Ix JH, Wassel CL, Kanaya AM, Vittinghoff E, Johnson KC, Koster A, Cauley JA, Harris TB, Cummings SR, Shlipak MG, Health ABC Study FT. Fetuin-A and Incident Diabetes Mellitus in Older Persons. JAMA. 2008;300(2):182-188. doi:10.1001/jama.300.2.182
Author Affiliations: Division of Nephrology and Hypertension, Department of Medicine, and Division of Preventive Medicine, Department of Family and Preventive Medicine, University of California, San Diego, and San Diego Veterans Affairs Healthcare System, San Diego, California (Dr Ix); Department of Epidemiology, University of Minnesota, Minneapolis (Ms Wassel); Department of Epidemiology and Biostatistics (Ms Wassel and Drs Vittinghoff and Shlipak) and Department of Medicine (Drs Kanaya, Cummings, and Shlipak), University of California, San Francisco; Department of Medicine, University of Tennessee Health Science Center, Memphis (Dr Johnson); Laboratory for Epidemiology, Demography, and Biometry, National Institute on Aging, Bethesda, Maryland (Drs Koster and Harris); Department of Epidemiology, University of Pittsburgh, Pittsburgh, Pennsylvania (Dr Cauley); Department of Medicine, California Pacific Medical Center and San Francisco Coordinating Center, San Francisco (Dr Cummings); and San Francisco Veterans Affairs Medical Center, San Francisco (Dr Shlipak).
Context Fetuin-A is a hepatic secretory protein that binds the insulin receptor and inhibits insulin action in vitro. In prior cross-sectional studies in humans, higher fetuin-A levels were associated with insulin resistance. However, the longitudinal association of fetuin-A with incident type 2 diabetes mellitus is unknown.
Objective To determine whether fetuin-A levels are associated with incident diabetes in older persons.
Design, Setting, and Participants Observational study among 3075 well-functioning persons aged 70 to 79 years. In this case-cohort study, we retrospectively measured fetuin-A levels in baseline serum among 406 randomly selected participants without prevalent diabetes, and all participants who developed incident diabetes mellitus during a 6-year follow-up (to August 31, 2005).
Main Outcome Measure Incident diabetes mellitus.
Results Incident diabetes developed in 135 participants (10.1 cases/1000 person-years). Participants with fetuin-A levels within the highest tertile (> 0.97 g/L) had an increased risk of incident diabetes (13.3 cases/1000 person-years) compared with participants in the lowest tertile (≤ 0.76 g/L) (6.5 cases/1000 person-years) in models adjusted for age, sex, race, waist circumference, body weight, physical activity, blood pressure level, fasting glucose level, high-density lipoprotein cholesterol concentration, triglyceride concentration, and C-reactive protein level (adjusted hazard ratio, 2.41; 95% confidence interval, 1.28-4.53; P = .007). The association was not affected by adipocytokine levels but was moderately attenuated by adjustment for visceral adiposity (adjusted hazard ratio of highest vs lowest tertile 1.72; 95% confidence interval, 0.98-3.05; P = .06).
Conclusion Among well-functioning older persons, serum fetuin-A is associated with incident diabetes, independent of other markers of insulin resistance.
Type 2 diabetes mellitus has become a global epidemic and the increased prevalence of obesity1 is a major contributing factor.2- 4 However, diabetes does not develop in all obese individuals and there is a strong genetic contribution to risk.5 Despite significant recent advances,6 mechanisms responsible for individual differences in clinical phenotype remain largely unknown. Recent research has identified proteins secreted from adipose tissue that regulate glucose metabolism, termed adipocytokines.7- 9 Study of these proteins has provided new insights to the biology of glucose regulation and has identified novel candidate therapeutic targets.
In contrast to adipocytokines, fetuin-A (α2–Heremans-Schmid glycoprotein [Ahsg]) is produced in hepatocytes and secreted into serum, where it is found in high concentrations.10 In vitro, fetuin-A reversibly binds the insulin receptor tyrosine kinase in muscle and fat and decreases downstream signal cascades, which results in insulin resistance in these target tissues.11- 14 In cross-sectional studies in humans, higher fetuin-A levels were associated with insulin resistance.15- 17 However, to our knowledge the temporal relationship of fetuin-A levels with incident diabetes has not been evaluated.
In this study, we measured baseline serum fetuin-A levels among participants in the Health, Aging, and Body Composition (Health ABC) Study who were diabetes-free. Participants were followed longitudinally for the occurrence of incident diabetes. We hypothesized that higher fetuin-A levels would be associated with incident diabetes and that the associations would be independent of previously recognized diabetes risk factors that are commonly ascertained in clinical practice.
At the baseline study visit (April 1997 to June 1998), Health ABC enrolled 3075 well-functioning men and women aged 70 to 79 years, recruited from a sample of Medicare beneficiaries at 2 clinical sites (Pittsburgh, Pennsylvania, and Memphis, Tennessee). Participant eligibility required self-reported ability in walking a quarter mile, climbing 10 steps, and performing basic activities of daily living. Venous blood samples were obtained after overnight (8-hour) fasts and stored at −70°C. Participants underwent a 1-day evaluation that included medical history, physical activity assessment, physical examination, and radiographic tests. All participants provided written informed consent, and the study was approved by the institutional review boards at the University of Pittsburgh and the University of Tennessee Health Science Center. In addition, the present study was approved by the Human Research Protection Program at the University of California, San Diego.
We used a case-cohort design for this study.18 Preliminary data in a separate cohort of participants of similar age demonstrated that individuals with prevalent diabetes had fetuin-A levels of 0.04 g/L higher on average, compared with individuals without diabetes (28% standardized effect size).17 Assuming a similar strength of association for incident diabetes, and knowing a priori that 135 incident diabetes cases accrued during follow-up in the Health ABC cohort, we calculated that a subcohort of 384 participants without prevalent diabetes was required to provide 80% power with a 2-sided α = .05. Because 19% of Health ABC participants had prevalent diabetes at baseline, we took a random sample of 500 participants, assumed that approximately 100 individuals (20%) would be excluded due to prevalent disease, and thereby derived the required number of nondiabetic subcohort participants required for comparison.
Thus, we randomly selected 500 participants from the parent Health ABC Study who were stratified equally in 4 sex and race strata (Figure 1). Random numbers were generated on a continuous standard uniform distribution (U [0,1]) and were assigned to each participant within each of the 4 race/sex strata. The participants were then sorted in ascending order by that assigned random number. One hundred twenty-five participants were chosen in order from each stratum until we reached the total subcohort sample size of 500.
From these 500 participants, 94 (19%) had prevalent diabetes at baseline, which was defined as any 1 or more of the following: (1) self-report of diabetes diagnosis; (2) use of hypoglycemic medications; (3) a fasting glucose level 126 mg/dL or greater; or (4) a 2-hour postchallenge plasma glucose level 200 mg/dL or greater (to convert glucose to mmol/L, multiply by 0.0555). These individuals were excluded and the resulting sample of 406 participants constituted the subcohort for this study.
All Health ABC participants were contacted by telephone every 6 months and assessed at annual clinic visits at which they were asked about interval diagnosis of diabetes. Medication use was recorded at each annual visit. Fasting blood samples were obtained for glucose measurement at the second, fourth, and sixth annual visits. Incident diabetes was defined by new diabetes diagnosis between baseline and study closeout (August 31, 2005), defined by self-report, new use of diabetes medications, or a fasting glucose level 126 mg/dL or greater. Any participant who was free of diabetes at baseline and who developed diabetes during follow-up represented a case for this study (Figure 1).
Cases could arise from either the randomly selected subcohort or the remainder of the Health ABC Study (as per typical case-cohort study design).18 We observed 135 cases of incident diabetes over the 6-year observation period (incidence rate 10.1 cases/1000 person-years). Among these cases, 23 individuals (17%) were also sampled within the subcohort. Therefore, the total study sample for this analysis consisted of 519 individuals.
Fetuin-A was measured in baseline serum among the 519 participants selected for this study in 2007 (7-9 years after the Health ABC Study enrollment) using a human fetuin-A enzyme linked immunosorbent assay kit (Epitope Diagnostics, San Diego, California). The assay used a 2-site “sandwich” technique with 2 polyclonal antibodies that bind to different epitopes of human fetuin-A. Measurements were performed at the Laboratory for Clinical Biochemistry Research, University of Vermont, Colchester. Fetuin-A was measured twice in the same sample for each participant and results were averaged. Intra-assay and interassay coefficients of variation were less than 5%. Among a 5% blind duplicate assessment at the Laboratory for Clinical Biochemistry Research, the intraindividual coefficients of variation were 13.5% to 20%.
Age, sex, and race were determined by self-report. Physical activity was assessed using self-report of walking and exercise and assigned kilocalories per week to activities. Height and weight were measured with participants wearing light-weight clothing and no shoes, and body mass index was calculated as weight in kilograms divided by height in meters squared. Waist circumference was assessed using a flexible measuring tape on bare skin at the level of maximal circumference, midway between the lower ribs and anterior superior iliac spine.
Regional adiposity was measured by computed tomography using scanners and standardized protocols of the Somatom Plus 4 (Siemens, Erlanger, Germany), the Picker PQ 2000S (Marconi Medical Systems, Cleveland, Ohio), or the 9800 Advantage (General Electric, Milwaukee, Wisconsin). Visceral and subcutaneous adiposity were measured at the midpoint between the fourth and fifth lumbar vertebrae with participants in the supine position. Fat area was calculated by multiplying the number of pixels of fat tissue by pixel area using interactive data language development software (RSI, Boulder, Colorado). Visceral adiposity was planimetrically distinguished from subcutaneous adiposity using the internal abdominal wall fascial plane.
Seated systolic and diastolic blood pressure levels were measured using manual sphygmomanometers by trained research technicians. Fasting and 2-hour postchallenge plasma glucose levels were measured by an automated glucose oxidase reaction (YSI 2300 glucose analyzer, Yellow Springs Instruments, Yellow Springs, Ohio). High-density lipoprotein cholesterol and triglyceride concentrations were measured using a Vitros 950 analyzer (Johnson & Johnson, New Brunswick, New Jersey). High-sensitivity C-reactive protein was measured by enzyme-linked immunosorbent assay (Calbiochem, San Diego) and was standardized to the World Health Organization's First International Reference Standard. Interleukin 6 was measured with the HS600 Quantikine kit (R&D Systems, Minneapolis, Minnesota), and the HSTA50 kit was used to measure tumor necrosis factor. Reliability, determined by blind duplicates, revealed mean coefficients of variation of 8.0%, 10.3%, and 15.8%, respectively.
Adiponectin and leptin were measured in duplicate by radioimmunoassay (Linco Research Inc, St Charles, Missouri) with intra-assay coefficients of variation of less than 3.6% and less than 7.5%, respectively. Plasminogen activator inhibitor 1 was measured by 2-site enzyme-linked immunosorbent assay according to previously published methods19 and had coefficients of variation of less than 3.5%. Creatinine was measured by colorimetry (Johnson & Johnson) and estimated glomerular filtration rate was calculated by the 4-variable Modification of Diet in Renal Disease equation.20
Because age-, sex-, and race-specific normal ranges of fetuin-A are unknown, and because the distribution of fetuin-A was positively skewed, we categorized participants into tertiles based on the distribution of fetuin-A in the subcohort. In companion analysis, we also evaluated fetuin-A (natural log-transformed) as a continuous predictor variable. Baseline differences in demographic and clinical variables were compared across fetuin-A tertiles by analysis of variance for continuous variables and the χ2 test or Fisher exact test for categorical variables.
To accommodate the stratified sampling design, diabetes incidence rates were estimated using all incident cases in the Health ABC cohort. To estimate the corresponding person-years at risk, we weighted follow-up times for the subcohort by the inverse of their probabilities of selection, which were known exactly and differed slightly by sex and race. The association of fetuin-A with incident diabetes was evaluated by a modified Cox regression that accounted for the case-cohort study design.21,22 Subcohort noncases and subcohort cases before their failure were weighted with the inverse of the sampling fraction (1/α). Cases outside the subcohort were not weighted before failure. All cases, either outside or inside the subcohort, were assigned a weight of 1 at the time of failure, as described by Barlow et al.21 Jackknife estimates of variance were used, which are equivalent to robust variance estimators from standard Cox models.23
Initial models were stratified on sex and race and age adjusted. Subsequent models were adjusted for measures of insulin resistance that are commonly available in clinical practice (age, sex, race, physical activity, systolic and diastolic blood pressure levels, fasting glucose level, waist circumference, body weight, high-density lipoprotein cholesterol concentration, triglyceride concentration, and C-reactive protein level). Subsequent models evaluated candidate mediators of the observed association that were selected a priori (visceral adiposity, plasminogen activator inhibitor 1, adiponectin, leptin, tumor necrosis factor, and IL-6). Multiplicative interaction terms were created to determine whether the association was similar by race, sex, and obesity status (body mass index ≥ 30).24 Proportional hazards assumptions were evaluated by formal hypothesis testing and by visual inspection of log-minus-log plots and Schoenfeld residuals vs survival time. No violations were observed. Two-tailed P values < .05 were considered statistically significant. Analyses were performed with SAS version 9.1 (SAS Institute, Cary, North Carolina) and SPlus version 6.1 (Insightful Corp, Seattle, Washington).
Among the randomly selected subcohort of 406 participants without diabetes, the mean (SD) age was 73 (3) years. The sample was 50% female and black, reflecting stratified sampling. The mean (SD) body mass index was 27 (5). The distribution of fetuin-A was positively skewed (Figure 2) with a median of 0.87 g/L (interquartile range, 0.71-1.04 g/L). When compared with participants with fetuin-A levels in the lowest third of the subcohort, participants with higher fetuin-A levels were more often white, had higher serum triglyceride levels, and had more visceral adiposity. Other measures of body composition, blood pressure levels, lipid parameters, and adipocytokine levels were similar across tertile groups. Bivariate associations were similar among participants who developed incident diabetes during follow-up (Table 1).
Between baseline and the end of the study, 60 of 406 subcohort participants died. Fetuin-A levels did not differ significantly between survivors and nonsurvivors (mean [SD] levels, 0.93 [0.40] and 0.96 [0.38 g/L], respectively; P = .56). Follow-up data were available for more than 98% of the remaining participants, and only 1% of visits were missed each year.
Among survivors, we observed 135 cases of incident diabetes (10.1 cases/1000 person-years). Of these, 71 cases (53%) were due to a fasting glucose level of 126 mg/dL or greater, 22 cases (16%) were due to new use of antidiabetic medication, and 42 cases (31%) were due to self-reported new diagnosis by a physician. We observed a graded increase in the incidence of diabetes with increased fetuin-A levels (Table 2). Participants in the highest tertile had more than twice the incidence rate compared with the lowest tertile in unadjusted analysis (13.3 vs 6.5 cases/1000 person-years, P = .005). We observed minimal change in this estimate after adjustment for demographic and other clinical predictors (adjusted hazard ratio, 2.41; 95% confidence interval, 1.28-4.53). Results were similar when body mass index replaced waist circumference and body weight as covariates in the model. Results were also similar between men and women, between black and white individuals, and among individuals with or without obesity (P values for interaction all > .27). When treated as a continuous predictor, each doubling in fetuin-A was associated with a 57% increased risk of incident diabetes (hazard ratio, 1.57; 95% confidence interval, 0.99-2.49; P = .05) in the final multivariate model (adjusted for age, sex, race, physical activity, body weight, waist circumference, systolic and diastolic blood pressure levels, fasting glucose level, high-density lipoprotein cholesterol concentration, triglyceride concentration, and C-reactive protein level).
In additional analysis, candidate mediators were added individually to the model to determine if they attenuated the association of fetuin-A with incident diabetes. We observed no significant attenuation on the basis of available measures of adipocytokines but observed moderate attenuation by visceral adiposity, which diminished the association by approximately one-third. Whereas the point estimate for hazards among the highest tertile remained 1.7-fold higher than the lowest tertile, the association was no longer statistically significant (P = .06) with adjustment for visceral adiposity (Table 3).
In this study, we observed that higher serum fetuin-A levels are associated with incident diabetes mellitus in humans. The association was independent of physical activity, inflammatory biomarkers, and other commonly available measures of insulin resistance and was similar irrespective of sex, race, and obesity status. Approximately one-third of this association appeared to be mediated by the quantity of visceral adiposity. Yet fetuin-A remained associated with a 1.7-fold risk of diabetes after adjustment for visceral adiposity, an association that was not statistically significant (P = .06).
In 1989, fetuin-A was reported as an endogenous inhibitor of insulin receptors through binding insulin receptor tyrosine kinase in adipocytes and skeletal muscle, which resulted in decreased rates of autophosphorylation and downstream insulin signaling cascades.11,12 This function was conserved across all species homologs evaluated to date in vitro25- 28 and has been confirmed in vivo in rats.28 The human fetuin-A gene resides on chromosome 3q27, which was previously identified as a metabolic syndrome and diabetes susceptibility locus.29,30 Fetuin-A knockout mice are insulin-sensitive; are resistant to weight gain; and have less adiposity, lower free fatty acid, and lower triglyceride levels compared with wild-type controls.13,14 We17 and others15,16 have previously demonstrated that higher fetuin-A levels are associated with insulin resistance/metabolic syndrome in cross-sectional studies. However, despite compelling in vitro and animal data, to our knowledge the association of fetuin-A with incident diabetes has not previously been evaluated. The data presented here are novel in demonstrating the longitudinal link between fetuin-A and incident diabetes in humans. Moreover, the data suggest that not only adipocyte-derived factors and pancreatic hormones, but also hepatocyte-derived proteins, such as fetuin-A, may regulate glucose metabolism in humans.
The association of fetuin-A with incident diabetes was partially attenuated with adjustment for visceral adiposity. Previous studies have demonstrated that fetuin-A stimulates adipogenesis in cell culture models,31 and conversely, fetuin-A knockout mice have less adiposity than wild-type controls.14 We hypothesize that the direct correlation of fetuin-A with visceral adiposity observed in this study is a consequence of higher fetuin-A levels and that accumulation of visceral adiposity may lie on a causal pathway between fetuin-A and incident diabetes. However, despite statistical adjustment for visceral adiposity, the association of fetuin-A with incident diabetes was only partially attenuated, which suggests that mechanisms other than accumulated visceral adiposity may likely also contribute to the link between fetuin-A and incident diabetes.
On the basis of these data, blockade of fetuin-A binding to the insulin receptor might be considered a novel therapeutic target for prevention or treatment of insulin-resistant states. However, fetuin-A levels have an inverse association with vascular calcification. Beyond the insulin-sensitive phenotype, the fetuin-A knockout mouse develops widespread soft-tissue calcification.32,33 Our group34 and others35- 37 have demonstrated that lower fetuin-A levels are associated cross-sectionally with vascular and cardiac valvular calcification in humans and are associated with mortality after myocardial infarction38 and in end-stage renal disease.37- 41 Therefore, any novel therapeutic that blocks fetuin-A should be closely evaluated for safety, particularly with respect to cardiovascular consequences.
Our study has limitations. Participation required that individuals be aged 70 to 79 years and of black or white race. The findings may not generalize to younger individuals or to other races or ethnicities. Two-hour oral glucose tolerance tests are more sensitive than fasting glucose for diagnosis of diabetes.42- 44 This measure was available at baseline, which provided accurate classification of diabetes status at entry, but not at follow-up. Unavailability of this measure at follow-up visits could have resulted in some incident diabetes cases being misclassified as normal and therefore should have biased our results toward the null hypothesis. Blind duplicate intraindividual coefficients of variation of fetuin-A were higher in this study than in previous articles.34,45 Again, measurement error misclassification should also have biased our results toward the null. Last, fetuin-A was measured at only 1 time point. Whether or not longitudinal trajectories of fetuin-A provide additional or more specific information regarding future diabetes risk is an important topic for future research.
In conclusion, fetuin-A is independently associated with incident diabetes in older individuals. Future studies should evaluate whether the results may generalize to middle-aged individuals in whom the incidence rate is highest.46 If confirmed in future studies, fetuin-A may ultimately prove useful as a target for therapeutics, and its study may provide novel insights to glucose metabolism in humans.
Corresponding Author: Joachim H. Ix, MD, MAS, Division of Nephrology and Hypertension, Department of Medicine, University of California, San Diego, and San Diego Veterans Affairs Healthcare System, 3350 La Jolla Village Dr, Mail Code 111-H, San Diego, CA 92161 (firstname.lastname@example.org).
Author Contributions: Ms Wassel had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Ix, Kanaya, Vittinghoff, Cummings, Shlipak.
Acquisition of data: Ix, Johnson, Cauley, Harris, Shlipak.
Analysis and interpretation of data: Ix, Wassel, Kanaya, Vittinghoff, Johnson, Koster, Shlipak.
Drafting of the manuscript: Ix.
Critical revision of the manuscript for important intellectual content: Ix, Wassel, Kanaya, Vittinghoff, Johnson, Koster, Cauley, Harris, Cummings, Shlipak.
Statistical analysis: Wassel, Vittinghoff.
Obtained funding: Ix, Johnson, Cauley, Harris, Cummings.
Financial Disclosures: None reported.
Funding/Support: This study was supported by an ADA-ASP Young Investigator Innovation Award in Geriatric Endocrinology sponsored by the Atlantic Philanthropies, American Diabetes Association, John A. Hartford Foundation, and Association of Subspecialty Professors (J.H.I.); an American Heart Association Fellow to Faculty Transition Award (J.H.I.); and contracts N01-AG-6-2101, N01-AG-6-2103, and N01-AG-6-2106 from the National Institute on Aging. The research was also supported in part by the Intramural Research Program of the National Institutes of Health.
Role of the Sponsors: The funding sources played no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data, nor in the preparation of the manuscript. The National Institute on Aging reviewed and approved this manuscript prior to submission.
Additional Contributions: We thank the other investigators, the staff, and the participants of the Health ABC Study for their valuable contributions and Frances A. Tylavsky, DrPH, Department of Preventive Medicine, University of Tennessee Health Science Center, Memphis, for her critical review of the manuscript.
We also thank the following Health ABC core investigators: Anne B. Newman, MD, MPH, and Piera Kost, University of Pittsburgh, Pittsburgh, Pennsylvania; Suzanne Satterfield, MD, DrPH, and Susan Thomas, University of Tennessee, Memphis; Stephen B. Kritchevsky, PhD, Wake Forest University, Winston-Salem, North Carolina; Michael C. Nevitt, PhD, and Susan M. Rubin, MPH, University of California, San Francisco; and Melissa E. Garcia, MPH, National Institute on Aging, Bethesda, Maryland.