Dietary Pattern, Sputum DNA Methylation, and Lung Health

Abstract

Background

We previously identified a sputum 12-gene methylation panel that predicts lung aging and risk for lung cancer.

Research Question

Can the sputum methylation panel be used as a readout to derive a dietary pattern beneficial for lung health? Is this dietary pattern associated with various subjective and objective lung health phenotypes? Does this relationship vary among people who currently smoke vs previously smoked?

Study Design and Methods

Using the Lovelace Smoker Cohort (LSC), we employed the least absolute shrinkage and selection operator regularized Poisson regression to define a dietary pattern for sputum. Associations of the dietary pattern with objective and subjective lung health measurements were examined using generalized linear and Cox models in the LSC and the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening trial.

Results

The Dietary Pattern for Healthy Lung (DiPHeaL) includes low consumption of processed meat, and high consumption of dark green vegetables, tea, alcohol, and fruit juice. In the LSC, a higher DiPHeaL score (1 SD) was associated with better FEV₁ (by 96.1 mL/s), FEV₁/FVC ratio (by 1.83%), and respiratory quality of life (by 4.9 for activity score), and decreased cardiopulmonary mortality (by 47%) in participants who previously smoked (all P values < .05), but not in participants who currently smoke. Moreover, effect sizes of the DiPHeaL score on respiratory quality of life measures were greater among participants who previously smoked with airway obstruction compared with those without. Associations with cardiovascular and respiratory mortality were replicated in PLCO participants who previously smoked . A higher DiPHeaL score was also associated with lower lung cancer incidence in participants who previously smoked, as well as reduced COPD incidence and lung cancer mortality regardless of smoking status in the PLCO.

Interpretation

We defined a novel dietary pattern for lung epigenetic aging, which linked to lung health measurements. Participants who previously smoked, especially those with airway obstruction, may benefit the most from nutritional modification.

Take-Home Points.

Study Question: Can a sputum methylation panel be used to derive a dietary pattern beneficial for lung health? Is this dietary pattern associated with various subjective and objective lung health phenotypes? Does this relationship vary among people who currently smoke vs people who previously smoked?

Results: We have derived a Dietary Pattern for Healthy Lung (DiPHeaL) for lung epigenetic aging, which is characterized by low consumption of processed meat and high consumption of dark green vegetables, tea, alcohol, and fruit juice. A higher DiPHeaL score (indicating better diet) was associated with better spirometry and respiratory quality of life, and decreased cardiopulmonary mortality in participants who previously smoked, but not in participants who currently smoke, in the Lovelace Smokers cohort. The effect sizes of the DiPHeaL score on respiratory quality of life measures were greater among former tobacco users with airway obstruction compared with those without. Associations with cardiovascular and respiratory mortality were replicated in Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening trial participants who previously smoked. A greater DiPHeaL score was also associated with lower lung cancer incidence in participants who previously smoked, as well as reduced COPD incidence and lung cancer mortality, regardless of smoking status in the PLCO trial.

Interpretation: Smoking cessation should remain the top approach to maintain lung health, and people who previously smoked, especially those with airway obstruction, may benefit from dietary modification to improve their lung health both objectively and subjectively.

Maintaining optimal lung health is essential to reducing the risk for COPD and lung cancer in people who consume cigarettes, 2 leading causes of disability and death in the United States and globally.^{1, 2, 3, 4} Lung health can be phenotypically assessed using psychometric (eg, St. George’s Respiratory Questionnaire [SGRQ]) or objective (eg, spirometry and lung disease diagnosis) measurements, or endotypically quantified using biomarkers (eg, sputum methylation) pathogenically linked to lung diseases.^{5, 6, 7, 8, 9} Compelling evidence indicates that gene methylation detected in exfoliated lung epithelial cells collected in sputum provides an assessment of field cancerization and predicts primary lung cancer incidence or its recurrence.^{10, 11, 12, 13, 14, 15, 16, 17} Moreover, in addition to being a validated lung cancer risk biomarker, sputum methylation may also serve as a lung epigenetic aging biomarker in people who ever smoked based on its associations with chronic bronchitis, accelerated decline of lung function, and increased all-cause mortality.^5,6

Dietary intake has attracted particular interest in translational epigenetics due to the mechanistic involvement of certain dietary components in the regulation of DNA methylation.^18,19 Extensive studies including our own have consistently shown that folate insufficiency results in promoter hypermethylation of tumor suppressor genes and/or hypomethylation of repetitive genomic elements.^{20, 21, 22} However, studies focusing on individual food items or nutrients are challenged by the fact that people consume a diverse array of food items, each comprising a multitude of distinct nutrients and bioactive constituents. The dietary patterns of people offer a comprehensive understanding of food consumption behaviors, which are shaped by a combination of culture, socioeconomic status, demographics, and long-term dietary habits.^23,24 So far, 1 study assessed the associations of established dietary patterns (Mediterranean-style diet score or Alternative Healthy Eating Index) with DNA methylation in blood and identified methylation of 30 CpG sites associated with better dietary pattern scores, among which methylation of 12 CpG sites were also associated with all-cause mortality in directions consistent with the protective role of these 2 dietary patterns.¹⁸ However, no specific dietary pattern has been developed for lung epigenetic aging.

We conducted a nutritional epidemiological study in the Lovelace Smokers Cohort (LSC) and the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening trial. Our goal is to derive a first-of-its-kind dietary pattern for lung epigenetic aging quantified as promoter hypermethylation of a 12-gene panel in sputum using least absolute shrinkage and selection operator (LASSO) regularized Poisson regression in the LSC, and then assess its associations with lung health measurements in these 2 large-scale prospective cohorts.

Study Design and Methods

Study Population

Lovelace Smokers Cohort

The LSC was established in 2001 to study sputum and blood biomarkers for lung cancer risk assessment and COPD development in people who currently smoke and who previously smoked enrolled from the greater Albuquerque area of New Mexico.²⁵ All participants enrolled were 40 to 75 years of age, consumed cigarettes with at least 10 pack-years of smoking history, and were free of prior lung cancer history. The current study includes 1,802 participants with Harvard Food Frequency Questionnaire (FFQ) data collected at baseline. This study was approved by the Western Institutional Review Board, and all participants signed consent forms.

Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial

The PLCO Cancer Screening trial enrolled 154,887 participants at 10 centers across the United States from 1993 to 2001, to test whether the screening examinations could lower the risk of death from prostate, lung, colorectal, and ovarian cancers.²⁶ All participants were aged 55 to 74 years without diagnosis of prostate, lung, colorectal, or ovarian cancer, or recent cancer screening for the organs of interest at enrollment. The current study excluded people who returned invalid dietary history questionnaires (n = 41,440), never smoked (n = 52,919), or had missing smoking status (n = 1,962), resulting in a total of 24,961 women and 33,605 men in analyses (e-Fig 1). Data access was under request PLCO-981 approved by the National Cancer Institute Cancer Data Access System on June 24, 2022. The University of New Mexico Health Sciences Center Institutional Review Board approved this analysis through an Exempt 4 mechanism (22-173).

Gene Promoter Methylation in Sputum in the LSC

Twelve genes, including P16, MGMT, RASSF1A, DAPK, GATA4, GATA5, PAX5α, PAX5β, SULF2, PCDH20, DAL1, and JPH3, were selected for methylation analysis in cytologically adequate sputum samples based on our previous studies establishing their associations with risk for lung health and their specificity for methylation in lung epithelial cells.^{10, 11, 12} Given the low percentage (< 3%) of lung epithelial cells in sputum samples that also varied significantly between people, a 2-stage nested methylation-specific polymerase chain reaction was used to detect methylated alleles.^{6,10, 11, 12,27} The methylation status of each gene was scored as 0 (unmethylated) or 1 (methylated). A composite methylation index was defined as the number of genes that are methylated, which provided the most optimal discrimination and improved the lung cancer prediction accuracy by 15%.^12,28 A higher methylation index predicted worse lung health.

Development of Dietary Pattern for Sputum Methylation in the LSC

The Harvard FFQ was used to collect average consumption frequencies and serving sizes of about 150 specific food items in the year before study entry. Our studies of 2 FFQs collected 1 decade apart in 28 LSC participants supported a relatively stable dietary pattern over a decade for the studied population.²⁹ Referring to the Dietary Guidelines for Americans, 2020-2025, and the MyPyramid Equivalents Database 2.0, food items were aggregated into 30 main groups according to the similarity of their nutritional composition (e-Table 1). The intake serving of each food item per month was estimated and then adjusted for total calorie intake, using the residual method. Monthly intake servings of the 30 food groups were calculated by summing intakes of specific food items within each respective food group. Then, z-score standardized transformation of food group intake servings was conducted to improve data normality and comparability.

The LASSO regularized Poisson regression with 10-fold cross-validation was applied on the above z-score standardized food group intakes to select a panel of food groups highly related to sputum methylation index. More specifically, the LSC data were separated into 10 equal parts with the model built in 80% of each part and then tested in the remaining 20% of each part. The mean squared error was averaged across all the folds. This method offers an effective way to achieve feature selection by shrinking the regression coefficients of less relevant food groups to zero, thereby facilitating the identification of a parsimonious set of influential dietary factors.³⁰ All selected food groups had nonzero effect coefficients from the LASSO regression. The derived dietary pattern (referred to as the Dietary Pattern for Healthy Lung [DiPHeaL] hereafter) was applied in further analyses in the form of a score (called the DiPHeaL score hereafter), which is a linear combination of the z-score standardized intakes of the selected food groups and coefficients derived in the Poisson model.

Estimation of the DiPHeaL Score in the PLCO

Participants in the PLCO reported their intake frequencies of commonly consumed food items over the previous year via diet history questionnaires. For each participant, monthly intake servings of the 5 food groups were summarized, calorie-adjusted, and z-score standardized and then included in the same equation derived in the LSC to derive the DiPHeaL scores.

Health Outcomes

From both cohorts, pulmonary and extrapulmonary comorbidity data were collected at baseline, and mortality and lung cancer incidence at follow-up. In addition, the LSC also provided respiratory quality of life data, using the SGRQ at baseline,³¹ and longitudinal spirometry (once every 18 months).^32,33 PLCO participants were followed up for incident COPD information by 2015. A detailed description of variables is given in the online article (e-Appendix 1). Lists of health outcome variables and sample sizes in both cohorts are available in Table 1 and e-Table 2.

Table 1.

Characteristics, Dietary Intake, and Health Outcomes of LSC and PLCO Participants

Variable	LSC	PLCO
No. of patients	1,802	58,566
Range of dietary score	–0.586, 0.340	–0.750, 0.812
Age at entry (mean ± SD), y	56.7 ± 9.3	65.5 ± 5.7
BMI (mean ± SD), kg/m2	28.2 ± 6.2	27.3 ± 4.8
Female sex	1,399 (77.6%)	24,961 (42.6%)
Ethnicity
Hispanic	299 (16.6%)	932 (1.6%)
Non-Hispanic White	1,381 (76.6%)	53,313 (91%)
Other racea	122 (6.8%)	4,298 (7.3%)
Missing	NA	23 (0%)
Formerly smoked	788 (43.7%)	48,124 (82.2%)
Pack-years (mean ± SD)	40.3 ± 20.9	34.8 ± 28.6
Some college or above	1,287 (71.4%)	32,943 (56.2%)
Dietary intake (50th [25th, 75th] percentile), servings/mob
Dark green vegetables	8.3 (3, 18.3)	5.1 (2.1, 12.3)
Alcohol	2.3 (0, 22.5)	6 (0.6, 32.7)
Tea	4 (0.50, 24)	8.3 (2.1, 22.1)
Fruit juice	4.5 (1.3, 16)	6.1 (0.8, 38.6)
Processed meat	5 (2.5, 12.5)	8.6 (4.1, 17.5)
Calories (50th [25th, 75th] percentile), kcal	1,718.4 (1,340.3, 2,185.8)	1,656.6 (1,256, 2,174.1)
Methylation index, mean ± SDc	2.33 ± 0.04	NA
Baseline health outcome measures
SGRQ score, mean ± SD
Symptom	30.3 ± 23.0	NA
Activity	31.3 ± 24.7	NA
Impact	11.2 ± 13.6	NA
Total	21.2 ± 17.4	NA
Baseline spirometry, mean ± SD
FEV1, mL/s	2,512.1 ± 748.8	NA
FEV1/FVC ratio, %	72.9 ± 10.6	NA
Baseline lung morbidities, event (rate)d
Emphysema	222 (1,232)	2,169 (370)
Chronic bronchitis	431 (2,392)	3,155 (539)
Dyspnea	989 (5,488)	NA
Prospective health outcome measures
Spirometry test (50th percentile [25th, 75th])	4 (2, 7)	NA
Mortality, event (rate)e
All-cause	322 (283)	22,495 (480)
Cardiopulmonary	134 (118)	NA
Respiratory	NA	2,807 (60)
Lung cancer	NA	2,496 (53)
Cardiovascular	NA	6,736 (144)
Incident lung cancer, event (rate)e	NA	2,010 (46)
Incident COPD, event (rate)e	NA	3,013 (1,584)

Data are presented as No. (%) unless otherwise indicated. Continuous variables are presented as mean ± SD or 50th (25th, 75th) percentile according to their distributions; categorical variables are presented as number (percentage); morbidities, mortalities, and lung cancer incidences are presented as event (rate). LSC = Lovelace Smokers Cohort; NA = not applicable; PLCO = Prostate, Lung, Colorectal, and Ovarian; SGRQ = St. George’s Respiratory Questionnaire.

^aCategories for other race include Native American, Asian, Black, other, and individuals who prefer not to specify.

^bRaw data on dietary intake servings per month are presented.

^cA total of 348 participants in the LSC had missing methylation data; sputum methylation data were not available in the PLCO.

^dRate for morbidities is cases per 10,000 people.

^eRate for mortalities and incidences for lung cancer and COPD is cases per 100,000 person-years. A total of 3,013 COPD incidences was identified from a subset of the PLCO trial (n = 11,115).

Statistical Analyses

Variables were summarized separately in the LSC and PLCO participants, using statistics based on their nature (ie, categorical vs continuous) and distribution (ie, skewed vs normal). In the LSC cohort, logistic and Poisson regression were used to evaluate effects of the DiPHeaL score on each individual gene methylation and the composite methylation index, respectively; a linear mixed-effects model with a subject-specific random intercept and slope was applied to assess associations of the DiPHeaL score with longitudinal spirometry (FEV₁ and FEV₁/FVC ratio); and linear regression was used to reveal the associations between the DiPHeaL score and SGRQ scores at baseline. Cox proportional hazard regression models were used to estimate the hazard ratios (HRs) and 95% CIs of mortality risk associated with the DiPHeaL score in both the LSC and PLCO, as well as of lung cancer and COPD incidence risks in the PLCO. Considering the large number of participants with invalid dietary information, sensitivity analyses were conducted to assess the likely impact that these missing dietary data might have on the associations of interest. In these analyses, we applied analysis weights that reflected the propensity for having missing dietary data to each PLCO participant to obtain estimates that accounted for the presence of missing data.^34,35 Age at enrollment, sex, ethnicity, BMI, smoking status, pack-years, and educational level were included for covariate adjustment. Because smoking is a primary risk factor for poor lung health, we stratified study participants as those who had formerly smoked and or who currently smoke and explored the effects of the DiPHeaL score among different subgroups. To further reveal the most sensitive population, we performed stratification analyses by smoking and airway obstruction (FEV₁/FVC ratio < 0.7) status. The statistical analyses were performed with the SAS program (version 9.4), except for LASSO regression and propensity weight calculation, which were performed with R software (version 4.2) with “glmnet” and “twang” packages, respectively. A P value < .05 was considered statistically significant.

Results

Dietary Pattern Associated With Sputum Methylation in the LSC

The optimal dietary pattern for sputum methylation was selected by the dimensional reduction method of LASSO, with the model tuning parameter lambda (λ) (0.09) automatically selected using cross-validation (Fig 1). At this optimal λ value, processed meat, dark green vegetables, alcohol, tea, and fruit juice were identified as significant stressors of sputum methylation with 0.0375, –0.0344, –0.0112, –0.0035, and –0.0024 as effect coefficients, respectively. We flipped the direction of effect coefficients to facilitate the interpretation and defined a dietary pattern, termed DiPHeaL, which is associated with lower sputum methylation and presumably better lung health. The DiPHeaL score was calculated as a linear combination of the z-score standardized intakes of the selected food groups. The DiPHeaL score was normally distributed in the LSC and PLCO participants (e-Fig 2). As shown in e-Figure 3 and e-Table 3, the DiPHeaL scores increased with decreasing methylation index in the LSC. Moreover, each standard deviation (SD) increase in the DiPHeaL score was significantly associated with a reduction in the odds of sputum methylation by 0.25 for DAPK, 0.16 for GATA5, and 0.13 for SULF2 (all P < .05). This suggests that DAPK is the most responsive gene to the dietary pattern identified in our study. We also found current smoking, male sex, and education less than college as determinants for lower DiPHeaL scores in both cohorts (all P < .05; e-Table 4).

Figure 1.

Sputum methylation-related food group selection, using least absolute shrinkage and selection operator (LASSO) regression in the Lovelace Smokers Cohort (LSC). The LASSO penalized regression model was conducted among LSC participants and included the sputum methylation index as the dependent variable and z-score standardized intakes of 30 food groups as the independent variables. The intake servings of 30 food groups were simultaneously included in the model. A, The vertical gray dotted lines with their upper and lower SD curves (error bars) represent the 10-fold cross-validation curve along the λ sequence, and the vertical red dotted lines represent the optimal value of λ with the minimum deviance. λ = 0.09 was selected as the optimal tuning parameter for LASSO regression. B, LASSO coefficient profiles of the 30 candidate food groups. Processed meat, dark green vegetables, tea, alcohol, and fruit juice were selected as the most significant stressors to sputum promoter methylation.

Association of the DiPHeaL Score With Lung Morbidities in Both Cohorts

In the LSC (e-Table 5), a higher DiPHeaL score was significantly associated with lower risk of emphysema, chronic bronchitis, and dyspnea in participants who had ever smoked at baseline (OR, 0.78, 0.82, and 0.83, respectively), and these associations were only significant (emphysema and dyspnea) or more robust (chronic bronchitis) in participants who had formerly smoked (OR, 0.60, 0.73, and 0.67, respectively). However, no significant associations were found for other comorbidities except diabetes. Consistently, significant associations of the DiPHeaL score with emphysema and chronic bronchitis were replicated in participants who had ever smoked and those who had formerly smoked in the PLCO, whereas no associations were observed in participants who currently smoke (e-Table 6). Sensitivity analysis considering propensity weights showed consistent findings (e-Table 7). Associations with other baseline comorbidities were mostly insignificant except for diabetes, gallbladder stones or inflammation, and stroke in participants who had ever smoked and who had formerly smoked in the PLCO. In addition, Cox regression revealed 4% decreased risk of incident COPD among all PLCO participants. When stratified by smoking status, significant relationships between DiPHeaL and COPD incidence risk were observed in both participants who formerly smoked and those who currently smoke (HR, 0.92 and 0.88, respectively; e-Table 6).

Association of the DiPHeaL Score With Respiratory Quality of Life at Baseline in the LSC

As summarized in Figure 2, greater DiPHeaL scores were associated with lower SGRQ scores among participants who had ever smoked and those who formerly smoked in the LSC (all P < .05), but not in participants who currently smoke. When further stratifying the participants by smoking and airway obstruction status, we observed stronger associations in participants who formerly smoked with airway obstruction than in those without. Among participants who formerly smoked with airway obstruction, differences in SGRQ total, symptom, and activity subdomain scores along with per SD increment in DiPHeaL score all exceeded the clinically significant difference (β, –5.19, –5.11, and –7.54, respectively), and the difference in psychosocial impact score was approaching clinical significance (β, –3.22). Interaction analyses revealed significant modified effects of smoking status and airway obstruction in participants who formerly smoked on respiratory quality of life (e-Table 8).

Figure 2.

Associations between sputum methylation-related dietary pattern and respiratory quality of life at baseline in the Lovelace Smokers Cohort (LSC). Respiratory quality of life was assessed in the LSC using the St. George’s Respiratory Questionnaire (SGRQ). Four scores were calculated for each participant, including scores for (A) symptoms, (B) activity, (C) psychosocial impact domains, and (D) a total score for overall quality of life. Linear models were applied with baseline SGRQ scores as dependent variables and the identified dietary score as independent variables. Age at entry, ethnicity, sex, BMI, educational level, smoking status, and pack-years were adjusted in the multivariable linear models. ∗P < .05. Per SD increase in DiPHeaL score was significantly associated with better respiratory quality of life (by 2.57 for symptom [A], 4.91 for activity [B], 1.83 for psychosocial impact [C], and 3.04 for total score [D]) in people who previously smoked. Stratified analyses identified a more pronounced impact on quality of life in people who previously smoked with airway obstruction compared with those without, with the magnitude of effects exceeding the clinically significant difference. DiPHeaL = Dietary Pattern for Healthy Lung.

Association of the DiPHeaL Score With Lung Function in the LSC

As shown in Figure 3 and e-Table 8, each SD increase in the DiPHeaL score was associated with higher FEV₁ (44.7 mL/s) and 0.69% higher FEV₁/FVC ratio in participants who had ever smoked in the LSC, and in participants who had formerly smoked (96.1 mL/s for FEV₁ and 1.8% for FEV₁/FVC ratio), but not in participants who currently smoke. Because participants with airway obstruction were quite vulnerable, we further stratified the study participants by smoking status and airway obstruction, and found that effects of the DiPHeaL score on FEV₁ and FEV₁/FVC ratio were much more pronounced among participants who formerly smoked with airway obstruction (215.4 mL/s for FEV₁ and 3.4% for FEV₁/FVC ratio; all P for interaction < .05). No significant associations were observed among participants who formerly smoked without airway obstruction (all P > .05).

Figure 3.

Impacts of dietary pattern for sputum methylation on longitudinal (A) FEV₁ and (B) FEV₁/FVC in the Lovelace Smokers Cohort (LSC). Linear mixed-effect models were used in the LSC to assess the effects of the dietary pattern on lung function measurements, with repeated measurements on lung function as dependent variables and the identified dietary score as independent variables. Fixed effects of baseline age, sex, ethnicity, education level, BMI, smoking status, pack-years, and random effects for intercept and time in the cohort were included in the linear mixed-effect models. ∗P < .05. Per SD increase in DiPHeaL score was significantly associated with higher FEV₁ (by 96.1 mL/s in participants who formerly smoked and 215 mL/s in participants who formerly smoked with airway obstruction, [A]) and higher FEV₁/FVC ratio (by 1.83% in participants who formerkyl smoked and 3.38% in participants who formerly smoked with airway obstruction, [B]). The interaction term between dietary score and time in cohort was not statistically significant, suggesting that the identified dietary pattern does not affect lung function decline (data not shown). DiPHeaL = Dietary Pattern for Healthy Lung.

Association of the DiPHeaL Score With Mortality in Both Cohorts

As shown in Table 2, per SD increase in the DiPHeaL score was associated with a 12% and 23% decreased risk of all-cause and cardiopulmonary mortality in the LSC. Stratification analysis by smoking status revealed enhanced associations between the DiPHeaL score and all-cause and cardiopulmonary mortality among participants who formerly smoked (HR, 0.74 and 0.53, respectively). No significant associations were observed among participants who currently smoke. Moreover, the associations between the DiPHeaL score and all-cause and cardiopulmonary mortality in participants who formerly smoked did not vary by airway obstruction status (data not shown).

Table 2.

Associations Between Sputum Methylation-Related Dietary Pattern and Cause-Specific Mortality in LSC and PLCO

Mortality	LSC			PLCO
	Cases/Noncases	HR (95% CI)	P Value	Cases/Noncases	HR (95% CI)	P Value
Among all participants
All-cause mortality	322/1,480	0.88 (0.78, 0.99)	.032	22,495/36,071	0.95 (0.94, 0.97)	< .001
Cardiopulmonary mortality	134/1,668	0.77 (0.65, 0.91)	.002	NA	NA	NA
Respiratory mortality	NA	NA	NA	2,807/55,759	0.94 (0.91, 0.98)	.004
Cardiovascular mortality	NA	NA	NA	6,736/51,830	0.96 (0.94, 0.98)	.001
Lung cancer mortality	NA	NA	NA	2,807/55,759	0.93 (0.89, 0.97)	.001
Among participants who formerly smoked
All-cause mortality	162/626	0.74 (0.62, 0.89)	.002	17,378/30,746	0.95 (0.94, 0.97)	< .001
Cardiopulmonary mortality	69/719	0.53 (0.41, 0.68)	< .001	NA	NA	NA
Respiratory mortality	NA	NA	NA	1,900/46,224	0.94 (0.90, 0.99)	.010
Cardiovascular mortality	NA	NA	NA	5,387/42,737	0.95 (0.92, 0.98)	< .001
Lung cancer mortality	NA	NA	NA	1,467/46,657	0.94 (0.89, 0.99)	.019
Among participants who currently smoke
All-cause mortality	160/854	1.00 (0.84, 1.18)	.957	5,117/5,325	0.97 (0.94, 0.99)	.014
Cardiopulmonary mortality	65/949	1.08 (0.83, 1.42)	.565	NA	NA	NA
Respiratory mortality	NA	NA	NA	907/9,535	0.96 (0.89, 1.03)	.217
Cardiovascular mortality	NA	NA	NA	1,349/9,093	1.00 (0.95, 1.06)	.964
Lung cancer mortality	NA	NA	NA	1,029/9,413	0.93 (0.87, 0.98)	.014

Cox proportional hazard regression models were applied with the identified dietary score as independent variables, and cause-specific mortalities as dependent variables. Because of data availability in 2 cohorts, only all-cause and cardiopulmonary mortalities were assessed in the LSC, whereas all-cause, respiratory, cardiovascular, and lung cancer mortalities were evaluated in the PLCO. In the LSC, respiratory and cardiovascular mortality were combined because of smaller sample size, whereas in the PLCO, these 2 were analyzed separately due to the large sample size. These models were adjusted for age at entry, ethnicity, sex, BMI, educational level, smoking status, and pack-years. HR = hazard ratio; LSC = Lovelace Smokers Cohort; NA = not applicable; PLCO = Prostate, Lung, Colorectal, and Ovarian.

These associations were further replicated in the PLCO. As illustrated in Table 2, per SD increase in the DiPHeaL score was correlated with a 5%, 6%, 7%, and 4% decreased risk of all-cause, respiratory, lung cancer, and cardiovascular mortality. Similar effects on respiratory and cardiovascular mortality were observed only among participants who formerly smoked (all P < .05), not among participants who currently smoke (all P > .05). As for all-cause and lung cancer-caused mortality, we found similar protection effects among participants who formerly smoked and those who currently smoke (all P < .05). Sensitivity analysis yielded consistent findings (e-Table 7), suggesting minimal impact of missing dietary data on the rigorousness of our results.

Association of the DiPHeaL Score With Lung Cancer Incidence

Considering the low lung cancer incidence in the LSC (n = 88), we explored the relationship between DiPHeaL score and lung cancer incidence only in the PLCO. Figure 4 showed 5% decreased incident lung cancer risk along with per SD increase in the DiPHeaL score risk among PLCO participants who ever smoked. When stratified by smoking status, a greater risk reduction was observed among participants who formerly smoked (HR, 0.92; P = .005), but not among those who currently smoke (P = .98). Sensitivity analysis showed concordant findings (e-Table 7).

Figure 4.

Hazard ratio (HR) (95% CI) for incident lung cancer risk according to sputum methylation-related dietary scores in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening trial. Cox proportional hazard regression models were applied with the identified dietary score as independent variables, and incident lung cancer risk as the dependent variable. Age at entry, ethnicity, sex, BMI, educational level, smoking status, and pack-years were adjusted in the above models. ∗P < .05. Per SD increase in DiPHeaL score was significantly associated with decreased lung cancer incidence by 2% in all participants and by 8% in participants who formerly smoked. DiPHeaL = Dietary Pattern for Healthy Lung.

Discussion

This study went beyond our previous understanding of individual food items and nutrients associated with sputum methylation and/or lung health,^20,21,29,36 and for the first time defined a dietary pattern called “DiPHeaL” specifically for lung epigenetic aging in a large cohort of people who ever smoked. This pattern was characterized by lower consumption of processed meat and higher consumption of dark green vegetables, alcohol, tea, and fruit juice. Additional analyses in 2 independent cohorts revealed consistent associations of greater DiPHeaL with better lung health measurements (spirometry; SGRQ; and self-reported dyspnea, emphysema, and chronic bronchitis) among all participants who had ever smoked and those who formerly smoked. Moreover, a higher DiPHeaL score was also associated with lower risk of all-cause and cardiopulmonary mortality among all participants who had ever smoked and those who formerly smoked, with the protective effects extending to participants who currently smoke for all-cause mortality in the PLCO. With sufficient power in the PLCO, we also demonstrated that a higher DiPHeaL score was also associated with a lower risk of lung cancer incidence and mortality among all participants who had ever smoked and those who had formerly smoked, with the protective effects extending to participants who currently smoke for lung cancer mortality. Thus, these findings strongly support our strategy of using a well-defined sputum methylation panel as a biomarker for lung epigenetic aging to derive a novel dietary pattern for malignant and nonmalignant lung outcomes.

Previous studies identified several prudent dietary patterns that were associated with better cross-sectional spirometric measurements^{37, 38, 39} and a reduced prospective risk of COPD incidence.⁴⁰ These prudent dietary patterns were identified using unsupervised principal component analyses and included high consumption of fresh fruit, vegetables, soybeans, meat, poultry, oily fish or seafood, eggs and whole meal cereals, and low consumption of white bread, added sugar, chips, and processed meat. Findings on dairy products differed across various studies. Our analyses benefit from an innovative statistical method, LASSO, which enables the extraction of a less biased data-driven pattern from information obtained through a relatively comprehensive food questionnaire. Another advantage is that we employed food group intake with negligible or mild correlations (e-Fig 4) as the independent variables and a validated lung epigenetic aging biomarker as the outcome to drive the selection. Compared with the prudent dietary patterns, our DiPHeaL pattern does not include oily fish, whole meal cereals, soybeans, meat, poultry, eggs, white bread, added sugar, full-fat dairy products, or chips. The DiPHeaL, however, contains tea and alcohol, which were not major contributors to the prudent pattern. The beneficial role of alcohol consumption for lung health in our study is consistent with the literature, which demonstrates better spirometry and lower COPD mortality associated with light to moderate alcohol intake.^{41, 42, 43, 44, 45} Moreover, alcohol seems to have a stronger impact on FVC vs FEV₁, with potential mechanisms involving muscle strength and antiinflammation.^{42, 43, 44} We additionally assessed relationships between the DiPHeaL and lung health measurements individuals who drink heavily (> 32 servings/mo for women, > 60 servings/mo for men), and found consistent protective effects of the DiPHeaL (data not shown).

Because few cohorts provided sputum methylation data, we could not replicate the DiPHeaL-methylation association in another population. However, the DiPHeaL score is calculated using calorie-adjusted and z-score standardized intakes of 5 food groups, and can be easily generalized to other cohorts which provided FFQ data and lung health measurements. Although the DiPHeaL pattern was derived in the LSC, which included more than 70% women, we successfully replicated its associations with lung health outcomes in the PLCO, which contains 42.6% women. Stratified analysis by sex in the PLCO yielded similar results between men and women (data not shown). Because the DiPHeaL was established as a lung epigenetic aging biomarker, it may not be associated with risk for extrapulmonary diseases. We explored the associations between the DiPHeaL score and baseline comorbidities in the LSC and PLCO (e-Tables 5 and 6). As expected, the DiPHeaL score was significantly associated with decreased risk for all lung morbidities among people who previously smoked, but was not associated with morbidities of liver, colon, kidney, peripheral blood vessels, bone and joints, and so on. It is interesting to note that a greater DiPHeaL score was associated with a lower risk of diabetes in both cohorts and of gallbladder stones and heart attack in the PLCO.

We compared the health effects of DiPHeaL on lung health measurements with those of the alternative Mediterranean diet (aMED; e-Table 9), which is well recognized for its benefits in chronic disease prevention (e-Table 10).^46,47 For cause-specific mortality and lung cancer incidence, both DiPHeaL and aMED showed comparable beneficial effects in the LSC and PLCO. However, in terms of lung-specific outcomes, such as respiratory quality of life and lung function at baseline, DiPHeaL and aMED showed similar effect sizes per SD increase among all LSC participants who ever smoked, whereas DiPHeaL demonstrated substantially higher effect sizes compared with aMED among participants who formerly smoked with airway obstruction. Regarding pulmonary comorbidities at baseline, DiPHeaL was consistently associated with a lower risk of emphysema, chronic bronchitis, and dyspnea in both cohorts, whereas no significant associations were observed between aMED and chronic bronchitis in the LSC. Notably, analyses of COPD in a PLCO subset revealed that only DiPHeaL had a beneficial impact on incident COPD in the PLCO, regardless of whether the analysis was stratified by smoking status. Thus, when focusing on lung-specific outcomes, DiPHeaL outperforms the traditional general dietary pattern, demonstrating stronger and more consistent results, particularly among participants who formerly smoked with airway obstruction.

The biological mechanism underlying this dietary pattern affecting lung epigenetic and phenotypic health in people who ever smoked should be multifaceted. First, dark green vegetables and fruit juice are rich in folate; betaine; choline; and vitamins B₂, B₆, and B₁₂, all important coenzymes in one-carbon metabolism that links nutritional biochemistry to epigenetics.^{20, 21, 22} Second, extensive studies, including ours, have identified evidence linking compromised DNA repair, especially toward double-strand DNA breaks, to the acquisition of DNA methylation in gene promoters.^25,48 The current study observed reduced nonhomologous end-joining capacity associated with poor DiPHeaL score among people who previously smoked (e-Table 11). This is consistent with the fact that extensive DNA damage created challenges in cells not only for restoring DNA integrity but also for maintaining the proper epigenetic modification of DNA and chromatin during DNA damage response. The third mechanism may be related to the antioxidant and antiinflammatory activities associated with this dietary pattern. Dark green vegetables, tea, and fruit juice contain antioxidant and antiinflammatory activities. Processed meat, however, contains high levels of pro-oxidants, such as nitrites^{49, 50, 51} and advanced glycation end-products.^52,53 The sputum DNA methylation panel of 12 tumor suppressor genes has potential in lung physiology, too. DAPK, the most responsive gene to the DiPHeaL, is known to express broadly in lung tissues⁵⁴ and to induce apoptosis,⁵⁵ which may help maintain lung hemostasis by eliminating damaged cells due to extrinsic and intrinsic stimuli. In addition, DAPK can activate multicellular antiinflammation through translational repression of inflammatory genes⁵⁶ and attenuation of inflammatory responses,⁵⁷ which is essential to the normally functioned lung.⁵⁸ The second responsive gene, GATA5, is expressed in airway smooth muscle cells and bronchial epithelium.^59,60 Its deficiency can cause airway hyperresponsiveness through downregulation of apolipoprotein E and upregulation of IL-13.⁶⁰ These findings indicate that the 12-gene panel is capturing underlying processes common to both malignant and nonmalignant lung conditions. However, the exact mechanisms underlying the links of dietary intake to promoter methylation remain inconclusive. Future studies addressing why those genes are preferentially affected may assist the discovery of mechanisms underlying observed associations.

Interpretation

To the best of our knowledge, we have developed the first dietary pattern based on lung epigenetic aging to guide the selection of food groups, which is further associated with multiple lung health measurements. Smoking cessation should remain the top priority for maintaining lung health; however, people who previously smoked, especially those with airway obstruction, may benefit from dietary modification to maintain their lung function as well as alleviate the psychometric impact of respiratory disease. Our study is limited by the observational design. Thus, future trials assessing efficacy of dietary intervention are warranted.

Funding/Support

This work was supported by NIEHS R01 ES035421, American Cancer Society Institutional Research Grant IRG–21-146-25-IRG, NCI P30 CA118100, NINR 1P20NR021824-01, and NHLBI 1R21HL173388-01.

Financial/Nonfinancial Disclosures

None declared.