Fibroblast growth factor 21 as a biomarker is superior to adiponectin in the prediction of incident diabetes

Abstract

INTRODUCTION: Adipokines are useful biomarkers for predicting diabetes risk as adipose tissue inflammation and dysregulated adipokine secretion are observed in obesity-related insulin resistance. We previously reported that serum adiponectin and tumor necrosis factor-alpha receptor 2 (TNF-α R2) levels independently predicted 5-year diabetes risk, and were comparable to 2 hours post oral glucose tolerance test (OGTT) glucose level (2hG), when used in combination1. Fibroblast growth factor 21 (FGF21), produced by the liver and also expressed in adipocytes, regulates glucose and lipid metabolism. Paradoxically, high circulating levels are found in prediabetes and type 2 diabetes, suggesting FGF21 resistance. We evaluated the use of FGF21 as a biomarker of diabetes risk and compared its performance with adiponectin, an anti-inflammatory adipokine, and several pro-inflammatory adipokines, when added to traditional risk factors, in diabetes prediction. HYPOTHESIS: FGF21 independently predicts incident diabetes and is superior to the previously studied obesity-related biomarkers. DESIGN AND METHODOLOGY: We studied 1380 non-diabetic subjects from the Hong Kong Cardiovascular Risk Factor Prevalence Study using the second visit (2000-2004) as baseline. All subjects’ glycemic status was tested by OGTT at baseline and follow-up visits in 2005-2008 and 2010-2012. Incident diabetes was defined as fasting glucose level (FG) ≥ 7 mmol/L or 2hG ≥ 11.1 mmol/L or use of anti-diabetic agents. Serum FGF21, adiponectin, TNF-α R2, lipocalin 2, leptin, interleukin-6, adipocyte-fatty acid binding protein (A-FABP) and high-sensitivity C-reactive protein (hsCRP) were measured at baseline. RESULTS: 123 participants developed diabetes over 9.0 years (median). Parameters including age, family history of diabetes, smoking, hypertension, body mass index and FG, which were either clinically relevant or independently predictive of diabetes, were used to construct a diabetes prediction model (DP). The levels of all biomarkers except lipocalin 2, at baseline, were significantly different between the groups with and without incident diabetes. On multivariable logistic regression analysis, FGF21 (p<0.001), A-FABP (p=0.004), hsCRP (p=0.009) and adiponectin (p=0.026) significantly predicted diabetes over 9 years. FGF21, with the best change in -2 log likelihood when added to the DP, improved the area under ROC curve (AUC) of the DP from 0.787 to 0.809 (P = 0.0078), rendering its performance comparable to the 'DP + 2hG' model (AUC = 0.830, P = 0.138) in ROC curve analysis. When an optimal cut-off at 178.2ng/ml for FGF21 was used, the performance of the prediction model remained similar (AUC = 0.813 [0.791-0.833]). CONCLUSIONS: As a biomarker, FGF21 appeared to be superior to adiponectin for diabetes prediction in this Chinese cohort and could be considered as an alternative to the OGTT.