Plasma phospholipid and dietary fatty acids as predictors of type 2 diabetes: interpreting the role of linoleic acid 1 3

Transcription

1 Plasma phospholipid and dietary fatty acids as predictors of type 2 diabetes: interpreting the role of linoleic acid 1 3 Allison M Hodge, Dallas R English, Kerin O Dea, Andrew J Sinclair, Maria Makrides, Robert A Gibson, and Graham G Giles ABSTRACT Background: Dietary fatty acids may be associated with diabetes but are difficult to measure accurately. Objective: We aimed to investigate the associations of fatty acids in plasma and diet with diabetes incidence. Design: This was a prospective case-cohort study of 3737 adults aged y. Fatty acid intake (/kj) and plasma phospholipid fatty acids (%) were measured at baseline, and diabetes incidence was assessed by self-report 4 y later. Logistic regression excluding (model 1) and including (model 2) body mass index and waist-hip ratio was used to calculate odds ratios (ORs) for plasma phospholipid and dietary fatty acids. Results: In plasma phospholipid, positive associations with diabetes were seen for stearic acid [OR model 1, highest versus lowest quintile: 4.14 (95% CI: 2.65, 6.49), P for trend ] and total saturated fatty acids [OR model 1: 3.76 (2.43, 5.81), P for trend ], whereas an inverse association was seen for linoleic acid [OR model 1: 0.22 (0.14, 0.36), P for trend ]. Dietary linoleic [OR model 1: 1.77 (1.19, 2.64), P for trend 0.002], palmitic [OR model 1: 1.65 (1.12, 2.43), P for trend 0.012], and stearic [OR model 1: 1.46 (1.00, 2.14), P for trend 0.030] acids were positively associated with diabetes incidence before adjustment for body size. Within each quintile of linoleic acid intake, cases had lower baseline plasma phospholipid linoleic acid proportions than did controls. Conclusions: Dietary saturated fat intake is inversely associated with diabetes risk. More research is required to determine whether linoleic acid is an appropriate dietary substitute. Am J Clin Nutr 2007;86: KEY WORDS Diabetes, dietary fats, fatty acids, phospholipids, prospective study INTRODUCTION Our understanding of associations between dietary fats and type 2 diabetes is limited by the accuracy of measurement of habitual fat intake. To overcome this, 3 prospective studies measured fatty acid biomarkers, but none of those studies simultaneously measured diet (1 3). Consistent observations across the 3 studies were that the incidence of diabetes is associated with lower proportions of linoleic acid in plasma phospholipids (1, 3), in cholesterol esters (1), or as serum esterified and nonesterified linoleic acid (2). Two of the studies also linked higher proportions of saturated fatty acids (SFAs) with diabetes (1, 3). These observations suggest that a low intake of linoleic acid may increase diabetes risk. A review of fat types and diabetes risk concluded that polyunsaturated fatty acids (PUFAs) could be beneficial (4), which is consistent with the above biomarker studies, although long-chain n 3 PUFAs may be particularly related to lower diabetes risk. Dietary SFAs are generally considered to have an adverse effect on insulin action and diabetes risk (4, 5), although the association between dietary and biomarker concentrations is not direct as the result of endogenous production (6). Our aim was to investigate prospectively associations between both plasma phospholipid and dietary fatty acids and diabetes. We specifically tested the hypotheses that SFAs would be positively associated with diabetes and that linoleic acid would be inversely associated with diabetes. We also assessed whether associations with biomarkers were similar to those with dietary intakes. SUBJECTS AND METHODS Subjects The Melbourne Collaborative Cohort Study (MCCS) recruited persons ( men) between 1990 and Persons aged y were invited; 0.7% of the participants fell outside this age range but are included in all analyses. The Cancer 1 From the Cancer Epidemiology Centre, The Cancer Council Victoria, Melbourne, Australia (AMH, DRE, and GGG); the School of Population Health, University of Melbourne, Melbourne, Australia (AMH, DRE, and GGG); the Department of Medicine, University of Melbourne, St Vincents Hospital, Melbourne, Australia (KO D); the School of Exercise and Nutrition Sciences, Deakin University, Melbourne, Australia (AJS); the Child Health Research Institute, Women s and Children s Hospital and Flinders Medical Centre, Adelaide, Australia (MM); the School of Paediatrics and Reproductive Health, University of Adelaide (MM and RAG); and the School of Agriculture, Food and Wine, University of Adelaide, Adelaide, Australia (RAG). 2 Cohort recruitment was funded by VicHealth and The Cancer Council Victoria. This study was funded by grants from the National Health and Medical Research Council (209057, ) and was further supported by infrastructure provided by The Cancer Council Victoria. 3 Reprints not available. Address correspondence to A Hodge, The Cancer Council Victoria, 1 Rathdowne Street, Carlton VIC 3053, Australia. allison.hodge@cancervic.org.au. Received August 22, Accepted for publication February 6, Am J Clin Nutr 2007;86: Printed in USA American Society for Nutrition 189

2 190 HODGE ET AL Council Victoria s Human Research Ethics Committee approved the study. Subjects gave written consent to participate and for the investigators to obtain access to their medical records. Case-cohort design We used a case-cohort design (7) for biomarker studies. Plasma phospholipid fatty acids were measured for all incident cases of diabetes and a random sample of the cohort (the subcohort), which included some randomly selected cases. We excluded participants with diabetes at baseline (self-reported or elevated plasma glucose), those who had had a heart attack or had angina before baseline, those who did not report diabetes at baseline but later reported a date of diabetes diagnosis before baseline, those with extreme self-reported energy intakes ( 1st percentile and 99th percentile), and those with missing values for relevant risk factors. A total of 3737 participants aged between 36 and 72 y, including 364 incident cases of type 2 diabetes, had complete data for these analyses. Baseline measurements Plasma glucose was measured by using a Kodak Ektachem analyzer (Rochester, NY). According to the World Health Organization criteria current at the time, elevated plasma glucose was defined as a concentration 7.8 mmol/l when fasting (68% of participants) and 11.1 mmol/l if not (8). Dietary data were collected with a self-administered 121-item food-frequency questionnaire specifically developed for the MCCS (9). Information on other risk factors was collected in face-to-face interviews (10 12). Height, weight, and waist and hip circumferences were measured directly; and body mass index (BMI, in kg/m 2 ) and waist-hip ratio (WHR) were calculated. Dietary fatty acid intake was calculated from the food-frequency questionnaire data by using Australian fatty acid composition data. The Australian database contains fatty acid data (g/100 g) to 2 decimal places for 1044 Australian foods that were selected as being of particular interest to the researchers who performed the assays; 11 SFAs, 7 monounsaturated fatty acids (MUFAs), 10 PUFAs, and 3 trans fatty acids are included (13). Analysis of the fatty acid composition of plasma phospholipids Blood was collected from all participants into sodium-heparin evacuated tubes, centrifuged immediately (3000 rpm, 15 min, 20 C), portioned into aliquots, and stored in liquid nitrogen. The fatty acid analysis has been described in detail elsewhere (6). Briefly, samples were realiquotted on ice under red light conditions before being refrozen and transported to the laboratory of one of us (RG) in Adelaide. Total lipids were extracted from plasma, and the extracts were separated by thin-layer chromatography into phospholipids, triacylglycerol, and cholesterol esters on silica gel plates (silica gel 60H; Merck, Darmstadt, Germany). Phospholipid fatty acid methyl esters were separated and quantified with a Hewlett-Packard (Palo Alto, CA) 5880 gas-liquid chromatograph by using a capillary column equipped with flame ionization detection and the Hewlett-Packard Chem-Station data system. Plasma insulin measurement Plasma insulin was measured in the plasma of fasted participants only (68%) by use of the AxSYM Microparticle Enzyme Immunoassay (Abbott, North Ryde, NSW, Australia). This assay has minimal cross-reactivity with proinsulin (0.016%) and no detectable cross-reactivity with C-peptide or glucagon. Ascertainment of new cases of diabetes Approximately 4 y after baseline, the participants completed a mailed, self-administered questionnaire that covered diagnosis of diabetes (10, 11). We attempted to verify with the person s nominated doctor any reports of diabetes diagnosed since baseline. Responses were available for 292 persons; 291 (84%) were confirmed as having type 2 diabetes. For the 52 persons with no response and for the 2 persons for whom doctors did not know diabetes type, we assumed a diagnosis of type 2 diabetes. Statistical analysis Means and SDs for each fatty acid in plasma phospholipid and diet were calculated by diabetes status at follow-up, and t tests were used to evaluate differences between the 2 groups. Age, country of birth, sex, physical activity score, 5-y weight change, education level, smoking, BMI, WHR, and family history of diabetes were considered as potential confounders. Weight change, education, and smoking were not associated with diabetes in the subcohort and were not included in subsequent models. Logistic regression models were computed first with age, sex, country of birth, physical activity, family history of diabetes, and alcohol intake (model 1) and then with all confounders plus BMI and WHR (model 2) for quintiles (based on the distributions in the subcohort) of plasma phospholipid fatty acid proportions and dietary fatty acids expressed as energy density. The following fatty acids and classes were analyzed: total SFAs, 15:0, 16:0, 18:0, total MUFAs, 16:1n 7, 18:1 n 9, total PUFAs, total n 6 fatty acids, 18:2n 6, 20:3n 6, 20:4n 6, total n 3 fatty acids, 18:3n 3, 20:5n 3, 22:5n 3, 22:6n 3, ratio of n 6 ton 3 fatty acids, total trans fatty acids, and total conjugated linoleic acid (plasma only). -Linolenic acid was not included because its measurement in plasma was not considered to be reliable owing to its extremely low concentrations in a substudy (6). Fatty acid ratios in plasma phospholipid reflecting product-precursor ratios of elongase and desaturase enzymes were also calculated and examined in the same way as the fatty acids. Additional analyses were performed with adjustment for insulin in subjects who were fasting at baseline. An interaction term for dietary linoleic acid and insulin was tested in model 1. Interactions between dietary linoleic acid and both BMI and age were also tested in view of the observations of van Dam (14), which showed an inverse association between linoleic acid and diabetes in younger, leaner persons. Fatty acid reliability study The reliability of plasma phospholipid and dietary fatty acid composition was assessed as described previously (6). RESULTS The baseline characteristics of the incident diabetes cases and controls in the subcohort are shown in Table 1. Persons who developed diabetes tended to be older, more obese, less active, and more likely to have a family history of diabetes and to originate from southern Europe.

3 FATTY ACIDS AND DIABETES 191 TABLE 1 Baseline characteristics of persons with (cases) or without (controls) incident diabetes, Melbourne Collaborative Cohort Study, Controls (n 3391) Cases (n 346) Age (y) BMI (kg/m 2 ) Waist-hip ratio Alcohol intake (g/d) Women [n (%)] 1918 (56.6) 170 (49.1) Country of birth Australia [n (%)] 2425 (71.5) 154 (44.5) United Kingdom [n (%)] 256 (7.6) 29 (8.4) Italy [n (%)] 397 (11.7) 90 (26.0) Greece [n (%)] 313 (9.2) 73 (21.0) Gained weight in past 5y[n (%)] 793 (23.4) 118 (34.1) Family history of diabetes [n (%)] 598 (17.6) 118 (34.1) Current smoker [n (%)] 371 (10.9) 46 (13.3) Primary education only [n (%)] 528 (15.6) 118 (34.1) Most active [n (%)] (24.4) 48 (13.9) 1 x SD (all such values). 2 Those in the highest of 4 groups for physical activity score. Mean plasma phospholipid fatty acid percentages by follow-up diabetes status are shown in Table 2. Persons who developed diabetes had higher proportions of 18:0, total saturated fat, 16:1n 7, 20:3n 6, 20:4n 6, total n 3 fatty acids, 20:5n 3, and 22:6n 3 and lower proportions of 15:0, total polyunsaturated fat, n 6 fatty acids, 18:2n 6, n 6:n 3, trans fats, and conjugated linoleic acid at baseline than did persons who did not develop diabetes. Mean daily intakes of total fat and fatty acid classes by diabetes status at follow-up are presented in Table 3. Persons who developed diabetes had higher intakes of total fat, total monounsaturated fats, 16:1n 7, 18:1n 9, total polyunsaturated fats, n 6 fats, 18:2n 6, 20:4n 6, n 3 fats, 18:3n 3, and trans fats and a lower intake of 15:0 at baseline than did persons who did not develop diabetes. The ORs for diabetes by quintile of plasma phospholipid fatty acids relative to the lowest quintile are shown in Table 4. After adjustment for age, sex, country of birth, physical activity, family history of diabetes, and alcohol intake (model 1), inverse associations were seen for 15:0, trans fatty acids, and 18: 2n 6. Positive associations were observed for 18:0, total SFAs, 16:1n 7, and 20:3n 6. After further adjustment for body size, these associations were attenuated but still highly significant. The ORs for elongation and desaturation product-precursor ratios are presented in Table 5. Strong positive associations were observed for stearoyl-coa desaturase (ratio of 16:1n 7 to 16:0) and elongase (ratio of 20:3n 6 to 18:2n 6), whereas inverse associations were seen for 5 desaturase (ratio of 20:4n 6 to 20:3 n 6) and the ratio of 18:1n 9 to 18:0, which also reflects stearoyl-coa desaturase-1. The ORs by quintile of dietary fatty acids are shown in Table 6. For model 1, the OR for the top quintile of dietary fat intake was elevated compared with the lowest. Both 16:0 and 18:0, but not total SFAs, were associated with higher risk in model 1. 16:1n 7 showed a weak positive association with diabetes. Positive associations were seen for 18:1n 9, MUFAs, 18:2n 6, total n 6 fatty acids, PUFAs, and 18:3n 3 in model 1; after adjustment for body size, however, these were no longer significant. The ratio of n 6ton 3 fatty acids showed a positive TABLE 2 Baseline plasma phospholipid fatty acid percentages of persons with (cases) or without (controls) incident diabetes after 4yoffollow-up, Melbourne Collaborative Cohort Study, Fatty acid Controls Cases P 2 % Saturated fatty acids : : : Monounsaturated fatty acids :1n :1n Polyunsaturated fatty acids Total n :2n :3n :4n Total n :3n :5n :5n :6n n 6:n Total trans Total CLA All values are x SD, n CLA, conjugated linoleic acid. 2 t Test with unequal variance.

4 192 HODGE ET AL TABLE 3 Baseline intakes of dietary fat and fatty acids of persons with (cases) or without (controls) incident diabetes after 4yoffollow-up, Melbourne Collaborative Cohort Study, Fatty acid Controls Cases P 2 g/d Total fat Saturated fatty acids : : : Monounsaturated fatty acids :1n :1n Polyunsaturated fatty acids Total n :2n :3n :4n Total n :3n :5n :5n :6n n 6:n Total trans All values are x SD; n t Test with unequal variance. association of borderline significance that was not attenuated by adjustment for body size. We observed previously that plasma and dietary linoleic acid were correlated in the subcohort (6); thus, our observations here that plasma phospholipid linoleic acid was associated with diabetes risk in the opposite direction (negative) to the association observed for diet (positive) is unlikely to be explained by measurement error in the dietary estimates. To further explore how these opposite associations could occur, we plotted plasma phospholipid linoleic acid for quintiles of dietary intake for persons who developed diabetes and for those who did not. We also fitted a linear regression model with plasma phospholipid linoleic acid as the dependent variable with quintiles of dietary linoleic acid and case-control status at follow-up as the independent variables. An interaction between dietary linoleic acid and diabetes status was also included to test for different slopes in cases and controls. As indicated in Figure 1, for both persons who developed diabetes and for those who did not, there were similar linear associations between dietary and plasma phospholipid linoleic acid; the interaction term was not significant (P 0.6 when quintile medians were modeled). However, within each quintile of reported dietary intake, persons who developed diabetes had lower mean plasma phospholipid linoleic acid proportions; the mean difference, estimated from the regression, was 1.8 (95% CI: 1.4, 2.1) percentage points. Models were recomputed for the fasting subgroup (n 2324, cases 224). There was a weak interaction between plasma insulin and linoleic acid intake (P for interaction 0.09), and the association between dietary linoleic acid and diabetes risk was most apparent in persons with plasma insulin concentrations at or above the median value ( 5.3 pmol/l). The OR for quintile 5 versus quintile 1 was 1.81 (95% CI: 1.01, 3.23; P for trend 0.02), compared with lower values (P for trend 0.30). There was no significant difference in the associations of dietary linoleic acid with incident diabetes across strata of age (P for interaction 0.292) or BMI (P for interaction 0.252). The 12-mo reliability coefficients (within-subject variation) for plasma phospholipid fatty acids ranged from 0.23 for palmitoleic acid to 0.89 for palmitic acid and from 0.32 for dietary palmitoleic acid to 0.49 for dietary oleic acid (6). DISCUSSION We found positive associations between the incidence of diabetes and SFAs in plasma phospholipid and diet. Plasma phospholipid linoleic acid was inversely, and dietary linoleic acid was positively, associated with diabetes risk. Persons who developed diabetes had lower plasma phospholipid linoleic acid proportions for each quintile of linoleic acid intake than did persons without diabetes. Our study has several strengths: it included both men and women and had a good follow-up rate. The trivial differences between those who did and did not complete follow-up mean that response bias was likely to be minimal. Plasma glucose was measured at baseline, and persons with elevated glucose were excluded. Although we did not attempt to verify negative reports of diabetes, participants were accurate in reporting diagnoses. Because we did not screen for diabetes at follow-up, some incident cases could be missed, but unless screening was associated with plasma phospholipid fatty acids, which is unlikely, this would not introduce bias (15). The use of a single plasma phospholipid fatty acid measurement to reflect long-term status is a potential limitation, but 12-mo reliability coefficients (withinsubject variation) for a small group from the MCCS suggest that reliability was reasonable for the fatty acids studied (6).

5 FATTY ACIDS AND DIABETES 193 TABLE 4 Odds ratios (and 95% CIs) for incident diabetes by quintile of plasma phospholipid fatty acid proportions, Melbourne Collaborative Cohort Study, Fatty acid and model Quintile 2 2 Quintile 3 Quintile 4 Quintile 5 P for trend 15:0 Model (0.54, 0.99) 0.36 (0.25, 0.52) 0.32 (0.22, 0.47) 0.26 (0.17, 0.40) Model (0.60, 1.15) 0.46 (0.31, 0.68) 0.45 (0.30, 0.68) 0.40 (0.26, 0.63) :0 Model (0.79, 1.65) 1.43 (0.99, 2.06) 1.37 (0.94, 2.00) 1.32 (0.89, 1.94) Model (0.74, 1.65) 1.44 (0.97, 2.13) 1.33 (0.89, 1.99) 1.27 (0.84, 1.92) :0 Model (0.87, 2.45) 1.99 (1.22, 3.23) 2.12 (1.32, 3.43) 4.14 (2.65, 6.49) Model (0.83, 2.45) 1.70 (1.02, 2.83) 1.34 (0.81, 2.23) 2.25 (1.39, 3.62) SFA Model (0.90, 2.42) 2.43 (1.53, 3.85) 3.48 (2.23, 5.42) 3.76 (2.43, 5.81) Model (0.66, 1.89) 1.62 (0.99, 2.63) 2.02 (1.26, 3.22) 1.88 (1.19, 2.99) trans Fatty acids Model (0.50, 0.95) 0.39 (0.27, 0.58) 0.38 (0.25, 0.57) 0.20 (0.12, 0.32) Model (0.53, 1.06) 0.49 (0.32, 0.74) 0.51 (0.33, 0.80) 0.30 (0.17, 0.51) :1n 7 Model (1.18, 2.87) 2.32 (1.48, 3.63) 3.43 (2.21, 5.32) 5.61 (3.65, 8.60) Model (1.03, 2.60) 1.79 (1.12, 2.85) 2.52 (1.60, 3.99) 3.55 (2.27, 5.54) :1n 9 Model (0.69, 1.43) 0.94 (0.65, 1.36) 0.63 (0.42, 0.94) 0.99 (0.69, 1.43) Model (0.81, 1.77) 1.09 (0.73, 1.63) 0.74 (0.48, 1.13) 1.19 (0.81, 1.76) MUFA Model (0.56, 1.18) 0.87 (0.60, 1.25) 0.70 (0.47, 1.02) 0.88 (0.60, 1.24) Model (0.56, 1.25) 0.89 (0.60, 1.33) 0.75 (0.50, 1.13) 0.93 (0.50, 1.13) CLA Model (0.50, 1.01) 0.92 (0.64, 1.33) 0.80 (0.54, 1.21) 0.84 (0.56, 1.27) Model (0.43, 0.92) 0.78 (0.52, 1.17) 0.69 (0.44, 1.06) 0.73 (0.47, 1.14) :2n 6 Model (0.34, 0.66) 0.53 (0.38, 0.74) 0.49 (0.34, 0.70) 0.22 (0.14, 0.36) Model (0.35, 0.71) 0.64 (0.45, 0.94) 0.65 (0.44, 0.94) 0.33 (0.20, 0.56) :3n 6 Model (1.09, 3.95) 2.69 (1.42, 5.04) 4.77 (2.66, 8.58) 9.65 (5.48, 16.97) Model (0.90, 3.39) 1.68 (0.88, 3.21) 2.93 (1.60, 5.39) 4.48 (2.48, 8.08) :4n 6 Model (0.94, 2.10) 1.42 (0.96, 2.09) 1.25 (0.85, 1.85) 1.32 (0.90, 1.94) Model (0.92, 2.19) 1.39 (0.91, 2.12) 1.43 (0.94, 2.17) 1.40 (0.93, 2.11) Total n 6 Model (0.58, 1.13) 0.84 (0.59, 1.18) 0.77 (0.53, 1.13) 0.66 (0.44, 0.98) Model (0.58, 1.18) 0.96 (0.66, 1.39) 0.90 (0.61, 1.34) 0.78 (0.51, 1.19) :3n 3 Model (0.79, 1.63) 0.94 (0.65, 1.36) 0.97 (0.67, 1.40) 0.88 (0.61, 1.27) Model (0.87, 1.88) 1.08 (0.72, 1.60) 1.10 (0.74, 1.63) 1.07 (0.72, 1.59) :5n 3 Model (0.68, 1.53) 1.20 (0.81, 1.77) 0.99 (0.67, 1.48) 1.38 (0.95, 2.01) Model (0.68, 1.62) 1.24 (0.81, 1.89) 0.95 (0.62, 1.45) 1.22 (0.81, 1.84) :5n 3 Model (0.66, 1.34) 0.76 (0.52, 1.10) 0.91 (0.64, 1.31) 0.89 (0.63, 1.28) Model (0.70, 1.49) 0.93 (0.62, 1.40) 0.99 (0.67, 1.46) 1.16 (0.79, 1.71) :6n 3 Model (0.67, 1.54) 1.41 (0.96, 2.07) 1.26 (0.85, 1.87) 1.30 (0.88, 1.92) Model (0.63, 1.52) 1.41 (0.93, 2.14) 1.21 (0.79, 1.85) 1.45 (0.95, 2.20) Total n 3 Model (0.87, 1.94) 1.08 (0.72, 1.62) 1.31 (0.87, 1.93) 1.28 (0.87, 1.89) Model (0.94, 2.22) 1.04 (0.67, 1.61) 1.38 (0.91, 2.09) 1.40 (0.92, 2.13) PUFA Model (0.70, 1.36) 0.83 (0.58, 1.19) 0.69 (0.47, 1.01) 0.79 (0.54, 1.15) Model (0.64, 1.31) 1.00 (0.68, 1.46) 0.93 (0.62, 1.40) 0.99 (0.66, 1.48) n 6:n 3 Model (0.75, 1.45) 0.80 (0.56, 1.16) 0.94 (0.66, 1.35) 0.81 (0.55, 1.20) Model (0.74, 1.50) 0.76 (0.51, 1.12) 0.98 (0.67, 1.44) 0.82 (0.54, 1.24) n SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; CLA, conjugated linoleic acid; PUFA, polyunsaturated fatty acids. 2 Quintile 1 was the referent. 3 Odds ratios by logistic regression; model 1 was adjusted for age, sex, country of birth, family history of diabetes, physical activity, and alcohol intake. 4 Odds ratios by logistic regression; model 2 was adjusted for age, sex, country of birth, family history of diabetes, physical activity, alcohol intake, BMI, and waist-hip ratio.

6 194 HODGE ET AL TABLE 5 Odds ratios (and 95% CIs) for incident diabetes by quintile of plasma phospholipid fatty acid ratios, Melbourne Collaborative Cohort Study, Fatty acid and model Quintile 2 2 Quintile 3 Quintile 4 Quintile 5 P for trend 16:1n 7/16:0 Model (1.33, 3.26) 2.32 (1.46, 3.67) 3.44 (2.19, 5.41) 6.70 (4.35, 10.32) Model (1.13, 2.88) 1.89 (1.17, 3.04) 2.55 (1.59, 4.08) 4.12 (2.63, 6.45) :1n 9/18:0 Model (0.55, 1.05) 0.57 (0.40, 0.82) 0.55 (0.38, 0.79) 0.49 (0.34, 0.71) Model (0.61, 1.25) 0.80 (0.55, 1.18) 0.71 (0.48, 1.05) 0.73 (0.49, 1.09) :3n 6/18:2n 6 Model (1.07, 4.78) 3.46 (1.71, 7.02) 7.23 (3.71, 14.11) (6.24, 23.17) Model (0.81, 3.74) 2.35 (1.14, 4.84) 4.12 (2.07, 8.20) 5.39 (2.74, 10.62) :4n 6/20:3n 6 Model (0.48, 0.89) 0.52 (0.38, 0.73) 0.35 (0.24, 0.51) 0.24 (0.16, 0.37) Model (0.59, 1.15) 0.75 (0.52, 1.07) 0.57 (0.38, 0.85) 0.46 (0.38, 0.85) :6n 3/22:5n 3 Model (0.60, 1.34) 1.25 (0.85, 1.82) 1.30 (0.89, 1.89) 1.12 (0.76, 1.66) Model (0.55, 1.32) 1.30 (0.87, 1.95) 1.32 (0.88, 1.97) 1.02 (0.68, 1.55) n Quintile 1 was the referent. 3 Odds ratios by logistic regression; model 1 was adjusted for age, sex, country of birth, family history of diabetes, physical activity, and alcohol intake. 4 Odds ratios by logistic regression; model 2 was adjusted for age, sex, country of birth, family history of diabetes, physical activity, alcohol intake, BMI, and waist-hip ratio. Our findings for plasma phospholipid fatty acids are generally consistent with the results of previous prospective studies. Positive associations between palmitoleic, dihomo- -linolenic, and SFAs in cholesterol esters and diabetes incidence, and an inverse association for linoleic acid, were reported previously (1, 3). For plasma phospholipids, positive associations were observed for SFAs, palmitic acid, and stearic acid, with a weakly inverse association for linoleic acid (1). An inverse association between serum linoleic acid and glucose intolerance was also reported previously (2). Fatty acid ratios that may indicate an increased activity of stearoyl-coa desaturase and 6 desaturase were positively associated with metabolic syndrome in men, whereas 5 desaturase activity showed an inverse association (16), which is consistent with our observations for diabetes incidence. The fatty acid composition of structural membrane lipids modifies insulin action in skeletal muscle, and SFAs are associated with insulin resistance (17), which is consistent with our findings that plasma phospholipid stearic acid and total SFAs are associated with increased diabetes risk. The results for dietary SFAs were consistent with those for plasma phospholipid SFAs and with the results of other studies (4). Although dietary intake of SFAs was not correlated with plasma phospholipid SFAs in the MCCS (6), our results suggest that reducing intakes of both palmitic and stearic acids could be beneficial. Short-term feeding studies show that SFA concentrations in platelet phospholipids can be changed by changing intakes of either palmitic or stearic acids (18, 19). The KANWU study showed that plasma phospholipid SFA concentrations could be reduced by reducing SFA intake and increasing MUFA intake, with a corresponding improvement in insulin sensitivity (20). In contrast with the case for other SFAs, our data show an inverse association between plasma phospholipid pentadecanoic acid and diabetes incidence. Pentadecanoic acid is a marker of milk fat intake (6, 21, 22), and evidence exists that milk fat is inversely associated with metabolic risk factors (22) and reduced incidence of the insulin resistance syndrome (23) and diabetes (24, 25). How can the opposite associations with risk of diabetes seen for diet and plasma phospholipid linoleic acid be reconciled (ie, how is it possible for cases to report higher intake but have lower plasma phospholipid proportions than controls)? Differential reporting of intake by those who subsequently developed diabetes would have meant that the slopes of the regression of plasma phospholipid linoleic acid on diet linoleic acid differed according to diabetes status at follow-up, but this was not the case. Thus, differential measurement error is an unlikely explanation. Nondifferential measurement error of linoleic acid in both diet and plasma would have attenuated the associations, in which case the conflicting associations are likely to be greater than what we observed. Our data showed clearly that for any level of dietary linoleic acid, cases had lower plasma phospholipid linoleic acid proportions than did controls, which suggests that some metabolic difference may exist in persons with pre-diabetes that explains the conflicting results. The relatively low proportions of plasma phospholipid linoleic acid in persons who went on to develop diabetes appeared to be balanced by higher proportions of some longer, less saturated metabolites of linoleic acid. There is evidence that 6 desaturase activity (which catalyzes the metabolism of 18:2n 6 to 18:3n 6) may be modified by insulin when insulin is low (3, 26). If 6 desaturase activity is higher at higher insulin concentrations (16), this could contribute to the relatively low proportion of linoleic acid and relatively high proportion of dihomo- -linolenic acid (via -linolenic acid) seen in persons who developed diabetes and who were likely to be relatively hyperinsulinemic (27, 28). We found no evidence that arachidonic acid, the next step in n 6 fatty acid metabolism after dihomo- -linolenic acid, was associated with diabetes incidence. Although MUFA have tended to be considered healthy fats, and in the KANWU study had a beneficial effect on insulin sensitivity (20), positive associations were seen between dietary MUFA intake and type 2 diabetes in our and other studies (14, 29, 30). This may reflect the fact that MUFAs in many Western

7 FATTY ACIDS AND DIABETES 195 TABLE 6 Odds ratios (and 95% CIs) for incident diabetes by quintile of dietary fatty acids, Melbourne Collaborative Cohort Study, Fatty acid and model Quintile 2 2 Quintile 3 Quintile 4 Quintile 5 P for trend Total fat Model (0.85, 1.89) 1.24 (0.841, 1.85) 1.22 (0.81, 1.83) 1.59 (1.08, 2.33) Model (0.82, 1.93) 1.15 (0.75, 1.75) 1.05 (0.66, 1.57) 1.12 (0.76, 1.73) :0 Model (0.76, 1.51) 1.13 (0.79, 1.60) 0.93 (0.64, 1.36) 0.75 (0.64, 1.36) Model (0.85, 1.76) 1.13 (0.78, 1.65) 1.06 (0.71, 1.58) 0.82 (0.54, 1.25) :0 Model (0.88, 1.95) 1.32 (0.89, 1.95) 1.45 (0.98, 2.14) 1.65 (1.12, 2.43) Model (0.89, 2.06) 1.16 (0.76, 1.78) 1.24 (0.82, 1.90) 1.22 (0.80, 1.85) :0 Model (0.75, 1.61) 1.08 (0.74, 1.58) 1.32 (0.90, 1.93) 1.46 (1.00, 2.14) Model (0.73, 1.67) 0.98 (0.65, 1.49) 1.01 (0.67, 1.53) 1.23 (0.81, 1.85) SFA Model (0.78, 1.65) 1.19 (0.82, 1.73) 1.34 (0.92, 1.96) 1.22 (0.83, 1.81) Model (0.75, 1.68) 1.11 (0.74, 1.66) 1.11 (0.74, 1.67) 1.04 (0.68, 1.58) trans Fatty acids Model (0.62, 1.33) 0.97 (0.66, 1.43) 0.84 (0.57, 1.24) 1.19 (0.83, 1.71) Model (0.62, 1.42) 0.91 (0.59, 1.38) 0.75 (0.49, 1.14) 0.97 (0.66, 1.44) :1n 7 Model (0.44, 1.03) 1.07 (0.73, 1.57) 1.05 (0.72, 1.53) 1.33 (0.92, Model (0.36, 0.90) 0.83 (0.55, 1.26) 0.87 (0.58, 1.30) 0.88 (0.60, 1.31) :1n 9 Model (0.69, 1.60) 1.23 (0.82, 1.85) 1.27 (0.85, 1.91) 1.51 (1.03, 2.23) Model (0.571,.41) 0.97 (0.62, 1.50) 0.90 (0.58, 1.39) 1.04 (0.69, 1.59) MUFA Model (0.71, 1.67) 1.20 (0.79, 1.82) 1.54 (1.03, 2.30) 1.52 (1.02, 2.26) Model (0.58, 1.45) 0.94 (0.60, 1.48) 1.10 (0.71, 1.70) 1.04 (0.68, 1.58) :2n 6 Model (0.84, 1.87) 1.30 (0.87, 1.94) 1.67 (1.13, 2.47) 1.77 (1.19, 2.64*) Model (0.84, 1.97) 1.13 (0.74, 1.74) 1.48 (0.97, 2.25) 1.41 (0.92, 2.17) :3n 6 Model (0.60, 1.28) 0.98 (0.68, 1.41) 0.96 (0.67, 1.39) 1.16 (0.81, 1.66) Model (0.55, 1.24) 0.88 (0.60, 1.30) 0.80 (0.54, 1.19) 0.92 (0.62, 1.35) :4n 6 Model (0.98, 2.13) 0.87 (0.57, 1.32) 1.34 (0.92, 1.96) 1.07 (0.73, 1.57) Model (0.88, 2.06) 0.81 (0.52, 1.27) 1.16 (0.78, 1.75) 0.81 (0.54, 1.23) Total n 6 Model (0.83, 1.86) 1.26 (0.84, 1.88) 1.70 (1.14, 2.50) 1.70 (1.21, 2.68) Model (0.83, 1.85) 1.10 (0.71, 1.68) 1.49 (0.98, 2.27) 1.42 (0.93, 2.18) :3n 3 Model (0.86, 1.90) 1.00 (0.66, 1.51) 1.33 (0.90, 1.96) 1.49 (1.02, 2.19) Model (0.70, 1.64) 0.89 (0.57, 1.38) 1.01 (0.66, 1.55) 1.14 (0.75, 1.73) :5n 3 Model (0.67, 1.40) 0.82 (0.56, 1.19) 0.94 (0.67, 1.40) 0.92 (0.67, 1.40) Model (0.68, 1.49) 0.73 (0.49, 1.10) 0.86 (0.62, 1.34) 0.68 (0.62, 1.34) :5n 3 Model (0.74, 1.55) 0.84 (0.57, 1.23) 0.99 (0.69, 1.44) 0.96 (0.69, 1.44) Model (0.74, 1.62) 0.76 (0.50, 1.15) 0.97 (0.65, 1.45) 0.81 (0.54, 1.21) :6n 3 Model (0.77, 1.62) 0.79 (0.54, 1.18) 1.16 (0.81, 1.67) 0.95 (0.69, 1.45) Model (0.75, 1.67) 0.69 (0.45, 1.05) 1.07 (0.72, 1.57) 0.77 (0.52, 1.16) Total n 3 Model (0.89, 1.98) 1.28 (0.86, 1.91) 1.06 (0.71, 1.60) 1.44 (0.97, 2.13) Model (0.72, 1.69) 1.06 (0.69, 1.63) 0.88 (0.57, 1.36) 0.97 (0.63, 1.48) PUFA Model (0.79, 1.75) 1.13 (0.75, 1.69) 1.67 (1.13, 2.45) 1.70 (1.14, 2.52) Model (0.78, 1.81) 0.91 (0.59, 1.41) 1.46 (0.97, 2.21) 1.29 (0.84, 1.97) n 6:n 3 Model (0.89, 1.84) 1.43 (0.99, 2.07) 1.35 (0.91, 1.98) 1.51 (1.02, 2.23) Model (0.90, 1.96) 1.42 (0.96, 2.12) 1.43 (0.94, 2.18) 1.56 (1.03, 2.36) n SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids. 2 Quintile 1 was the referent. 3 Odds ratios by logistic regression; model 1 was adjusted for age, sex, country of birth, family history of diabetes, physical activity, and alcohol intake. 4 Odds ratios by logistic regression; model 2 was adjusted for age, sex, country of birth, family history of diabetes, physical activity, alcohol intake, BMI, and waist-hip ratio.

8 196 HODGE ET AL FIGURE 1. Distribution of plasma phospholipid linoleic acid percentages by quintile (Q) of linoleic acid intake at baseline in persons who remained healthy (controls) or who developed diabetes at follow-up 4 y later (cases). In a model with plasma linoleic acid percentage as the dependent variable and dietary linoleic acid modeled as the median of quintiles, there was no interaction (P 0.6) between diabetes status at follow-up and linoleic acid intake. The proportion of plasma linoleic acid was higher by (95% CI: 0.020, 0.025) for each increment in dietary intake and lower by (95% CI: 0.015, 0.021) in persons who developed diabetes than in those who did not. countries are derived largely from meat and hence are correlated with SFAs (4, 30) or that benefits relating to oleic acid, the principal dietary MUFA, are attributable to other components of olive oil or characteristics of a diet high in olive oil (31). Both plasma phospholipid and dietary 16:1n 7 were positively associated with diabetes risk, as was the ratio of 16:1n 7 to 16:0, which reflects stearoyl-coa desaturase activity (16). On the other hand, the ratio of 18:1n 9 to 18:0, which also reflects stearoyl-coa desaturase activity, was inversely related with diabetes incidence. In Swedish men, only the ratio of 16:1 n-7 to 16:0, but not that of 18:1n 9 to 18:0, was associated with development of the metabolic syndrome over 20 y of follow-up (16). An increased ratio of 16:1n 7 to 16:0 has been observed in other insulin-resistant states, but it is not known whether the association is a causal one (32). Recent studies in mice suggest that a lack of stearoyl-coa desaturase prevents dietary induced obesity and suggest this enzyme as a target for antiobesity drugs (33). Avoidance of saturated fats and widespread adoption of PUFA-rich fats and oils has led to relatively high intakes of n 6 fats and ratios of n 6ton 3 fatty acids of around 10:1 or higher in the United States and 9:1 in the MCCS. Such ratios are in contrast with the recommended ratio of 4:1, which is based on estimates from traditional diets (34). The ratio of n 6 ton 3 fatty acids may contribute to insulin resistance (17, 35) and a range of other health conditions (36). It may be important that existing recommendations to limit SFA (37) intake do not lead to adverse effects of n 6 fatty acid consumption. Further intervention studies with insulin resistance as the outcome may clarify the most appropriate fat to substitute for SFAs (38). The inverse association we observed between plasma phospholipid trans fatty acids and diabetes conflicts with 2 studies suggesting that dietary intake of trans fats has adverse effects on insulin sensitivity and diabetes risk (4, 29). The Iowa Women s Health Study, however, found an inverse association between trans fatty acid intake and diabetes (39), and the difficulty of measuring trans fatty acid intake as the result of changes in dietary composition over time in the United States has been noted (40). Plasma phospholipid concentrations of trans fatty acids appear to reflect dietary intake (41), but we did not observe this association in the MCCS, possibly because of the very low levels of both dietary and plasma phospholipid trans fatty acids (6). The lower proportion of linoleic acid in plasma phospholipid of those who developed diabetes relative to those who did not, given the correlation between phospholipid and dietary linoleic acid, suggests that increasing linoleic acid intake could reduce diabetes risk. However, our analysis of dietary intakes does not support this. If there is something about the pre-diabetic state that causes differences in plasma phospholipid fatty acid composition, prospective biomarker studies of diabetes will require careful interpretation. Reduction in dietary saturated fat intake may reduce diabetes risk. However, it is not clear how our results regarding linoleic acid should be interpreted, and recommendations to limit SFA intake should take into account the possible importance of the n 6:n 3 PUFA ratio. More work is required to determine the most appropriate dietary substitute for SFAs. The authors contributions were as follows AMH: design of analysis, statistical analysis, writing of the manuscript; DRE: interpretation of results and revision of manuscript drafts; RAG: responsibility for conduct and supervision of plasma fatty acid analyses, input into data interpretation, and commenting on manuscript drafts; AJS: interpretation of data and commenting on manuscript drafts; MM: responsibility for conduct and supervision of plasma fatty acid analyses, input into data interpretation, and commenting on manuscript drafts; KO D: design of original MCCS and commenting on manuscript drafts; GGG: design and conduct of original MCCS and responsible for grants used to fund fatty acid assays. None of the authors had a conflict of interest. REFERENCES 1. Wang L, Folsom AR, Zheng ZJ, Pankow JS, Eckfeldt JH. Plasma fatty acid composition and incidence of diabetes in middle-aged adults: the

9 FATTY ACIDS AND DIABETES 197 Atherosclerosis Risk in Communities (ARIC) Study. Am J Clin Nutr 2003;78: Laaksonen DE, Lakka TA, Lakka HM, et al. Serum fatty acid composition predicts development of impaired fasting glycaemia and diabetes in middle-aged men. Diabet Med 2002;19: Vessby B, Aro A, Skarfors E, Berglund L, Salminen I, Lithell H. The risk to develop NIDDM is related to the fatty acid composition of serum cholesterol esters. Diabetes 1994;43: Hu FB, van Dam RM, Liu S. Diet and risk of type II diabetes: the roles of types of fat and carbohydrate. Diabetologia 2001;44: Storlien LH, Kriketos AD, Calvert GD, Baur LA, Jenkins AB. Fatty acids, triglycerides and syndromes of insulin resistance. Prostaglandins Leuk Essent Fatty Acids 1997;57: Hodge AM, Simpson JA, Gibson RA, et al. Plasma phospholipid fatty acid composition as a biomarker of habitual dietary fat intake in an ethnically diverse cohort. Nutr Metab Cardiovasc Dis 2006; Sep 5 [Epub ahead of print]. 7. MacInnis RJ, English DR. Making the most of cohort studies: choosing between case-cohort and nested case-control designs. Australasian Epidemiol 2004;11: World Health Organization. Diabetes mellitus: report of a WHO study group. Geneva, Switzerland: WHO, Ireland P, Jolley D, Giles G, et al. Development of the Melbourne FFQ: a food frequency questionnaire for use in an Australian prospective study involving an ethnically diverse cohort. Asia Pac J Clin Nutr 1994;3: Hodge AM, English DR, O Dea K, Giles GG. Glycemic index and dietary fiber and the risk of type 2 diabetes. Diabetes Care 2004;27: Hodge AM, English DR, O Dea K, Giles GG. Increased diabetes incidence in Greek and Italian migrants to Australia: how much can be explained by known risk factors? Diabetes Care 2004;27: Hodge AM, English DR, O Dea K, Giles GG. Alcohol intake, consumption pattern and beverage type, and the risk of type 2 diabetes. Diabet Med 2006;23: RMIT Lipid Research Group. Fatty acid compositional database. Xyris software. Brisbane, United Kingdom: RMIT Lipid Research Group, van Dam RM, Willett WC, Rimm EB, Stampfer M, Hu FB. Dietary fat and meat intake in relation to risk of type 2 diabetes in men. Diabetes Care 2002;25: Rothman KJ, Greenland S, eds. Modern epidemiology. 2nd ed. Philadelphia, PA: Lippincott-Raven, Warensjo E, Riserus U, Vessby B. Fatty acid composition of serum lipids predicts the development of the metabolic syndrome in men. Diabetologia 2005;48: Storlein LH, Baur LA, Kritekos AD, et al. Dietary fats and insulin action. Diabetologia 1996;39: Kelly FD, Sinclair AJ, Mann NJ, Turner AH, Abedin L, Li D. A stearic acid-rich diet improves thrombogenic and atherogenic risk factor profiles in healthy males. Eur J Clin Nutr 2001;55: Kelly FD, Sinclair AJ, Mann NJ, et al. Short-term diets enriched in stearic or palmitic acids do not alter plasma lipids, platelet aggregation or platelet activation status. Eur J Clin Nutr 2002;56: Vessby B, Uusitupa M, Hermansen K, et al. Substituting dietary saturated for monounsaturated fat impairs insulin sensitivity in healthy men and women: The KANWU Study. Diabetologia 2001;44: Wolk A, Furuheim M, Vessby B. Fatty acid composition of adipose tissue and serum lipids are valid biological markers of dairy fat intake in men. J Nutr 2001;131: Smedman AE, Gustafsson IB, Berglund LG, Vessby BO. Pentadecanoic acid in serum as a marker for intake of milk fat: relations between intake of milk fat and metabolic risk factors. Am J Clin Nutr 1999;69: Pereira MA, Jacobs DR Jr, Van Horn L, Slattery ML, Kartashov AI, Ludwig DS. Dairy consumption, obesity, and the insulin resistance syndrome in young adults: the CARDIA Study. JAMA 2002;287: Choi HK, Willett WC, Stampfer MJ, Rimm E, Hu FB. Dairy consumption and risk of type 2 diabetes mellitus in men: a prospective study. Arch Intern Med 2005;165: Liu S, Choi HK, Ford E, et al. A prospective study of dairy intake and the risk of type 2 diabetes in women. Diabetes Care 2006;29: Salomaa V, Ahola I, Tuomilehto J, et al. Fatty acid composition of serum cholesterol esters in different degrees of glucose intolerance. Metabolism 1990;39: Piche ME, Arcand-Bosse JF, Despres JP, Perusse L, Lemieux S, Weisnagel SJ. What is a normal glucose value? Differences in indexes of plasma glucose homeostasis in subjects with normal fasting glucose. Diabetes Care 2004;27: Piche ME, Lemieux S, Perusse L, Weisnagel SJ. High normal 2-hour plasma glucose is associated with insulin sensitivity and secretion that may predispose to type 2 diabetes. Diabetologia 2005;48: Salmerón J, Hu FB, Manson JAE, et al. Dietary fat intake and risk of type 2 diabetes in women. Am J Clin Nutr 2001;73: Feskens E, Virtanen S, Rasanen L, et al. Dietary factors determining diabetes and impaired glucose tolerance. A 20-year follow-up of the Finnish and Dutch cohorts of the Seven Countries Study. Diabetes Care 1995;18: Simonsen NR, Fernandez-Crehuet Navajas J, Martin-Moreno JM, et al. Tissue stores of individual monounsaturated fatty acids and breast cancer: the EURAMIC study. European Community Multicenter Study on Antioxidants, Myocardial Infarction, and Breast Cancer. Am J Clin Nutr 1998;68: Vessby B, Gustafsson IB, Tengblad S, Boberg M, Andersson A. Desaturation and elongation of fatty acids and insulin action. Ann N Y Acad Sci 2002;967: Dobrzyn A, Ntambi JM. Stearoyl-CoA desaturase as a new drug target for obesity treatment. Obes Rev 2005;6: Kris-Etherton PM, Taylor DS, Yu-Poth S, et al. Polyunsaturated fatty acids in the food chain in the United States. Am J Clin Nutr 2000; 71(suppl):179S 88S. 35. Raheja BS, Sadikot SM, Phatak RB, Rao MB. Significance of the N-6/ N-3 ratio for insulin action in diabetes. Ann N Y Acad Sci 1993;683: Connor WE. Importance of n-3 fatty acids in health and disease. Am J Clin Nutr 2000;71(suppl):171S 5S. 37. Mann JI, De Leeuw I, Hermansen K, et al. Evidence-based nutritional approaches to the treatment and prevention of diabetes mellitus. Nutr Metab Cardiovasc Dis 2004;14: McAuley K, Mann J. Nutritional determinants of insulin resistance. J Lipid Res 2006;47: Meyer KA, Kushi LH, Jacobs DR Jr, Folsom AR. Dietary fat and incidence of type 2 diabetes in older Iowa women. Diabetes Care 2001;24: Clandinin MT, Wilke MS. Do trans fatty acids increase the incidence of type 2 diabetes? Am J Clin Nutr 2001;73: Mansour MP, Li D, Sinclair AJ. The occurrence of trans-18:1 isomers in plasma lipids classes in humans. Eur J Clin Nutr 2001;55:59 64.