CVD is the leading cause of morbidity and mortality in most of the world, despite efforts to reduce risk through effective LDL-lowering therapy(Reference Cannon, Braunwald and McCabe1, Reference LaRosa, Grundy and Waters2). To further reduce CVD burden, attention has expanded toward HDL as a therapeutic target. A low HDL-cholesterol (HDL-C) level is a significant independent risk factor for CVD and high HDL-C is usually cardioprotective(3, Reference Chapman, Assmann and Fruchart4). While numerous efforts are underway to develop new pharmacological approaches to increase HDL levels, lifestyle interventions are also promising (for a review, see Singh et al. (Reference Singh, Shishehbor and Ansell5)).
Abundant evidence indicates that regular light or moderate alcohol consumption reduces the risk of CHD and all-cause mortality (for a review, see O'Keefe et al. (Reference O'Keefe, Bybee and Lavie6)). At least half of alcohol's cardiovascular benefits are attributed to increased HDL-cholesterol(Reference Criqui, Cowan and Tyroler7); HDL rises in a dose-dependent fashion with a 5 % increase with one drink per d and up to 12 % increase with two or three drinks per d over a 3-week time period(Reference Asztalos, Cupples and Demissie8–Reference Valmadrid, Klein and Moss13).
Lipoprotein subclass composition may help explain CVD risk, though few large studies have investigated the relationship between NMR-measured HDL particle size and CVD risk. Results from existing studies of HDL size and CVD risk have been mixed. Early studies suggest that high levels of large HDL confer decreased risk and high levels of small HDL increase risk(Reference Asztalos, Cupples and Demissie8–Reference Freedman, Otvos and Jeyarajah10, Reference Hattori, Kujiraoka and Egashira14–Reference Rosenson, Otvos and Freedman17); more recent studies have found the reverse(Reference Otvos, Collins and Freedman18–Reference Navab, Ananthramaiah and Reddy21), with risk highest at very high levels of HDL size(Reference van der Steeg, Holme and Boekholdt22).
We hypothesised that, in addition to increasing total HDL, regular alcohol consumption induces a favourable shift in HDL particle size, which might help explain alcohol's protective role in CVD risk. We tested this hypothesis in a cross-sectional study using lipoprotein data from NMR spectroscopy to examine the quantitative association of alcohol consumption with the amount, size, and subclass proportions of HDL-C in a population of relatively healthy community-dwelling older men and women with a large proportion of daily or near-daily drinkers.
Methods
Study population
The Rancho Bernardo Study is a longitudinal population-based study of healthy ageing and CVD risk factors. Study participants are mostly white and middle- to upper-middle class; all are ambulatory. At the initial visit in 1972–4, 82 % of adult Rancho Bernardo residents enrolled in the study. In 1984–7, 81 % (n 2480) of surviving community-dwelling participants returned for a follow-up visit at which time they completed a comprehensive questionnaire and provided a blood sample for laboratory analysis. After exclusions (age < 50 years, current cholesterol medication, or TAG>4000 mg/l), 2174 were eligible for the present study. Of these, two were excluded for missing alcohol information and one for implausibly high reported alcohol intake, leaving a final sample of 2171. The study complies with the Declaration of Helsinki and was approved by the institutional review board of the University of California, San Diego. All study participants gave written informed consent.
Alcohol consumption
Usual alcohol consumption and weekly number of drinks of beer, wine and liquor were reported on a standardised self-administered questionnaire. One drink was equivalent to roughly 12 g alcohol. Participants also were asked, ‘How often do you usually consume alcoholic beverages?’ with possible responses: ‘never’, ‘1 time/month’, ‘1–2 times/month’, ‘1–2 times/week’, ‘3–4 times/week’ and ‘almost daily or daily’. Before analysis, the less frequent categories were merged to form the following categories of alcohol frequency: none, < 3 times/week, 3–4 times/week, and 5+times/week.
Lipoprotein and lipid measurements
Lipoprotein subclasses were determined by standard NMR spectroscopy at LipoScience, Inc. (Raleigh, NC, USA) in 1999 from plasma samples collected at the 1984–7 follow-up visit and stored at − 70°C. Proton NMR spectra of freshly thawed samples (0·25 ml) were acquired in duplicate at 47°C using a dedicated 400 MHz NMR analyser (LipoScience, Inc.). Lipoprotein particles of different size give rise to distinguishable lipid methyl group NMR signals, the intensities of which are proportional to the lipid mass of the particles(Reference Otvos, Jeyarajah and Bennet23, Reference Freedman, Otvos and Jeyarajah24). The NMR signal amplitudes were converted to mass concentration units (mg/l) of cholesterol (for LDL and HDL subclasses) or TAG (for VLDL subclasses) to allow comparison with chemically measured lipid fractions.
The NMR-determined subclasses were classified by diameter range. Diameter range was determined by calibration with purified lipoprotein subfractions isolated by ultracentrifugation and/or agarose gel filtration chromatography. For the present study we examined the HDL subclass categories: small HDL (7·3–8·2 nm), intermediate HDL (8·2–8·8 nm) and large HDL (8·8–13 nm). Summation of subclasses provides level of total HDL. Average HDL particle size was calculated by weighting the relative mass percentage of each subclass by its diameter. The HDL subclasses correspond to HDL subclasses by gradient gel electrophoresis as follows: large HDL ≈ HDL2b+HDL2a; intermediate HDL ≈ HDL3a; small HDL ≈ HDL3b+HDL3c.
Total fasting plasma cholesterol and TAG levels were measured by enzymic techniques using an ABA-200 biochromatic analyser in a Centers for Disease Control (CDC)-certified laboratory (Abbott Laboratories, Irving, TX, USA). HDL level was measured after precipitation of the other lipoproteins with heparin and manganese chloride. LDL levels were calculated using the Friedewald equation(Reference Friedewald, Levy and Fredrickson25).
Other covariates
Height, weight and waist circumference were measured in the clinic by trained nurses with participants wearing light clothing and no shoes. BMI (kg/m2) was used as an estimate of overall adiposity. Diabetes was defined by the 1999 WHO criteria: fasting blood glucose ≥ 126 mg/dl ( ≥ 1260 mg/l), 2 h post-challenge glucose level ≥ 200 mg/dl ( ≥ 2000 mg/l), history of diabetes diagnosed by a physician, or treatment with an oral hypoglycaemic agent or insulin(26). Participants self-reported their use of postmenopausal hormone replacement therapy (HRT), smoking and physical activity in a validated questionnaire. HRT and other medications were validated by a nurse who examined pills and prescriptions brought to the clinic.
Statistical analyses
Data were analysed with SPSS (version 16.0; SPSS Inc., Chicago, IL, USA). ANOVA, general linear models, and χ2 analysis were used for descriptive statistics across four alcohol-use groups (none, < 3 times per week, 3–4 times per week, 5+times per week). Mean HDL particle sizes, subclass levels and other characteristics were compared among alcohol-use groups by univariate analysis for linear and quadratic trend. The association between alcohol consumption (both as a continuous variable and as a categorical variable based on alcohol frequency) and HDL subclasses was assessed with multiple linear regression; current HRT use in women (categorical, yes or no), BMI (as a continuous variable), diabetes (categorical, yes or no, based on 1999 WHO criteria), smoking (categorical, yes or no), exercise (categorical, moderate activity 3 or more d per week, yes or no) and age (as a continuous variable) were evaluated as covariates. Because HDL size and the intermediate and large HDL subclasses were skewed, median levels are presented for those variables. Non-parametric analyses with Kruskal–Wallis tests revealed qualitatively similar findings as parametric analyses; P values are presented for parametric analysis of linear and quadratic trend only. All analyses were stratified by sex. Statistical significance was designated as P < 0·05.
Results
A total of 1197 women (mean age 72 years) and 974 men (mean age 73 years) participated in the present study. Overall, 8 % of participants had a BMI greater than 30 kg/m2, 17 % met Adult Treatment Panel III criteria for central obesity (waist girth ≥ 102 cm for men and ≥ 88 cm for women) and 15 % had diabetes. Only 13 % reported current smoking and of those approximately half smoked one pack per d or more; 81 % reported exercising three or more times per week. Current postmenopausal HRT was reported by 27 % of women; 19 % were taking oestrogen alone and 8 % were taking a combined oestrogen–progestin therapy.
Overall, 958 (44 %) participants reported drinking at least five times per week (38 % of women and 51 % of men); 12 % abstained from alcohol. Table 1 shows selected baseline characteristics of study participants across alcohol frequency categories. In both sexes, smoking behaviours and exercise frequency increased linearly with alcohol frequency while diabetes prevalence and fasting insulin levels decreased with increasing alcohol frequency. Women on HRT reported higher frequency of alcohol consumption and a corresponding significantly higher alcohol consumption in grams. BMI and waist circumference were lower among those with higher alcohol consumption in women but not men. Waist:hip ratio and creatinine levels did not differ by alcohol intake.
(Mean values and standard deviations, medians and interquartile ranges or percentages)
HOMA-IR, homeostasis model of insulin resistance; SGOT, serum glutamic oxaloacetic transaminase; SGPT, serum glutamic pyruvic transaminase.
* Asymmetric variables log-transformed for analysis.
† HOMA-IR available in a subset of 1552 men and women.
We assessed chemically measured TAG, total cholesterol, LDL and HDL and NMR-determined HDL subclasses including large, intermediate and small HDL and overall HDL size. The mean values of enzymically derived HDL-C (620 mg/l) and HDL measured by proton NMR (HDL-P) (480 mg/l) differed appreciably, with HDL-P approximately 21 % lower than HDL-C. However, the correlation between the ascertainment methods was high (R 0·85; P < 0·01) (data not shown). Table 2 shows unadjusted levels of lipids and HDL subclasses by category of alcohol frequency. TAG levels were progressively lower as the frequency of alcohol consumption increased; however, when alcohol consumption was assessed as a continuous variable (number of drinks per week), TAG levels and alcohol consumption were negatively correlated among individuals who drank ≤ 14 drinks per week (R − 0·051; P = 0·033) and positively correlated among those who drank>14 drinks per week (R 0·108; P = 0·024) (data not shown). Total cholesterol levels were higher with increasing alcohol frequency in both men and women. While LDL-cholesterol (LDL-C) changes were not significant for men and followed an inverted U-shaped pattern for women (P quadratic ≤ 0·001) in that non-drinkers and daily drinkers had the lowest levels, an analysis of the LDL particle number showed that increased alcohol consumption was associated with increased particle number for both men and women (P ≤ 0·01). LDL-C and HDL-C levels were modestly inversely correlated with each other for men (R − 0·12; P < 0·01) and women (R − 0·27; P < 0·01). The LDL-C:HDL-C ratio decreased with increasing alcohol consumption for both sexes but this decrease was primarily due to an alcohol-related increase in HDL. Qualitative results for these analyses were similar when women were stratified by current HRT use.
(Mean values and standard deviations, medians and interquartile ranges or percentages)
LDL-C, LDL-cholesterol; HDL-C, HDL-cholesterol.
* Log transformed for analysis.
† P values for one-way ANOVA weighted linear trend or quadratic trend.
Levels of each HDL measure were significantly related to alcohol frequency. Levels of HDL-C (P < 0·001), large and intermediate HDL-P (P < 0·001), large HDL-P percentage (P < 0·001 in women, P = 0·002 in men), total and large HDL-P for men and women not on HRT (P < 0·001), large HDL-P percentage (P < 0·001 in women, P = 0·002 in men), small HDL-P in men (P = 0·013), and HDL size (P < 0·001) were higher with increased alcohol consumption; the opposite was true for small HDL-P percentage (P < 0·001). Small HDL-P in women followed an inverted U-shaped pattern in which levels were highest for non-drinkers and for those who drank most frequently (quadratic P < 0·001). Total and large HDL-P followed a J-shaped pattern for women on HRT. Levels of HDL-C, large HDL-P and intermediate HDL-P were higher in women compared with men across all alcohol categories. Figure 1 graphically compares HDL subclass proportions by alcohol frequency, sex and current HRT.
HDL subclass distribution by sex and hormone replacement therapy (HRT). (
), Large HDL particles; (
), medium HDL particles; (
), small HDL particles.
We defined a multivariate model of various HDL subclasses as a linear function of both alcohol frequency and the number of alcoholic drinks per week (where an alcoholic drink contains 12 g alcohol), HRT, BMI, diabetes, smoking, exercise and age. Results were similar for both analyses. Results with alcohol as a continuous variable are shown in Table 3. Alcohol and HRT were associated with higher levels of HDL-C, medium HDL-P, large HDL-P and HDL size. While small HDL-P levels were lower among those currently using HRT and only modestly lower among those with higher levels of alcohol consumption, the percentage of small HDL-P was significantly lower with hormone use and increasing alcohol consumption (P < 0·001) (data not shown). BMI, diabetes, smoking, exercise and age also influenced HDL subclasses to varying degrees, as represented by the size of β coefficients in Table 3.
Standardised β coefficients of multivariate analysis for HDL subclasses, proportions and size by sex
* P < 0·05, ** P < 0·01, *** P < 0·001.
† Coefficient β represents association of dichotomous variable and alcohol intake with standardised outcome.
Discussion
It is known that both alcohol intake and high levels of HDL-C are associated with decreased risk of CVD (for reviews, see Singh et al. (Reference Singh, Shishehbor and Ansell5) and O'Keefe et al. (Reference O'Keefe, Bybee and Lavie6)). In the present study of relatively healthy community-dwelling older adults, of whom 44 % were daily or near-daily alcohol drinkers, alcohol consumption was associated with 20 % higher total HDL-C, with a higher total amount and percentage of large and medium HDL-P, and with a lower percentage of small HDL-P. These trends were identical for men and women, and independent of HRT use, itself a modulator of HDL amount and type(Reference Mackey, Kuller and Sutton-Tyrrell27–Reference Vadlamudi, MacLean and Israel29). The size of the alcohol-associated HDL changes reported here are equivalent to those of niacin, the current treatment of choice to raise HDL-C. Niacin raises HDL by approximately 20–30 %(Reference Singh, Shishehbor and Ansell5), presumably through increases in apo A-1 production and large HDL with little effect on small HDL(Reference Morgan, Carey and Lincoff30). Consistent with previous studies, moderate alcohol consumption was associated with decreased TAG levels whereas reported consumption greater than two drinks per d was associated with higher TAG levels(Reference O'Keefe, Bybee and Lavie6).
Previous research
We are aware of five other epidemiological studies that examined the association between alcohol intake and HDL subclasses; only one of these assessed subclasses by NMR spectroscopy. The first, a cross-sectional study of 151 male and 146 female, mostly Mormon, participants found that among participants aged ≥ 18 years, alcohol was associated with an increase in the large lipid-rich HDL2b in both sexes and a selective increase in intermediate-sized HDL3a/2a region in men only(Reference Williams, Vranizan and Austin31). In a case–control study of 340 hospitalised patients with a myocardial infarction and an equal number of matched controls, those who reported drinking the most alcohol had higher levels of total, large HDL2 and small HDL3(Reference Gaziano, Buring and Breslow11). In the third study of 279 men, participants who drank more than 5 g alcohol per d had a higher proportion of large lipid-rich HDL2 cholesterol compared with small lipid-poor HDL3 than those who drank less than 5 g alcohol per d(Reference Schafer, Parlesak and Eckoldt32). These three studies used gradient gel electrophoresis to measure HDL subclasses. Among participants in the Atherosclerosis Risk in Communities (ARIC) Study, alcohol consumption was associated with higher levels of total HDL and HDL3 cholesterol in African-American participants and greater levels of total HDL, HDL2 and HDL3 in white participants measured by dextran-sulfate and Mg precipitation(Reference Volcik, Ballantyne and Fuchs33). Only one study, the Cardiovascular Health Study of 1850 participants aged 65 years and older, used NMR. They found that alcohol intake was associated with decreased small HDL and increased medium and large HDL measured by NMR(Reference Mukamal, Mackey and Kuller34).
Assuming that high levels of large HDL are protective, these findings support the hypothesis that alcohol may decrease CVD risk by inducing favourable changes in HDL subclasses. This interpretation is based on the assumption that small HDL contributes to increased cardiovascular risk and large HDL contributes to decreased risk. However, some mechanistic studies suggest that small HDL particles may be anti-inflammatory and thereby decrease CVD risk whereas large HDL particles may be pro-inflammatory and increase risk for recurrent CVD(Reference Ansell, Fonarow and Fogelman19–Reference van der Steeg, Holme and Boekholdt22, Reference Schafer, Parlesak and Eckoldt32). Importantly, the latter results are from studies conducted primarily in individuals with prevalent CVD or at high risk for CVD whereas Rancho Bernardo participants are community dwelling and unselected for CVD or CVD risk. Perhaps a pro-inflammatory state such as CVD or consumption of alcohol, which has anti-inflammatory effects(Reference Albert, Glynn and Ridker35, Reference Sierksma, van der Gaag and Kluft36), alters the functionality of HDL subclasses. More research is needed to better understand the role of HDL size in CVD risk.
Previous work suggests that alcohol influences HDL through several pathways including: increased production of apo A-1, a HDL precursor(Reference Gaziano, Buring and Breslow11–Reference Valmadrid, Klein and Moss13, Reference Volcik, Ballantyne and Fuchs33); increased muscle ATP-binding cassette, subfamily A (ABCA1) which may be important in recycling preformed HDL through reverse cholesterol transport(Reference Hoang, Tefft and Duffy37); decreased cholesteryl ester transfer protein (CETP)(Reference Forrester, Makkar and Shah38–Reference Serdyuk, Metelskaya and Ozerova41). The blunting of the LDL-C:HDL-C ratio associated with increased alcohol intake in the present study combined with the increased number of large HDL particles is consistent with alcohol-induced decreased CETP activity. Lowered CETP level is associated with increased total and large HDL-P(Reference Brousseau, O'Connor and Ordovas42, Reference Ordovas, Cupples and Corella43) as well as potent anti-atherosclerotic activity in several studies(Reference Forrester, Makkar and Shah38); however, failure of the CETP-inhibitor torcetrapib in a clinical trial(Reference Barter, Caulfield and Eriksson44–Reference Nissen, Tardif and Nicholls47) has raised doubts regarding the role of CETP and HDL-C in cardiovascular risk reduction (for a review, see Joy & Hegele(Reference Joy and Hegele48)). In the Investigation of Lipid Level management to Understand its impact IN ATherosclerotic Events (ILLUMINATE) trial torcetrapib had no impact on atherosclerosis(Reference Barter, Caulfield and Eriksson44–Reference Nissen, Tardif and Nicholls47) and led to elevation in blood pressure, aldosterone levels, and morbidity and mortality. Further analyses suggested that the elevations in blood pressure and aldosterone were probably not due to CETP inhibition, and it may be that these unique side effects contributed to torcetrapib's failure(Reference Rader49). More research is necessary to better understand how alcohol affects CETP inhibition and other pathways to alter lipoprotein distribution and how alcohol-induced lipoprotein changes, which are far more modest than those induced by torcetrapib, affect CVD.
Limitations
The present study has limitations. The Rancho Bernardo Study cohort is almost entirely white, middle- to upper-middle-class, and older; therefore, these findings may not apply to other age, ethnic and socio-economic groups. Although generalisability is reduced, the homogeneity reduces confounding of socio-economic status and ethnicity. While it is possible that self-reported alcohol intake resulted in over- or underestimating actual intake, relatively high alcohol intakes were freely reported based on a standard questionnaire, probably reflecting normative behaviour for this age and socio-economic group. Although we adjusted for many of the most likely confounders, it is possible that the association between alcohol intake and total HDL and HDL subclasses reflects residual confounding due to unascertained differences between drinkers and non-drinkers. For example, dietary differences could be an important source of confounding. However, there is limited evidence in the literature that common diet items other than alcohol have a meaningful effect on HDL-C. The only nutrient proposed to have benefit is n-3 fatty acids, largely from fatty fish. Although marine-derived n-3 fatty acids have only modest effects on total HDL-C, there is some evidence that they also alter HDL subfractions toward a more favourable, cardioprotective profile (increased HDL2 (large HDL), decreased HDL3 (small HDL))(Reference Mori, Burke and Puddey50, Reference Lungershausen, Abbey and Nestel51). The current enthusiasm for consuming fish for health followed the era when the data of the present study were collected. Though fish consumption may affect HDL composition, there is little reason to believe that those who drink more alcohol also consume more fatty fish. Finally, the present study is cross-sectional and therefore causality cannot be assumed, but the strong dose–response association is suggestive.
Other studies have found that LDL subclasses determined by NMR spectroscopy are highly correlated with the subclasses determined by gradient gel electrophoresis(Reference Blake, Otvos and Rifai52, Reference Grundy, Vega and Otvos53), but the correlation among HDL subclasses has not been adequately evaluated. The NMR-based approach to the lipoprotein subclasses requires ongoing validation.
While our questionnaire did not distinguish regular from binge drinking, binge drinking declines with age and is likely to be low in this cohort where daily drinking was common(Reference Naimi, Brewer and Mokdad54). Because most alcohol consumed in this cohort was wine (32 %) or mixed drinks (56 %), not beer (11 %) or hard liquor ( < 1 %), we had limited ability to determine whether beverage type matters. Although wine, especially red wine, is described as a particularly healthy form of alcohol, in epidemiological studies any type of alcohol is usually associated with less CVD(Reference Mukamal, Conigrave and Mittleman55, Reference Mukamal, Jensen and Gronbaek56).
In summary, both alcohol consumption and high levels of HDL-C are associated with a decreased risk of CVD. The present results suggest that one way that alcohol might be cardioprotective is through an increase in overall HDL-C coupled with potentially favourable changes in HDL subclasses, though more research is needed to better understand the role of lipoprotein size in CVD risk as well as how an alcohol-related lipoprotein profile affects CVD risk and outcomes.
Acknowledgements
The present study was supported by grants from the Medical Alumni Endowment Fund at the University of North Carolina at Chapel Hill; the National Institute of Diabetes & Digestive & Kidney Diseases, a component of the National Institutes of Health (grant no. DK31801); and the National Institutes of Health/National Institute on Aging (grant no. AG07181 and no. AG028507).
We are thankful to the Rancho Bernardo Study participants for their ongoing participation and commitment to the study. We also thank Jaclyn (Nikki) Bergstrom for her assistance with statistical analysis and LipoScience, Inc. for NMR spectroscopy lipoprotein analysis of participant blood samples.
N. D. M. was responsible for the conception and design of the study, analysis and interpretation of data, drafting and revising the manuscript, and final approval of the submitted manuscript; G. A. L. was responsible for the conception and design of the study, analysing and interpreting data, writing and revising the manuscript, and the final approval of the manuscript; D. v. M. was responsible for data analysis and interpretation, writing and revising the manuscript, and final approval of the manuscript; S. C. S. Jr was responsible for writing and revising the manuscript, and final approval of the manuscript; E. B.-C. was responsible for the conception and design of the study, analysis and interpretation of data, writing and revising the manuscript, and final approval of the manuscript.
There are no conflicts of interest to disclose.
