	< 0.001	−50%	−56% to −45%	< 0.001
CYP3A5*1/*3 (vs *3/*3)	−41%	−44% to −38%	< 0.001	−43%	−45% to −40%	< 0.001
CYP3A5*1/*6 (vs *3/*3)	−39%	−51% to −24%	< 0.001	−49%	−59% to −38%	< 0.001
CYP3A5*3/*6 (vs *3/*3)	−18%	−30% to −6%	0.006	−26%	−36% to −15%	< 0.001
CYP3A5*6/*6 (vs *3/*3)	−36%	−67% to +24%	0.183	−47%	−72% to +3%	0.061
Ancestry adjustment
African origina	—	—	—	+21%	+14% to +29%	< 0.001
HLA PC1	+1.5%	+0.7% to +2.4%	< 0.001	—	—	—
HLA PC2	+0.7%	−0.1% to +1.6%	0.082	—	—	—
HLA PC3	−1.1%	−2.0% to −0.2%	0.014	—	—	—
HLA PC4	+0.8%	−0.2% to +1.7%	0.101	—	—	—
HLA PC5	+0.4%	−0.5% to +1.3%	0.381	—	—	—
Clinical covariates
Time since transplant (days)	+0.2%	+0.2% to +0.3%	< 0.001	+0.2%	+0.2% to +0.3%	< 0.001
(Time since transplant)2	−0.0005%	−0.0006% to −0.0004%	< 0.001	−0.0005%	−0.0006% to −0.0004%	< 0.001
Age at AUC (years)	+0.7%	+0.6% to +0.9%	< 0.001	+0.6%	+0.5% to +0.8%	< 0.001
Hematocrit (per 1%)	+2.4%	+2.1% to +2.7%	< 0.001	+2.4%	+2.1% to +2.6%	< 0.001
Body weight (per kg)	−0.3%	−0.5% to −0.2%	< 0.001	−0.4%	−0.5% to −0.2%	< 0.001
Time between Hct and AUC (days)	−0.05%	−0.2% to +0.1%	0.616	−0.04%	−0.2% to +0.1%	0.673
Female sex (vs. male)	+10.0%	+6.2% to +13.6%	< 0.001	+8.1%	+4.4% to +11.6%	< 0.001

Genotype was modeled as a categorical variable with CYP3A5*3/*3 as the reference. Both models were adjusted for time since transplant (linear and quadratic), age, sex, hematocrit, body weight, and timing of hematocrit measurement. A random intercept per patient was included.

^a African origin: Individuals born in either Afro‐Caribbean regions (Suriname, Curaçao, the Netherlands Antilles, Dominican Republic, or Guadeloupe) or Sub‐Saharan Africa (Ghana, Nigeria, Eritrea, Ethiopia, Angola, Somalia, Sierra Leone, Sudan, Zimbabwe, South Africa, Guinea, Democratic Republic of Congo, Senegal, or Cape Verde).

Ancestry was significantly associated with tacrolimus exposure only when using country of birth. Patients of African origin had 23% higher AUC_0‐24 compared to patients of African‐origin (P < 0.001). In contrast, none of the five HLA‐based ancestry principal components showed an association of similar magnitude. Analyses using dose‐normalized C₀ yielded consistent findings: in the HLA‐PCA‐based adjusted model, exposure was 46% lower in *1/*6 (P < 0.001), 21% in *3/*6 (P = 0.007), and 26% in *6/*6 (P = 0.437). Complete outputs for all model variants are provided in Tables S15 – S19 .

DISCUSSION

In this real‐world cohort of kidney transplant recipients representing 67 countries across all UN regions, ³³ we demonstrate that carriers of the CYP3A5*6 allele show significantly lower tacrolimus dose‐normalized exposure compared to CYP3A5*3 carriers during the first 13 months post transplantation. Interestingly, both variants are classified as loss‐of‐function alleles. ⁸ , ¹⁰ , ¹³ Dose‐normalized AUC_0‐24 was consistently 17% lower in CYP3A5*6 carriers compared with CYP3A5*3 carriers across the first year post‐transplant, and differences persisted even after adjusting for clinical covariates and ancestry, whether defined by HLA‐based principal components or country of birth. Notably, all CYP3A5*6 diplotypes (*1/*6, *3/*6, and *6/*6) were associated with markedly lower dose‐normalized AUC_0‐24 compared to *3/*3, the most common genotype in European populations, ¹¹ although the estimate for *6/*6 was less precise due to limited sample size. Sensitivity analyses using tacrolimus trough levels (C₀), which are more commonly employed in clinical practice, yielded congruent results, showing 20% lower exposure in *6 carriers compared to noncarriers. Stratified analyses further confirmed this effect across Afro‐Caribbean, Sub‐Saharan African, and European origin subgroups, underscoring the allele’s functional impact independent of ancestry.

Our findings also indicate that African origin, defined by country of birth, significantly influences tacrolimus AUC_0‐24, independently of CYP3A5 diplotype. This finding is particularly important given the underrepresentation of African‐ancestry populations in pharmacogenomic studies. ⁶ , ¹⁵ , ³⁴ Interestingly, this ancestry association was only prominent in models that included full diplotype information and not in those comparing *6 vs. *3 carriers. This suggests that diplotypes more closely align with underlying population structure, allowing the independent contribution of country of birth to emerge. In contrast, haplotype‐level comparisons may mask such distinctions by grouping individuals with diverse backgrounds together. The absence of a similar effect when using HLA‐derived principal components further supports the interpretation that country of birth serves not only as a proxy for genetic ancestry but also for associated sociodemographic and environmental exposures, such as dietary patterns, infectious burden, healthcare access, or microbiome differences, that are not encoded in the genome. In contrast, HLA‐based PCA primarily reflects underlying genomic structure. Incorporating both genetic ancestry and origin proxies in pharmacogenomic analyses may therefore better capture the multifactorial drivers of interindividual variability in tacrolimus pharmacokinetics.

Origin distributions in our study population were notably skewed: 77% of CYP3A5*3 carriers were of European origin, compared to 66% of CYP3A5*6 carriers from African‐designated regions (e.g., Somalia, Suriname, Ghana). CYP3A5*6 is prevalent in Afro‐Caribbean and Sub‐Saharan African populations. ¹¹ Its elevated frequency in our cohort compared to European reference populations likely reflects migration from Suriname and Morocco, two of the most common countries of origin among immigrants in the Netherlands. ³⁵ Although Suriname is geographically located in South America, it has strong cultural and historical ties to the Caribbean, ³⁶ particularly due to its significant Maroon population, ³⁷ descendants of enslaved Africans who escaped slavery, which may explain the high frequency of CYP3A5*6. Similarly, historical Sub‐Saharan African migration to Morocco, largely driven by the trans‐Saharan slave trade, ³⁸ may account for the presence of this variant in individuals of Moroccan descent.

In a previous study, the CYP3A5*6 variant was associated with a smaller increase in dose‐normalized tacrolimus trough levels compared to CYP3A5*3, ³⁹ supporting our finding of a smaller effect on tacrolimus exposure in CYP3A5*6 carriers. This study also demonstrated that CYP3A5*3 was associated with significantly higher dose‐normalized tacrolimus trough concentrations in European‐ancestry populations compared to African‐ancestry populations. Consistent with the higher frequency of CYP3A5*6, African‐ancestry populations also exhibit a higher prevalence of CYP3A5*1 (MAF: 48% in Sub‐Saharan Africa and 45% in Afro‐Caribbean populations), ¹¹ which is associated with increased tacrolimus metabolism and approximately twofold higher dose requirements. ¹³ In our study, diplotypes involving CYP3A5*1, including *1/*1, *1/*3, and *1/*6, were associated with significantly lower dose‐corrected AUC_0‐24 compared to CYP3A5*3/*3 (reductions of 47%, 41%, and 39%, respectively). This higher metabolic capacity, driven by both CYP3A5*1 and CYP3A5*6, likely explains the significantly lower non‐dose‐normalized tacrolimus exposure observed in CYP3A5*6 carriers in our study during the early post‐transplant phase, with no significant differences at later time intervals. These findings have direct clinical implications. First, while therapeutic drug monitoring helps mitigate exposure variability over time, lower tacrolimus exposure in CYP3A5*6 carriers during the early post‐transplant phase is clinically relevant, as it may delay target attainment and contribute to earlier rejection. ⁴⁰ Second, the observation that two alleles both classified as LOF differ in effect size highlights the risk of misclassification when only CYP3A5*3 is genotyped, particularly in African‐ancestry patients. African‐ancestry patients have historically shown higher rates of acute rejection and lower long‐term graft survival compared to their European‐ancestry counterparts. ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ While multiple factors contribute to these disparities, including socioeconomic variables, access to care, and immunological risk profile, differential drug metabolism represents a key and potentially modifiable contributor.

Current clinical genotyping approaches that test only for CYP3A5*3 may wrongly assign metabolizer status in African‐ancestry populations. For example, a patient with a *6/*6 genotype may be incorrectly labeled as an expresser (due to the absence of *3), resulting in incorrect dosing, particularly in early treatment, and may contribute to early variability in tacrolimus exposure. This concern is supported by a UK kidney transplant study demonstrating significantly worse long‐term graft survival in Black vs. non‐Black recipients. ²² Although the study assessed CYP3A5 expresser status, genotyping was limited to CYP3A5*3, potentially overlooking other clinically relevant variants such as CYP3A5*6.

A biological explanation for the observed difference between CYP3A5*6 and CYP3A5*3 carriers lies in their distinct molecular mechanisms of loss of function. The CYP3A5*6 variant (30597G>A) introduces a silent mutation in exon 7, disrupting an exonic splicing enhancer, which may lead to exon 7 skipping during mRNA maturation. ⁴¹ This can result in a frameshift and truncated protein, but the splicing defect may not be absolute. Mutations in exonic splicing enhancers can lead to partial or complete exon skipping, as observed in the tumor suppressor gene BRCA1. ⁴² A comparable mechanism has been reported for CYP2D6, where a common variant (rs16947) also disrupts an exonic splicing enhancer in exon 6, leading to enhanced exon skipping. ⁴³ These data support that disruption of exonic splicing enhancers may lead to residual enzymatic activity, as observed with CYP3A5*6 in our data. In contrast, the CYP3A5*3 allele (6986A>G) results in a premature stop codon, truncation of the protein, and complete loss of enzyme activity. ⁴¹

This study has several limitations. First, we did not have genome‐wide genotyping data for ancestry inference. However, we adjusted for ancestry using both genetic (HLA PCA) and demographic (country of birth) data. The high polymorphism and population‐specific distribution of HLA alleles make them valuable for capturing genetic ancestry. ³⁰ , ³¹ Notably, this is also the first pharmacogenomic study to apply HLA‐based principal components for ancestry adjustment. These ancestry proxies may additionally help account for unmeasured environmental and genetic confounding. Second, while we used real‐world data from a large cohort with unique longitudinal AUC_0‐24 data, including one of the largest numbers of CYP3A5*6 carriers, the absolute number remains modest, potentially limiting power in ancestry‐stratified analyses. Third, we had no data on the occurrence of other pharmacogenomic factors influencing tacrolimus metabolism in African‐ancestry populations, such as CYP3A5*7 ¹⁴ , ³⁹ , ⁴⁴ or rare CYP3A5 variants, as well as genetic variants in other genes that have been inconsistently associated with variability in tacrolimus PK, like CYP3A4, ABCB1, and PPARA. ¹⁴ Of these genetic variants, CYP3A4*22 is the most investigated contributor to tacrolimus clearance. ¹³ It has a minor allele frequency (MAF) of 5% ⁴⁵ in Europeans but is nearly absent in African populations (< 0.1%). ⁴⁵ Considering the prevalence of *22 carriers is 5% in the European population and the 17% lower tacrolimus clearance, the absolute contribution to the mean clearance of CYP3A5*3/*3 is less than 1% (0.05 × 17). Therefore, when comparing the *6 carriers to the *3 carriers, the effect of CYP3A4*22 will be marginal. Additionally, CYP3A4 gain‐of‐function variants such as CYP3A4*18 have been reported in African populations. However, their effects appear substrate‐specific, and they are characterized by an MAF < 1%. ⁴⁶ Additionally, we did not consider non‐genetic contributors such as indigenous diet patterns, ⁴⁷ food‐drug interactions ⁴⁸ , the gut microbiome, ⁴⁹ or concomitant drug use, ⁵⁰ all of which have been related to variability in drug exposure. However, AUC monitoring at our center is only performed in clinically stable patients and not during known drug interactions, further limiting the risk of systematic bias. Nevertheless, residual confounding was minimized by ancestry adjustment, and the large dataset of 4,177 AUC_0‐24 measurements, representing the largest AUC‐based pharmacogenomic study of tacrolimus to date, further enhances robustness. Additionally, our use of AUC_0‐24 rather than C₀ as the primary PK parameter provides a more comprehensive assessment of tacrolimus exposure. ¹³

Our findings likely extend beyond CYP3A5 in highlighting a fundamental limitation in pharmacogenomics: Eurocentric bias not only contributes to unequal health outcomes but also overlooks scientific opportunities. ⁴ African populations harbor the largest genetic diversity on Earth. ³⁴ Many variant alleles, like CYP3A5*6, are virtually absent in European populations and are therefore overlooked in Eurocentric pharmacogenomic research. As a result, population‐enriched variants in African genomes remain undercharacterized. ⁴ In transplant pharmacology, this “African gap” has been explicitly recognized: data on calcineurin inhibitor pharmacogenomics in African‐descent populations remain sparse, with most studies too small or heterogeneous to guide dosing strategies reliably. ¹⁵ Notably, the two largest randomized controlled trials evaluating CYP3A5‐guided tacrolimus dosing were conducted in predominantly European cohorts. ⁵¹ , ⁵² As noted by Shuker et al., ⁵² the omission of other genetic variants in CYP3A5, such as *6, may partially account for the unexplained variability in tacrolimus exposure observed in their study. Clinicians are left to extrapolate results from European cohorts to other populations. Bridging the ethnicity gap in pharmacogenomics requires a structural shift: expanding genotyping panels and clinical guidelines to include population‐enriched variants like CYP3A5*6, incorporating African genome data into dosing algorithms, and prioritizing large‐scale, inclusive research across Africa and the African diaspora. Our study contributes to this imperative by demonstrating how a variant allele that is nearly absent in European datasets can meaningfully influence immunosuppressive therapy, potentially improve transplant equity, and represent a tangible opportunity to advance health equity and improve personalized medicine for all.

FUNDING

This study received no external funding.

CONFLICTS OF INTEREST

In the last 3 years, T.V.G. Has received lecture fees from Otsuka, Chiesi, and Astellas, in addition to consulting fees from MaxyMune Tx, Roche Diagnostics, Thermo Fisher, CSL Behring, and Astellas. J.J.S. has received lecture fees from Lundbeck, paid directly to his employer (Leiden University Medical Center). All other authors declared no competing interests for this work.

AUTHOR CONTRIBUTIONS

A.D.L., D.J.A.R.M., S.B., T.V.G., A.P.J.V., D.L.R., and J.J.S. wrote the manuscript. A.D.L., D.J.A.R.M., S.B., T.V.G., S.M., and J.J.S. designed the research. A.D.L., Y.B., D.J.A.R.M., D.K.V., D.H., and S.B. performed the research. A.D.L., D.J.A.R.M., S.B., T.V.G., A.P.J.V., and J.J.S. analyzed the data.

Supporting information

Data S1