点击显示 收起
1 From the Departments of Nutrition (QS, HC, and FBH) and Epidemiology (QS, SEH, and FBH), Harvard School of Public Health, Boston, MA, and the Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA (JM, SEH, and FBH)
2 Supported by research grants CA49449, CA42182, HL24074, HL34594, and CA87969 from the National Institutes of Health. FBH is a recipient of the American Heart Association Established Investigator Award. 3 Address reprint requests to Q Sun, Department of Nutrition, Harvard School of Public Health, 665 Huntington Avenue, Boston, MA 02115. E-mail: qisun{at}hsph.harvard.edu. 4 Address correspondence to FB Hu, Department of Nutrition, Harvard School of Public Health, 665 Huntington Avenue, Boston, MA 02115. E-mail: frank.hu{at}channing.harvard.edu.
ABSTRACT
Background: Erythrocyte fatty acids may be superior to plasma fatty acids for reflecting long-term fatty acid intake because of less sensitivity to recent intake and a slower turnover rate.
Objective: The objective was to compare the fatty acid content of erythrocytes with that of plasma with respect to their abilities to reflect usual fatty acid intake.
Design: Fatty acids in plasma and erythrocytes were measured by capillary gas-liquid chromatography in 306 US women aged 43–69 y. Fatty acid intake was assessed with a food-frequency questionnaire, which was validated for measuring intakes of various fatty acids.
Results: Docosahexaenoic acid (DHA, 22:6n–3) in erythrocytes and plasma provided the strongest correlations with its intake, but erythrocyte DHA concentrations [Spearman's partial correlation coefficient (rs) = 0.56] were better than plasma DHA concentrations (rs = 0.48) as a biomarker. Total trans fatty acids (rs = 0.43) and total 18:1 trans isomers (rs = 0.42) in erythrocytes were also more strongly correlated with intake than were those in plasma (rs = 0.30 and rs = 0.29, respectively). Moderate correlations were observed for linoleic acid (18:2n–6; erythrocytes, rs = 0.24; plasma, rs = 0.25), -linolenic acid (18:3n–3; erythrocytes, rs = 0.18; plasma, rs = 0.23), and eicosapentaenoic acid (20:5n–3; erythrocytes, rs = 0.38; plasma, rs = 0.21). For polyunsaturated and trans fatty acids, correlations between intakes and biomarkers improved moderately when average intakes over previous years were used.
Conclusion: Erythrocyte n–3 fatty acids of marine origin and trans fatty acid content are suitable biomarkers for long-term intake.
Key Words: Fatty acids • erythrocytes • plasma • biological markers • food-frequency questionnaires • US women
INTRODUCTION
Precise assessment of fatty acid intake is essential for nutritional epidemiologic studies. In contrast with self-report methods, eg, food-frequency questionnaires (FFQs) and diet records, biomarkers of dietary fatty acids have unique strengths. Biomarkers are objective and do not rely on the accuracy of memories, awareness of fat intake, or willingness to report details of diet (1, 2). Also, nutrient databases may not adequately reflect temporal changes in food composition, which could be readily accommodated by biomarkers (2). Therefore, measurement errors of biomarkers are largely independent of those of self-report methods. For these reasons, biomarkers of fatty acid intake have been widely used in epidemiologic studies to validate FFQs (3-8), to evaluate compliance with dietary interventions (9, 10), or to predict risk of diseases (11-14).
Fatty acids that are of largely exogenous origin, ie, n–3, n–6, trans, and odd-numbered fatty acids, could provide the best quantitative estimate of their intakes (15). These fatty acids can be measured in various blood fractions and tissues, eg, plasma or serum, erythrocytes, and adipose tissue (1, 15). The half-life of linoleic acid in adipose tissue has been estimated to be 680 d (16), which indicates that fatty acids in this tissue could be used to reflect long-term fat intake. Studies have corroborated that polyunsaturated fatty acids (PUFAs) and trans fatty acids in adipose tissue were reasonably correlated with intake measured by FFQs (4, 5, 17, 18). However, the availability of adipose tissue limits its use in epidemiologic studies (15). Instead, blood specimens are more available and widely used. One well-conducted controlled dietary trial clearly showed that serum n–3 fatty acid concentrations responded more quickly than did erythrocyte n–3 fatty acid concentrations to recent dietary supplementation with fish oil (19). Whether erythrocytes better reflect long-term dietary fatty acid intakes than does plasma or serum is of interest. Several studies have examined plasma or erythrocytes separately, and a range of correlations (0.20–0.80) was observed between concentrations of PUFAs and trans fatty acids in these specimens and intakes measured by various FFQs (3, 20-25). However, few studies have been conducted to compare correlations with plasma and erythrocytes within the same population. True between-person variation in diet, laboratory measurement errors, different dietary measurement methods used, and biological variability all contributed to the correlations observed in different studies, which makes it hard to draw conclusions about the relative performance of these 2 specimens on the basis of current literature.
To overcome these limitations, in a US population of women we measured >30 fatty acids in both plasma and erythrocytes under identical conditions and compared them with respect to their ability to reflect long-term fatty acid intake assessed by validated FFQs.
SUBJECTS AND METHODS
Study population
The Nurses' Health Study, initiated in 1976, consisted of 121 700 female registered nurses aged 30–55 y living in 1 of 11 US states. Data on the occurrence of cardiovascular diseases and other illnesses and major risk factors have been collected with biennial questionnaires since baseline. Between 1989 and 1990, blood samples were collected from 32 826 women. Within this subcohort, 167 cases of coronary heart disease newly diagnosed between 1990 and 1996 were identified, and 334 control subjects matched for age, smoking, fasting status, and time of blood drawing were selected. All cases and controls were free of major cancers and cardiovascular diseases at the time of blood drawing. This analysis was conducted in the control group. Of the 334 participants, 7 had missing plasma or erythrocyte measurements, 14 had missing information on intake of fat, and 7 were currently using fish-oil supplements. After these participants were excluded, 306 were available for analysis.
All participants gave written informed consent. The study protocol was approved by the Institutional Review Board of the Brigham and Women's Hospital and the Human Subjects Committee Review Board of Harvard School of Public Health.
Dietary assessment
Fatty acid intake assessed with a semiquantitative FFQ in 1990 was primarily used as usual intake in this study; a detailed description of the FFQ was published elsewhere (1). In the Nurses' Health Study, diet has been assessed with the FFQs since 1980 and updated every 4 y. The original 61-item FFQ used in 1980 was expanded to include >130 food items in the 1984, 1986, and 1990 FFQs. These FFQs inquired about food consumption in the previous year and about the use of cooking oil and fat for frying and baking and the addition of margarine and butter to food. For each food item, a standard portion size was specified, and the participants were asked how often, on average, they consumed foods of that specified amount during the previous year. There were 9 possible coding responses, ranging from "never or less than once per month" to "6 or more times per day." Fatty acid intake was calculated by multiplying the frequency of consumption of each food by the fatty acid composition in the specified amount of that food. The contributions across all foods were then summed for each fatty acid. The food-composition database was primarily based on US Department of Agriculture (USDA) publications (26). Food-composition data from other publications and individual laboratories were also used to supplement the USDA data (27-29). The intake of major saturated fatty acids (12:0, 14:0, 16:0, and 18:0), monounsaturated fatty acids (MUFAs; 16:1n–7, 18:1n–9, and 20:1n–12), PUFAs (18:2n–6, 20:4n–6, and 18:3n–3), and trans fatty acids (16:1, 18:1, and 18:2 trans isomers) was calculated. In our FFQs, 3 questions inquired about intake of canned tuna fish, dark meat fish, and other fish. The assessment of intake of n–3 fatty acids of marine origin, ie, eicosapentaenoic acid (EPA, 20:5n–3), docosapentaenoic acid (22:5n–3), and docosahexaenoic acid (DHA, 22:6n–3), was based on these questions.
Both the 61-item FFQ and the expanded FFQs used in the Nurses' Health Study were validated against multiple-week diet records and biomarkers in adipose tissue (1, 4, 17, 30). For example, among 92 Nurses' Health Study participants, correlation coefficients between intakes assessed by the 1986 FFQ and multiple diet records were 0.57 for total fat, 0.68 for total saturated fatty acids, 0.48 for total PUFAs, and 0.58 for total MUFAs (1).
Blood sample collection and analysis
In 1989 and 1990, interested women in the Nurses' Health Study were sent supplies needed to collect blood samples. The samples were sent back on ice by a prepaid overnight courier. Ninety-seven percent of the samples were received within 24 h of blood drawing. Immediately on arrival, the samples were centrifuged (1200 x g for 15 min at room temperature) and then divided into aliquots of plasma, erythrocytes, and buffy coat fractions. These aliquots were stored in liquid nitrogen freezers at –130 °C or colder until analysis in 2000 and 2002. The methods used to collect and store blood samples have been proven reliable (31). Fatty acids in serum phospholipids stored at –80 °C for 7–12 y showed minimal degradation over time (32). Our blood samples were stored at a much lower temperature (–130 °C), which was intended to minimize any influences on fatty acid concentrations caused by the long-term storage.
Fatty acid concentrations were determined by gas-liquid chromatography. The methods were described elsewhere (20). Briefly, fatty acids in plasma and erythrocytes were first extracted into isopropanol and hexane and then transmethylated with methanol and sulfuric acid. Fatty acid methyl esters were evaporated and redissolved in isooctane and then measured by gas-liquid chromatography. Individual peaks were identified by comparison with known standards, and each peak was quantified by calculating the area under the peak. The concentration of each individual fatty acid was expressed as a percentage of total area under the peaks.
Of 51 fatty acids identified, the 37 fatty acids in both plasma and erythrocytes that had meaningful concentrations (mean concentration >0.01%) are reported here. For the current analysis, MUFAs and PUFAs include cis isomers only. Trans isomers were reported separately. Within-run CV percentages were assessed by repeatedly analyzing pooled samples. The CV percentages of the most abundant fatty acids in plasma were generally lower than those in erythrocytes, although they were all reasonably low. For example, the CV percentage for palmitic acid (16:0) was 1.4% (plasma) compared with 3.3% (erythrocytes); for oleic acid (18:1n–9), 2.4% (plasma) compared with 2.8% (erythrocytes); for linoleic acid (18:2n–6), 1.8% (plasma) compared with 2.8% (erythrocytes); and for DHA, 3.4% (plasma) compared with 7.2% (erythrocytes). The CV percentages of trans fatty acids, for which the concentrations were relatively low, were higher than those of the more abundant fatty acids. The average CV percentage of 18:1 trans isomers was 8.0% for plasma and 7.6% for erythrocytes and of 18:2 trans isomers was 6.9% for plasma and 10.0% for erythrocytes.
Statistical analyses
Fatty acid intake measured with the 1990 FFQ was expressed as a percentage of total fat intake to be comparable with the biomarker measurements. For FFQ measurements and biomarkers, total intakes of saturated fatty acids, MUFAs, PUFAs, and trans fatty acids were calculated by summing the concentrations of individual fatty acids of the same class, if they were detectable and available. Crude and partial Spearman's rank-correlation coefficients (rs), adjusted for age at blood collection, fasting status (yes or no), BMI (in kg/m2), current weight (in kg), postmenopausal status (yes or no), postmenopausal hormone use (never, past, and current), smoking status (never smoker, past smoker, current smoker of 1–14 cigarettes/d, current smoker of 15–24 cigarettes/d, and current smoker of 25 cigarettes/d), and the periods during which the blood samples were assayed, were calculated to determine correlations between fatty acid composition in plasma or erythrocytes and intake. We used t tests to examine the significance of Spearman's partial correlation coefficients (33). Correlation coefficients were considered significant at the 0.05 level. To compare the Spearman's partial correlation coefficients between plasma and erythrocytes, we first obtained the residuals of biomarkers and dietary intakes corrected for the covariates. The residuals were ranked and then converted to probit scale to normalize the ranks (34). On the basis of these transformed ranks, Wolfe's test for comparing dependent correlation coefficients was applied to test the hypothesis that correlation coefficients with dietary intakes were equal for plasma fatty acids and erythrocyte fatty acids (35).
Intake of fatty acids was also measured in 1984 and 1986 with the use of FFQs that were similar to the 1990 FFQ. Spearman's partial rank-correlation coefficients between intake assessed in these years and the biomarker concentrations were calculated. We also calculated average fatty acid intakes by using the 1984, 1986, and 1990 FFQ measures. Correlations among fatty acids within plasma and erythrocyte measurements and correlations between plasma and erythrocyte fatty acids were also calculated.
Multivariate linear regression was used to detect linear trends of biomarker concentrations across deciles of average intake calculated from all 3 FFQs. Fatty acid concentration was entered into the model as a dependent variable; deciles of average intake and total energy intake, age, smoking status, BMI, current weight, postmenopausal hormone use, fasting status, and period of blood assay were entered as independent variables. Least-squares means of biomarkers were calculated for each decile of intake. Robust estimators of variance for these means were calculated to allow for the deviation from assumption of normal distribution of dependent variables (36). P values for linear trend were calculated by entering the median values of each decile of intake into the models as a continuous variable. All P values were 2-sided. Data were analyzed with the Statistical Analysis System software package (version 9.1; SAS Institute, Cary, NC).
RESULTS
Baseline demographic characteristics and intakes of the study participants are shown in Table 1. As expected, the women included in this study were older and more likely to smoke than were the overall Nurses' Health Study population because they were selected as controls matched for the age and smoking status of the cases with myocardial infarction. Intakes of fatty acids and plasma and erythrocyte fatty acid concentrations are shown in Table 2.
View this table:
TABLE 1. Baseline characteristics of study participants in the Nurses' Health Study, 19901
View this table:
TABLE 2. Fatty acid composition in plasma and erythrocytes and intake of fatty acids measured with a food-frequency questionnaire at baseline: the Nurses' Health Study, 19901
Spearman's partial correlation coefficients of plasma or erythrocyte fatty acids with intake in 1990 adjusted for age, fasting status, BMI, postmenopausal status, smoking status, and other covariates are shown in Table 3. We did not observe substantial differences between crude and adjusted correlation coefficients. For both plasma and erythrocytes, correlation coefficients with dietary saturated fatty acids and MUFAs were weak (rs 0.20). Of the PUFAs, DHA had the strongest correlations, albeit the correlation was stronger for erythrocytes (rs = 0.56) than for plasma (rs = 0.48). Similarly, EPA in erythrocytes was more strongly correlated with intake (rs = 0.38) than was EPA in plasma (rs = 0.21). Linoleic acid provided the third strongest correlations with intake; the correlation coefficients for plasma and erythrocytes barely differed (rs = 0.25 and rs = 0.24, respectively). The correlation coefficients for arachidonic acid (20:4n–6) and docosapentaenoic acid were close to zero. Total trans fatty acids (rs = 0.43) and total 18:1 trans isomers (rs = 0.42) in erythrocytes were more strongly correlated with intake than were those in plasma (rs = 0.30 and rs = 0.29, respectively). The correlations for the other trans isomers were not different between erythrocytes and plasma.
View this table:
TABLE 3. Spearman's correlation coefficients between fatty acid composition in plasma and erythrocytes and intake measured with the 1990 food-frequency questionnaire: the Nurses' Health Study, 19901
To examine the time integration of the biomarkers, we further calculated Spearman's partial correlation coefficients between intake measured in 1984, 1986, and 1990 and the cumulative average intake from 1984 to 1990 and the biomarker concentrations (Table 4). We restricted our analysis to PUFAs and trans fatty acids, for which reasonable correlation coefficients were observed in Table 3. For most PUFAs, biomarkers were most strongly correlated with intake measured in 1990 when blood samples were collected. The correlations for intake measured in 1984 were mildly attenuated. For example, the correlation coefficient between erythrocyte DHA content and intake was 0.56 for the 1990, 0.43 for the 1986, and 0.41 for the 1984 questionnaire. A similar pattern was observed for erythrocyte EPA content. Plasma EPA was an exception; the strongest correlation was between plasma EPA and intake in 1984. For trans fatty acids in erythrocytes, the biomarkers were most strongly correlated with intake in 1990, although we did not observe the same pattern for plasma. Correlations between these biomarkers and average intakes during 1984–1990 were not different from, or even stronger than, those for intakes from the 1990 FFQ. In particular, modestly improved correlations were seen for EPA, linoleic acid, and trans fatty acids.
View this table:
TABLE 4. Spearman's partial correlation coefficients between fatty acid composition in plasma and erythrocytes and intake measured with a food-frequency questionnaire (FFQ) during various periods: the Nurses' Health Study, 1984–19901
Linear trends of fatty acids in plasma and erythrocytes across deciles of average dietary fat from 1984 to 1990 (percentage of total fat intake) are shown in Figure 1. After adjustment for total energy and other covariates, clear dose-response relations were observed between DHA, linoleic acid, and trans fatty acids in plasma and erythrocytes and their intakes. P values for linear trend were all <0.001. Linoleic acid concentrations in plasma and erythrocytes plateaued at a high intake; this may partly explain the lower correlations observed for linoleic acid than for n–3 and trans fatty acids.
FIGURE 1.. Relation between cumulative average intake of docosahexaenoic acid (DHA, 22:6n–3), linoleic acid (18:2n–6), and total trans fatty acids during 1984–1990 and their concentrations in plasma and erythrocytes: the Nurses' Health Study, 1990. n = 306. Deciles of dietary intake were adjusted for total energy intake, age, smoking, BMI, current weight, postmenopausal hormone use, fasting status, and period of blood assay. Fatty acid concentrations are least-squares means; bars represent 95% CIs (
Spearman's correlation coefficients between plasma and erythrocyte fatty acids are shown in Figure 2. For each fatty acid or fatty acid group, the plasma content was closely correlated with the erythrocyte content, especially for the fatty acids that are largely exogenously produced. The average correlation coefficient was 0.72. Within plasma and erythrocytes, the fatty acids that accounted for a high proportion of total fatty acids were inversely correlated with each other (data not shown); trans 18:1 isomers were strongly positively correlated with trans 18:2 isomers in both plasma (rs = 0.77) and erythrocytes (rs = 0.66).
FIGURE 2.. Spearman's partial correlation coefficients between plasma and erythrocyte fatty acid concentrations (Nurses' Health Study, 1990) adjusted for age at blood drawing, BMI, current weight, smoking status, postmenopausal status, postmenopausal hormone use, and fasting status at blood drawing. n = 306; n = 87 for eicosapentaenoic acid (20:5n–3). P < 0.01 (t test).
DISCUSSION
We observed moderate-to-strong correlations between n–3 fatty acids of marine origin and trans fatty acids in erythrocytes and plasma and corresponding intakes measured with validated FFQs; overall, correlations with intake were stronger for erythrocytes than for plasma. Our study population did not differ from the whole Nurses' Health Study cohort with respect to age, BMI, diet, and other characteristics that could influence the correlations between biomarkers and intake. Although the study participants were older and more likely to be smokers and postmenopausal, these characteristics did not substantially alter the correlations in the analysis.
As expected, saturated fatty acids and MUFAs in plasma and erythrocytes did not reflect intake, probably because these 2 classes of fatty acids could be endogenously synthesized from carbohydrates. In line with previous studies (20-25, 37), n–3 fatty acids of marine origin and trans fatty acids in plasma or erythrocytes provided the strongest correlations with intake. In general, erythrocyte fatty acids were more strongly correlated with intake than were plasma fatty acids. For example, correlation coefficients of 0.23–0.55 were observed for EPA in plasma fractions (20, 21, 37, 38), whereas relatively higher correlation coefficients (0.36-0.58) were observed for EPA in erythrocytes (22-25). However, it is complicated to make direct comparisons between plasma and erythrocytes on the basis of these data, because the dietary assessment methods used, population characteristics, and true between-person variation of intake and biomarker concentrations can be quite different among these studies. All of these factors could influence the correlation coefficients; therefore, any differences between plasma and erythrocytes could result from either these factors or the inherent differences between these 2 specimens (3, 8).
To our knowledge, the current study is the first to examine the performance of plasma and erythrocytes to reflect intake measured with FFQs within the same population. The half-life of erythrocytes is 120 d, which is much longer than that of plasma lipoproteins. Erythrocytes are therefore hypothesized to be a more appropriate specimen to reflect long-term intake than is plasma or serum. In one controlled dietary trial, Katan et al (19) clearly showed that serum cholesteryl esters were more sensitive to recent diet than were erythrocytes. In serum cholesteryl esters the incorporation half-life of EPA was 4.8 d, and the concentration of EPA peaked after 1–2 mo of fish-oil supplementation. In contrast, in erythrocytes the incorporation half-life of EPA was 4 wk, and the concentration of EPA plateaued at 6 mo of supplementation. The authors concluded that serum may reflect intake over the past weeks, whereas erythrocytes reflect intake over the past months (19). This is the likely explanation for our findings of higher correlations for n–3 fatty acids of marine origin and trans fatty acids in erythrocytes than in plasma.
The half-life of fatty acids in adipose tissue was estimated to be 680 d (16), which suggested that adipose tissue better reflected the long-term intake of fatty acids than did other specimens. One controlled dietary trial showed evidence supporting this hypothesis (19). Similarly, Baylin et al (20) showed stronger correlations for n–6 fatty acids, linolenic acid, and trans fatty acids in adipose tissue than in plasma or whole blood (20). In a Norwegian population whose intake of n–3 fatty acids of marine origin was high, total serum reflected long-term intake as strongly as did adipose tissue (8). In another study, the n–3 fatty acid concentration in erythrocytes was even better correlated with intake than was the n–3 fatty acid concentration in adipose tissue (25). The much lower concentrations of n–3 fatty acids in adipose tissue in this population may have resulted in less precise measurements and may partly explain this finding (25). In a group of Nurses' Health Study participants, Garland et al (17) used the same FFQs that were used in the present study to compare fatty acids in adipose tissue with fatty acid intake. The correlations for trans fatty acids reported in that study (r = 0.43 for 18:1 trans isomers, r = 0.22 for 18:2 trans isomers, and r = 0.40 for total trans fatty acids) were not significantly different from the correlations of trans fatty acids in erythrocytes in the current study. In another study conducted in women living in the Boston, Massachusetts, area who had characteristics similar to our study participants, London et al (30) showed a correlation coefficient of 0.48 for n–3 fatty acids of marine origin (EPA plus DHA) in adipose tissue and intake derived from the same FFQs. This correlation was not different from the correlations for EPA and DHA in erythrocytes observed in our study (rs = 0.38 for EPA and rs = 0.56 for DHA). These observations indicated that erythrocytes might reflect the long-term intake of n–3 fatty acids of marine origin and of trans fatty acids to an extent comparable with that of adipose tissue. In consideration of the limited availability of fat aspirate samples in epidemiologic studies, erythrocytes may warrant more attention as a medium to reflect long-term fatty acid intake.
The PUFAs in plasma and erythrocytes tended to be most strongly correlated with intake estimated by the 1990 FFQ. The correlations were slightly weakened when intake was estimated by the 1984 FFQ. In the current study the FFQs inquired about intakes over the previous year. If the assumption were true that intake of fatty acids is constant over time, we should have observed similar correlations over the previous 6 y. The mild gradient of correlations we observed over time may have been due to changes in dietary intake over time. In general, intraindividual variation in intake weakens the correlation coefficients between biomarkers and diet. One way to minimize this type of measurement error is to obtain long-term average intakes from repeated dietary measurements over previous years. In our study, the average intakes of polyunsaturated and trans fatty acids calculated from the 1984, 1986, and 1990 FFQs correlated better with biomarker concentrations than did those calculated from the 1990 FFQ alone. In a prospective analysis of dietary fatty acids and coronary heart disease risk, we showed that the analyses using repeated measures of diet yielded stronger associations than did those using only baseline diet or most recent diet (39).
Perfect correlations between fatty acids in tissues and intakes measured by FFQs are unrealistic. Persons with the same fatty acid intake may not have the same concentration of that fatty acid in tissues (4, 40). Nondietary factors, such as absorption, metabolism, and genetic and lifestyle determinants, can affect fatty acid concentrations in human tissues (1). In addition to the differences in fatty acid composition in blood fractions (plasma and erythrocytes) and dietary fat, we also observed considerable differences between these 2 blood fractions. These differences may have been due to the endogenous synthesis of some fatty acids, the different physiologic functions of certain fatty acids in different blood fractions, or different roles of these fractions as vehicles for fatty acid transport (1). However, because measurement errors of the biomarkers and the FFQs are independent, the moderate-to-strong correlation coefficients observed in the current study support the relative ability of biomarkers to reflect usual fatty acid intake.
In summary, we observed moderate-to-strong correlations between n–3 fatty acids of marine origin and trans fatty acids in erythrocytes and corresponding intakes, especially when multiple FFQs in preceding years were used to calculate long-term usual dietary intakes. Correlations with intake were generally stronger for erythrocytes than for plasma, which suggests that erythrocytes may be considered as an alternative to adipose tissue for these measurements.
ACKNOWLEDGMENTS
We are indebted to Frank Sacks and Walter Willett for their valuable comments and for editing this manuscript and to Bernard Rosner for providing critical statistical guidance.
The authors' responsibilities were as follows—QS: analyzed the data and drafted the manuscript; JM and SEH: designed the study and directed the blood sample assays; HC: designed the study, assayed the biomarkers, and prepared the data; FBH: designed the study's analytic strategy and supervised the data analysis; and all authors: contributed to the revision of the manuscript. None of the authors had any financial or personal conflict of interest to disclose.
REFERENCES