Change in plasma α-tocopherol associations with attenuated pulmonary function decline and with CYP4F2 missense variation

ABSTRACT

Background

Vitamin E (vitE) is hypothesized to attenuate age-related decline in pulmonary function.

Objectives

We investigated the association between change in plasma vitE (∆vitE) and pulmonary function decline [forced expiratory volume in the first second (FEV₁)] and examined genetic and nongenetic factors associated with ∆vitE.

Methods

We studied 1144 men randomly assigned to vitE in SELECT (Selenium and Vitamin E Cancer Prevention Trial). ∆vitE was the difference between baseline and year 3 vitE concentrations measured with GC-MS. FEV₁ was measured longitudinally by spirometry. We genotyped 555 men (vitE-only arm) using the Illumina Expanded Multi-Ethnic Genotyping Array (MEGA^ex). We used mixed-effects linear regression modeling to examine the ∆vitE–FEV₁ association.

Results

Higher ∆vitE was associated with lower baseline α-tocopherol (α-TOH), higher baseline γ-tocopherol, higher baseline free cholesterol, European ancestry (as opposed to African) (all P < 0.05), and the minor allele of a missense variant in cytochrome P450 family 4 subfamily F member 2 (CYP4F2) (rs2108622-T; 2.4 µmol/L higher ∆vitE, SE: 0.8 µmol/L; P = 0.0032). Higher ∆vitE was associated with attenuated FEV₁ decline, with stronger effects in adherent participants (≥80% of supplements consumed): a statistically significant ∆vitE × time interaction (P = 0.014) indicated that a 1-unit increase in ∆vitE was associated with a 2.2-mL/y attenuation in FEV₁ decline (SE: 0.9 mL/y). The effect size for 1 SD higher ∆vitE (+4 µmol/mmol free-cholesterol-adjusted α-TOH) was roughly one-quarter of the effect of 1 y of aging, but in the opposite direction. The ∆vitE–FEV₁ association was similar in never smokers (2.4-mL/y attenuated FEV₁ decline, SE: 1.0 mL/y; P = 0.017, n = 364), and current smokers (2.8-mL/y, SE: 1.6 mL/y; P = 0.079, n = 214), but there was little to no effect in former smokers (−0.64-mL/y, SE: 0.9 mL/y; P = 0.45, n = 564).

Conclusions

Greater response to vitE supplementation was associated with attenuated FEV₁ decline. The response to supplementation differed by rs2108622 such that individuals with the C allele, compared with the T allele, may need a higher dietary intake to reach the same plasma vitE concentration.

Introduction

Vitamin E (vitE) is a lipid-soluble antioxidant nutrient that scavenges free radicals that can otherwise damage human tissues (1, 2). Among the 8 forms of vitE, α-tocopherol (α-TOH) is the most abundant form in blood circulation owing to preferential postabsorption retention, whereas γ-tocopherol (γ-TOH) is the major form in the US diet (e.g., rich in vegetable oil and nuts).

There is evidence from observational studies that genetic and nongenetic differences may contribute to interindividual variation in plasma vitE concentrations (3–10). For example, genes related to vitE metabolism are highly polymorphic (3). Genome-wide association studies (GWASs) identified several genetic variants associated with cross-sectional plasma vitE (11–13), and 1 study in male heavy smokers reported 2 genome-wide significant variants associated with postsupplementation plasma vitE (14). Nongenetic factors, such as age and blood lipid concentrations, also contribute to interindividual differences in plasma vitE (4–10). Thus, the efficacy of vitE supplementation likely differs across individuals. To our knowledge, the study presented herein is the first to investigate the influence of both genetic and nongenetic factors on change in plasma vitE with supplementation (i.e., plasma “response”).

For >2 decades, vitE has been posited to have beneficial effects on the lungs given its antioxidant potential. Pulmonary function follows a normal trajectory of decline with aging, but other factors, primarily cigarette smoking, can accelerate this decline (15). Pulmonary function tests (PFTs), such as spirometry, are noninvasive tests that provide insight into a person's lung health by measuring air flow and volume. On a population level, cross-sectional positive associations of dietary intake of vitE and vitE biomarkers with pulmonary function were reported in some (16–28), but not all (19, 24, 29–36) studies. Only 2 studies examined the effect of vitE on longitudinal pulmonary function; 1 reported a positive association of serum vitE (37) and the other reported little to no association of dietary vitE (17). To the best of our knowledge, the RAS (Respiratory Ancillary Study) to SELECT (Selenium and Vitamin E Cancer Prevention Trial), with a generally healthy male population, is the only randomized controlled trial (RCT) to robustly evaluate the effect of vitE supplementation on longitudinal change in pulmonary function (38, 39). In the intention-to-treat analysis in the RAS, neither vitE nor selenium had a statistically significant association with pulmonary function decline (38). Given previous reports of variation in plasma vitE concentrations in response to the same dose of vitE supplementation (14), detecting the true association of vitE with pulmonary function may require accounting for factors that affect the plasma vitE concentration.

In this study, we investigated the association of change in plasma vitE concentration (hereafter, ∆vitE), as a biomarker reflecting the response to vitE supplementation, with age-related decline in pulmonary function, using the longitudinal data collected in the RAS. We hypothesized that a greater response to vitE supplementation (i.e., higher ∆vitE) would lead to attenuated pulmonary function decline. Furthermore, given the prior evidence from GWASs (14) and epidemiologic studies on vitE nutriture (4–10), we examined both genetic and nongenetic factors contributing to interindividual differences in plasma response to vitE supplementation to aid in identifying subpopulations who may benefit the most from vitE supplementation.

Methods

Study design and participants

This study capitalizes on data from the RAS to SELECT, which was a phase 3 double-blinded RCT of vitE and/or selenium supplementation for prostate cancer prevention. Details of the RAS and SELECT are published elsewhere (38, 40). Briefly, SELECT recruited 35,533 men aged ≥50 y for African Americans or >55 y for all other men in the United States, Canada, and Puerto Rico. Eligibility criteria for SELECT included no prior prostate cancer diagnosis or suspicion of cancer, a serum prostate-specific antigen concentration ≤4 ng/mL, no more than 175 mg/d acetylsalicylic acid or 81 mg/d of acetylsalicylic acid if combined with clopidogrel bisulfate for anticoagulant therapy, a normal blood pressure, and no history of hemorrhagic stroke (40). The safety and efficacy of the supplementation for prostate cancer prevention were evaluated in SELECT (39).

The RAS included 2921 men from 16 SELECT sites and preserved the randomized design. The 4 study arms were vitE (400 IU/d of synthetic all rac-α-tocopheryl acetate) + selenium (Se) (200 μg/d of l-selenomethionine) (vitE + Se arm); vitE + Se placebo (vitE arm); vitE placebo + Se (Se arm); and vitE placebo + Se placebo (double placebo arm). The primary endpoint of the RAS was the annual decline in forced expiratory volume in the first second (FEV₁), a measurement obtained by spirometry that reflects the volume of air exhaled from the lungs in the first second of a forced expiratory maneuver. The intention-to-treat analysis of the annual decline in FEV₁ found no statistically significant differences by intervention arm (38). The present study investigates the biological vitE dose (i.e., change in plasma vitE concentration with supplementation) as the exposure. The Cornell University Institutional Review Board (IRB) and the IRB at each of the 16 RAS study sites approved the RAS. This study analyzed data on 1144 participants who self-reported as European or African American in the 2 vitE arms (vitE or vitE + Se) with baseline (preintervention) and year 3 (on intervention) plasma vitE measures and ≥1 pulmonary function measurement (Figure 1).

FIGURE 1.

Study sample size flowchart. PFT, pulmonary function test; RAS, Respiratory Ancillary Study; SELECT, Selenium and Vitamin E Cancer Prevention Trial; vitE, vitamin E.

Pulmonary function

PFTs were performed using the EasyOne handheld spirometer (ndd Medical Technologies) , which was previously tested for validity and reliability (41). Only PFTs meeting the standardization of spirometry criteria of the American Thoracic Society (42) were included; repeated PFTs had to be ≥2 y apart (from first to last PFT) to be included.

VitE

Plasma vitE was assayed for men in the 2 vitE arms and a random sample of men in the double placebo arm. Blood samples were collected at SELECT baseline (before the start of supplements) and at the year 3 annual visit. These blood samples were then stored at −80°C for a maximum of 9 y (see Supplemental Methods). Plasma α-TOH, γ-TOH, and free unesterified cholesterol were assayed via GC-MS on a Hewlett Packard 6890 gas chromatograph coupled with a Hewlett Packard 6890 mass spectrometer. Free cholesterol was measured as a proxy for total cholesterol given that the ratio of free to total cholesterol is relatively constant (43). Hereafter, free cholesterol–adjusted TOHs refer to the TOH concentrations adjusted for free unesterified cholesterol (i.e., TOH/cholesterol). Both raw TOH and free cholesterol–adjusted TOH were used in this study because as a lipid-soluble vitamin, plasma vitE concentration varies by blood lipid concentrations (44).

Genotyping

Genome-wide genotypes were assayed using the Illumina Infinium^® Expanded Multi-Ethnic Genotyping Array (MEGA^ex) (Illumina Inc.) for individuals in the vitE-only arm. Quality control (QC) was applied on the genotyped individuals and the single nucleotide polymorphisms (SNPs). At the individual level, the QC filters included 1) missing call rate > 3%, 2) duplicate sample (identity-by-state > 0.9), 3) first-degree relative (identity-by-descent > 0.4), and 4) excessive homozygosity. At the SNP level, the QC filters included 1) missing rate > 3%, 2) Hardy–Weinberg equilibrium P value < 1 × 10⁻⁴, 3) duplicate SNPs, and 4) minor allele frequency (MAF) ≤0.03. Imputation was performed based on the Haplotype Reference Consortium panel (version r1.1, 2016) (45). For the genetic association analysis, the final sample size was 555 [417 European ancestry (EA) participants and 138 African ancestry (AA) participants] with all covariate data available. In the genetic analysis, EA and AA refer to genetically confirmed ancestry, which was measured in participants in the vitE-only arm; in the nongenetic analyses, EA and AA refer to self-reported race, which was measured in all RAS participants.

Statistical analysis of nongenetic factors and ∆vitE

We completed a systematic search of the literature for reports of nongenetic factors associated with plasma vitE concentration. For each factor identified, a further literature review sought to identify the biological plausibility of the putative association (Supplemental Methods), leading to a final set of factors that were tested in a joint model. The variables tested were race, intervention arm, smoking status, and baseline values for age, BMI, vitE (i.e., α-TOH) concentration, γ-TOH concentration, free cholesterol concentration, plasma Se concentration, and dietary intakes of vitamin C (vitC), fiber, and alcohol. The dietary information was collected by a 120-item FFQ administered at study baseline (46) to assess average food intake during the preceding 12 mo, developed by the Nutrition Assessment Shared Resource of Fred Hutchinson Cancer Research Center. Nutrient calculations were performed using the Nutrient Data System for Research software (version 2010), which was developed by the Nutrition Coordinating Center at University of Minnesota. Smoking status was categorized as never, former (stopped smoking before the study), and current smokers (smoking during the study). Linear regression models in SAS version 9.4 (SAS Institute) tested the association of nongenetic factors with both raw ΔvitE and free cholesterol–adjusted ΔvitE. The total R² denotes the percentage of variability in ∆vitE explained by the full model, and the individual R² denotes the estimated variability explained by each individual factor [race, baseline tocopherol (TOH) concentrations, etc.], which was calculated as the difference between total R² with and without the factor in the model.

Statistical analysis of genetic variants and ∆vitE

We conducted GWAS analyses to identify SNPs associated with ∆vitE. We tested SNP associations under an additive model using imputed genotype dosage values (a fractional value between 0 and 2 indicating the expected number of alternative allele copies) to account for any imputation uncertainty (47). GWAS analyses stratified by ancestry (n_EA = 417, n_AA = 138) were performed in rvtests (48) and then meta-analyzed using METAL (49) for each of 4 vitE phenotypes (raw ΔvitE, free cholesterol–adjusted ΔvitE, raw Δγ-TOH, and free cholesterol–adjusted Δγ-TOH) in models adjusted for age, baseline TOH concentrations, and principal components to account for population substructure. We used the GWAS results to conduct in silico look-up for 2 SNPs that were discovered in a previous GWAS of plasma vitE in heavy smokers (P < 5 × 10⁻⁷, accounting for 549,989 tested SNPs in the previous study) (14); we applied a Bonferroni-corrected P value threshold of 0.025 (α = 0.05/2 SNPs). The SNP that we replicated was further examined in models that included additional adjustment for the set of previously reported nongenetic covariates (Supplemental Methods), which were not included in our parsimonious GWAS model that was designed to optimize statistical power.

Statistical analysis of plasma vitE–pulmonary function association

We examined the association between the biological dose of vitE (i.e., change in plasma vitE concentration in response to vitE supplementation) and annual rate of change in FEV₁ among men in the 2 vitE arms. This analysis will hereafter be referred to as a dose-response analysis. Unless otherwise specified, these analyses focus on α-TOH, the primary form of vitE. To account for the repeated PFTs, we used a mixed-effects linear regression model where FEV₁ at each time point was the dependent variable and time was an independent variable that reflected the effect of 1 y of aging on change in pulmonary function (i.e., annual rate of change in FEV₁). The association of ∆vitE with annual rate of change in FEV₁ was modeled as an interaction term between ∆vitE and time (see Equation 1). We adjusted for covariates, including PFT-related factors [age, height, self-reported race, smoking status, and cigarettes per day (current smokers only)] and vitE status–related factors [treatment arm (vitE arm, vitE + Se arm), baseline vitE, baseline γ-TOH, and baseline free cholesterol concentrations].

(1)

where i = i^th individual, j = j^th PFT measure.

Sensitivity analyses focused on subgroups within which we hypothesized a stronger ∆vitE–FEV₁ association. First, we limited analyses to participants who were adherent to the intervention, defined as using ≥80% of the vitE supplement pills. The adherence rate in the RAS for vitE supplementation at year 3 was 86%–88%, as previously reported (38). Next, we limited analyses to adherent men who were responders to the vitE supplementation. Given that there was a decrease in plasma vitE concentration over 3 y, on average, in a subset of randomly selected participants in the double placebo arm (n = 91; mean ΔvitE = −3.8 µmol/L), responders to the vitE supplementation in the 2 intervention arms were defined as participants with a plasma ΔvitE value higher than the mean ΔvitE (−3.8 µmol/L) in the double placebo arm. We chose the mean ΔvitE in the double placebo arm as the reference point, instead of 0 (i.e., no change), because the change in plasma vitE over 3 y without a supplement is best estimated by the ΔvitE value in the double placebo group (Figure 2).

FIGURE 2.

Distribution of the change in plasma α-TOH (µmol/L) from baseline to year 3 in the Respiratory Ancillary Study. The black vertical line denotes the mean change in plasma α-TOH (−3.8 µmol/L) in the placebo arm. VitE groups include participants in the vitE supplementation–only arm and vitE + Selenium supplementation arm. Responders in the vitE groups are defined as having a change in plasma α-TOH > −3.8 µmol/L. Mean ± SD change in plasma α-TOH is shown in each panel as well as the proportions of participants who had a change > −3.8 µmol/L. VitE, vitamin E; α-TOH, α-tocopherol.

For all analyses, we set a 2-sided P ≤ 0.05 as the threshold for statistically significant findings. Nonlinearity was examined by including a square term for ΔvitE. The influence of 1 outlier for ∆vitE was examined by comparing the results with and without the outlier. Possible effect modification of the ∆vitE–FEV₁ association by smoking status, by treatment arm, and by race was examined in models that included a 3-way interaction term (effect modifier × ΔvitE × time). Statistical assumptions were examined in these models for linearity, normality of the residuals, homoscedasticity, multicollinearity, and independence of residual error terms.

Results

Participant characteristics

A total of 1144 participants randomly assigned to vitE were included in the model of annual rate of change in FEV₁: n_EA = 874 (76.4%) and n_AA = 270 (23.6%). The mean age of included participants was 63 y, and the majority of participants had a history of smoking (49% former smokers, 19% current smokers) (Table 1) and were overweight (48%) or obese (34%). Six percent of participants were at risk of vitE deficiency at baseline using the clinical threshold of 12 µmol/L (50); this dropped to 2% in the vitE arms after 3 y of supplementation. The 2 vitE intervention arms were similar in terms of baseline characteristics and change in plasma TOH concentrations (Table 1, Supplemental Table 1).

TABLE 1.

Characteristics of male participants with vitE supplementation in the Respiratory Ancillary Study¹

	Overall (n = 1144)	VitE + Selenium (n = 565)	VitE + Placebo (n = 579)
Age,2 y	62.6 ± 6.4	62.7 ± 6.3	62.5 ± 6.4
Race,2n (%)
EA	874 (76)	435 (77)	439 (76)
AA	270 (24)	130 (23)	140 (24)
Smoking status,3n (%)
Never	364 (32)	181 (32)	183 (32)
Former	564 (49)	277 (49)	287 (50)
Current	216 (19)	107 (19)	109 (19)
BMI,2 kg/m2	28.9 ± 4.6	28.4 ± 4.6	29.3 ± 4.6
Height,2 cm	176.6 ± 7.1	176.6 ± 7.0	176.7 ± 7.1
FEV1,2 mL	2986.8 ± 680.8	2990.4 ± 669.1	2983.2 ± 692.5

¹ n = 1144. Values are mean ± SD, unless otherwise indicated. AA, African ancestry; EA, European ancestry; FEV₁, forced expiratory volume in the first second; vitE, vitamin E.

²Measured at baseline.

³Smoking status refers to the status throughout the study such that former smoker is defined as someone who had quit smoking before enrolling in the study, whereas current smoker is defined as someone who reported smoking during the study.

On average, participants in the vitE arms showed an increase in plasma vitE after 3 y of supplementation; the mean ± SD ∆vitE was 8.2 ± 12.5 µmol/L in the vitE arm and 7.4 ± 14.5 µmol/L in the vitE + Se arm (Supplemental Table 1). In contrast, plasma vitE among participants in the double placebo arm decreased on average [mean ± SD ΔvitE: −3.8 ± 8.5 µmol/L, P = 4.95 × 10⁻⁵ (significantly different from the null: ΔvitE = 0)] (Figure 2). Participants in the double placebo arm who reported use of prestudy vitE supplementation (i.e., >50 IU/d vitE) (16%) had a greater decrease in their plasma vitE concentration over 3 y (mean ΔvitE: −10.7 µmol/L compared with −2.4 µmol/L for participants not reporting prestudy vitE supplementation). For participants in the vitE arms, the increase in plasma vitE concentration was lower in participants who took prestudy vitE supplements (21%) (mean ΔvitE: +2.7 µmol/L compared with +9.2 µmol/L for those who did not take prestudy vitE supplementation).

There was a notable change in plasma γ-TOH in participants in the vitE arms; plasma γ-TOH decreased by a mean of 1.7 µmol/L in both vitE arms (SD = 2.4 and 2.6 µmol/L in the vitE arm and the vitE + Se arm, respectively), but there was little to no change in the double placebo arm [mean ± SD Δ: +0.5 ± 2.9 µmol/L, P = 0.11 (not significantly different from the null hypothesis: Δγ-TOH = 0 in the double placebo arm)] (Supplemental Table 1). Whereas there was little to no correlation between baseline concentrations of vitE and γ-TOH in participants in the 2 vitE arms (Pearson r = 0.018, P = 0.53), there was an inverse correlation between ΔvitE and Δγ-TOH (Pearson r = −0.22, P < 0.0001).

The mean ± SD increase in ΔvitE after 3 y of supplementation was significantly higher in EA than in AA men in the vitE-only arm (P_{EA vs. AA} = 0.010; 9.0 ± 12.5 µmol/L in EA compared with 5.9 ± 12.1 µmol/L in AA), with difference in a similar direction but of lower magnitude in the vitE + Se arm (P_{EA vs. AA} = 0.60; 7.5 ± 15.2 µmol/L in EA compared with 6.9 ± 11.9 µmol/L in AA). The higher increase in plasma vitE concentration in EA men was not due to differences by race in prestudy vitE supplement use (i.e., a higher increase due to less prestudy vitE use). Indeed, the proportion of EA men reporting prestudy vitE supplement use was higher (23%) than for AA men (13%).

Nongenetic factors and plasma ΔvitE

Plasma ΔvitE values ranged from −55 µmol/L to +109 µmol/L in the 2 vitE arms (Figure 2) (mean ± SD: 7.8 ± 13.5 µmol/L, with no difference between arms; P = 0.28); Supplemental Table 1 and Supplemental Table 2 provide further descriptive data on plasma ΔvitE by treatment arm, race, age, and smoking status. Higher vitE at study baseline was associated with a lower increase in ΔvitE (P < 0.0001), whereas higher free cholesterol and higher γ-TOH at baseline were associated with a higher increase in ΔvitE (P < 0.0001) (Table 2). EA men had a higher increase in ΔvitE than AA men (P < 0.0001), and generally this pattern persisted regardless of smoking status (Supplemental Table 3). We found little to no association of ΔvitE with age, BMI, smoking status, and other nutrition factors, including baseline plasma Se concentration or dietary intake of vitC, fiber, or alcohol (P > 0.05), in this study. Overall, we saw consistent findings between models examining raw and free cholesterol–adjusted ΔvitE (Table 2). The fully adjusted model explained 18.7% of the variability (R²) in ΔvitE in the 2 vitE arms (n = 1144), and baseline concentrations of vitE, γ-TOH, free cholesterol, and race explained 10.9%, 1.7%, 1.2%, and 1.6% of the variability, respectively.

TABLE 2.

Factors associated with plasma change in vitE after 3-y vitE supplementation in the Respiratory Ancillary Study¹

	ΔRaw vitE2		ΔAdjusted vitE3
	β ± SE4	P value5	β ± SE4	P value5
Baseline α-TOH6	−0.64 ± 0.052	<0.0001*	−0.38 ± 0.039	<0.0001*
Baseline γ-TOH6	0.82 ± 0.17	<0.0001*	1.31 ± 0.14	<0.0001*
Cholesterol, mmol/L	0.0024 ± 0.00058	<0.0001*
Race
EA	4.43 ± 0.93	<0.0001*	1.71 ± 0.30	<0.0001*
AA	Ref.	Ref.	Ref.	Ref.
Treatment arm
VitE + Selenium	−0.30 ± 0.73	0.68	0.30 ± 0.24	0.21
VitE + Placebo	Ref.	Ref.	Ref.	Ref.
Age, y	−0.078 ± 0.059	0.19	0.0082 ± 0.019	0.67
BMI, kg/m2	0.044 ± 0.080	0.58	0.013 ± 0.026	0.63
Smoking status7
Current smokers	−1.16 ± 1.09	0.29	−0.45 ± 0.35	0.21
Former smokers	−0.11 ± 0.83	0.89	−0.039 ± 0.27	0.88
Never smokers	Ref.	Ref.	Ref.	Ref.
Baseline selenium concentration, parts per billion	−0.0024 ± 0.0056	0.67	−0.00064 ± 0.0018	0.72
Total vitamin C intake, mg	−0.00041 ± 0.0014	0.78	0.00025 ± 0.00046	0.60
Dietary fiber intake, g	0.043 ± 0.035	0.22	0.0052 ± 0.011	0.65
Alcohol intake, g	−0.0016 ± 0.018	0.93	−0.0049 ± 0.0060	0.41

¹ n = 1135. All factors shown were modeled in the same multiple linear regression model. Participants in the vitE + selenium arm and in the vitE + placebo arm were included. Nine participants had missing dietary vitamin C, fiber, and alcohol intake and thus the sample size for this analysis was 1135. AA, African ancestry; EA, European ancestry; vitE, vitamin E (i.e., in the form of α-tocopherol); α-TOH, α-tocopherol; γ-TOH, γ-tocopherol.

²Total variability explained, R² = 18.7%.

³ΔAdjusted vitE refers to the change in free cholesterol–adjusted α-TOH concentration from baseline to year 3. Total variability explained, R² = 19.5%.

⁴β is the coefficient of the association of each factor with the change in α-TOH phenotype.

⁵For each categorical variable, the P value of each categorical level (e.g., current smokers or former smokers) indicates whether the coefficient significantly differed from that of the reference group (e.g., never smokers). For each continuous variable, the P value indicates whether the coefficient significantly differs from 0. Significant p-value < 0.05 is denoted with *.

⁶The unit for baseline α-TOH and γ-TOH concentrations is µmol/L when the outcome is the change in plasma raw α-TOH concentration, and the unit is µmol/mmol free cholesterol when the outcome is the change in plasma free cholesterol–adjusted α-TOH concentration.

⁷The P values of smoking status on ΔRaw vitE and ΔAdjusted vitE are 0.53 and 0.40, respectively, based on Type III tests, which examined the significance of the variable of interest (e.g., smoking status) when all other variables were in the model.

Genetic factors and plasma ΔvitE

GWAS analyses revealed no genome-wide significant variants (P < 5 × 10⁻⁸) for change in TOH concentrations (either α- or γ-TOH) in the vitE-only arm (n = 555, results not shown). We used the GWAS results to study 2 SNPs (rs964184 and rs2108622) that were reported in a prior GWAS of circulating vitE measured after 3 y of vitE supplementation in male current smokers, which was not previously tested for independent replication (14). As shown in Table 3, we replicated 1 SNP, rs2108622, because it was associated with ΔvitE after 3 y of vitE supplementation in the RAS (a healthy male population), with a P value passing the Bonferroni-corrected threshold for replication testing (P < 0.025). Rs2108622, a missense SNP in the cytochrome P450 family 4 subfamily F member 2 (CYP4F2) gene on chromosome 19, was directly genotyped on the MEGA^ex array, so its dosages were 0 (major allele homozygote), 1 (heterozygote), and 2 (minor allele homozygote). When the genotype was modeled as a categorical variable, heterozygotes for the rs2108622 minor allele (CT genotype) had a +1.75 μmol/L ΔvitE (SE: 1.06 μmol/L; P = 0.10) compared with homozygotes for the major allele (CC genotype), whereas homozygotes for the minor allele (TT genotype) had a +5.66 μmol/L ΔvitE (SE: 1.93 μmol/L; P = 0.0034) compared with the CC genotype.

TABLE 3.

The association of rs2108622-T with plasma change in α-TOH for male participants randomly assigned to the vitE-only arm (400 IU/d all rac-α-tocopheryl acetate) in the Respiratory Ancillary Study¹

	ΔRaw vitE		ΔAdjusted vitE2
	β ± SE3	P value4	β ± SE3	P value4
Unadjusted model
Overall (n = 555)	2.46 ± 0.83	0.0034*	0.48 ± 0.28	0.080
EA (n = 417)	2.27 ± 0.93	0.015*	0.34 ± 0.30	0.25
AA (n = 138)	1.02 ± 2.25	0.65	0.30 ± 0.80	0.70
Adjusted model5
Overall (n = 548)6	2.36 ± 0.80	0.0032*	0.25 ± 0.26	0.33
EA (n = 413)	2.42 ± 0.87	0.0058*	0.23 ± 0.28	0.40
AA (n = 135)	1.42 ± 2.05	0.49	0.43 ± 0.74	0.56

¹ n = 555. AA, African ancestry; EA, European ancestry; vitE, vitamin E (i.e., in the form of α-tocopherol); α-TOH, α-tocopherol.

²ΔAdjusted vitE refers to the change in free cholesterol–adjusted α-TOH concentration from baseline to year 3.

³β is the coefficient of the association of the rs2108622-T allele with change in α-TOH phenotype in the multiple linear regression model. No interaction with ancestry was found (P ≥ 0.61 for raw or adjusted ΔvitE), thus the overall findings with EAs and AAs combined are presented.

⁴ P value of the association coefficient. Significant p-value < 0.025 (Bonferroni correction for testing 2 SNPs) is denoted with *.

⁵Adjusted for age at baseline, BMI at baseline, ancestry, smoking status, plasma selenium concentration at baseline, total vitamin C intake at baseline, dietary fiber intake at baseline, alcohol intake at baseline, baseline α-TOH, baseline γ-TOH, and baseline free cholesterol (only in the model with raw α-TOH). Seven participants had missing dietary vitamin C, fiber, and alcohol intake, thus the sample size for the adjusted model was 548.

⁶Total variability explained, R² = 18.3% for raw change in vitE (19.2% for change in free cholesterol–adjusted vitE); baseline concentrations of vitE, γ-TOH, free cholesterol, ancestry, and rs2108622-T explained 9.2%, 1.6%, 1.3%, 1.7%, and 1.3% of the variability, respectively.

The mean ΔvitE for participants with 0, 1, and 2 minor alleles of rs2108622 was 7.6 µmol/L, 9.5 µmol/L, and 12.7 µmol/L in EAs, respectively, and 5.8 µmol/L, 6.0 µmol/L, and 12.9 µmol/L in AAs, respectively. The ancestry-specific analyses revealed a consistent direction of association for rs2108622 with ΔvitE, and no interaction with ancestry was found (P ≥ 0.61 for raw or adjusted ΔvitE). Thus, findings with EAs and AAs combined are presented herein. In the fully adjusted model where the number of SNP alleles was modeled as a continuous variable, each additional copy of the rs2108622-T minor allele (MAF = 31% in EAs and 13% in AAs) was associated with a 2.36-µmol/L higher increase in ΔvitE (P = 0.0032) (Table 3), than for the C major allele. This effect magnitude is ∼20% of 1 SD variation of ΔvitE in the vitE-only arm (SD = 12.5 µmol/L) (Supplemental Table 1). No substantial violation of statistical assumptions (i.e., linearity, normality, homoscedasticity, no multicollinearity, and independence) was observed in regression models of ΔvitE regressed against genetic and nongenetics factors. In addition, the rs2108622 SNP explained about one-third of the variability (R²) in ΔvitE that was initially explained by ancestry (without rs2108622 in the model, ancestry explained ∼2.5% of ΔvitE variability; with rs2108622 in the model, ancestry explained ∼1.7% of ΔvitE variability). A consistent direction of association was observed in models of free cholesterol–adjusted ΔvitE, but findings were not statistically significant (P = 0.33). There was no evidence for an rs2108622 × smoking interaction (P = 0.65). The fully adjusted model explained 18.3% of the variability in plasma ΔvitE in the vitE arm (n = 555): baseline concentrations of vitE, γ-TOH, free cholesterol, ancestry, and rs2108622-T explained 9.2%, 1.6%, 1.3%, 1.7%, and 1.3% of the variability, respectively.

Plasma ΔvitE and longitudinal pulmonary function decline

Positive change in plasma ΔvitE was associated with attenuated FEV₁ decline in adherent participants who responded to the vitE supplement (ΔvitE > −3.8 µmol/L) (Table 4), and there was little to no association of Δγ-TOH (Supplemental Table 4). In the full sample, an increase of 1 µmol/mmol in free cholesterol–adjusted ΔvitE was associated with a 0.96-mL/y (SE: 0.60 mL/y) attenuation in annual FEV₁ decline (P = 0.11). The effect size increased in magnitude in sensitivity analyses. Among men adherent to the intervention, an increase of 1 µmol/mmol in free cholesterol–adjusted ΔvitE was associated with a +1.36 mL/y (SE: 0.78 mL/y) attenuation in annual FEV₁ decline (P = 0.082). The magnitude of effect was greater among adherent participants who responded to the vitE supplementation (+2.22 mL/y, SE: 0.90 mL/y; P = 0.014). Overall, there was evidence that an increase in free cholesterol–adjusted ΔvitE was associated with attenuation in FEV₁ decline, with no evidence of nonlinearity (P_{non-linearity} = 0.69). All statistical assumptions (i.e., linearity, normality, homoscedasticity, no multicollinearity, and independence) were tested and no violations were observed. In summary, in adherent participants who responded to the intervention, a 1-SD higher ΔvitE (∼4 µmol/mmol free cholesterol) was associated with an attenuation of ∼9 mL/y in FEV₁ decline. Consistent trends were observed for ΔvitE without adjusting for free cholesterol (Table 4).

TABLE 4.

Association of plasma change in vitE after 3-y vitE supplementation with, and its interaction with smoking status on, annual rate of change in FEV₁ in the Respiratory Ancillary Study¹

	ΔRaw vitE			ΔAdjusted vitE2
	β ± SE3	95% CI	P value4	β ± SE3	95% CI	P value4
Main effect model (n = 1142)5
Full sample	0.23 ± 0.20	(−0.17, 0.63)	0.26	0.96 ± 0.60	(−0.23, 2.14)	0.11
Adherent sample6 (n = 775)	0.24 ± 0.25	(−0.26, 0.73)	0.35	1.36 ± 0.78	(−0.17, 2.89)	0.082
Adherent sample who responded to the vitE supplement6,7 (n = 680)	0.67 ± 0.31	(0.054, 1.28)	0.033*	2.22 ± 0.90	(0.45, 3.99)	0.014*
Smoking interaction model (n = 1142)5		P interaction = 0.0138			P interaction = 0.0328
Never smokers (n = 364)	0.83 ± 0.35	(0.15, 1.51)	0.017*	2.43 ± 1.01	(0.44, 4.42)	0.017*
Former smokers (n = 564)	−0.41 ± 0.30	(−0.99, 0.17)	0.17	−0.64 ± 0.86	(−2.32, 1.04)	0.45
Current smokers (n = 214)	0.73 ± 0.48	(−0.20, 1.67)	0.12	2.75 ± 1.57	(−0.32, 5.82)	0.079

¹All mixed-effects linear regression models were adjusted for age at baseline, height, race, smoking status, smoking dose at baseline, treatment arm, baseline α-TOH, baseline γ-tocopherol, and baseline free cholesterol (only in the model with raw α-TOH). vitE, vitamin E (i.e., in the form of α-tocopherol); α-TOH, α-tocopherol.

²ΔAdjusted vitE refers to the change in free cholesterol–adjusted α-TOH concentration from baseline to year 3.

³β is the coefficient of the ΔvitE × time term in the main effect model. For the smoking interaction model, β is the coefficient of the ΔvitE × time term for the reference group when each smoking status is set as the reference.

⁴ P value of the ΔvitE × time term to test the association of ΔvitE with rate of change in FEV₁. For the smoking interaction model that includes all smoking status variables, this is the P value for the ΔvitE × time term for each smoking group (when it is set as the reference) to test whether the coefficient significantly differs from 0. Significant p-value < 0.05 is denoted with *.

⁵Two current smokers had missing data for smoking dose and thus were excluded in the statistical model.

⁶Adherent participants (defined as taking ≥80% of the supplement pills) during the 3 y of supplementation.

⁷Responder is defined as Δraw vitE higher than the mean change in the placebo arm (−3.8 µmol/L) after 3-y supplementation.

⁸The P_interaction value is for the 3-way interaction term (ΔvitE × time × smoking status) to test whether the association of ΔvitE with rate of change in FEV₁ differs by smoking status.

Smoking status modified the association of plasma ΔvitE with FEV₁ decline (P_{smoking interaction} < 0.05) (Table 4). An increase in free cholesterol–adjusted ΔvitE was statistically significantly associated with attenuation in FEV₁ decline in never smokers (P = 0.017) and attenuation of similar magnitude in current smokers (P = 0.079), but, contrary to expectation, little to no association in former smokers (P = 0.45) (Table 4). To put the effect magnitude in context, 1 SD higher free cholesterol–adjusted ΔvitE was associated with 9.7 mL/y and 11.0 mL/y attenuation in FEV₁ decline in never and current smokers, respectively. The same trend was observed for raw ΔvitE (Table 4) and in a sensitivity analysis limited to adherent participants (Supplemental Table 5). Neither race (AA, EA) nor treatment arm (vitE, vitE + Se) modified the association of ΔvitE with FEV₁ decline (P_interaction = 0.85 and 0.43 for race and treatment arm, respectively).

Discussion

The present study reports associations of both genetic and nongenetic factors contributing to individual response to vitE supplementation in an RCT. RCTs provide high-quality evidence about the causal effects of nutrition on chronic disease outcomes, yet many RCTs do not confirm nutrient–disease associations reported in observational studies and/or animal studies (39, 51, 52). The inconsistent findings could be due in part to a failure to consider interindividual variation in response to supplementation (53), which may lead to diminished efficacy of the supplement intervention. This study examined individual variation in the response to vitE supplementation in a healthy population; ΔvitE ranged widely from −10 µmol/L (5^th percentile) to 31 µmol/L (95^th percentile) despite the same dose of vitE supplementation being given. Importantly, we found that the variation in ΔvitE was associated with baseline nutriture status and genetic factors.

In analyses of genetic factors, the minor allele (T) of rs2108622 on CYP4F2 was associated with greater response to vitE supplementation (P = 0.0032). This SNP passed genome-wide significance in a previous GWAS of ∆vitE in a male heavy smoker population (14), but the finding was not tested for independent replication before the study reported herein. Given our population of healthy males, this replication provides important evidence in support of the generalizability of this finding. In our study, rs2108622 explained about one-third of the variability in ΔvitE that was initially explained by ancestry. Rs2108622 also explained about the same proportion of variability in ΔvitE as explained by baseline cholesterol status, which has a well-established link with vitE concentration (54). The enzyme encoded by the CYP4F2 gene is reported to be the only one among the cytochrome P450 enzymes that has ω-hydroxylase activity for vitE catabolism (e.g., for α- and γ-TOH) in the liver (55, 56). The rs2108622-T allele in CYP4F2 results in a methionine, instead of a valine, in the enzyme. Given the CYP4F2 enzyme goes through faster degradation by the proteasome when methionine is present (57), individuals with the T allele would have more vitE in the liver available for α-TTP transport into the circulation, leading to higher plasma vitE. Consistent with this proposed mechanism, we observed that the EA population, with a higher frequency of the T allele (31% compared with 13% in AA), had a higher increase in ∆vitE after supplementation. The association of rs2108622-T with higher ΔvitE was also reported in a candidate gene study from an RCT in patients with nonalcoholic fatty liver disease (58).

In light of the interindividual variation in response to vitE supplementation, we hypothesized that the effect of vitE supplementation on longitudinal pulmonary function would vary across participants. To address this question, we used plasma vitE as a biomarker to quantify the treatment exposure. We hypothesized that the ΔvitE–FEV₁ associations would be stronger using this more direct measure of treatment exposure than the intention-to-treat analysis based on the treatment assignment. The previous intention-to-treat analysis in the RAS (38) found that the mean rate of decline in FEV₁ in the vitE arm was −33 mL/y (95% CI: −40, −26 mL/y) compared with −39 mL/y (95% CI: −46, −32 mL/y) in the double placebo arm, but the association of vitE supplementation with longitudinal pulmonary function did not reach the statistical significance threshold (P = 0.19). In this study, using the change in a vitE biomarker instead of the treatment assignment as the exposure, we showed that ΔvitE was associated with an attenuated decline in FEV₁, particularly in adherent vitE supplement responders. A 1-SD higher ΔvitE attenuated FEV₁ decline in generally healthy males; the effect size was about one-quarter of the magnitude of the effect of a year of aging, but in the opposite direction. Put simply, higher ΔvitE attenuated the age-related decline in pulmonary function. The stronger ΔvitE–FEV₁ association among men who were adherent suggested that we would have observed a greater effect of vitE in the overall study had optimal adherence been achieved. The finding was consistent for both raw ∆vitE and cholesterol-adjusted ∆vitE, although not all results were statistically significant. Overall, our findings show that it is informative to consider interindividual response to nutrition interventions within the RCT design, and that doing so may reveal treatment effects that are otherwise missed.

Whereas this study showed a beneficial effect of vitE in never and current smokers, there was little to no effect in former smokers. Similarly, the only prior longitudinal study of a vitE biomarker and FEV₁ found a protective association limited to heavy smokers (37), but the participants (age 20–44 y) were much younger than in the present study (age ≥ 50 y). The finding in the RAS is consistent with a cross-sectional study of a vitE biomarker and FEV₁ in the Third NHANES, which studied a sample population that is representative of the general US adult population (24). Like our study, the NHANES study found a protective effect of vitE on pulmonary function in current and never smokers, with a much smaller effect in former smokers (24). Another study found a higher self-reported fruit intake (typically higher in vitC) was associated with an attenuation in pulmonary function decline in former smokers only, with negligible associations seen in current and never smokers (59). A possible explanation is that differences in the nutrient–pulmonary function decline associations by smoking groups reflect differences in how nutrients with antioxidant properties are used under different physiologic conditions. Given that vitE is at the frontline against free radicals in the redox reaction, whereas vitC recycles vitE from its oxidized form afterwards (60), it is possible that vitE serves as the frontline antioxidant to scavenge free radicals in lungs when there is acute oxidative stress (e.g., direct cigarette exposure in current smokers, direct exposure to air pollution or second-hand smoke in never smokers). Further mechanistic studies are needed to support epidemiologic inference and/or to allow design of more specific biomarker-based measurement in epidemiologic studies.

Several limitations deserve mention. We had limited power to detect significant associations at the genome-wide threshold (P < 5 × 10⁻⁸) with a total of 555 men in the vitE arm of the RAS. A GWAS with a larger sample size is needed in the future to detect more genetic variants associated with response to vitE supplementation. Nevertheless, we leveraged the genotype data in the RAS to replicate 1 missense SNP (rs2108622) reported in a previous GWAS (14). Second, although our study did not measure all nongenetic factors identified in the prior literature related to plasma vitE, such as waist circumference (WC) and waist-to-hip ratio (WHR) (61), the factors we investigated were representative of unmeasured ones [e.g., r > 0.60 for the correlations of BMI with WC and WHR (62, 63)]. Third, it is possible that there was a synergistic effect between vitE and Se supplementation for individuals in the vitE + Se arm. Our study focused on the effect of vitE supplementation instead of Se, because the magnitudes reported in the original intention-to-treat analysis (38) provided compelling evidence for an association of vitE supplementation with pulmonary function, compared with double placebo (with P_difference = 0.19, and 6 mL/y attenuation in FEV₁ decline in the vitE arm), whereas the difference was far smaller in magnitude for the Se arm compared with double placebo (with P_difference = 0.75 and only 1 mL/y attenuation in decline in the Se arm). Finally, because the RAS participants were overall healthy males ≥ 50 y old in the United States, Canada, and Puerto Rico, the generalizability of the study may be limited, and whether the same findings would apply to females, patients with chronic obstructive pulmonary disease or other diseases, or other populations beyond EA and AA needs further investigation.

Our study examined the effect of changes in plasma vitE in response to supplementation on longitudinal pulmonary function in healthy males; this study adds new information to the current epidemiologic literature on vitE and pulmonary health. Most prior studies were cross-sectional, and to our knowledge no longitudinal study has yet examined the effect of change in plasma vitE in response to vitE supplementation on pulmonary function. Further research is needed in more diverse study populations.

To summarize, this study provides convincing evidence that interindividual differences in genetic factors and baseline nutriture contribute to the wide variation in plasma response to vitE supplementation. Individuals with the rs2108622-T minor allele may need less vitE supplementation (i.e., lower dosage) than individuals without this allele in order to achieve the same magnitude of increase in plasma vitE and reap its health benefits. Further, our study found a protective role of vitE on longitudinal pulmonary function such that an increase in ∆vitE was associated with an attenuation in FEV₁ decline, with evidence of smoking stratification. Further studies are needed to investigate the effect modification by smoking status identified.

Notes

Supported by National Heart, Lung, and Blood Institute (NHLBI) grants R01 HL071022 (to PAC) and R21 HL125574 and R01 HL149352 (to PAC and DBH) and National Cancer Institute/Division of Cancer Prevention grants U10 CA37429 and 5UM1CA182883. Genotyping services were provided through the DNA Resequencing and Genotyping Service by the Northwest Genomics Center at the University of Washington, Department of Genome Sciences, under US Federal Government contract number HHSN268201100037C from the NHLBI. This research was also supported, in part, by the Cornell University Center for Vertebrate Genomics Scholarship Award (to JX) and the American Society for Nutrition Pfizer Predoctoral Fellowship (to JX). The study sponsors were not involved in study design, data collection, data analysis, data interpretation, report writing, or decisions to submit the paper for publication.

Supplemental Methods and Supplemental Tables 1–5 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.

RSP and ARK are deceased.

Abbreviations used: AA, African ancestry; CYP4F2, cytochrome P450 family 4 subfamily F member 2; EA, European ancestry; FEV₁, forced expiratory volume in the first second; GWAS, genome-wide association study; IRB, Institutional Review Board; MAF, minor allele frequency; MEGA^ex, Expanded Multi-Ethnic Genotyping Array; PFT, pulmonary function test; QC, quality control; RAS, Respiratory Ancillary Study; RCT, randomized controlled trial; SELECT, Selenium and Vitamin E Cancer Prevention Trial; SNP, single nucleotide polymorphism; TOH, tocopherol; vitC, vitamin C; vitE, vitamin E; WC, waist circumference; WHR, waist-to-hip ratio; α-TOH, α-tocopherol; γ-TOH, γ-tocopherol; ∆vitE, change in plasma vitamin E.

Data Availability

Data described in the article, code book, and analytic code will be made available upon request pending application and approval.