Cross-sectional Analysis of AGE-CML, sRAGE, and esRAGE with Diabetes and Cardiometabolic Risk Factors in a Community-Based Cohort

Abstract

BACKGROUND

Advanced glycation end products (AGEs) and their receptors are regarded as central to the development of diabetic complications, but associations with diabetes and cardiometabolic outcomes in previous studies are mixed.

METHODS

Using ELISA assays, we measured N(6)-carboxymethyllysine (AGE-CML), soluble receptor for AGEs (sRAGE), and endogenous secreted receptor for AGEs (esRAGE) in 1874 participants from the Atherosclerosis Risk in Communities study. We conducted a cross-sectional analysis to evaluate associations of these biomarkers with demographics, diabetes, hyperglycemia, cardiometabolic measures, and genetic variants in the gene encoding RAGE, AGER (advanced glycosylation end-product specific receptor).

RESULTS

After adjustment for demographics and body mass index (BMI), there were no significant differences in AGE-CML, sRAGE, or esRAGE by diabetes or hemoglobin A_1c. Black race and AGER genetic variants were strongly associated with lower sRAGE and esRAGE even after adjustment [percent difference (95% CI) in black vs whites in sRAGE: −29.17 (−34.86 to −23.48), esRAGE: −26.97 (−33.11 to −20.84); with rs2070600 in sRAGE: −30.13 (−40.98 to −19.29), and esRAGE: −30.32 (−42.42 to −18.21); with rs2071288 in sRAGE: −20.03 (−34.87 to −5.18), and esRAGE: −37.70 (−55.75 to −19.65)]. Estimated glomerular filtration rate and albuminuria significantly correlated with sRAGE and esRAGE. BMI and C-reactive protein significantly negatively correlated with AGE-CML, sRAGE, and esRAGE. AGE-CML was modestly correlated with fructosamine and glycated albumin.

CONCLUSIONS

AGE-CML, sRAGE, and esRAGE were more related to genetic, kidney, and inflammatory measures than to diabetes in this community-based population. Our results suggest that, when measured by ELISA, these biomarkers lack specificity and are of limited value in evaluating the role of these compounds in diabetes.

Advanced glycation end products (AGEs)⁵ are proteins, nucleic acids, or lipids that are nonenzymatically glycated by aldose sugars such as glucose. They are synthesized endogenously in the blood and in tissues, inhaled through smoking, or ingested from foods that have undergone the Maillard reaction, a browning reaction that occurs during cooking (1, 2). AGE levels increase with blood glucose concentration and oxidative stress (1). AGEs are thought to be central in diabetes, a disorder defined by chronically increased glucose concentrations.

AGEs can induce a signal transduction cascade leading to an inflammatory response (3). The most common of the multiple forms of AGEs is N(6)-carboxymethyllysine (AGE-CML) (4). AGEs bind to various receptors, including the membrane-bound receptor for AGEs (RAGE), which activates the proinflammatory nuclear factor κ-light-chain-enhancer of activated B cells (NF-κB) cascade. This interaction leads to increased expression of RAGE, which induces more NF-κB, causing chronic inflammation (3). RAGE is also present in 2 forms that lack the NF-κB domain and hence do not induce inflammation, acting as decoy receptors: a soluble form that circulates in the blood (sRAGE), and an alternatively spliced isoform of RAGE, endogenous secreted RAGE (esRAGE) (1, 2, 5). A prevailing hypothesis is that when sRAGE and esRAGE levels are increased enough to outcompete proinflammatory RAGE for AGE binding, less inflammation occurs, leading to a decreased risk of inflammatory disease.

AGE-CML, sRAGE, and esRAGE have been identified as potentially useful biomarkers in diabetes, but associations of these parameters with complications of diabetes are unclear and controversial (6–8). AGE-CML has been shown to be positively associated with incident and prevalent diabetes and also cardiovascular disease risk factors (9–11). The direction of association for sRAGE with diabetes and cardiovascular risk factors, however, has varied substantially by population type. In patient populations with a history of diabetes, sRAGE has been positively associated with adverse outcomes. However, in community-based populations, the opposite is observed, with sRAGE being strongly inversely associated with clinical events. Specifically among those with diabetes, sRAGE studies have generally shown a positive association with risk of cardiovascular disease (7, 8, 12), mortality (12–13), and renal outcomes (13). However, among community-based populations, sRAGE has been reported to be inversely associated with diabetes (14), body mass index (BMI), LDL-cholesterol (15), inflammation (16), cardiovascular events (17–20), and chronic kidney disease (CKD) (21). esRAGE has been studied less, although higher levels have been associated with an increased risk of major adverse cardiovascular events in people with diabetes (22). The variable findings for sRAGE have raised questions regarding its utility as a diabetes biomarker (23). Further complicating the field is that measurement of AGE products is challenging and methods have differed across studies. Mass spectrometry can provide the most precise measurements but is resource intensive and problematic in large-scale studies. Immunologic techniques such as ELISA have been employed more commonly but are much less specific, capturing other molecules than the target.

Other potential contributors to variation in previous studies include genetic and racial contributions to sRAGE, as sRAGE levels differ by race (14, 17) and 2 variants in the gene that encodes RAGE, AGER (advanced glycosylation end-product specific receptor), explain 26% of the variation in whites and 21.5% of the variation in blacks, but do not explain differences in sRAGE levels by race (24). Thus, inadequate control for race and genetic factors may have contributed to the variable findings in the literature.

We undertook this study to characterize the associations of 3 of the most commonly studied serum AGE-related biomarkers—AGE-CML, sRAGE, and esRAGE. We measured these 3 biomarkers using commercially available ELISA methods and examined the associations with diabetes, measures of hyperglycemia, demographics, and cardiometabolic factors in the community-based Atherosclerosis Risk in Communities (ARIC) study cohort.

Materials and Methods

STUDY POPULATION

We conducted a cross-sectional study of a subset of participants in the ARIC cohort who also took part in the Carotid Magnetic Resonance Imaging (CARMRI) study that was nested within the overall ARIC cohort. The ARIC study is a prospective epidemiologic cohort that began with 15792 individuals recruited between 1987 and 1989 from 4 study centers across the US (Forsyth County, NC; Jackson, MS; Minneapolis, MN; and Washington County, MD) (25). The ARIC CARMRI ancillary study, consisting of a weighted sample of 2066 ARIC participants, took place in 2004 and 2005 (26). Participants in ARIC CARMRI underwent interviews, carotid artery MRI, and physical exams, and provided blood and urine samples. The present study was limited to 1865 individuals who participated in the CARMRI study, had valid AGE-CML, sRAGE, and esRAGE biomarker measurements, and had complete data for other variables of interest (90.3% of the CARMRI study sample).

The ARIC CARMRI study contains an oversampling of individuals with thicker carotid arteries and thus is a nonrandom sample of ARIC participants. All analyses in the present study were weighted by the inverse of the sample fractions in the study sampling strata using methods for the analysis of complex sample survey data, allowing for generalizability to the full ARIC (26). Institutional review boards at each study site approved the study protocol and written informed consent was obtained from all participants.

BIOMARKER MEASUREMENTS

Serum samples were collected from 2004 to 2005 as part of the CARMRI study and stored at −80 °C. AGE-CML, sRAGE, and esRAGE were measured in stored serum samples in 2012 and 2013 using ELISA methods. AGE-CML concentrations, which correspond to the burden of AGE products, were measured with the AGE-CML kit from Microcoat. Assay imprecision in 58 runs (2 runs per day) at a concentration of 24.7 ng/mL was a within-run CV of 5.1%, between-run CV of 2.7%, and a total CV of 5.8%. sRAGE concentrations were measured with the Quantikine Human RAGE Immunoassay kit (R&D Systems Inc.). Assay imprecision in 56 runs at a concentration of 467 pg/mL was a total CV of 9.4%; for a higher control at 3137 pg/mL the total was CV 8.7%. esRAGE concentrations were quantified with the esRAGE ELISA Kit (B-Bridge International, Inc.). Assay imprecision in 55 runs at a concentration of 0.317 ng/mL was a total CV of 11.9%; at a concentration of 0.481 ng/mL the total CV was 10.5%. Formal long-term stability data are not available, but these biomarkers are thought to be highly stable and have been measured in long-term stored samples in numerous prior studies. All measurements were performed in the Clinical Chemistry Research Laboratory at the University of Maryland School of Medicine, which is CLIA licensed and accredited by the College of American Pathologists.

ASSESSMENT OF DIABETES AND MEASURES OF HYPERGLYCEMIA

Individuals were defined as having diagnosed diabetes if they self-reported a physician diagnosis of diabetes or were taking glucose-lowering medication at or before the CARMRI visit. We further categorized individuals with and without diabetes according to clinically relevant categories of hemoglobin A_1c (Hb A_1c): <5.7, 5.7–6.4, ≥6.5% in persons without diagnosed diabetes and <7% or ≥7% in persons with diagnosed diabetes. Hb A_1c was measured in whole blood using the Tinaquant II immunoassay method (Roche Diagnostics) on the Roche Hitachi 911 analyzer, a Diabetes Control and Complications Trial (DCCT)-aligned assay. Glucose was also measured on the Roche Hitachi 911 analyzer using the hexokinase method (Roche Diagnostics). Fructosamine, glycated albumin, and serum albumin were measured in 2009 in stored serum samples with a Roche Modular P800 system (Roche Diagnostics) (26). Glycated albumin was calculated as a percentage of total albumin: {[(glycated albumin concentration in g/dL/serum albumin concentration in g/dL)/1.14] × 100} + 2.9.

OTHER VARIABLES OF INTEREST

BMI was calculated from height and weight measured during the physical exam. Family (parental) history of diabetes was collected by self-report. Information on diabetes and blood pressure–lowering medication use was obtained from self-reports and recorded medications, brought to the visit by the participants. We defined hypertension as current blood pressure–lowering medication use, mean systolic blood pressure ≥140 mmHg, or mean diastolic blood pressure ≥90 mmHg. HDL-cholesterol and total cholesterol were assayed from blood samples using standard techniques, and LDL-cholesterol was calculated using the Friedewald formula. Triglycerides, serum creatinine, and high sensitivity C-reactive protein (CRP) were measured in serum. Creatinine and albumin were measured from urine collected in sterile containers using a clean-catch method. We calculated estimated glomerular filtration rate (eGFR) from serum creatinine and demographic information using the CKD Epidemiology Collaboration (CKD-EPI) formula (27). eGFR stages were based on the Kidney Disease Improving Global Outcomes (KDIGO) guidelines (28): G1, eGFR ≥90 mL/min/1.73 m²; G2, eGFR 60–89 mL/min/1.73 m²; G3a, 45–59 mL/min/1.73 m²; G3b, 30–44 mL/min/1.73 m²; G4, 15–29 mL/min/1.73 m²; G5, <15 mL/min/1.73 m². Albuminuria was defined as the ratio of urine albumin to creatinine: A1, <30 mg/g, A2, 30–300 mg/g, A3, >300 mg/g. CKD risk levels were determined from the KDIGO guidelines: (a) low risk: G1–G2 and A1; (b) moderate risk: G1–G2 and A2 or G3a and A1; (c) high to very high risk: G3a and A2, A3, G3b, G4, or G5. Dietary intake was assessed using a 66-item food frequency questionnaire (29–31). A high AGE-diet score was calculated as the sum of the frequency of consumption of high AGE containing foods (butter, chicken or turkey with skin, hamburger, hot dogs, processed meat, bacon, and fried food) (32). The high AGE-diet score ranged from 0.66–21.43 servings per day with a median value of 9.73. We compared participants consuming a high level of AGE-containing foods (high AGE-diet score at or above the median) to those consuming a non–high AGE-diet (high AGE-diet score below the median).

Genetic data were obtained from 1000 Genomes Phase I imputed genome-wide association study data genotyped using the Affymetrix 6.0 array (24). We extracted imputation dosages for 2 single nucleotide polymorphisms (SNPs) in the AGER gene previously reported to be associated with sRAGE (rs2070600 and rs2071288), and assigned genotypes based on the most similar number of minor alleles to the dose (24).

STATISTICAL ANALYSES

All analyses accounted for the complex sample survey design using the svy command in Stata version 13.1 (StataCorp). We summarized median biomarker levels by diabetes categories and also by race, diabetes duration, medication use, and high AGE-diet score, and evaluated differences in medians by using the K-sample equality of medians test and Wilcoxon rank sum test. In the overall study population, we calculated Pearson correlations among AGE-CML, sRAGE, and esRAGE and with fasting glucose, Hb A_1c, and fructosamine.

We used linear regression models to examine the association between a diagnosis of diabetes, Hb A_1c, fructosamine, glycated albumin, and the other cardiometabolic measures with log-transformed AGE-related biomarkers (AGE-CML, sRAGE, and esRAGE) with adjustment for age, sex, race, and BMI, except when these variables were the exposure of interest, in which case the model was adjusted for the other variables only (model 1). Effect estimates from the linear models were converted to percent differences in AGE-CML, sRAGE, or esRAGE for the presence of each categorical predictor variable by exponentiating the effect estimate, subtracting it from 1, and multiplying by 100, i.e., 100 × (e^β − 1). We conducted sensitivity analyses with additional adjustment for CRP (model 2) and 2 SNPs in the AGER gene (rs2070600 and rs2071288) (model 3). We performed race-stratified analyses and tested for effect modification by race. We conducted a sensitivity analysis of the association between diabetes and AGE-CML after additional adjustment for serum albumin. We also used linear splines with 4 knots equally spaced across the data to allow for more flexible associations between AGE-CML, sRAGE, and esRAGE and continuous exposures (HDL, LDL, total cholesterol, triglycerides, CRP, eGFR, albuminuria, BMI, and age). In the spline models, the data were truncated at the 1st and 99th percentiles to reduce the influence of extreme points (33).

Results

In crude analyses, AGE-CML did not differ by diabetes categories, whereas sRAGE and esRAGE were lowest among persons with diagnosed diabetes and poor glycemic control (Hb A_1c >7%) (Table 1). In persons without a diagnosis of diabetes, there were significant inverse associations of lower sRAGE and esRAGE with Hb A_1c clinical categories (<5.7%, 5.7% to <6.5%, and ≥6.5%). AGE-CML was weakly correlated with both sRAGE (r = 0.12) and esRAGE (r = 0.13). As expected, sRAGE and esRAGE were strongly correlated with each other (r = 0.83) (see Fig. 1 in the Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/vol63/issue5). AGE-CML, sRAGE, and esRAGE were all lower in those persons with highest diabetes duration, although this trend was not significant (Table 2). Overall correlations of AGE-CML, sRAGE, and esRAGE with fasting glucose, Hb A_1c, and fructosamine were generally quite low (all correlations <0.10) with the exception of fructosamine and glycated albumin with AGE-CML which had moderate correlations of r = 0.23 and r = 0.18, respectively (see Fig. 2 in the online Data Supplement).

Table 1.

Median (25th, 75th percentiles) of AGE-CML, sRAGE, esRAGE fasting glucose, and Hb A_1c overall and according to diagnosed diabetes status and Hb A_1c category.^a

	Overall (n = 1874)	No diagnosed diabetes (n = 1409)				Diagnosed diabetesb (n = 465)
		Hb A1c <5.7% (n = 751)	Hb A1c 5.7%–6.4% (n = 467)	Hb A1c ≥6.5% (n = 65)	P valuec	Hb A1c <7% (n = 324)	Hb A1c ≥7% (n = 103)	P valuec
Fasting glucose, mg/dl, mmol/L	103.0 (94.0, 115.0), 5.72 (5.22, 6.38)	98.0 (91.0, 106.0), 5.44 (5.05, 5.88)	105.0 (96.0, 114.0), 5.83 (5.33, 6.33)	113.0 (103.0, 135.0), 6.27 (5.72, 7.49)	<0.001	116.0 (103.0, 140.0), 6.44 (5.72, 7.77)	159.0 (134.0, 190.0), 8.83 (7.44, 10.55)	<0.001
Hb A1c, %	5.6 (5.4, 6.0)	5.4 (5.2, 5.5)	5.9 (5.7, 6.0)	6.7 (6.6, 7.2)	<0.001	6.0 (5.7, 6.5)	7.4 (7.2, 8.2)	<0.001
AGE-CML, ng/mL	36.9 (31.8, 42.7)	37.7 (32.2, 43.9)	36.6 (31.7, 42.3)	36.2 (32.3, 40.7)	0.68	36.0 (30.5, 42.2)	34.1 (29.5, 42.5)	0.45
sRAGE, pg/mL	1221.4 (916.0, 1658.8)	1313.9 (1011.8, 1723.0)	1204.2 (889.5, 1637.5)	944.4 (773.4, 1582.9)	0.003	1159.0 (833.9, 1572.7)	1093.7 (814.9, 1608.9)	0.57
esRAGE, ng/mL	0.54 (0.40, 0.73)	0.57 (0.43, 0.74)	0.52 (0.40, 0.72)	0.44 (0.33, 0.65)	0.01	0.51 (0.37, 0.69)	0.47 (0.34, 0.67)	0.16

^aData are weighted, missing Hb A_1c values for 164 individuals.

^bDiagnosed diabetes was defined as self-reported diagnosis of diabetes or taking diabetes medication.

^cP value for difference in median biomarker levels by diabetes status. Wilcoxon rank sum test was used to compare medians from 2 groups in diagnosed diabetes, K-sample equality of median test was used to compare 3 medians among no diagnosed diabetes.

Table 2.

Median (25th, 75th percentiles) of AGE-CML, sRAGE, esRAGE, fasting glucose, and Hb A_1c in persons with diagnosed diabetes by duration of diabetes.^a

	Diagnosed diabetesb duration			P valuec
	<5 years (n = 128)	5–10 years (n = 104)	≥10 years (n = 222)
Fasting glucose, mg/dl, mmol/L	98.00 (91.00, 106.00), 5.44 (5.05, 5.88)	105.00 (96.00, 114.00), 5.83 (5.33, 6.33)	113.00 (103.00, 135.00), 6.27 (5.72, 7.49)	<0.001
Hb A1c, %	5.40 (5.20, 5.50)	5.90 (5.70, 6.00)	6.70 (6.60, 7.20)	<0.001
AGE-CML, ng/mL	37.72 (32.19, 43.90)	36.57 (31.73, 42.32)	36.15 (32.25, 40.66)	0.10
sRAGE, pg/mL	1313.86 (1011.77, 1723.04)	1204.24 (889.51, 1637.48)	944.43 (773.43, 1582.90)	0.33
esRAGE, ng/mL	0.57 (0.43, 0.74)	0.52 (0.40, 0.72)	0.44 (0.33, 0.65)	0.41

^aAnalyses are weighted, 9 individuals with missing diagnosed diabetes duration data (n = 454).

^bDiagnosed diabetes was defined as self-reported diagnosis of diabetes or taking diabetes medication.

^cP value for difference in median biomarker levels by diabetes status. Wilcoxon rank sum test was used to compare medians from 2 groups in diagnosed diabetes, K-sample equality of median test was used to compare 3 medians among no diagnosed diabetes.

In models adjusted for age, sex, race, and BMI (model 1), AGE-CML, sRAGE, and esRAGE were not significantly different in persons with diagnosed diabetes compared to those without (Table 3). AGE-CML, sRAGE, and esRAGE did not differ in participants with Hb A_1c ≥6.5% vs <6.5%, but AGE-CML and sRAGE were positively associated with fructosamine and glycated albumin. In all models, much lower concentrations of sRAGE and esRAGE were observed in blacks as compared to whites. The SNPs previously reported to be associated with sRAGE (rs2070600 and rs2071288) also showed large differences in sRAGE and esRAGE concentrations, with substantially lower concentrations of sRAGE and esRAGE in individuals with 1 or 2 copies of either risk allele, with differences in the same direction but attenuated for AGE-CML.

Table 3.

Adjusted percent difference in AGE-CML, sRAGE, and esRAGE according to the presence vs absence of diabetes and cardiometabolic disease risk factors (model 1, n = 1874).^a

	AGE-CML, percent difference (95% CI)	sRAGE, percent difference (95% CI)	esRAGE, percent difference (95% CI)
Diagnosed diabetesb vs no diagnosed diabetes	−0.46 (−3.52 to 2.61)	−1.93 (−7.89 to 4.02)	0.27 (−5.93 to 6.47)
Age
65–70 vs <65 years	−1.34 (−6.24 to 3.56)	−5.86 (−14.90 to 3.18)	−5.55 (−14.93 to 3.82)
70–75 vs <65 years	−4.36 (−12.02 to 3.30)	−5.39 (−19.87 to 9.08)	−9.89 (−24.74 to 4.95)
>75 years vs <65 years	−2.01 (−13.45 to 9.42)	−7.14 (−29.64 to 15.35)	−7.56 (−30.29 to 15.17)
Female vs male	0.80 (−1.69 to 3.29)	8.71 (3.94–13.47)c	9.46 (4.64–14.28)c
Black race vs white race	4.13 (1.22–7.05)c	−29.17 (−34.86 to −23.48)c	−26.97 (−33.11 to −20.84)c
Increased Hb A1c, ≥6.5% vs not increased	−2.88 (−6.46 to 0.70)	2.48 (−5.01 to 9.97)	1.43 (−6.70 to 9.57)
Fructosamine
Q2, 218.1–231.7 μmol/L vs Q1, <218.1 μmol/L	7.49 (4.17–10.80)c	−0.09 (−6.98 to 6.79)	−1.96 (−8.85 to 4.92)
Q3, 231.7–247.5 3μmol/L vs Q1, <218.1 μmol/L	16.46 (12.98–19.94)c	7.56 (0.86–14.27)c	4.02 (−2.73 to 10.76)
Q4, ≥247.5 μmol/L vs Q1, <218.1 μmol/L	19.41 (15.86–22.96)c	7.86 (0.72–14.99)c	4.39 (−3.01 to 11.80)
Glycated albumin
Q2, 12.7–13.7% vs Q1, <12.7%	8.82 (5.31–12.33)c	0.30 (−6.70 to 7.31)	0.10 (−6.98 to 7.19)
Q3, 13.7–14.9% vs Q1, <12.7%	14.35 (10.98–17.73)c	2.74 (−4.00 to 9.49)	3.19 (−3.76 to 10.14)
Q4, ≥14.9% vs Q1, <12.7%	15.38 (11.85–18.92)c	7.65 (0.79–14.51)c	5.79 (−1.62 to 13.20)
Serum albumin
Q2, 3.93–4.13 g/dL vs Q1, <3.93 g/dL	4.00 (0.39–7.61)c	−5.33 (−12.20 to 1.54)	−5.54 (−12.55 to 1.47)
Q3, 4.13–4.3 g/dL vs Q1, <3.93 g/dL	1.25 (−2.19 to 4.69)	−8.11 (−14.60 to −1.61)c	−4.34 (−11.30 to 2.63)
Q4, ≥4.3 g/dL vs Q1, <3.93 g/dL	5.10 (1.17–9.03)c	−15.63 (−23.13 to −8.14)c	−14.19 (−21.64 to −6.75)c
Hypertensiond vs no hypertension	3.95 (1.13–6.77)c	0.45 (−4.80 to 5.70)	−1.63 (−6.87 to 3.62)
Low HDL-cholesterol, <40 mg/dL (1.14 mmol/L) for women, <50 mg/dL (1.43 mmol/L) for men vs not low	−4.53 (−7.20 to −1.86)c	3.89 (−1.08 to 8.86)	0.35 (−4.74 to 5.44)
High LDL-cholesterol, ≥100 mg/dL (2.86 mmol/L) vs not high	0.31 (−2.39 to 3.00)	−1.21 (−6.08 to 3.67)	−1.40 (−6.40 to 3.59)
High total cholesterol, ≥200 mg/dL (5.71 mmol/L) vs not high	0.90 (−1.76 to 3.55)	−5.94 (−11.07 to −0.81)c	−4.42 (−9.63 to 0.80)
High triglycerides, ≥150 mg/dL (1.69 mmol/L) vs not high	−4.06 (−6.67 to −1.44)c	0.33 (−4.70 to 5.35)	−0.30 (−5.41 to 4.81)
CRP
1–3 mg/L vs <1 mg/L	−4.54 (−7.89 to −1.20)c	−5.86 (−11.83 to 0.10)c	−3.37 (−9.31 to 2.57)
>3 mg/L vs <1 mg/L	−4.80 (−8.17 to −1.44)c	−11.27 (−17.76 to −4.79)c	−9.44 (−15.88 to −2.99)c
Low eGFR, <60 mL/min/1.73 m2 vs not low	10.27 (6.83–13.71)c	19.72 (13.48–25.97)c	15.65 (9.43–21.86)c
eGFRe
G2 vs G1	2.23 (−0.88 to 5.34)	3.58 (−2.60 to 9.76)	0.50 (−5.91 to 6.91)
G3a vs G1	11.17 (6.64–15.69)c	19.91 (11.56–28.26)c	11.80 (3.22–20.38)c
G3b, G34, or G5 vs G1	15.63 (8.98–22.28)c	33.89 (20.24–47.53)c	30.85 (18.65–43.06)c
Albuminuria
30–300 mg/g vs <30 mg/g	6.30 (2.35–10.24)c	5.37 (−1.87 to 12.61)	3.00 (−4.56 to 10.57)
>300 mg/g vs <30 mg/g	−6.39 (−16.19 to 3.42)	39.37 (24.96–53.79)c	31.56 (17.37–45.75)c
Risk of CKDf
Moderate risk vs low risk	7.31 (3.93–10.69)c	8.90 (3.01–14.79)c	6.67 (0.72–12.62)c
High to very high risk vs Low risk	11.66 (6.83–16.50)c	27.31 (18.09–36.53)c	21.87 (12.68–31.06)c
BMI
25–30 vs <25 kg/m2	−4.05 (−7.26 to −0.85)c	−9.09 (−15.38 to −2.80)c	−10.42 (−16.77 to −4.08)c
≥30 vs <25 kg/m2	−13.11 (−16.44 to −9.78)c	−14.57 (−21.20 to −7.93)c	−16.09 (−22.80 to −9.38)c
High AGE-diet score, ≥mediang	−2.11 (−5.47 to 1.25)	−3.10 (−9.66 to 3.46)	−0.17 (−6.89 to 6.54)
Family history of diabetes vs none	0.72 (−2.21 to 3.65)	−0.31 (−5.68 to 5.07)	−0.82 (−6.44 to 4.80)
rs2070600h	−5.99 (−12.19 to 0.21)	−30.13 (−40.98 to −19.29)c	−30.32 (−42.42 to −18.21)c
rs2071288h	−4.37 (−11.56 to 2.81)	−20.03 (−34.87 to −5.18)c	−37.70 (−55.75 to −19.65)c

^aAdjusted for age, sex, race, and BMI, except when 1 of these is the variable of interest, then the model was adjusted for the other 2 variables only. Hb A_1c n = 1710; fructosamine n = 1861.

^bDiagnosed diabetes was defined as self-reported diagnosis of diabetes or taking diabetes medication.

^cP <0.05.

^dHypertension was defined as current hypertension medication use, diastolic blood pressure ≥90 mmHg, or systolic blood pressure ≥140 mmHg.

^eeGFR stages based on the KDIGO guidelines; G1: eGFR ≥90 mL/min/1.73 m2, G2: eGFR 60–89 mL/min/1.73 m2, G3a: 45–59 mL/min/1.73 m2, G3b: 30–44 mL/min/1.73 m2, G5: 15–29 mL/min/1.73 m2, G5: <15 mL/min/1.73 m2.

^fBased on KDIGO guidelines: low risk of CKD, eGFR G1 or G2 and albuminuria A1 (<30 mg/g); moderate risk of CKD, eGFR G1 or G2 and albuminuria A2 (30–300 mg/g), or eGFR G3a and albuminuria A1; high to very high risk of CKD, eGFR 3a and albuminuria A2, or albuminuria A3 (>300 mg/g), or eGFR G3b, or eGFR G4, or eGFR G5.

^gHigh AGE-diet score was calculated based on the sum of frequency (servings per day) of consuming high AGE-foods including butter, chicken, or turkey with skin, hamburger, hot dogs, processed meat, bacon, and fried food.

^hOne or 2 copies of minor allele vs 0 copies; n = 1600.

Measures of kidney function were strongly associated with sRAGE and esRAGE, with higher sRAGE and esRAGE concentrations in those with worse kidney parameters (lower eGFR, lower serum albumin, higher albuminuria, or higher predicted CKD risk based on KDIGO eGFR and albuminuria stages). AGE-CML concentrations were also higher in those with worse kidney function, but the associations were less robust than those for sRAGE and esRAGE. AGE-CML was not strongly correlated with serum albumin (r = 0.08), and a spline model showed a nonlinear association between AGE-CML and serum albumin (see online Supplemental Fig. 3).

Some additional differences were seen for other variables, especially overweight (BMI 25–30 kg/m²) and obesity (BMI ≥30 kg/m²), which were associated with lower AGE-CML, sRAGE, and esRAGE. High CRP (>3 mg/L) was also significantly associated with lower AGE-CML, sRAGE, and esRAGE. Female sex was associated with increased levels of sRAGE and esRAGE but not AGE-CML. Associations with hypertension, low HDL-cholesterol, high LDL-cholesterol, high total cholesterol, high triglycerides, and a family history of diabetes were inconsistent or not observed. In sensitivity analyses, additional adjustment for AGER SNPs (model 3) did not appreciably change our results, nor did adjustment for CRP (model 2) (see online Supplemental Table 1). Percent difference in AGE-CML by diagnosed diabetes status did not significantly differ after controlling for serum albumin: −0.78 (95% CI, −3.85 to 2.28). The shape of the continuous associations between AGE-CML, sRAGE, and esRAGE with CRP, eGFR, albuminuria, and BMI are shown visually in online Supplemental Fig. 3. Most associations were relatively linear, although some, such as log albuminuria, have different shapes near the tails of the distributions. Whereas fasting glucose and Hb A_1c levels differed substantially by medication use in the study population, we did not observe substantial differences in AGE-CML, sRAGE, or esRAGE according to glucose-, lipid- or blood pressure–lowering medication use (see online Supplemental Table 2). There were no significant differences in AGE-CML, sRAGE, or esRAGE according to categories of high AGE-diet score in crude analyses (see online Supplemental Table 3) nor after adjustment for age, sex, race and BMI (Table 3). Similarly, Pearson correlations of the high AGE-diet score with AGE-CML, sRAGE, and esRAGE were all very weak (Pearson correlations all |r| <0.1).

When stratified by race, the adjusted associations between AGE-CML, sRAGE, and esRAGE, and diabetes remained nonsignificant, although the interaction was significant for esRAGE (P = 0.035) (see online Supplemental Table 4). The only diabetes or cardiometabolic risk factor association that differed significantly by race was albuminuria in association with esRAGE, with a significant association in blacks but not whites (P for interaction =0.028). The associations of several diabetes and cardiometabolic risk factors with the SNPs, CRP, and eGFR, differed in magnitude by race but the interactions were not significant.

Discussion

In this community-based population, diabetes was not associated with serum AGE-CML, sRAGE, or esRAGE after adjustment for basic demographic information and BMI. In contrast, negative associations with race, SNPs in the RAGE gene (rs2070600 and rs2071288), BMI, CRP, male sex, and positive associations with markers of kidney disease were robust to adjustment. Our findings suggest that serum concentrations of AGE-CML, sRAGE, and esRAGE, when measured using standard ELISAs, may not be robust markers of the metabolic consequences of diabetes. Positive associations of AGE-CML with fructosamine and glycated albumin, which are short-term (2–4 weeks) markers of hyperglycemia, suggest serum AGE-CML may also reflect short-term glycation.

SNPs in the AGER gene that were associated with sRAGE in previous studies (24) also showed a strong association with sRAGE and esRAGE in our study. In a previous analysis in ARIC, these SNPs accounted for up to 26% of the variation in sRAGE concentrations but were not associated with clinical outcomes including diabetes, coronary heart disease, and CKD (24). Because these SNPs are known to be associated with AGE-related biomarkers, and adjusting for these variants in our analysis did not substantially affect results, our study supports the hypothesis that the SNPs do not play a causative role in diabetes and cardiovascular disease risk factors but rather independently affect levels of AGE-related biomarkers.

Race has also been associated with sRAGE and esRAGE levels and This association was one of the strongest associations observed in our study (14, 17, 34). Controlling for race along with age and sex did not attenuate the unadjusted association between diabetes and AGE-CML, sRAGE, and esRAGE, but the association between esRAGE and diabetes and albuminuria indicates potential effect modification. Consistent with prior studies, BMI and CRP were strong confounders of the association between diabetes and all 3 biomarkers in our study. Differences in models controlling for race, BMI, and CRP may be a source of differential results seen in previous studies.

We observed that kidney-related factors (eGFR, albuminuria, and CKD risk based on eGFR and albuminuria stages) were associated with higher AGE-CML, sRAGE, and esRAGE. This is consistent with a previous study in ARIC examining sRAGE and incident CKD before and after adjustment for baseline eGFR (21) as well as other reports (35, 36). It is possible that impaired renal clearance of sRAGE in the setting of impaired kidney function results in high circulating concentrations of sRAGE. Our findings support the hypothesis that observed associations between sRAGE and kidney disease, and related outcomes may occur through a mechanism independent of diabetes, likely as a result of impaired kidney filtration rather than as a complication of hyperglycemia. It is critical that future studies of sRAGE adequately account for the potential confounding effect of kidney function.

Prior studies have suggested strong positive associations of AGE-CML and negative associations of sRAGE and esRAGE with diabetes status (9, 14). Our results show a lack of association with diabetes, but associations with measures of hyperglycemia were mixed. Hb A_1c showed no association, but fructosamine and glycated albumin showed positive associations with AGE-CML and, to a lesser degree, sRAGE. AGE-CML was significantly but weakly associated with serum albumin, but adjusting for serum albumin did not attenuate the association between AGE-CML and diabetes, Hb A_1c, fructosamine, or glycated albumin (data not shown), indicating that serum albumin is not likely a strong confounder. Because AGEs form in both tissues and serum, however, it is possible that tissue albumin levels (unmeasured in the present study) may have a stronger effect on AGE-CML than serum concentrations.

The lack of consistent associations of AGE-CML, sRAGE, or esRAGE with diabetes and its complications in the literature may reflect difficulties in measuring components of the complex AGE biology and differences across studies in selection of molecules and tissue sources for measurement. AGEs bind to many receptors other than RAGE such as AGE receptor 1 and AGE receptor 2 (4), and RAGE has alternative ligands such as S100/calgranulin and high-mobility group protein B1 (5). Each of these molecules may compete with each other for ligands and receptors, affecting the amount of proinflammatory stimuli present in the cell. Thus, levels of sRAGE may depend on multiple molecules in the AGE-RAGE pathway, but many studies have measured either sRAGE individually or a combination of sRAGE, CML-AGE, and esRAGE, which only captures a portion of the AGE-RAGE pathway. Levels of RAGE would be useful to measure because sRAGE levels are dependent on RAGE levels, but methods to detect RAGE in humans do not currently exist (6). The alternative ligands for RAGE, alternative receptors for AGE, and the enzyme that cleaves RAGE into sRAGE would also be useful measures, because they affect the relative levels of sRAGE. In addition, different studies have measured sRAGE from different tissue sources, and it is not clear that all of these sources have the same biological relevance. One of the earliest studies to examine AGEs and diabetes was the DCCT, which found a positive association between several types of AGEs and Hb A_1c and diabetes complications in participants with type 1 diabetes (37). In contrast with our null and negative associations, the DCCT investigators obtained skin biopsies and measured AGEs in tissue using HPLC techniques (37).

Variable sRAGE and esRAGE findings in epidemiologic studies could also be related to assay diversity in research studies. Most recent studies have used ELISAs in serum or plasma, which use antibodies to selectively bind to epitopes on molecules of interest. These epitopes may occur on multiple molecular structures extracted from the serum, making the test nonspecific for AGEs (38, 39). In addition, ELISAs lack standardization. AGEs can also be measured by skin autofluorescence (40). This method is noninvasive, but is neither specific for individual AGEs, nor quantitative (38–41). Gas or LC-MS is sensitive, but not without technical limitations (23). Taken together, all of the issues in measurement make biologically meaningful assessment of AGE-related biomarkers difficult.

Our study has several limitations. Because of the cross-sectional nature, we could not determine the temporality of any of the observed associations. Also, we performed only single measurements of serum AGE-related biomarkers using ELISA assays.

In summary, we found no cross-sectional association of AGE-CML, sRAGE, or esRAGE with diabetes after adjustment for demographic factors and BMI. The factors most strongly associated with sRAGE and esRAGE were race, AGER SNPs, CRP, BMI, and kidney function. AGE-CML was associated with CRP, BMI, and kidney parameters but also fructosamine and glycated albumin, but not Hb A_1c. Future studies should recognize the limitations of ELISA methods in measurement of AGE-related biomarkers and fully account for the potential confounding effects of adiposity, inflammation, and kidney filtration. Our results reinforce the need for more specific analytical techniques (e.g., mass spectrometry) to enhance our understanding of the role of AGEs and RAGE in diabetes and other cardiometabolic disorders.