Advanced glycation end products measured by skin autofluorescence and subclinical cardiovascular disease: the Rotterdam Study

Participants were from the population-based cohort, the Rotterdam Study (RS), which invited all inhabitants of the Ommoord suburb of Rotterdam to participate [24]. The first subcohort (RS-I) was initiated in 1990 and included 7983 participants aged 55 years and older. The second subcohort (RS-II) started with 3011 new participants aged 55 years and older in 2000. The third subcohort (RS-III) included 3932 new participants aged 45 years and older in 2006. All participants were examined at baseline and were followed up every 3–6 years at home and at the Rotterdam Study research center. The Rotterdam Study was approved by the institutional review board (Medical Ethics Committee) of Erasmus Medical Center and by the review board of The Netherlands Ministry of Health, Welfare and Sports. All participants in the present analysis provided written informed consent.

SAF was introduced to the Rotterdam Study in the middle of follow-up visits (RS-I-6th, RS-II-4th, and RS-III-2nd visits) to the RS research center from 2012 to 2016 and was measured in 3009 participants. No apparent selection bias seemed present. After excluding 8 participants with SAF outside mean ± 4SD, 3001 participants were eligible for analyses and those with available data on outcomes were included for each analysis. Population inclusion and exclusion details are shown in Additional file 1: Figure S1.

During visits to the RS research center, SAF at the inner side of the dominant forearm was measured using an AGE Reader (Diagnoptics B.V., Groningen, The Netherlands) based on the fluorescent property of some AGEs. It is expressed in arbitrary units (A.U.) based on the ratio between the strength of emission and excitation light. The measurement was previously validated against liquid chromatography-mass spectrometry measurements [25]. Three measurements were taken consecutively and showed good agreement (Pearson correlation coefficients 0.87–0.91). The mean was used for analyses. The device was calibrated by measuring the autofluorescence of the standard and requires re-calibration if the difference with standard autofluorescence is more than 10%. Details of the measurement have been described previously [26]. 1 A. U. corresponds to ~ 2 SD difference in SAF in our population. In healthy Dutch individuals younger than 70 years, SAF showed a slow yearly progression (~ 0.024 A. U.) and 1 A. U. difference of SAF corresponds to differences of ~ 40 years difference. However, smokers had a much higher SAF (~ 0.16 A. U.) than non-smokers [5], so as the presence of other major risk factors including diabetes and CKD. The yearly progression of SAF was also a bit faster when kidney diseases and diabetes were present [5, 27]. A quick increase of around 0.4 A. U. SAF was observed after acute renal failure [28]. SAF showed no seasonal variation in a German study and our study population [29]. To minimize the potential influence of skin creams, blood flow, and skin color, SAF was measured at the inner side of the forearm. Participants were asked not to use sunscreen and skin creams 2 days before the measurements and had tests without performing physical exercise.

To confirm an association of SAF with CVD, we examined the association of SAF with a history of myocardial infarction (MI), a major type of CVD. A history of MI was defined as having had MI at study entry or had an MI event during follow-up before SAF was measured during the RS-I-6th, RS-II-4th, and RS-III-2nd visits. Data collection on MI was explained in detail elsewhere [30]. Briefly, in RS-I, the presence of MI at study entry was based on verification of self-reported MI or electrocardiogram abnormalities suggestive of previous MI. In RS-II and RS-III, the medical records of all participants were screened for prevalent MI at entry. The presence of MI was based on clinical information and ICD-10 coding from the medical records, adjudicated by two independent cardiovascular researchers. During follow-up, medical records and hospital discharge diagnoses of all participants were continuously screened for MI events.

Physical activity, smoking habits, alcohol intake, the highest level of education acquired, medications used in the past week, and medical history were collected from home interviews. Physical activity was expressed in metabolic equivalents METhours/week. Smoking status was categorized as never, past, or current smoker based on the smoking habit of cigarette, cigar, or pipe. Alcohol intake was harmonized to grams of alcohol/day. Education level was harmonized across all three RS subcohorts according to the UNESCO classification. Weight and height were measured at the research center. Fasting serum glucose, creatinine, total cholesterol, and HDL-cholesterol were measured using standard techniques. The estimated glomerular filtration rate (eGFR, mL/min per 1.73 m²) was calculated using the CKD-EPI equation. CKD was defined as eGFR ≤ 60 mL/min per 1.73 m². Diabetes was defined as a fasting glucose of ≥ 7.0 mmol/L or a non-fasting glucose level of ≥ 11.0 mmol/L and/or the use of glucose-lowering medication, and was verified with medical records [38]. Dyslipidemia was defined as total cholesterol > 6.5 mmol/L, or use of lipid-lowering medications, or HDL < 1.0 mmol/L for men and HDL < 1.2 mmol/L for women.

Population characteristics of the total population and SAF tertile groups were summarized. To balance sex and age representation among tertile groups, we obtained tertiles of residuals of SAF regressed on age in men and women separately and created sex-specific and age adjusted SAF tertile groups. SAF as a continuous variable, z-scores of IMT, blood pressure, and PWV were used in regression analyses. CAC scores were categorized into clinically meaningful categories (< 100, and ≥ 100), and also analyzed in the natural logarithm of (CAC + 1) for analyses due to their skewed distribution and excessive zeros. The associations between SAF and endophenotypes were studied using linear regression for continuous variables or (binary or multinomial) logistic regression for categorical variables, with SAF being the independent variable regardless of the potential causal direction. Three models were constructed to adjust the associations for potential confounders that are risk factors for AGE accumulation and cardiovascular diseases. Model 1 was adjusted for age, sex, and RS subcohorts. Model 2 was additionally adjusted for cardiovascular risk factors including BMI and the presence of dyslipidemia and hypertension. To explore the influence of covariates that are potential confounders but also important determinants of SAF, we additionally adjusted for smoking, diabetes, and eGFR in model 3. Theoretically, AGEs may also act as intermediates linking risk factors such as diabetes and smoking to cardiovascular risk. The directed acyclic graph (Additional file 1: Figure S3) shows the potential relationship of included confounders with SAF and cardiovascular outcomes. Additionally, associations of conventional cardiovascular risk factors with SAF were provided. Though not the focus of this study, we also investigated the associations of continuous and categorical SAF with prevalent MI using logistic regression models. Underlying assumptions for linear and logistic regression analyses were checked, including heteroscedasticity, collinearity, nonlinear effect, and influential data points, by reviewing variation inflation factors, residual plots, and probability plots. To reduce reverse causation as much as possible, covariates measured during visits close to the assessment of outcomes and age at SAF measurement time were used for adjustment in regression models. Specifically, covariates were derived from RS-I-5, RS-II-3, and RS-III-2 visits for analyses on MI history, hypertension, IMT, and plaque presence; from RS-I-3, RS-II-1, and RS-III-1 visits for analyses on CAC and PWV.

Missing data in covariates (ranging from 0.8 to 4.1%) were imputed by multiple imputations using logistic regression and predictive mean matching methods for categorical and continuous variables respectively for 10 imputations using the MICE package in R. The results were pooled from analyses of the imputed datasets. A two-sided p < 0.05 was considered statistically significant. All analyses were conducted using R Studio (version 4.1.2).

A higher SAF was associated with an MI history, independent of other established cardiovascular risk factors (adjusted odds ratio of MI for one unit higher SAF 1.57 (95% CI 1.16, 2.12)) (Table 2). The highest SAF tertile had an odds ratio of 1.87 (95% CI 1.28, 2.73) compared to the lowest tertile. No interaction of SAF with sex, diabetes, and CKD was observed (p > 0.1), but the interaction of SAF and diabetes had a p-value of 0.13, which is close to 0.1. Subsequent stratified analysis showed a more pronounced association in participants who had diabetes (OR 2.83 (95% CI 1.42, 5.68)) than those without (OR 1.39 (95% CI 0.98, 1.95)) in model 3 (Additional file 1: Figure S4).

Carotid plaques were detected in 2469 of the 2882 participants with images of more than 4 carotid arteries (Table 1). Higher SAF was associated with plaque presence after adjusting for age, sex, and RS subcohorts (OR 2.24 (95% CI 1.76, 2.85)) for one unit higher SAF), but the association was attenuated in model 3 (OR 1.20 (95% CI 0.92, 1.57)) (Table 3). Plaque severity was classified as mild, moderate, and severe in 494 (17.1%), 561 (19.5%), and 1414 (49.1%) participants, respectively (Additional file 1: Table S3). Using participants who did not have plaques as the reference group, a higher SAF was only associated with the severe plaque burden (OR 1.45 (95% CI 1.09, 1.93) for one unit higher SAF) but not with the moderate (OR 0.87 (95% CI 0.64, 1.20)) and mild (OR 1.17 (95% CI 0.85, 1.60)) plaque burden in model 3. Sex modified the association of SAF with plaque presence (p for interaction = 0.03). The association of SAF with plaque presence (yes/no) was not significant in either sex in model 3, but the association with severe plaque burden appeared to be larger in men than in women. (Additional file 1: Figure S5).

In model 3, one unit higher SAF was associated with 0.08 SD (95% CI 0.01, 0.15) higher max IMT (Table 3). Significant interactions were noticed including SAF with sex (p for interaction = 0.01), diabetes (p for interaction = 0.02), and CKD (p for interaction = 0.01). Stratified analysis showed that the association of SAF with IMT was attenuated substantially after adjusting for age and pertained in men but not women, and in individuals with either diabetes or CKD but not those without these diseases (Fig. 2). Adjusting for other risk factors further weakened the association.

The CAC score was measured by EBT (N = 455) or MDCT (N = 318). In model 1, one unit higher SAF was associated with a 0.55 (95% CI: 0.1, 0.99) higher EBT-CAC score in natural logarithm, and a 0.77 (95% CI: 0.25, 1.29) higher MDCT-CAC score in natural logarithm (Table 3). The associations attenuated in model 3 to 0.28 (95% CI −0.15, 0.71) for EBT-CAC and 0.55 (95% CI 0.04, 1.06) for MDCT-CAC. No significant interactions with sex, diabetes, or CKD were noticed. For categorical CAC scores, taking the category of absent or mild CAC score (CAC < 100) as the reference, a higher SAF was associated with a moderate or severe MDCT-CAC score (CAC ≥ 100), OR 2.2 (95% CI 1.39, 3.48). No significant association was observed for the categorical EBT-CAC score (OR 1.31 (95% CI 0.87, 1.97)).

In model 3, one unit higher SAF was associated with a 0.08 SD (95%CI 0.01, 0.15)) higher pulse wave velocity, an indicator of arterial stiffness (Table 3). A significant interaction between SAF and diabetes (p for interaction = 0.001) was observed (Additional file 1: Figure S6). A more pronounced association was noted for people who had diabetes than those who did not have diabetes, with an adjusted difference in PWV of 0.36 SD (95% CI 0.06, 0.65) vs. 0.05 SD (95% CI −0.03, 0.13) for one unit higher of SAF.

Hypertension was present in 2166 out of the 2996 participants with available information. No association was observed between SAF and hypertension in model 3 (OR 0.99 (95% CI 0.81, 1.21) (Table 3).

The associations of SAF with all endophenotypes remained similar after excluding participants with an MI history (Additional file 1: Table S4).

Given a likely role of AGEs in arterial stiffness, an association of a higher SAF with hypertension would be expected. However, the current study fuels the debate of whether AGEs are involved in hypertension by showing no association [50]. DBP was even inversely associated with SAF, which was also observed in a younger Dutch cohort NQplus in men but not in women [51]. It is widely perceived that DBP tends to decrease after around 50–60 years of age mainly due to stiffness of large arteries [52]. As such, our observations may imply that AGEs contribute to large artery stiffness by forming crosslinks on elastin or collagen during the aging process. The results may not be extrapolated to younger populations because the majority of our study participants were at the later part of the blood pressure trajectory where DBP tends to decrease and the increase of SBP with age slows down [53]. In fact, the Dutch LifeLines cohort covering a wider age range (n = 78671) reported that higher SAF was associated with hypertension in men [54]. Nevertheless, a null association between SAF and BP was also observed in primary school children [55]. We encourage longitudinal studies to investigate the link between AGEs and hemodynamic patterns to address their relevance to cardiovascular risk.

The association of SAF with IMT showed subgroup differences. It was stronger in people who had diabetes or CKD and was insignificant in those without and was seen in men but not women, arguing for a more pronounced impact of AGEs on IMT in diabetes or CKD, and in men.

Diabetes and CKD are two major risk factors for AGE accumulation and CVD. Our observation may indicate that AGE accumulation is involved in arterial wall thickening and further contributes to an elevated CVD risk in diabetes or CKD. In line with our finding, a study in people with type 1 diabetes showed that, compared to regular glucose management, 6.5 years of intensive glucose management resulted in a lower IMT [13] 6 years after the intervention and the levels of skin AGEs and IMT correlated [12]. Likewise, we observed a larger association between SAF and PWV in people with diabetes than in those without but the large time gap between measurements limits the interpretation of results.

Sex differences in the association between SAF and IMT suggest that SAF is more relevant for IMT in men. This difference may be attributed to the better risk profile of women. Moreover, women generally have a lower IMT than men before menopause potentially due to a lower grade of chronic inflammation [56], protection from estrogen [57], or a better risk profile [58]. The sex disparity diminishes gradually afterward [59]. Although all participants in our study were postmenopausal, the long half-life of skin AGEs (median 15 years) may capture associations of skin AGEs and IMT from earlier life stages, i.e., a potential absence of association in women before menopause. Similarly, we observed an association of SAF with severe plaque burden only in men. Future longitudinal studies could investigate whether these differences translate to sex disparities in CVD risk [39]. Our results shed light on how subclinical cardiovascular pathological changes in relation to SAF might develop differently in men and women and individuals with or without diabetes or CKD.

Strengths of our study include the analysis in a well-characterized and densely phenotyped general population with well-defined outcomes, allowing for investigating multiple CVD-related endophenotypes within one general population of adequate sample size and controlling for multiple confounders. SAF, as a marker for long-term AGEs accumulation, can capture the accumulated AGE burden in tissues during a longer period than circulating markers.

Several limitations are also present in the current study. Our results were restricted to the older Dutch population and the generalization of results needs caution. The small group of participants with diabetes or CKD and CAC data also limits the power for statistical inference. Further, we were not able to rule out influences from survival selection and were only able to investigate the association of SAF with a history of non-fatal MI as the clinical outcome of fatal MI before SAF measurement was automatically precluded. Nevertheless, in both cases, selection would likely have biased the associations toward the null. In addition, reverse causation might have biased the estimations because some outcomes were assessed before SAF measurement although SAF showed a slow yearly progression [5]. For this reason, we repeated our analyses in participants without a history of MI as a sensitivity analysis. In addition, progression in endophenotypes measured earlier was not accounted for, which may have concealed some of the associations. Studies in CKD patients showed that one year increase in SAF was predictive of cardiovascular events and all-cause mortality [27, 60]. Future studies with repeated measurement of SAF and longitudinal data on CVD incidence may investigate if changes in SAF are related to cardiovascular risk in the general population. Although we were able to control for many risk factors, given the intertwined relationship of SAF with many risk factors, residual confounding could not be ruled out completely. Finally, the cross-sectional design does not allow for the interpretation of causality.