This is a cross-sectional study of a single-center cohort in which subjects were enrolled consecutively from the NAFLD and Health Examination Center of the First Affiliated Hospital of Sun Yat-sen University, China. The subjects were enrolled from January 1, 2009, to December 30, 2019, with the following inclusion criteria: (1) subjects aged 18 to 80 years; (2) patients with complete serum FFA levels, glucose, lipid, and insulin measurement results; and (3) assessment of NAFLD, FFA levels, and oral glucose tolerance test (OGTT) performed within 2 weeks.

The exclusion criteria were as follows: (1) diagnosis or any evidence of viral hepatitis, autoimmune liver disease, HCC, or other end-stage liver diseases; (2) diagnosis of other types of diabetes; (3) pregnancy and breastfeeding; (4) a previous history of alcohol consumption of >30 g per day in men or >20 g per day in women; (5) the use of statins, fibrates, or anti-diabetic drugs in the past 6 months; and (6) trained athlete status.

The protocol was approved by the Independent Ethics Committee for Clinical Research and Animal Trials of the First Affiliated Hospital, Sun Yat-sen University (approval number: [2017]104), and informed consent was collected from all participants. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Questionnaire interviews were conducted to collect patient histories, including patient demographics, past medical history, smoking status and alcohol consumption. All subjects underwent anthropometric measurements, including body weight, body height, waist circumference (WC) and blood pressure. Body mass index (BMI) was defined as the body weight in kilograms divided by the square of the body height in meters. WC was measured in centimeters at the midpoint between the lower margin of the rib cage and the top of the iliac crest using a nonelastic measuring tape. Sitting blood pressure was measured twice by physicians using an Omron (J710; Omron, Kyoto, Japan) electronic monitor, which was applied to the right upper arm after a 15-minute rest. Obesity was defined as having a BMI ≥25 kg/m², and abdominal obesity was defined as having a WC ≥90 cm for men and WC ≥80 cm for women [12]. Hypertension was diagnosed in subjects with a systolic blood pressure ≥140 mm Hg, a diastolic blood pressure ≥90 mm Hg or a previous diagnosis of hypertension [13].

After fasting overnight for ≥8 hours, blood samples were taken, and the following biochemical parameters were assayed using the Abbott c8000 Automatic Biochemistry Analyzer (Abbott, Abbott Park, IL, USA): FFA (acyl CoA synthetase-acyl CoA oxidase [ACS-ACOD] method), lipid profiles, uric acid, alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transpeptidase (GGT), alkaline phosphatase, fasting plasma glucose (FPG), fasting insulin (FINS), and glycosylated hemoglobin (HbA1c). The 2-hour plasma glucose (2hPG) levels were measured during the 75 g OGTT. Serum lipid levels were determined directly using the enzymatic colorimetric method with the following Beckman Coulter reagent test kits: total cholesterol (OSR6216), triglycerides (OSR61118), low-density lipoprotein cholesterol (LDL-C; OSR6283), and high-density lipoprotein cholesterol (HDL-C; OSR6287). The HbA1c test was performed with the National Glycohemoglobin Standardization Program (NGSP) method in our study (Reagent test kit: VARIANT II TURBO HbA1c Kit 2.0; Bio-Rad Laboratories, Hercules, CA, USA). All laboratory measurements were performed within one day using the same fasting samples, with the exception of the OGTT.

Based on the homeostasis model assessment of insulin resistance (HOMA-IR) index, IR values were calculated using the equation FINS (µU/mL)×FPG (mmol/L)/22.5, and IR was confirmed if the HOMA-IR level was over 2.69 [14]. Additionally, the fasting adipose tissue insulin resistance (Adipo-IR) index was calculated with the equation FFA (mmol/L)×FINS (pmol/L) [15]. Prediabetes, defined as an impaired fasting glucose level, was determined by a fasting glucose value lower than 5.6 to 7.0 mmol/L and/or a 120-minute blood glucose level between 7.8 and 11.1 mmol/L observed in the OGTT (impaired glucose tolerance) or an HbA1c level between 5.7% and 6.4%. T2DM was defined by a fasting glucose value of ≥7 mmol/L, a 120-minute OGTT value of ≥11.1 mmol/L, and/or HbA1c ≥6.5% (48 mmol/mol) after a clinical physician excluded other types of diabetes [16].

Because there can be discrepancies in the liver and kidney echo intensity, fatty liver was assessed in all subjects by abdominal ultrasonography based on the following criteria: the presence of posterior attenuation of the ultrasound beam, vessel blurring, difficult visualization of the gallbladder wall, and difficult visualization of the diaphragm [17]. A total of 979 subjects also underwent liver and pancreatic fat content (PFC) quantification via magnetic resonance imaging (MRI) of the upper abdomen with a 3.0-Tesla MRI scanner (Siemens 3.0T MAGNETOM Verio; Siemens, Munchen, Germany), which was performed within 2 weeks after the patient’s biochemical measurements were taken. The scanning protocol and imaging parameters, described in detail in our previous study [18], were as follows: TE1, 2.5 ms; TE2, 3.7 ms; repetition time, 5.47 ms; flip angle, 5°; receiver bandwidth, ±504.0 kHz per pixel; and slice thickness, 3.0 mm. The fat content was calculated for each patient using an irregularly shaped region of interest that covered the entire liver in 21 consecutive slices (maximum area centered). Based on the MRI proton density fat fraction (MRI-PDFF), the liver fat content (LFC) was classified as absent (<5%), mild (5% to 10%), moderate (10% to 25%), and severe (>25%) steatosis [19-21]. Pancreatic fat infiltration was defined as an average PFC ≥5%.

The software program R version 3.5.1 (http://www.Rproject.org), was used for statistical analyses. Continuous variables are shown as the mean±standard deviation or the median with interquartile range. Student’s t-test and the Mann-Whitney U test were used to compare continuous variables. The chi-square test was applied to compare categorical variables. Univariate and multivariate backward stepwise logistic regression was used to evaluate whether factors were independently associated with IR, prediabetes, and T2DM. We explored the nonlinear relationship between FFA levels and several parameters using smooth curve fitting analysis. Receiver operating characteristic (ROC) curve analysis was conducted for the predictive factors. A two-tailed P value less than 0.05 was considered statistically significant.

In total, 4,010 subjects were enrolled in our study, including 2,220 subjects in the non-NAFLD group, 1,790 subjects in the NAFLD group and 941 subjects in the NAFLD subgroup defined by MRI-PDFF. As shown in Table 1, the ages and sex proportions were comparable among the groups. IR (40.2% vs. 10.2%), prediabetes (14.3% vs. 11.5%), and T2DM (31.1% vs. 16.6%) were more common in the NAFLD group than in the non-NAFLD group. Compared with the non-NAFLD subjects, the NAFLD subjects presented with higher BMIs (25.7 kg/m² [23.7 to 28.0] vs. 21.9 kg/m² [19.7 to 24.4], P<0.001) and WCs (89 cm [83 to 95] vs. 77 cm [71 to 85], P<0.001), as expected. Additionally, we found that compared with individuals without NAFLD, individuals with NAFLD tended to have higher blood pressure, ALT, AST, GGT, and uric acid levels and significant differences in their plasma lipid profiles, including triglycerides, total cholesterol, HDL-C, and LDL-C. Notably, there were no significant differences in the FFA levels between the non-NAFLD and NAFLD groups (561±234 μmol/L vs. 562±218 μmol/L). Of the parameters associated with diabetes, FPG, FINS, HbA1c, HOMA-IR, and Adipo-IR index were significantly higher in the patients with NAFLD than in the subjects without NAFLD (all P<0.05); however, there were no significant differences in the 2hPG levels. Additionally, NAFLD subjects defined by MRI-PDFF showed consistent results when compared to the non-NAFLD group and presented similar trends as those of the total NAFLD subjects.

To explore the nonlinear association of FFA levels with traditional serum glucose and lipid metabolism markers in non-NAFLD and NAFLD subjects, we performed smooth curve fitting analyses after adjusting for age, sex, and BMI (Fig. 1). Among the NAFLD patients, plasma triglyceride concentrations fluctuated between 0 and 810 μmol/L as the FFA levels increased, increased sharply when the FFA levels reached 810 μmol/L, and then decreased when the FFA levels reached 1,121 μmol/L (Fig. 1A). In the non-NAFLD group, the triglyceride concentrations showed a stable, but not obvious, decreasing trend as the FFA levels increased. Total serum cholesterol levels in both groups displayed an increasing trend as the FFA levels increased; the curves in the NAFLD patients were sharper when the FFA levels were below 612 μmol/L but then became smooth (Fig. 1B). However, when the FFA levels increased to over 671 μmol/L in the NAFLD group, there was an obvious inverse U-shaped trend in the LDL-C concentration, whereas there was a positive linear correlation in the non-NAFLD group (Fig. 1C). For plasma HDL-C, both groups exhibited a similar and stable increasing trend, but the concentration was noticeably lower in the NAFLD patients (Fig. 1D). As shown in Fig. 1E, the HOMA-IR index rose more sharply as FFA levels increased in the NAFLD group and presented a J-shaped curve in general. However, there was an inverse J-shaped relationship when FFA levels increased between 500 and 1,000 μmol/L in the NAFLD group. As shown in Fig. 1F, compared to the FPG levels of the non-NAFLD subjects, which increased stably as FFA levels increased, the FPG levels of the NAFLD patients showed a decreasing trend as FFA levels increased; however, their levels increased sharply when FFA levels exceeded 748 μmol/L. The results of the NAFLD patients defined by the MRI-PDFF subgroup exhibited trends resembling those of the total NAFLD group in all of the above analyses.

There were no significant differences in the FFA levels of subjects with or without obesity and subjects with or without NAFLD (Fig. 2A). The NAFLD patients with abdominal obesity presented higher levels of FFAs than those with no abdominal obesity (574±214 μmol/L vs. 555±231 μmol/L, P<0.05), and this difference was not apparent in the non-NAFLD subjects (Fig. 2B). In total, 979 subjects underwent an MRI scan to measure the fat content of their liver and pancreas. The subjects with LFC >25%, which indicates severe steatosis, showed higher FFA levels than the subjects with moderate, mild and no steatosis (638±227 μmol/L vs. 571±201 μmol/L vs. 567±213 μmol/L vs. 550±183 μmol/L, respectively; P<0.05) (Fig. 2C). Notably, higher levels of FFAs were detected in the NAFLD patients who exhibited pancreatic fat infiltration (PFC ≥5%; 651±181 μmol/L vs. 558±213 μmol/L, P<0.05) (Fig. 2D). Besides, FFA levels were significantly correlated with LFC (r=0.08, P=0.014) and PFC (r=0.18, P<0.001). In addition, a stepwise increase in FFA levels was found from a normal status to the IR, prediabetes, and T2DM stages within the NAFLD group and the NAFLD subgroup defined by MRI-PDFF, while a significant difference was observed between a normal status and the other three stages in the non-NAFLD group (Fig. 2E). The NAFLD subgroup defined by MRI-PDFF showed similar results as those of the total NAFLD group. With the purpose of exploring subtle differences in FFA levels in subjects with LFC data, a comparison of FFA levels among the 10 quantiles of LFC was conducted, as shown in Supplementary Fig. 1. Significant differences in FFA levels were found among quantile 7 (LFC, 11.2% to 14.1%), quantile 9 (LFC, 17.2% to 22.4%), and quantile 10 (LFC >22.4%). However, there was no significant correlation between FFA levels and quantiles of LFC in non-NAFLD and NAFLD subjects.

Trend regression analyses were used to identify the role of FFA levels in predicting diabetes development. All the subjects were categorized based on the 25th (397 μmol/L), 50th (510 μmol/L), and 75th (647 μmol/L) FFA level percentiles (Table 2). In addition to the crude model, model 1 (adjusted for age and sex) and model 2 (adjusted for various biochemical parameters) were also built. In the NAFLD group, the subjects with the highest quartile of FFA levels had higher risk ratios for IR, prediabetes, and T2DM (odds ratios, 9.24 [6.43 to 13.36], 10.48 [5.66 to 19.39], and 19.43 [12.75 to 29.81], respectively; all P<0.05), even after adjusting for various confounders. Furthermore, significant differences were observed in the tests for trends. Analysis of the NAFLD subgroup defined by MRI-PDFF presented similar results. In contrast, in the non-NAFLD group, although there was a higher risk ratio in the subjects with a higher FFA level quartile, none of the analyses indicated significant trends after adjusting for confounders. Characteristic comparison among FFA levels quartiles is shown in Supplementary Table 1.

The independent risk factors for IR, prediabetes, and T2DM were identified via univariate and multivariate regression analysis (Supplementary Tables 2-4). For the purpose of evaluating the efficacy of FFA levels in predicting IR, prediabetes, and T2DM, ROC curves were drawn, as shown in Fig. 3. For predicting IR, prediabetes and T2DM, the areas under the curve (AUCs) for all of the independent risk factors combined, including FFA levels, were 0.82, 0.83, and 0.79 in non-NAFLD subjects, respectively (Fig. 3A, B, and C). While in NAFLD subjects, the AUCs for the combined independent risk factors were 0.71, 0.73, and 0.86, respectively (Fig. 3D, E, and F). For subgroup analysis in the NAFLD subgroup defined by MRI-PDFF, FFA levels were still an independent risk factor, although some other risk factors were identified. The AUCs for FFA levels and combined independent risk factors were similar to those for the total NAFLD group in predicting IR and T2DM. In predicting prediabetes, the AUC for the combined independent risk factors was lower than that of the total NAFLD group (Fig. 3G, H, and I). As shown in Table 3, the NAFLD subjects had lower cut-off values for FFA levels for predicting IR, prediabetes, and T2DM compared to the values for the non-NAFLD subjects, based on the ROC analysis (573 μmol/L vs. 815 μmol/L, 697 μmol/L vs. 938 μmol/L, and 715 μmol/L vs. 1,049 μmol/L, respectively). Additionally, cut-off values in the NAFLD subgroup defined by MRI-PDFF were found to be similar to those in the total NAFLD group and were still lower than those in the non-NAFLD group.