INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Eicosapentaenoic acid (EPA, C20:5) and docosahexaenoic acid (DHA, C22:6) are polyunsaturated fatty acids of the n-3 (omega-3) family. Both contribute to the structure of cell membranes and both can modulate some cellular functions (1). Since the human body cannot efficiently synthesize either EPA or DHA from C18 precursors, the two fatty acids are considered nutritionally essential (2). Their primary and almost exclusive exogenous source is seafood.

Supplementation of the diet with high doses of EPA and DHA has been found to attenuate several components of the inflammatory response (3), and to benefit some patients with chronic inflammatory conditions such as rheumatoid arthritis (11) and ulcerative colitis (12). We therefore hypothesized that greater dietary intake of the two fatty acids would attenuate the inflammatory response of the lung to cigarette smoking, thereby protecting some smokers against chronic obstructive pulmonary disease (COPD).

We and others have previously reported an inverse association between the intake of n-3 fatty acids or fish consumption and COPD among cigarette smokers (13, 14). We report here tests of the hypothesis that the prevalence odds of smoking- related COPD are inversely related to the proportion of EPA and DHA in the phospholipid and cholesterol ester fractions of the plasma.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Source Population

The subject sample for the study was derived from the Atherosclerosis Risk in Communities (ARIC) Study, a population-based study of atherosclerosis and cardiovascular diseases (15). Nearly 16,000 middle-aged men and women were recruited for that study in four communities in the United States from 1987 to 1989 and underwent a 3 to 4 h comprehensive examination in study clinics. The present analysis was restricted to study participants from suburban Minneapolis, because measurements of plasma fatty acids were made only in that population (16). We estimated that 67% of eligible residents in sampled households took part in the baseline examination (17).

Plasma Fatty Acids

Fasting blood was collected according to a studywide protocol. The sample for fatty acid analysis was collected in a 10-ml tube containing ethylenediamine tetraacetic acid (EDTA). The tube was refrigerated at the study clinic and sent by courier within 3 h to the University of Minnesota Hospital and Clinic Laboratory. Upon arrival, the blood was centrifuged at 800 × g for 10 min. The plasma was separated and divided into two 1.5-ml aliquots, and was frozen at 70° C. A single technician analyzed the fatty acid composition of the plasma approximately 2 yr later according to the following procedure.

After thawing, 0.5 ml of plasma was extracted with 0.5 ml of methanol followed by 1.0 ml of chloroform under a nitrogen atmosphere, and the lipid extract was filtered to remove protein. The phospholipid and cholesterol ester fractions were separated by thin-layer chromatography, using a silica-gel plate and two-stage mobile-phase development, with use of 80:20:1 (vol/vol) and 40:60:1 (vol/vol) petroleum ether, diethyl ether, and glacial acetic acid, respectively. The chromatography plate was dried between exposure to development solvents, and the second mobile phase was allowed to migrate only half the length of the plate. After the plate had been redried, one lane was sprayed with dichlorofluorescein to visualize the phospholipid, cholesterol ester, triglyceride, and fatty acid bands under ultraviolet light. The phospholipid and cholesterol ester bands were scraped into separate test tubes and the lipids were converted to methyl esters of fatty acids by boron trifluoride catalysis (18). The methyl esters were then separated and measured with a Model 5890 gas chromatograph (Hewlett-Packard, Avondale, PA) equipped with a 30-m FFAP WCOT glass capillary column (J&W Scientific, Folsom, CA) and a flame-ionization detector.

The identity of 28 individual fatty acid peaks revealed by gas chromatography was determined by comparing each peak's retention time to the retention times of fatty acids in synthetic standards of known fatty acid compositions. The relative amount of each fatty acid (as a percent of all fatty acids) was calculated by integrating the area under the peak and dividing the result by the total area for all fatty acids (×100). To minimize transcription errors, the data from the gas chromatogram was electronically transferred to a VAX computer (Digital Equipment Corporation, Minneapolis, MN) for data analysis.

The estimated short-term (within weeks) reliability coefficients for EPA and DHA ranged from 0.31 (EPA in phospholipids) to 0.58 (DHA in phospholipids) (19). The long-term reliability coefficients (within 3 yr) ranged from 0.35 (EPA in cholesterol esters) to 0.51 (EPA in phospholipids) (19). In three of four comparisons the reliability coefficient for DHA was larger than the reliability coefficient for EPA.

Smoking Exposure

Participants provided information about their smoking habits (whether they were current smokers, ex-smokers, or never-smokers). Former smokers and current smokers estimated the numbers of years they had smoked and the lifetime average number of cigarettes they had smoked per day. Current smokers also reported the average number of cigarettes they were smoking at the time of the study. To quantify lifetime smoking exposure, we used the construct of pack-yr of smoking, defined as the lifetime average number of cigarettes smoked per day divided by 20 and multiplied by the number of years the subject had been smoking.

Spirometric Measurements

Trained clinic technicians used a Collins Survey II volume-displacement spirometer (Warren E. Collins, Braintree, MA) which was connected to a PC/XT computer (IBM, Armonk, NY) through the Pulmo- Screen II software system (S&M Instrument, Doylestown, PA). The measurement protocol was based on guidelines of the Epidemiology Standardization Project (20) and the American Thoracic Society (21).

From among at least five forced expirations, the spirometry technician attempted to obtain three acceptable spirograms, of which at least two had similar results (within 5%) for FEV₁ and FVC. The largest value of FEV₁ and FVC from any of the participant's acceptable tests was selected. We also computed predicted values of FEV₁ and FVC, taking into account the subject's sex, race, age, and height (22, 23).

Other Measurements

Height (rounded to the nearest centimeter) and weight (rounded to the nearest pound) were measured while the participant was wearing an examination gown and not wearing shoes. Structured interviews assessed the presence or absence of cough and production of phlegm, including questions about the frequency and chronicity of these symptoms. Participants also reported whether a doctor had told them that they had emphysema.

COPD Definitions

Because COPD is a heterogenous entity, we used clinical indicators, spirometric indicators, and a combination of both in defining it. We defined chronic cough (or chronic production of phlegm) as coughing on most days for at least three consecutive months of the year for at least 2 yr. When both chronic cough and chronic production of phlegm were reported, we classified the participant as having chronic bronchitis. We also used two independent definitions of spirometry-based COPD: an FEV₁  65% predicted and an FEV₁/FVC × 100  65%. Additionally, we used a combined, sensitive definition of COPD by identifying subjects with at least one of the following: chronic bronchitis, FEV₁  65% predicted, FEV₁/FVC × 100  65%, or self- reported physician-diagnosed emphysema. There were too few cases of self-reported emphysema to consider this entity separately.

Analysis

Of 2,525 current or former smokers in the Minneapolis field center of the ARIC Study, 2,349 contributed data to the cross-sectional analysis described here. We excluded participants for whom we had no data about plasma fatty acids (n = 45), those who reported taking lipid-lowering medications (n = 51), and those who did not fast for 12 h (n = 80).

In a previous communication we reported an extremely high correlation between questionnaire-assessed dietary intake of EPA and DHA that precluded fatty acid-specific analysis (13). In the present study, the proportion of EPA in both the plasma phospholipid and cholesterol ester fractions correlated only moderately with the proportion of DHA in the corresponding fraction (the Pearson correlation coefficients were 0.36 and 0.45 in the phospholipid and cholesterol esters fractions, respectively), and each fatty acid could therefore be analyzed separately. Our analyses indicated that the relation of EPA, but not of DHA, to prevalent COPD was compatible with the null theory. Because of space constraints (and because the measurement of DHA appeared to be more reliable [19]), we focus on DHA and present only one typical null result for EPA.

The fatty acid content of plasma fractions was measured on a continuous scale (as a proportion). However, we chose not to impose the assumption of a linear relation of fatty acid content with the log odds of prevalent COPD, and instead classified the participants into four quartiles on the basis of the proportion of DHA (or EPA) in their plasma. The lowest quartile of the distribution served as a reference for the other three quartiles.

The unconditional relation of possible determinants of COPD to the quartile distribution of DHA was described by mean values, percentiles, or proportions. Univariate and multivariate associations of the DHA quartile with the various COPD case definitions were examined with unconditional logistic regression models, and linear trends across the quartiles were tested by modeling a four-value ordinal variable. In the case of continuous measurements of lung function (FEV₁, FEV₁/FVC), we also tested the study hypothesis by analysis of covariance (ANCOVA), computing adjusted mean values for each DHA quartile.

Because confounding by smoking exposure was our primary concern, we used two approaches to control for the effect of smoking on COPD and lung function. First, in our main multivariate analysis, we simultaneously modeled three indices of smoking exposure: pack-yr of smoking, the average lifetime number of cigarettes smoked per day, and the current number of cigarettes smoked per day (assigning a value of 0 to former smokers). Although the three measures were correlated, each proved to contribute to the fit of most models. Second, we repeated the analyses using data from a restricted group of current heavy smokers (20 or more cigarettes per day). Although precision was compromised by the small sample size represented by this group, the threat of confounding by smoking exposure was presumably less severe in such a relatively homogenous group of heavy smokers.

Although we had no prior hypotheses about other fatty acids, we replicated the analysis for all of the fatty acids in the data set. SAS software (SAS Institute, Cary, NC) was used for computations.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We found that n-3 fatty acids composed a very small proportion of total fatty acids in both the phospholipid and cholesterol ester fractions of the plasma. The average proportion of DHA in plasma phospholipids and plasma cholesterol esters was 2.7% and 0.5%, respectively. The average proportion of EPA was almost identical in the two fractions (0.55%). The proportion of DHA in one plasma fraction correlated highly with the proportion in the other fraction (Pearson correlation coefficient = 0.95). Therefore, analysis of either plasma fraction yielded very similar results. Because of space constraints, only results for plasma phospholipids are shown here.

The natural allocation of subjects into quartiles of DHA cannot be viewed as "successful randomization" (Table 1). In particular, there was an inverse relation between the quartile of DHA and smoking exposure in that the higher the DHA quartile, the lower the proportion of current smokers and the median value of pack-yr of smoking. This relation might reflect a shared antecedent cause, an effect of smoking on DHA intake or metabolism, or pure coincidence.

Table 2 shows the overall prevalence of COPD indicators among 2,349 current or former smokers, and the prevalence of COPD in quartiles of DHA. Although various disease indicators sometimes identified the same individuals, the overlap among the case groups was generally modest. Few participants reported physician-diagnosed emphysema. The prevalence of COPD was generally inversely related to the DHA quartile, in part because for any two quartiles, participants in the higher quartile had less exposure to smoking. Clearly, the hypothesis that differences in smoking exposure explain all of the pattern in Table 2 required multivariate modeling.

Table 3 shows the univariate odds ratios (ORs) of prevalent COPD for DHA quartiles, and the fate of those estimates in multivariate models. As was expected, in all cases adjustment attenuated the univariate estimates, although the magnitude of the attenuation varied. Generally, our list of putative confounders accounted for 20 to 40% of the magnitude of the univariate effect, and rarely produced adjusted estimates that were compatible with the null. Departure from the null is also evident within each DHA quartile (i.e., down the columns of Table 3). Compared with Quartile I, the ORs for Quartiles II, III, and IV ranged from 0.65 to 1.09, 0.23 to 0.93, and 0.48 to 0.85, respectively. In general, the associations of DHA with clinical indicators of COPD were weaker than the associations with spirometric indicators. Adjusted mean values of FEV₁ in ANCOVA models were 2,994, 3,066, 3,107, and 3,097 ml for subjects in Quartiles I through IV, respectively. The corresponding mean values of FEV₁/FVC (lowest to highest DHA quartile) were 72.0%, 73.5%, 73.7%, and 73.8%.

Restriction of the analysis to 543 current smokers of 20 or more cigarettes per day yielded the data in Table 4. In that group, the relation of DHA quartile to prevalent COPD was compatible with the null hypothesis for two indicators of the disease (chronic production of phlegm and chronic bronchitis), but showed departure from the null for the other four indicators. Point estimates for the magnitude of the uppermost quartile effect ranged from as strong as 0.36 to as weak as 0.85 (and 1.09 in one case). Adjusted mean values of FEV₁ (lowest to highest quartile) among current heavy smokers were 2,706, 2,785, 2,801, and 2,854 ml. The corresponding mean values of FEV₁/FVC × 100 in this group were 69.4%, 70.5%, 71.9%, and 71.2%.

Table 5 summarizes the results of multiple regression of the log odds of COPD (combined definition) on the quartile distribution of other fatty acids in the data set, including EPA (C20:5). We highlighted associations that we interpreted as departures from the null on the basis of the data in Table 5 and in regression models for other COPD indicators (not shown).

Relatively few never-smokers were classified as having COPD. For example, only 60 subjects were classified as having chronic bronchitis, and only 92 subjects met the combined definition of having COPD. Neither EPA nor DHA showed convincing associations with any COPD indicator among never-smokers (data not shown).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Chronic inflammation in response to smoking is one presumed causal pathway of COPD (24, 25). Inhalation of cigarette smoke leads to retention of neutrophils in the lungs (26), to a higher concentration of leukotriene B₄ (LTB₄), an inflammatory agent, in the sputum (27), and to greater release of free radicals, such as superoxide anion, by alveolar macrophages (AM) (28). LTB₄ and free radicals are postulated stimulants of mucus secretion in the airways (25), and free radicals might also be involved in the pathogenesis of emphysema (29). Other putative mediators of pulmonary inflammation are interleukin-1 (IL-1), tumor necrosis factor (TNF), platelet-activating factor (PAF), and platelet-derived growth factors (PDGF) (25, 30). If inflammatory reaction to repetitive smoke inhalation is indeed a cause of COPD, then suppression of pulmonary inflammation might decrease the risk of the disease in some smokers.

In numerous human feeding experiments, high doses of EPA and DHA attenuated several components of the body's inflammatory response, including those that might be triggered by cigarette smoking. High intake of the two fatty acids (or one of them) attenuated the chemotactic responsiveness of neutrophils and monocytes (3) and decreased the secretion of LTB₄ (3, 6), free radicals (7), IL-1 (8), TNF (8), PAF (9), and PDGF (10). It should be noted that the typical supplement in these experiments contained several grams of n-3 fatty acids, a quantity which is 10- to 30-fold greater than the average intake in the American diet. It is unknown, however, whether the effect of n-3 fatty acids on the inflammatory response is constrained by some minimal threshold dose.

The proportionate distribution of DHA in plasma lipids is likely to be a surrogate measure of the DHA content of other, COPD relevant, biologic structures, rather than having an independent protective role. If causality indeed underlies the association of DHA levels with COPD, the putative preventive factor is the DHA content of cell membranes, and particularly membranes of circulating white blood cells and resident AM. In animal feeding models, diets containing EPA and DHA increased the content of both fatty acids in alveolar type II cells (31), suppressed the biosynthesis of proinflammatory eicosanoids in the lung (32), and inhibited the accumulation of neutrophils in that organ (32). Similarly, in vitro experiments have shown that exposure of rat AM (33) or isolated rabbit lungs (34) to n-3 fatty acids can attenuate the pulmonary inflammatory response.

Some readers might regard the results we present here as equivocal in light of the width of several confidence intervals, (CIs), the results of several tests of statistical significance, and some inconsistency in the findings for the group of current heavy smokers. It might also be noted that our statistical tests of six definitions of COPD were not completely independent (in a statistical sense) because some of the case definitions overlapped and so did the actual case groups. We offer two answers to these arguments. First, in the absence of randomization to exposure status, it is not at all clear that tests of statistical significance or CIs are much more informative than the estimates themselves. Second, the issue at hand is whether the causal null theory survived empirical testing and whether we should still prefer it in light of all of the data. With inferential statistics or without them, it is not that easy to reconcile many of the results with the causal null being true.

It may be argued that our null results for never-smokers undermine the validity of the results for smokers because residual confounding by smoking exposure is nonexistent among never-smokers (whereas it could have accounted for the associations in smokers). This argument, however, subscribes to two background theories, neither of which is necessarily true. First, it is presupposed that COPD caused by active smoking and COPD arising through other mechanisms (e.g., air pollution) have similar pathogeneses and should therefore respond similarly to protective factors. There seems to be a substantial difference, however, between active inhalation of cigarette smoke on a daily basis and breathing of air even when the latter is considered polluted by contemporary standards. COPD is overwhelmingly a disease of active cigarette smokers, not of individuals who never smoked. Second, the expectation of homogenous effects across varying initial conditions (such as never-smoking versus ever-smoking) is at odds with both deterministic and indeterministic views of causation.

Notwithstanding these considerations, numerous competing explanations may account for the observed associations of the proportion of DHA in plasma fractions with prevalence of COPD, including residual confounding by smoking exposure (or by other measured covariates), confounding by unknown causes of COPD, differential measurement error, prevalence- incidence bias, reverse causality, and model misspecification, although some of these theoretical explanations (e.g., prevalence-incidence bias and reverse causality) seem unlikely.

Despite the null results for EPA, we cannot easily dismiss a role of this fatty acid, either in the diet or in cellular function, in protecting against COPD. First, because the human body can metabolize EPA to DHA (1), some of the EPA in the diet may eventually be metabolized to DHA, thereby contributing to putative protection against COPD. Second, the estimated intake of EPA based on the dietary questionnaire used in our study correlated more strongly with plasma DHA than with plasma EPA (16). Third, most of our reliability measures suggest a greater component of intrasubject variance plus method variance for the measurement of plasma EPA than for that of DHA (19). Hence, tests of the null theory for EPA were methodologically less rigorous than tests of the null theory for DHA.

Our findings indicate that n-3 fatty acids might be useful in preventing or treating respiratory diseases in which chronic inflammation plays a significant role. This avenue of research has not yet been fully explored (35). Further testing of this paradigm and of our specific hypothesis in experimental models seem warranted.