Cytochrome P450 Family 4F2 and 4F11 Haplotype Mapping and Association with Hepatic Gene Expression and Vitamin K Hydroxylation Activity

Abstract

This study evaluated the underlying mechanistic links between genetic variability in vitamin K metabolic pathway genes (CYP4F2 and CYP4F11) and phylloquinone hydroxylation activity using genotype- and haplotype-based approaches. Specifically, we characterized genetic variability in the CYP4F2/CYP4F11 locus and compared common single allele genotypes and common haplotypes as predictors of hepatic gene expression, enzyme abundance, and phylloquinone (VK₁) ω-hydroxylation kinetics. We measured CYP4F2 and CYP4F11 mRNA levels, CYP4F2 and CYP4F11 protein abundances, and the VK₁ concentration-dependent ω-hydroxylation rate in matched human liver nucleic acid and microsome samples, utilizing a novel in vitro population modeling approach. Results indicate that accounting for the CYP4F2*3 allele alone is sufficient to capture most of the genetic-derived variability in the observed phenotypes. Additionally, our findings highlight the important contribution that CYP4F11 makes toward vitamin K metabolism in the human liver.

The cytochrome P450 4 (CYP4) family of enzymes plays an important role in the metabolism of fatty acids and some fat-soluble vitamins (i.e., vitamin E and K),¹ generating in some instances important signaling molecules, such as arachidonic acid-derived eicosanoids (e.g., prostaglandins, leukotrienes, and thromboxane),² which are involved in the regulation of a variety of physiologic processes (i.e., immune response, cardiovascular health, and hemostasis). Importantly, several studies have shown that nucleotide variation in CYP4 genes contributes to interindividual differences in disease susceptibility or altered drug pharmacodynamics. Notable examples include the CYP4A11 variant rs1126742 (A > G) allele, which is associated with hypertension in White ancestry populations³ and proposed to be caused by diminished 20-hydroxyeicosatetraenoic acid (20-HETE) synthesis from arachidonic acid, the CYP4F3 variant rs4646904 (G > A) allele associated with ulcerative colitis and lung cancer in smokers^4,5 and the CYP4 V2 variant rs13146272 (C > A) allele associated with the risk of deep venous thrombosis and tamoxifen-induced venous thrombosis.⁶⁻⁸ Of particular interest for this study are the downstream effects of genetic variation at the CYP4F2 and CYP4F11 gene loci on vitamin K metabolism, associated pathophysiology, and the pharmacological response to oral vitamin K antagonists such as warfarin.

Vitamin K (VK) refers to a group of lipid-soluble molecules recognized in significant part for their role in regulating blood coagulation. VK exists in two forms, phylloquinone (VK₁), found predominantly in leafy greens, and a class of molecules called menaquinones (including menaquinone-4, i.e., VK₂), found in meats and fermented foods.⁹ VK serves as a cofactor for the post-translational γ-carboxylation of glutamic acid residues of Gla proteins, including key clotting factors.¹⁰ In addition to supporting hemostasis, VK and Gla protein production play an active role in cardiovascular health, bone remineralization, renal function, and cell growth.¹¹ CYP4F2 and CYP4F11 are the principal catalysts of the metabolic elimination of VK and have a direct effect on tissue (e.g., hepatic) exposure to VK, with CYP4F2 playing the dominant role in the initiation of VK catabolism by ω-hydroxylation.¹²

The VK antagonist, warfarin, is a widely used oral anticoagulant with proven efficacy in treating certain thromboembolic disorders and preventing deep vein thrombosis and pulmonary embolism. Despite its efficacy, long-term use of warfarin is associated with morbidity and mortality primarily from major bleeding events that can arise without warning, due to the drug’s narrow therapeutic window.¹³ Variation in two genes responsible for warfarin pharmacokinetics (CYP2C9) and pharmacodynamics (VKORC1) has repeatedly been found to be significantly associated with warfarin dose response, explaining approximately 15 and 25% of the therapeutic dose variability,^14,15 respectively. GWAS studies also attribute a CYP4F2 coding variant, rs2108622 (C > T, V433M, CYP4F2*3), to 1–7% of the variability in warfarin dose–response.¹⁶ Importantly, a previous study evaluating genotype-phenotype relationships found a significant association between the rs2108622 TT genotype and lower microsomal CYP4F2 protein concentration, paired with reduced VK1 oxidation.^12,17 However, in a follow-up study, microsomal CYP4F2 protein content only explained 13% of the variability seen in VK2 ω-hydroxylation rate in vitro,¹² suggesting that other factors (genetic or nongenetic) besides CYP4F2 tissue abundance may control VK catabolism. In addition, a substrate-dependent genotype association was observed in a study evaluating the effect of the CYP4F2*3 variant on 20-HETE production and leukotriene B4 (LTB4) metabolism, where CYP4F2*3 was associated with a 50% reduction in 20-HETE production but no change in LTB4 ω-hydroxylation.¹⁸

A comprehensive analysis investigating variation in the CYP4F2/CYP4F11/CYP4F12 loci (select single nucleotide variants (SNVs) and extended haplotypes) found significant associations with mRNA expression and that this variability was also associated with warfarin dose–response.¹⁹ The study observed potentially competing effects of different SNVs within the same gene cluster, resulting in the “canceling out” of independent SNVs associations with mRNA expression and warfarin pharmacodynamics. Specifically, the CYP4F11 variant rs1060467 was associated with a decrease in the therapeutic warfarin dose; CYP4F2*3 was associated with an increase in the therapeutic warfarin dose; and a haplotype that combines both these variants was not associated.¹⁹ The authors noted that although the absolute effect of SNVs in CYP4F2 and CYP4F11 on warfarin dose–response variability in their study was small, populations with different linkage-disequilibrium patterns may yield haplotypes with larger effects.

Taking all of this into account, the objective of this study was to reassess, using a well-characterized human liver bank, the relationship between the VK metabolic pathway gene variation and phenotypic markers of VK metabolic activity. We hypothesized that variation at the CYP4F2 and CYP4F11 gene loci contributes to interindividual variability in hepatic protein expression and intrinsic clearance of VK₁ in a haplotype/diplotype-dependent manner. Specifically, we sought to determine whether stratification of livers by CYP4F2/CYP4F11 diplotypes can improve our ability to identify poor and extensive metabolizers of VK₁, compared with single allele genotyping alone.

Materials and Methods

Chemicals

Phylloquinone, lapachol, NADPH, potassium phosphate, EDTA, and methyl-β-cyclodextrin (MβCDX) were purchased from Sigma-Aldrich (St. Louis, MO). Acetonitrile, UPLC-grade water, hexane, iodoacetamide, trypsin, and dithiothreitol were purchased from Thermo Fisher Scientific (Waltham, MA). Unlabeled and ¹³C,¹⁵N stable isotopically labeled (SIL) peptides were purchased from New England Peptide (Harvard, MA) and Thermo Fisher (Waltham, MA), respectively.

CYP4F2 and CYP4F11 DNA Sequencing, RNA Quantification, and Variant Selection

The Pharmacogenetic Research Network (PGRN) next-generation sequencing platform PGRNseq version1.1 was used to identify CYP4F2 and CYP4F11 genomic variation following previously described methods.^20,21 Liver RNA was isolated, purified, and quantified previously using a formerly described and validated method.²⁰

Linkage Disequilibrium Analysis and Haplotype and Diplotype Classification

A total of 32 single nucleotide variants within the CYP4F2 and CYP4F11 gene loci were identified. Those with a minor allele frequency of less than 0.05 (n = 19) were excluded from further evaluation and the remainder (n = 13) were included for statistical analysis. Each variant was assessed for Hardy–Weinberg equilibrium (HWE) using a Chi-Square test (χ²) with the threshold p-value set to 0.01 for inclusion in haplotype analysis. Haplotypes and diplotypes were generated for 274 resequenced human liver DNA samples using the statistical deconvolution software PHASE 2.1.1;²² individuals with a haplotype pair probability of <90% were excluded from phenotype association tests. Linkage disequilibrium (LD) was calculated between genetic variants found across CYP4F2 and CYP4F11 loci. Pairwise linkage disequilibrium relationships were both calculated and plotted in a correlation matrix using the software Haploview (version 4.2);²³ LD block and D′ values were generated using the default algorithm. A total of 20 unique haplotypes and 50 unique diplotypes were identified in the liver bank population; a summary of variant information for each haplotype and diplotype can be found in Tables 1 and 2.

Table 1. Characteristics of Select CYP4F2 and CYP4F11 Variants.

RSID	Gene	Location (Chr. 19)	MAFa	Alleles	Amino Acid Δ	Mutation Type
rs2108622 (*3)	CYP4F2	15990431	0.256	C > T	Val/Met	Missense
rs2074900	CYP4F2	15996820	0.323	G > A	His	Synonymous
rs3093160	CYP4F2	15996907	0.147	C > T		Intron
rs3093153	CYP4F2	16001215	0.059	C > A	GLY/VAL	Missense
rs3093114	CYP4F2	16006413	0.152	G > A	ALA	Synonymous
rs3093106	CYP4F2	16008257	0.161	T > C	PRO	Synonymous
rs3093105 (*2)	CYP4F2	16008388	0.159	A > C	TRP/GLY	Missense
rs3093103	CYP4F2	16008434	0.152	A > G		Intron
rs3093100	CYP4F2	16008469	0.159	C > G		Intron
rs1060463	CYP4F11	16025176	0.412	T > C	ASP/ASN	Missense
rs8104361	CYP4F11	16034714	0.203	G > A	CYS/ARG	Missense
rs3765070	CYP4F11	16040292	0.412	G > A	ILE	Synonymous
rs2305801	CYP4F11	16045141	0.206	C > T	GLY	Synonymous

^aMinor allele frequency (MAF) > 0.05 for all variants. Star allele nomenclature denoted next to RSID if available.

Table 2. Population Michaelis–Menten Model Parameter Estimate Comparisons^a.

	(1) Base Model		(2) Genotype Covariate Model		(3) Diplotype Covariate Model		Change (%)
	Estimate	95% CI	Estimate	95% CI	Estimate	95% CI
θTypical,Vmax	0.14***	[0.12–0.18]
θTypical,Km	4.22***	[3.6–4.95]
Genotype Estimates
Vmax *1/*1			0.17 ***	[0.13–0.23]
Vmax *1/*3			0.17	[0.1–0.8]			–2% ↓
Vmax *3/*3			0.07 **	[0.04–0.13]			–57% ↓
Km *1/*1			4.25 ***	[3.5–5.15]	4.24 ***	[3.48–5.17]
Km *1/*3			2.89	[1.96–4.25]	2.89	[1.94–4.31]	–32% ↓
Km *3/*3			8.66 **	[5.5–13.65]	8.6 **	[5.4–13.74]	103% ↑
Diplotype Estimates
Vmax13/13					0.17 ***	[0.11–0.27]
Vmax13/17					0.17	[0.07–0.41]	–2% ↓
Vmax1/17					0.32	[0.09–1.1]	82% ↑
Vmax13/20					0.18	[0.08–0.39]	5% ↑
Vmax6/13					0.49 *	[0.21–1.15]	181% ↑
Vmax1/20					0.11	[0.03–0.39]	–36% ↓
Vmax6/20					0.17	[0.05–0.61]	0%
Vmax20/20					0.07 *	[0.03–0.14]	–62% ↓
Vmax1/13					0.28	[0.1–0.8]	61% ↑
Vmax6/6					0.14	[0.07–0.29]	–21% ↓
Vmax17/20					0.08	[0.03–0.18]	–63% ↓
Vmax4/6					0.15	[0.06–0.38]	–14% ↓
Vmax4/4					0.21	[0.06–0.73]	21% ↑
Vmax1/6					0.14	[0.05–0.35]	–21% ↓
Vmax12/13					0.15	[0.07–0.33]	–12% ↓
Vmax1/4					0.07	[0.02–0.26]	–58% ↓

Residual (Power model)	0.848	0.835	0.838
Subjects (n)	88	88	88
logLik	455.057	471.942	486.917
AIC	–898.113	–923.884	–917.833

^a*** p < 0.001; ** p < 0.01; * p < 0.05. Units for V_max and K_m are PAR/min/mg microsomal protein and μM, respectively. Log-likelihood; logLik, Akaike information criterion; AIC. Diplotype groups containing only <2 subjects were omitted from analysis (6/17, 1/12, 17/17, 4/13, and 12/12). Parameters reported as Estimate [95% CI].

Human Liver Microsome (HLM) Preparation

Liver microsomes were prepared using an adaptation of a previously described method.¹⁷ Human liver tissue (1.6–1.7 g frozen weight) was thawed on ice in prechilled buffer (0.25 M sucrose, 1 mM EDTA, 50 mM potassium phosphate, pH 7.4). Fat and connective tissue were removed, and the remaining tissue was cut into fine pieces before homogenization in an Omni Bead Mill Homogenizer (Omni International, Kennesaw GA). The homogenate was fractionated by centrifugation (15,000g for 30 min at 4 °C), and the supernatant was isolated. The microsomal pellet was collected upon further fractionation by ultracentrifugation (120,000g for 70 min at 4 °C). The pellet was resuspended in wash buffer (10 mM potassium phosphate, 0.10 mM potassium chloride, 1 mM EDTA), rehomogenized using a Dounce homogenizer, and repelleted by ultracentrifugation (120,000g, 70 min at 4 °C). The final pellet was resuspended in a minimal volume (200–250 μL) of buffer (0.25 M sucrose, 10 mM EDTA, and 50 mM potassium phosphate, pH 7.4) and stored at −80 °C. A bicinchoninic acid assay (BCA) was used to determine total microsomal protein content.²⁴

Quantification of Hepatic CYP4F2 and CYP4F11 Proteins

The abundance of hepatic CYP4F2 and CYP4F11 in liver microsomes was quantified using an adaption of a previously validated LC-MS/MS-based targeted proteomics method.¹² Surrogate peptides for CYP4F proteins were selected based on in silico criteria.¹² Peptides potentially affected by nonsynonymous single-nucleotide polymorphisms (SNPs), or those susceptible to degradation, were not selected. Peptides with contiguous arginine and lysine sequences (RR, RK, KR, and KK) were also excluded from the peptide selection pool. Of the remaining peptide pools, a genome wide BLAST search was performed to identify those specific for human CYP4F2 and CYP4F11. Unique peptide sequences for synthetic CYP4F2 (SVINASAAIAPK) and CYP4F11 (TLTQLVTTYPQGFK) and stable-isotope-labeled (SIL) forms of these peptides (labeled lysine residue; two nitrogen isotopes ¹⁵N and six carbon isotopes ¹³C atoms) were prepared by Thermo Fisher (Waltham, MA) and used for quantitative analyses. Human liver microsomes (40 μL, 2 mg of protein/ml) were incubated with 4 μL of dithiothreitol (100 mM) and 40 μL of ammonium bicarbonate buffer (100 mM, pH 7.8) for 5 min at 95 °C (denaturation and reduction). Samples were cooled to room temperature, and cysteines were alkylated by addition of 4 μL of iodoacetamide (200 mM) for 20 min at room temperature in the dark. 10 μL of a trypsin solution were added (0.16 mg/mL), and samples were digested at 37 °C for 24 h. At the end of the digestion period, the reaction was quenched by adding 30 μL of a 50 nM internal standard solution (CYP4F2 and CYP4F11 SIL peptides dissolved in a 50% acetonitrile, 0.1% formic acid, v/v), and the mixture centrifuged at 5,000g for 5 min at 4 °C and supernatant stored at −80 °C until analysis. Derivatized sample peptides and internal standards were analyzed using electrospray ionization (positive) under multiple reaction monitoring (MRM) conditions using a Xevo TQ-XS triple-quadrupole mass spectrometer coupled to an Acquity liquid chromatography system (Waters). Ten microliters of sample were injected onto a C18 1.7 μm, 100 mm × 2.0 mm UPLC column (Waters).

LC conditions were as follows: the mobile phase consisted of 0.1% formic acid (v/v) in water (A) and acetonitrile (B), at a flow rate of 0.3 mL/min. The mobile phase gradient was as follows: 0–2 min (hold 10% B, v/v), 2–4 min (10–90% B, v/v), 4–6 min (hold 90% B, v/v), 6–7.5 min (90–10% B, v/v), 7.5–8 min (10% B, v/v). MRM mass transitions that were monitored and the peptide retention times are listed, with m/z fragments in brackets: CYP4F2 retention time = 3.71 min, m/z 571 > [499.3, 657.4, and 728.4]; CYP4F11 retention time = 3.83 min, m/z 789 > [576.4, 941, and 1040.6] (parent z = 2, fragment z = 1); CYP4F2 SIL retention time = 3.68 min, m/z 575.3 > [507.3, 665.4, and 736.4]; CYP4F11 SIL retention time = 3.85 min, m/z 802.9 > [584.4, 949.6, and 1048.6] (parent z = 2, fragment z = 1). Instrument settings for CYP4F2 and CYP4F11 peptides were as follows: dwell time, 10 ms for both; cone, 30 V; and collision energy, 20 eV. A standard curve was constructed using the PAR of an unlabeled peptide standard to a labeled peptide standard for both CYP4F2 and CYP4F11 with a range of 0.10–100 pmol/mL (R² > 0.99). Protein concentrations were quantified by calculating peak area ratios (PAR) of unlabeled and SIL (IS) peptide fragment transitions and interpolating concentrations against the linear peptide calibration curves. Quantified samples were normalized to total microsomal protein content per mL of microsomal solution for final units of (pmol CYP)/(mg microsomal protein).

Microsomal Activity Assay: ω-Hydroxy-Phylloquinone Formation

Combined CYP4F2 and CYP4F11 mediated catalytic activity of selected human liver microsomes (HLMs) was measured by monitoring the formation of ω-hydroxyphylloquinone from VK₁ over time. VK₁ incubations were performed in a matrix of 0.5 mL of 0.1 M KPi buffer (pH 7.4) with 1 mM EDTA, 1 mM MβCDX, and 1–2 mg/mL HLM. VK₁ was dissolved in 2-propanol and added to buffer and microsomes to achieve the appropriate nominal concentrations, with a final organic solvent concentration <0.7%. These incubation mixtures were preincubated in a water bath for 3 min at 37 °C, and the VK₁ hydroxylation reaction was initiated by the addition of NADPH (1 mM final concentration). After a 30 min incubation, the reactions were quenched by the addition of 2 mL of chilled hexanes followed by 50 μL of internal standard (1 μM, lapachol). The reaction product in quenched incubation mixtures was extracted (LLE) with hexanes (2 × 2 mL). The hexane layer was evaporated under N₂-gas and the residue reconstituted in 200 μL of 50:50 water and 2-propanol for analysis.

UPLC-MS/MS Analysis of ω-Hydroxy-Phylloquinone

Microsomal production of ω-hydroxy VK₁ was quantified by ultraperformance liquid chromatography tandem mass spectrometry (UPLC-MS/MS) using previously published multiple reaction monitoring conditions¹² (Supplemental Figure 1). In summary, liquid chromatography–mass spectrometry analyses were conducted on an ACQUITY Ultra Performance LC (UPLC) system with integrated Waters autoinjector coupled to a Waters Xevo TQ-XS tandem quadrupole mass spectrometer (Waters Co., Milford MA). MS/MS analysis was performed under atmospheric pressure negative chemical ionization (APCI^–) with multiple reaction monitoring (MRM). The instrument source temperature was set at 150 °C with a probe temperature set to 650 °C. The following mass transitions (m/z) for each analyte were monitored on separate ion channels: m/z 466.3 > 185 (ω-hydroxy VK₁); m/z 241 > 186 (lapachol). The cone voltages were set at 58 and 40 V for ω-hydroxy VK₁ and lapachol, respectively. Collision energies were set to 30 and 20 V for ω-hydroxy VK₁ and lapachol, respectively.

Analytes were separated on an Acquity UPLC C18 1.7 μm, 100 × 2.0 mm column (Waters) with a water (A) and methanol (B) mobile phase. The mobile phase gradient ran as follows: solvent B held at 40% for 0–1 min, ramped up to 98% between 1 and 3.5 min, held at 98% B from 3.5 to 8.0 min, ramped down to 40% B from 8 to 8.5 min, held at 40% B from 8.5 to 9.5 min. Retention times for ω-hydroxy VK₁ and lapachol were 5.27 and 3.92 min, respectively. ω-Hydroxy VK₁ identity was confirmed by high-resolution mass spectrometry using an AB Sciex 5600 Q-TOF LC-MS/MS system (data not shown), and its formation in human liver microsomes confirmed using the probe CYP4F inhibitor HET0016 (Supplemental Figure 2). No authentic standard was available, and thus data are reported as peak area ratios (PAR) for the metabolite and internal standard. Product formation linearity over time (0–40 min) was confirmed at low and high substrate concentrations prior to setting the VK₁ incubation conditions. Analyte detection proportionality under the LC-MS/MS conditions was confirmed across a 250-fold serial dilution of the maximum amount of ω-hydroxy VK₁ formed during the 30 min of the HLM incubation period. This encompassed all of PAR values generated from individual HLM incubations.

In Vitro Study Design

In vitro incubation concentrations were selected using a strategic sparse sampling approach, where 2 sets of 3 nonoverlapping substrate concentrations spanning the range of the Michaelis–Menten profile (Set 1: 0.5, 5, and 50 μM, Set 2: 1, 10, and 100 μM VK₁). For each genotype/diplotype group, half of the subjects were assigned to set 1 and the other half to set 2. Using an optimal design estimator (PopED) we confirmed from preliminary data of a sparse sampling design (n = 3 pts per subject) that a sampling efficiency ratio at the proposed substrate concentrations relative to a rich sampling design (n = 6–9 pts per subject) was between 1 and 1.2, demonstrating equivalency and comparable relative standard error for both set 1 and set 2 concentrations using the D-optimality criterion.

Statistical Analysis

Statistical analysis was performed using R version 4.0 via RStudio. For liver bank mRNA analysis, CYP4F11 mRNA abundance (fragments per kilobase per million bases read, FPKM) were square root transformed to obtain normality; CYP4F2 mRNA expression data were normally distributed without transformation (Supplemental Figure 3). mRNA data were further normalized to the median expression from the individual genes. To evaluate the association between genotype and mRNA expression, a one-way analysis of variance (ANOVA) model was generated with a significance threshold set to a p-value of 0.05 and expression levels were compared across alleles; post hoc analysis was conducted using a Tukey HSD test. Parametric test assumptions of normality and homogeneity of variance were assessed using a Shapiro-Wilk test and Bartlett test, respectively. Wilcox/Mann–Whitney test was used to compare mean mRNA expression by sex. The impact of age on mRNA and protein expression was evaluated by comparing mean abundances across binned age groups (≤18 year, 19–55 years, and 55+ years) using one-way ANOVA and posthoc Tukey HSD analysis. Genotype and diplotype groups containing only 1 individual were excluded from mRNA statistical analysis. Linear regression models were used to evaluate additive inherited allele effects and covariate interactions on parameter estimates. Coding of individual genotypes for additive inherited allele effects was as follows: WT = 0, heterozygotes = 1, homozygous variants = 2. Coding for haplotypes and diplotype was as follows: 0 = zero copies of specified variant combination, 1 = one copy of specified variant combination, 2 = two copies of specified variant combination. Diplotype groups with n < 2 (6/17, 1/12, 17/17, 4/13, and 12/12) are reported in summary tables but excluded from statistical analysis because of low count. VK₁ hydroxylation activity of the different liver microsomal preparations was evaluated using nonlinear mixed effect modeling (see the “Population Michaelis–Menten Modeling” section described below). Age, sex, CYP4F2/4F11 protein, and cytochrome P450 reductase abundance data were evaluated as additional covariates during the model optimization steps. Estimates in the text are reported as the mean ± standard deviation unless otherwise stated. Posthoc analysis of diplotype associations were conducted using Dunnett’s test for multiple comparisons and diplotype 13/13 as the reference group. The intrinsic clearance estimates adjusted for CYP4F2 and CYP4F11 protein content were evaluated using multiple linear regression; reported p-values were adjusted for false discovery rate using the Benjamin–Hochberg correction method.²⁵

Population Michaelis–Menten Modeling

Population Michaelis–Menten Modeling (PopMM) was performed in R (v4.2.0) using the nonlinear mixed effect modeling package NLME²⁶ and supporting packages for data manipulation and data visualization. The final model was built by fitting a base Michaelis–Menten model with a random-effect structure and covariates. The model was derived from the Michaelis–Menten equation with modifications to V_max and K_m to incorporate both fixed and random-effects. Product formation rate for the ith individual at the jth substrate incubation concentration, defined as (V_ij), was modeled as a function of Φ_i and substrate concentration (S_i) plus residual error (ε_ij); where Φ_i is a matrix of parameters (Φ_1i and Φ_2i) each being vectors of fixed and random effects whose sum equates to the predicted individual-level estimates of V_max and K_m, respectively (eq 1). Individual estimates of V_max (Φ_1i) and K_m (Φ_2i) were defined as the natural log of the sum of a typical population value for V_max (θ_1,Vmax) or K_m (θ_1,Km) and a fixed covariate effect (θ_2,Vmax or θ_2,Km) based on individual CYP4F2 genotype or diplotype (X_i), plus an individual specific random effect on both V_max and K_m (η_1i, and η_2i, respectively) (eq 2). Model parameters (θ, X, and η) were incorporated using an exponential variation model to account for non-normal distribution of population estimates. Individual level random effects (η_i) and residual error (ε_ij) were assumed to be normally distributed with a mean of 0 and ε_ij having a variance of σ². V_max and K_m were assumed to be uncorrelated (independent), therefore variance (ψ_i) for η_i was structured as a diagonal matrix with covariance set to 0.

1

2

Intrinsic clearance (CL_int) estimates were calculated as the quotient of V_max/K_m, with population level intrinsic clearance defined as (θ_1,Vmax)/(θ_1,Km), and genotype (or diplotype) level defined as the sum of θ_1,Vmax and θ_2,Vmax for a given genotype divided by the sum of θ_1,Km and θ_2,Km for a given genotype: (θ_1,Vmax + θ_{2,Vmax,Genotype})/(θ_1,Km + θ_{2,Km,Genotype}). Selection of the appropriate covariates and error model were based on Akaike criterion (AIC), F-tests (ANOVA), and diagnostic plots. The final model was weighted using a power residual error structure based on optimization of AIC and BIC and log-likelihood criteria. The impact of genotype (and diplotype) on intrinsic clearance was evaluated using multiple regression of the PopMM derived intrinsic clearance estimates; with intrinsic clearance defined as Φ₁/Φ₂ as described in eq 3. For this analysis, intrinsic clearance estimates were natural log-transformed to correct for non-normal distribution. The reference groups (intercept) for the genotype and diplotype multiple regression models were CYP4F2*1/*1 and CYP4F diplotype 13/13, respectively.

3

Proteomic Based Prediction of In Vitro Intrinsic Clearance

The ω-hydroxy vitamin K₁ formation rate (v) and VK₁ intrinsic clearance (CL_int) for the observed data were predicted based on CYP4F2 and CYP4F11 protein abundance using model-derived estimates of V_max and K_m. An estimate of catalytic turnover (k_cat) was calculated by dividing model-derived V_max values for each CYP4F2*3 genotype by the total CYP4F (CYP4F2 + CY4F11) protein content (for that specific genotype); individual predictions were made using the following: substrate concentration used in the experiment, individual CYP4F2 protein abundance, individual CYP4F11 protein abundance, genotype-dependent catalytic turnover value (k_cat,Genotype), and a genotype-dependent Michaelis–Menten constant (K_m,Genotype) as independent variables as shown in eqs 4 and 5. A linear regression model was fit to the predicted vs. observed data and used to evaluate the predictive value of experimentally determined CYP4F protein abundance data toward metabolic activity.

4

5

Results

Study Population

Human liver tissue samples (n = 274) were obtained from two liver banks: University of Washington Human Liver Bank, Seattle, WA (n = 58) and St. Jude Liver Resource at the St. Jude Children’s Research Hospital in Memphis, TN (n = 216).²⁷ Of these 274 subjects, demographic information was available at the following percentages: age (77%, n = 213); sex (100%, n = 274), 58% male, and 41% female. For subjects included in the proteomic and activity analysis (n = 88) demographic data were available at the following percentages: ages (84%), with mean and standard deviation of 39 ± 20 years; sex (100%) with 59% male and 41% female; race/ethnicity (100%) with White = 96%, African American/Black = 1%, Asian = 1%, and Hispanic/Latino = 2%. Institutional review boards at both sites approved the collection and use of these samples for research purposes, and all links between archived tissues and donors were previously destroyed.

CYP4F2 and CYP4F11 SNV Identification

A total of 13 common gene variants (5 missense, 5 synonymous, and 3 intron) in the CYP4F2-CYP4F11 locus were detected in the liver bank population (Supplemental Figure 4). Of these, 9 variants were found in the CYP4F2 gene (rs2108622, rs2074900, rs3093160, rs3093153, rs3093114, rs3093106, rs3093105, rs3093103, rs3093100) and 4 in the CYP4F11 gene (rs1060463, rs8104361, rs3765070, rs2305801). For CYP4F2, rs2108622 (CYP4F2*3) was the most prominent nonsynonymous variant with a MAF = 0.26 and for CYP4F11 rs1060463 was the most prominent nonsynonymous variant with a MAF = 0.41. Details of all of the variants characterized can be found in Table 1.

Evaluation of CYP4F2 and CYP4F11 Hepatic mRNA Abundance

CYP4F2 and CYP4F11 mRNA abundance in the combined St. Jude and University of Washington Liver Banks (n = 274) was evaluated. Overall, the mean CYP4F2 mRNA expression was higher than CYP4F11 mRNA expression (48 ± 20 FPKM vs 23 ± 28 FPKM, respectively). A summary table of the population characteristics stratified by CYP4F2*3 genotype can be found in Supplemental Table 1. Associations between SNVs found in the CYP4F2 and CYP4F11 genes and hepatic mRNA abundances were investigated. There were no statistically significant associations between CYP4F2 mRNA abundance and CYP4F2 variants (Figure 1), including the CYP4F2*2 and CYP4F2*3 alleles. The same was true for variants in the CYP4F11 gene and its mRNA abundance (Figure 2). However, rs3093153 (missense mutation found in the CYP4F2 loci) showed a statistically significant association with CYP4F11 mRNA expression (ANOVA p = 0.014) (Supplemental Figure 5); post hoc analysis revealed higher hepatic CYP4F11 mRNA abundance when comparing heterozygous variant to homozygous reference allele (8.1% increase, Tukey HSD p = 0.036); additive inheritance effects could not be evaluated due to the low sample size in the homozygous variant group (n = 1). There were no allele dependent effects observed between CYP4F11 variants (rs1060463, rs2305801, rs3765070, and rs8104361) and CYP4F2 mRNA abundance (Supplemental Figure 6).

Figure 1.

Human liver bank (n = 274) hepatic CYP4F2 mRNA abundance stratified by variants found at the CYP4F2 locus. The y-axis represents normalized mRNA expression values, and the x-axis discrete number of variant alleles. p-values comparing mean expression across allele copy groups were obtained using a one-way ANOVA.

Figure 2.

Human liver bank (n = 274) hepatic CYP4F11 mRNA abundance stratified by variants found at the CYP4F11 locus. The y-axis represents normalized mRNA abundance values, and the x-axis discrete number of variant alleles. p-values comparing mean expression across allele copy groups were obtained using a one-way ANOVA.

Impact of Age and Sex on Hepatic mRNA Expression

Sex was identified as a significant covariate with females having modestly higher hepatic CYP4F2 mRNA abundance than males, with normalized FPKM values of 1.08 ± 0.41 and 0.94 ± 0.41, respectively (Wilcox p = 0.01), and higher CYP4F11 mRNA abundance than males at 1.02 ± 0.17 vs 1.00 ± 0.17, respectively (Wilcox p = 0.05) as shown in Supplemental Figure 7. Additionally, the age was also evaluated for its association with hepatic CYP4F2 and CYP4F11 mRNA abundance. Available age data (n = 212) was binned into 3 age categories: ≤18 years, 19–55 years, and 55+ years, then normalized mRNA abundance was compared by age group (Figure 3). There was a statistically significant difference in CYP4F2 mRNA abundance between the ≤18 years and 19–55 years age categories (23% lower, Tukey p = 0.01), as well as the ≤18 years and 55+ years age categories (23% lower, Tukey p = 0.01). Similarly, there was a statistically significant difference in CYP4F11 mRNA abundance between the ≤18 years and 19–55 years age categories (9.5% lower, Tukey p = 0.01) as well as the ≤18 years and 55+ years age category (8.9% lower, Tukey p = 0.02). There was no statistically significant difference between the 19–55 years and 55+ years age categories for both CYP4F2 and CYP4F11. Additionally, the combined impact of sex and age on mRNA abundance was evaluated by using multiple linear regression. There were no statistically significant interactions between sex and age when used as predictors of CYP4F2 or CYP4F11 mRNA abundance (CYP4F2, ANOVA p = 0.29; CYP4F11, ANOVA p = 0.34).

Figure 3.

CYP4F2 and CYP4F11 mRNA abundance by age category. The y-axis represents normalized mRNA abundance values. Comparison of mean mRNA abundances across age groups was evaluated using a one-way ANOVA. In the left panel a statistically significant difference in mean normalized CYP4F2 mRNA abundance was found between the following groups: ≤18 years and 19–55 years age groups, 0.82 ± 0.42 vs 1.07 ± 0.43, respectively (posthoc Tukey p = 0.01); ≤18 years and 55+ years age groups at 0.82 ± 0.42 vs 1.07 ± 0.43, respectively (posthoc Tukey p = 0.01). In the right panel a statistically significant difference in mean normalized CYP4F11 mRNA abundance was found between the following groups: ≤18 years and 19–55 years age groups, 0.94 ± 0.17 vs 1.04 ± 0.17, respectively (posthoc Tukey p = 0.01); ≤18 years and 55+ years age groups at 0.94 ± 0.17 vs 1.04 ± 0.16, respectively (posthoc Tukey p = 0.02).

Haplotype-Mapping of CYP4F2 and CYP4F11 Loci

Linkage disequilibrium (LD) analysis was conducted to determine associations among SNVs across CYP4F2 and CYP4F11 genes. Genomic crosstalk between CYP4F2 and CYP4F11 loci was explored by generating a single haplotype block spanning both regions, yielding 20 unique haplotypes incorporating the 13 previously mentioned SNVs. Strong associations were present between all SNVs across the loci; a visual representation of this can be found in Figure 4. The population frequencies and associations between the CYP4F2/4F11 haplotypes are summarized in Table 3. For our investigation, we assigned haplotype 13 as the reference CYP4F gene locus sequence based on its frequency in the population (31%, highest observed).

Figure 4.

Linkage-disequilibrium (LD) plot for pairwise D′ between variants across CYP4F2 and CYP4F11 (Chromosome 19). D′ values are indicated as percentages within squares with strong LD specified by red, light gray or pink indicates weak association, and white indicates no association.

Table 3. Liver Bank Haplotype Summary.

			Frequency
Haplotype Code	Haplotype	RSID	n	(%)
1	CGCCGTAACTGGC	rs1060463	50	8.96
2	CGCCGTAACTGAC	rs1060463, rs3765070	2	0.36
3	CGCCGTAACCGGT	rs2305801	1	0.18
4	CGCCGTAACCGAC	rs3765070	30	5.38
5	CGCCGTAACCAAC	rs8104361, rs3765070	6	1.08
6	CGCCGTAACCAAT	rs8104361, rs3765070, rs2305801	105	18.82
7	CGCCGCAACCGGT	rs3093106, rs2305801	1	0.18
8	CGCCGCCAGTGGC	rs3093106, rs3093105, rs3093100, rs1060463	2	0.36
9	CGCCGCCAGCGAT	rs3093106, rs3093105, rs3093100, rs3765070, rs2305801	2	0.36
10	CGCCACCGGTGGC	rs3093114, rs3093106, rs3093105, rs3093103, rs3093100, rs1060463	2	0.36
11	CGCCACCGGCGAC	rs3093114, rs3093106, rs3093105, rs3093103, rs3093100, rs3765070	1	0.18
12	CGCAGTAACTGGC	rs3093153, rs1060463	33	5.91
13	CACCGTAACTGGC	rs2074900, rs1060463	172	30.82
14	CACCGTAACTGGT	rs2074900, rs1060463, rs2305801	2	0.36
15	CACCGTAACCGAC	rs2074900, rs3765070	4	0.72
16	CACCGTAACCAAT	rs2074900, rs8104361, rs3765070, rs2305801	2	0.36
17	TGCCGTAACTGGC	rs2108622, rs1060463	59	10.57
18	TGCCGTAACTGGT	rs2108622, rs1060463, rs2305801	2	0.36
19	TGTCACCGGTGGC	rs2108622, rs3093160, rs3093114, rs3093106, rs3093105, rs3093103, rs3093100, rs1060463	4	0.72
20	TGTCACCGGCGAC	rs2108622, rs3093160, rs3093114, rs3093106, rs3093105, rs3093103, rs3093100, rs3765070	78	13.98

Hepatic CYP4F2 and CYP4F11 Proteomic Analysis

Individual human liver microsome donors (n = 88) were evaluated for CYP4F2 and CYP4F11 protein abundance using quantitative proteomics (Supplemental Figure 8 and Supplemental Table 2). Overall, the fractional contribution of CYP4F2 to the combined abundance was greater than CYP4F11 across most of the donors (n = 84) with the exception of 4 subjects (HL-149, SJLB335, SJLB174, and SJLB767), all of whom were CYP4F2*3 carriers and had CYP4F11 protein abundance accounting for 52–59% of the combined CYP4F protein pool. CYP4F2*1 homozygotes had the greatest CYP4F2 contribution accounting for 83% of the combined CYP4F protein pool, followed by CYP4F2*3 heterozygotes at 77% and CYP4F2*3 homozygotes at 61% fractional abundance on average (Figure 5). Additionally, when stratified by diplotype, carriers of haplotype 4 or 6 had the highest fractional CYP4F2 contribution accounting for 82–87% of the combined CYP4F protein pool; however, total protein content varied substantially (Figure 6). Diplotype 6/13 had the highest individual and combined CYP4F protein abundance out of all the groups analyzed, with a 40% higher combined protein abundance (21.73 ± 5.59 pmol/mg microsomal protein) compared to the reference group 13/13 (15.31 ± 6.4 pmol/mg microsomal protein). Additionally, diplotype 20/20 (*3/*3 carrier) had the lowest combined CYP4F protein abundance (4.97 ± 2.47 pmol/mg microsomal protein). Both CYP4F2 and CYP4F11 microsomal protein abundances were evaluated by age group (≤18 years, 19–55 years, and 55+ years) as shown in Supplemental Figure 9. Overall, there were no statistically significant differences in mean protein abundance for CYP4F2 and CYP4F11 across age groups (ANOVA, p = 0.11 and 0.16, respectively), nor was there interaction between sex and age on protein abundance (CYP4F2, ANOVA p = 0.57; CYP4F11, ANOVA p = 0.94).

Figure 5.

Cumulative percent of total protein abundance for CYP4F2 and CYP4F11 by CYP4F2*3 genotype.

Figure 6.

Fractional contribution of CYP4F2 and CYP4F11 toward combined microsomal protein abundance grouped by CYP4F2*3 genotype (top panel), and CYP4F diplotype (bottom-panel). Diplotype groups containing only <2 subjects were omitted from analysis (6/17, 1/12, 17/17, 4/13, and 12/12).

Population Michaelis–Menten Analysis

In a subpopulation of individuals from the UW and St. Jude Liver Banks (n = 88; selected for coverage of the common haplotypes) individual and population Michaelis–Menten (PopMM) parameters were estimated using a validated nonlinear mixed effect modeling method we have described previously.²⁸ ω-Hydroxy VK₁ formation over time was monitored across a range of substrate concentrations using a sparse sampling approach (Supplemental Figure 10). After fitting a base model incorporating random effects on V_max and K_m, the typical population values were 0.14 PAR/min/mg protein, 95% CI [0.12–0.18], and 4.22 μM, 95% CI [3.6–4.95]. A power residual error structure that provided the greatest improvement in objective values and corrected for the heteroscedastic nature of the observed data was selected to characterize residual variability. A summary of model diagnostics including residual error plots, individual, and population predicted versus observed values can be found in Supplemental Figure 11. Covariate selection was based on random-effect correlations for both V_max and K_m (Supplemental Figures 12 and 13). Two additional models, a CYP4F2 genotype covariate model and a CYP4F diplotype covariate model, were also evaluated for goodness of fit compared to the base model (Table 2). Incorporating CYP4F2*3 genotype as a fixed effect on V_max and K_m improved the model criterion (ΔAIC = −25.77, ANOVA p < 0.0001), and yielded the following respective V_max and K_m estimates: CYP4F2*1/*1: 0.17 PAR/min/mg protein, 95% CI [0.13–0.23], and 4.25 μM, 95% CI [3.5–5.15]; CYP4F2*1/*3, 0.17 PAR/min/mg protein, 95% CI [0.1–0.28], and 4.17 μM, 95% CI [2.48–7]; CYP4F2*3/*3, 0.07 PAR/min/mg protein, 95% CI [0.04–0.13], and 8.66 μM, 95% CI [5.5–13.65]. Importantly, CYP4F2*3/*3 was associated with a significantly lower V_max (−57%, p = 0.004) and higher K_m (+103%, p = 0.002). However, there were no significant differences in either V_max or K_m for the CYP4F2*1/*3 heterozygous variant group compared to the CYP4F2*1/*1 reference group.

The third model evaluated used the CYP4F diplotype as an explanatory covariate for deviations in V_max estimates. A total of 50 unique diplotype groups were identified in the liver bank population, of which 21 were characterized in vitro. Incorporation of CYP4F diplotype as a fixed-effect on V_max while maintaining CYP4F2*3 genotype as a fixed-effect on K_m improved model criterion compared to the base model (ΔAIC = −19.72, ANOVA p < 0.0001); however, it did not provide greater improvement in objective values compared to the incorporation of CYP4F2*3 genotype as a covariate alone (AIC_{Genotype-model} = −923.80, AIC_{Diplotype-Model} = −917.83, ANOVA p = 0.038 in favor of genotype model). Of the diplotypes analyzed, there were 2 haplotype pairs (20/20 and 6/13) with statistically significant differences in Michaelis–Menten parameters relative to the reference group (13/13). Diplotype 20/20 was associated with a 62% lower V_max compared to the reference group at 0.07 PAR/min/mg protein, 95% CI [0.03–0.14], and 0.17 PAR/min/mg protein, 95% CI [0.11–0.27], respectively, while diplotype 6/13 was associated with 181% higher V_max compared to the reference group (0.49 PAR/min/mg protein, 95% CI [0.21–1.15], and 0.17 PAR/min/mg protein, 95% CI [0.11–0.27], respectively). The magnitude and significance of the impact that the CYP4F2*3 genotype had on K_m remained constant after adjusting for diplotype effects.

To further assess the impact that the CYP4F2*3 variant has on metabolic activity, the genotype based PopMM model was adjusted to account for differential protein expression, by incorporating both CYP4F2 and CYP4F11 protein abundance as covariates affecting V_max. After adjusting the model for protein content, both CYP4F2 and CYP4F11 protein abundance were found to be a significant predictor of V_max, however incorporation of both proteins’ abundances as predictors did not significantly improve model performance compared to incorporation of CYP4F2 abundance alone; AIC = −734.47 (CYP4F2 + CYP4F11) vs AIC = −732.90 (CYP4F2 only). Additionally, after protein abundance adjustment, the CYP4F2*3 variant no longer posed as a significant predictor of V_max, suggesting that variability in V_max can be explained by variability in CYP4F2 abundance rather than variability in catalytic turnover (k_cat). Adjusting for protein content had no impact on the significant CYP4F2*3 related changes in K_m estimates. Additional models were evaluated and compared to identify the optimal covariates of response. Adding CYP4F2 protein abundance and cytochrome P450 reductase mRNA abundance as covariates for V_max while maintaining an allele-dependent CYP4F2*3 effect on K_m provided the best performing model for the observed data: AIC = −742.87.

Evaluation of In Vitro Hepatic Intrinsic Clearance

Population estimates of intrinsic clearance generated using the PopMM model were bimodal and non-normally distributed. Multiple linear regression was performed on the natural log-transformed estimates of intrinsic clearance, which corrected for the assumption of normality needed for the linear regression model (Shapiro–Wilks, p = 0.367 for natural log-transformed data). CYP4F2*3/*3 genotype was associated with a 65% lower intrinsic clearance compared to the reference CYP4F2*1/*1 genotype; however, the intrinsic clearance of the CYP4F2*1/*3 genotype group was not different from that of the CYP4F2*1/*1 group. Of the CYP4F diplotypes analyzed, 3 haplotype pairs were associated with statistically significant changes in intrinsic clearance compared to the reference diplotype group: dipolotypes 6/13, 17/20, and 20/20. Diplotypes 17/20 and 20/20 (both CYP4F2*3/*3 carriers) were associated with a 63% and 68% lower hepatic intrinsic clearance, respectively. In contrast, diplotype 6/13 was associated with a 175% higher hepatic intrinsic clearance compared to reference diplotype 13/13 (Table 4 and Figure 7). Although the genotype-based multiple regression model provided better performance than the diplotype-based model, the difference was marginal (ΔAIC = −8.12, ΔResidual Sum of Squares = 19.55, p = 0.167).

Table 4. Intrinsic Clearance Estimates Comparisons^a.

	(1) Genotype Covariate Model		(2) Diplotype Covariate Model
	Estimatea	95% CI	Estimatea	95% CI	Change (%)
β0
Reference	1.18	[0.92–1.51]	1.17	[0.75–1.83]
β1,Genotype
Genotype *1/*3	1.32	[0.82–2.14]			12% ↑
Genotype *3/*3	0.42 ***	[0.25–0.7]			–65% ↓
β1,Diplotype
Diplotype13/17			1.33	[0.54–3.25]	13% ↑
Diplotype1/17			2.43	[0.66–8.97]	107% ↑
Diplotype13/20			1.41	[0.64–3.11]	20% ↑
Diplotype6/13			3.23 *	[1.32–7.9]	175% ↑
Diplotype1/20			0.9	[0.24–3.32]	–23% ↓
Diplotype6/20			1.34	[0.36–4.95]	15% ↑
Diplotype20/20			0.38 **	[0.18–0.78]	–68% ↓
Diplotype1/13			1.88	[0.63–5.63]	60% ↑
Diplotype6/6			0.95	[0.45–2.03]	–19% ↓
Diplotype17/20			0.43 *	[0.19–1]	–63% ↓
Diplotype4/6			1.03	[0.39–2.74]	–12% ↓
Diplotype4/4			1.43	[0.39–5.28]	22% ↑
Diplotype1/6			0.95	[0.36–2.52]	–19% ↓
Diplotype12/13			1.04	[0.47–2.3]	–11% ↓
Diplotype1/4			0.52	[0.14–1.93]	–55% ↓

Subjects (n)	88	88
R2	0.174	0.398
logLik	–116.034	–102.094
AIC	240.068	248.188

^a***p < 0.001; **p < 0.01; *p < 0.05. Units for intrinsic clearance are PAR/mg microsomal protein/μM phylloquinone/min. Genotype covariate model reference = *1/*1. Diplotype covariate model reference = 13/13. Log-likelihood; logLik, Akaike information criterion; AIC. Diplotype groups containing only <2 subjects were omitted from analysis (6/17, 1/12, 17/17, 4/13, and 12/12). Parameters reported as Estimate [95% CI].

Figure 7.

Distribution of intrinsic clearance (CL_int) estimates stratified by (A) Genotype, and (B) Diplotype. Units for intrinsic clearance are defined as PAR/mg microsomal protein/μM phylloquinone/min. The dotted line represents intrinsic clearance estimate for reference CYP4F2 genotype (*1/*1).

Additionally, model derived estimates of in vitro activity were predicted and compared to observed data by evaluating the ω-hydroxy VK₁ formation rate (v) and VK₁ intrinsic clearance (CL_int) based on CYP4F2 and CYP4F11 protein abundance. Overall, incorporating CYP4F2 and CYP4F11 protein abundance data in conjunction with model-derived estimates of k_cat and K_m, 49% of the variability in experimentally observed reaction rates and 44% of the variability in experimentally observed CL_int values (v/[S], where S ≪ K_m) could be explained (Figure 8). Finally, to investigate the impact that the CYP4F2*3 and haplotype 6 have on CL_int and catalytic efficiency (k_cat/K_m), multiple linear regression analysis was performed. To correct for the non-normal distribution of CL_int, values were log-transformed prior to regression. The impact of genotype on catalytic efficiency was determined by adjusting the regression model to account for both CYP4F2 and CYP4F11 protein abundances on CL_int estimates by including both proteins as covariates. Unadjusted and untransformed data were also provided as a comparison (Supplemental Figure 14); objective function and goodness of fit values (AIC, log-likelihood, R²) are provided for model comparison purposes in Supplemental Tables 3 and 4. Overall, the CYP4F2*3 variant was no longer a significant predictor of metabolic activity when variability in CYP4F2 and CYP4F11 protein abundance was accounted for (Supplemental Figure 14D). Similarly, for haplotype 6, when log transformed CL_int for diplotypes 13/13 (reference), 6/13 (heterozygous variant haplotype), and 6/6 (homozygous variant haplotype) were analyzed, diplotype 6/13 was no longer a significant predictor of response after accounting for CYP4F2 and CYP4F11 protein abundance (Supplemental Figure 14H).

Figure 8.

Predicted vs experimentally observed reaction rate for ω-hydroxy phylloquinone (VK₁) formation (A) and CL_int (B) in human liver microsomes based on individual CYP4F2 and CYP4F11 protein abundances. Observed ω-hydroxy VK₁ formation rate for individual human liver microsomes was predicted using the Michaelis–Menten estimates generated from the PopMM analysis. Observed intrinsic clearance was defined as v/[S], where S is a substrate concentration between 3- and 18-fold below expected K_m value. The solid red line represents the line of unity, and the dotted black line represents a linear regression curve with 95% CI (shaded region). Units for parameter estimates are the following: reaction rate (v), PAR/min/mg microsomal protein; CL_int, PAR/mg microsomal protein/μM phylloquinone/min.

Correlation between mRNA, Protein, and Metabolic Activity

To identify possible mechanistic links between CYP4F2 and CYP4F11 mRNA abundances, CYP4F2 and CYP4F11 protein abundances, and VK₁ ω-hydroxylation activity, Pearson correlations were estimated and compared. While CYP4F2 mRNA abundance was not correlated with CYP4F2 protein abundance (R = 0.19, p = 0.078), there were significant positive correlations (Pearson’s R > 0.3, p < 0.005) between CYP4F11 mRNA abundance and CYP4F11 protein levels. Both CYP4F2 and CYP4F11 protein abundances were strongly correlated with ω-hydroxy VK₁ intrinsic clearance, as shown in Figure 9. There were no observable differences in CYP4F2 mRNA and CYP4F11 abundance across CYP4F2*3 genotypes (Kruskal–Wallis, p = 0.81 and 0.59 for CYP4F2 and CYP4F11 mRNA, respectively). In contrast, a significant allele-dependent lower CYP4F2 protein abundance was observed for CYP4F2*3 carriers, with heterozygous carriers (*1/*3) associated with a 19.4% lower abundance (10.3 pmol/mg microsomal protein), and homozygous variant carriers (*3/*3) associated with a 73% lower abundance (Figure 10). However, when total CYP4F2+CYP4F11 protein abundance was accounted for, there was no statistically significant difference in abundance between *1/*1 and *1/*3 carriers and an overall 63% lower abundance between *1/*1 and *3/*3 (Figure 11).

Figure 9.

Pearson correlation plots across CYP4F2 and CYP4F11 mRNA and protein abundance derived from 88 human liver tissue donors. (A,B) mRNA vs protein abundance, and (C,D) protein abundance vs intrinsic clearance (CL_int). Blue circles represent observed data, the solid red line represents unweighted linear regression fit, and the gray band represents 95% confidence interval. Units for CL_int are PAR/mg microsomal protein/μM phylloquinone/min.

Figure 10.

CYP4F2 and CYP4F11 mRNA and protein abundance derived from 88 human liver tissue donors and relationship to corresponding CYP4F2*3 genotype (n: *1/*1 = 53, *1/*3 = 19, *3/*3 = 16). (A and C) mRNA abundance; (B and D) protein abundance. Analysis of variance between CYP4F2 genotypes was performed using a Kruskal–Wallis test.

Figure 11.

Combined CYP4F2 and CYP4F11 protein abundance, and associations with (A) CYP4F2 genotype and (B) intrinsic clearance. Samples were derived from 88 human liver tissue donors. Estimates for total protein abundance (mean ± SD) for each CYP4F2*3 genotype are as follows: *1/*1, 15.3 ± 5.3 (n = 53); *1/*3, 13.4 ± 4.2 (n = 19), *3/*3, 5.6 ± 2.1 (n = 16) pmol/mg microsomal protein. Analysis of variance for protein abundance across CYP4F2 genotypes was performed using a Kruskal–Wallis test. Blue circles represent observed data, solid red line represents unweighted linear regression fit, and gray band represents 95% confidence interval. Units for CL_int are PAR/mg microsomal protein/μM phylloquinone/min.

Discussion

In this study we assessed the relationship between common variation in the CYP4F2 and CYP4F11 genes, protein abundances, and metabolic activity toward VK₁ (phylloquinone), with a specific emphasis on variant haplotypes. To do so, we first characterized variability across CYP4F2 and CYP4F11 gene loci and identified 13 genetic variants (9 CYP4F2, and 4 CYP4F11) with minor allele frequencies greater than 5%. We then performed haplotype mapping analysis across the CYP4F2 and CYP4F11 locus, which revealed 50 unique CYP4F diplotypes made up of combinations of 20 unique CYP4F haplotypes. We then evaluated correlations between mRNA, protein, and metabolic activity using a genotype- and diplotype-based approach. We hypothesized that stratification of livers by CYP4F haplotypes would yield unique and stronger associations than single allele genotypes. However, our data suggest that CYP4F2*3 genotype alone is sufficient to account for most of the genetic associated variability observed in VK₁ metabolism. The ω-hydroxy VK₁ intrinsic formation clearance of diplotypes 17/20 and 20/20 were similarly lower (63% and 68%) than that of the reference diplotype (13/13); this is a result of a combination of both a decrease in CYP4F2 protein expression and an increase in apparent K_m. Diplotype 17/20 and 20/20 are both CYP4F2*3/*3 containing groups, with diplotype 17/20 having a single copy of 8 additional CYP4F2/4F11 genetic variants on the *3 background (rs3093160, rs3093114, rs3093106 (*2), rs3093105, rs3093103, rs3093100, rs1060463, and rs3765070), and diplotype 20/20 having two copies of the additional 8 variants on the *3 background.

The abundance and activity of CYP4F2 protein containing both W12G (*2) and V433 M (*3) amino acid substitutions was not different from protein with only V433M, implying that M433 causes the change in enzyme abundance and function, which is in agreement with experimental results with recombinantly expressed reference and variant proteins.¹⁷ Hepatic CYP4F2 and CYP4F11 mRNA abundances were found to be poor predictors of metabolic activity. Additionally, contrary to a previous report¹⁹ we found no significant association between rs2108622 (CYP4F2*3) and hepatic CYP4F2 or CYP4F11 mRNA abundance. However, in our analysis, we did identify that donors ≤18 years of age had lower CYP4F2 and CYP4F11 mRNA abundance compared to adults. While the underlying mechanism of this observation in our system is unclear, there is evidence of postnatal hepatic CYP4F maturation in mouse models demonstrating a similar effect.²⁹

Interestingly, our study did reveal a unique haplotype-phenotype relationship. Of the 20 unique haplotypes found in the sample population, 6 exist commonly at a frequency of >5% (haplotypes 1, 6, 12, 13, 17, and 20). A total of 5 out of 6 of these haplotypes fall under previously established CYP4F2 star allele nomenclature, as noted by the Pharmacogene Variation Consortium (PharmVar) under the following definitions: Haplotypes 1 and 13 = CYP4F2*1, Haplotype 12 = CYP4F2*6, Haplotype 17 = CYP4F2*3, and Haplotype 20 = CYP4F2*4. However, a novel CYP4F11 variant combination (Haplotype 6) was identified and shown to exhibit a differential protein abundance and activity compared to the reference group. The intrinsic clearance of diplotype 6/13 was 175% higher than that of the population reference diplotype (13/13). CYP4F2 protein abundance in diplotype 6/13 was also 45% higher than that of the reference (18.07 ± 4.86 vs 12.44 ± 5.24 pmol/mg microsomal protein, respectively) and 28% higher for CYP4F11 than the reference diplotype (3.66 ± 0.87 vs 2.87 ± 1.59 pmol/mg microsomal protein, respectively). However, these differences were not observed for the homozygous diplotype 6/6 group. Individuals in the diplotype 6/13 group are carriers of 2 CYP4F11 missense mutations and 3 synonymous mutations (rs1060463, rs8104361, rs3765070, rs2305801, and rs2074900, respectively). How these changes might affect CYP4F2 abundance is unclear but possibly by some cis-acting process.

Curiously, there was lack of a CYP4F2*3 allele effect on metabolic activity for heterozygous carriers (*1/*3) compared to homozygous reference allele carriers (*1/*1), despite the presence of a clear allele dependent decrease in CYP4F2 protein abundance. This can be explained in part by the variability in CYP4F11 protein abundance (Figure 10D) in the *1/*3 group. Moreover, when accounting for combined CYP4F2 and CYP4F11 protein content, the allele dependent decrease in protein abundance between CYP4F2*3 heterozygous carriers (*1/*3) and homozygous reference (*1/*1) carriers is smaller and no longer statistically significant (Figure 11). Additionally, when intrinsic clearance was adjusted for CYP4F protein content (Supplemental Figure 14), the data indicates that there were no significant changes in apparent catalytic turnover at the genotype level for the CYP4F2*3 variant nor haplotype 6 and that genetic variability in CYP4F2 at the genotype and diplotype level was associated with variable protein abundance, which in turn is a strong predictor of metabolic activity. This, taken together with results from a previous study³⁰ which identified recombinant CYP4F11 as having similar catalytic efficiency and K_m as CYP4F2 for VK₂ (MK₄) ω-hydroxylation, provides evidence that the lack of an allele dependent CYP4F2*3 effect on the intrinsic clearance in the set of study samples was due to comparable levels of total CYP4F2+CYP4F11 protein content in the CYP4F2*1/*3 and CYP4F2*1/*1 groups (13.4 ± 4.24 vs 15.3 ± 5.31 pmol/mg microsomal protein, respectively; p = 0.083).

A notable observation from this study was the wide range of CYP4F2 and CYP4F11 protein abundances (25-fold combined range) and the corresponding aggregate metabolic activity (80-fold range). Although some of this variability can be explained by CYP4F2*3 genotype (as described in this study), a considerable amount of within genotype variability can be observed for the wild-type (*1/*1) group that is not explained by underlying CYP4F haplotype. We did observe a significant contribution of cytochrome P450 reductase (CPR) protein abundance (Supplemental Figure 13B) to the variability in metabolic activity toward VK₁, but the absolute contribution to the observed variability was small. This suggests that nongenetic (environmental) factors may be involved and contributing to the observed phenotypic behavior, particularly for individuals who are carriers of fully functional alleles.

A limitation of this study was the lack of a commercially available ω-hydroxy VK₁ standard for metabolite quantitation. However, we expect proportionality between measured PAR values and absolute levels of ω-hydroxy VK₁, based on results of control experiments. Additionally, sample size limitations for diplotype groups restricted our ability to probe the effect size of the CYP4F2*3 variant alone (diplotype 17/17) or CYP4F2*2 variant alone (diplotype 12/12). It is possible that variants in LD with the CYP4F2*3 allele seen in haplotype 20 have an undetected effect on catalytic activity that would only emerge by comparison with the diplotype 17/17 group, but if true, it would likely be of a small magnitude.

In conclusion, we effectively evaluated the underlying mechanistic links between genetic variability in VK metabolic pathway genes (CYP4F2 and CYP4F11) and metabolic activity using a diplotype-based approach. Our study highlights the challenge of determining the effect size of causal variants given complex genomics traits such as linkage disequilibrium and demonstrates the practical application of PopMM modeling as an approach for characterizing genomic traits in vitro. Results suggest that testing for the CYP4F2*3 allele alone is sufficient for in vivo predictions such as the warfarin dose needed to achieve a therapeutic INR. Results also indicate that both CYP4F2 and CYP4F11 enzymes contribute importantly to the metabolism of VK₁ in the human liver. Quantitation of these proteins in blood, using liquid biopsies,³¹ might enhance the predictability of the warfarin dose response relationship and impact on Precision Medicine.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsptsci.3c00287.

• UPLC-MS/MS chromatograms, distribution of hepatic mRNA expression, localization of common CYP4F2 and CYP4F11 gene variants on chromosome 19, hepatic CYP4F11 mRNA expression stratified by variants found at the CYP4F2 and CYP4F11 loci, human liver bank (n = 274) hepatic CYP4F2 mRNA abundance stratified by sex, fractional contribution of CYP4F2 and CYP4F11 towards total protein abundance, CYP4F2 and CYP4F11 protein abundance by age category, NADPH-dependent formation of ω-hydroxy metabolite of vitamin K₁, goodness-of-fit plots for the final PopMM model, random-effects correlation plots of base PopMM model for natural log estimate of K_m and of V_max, multiple linear regression estimates, human liver bank mRNA analysis population demographic summary, summary of CYP4F2 and CYP4F11 protein abundances, Genotype-specific CLint estimates adjusted for total CYP4F2 and CYP4F11 protein abundance, haplotype 6 containing diplotype CL_int estimates (PDF)

Author Contributions

Participated in research design: A.N.A., K.G.C., A.E.R., B.P., M.G.M., K.E.T. Performed data analysis: A.N.A., K.G.C. Wrote or contributed to the writing of the manuscript: A.N.A., A.E.R., K.E.T.

This work was supported in part by funding from the National Institutes of Health, P01 GM116691.

The authors declare no competing financial interest.