Kanaya AM, Wassel Fyr C, Vittinghoff E, Harris TB, Park SW, Goodpaster BH, Tylavsky F, Cummings SR. Adipocytokines and Incident Diabetes Mellitus in Older AdultsThe Independent Effect of Plasminogen Activator Inhibitor 1. Arch Intern Med. 2006;166(3):350-356. doi:10.1001/archinte.166.3.350
Copyright 2006 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2006
Adipose tissue produces “adipocytokines” of uncertain clinical significance.
We analyzed the relationships among adiposity, adipocytokines, glycemia, and incident diabetes mellitus in 2356 white and black adults aged 70 to 79 years in the Health, Aging, and Body Composition Study who did not have diabetes at baseline. We measured the levels of adipocytokines adiponectin, leptin, interleukin 6, tumor necrosis factor α, and plasminogen activator inhibitor 1. Regional fat area was determined by means of computed tomography. New diabetes was defined as a self-reported diagnosis of diabetes or as a fasting plasma glucose level of 126 mg/dL or greater (≥7.0 mmol/L) at the second, fourth, or sixth annual examination.
A total of 143 participants (14.1 cases per 1000 person-years) developed diabetes across 5 years. Visceral fat area (odds ratio [OR], 1.33; 95% confidence interval [CI], 1.10-1.60 per standard deviation increase) and body mass index (white individuals: OR, 1.65; 95% CI, 1.26-2.15 per standard deviation increase; black individuals: OR, 1.22; 95% CI, 0.99-1.51 per standard deviation increase) independently predicted incident diabetes. Adiponectin, leptin, and plasminogen activator inhibitor 1 attenuated the relationship between adiposity and diabetes. After controlling for body mass index, visceral fat, fasting glucose, fasting insulin, high-density lipoprotein cholesterol, triglycerides, and hypertension at baseline, plasminogen activator inhibitor 1 was the only adipocytokine independently associated with increased odds of diabetes (OR, 1.35; 95% CI, 1.01-1.81). Fasting glucose level at baseline remained a strong predictor of incident diabetes, whereas associations with body mass index and visceral fat were attenuated.
Adipocytokines and glycemia partially account for the relationship between adiposity and risk of type 2 diabetes due to adiposity. Plasminogen activator inhibitor 1 may be a useful predictor of diabetes in addition to measurements of body fat.
Obesity is a well-established risk factor for type 2 diabetes mellitus. However, abdominal adiposity is more strongly associated with diabetes than is overall obesity measured by means of body mass index (BMI) (calculated as weight in kilograms divided by the square of height in meters).1,2 There is some controversy over which abdominal adipose tissue depots are most closely linked to metabolic abnormalities because visceral adipose tissue,3- 5 subcutaneous abdominal fat,5- 7 and intermuscular fat5,7,8 have all been correlated with insulin resistance, dyslipidemia, and diabetes.
In the past decade, adipose tissue has been discovered to secrete several novel proteins and cytokines, collectively termed adipocytokines, which may link insulin resistance or obesity (overall or central) to type 2 diabetes. Increased levels of interleukin 6 (IL-6),9- 11 leptin,12 plasminogen activator inhibitor 1 (PAI-1),13,14 and tumor necrosis factor α (TNF-α)10 and decreased levels of adiponectin15- 21 have all been associated with the development of type 2 diabetes. However, most of these studies used surrogate measures of adiposity and evaluated the effect of a single adipocytokine on diabetes.
We evaluated the prospective association of total body fat and regional fat distribution (visceral, subcutaneous abdominal, and thigh intermuscular fat) with incident type 2 diabetes in older adults. We determined whether the relationship between adiposity and diabetes could be explained by the joint effects of these adipocytokines. We also determined whether these relationships among adiposity, adipocytokines, glycemia, and diabetes varied by sex or race.
Participants enrolled in the Health, Aging, and Body Composition (Health ABC) Study were well-functioning men and women aged 70 to 79 years who were recruited at 2 clinical sites (Pittsburgh, Pa, and Memphis, Tenn) (N = 3075). To be eligible for the study, participants had to report no difficulty in walking a quarter mile, climbing 10 steps, or performing basic activities of daily living. We performed a prospective cohort study using stored serum specimens, physical examination measurements, radiographic tests, and questionnaire data gathered at the baseline visit (April 28, 1997, to June 22, 1998). We excluded 719 individuals who had diabetes at baseline defined by self-report of a diabetes diagnosis, use of a hypoglycemic medication, a fasting plasma glucose level of 126 mg/dL or greater (≥7.0 mmol/L), or a 2-hour postchallenge plasma glucose level of 200 mg/dL or greater (≥11.1 mmol/L).
Weight was measured using a standard balance beam scale, and height was measured using a stadiometer. Total adiposity was estimated using BMI. Total fat mass was measured using whole-body dual-energy x-ray absorptiometry (QDR 4500A; Hologic Inc, Bedford, Mass). Arm, leg, trunk, and total body fat were measured using dual-energy x-ray absorptiometry, and total percentage of body fat was calculated.
Regional adiposity was measured by means of computed tomography using the Somatom Plus 4 (Siemens, Erlangen, Germany), Picker PQ 2000S (Marconi Medical Systems, Cleveland, Ohio), or 9800 Advantage (General Electric, Milwaukee, Wis) scanner and standardized protocols. Visceral fat and subcutaneous abdominal fat were measured at the L4-L5 level with participants in the supine position. Fat areas were calculated by multiplying the number of pixels of a given tissue type by the pixel area using interactive data language development software (RSI, Boulder, Colo). Visceral fat was manually distinguished from subcutaneous fat using the internal abdominal wall fascial plane. Waist circumference was measured using a flexible tape measure at the site of maximum circumference midway between the lower ribs and the anterior superior iliac spine. Intermuscular fat area and thigh subcutaneous fat area were measured using computed tomography at the mid-thigh level between the greater trochanter and the intercondyloid fossa. Intermuscular adipose tissue was distinguished from subcutaneous adipose tissue by the deep fascial plane surrounding the thigh muscles.22 Thirteen participants were missing information for total fat mass, 82 for total percentage fat, 35 for visceral fat, 53 for abdominal subcutaneous fat, 39 for thigh intermuscular fat, and 41 for thigh subcutaneous fat.
Participants underwent venipuncture at the baseline visit after an overnight fast, and serum samples were frozen at −70°C. Adiponectin and leptin were measured in duplicate by means of radioimmunoassay (Linco Research Inc, St Charles, Mo), with an intra-assay coefficient of variation of 1.8% to 3.6% for adiponectin and 3.7% to 7.5% for leptin. Both IL-6 and TNF-α were measured in duplicate using an ultrasensitive enzyme-linked immunosorbent assay (R&D Systems, Minneapolis, Minn). The lower limit of detection was less than 0.10 pg/mL for IL-6 and 0.18 pg/mL for TNF-α, with coefficients of variation of 6.3% and 16.0%, respectively. The level of PAI-1 was measured in citrated plasma samples using a 2-site enzyme-linked immunosorbent assay according to previously published methods.23 The PAI-1 assay is sensitive to free PAI-1 (latent and active) but not to PAI-1 in complex with tissue plasminogen activator, and it has a coefficient of variation of 3.5%. A total of 153 persons were missing data for leptin, 126 for IL-6, 53 for PAI-1, 158 for TNF-α, and 25 for adiponectin. The number of participants missing data for these adipocytokines did not differ by incident diabetes status or sex.
Questionnaire variables gathered at the time of the first annual visit included self-identified racial group, age, and sex. Participants reported smoking history (never, former, or current smoker), and physical activity was assessed using self-report of walking and exercise, assigning kilocalories per week to activities. Seated systolic and diastolic blood pressures were measured using a manual sphygmomanometer. Fasting lipoprotein levels (Vitros 950 analyzer; Johnson & Johnson, New Brunswick, NJ) and fasting and 2-hour postchallenge plasma glucose levels using an automated glucose oxidase reaction (YSI 2300 glucose analyzer; Yellow Springs Instrument, Yellow Springs, Ohio) were measured. Fasting serum insulin levels were measured by radioimmunoassay (Pharmacia, Uppsala, Sweden), and they were used as a surrogate measure of insulin resistance.24
Each participant was contacted by telephone every 6 months and attended an annual clinic visit in which they were asked to self-report a new diagnosis of diabetes. Fasting blood samples were obtained for measuring glucose levels at the second, fourth, and sixth annual examinations. Participant retention in the Health ABC Study was high. Only 51 participants (1.6%) were lost to follow-up during the 5-year study, and approximately 1% of visits were missed each year. Incident diabetes was defined by self-report of a new diabetes diagnosis, use of a diabetes medication, or a fasting glucose level of 126 mg/dL or greater (≥7.0 mmol/L), depending on the variables assessed at each clinical visit.
Baseline characteristics of the participants with and without incident diabetes were compared using χ2, unpaired t, or Wilcoxon tests as appropriate. We log-transformed all adipocytokine values to normalize their distribution. Spearman rank correlation coefficients between adiposity measures and adipocytokines were calculated.
Because the development of type 2 diabetes was interval censored between regularly spaced clinical visits, we used continuation ratio models fit by pooled logistic regression25 to identify independent baseline predictors of this end point. We used natural splines to examine the need for linearizing transformations of the continuous predictors: adiposity and adipocytokine levels. To compare strengths of association, the different adiposity measures and adipocytokines were each divided by their standard deviations.
In model building, we evaluated a single measure of total adiposity and added regional measures of adiposity. Because some of the regional adiposity variables were strongly correlated with each other or with measures of total adiposity, we cautiously evaluated models when 2 collinear variables were introduced. We backward selected regional fat variables that were not independently associated with incident diabetes from these models using P<.10 for significance. To examine whether adipocytokines lie on the causal pathway between obesity measures and incident diabetes, we determined whether the coefficients for adiposity were attenuated after adjustment for these adipocytokines. Primary analyses were performed using SAS version 8.2 (SAS Institute Inc, Cary, NC), whereas the exploratory analysis using natural splines was implemented using S-PLUS version 6.1 (Insightful Corp, Seattle, Wash).
A total of 143 (14.1 cases per 1000 person-years) participants developed type 2 diabetes across 5 years of follow-up. Black women had the highest rate of incident diabetes (19.9 cases per 1000 person-years), whereas white women had the lowest rate (9.5 cases per 1000 person-years). Incident diabetes rates were similar in black vs white men (17.1 vs 13.5 cases per 1000 person-years; P = .16). Participants who developed diabetes were more likely to be black; to have higher triglyceride, fasting insulin, fasting glucose, and 2-hour postchallenge glucose tolerance levels; and to have lower high-density lipoprotein cholesterol levels than individuals who did not develop diabetes (Table 1).
All measures of adiposity were significantly higher among participants who developed diabetes compared with individuals without diabetes (Table 2). We examined correlations between our measures of total and regional adiposity and the adipocytokines (Table 3). The total adiposity variables were highly correlated with each other, as expected (ρ = 0.85; P<.001 for total fat mass and BMI), but other regional fat variables were only moderately correlated (ρ = 0.33-0.55). The adipocytokines were not well correlated with each other (ρ = 0.01-0.28).
We combined a measure of total adiposity and all significant measures of regional adiposity in models that were adjusted for age, sex, and race. The 2 fat depots that were independently associated with incident diabetes were BMI (OR, 1.36; 95% CI, 1.14-1.62 per SD) and abdominal visceral fat (OR, 1.39; 95% CI, 1.16-1.66 per SD). We found that race significantly modified the relationship between BMI and incident diabetes (OR, 1.65; 95% CI, 1.26-2.15 per SD of BMI for white individuals and OR, 1.22; 95% CI, 0.99-1.51 per SD of BMI for black individuals, P = 0.04 for interaction) but that no interaction was present for visceral adiposity.
Median plasma leptin, IL-6, and PAI-1 levels were higher and the median adiponectin level was lower in incident cases of diabetes than in those without diabetes (Table 2). In separate models, only the associations between adiponectin, leptin, PAI-1, and incident diabetes remained significant after adjusting for age, sex, and race.
We constructed multivariate models to determine whether these adipocytokines explained the association between total or abdominal visceral adiposity and incident diabetes. Table 4 provides the association between visceral adiposity and BMI (stratified by race) and incident diabetes after successive adjustment for explanatory variables. First we adjusted for individual adipocytokines, then for baseline fasting plasma glucose and baseline fasting insulin values, and finally for other variables associated with the metabolic syndrome. The relationship between visceral adiposity and diabetes was attenuated with the addition of the adipocytokines (from OR, 1.33; 95% CI, 1.10-1.60 to OR, 1.21; 95% CI, 0.97-1.49 per SD). The relationship between BMI and diabetes in white individuals was also proportionately attenuated with the addition of adipocytokines (from OR, 1.65; 95% CI, 1.26-2.15 to OR, 1.42; 95% CI, 1.02-1.97 per SD). Addition of the fasting glucose, insulin, and metabolic syndrome variables did not further attenuate the relationship between visceral fat and diabetes. However, these additional variables reduced the relationship between BMI and diabetes by half (from OR, 1.42 to OR, 1.24, in white individuals).
The final multivariate model is given in Table 5. Only age and fasting glucose, fasting insulin, and PAI-1 levels remained independently associated with incident diabetes. The association between PAI-1 level and diabetes did not vary by sex or race.
Overall adiposity measured by BMI and abdominal visceral adiposity were independently associated with the development of type 2 diabetes in older men and women. The relationship between these adiposity variables and diabetes was partially explained by 3 adipocytokines: PAI-1, leptin, and adiponectin. Of these 3 adipocytokines, only PAI-1 remained independently associated with incident type 2 diabetes after further adjustment for variables associated with the metabolic syndrome. This relationship of PAI-1 to diabetes did not vary by sex or race.
Obesity is a well-recognized risk factor for diabetes, and adipose tissue deposited around the abdominal viscera has been found to be the culprit in the development of type 2 diabetes.1,26 The present study found an independent relationship between abdominal visceral fat and BMI and incident diabetes among white adults and between only visceral adiposity and diabetes among black adults. Several adipocytokines produced by adipocytes or surrounding vascular stromal cells or inflammatory cells in adipose tissue have been associated with insulin resistance and diabetes.27 We are aware of only 1 study that investigated the joint effects of adipocytokines and other inflammatory markers in the development of diabetes. In a nested case-control study28 from the Pima Indian cohort, only adiponectin remained significantly associated with decreased odds of diabetes after adjustment for IL-6, TNF-α, C-reactive protein, and other endothelial markers. This study did not include PAI-1, and the relative contribution of adiponectin apart from adiposity or visceral adiposity was not assessed in this study because the case and control groups were matched for BMI.
Plasminogen activator inhibitor 1 is a prothrombotic factor secreted from endothelial cells, mononuclear cells, hepatocytes, fibroblasts, and adipocytes29 that negatively regulates fibrinolysis by inhibiting tissue plasminogen activator. The connection between PAI-1 and cardiovascular disease is more firmly established,30,31 and PAI-1 has also been associated with insulin resistance,32- 34 obesity,35 glucose intolerance,32,36 and type 2 diabetes37,38 in cross-sectional studies. In addition, adipocytes from the visceral fat depot produces significantly more PAI-1 messenger RNA than adipocytes derived from subcutaneous tissue from the same individual.34,39 Thus, PAI-1 has been suspected to provide a link between the increased predisposition of atherosclerosis in patients and the metabolic syndrome.40 All the lifestyle41- 43 and drug treatment41,44 strategies shown to prevent diabetes also decrease plasma PAI-1 levels,45- 47 providing additional support for the possible causative role of PAI-1 in diabetes development.
Only 2 previous studies have examined the prospective association between PAI-1 and the development of type 2 diabetes. The first study13 followed 1047 middle-aged adults enrolled in the Insulin Resistance and Atherosclerosis Study (IRAS) for 5 years and found that PAI-1 increased the odds of diabetes independent of age, sex, ethnicity, smoking, BMI, insulin sensitivity, physical activity, and family history of diabetes (OR, 1.61; 95% CI, 1.20-2.16). In the IRAS and the Health ABC Study, PAI-1 concentration was measured using similar methods by the same laboratory, and median values of PAI-1 between the 2 cohorts were similar (diabetic participants: 28 ng/mL in the Health ABC Study and 24 ng/mL in the IRAS; nondiabetic participants: 19 ng/mL in the Health ABC Study and 16 ng/mL in the IRAS). The second, smaller study, the Northern Sweden MONICA (Monitoring of Trends and Determinants in Cardiovascular Disease) study,14 evaluated the effects of PAI-1 activity on future diabetes. Of 551 individuals followed for 9 years, only 15 developed diabetes. The odds of diabetes tended to increase with increasing quartile of PAI-1 activity. However, the association between PAI-1 activity and diabetes was completely attenuated after the addition of variables associated with the metabolic syndrome,14 most likely owing to the small sample for analysis. The present study supports these previous findings but further suggests that PAI-1 acts in independent pathways to visceral fat, adiponectin, and leptin in its association with diabetes.
Although much is known about the various factors that enhance PAI-1 gene transcription and expression, including insulin, glucose, triglycerides, nonesterified fatty acids, angiotensin II, glucocorticoids, growth factors, and cytokines such as TNF-α,45 the precise mechanism relating PAI-1 to diabetes onset remains illusive. Potential biological mechanisms that link PAI-1 to diabetes involve hyperinsulinemia and insulin resistance. Insulin stimulates PAI-1 release from fat and other tissues through the mitogen-activated protein kinase pathway.48 This pathway is up-regulated in insulin resistance and hyperglycemia.49 Therefore, a combination of hyperglycemia, hyperinsulinemia, insulin resistance, and visceral adiposity contributes to elevated PAI-1 levels. Plasminogen activator inhibitor 1 may promote further insulin resistance, but our results suggest that there may be another independent mechanism by which PAI-1 affects diabetes.
Although we followed a large cohort of older black and white adults with precise measurements of body fat, we were limited in the classification of incident diabetes by self-report of a new diagnosis or by fasting plasma glucose variables alone. The 2-hour oral glucose tolerance test is more sensitive than the fasting plasma glucose test for the diagnosis of diabetes, especially in elderly individuals.50- 52 However, this misclassification of incident diabetes should attenuate estimates of the true relationship between adipocytokines and incident diabetes, biasing our results toward the null. A fasting plasma glucose test and an oral glucose tolerance test were performed on all the participants at the baseline visit; thus, we accurately classified all persons with prevalent diabetes at baseline and removed them from our analysis. Second, given the older age criterion in the Health ABC Study, these results may not generalize well to younger individuals. However, in a younger population we would expect results to differ only in the magnitude and not in the overall quality or direction of association, as was shown by the results of the IRAS, a much younger population. Last, we did not have measures of free fatty acids in this cohort. Studies53,54 have found that higher concentrations of free fatty acids observed in insulin resistance can contribute to elevations in PAI-1 levels. Therefore, we cannot assess whether the observed association between PAI-1 and diabetes is independent of circulating free fatty acids levels.
The adipocytokines were measured once from specimens collected during the baseline clinical visit, and most adipocytokines were measured within a year of collection, minimizing concerns regarding long-term cryopreservation. Adiponectin was assayed in 2002-2003 from serum specimens that were frozen approximately 6 to 7 years earlier. But adiponectin is known to circulate in serum as a low-molecular-weight hexamer and a high-molecular-weight structure,55 with the high-molecular-weight form being more closely related to glucose homeostasis.56 Because we measured the total level of serum adiponectin, we may have missed important relationships with the active high-molecular-weight form and incident diabetes.
In conclusion, abdominal visceral obesity and BMI are important predictors of diabetes in older adults, but the association between measures of adiposity and diabetes is partially explained by adipocytokines. Plasminogen activator inhibitor 1 predicts type 2 diabetes independently of the metabolic syndrome and adipocytokine pathways and may improve our ability to identify individuals at risk for diabetes. However, the lack of standardization and the expense of the assay preclude PAI-1 from being a clinically useful measure currently.
Correspondence: Alka M. Kanaya, MD, 1701 Divisadero St, University of California, San Francisco, Suite 500, San Francisco, CA 94143-1732 (Alka.Kanaya@ucsf.edu).
Accepted for Publication: August 16, 2005.
Financial Disclosure: None.
Funding/Support: This study was funded in part by grants 5 K12 AR47659 and P30-AG15272 under the Resource Centers for Minority Aging Research Program, Los Angeles, Calif (Dr Kanaya) and by contracts N01-AG-6-2101, N01-AG-6-2103, and N01-AG-6-2106 from the National Institute on Aging, Bethesda, Md (Health ABC Study).
Role of the Sponsor: This study had substantial involvement from National Institute on Aging staff in data collection, analysis, and interpretation of these results.
Previous Presentation: This study was presented as an oral abstract at the European Diabetes Epidemiology Group meeting; April 25, 2004; Vietri Sul Mare, Italy.
Acknowledgment: We thank the investigators of the Health, Aging, and Body Composition Study.