《1. Introduction》

1. Introduction

The treatment and post-onset rehabilitation and care of cardiovascular diseases (CVDs) comprise a heavy economic burden [1], as CVDs rank as the leading cause of disability and mortality worldwide. According to the ‘‘Global Burden of Disease Study 2019” [2], there were 523 million prevalent cases of CVDs in 2019, and the number of deaths due to CVDs reached 18.6 million. Globally, the years lived with disability due to CVDs doubled from 17.7 million in 1990 to 34.4 million in 2019. Thus, it is essential to upscale the primary prevention of CVDs and identify novel biomarkers and tools to facilitate risk assessment and targeted intervention.

Glycosylation, the covalent attachment of sugar moieties to proteins, is the most common and diverse form of posttranscriptional modification, providing a novel dimension for early diagnosis, targeted prevention, and precision medicine by means of glycomedicine [3]. After nucleic acids and proteins, the first and second alphabets of the life code in the central dogma law, glycans could be the third alphabet of the cellular life under the governance of so-called paracentral dogma in cellular materialisty [3]. Glycans are regulated by glycosyltransferases and glycosidases [4], and participate in key biological processes (i.e., cell adhesion, molecular trafficking and clearance, receptor activation, signal transduction, and endocytosis). It is notable that the interaction between the N-glycosylation pattern and immunoglobulin G (IgG) has a direct impact on the regulation of inflammation, as the attached glycans can shift the role of IgG in pro- and antiinflammatory functions [5]. The core of the glycan domain consists of N-acetylglucosamine (GlcNAc) and mannose residues, which may contain an added bisecting GlcNAc residue linked either to the first mannose or to terminal galactose or sialic acid residues [6–8]. 

Previous studies have reported the association of IgG N-glycosylation profiles with aging [9], inflammation [10], and cardiometabolic diseases including hypertension [11], diabetes [12], and dyslipidaemia [13], which are important traditional risk factors of CVDs. One study confirmed that the IgG N-glycosylation profile is associated with atherogenic lipoprotein and subclinical atherosclerosis carotid plaque [14]. However, the link between the IgG N-glycosylation profile and cardiovascular risk or diseases remains unknown. The concept of IgG glycan age has been proposed to explain biological age [15,16]. In addition, a recent study developed an aging clock to track cardiovascular aging using inflammatory biomarkers [17]. However, the capacity of the IgG N-glycosylation profile to indicate cardiovascular health has not been elucidated. 

Unlike conventional statistical models, which input limited variables and show relatively low efficiency for predictive performance, artificial intelligence (AI) approaches can incorporate high-dimensional and multivariate data to solve challenging issues [18]. Machine learning (ML) is a subdomain of AI involving the automated discovery of patterns within data [19,20]. For different ML-based models, supervised learning tools, such as random forest, the least absolute shrinkage and selection operator (LASSO), and recursive feature elimination (RFE), can learn complicated structures by incorporating numerous variables with multidimensional data [21]. Furthermore, due to their outstanding performance, ML techniques have been increasingly applied to clinically relevant domains, such as predictors selection, predictive diagnostics, targeted prevention, and personalized medical services [9,17,18]. 

In this study, we applied an ML approach for feature selection and further developed and validated an IgG N-glycosylation cardiovascular age (GlyCage) index to track cardiovascular risk, thereby providing a novel tool to evaluate cardiovascular aging beyond calendar age.

《2. Methods》

2. Methods

《2.1. Study population》

2.1. Study population

The data used in this analysis were collected from the Busselton Healthy and Ageing Study (BHAS). The study design and recruitment procedures of the BHAS have been described elsewhere [22]. A total of 5107 participants were recruited by the BHAS from May 2010 to December 2015 [23]. For the present cross-sectional study, 1465 participants—that is, those whose complete information was available, including the IgG N-glycome, medical history of CVDs, demographic data, anthropometric measurements, and biochemistry tests—were included in the final analysis. This study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Busselton Population Medical Research Institute (Amendment to BSN 22/10) and Edith Cowan University Human Research Ethics Committee (2019-00947- ZHENG and 2021-03164-WU). All participants gave their informed consent to participate.

《2.2. Measurement and definition》

2.2. Measurement and definition

The demographic characteristics, lifestyle factors, information of medication use, and medical history of diseases were assessed by the BHAS participant questionnaire [22]. Educational level was categorized as primary school or below, secondary school, or above secondary school. Marital status was classified as married and others. Smoking status was defined as currently smoking or not.

Body mass index (BMI) was measured by means of standard anthropometric techniques and calculated as weight in kilograms divided by height in meters squared. Systolic and diastolic blood pressures were measured using a vascular profiler (Omron VP1000, Japan) in all four limbs with the participant supine, after 5 min of rest. Tests for total cholesterol, high-density lipoprotein (HDL) cholesterol, fasting plasma glucose, glycated hemoglobin A1c (HbA1c), and C-reactive protein (CRP) were conducted on fresh specimens by PathWest laboratories (Busselton and QEII Medical Centre, Australia).

《2.3. Definition of ten-year cardiovascular risk and CVDs》

2.3. Definition of ten-year cardiovascular risk and CVDs

The ten-year cardiovascular risk was orthodoxly defined using eight clinical factors (i.e., age, sex, total cholesterol, HDL cholesterol, systolic blood pressure, use of anti-hypertensive medication, diabetes status, and smoking status) for individuals from 30 to 74 years without CVDs, with reference to the established Framingham Heart Study score [24]. In brief, the Framingham ten-year cardiovascular risk was calculated as 1 – for males and 1 –  for females, where β is the regression coefficient and X is the level of each risk factor mentioned above. The detailed sex-specific formula is provided in Table S1 in Appendix A. Individuals with a tenyear cardiovascular risk ≥ 20% were defined as the high-risk group.   

CVDs were defined as angina, myocardial infarction or heart attack, transient ischemic attack (TIA), stroke, experience of coronary angioplasty or stent, coronary bypass, and implant of cardiac pacemaker in this study. The composition of CVDs among the sample of this study is shown in Table S2 in Appendix A.

《2.4. IgG isolation and N-glycan analysis》

2.4. IgG isolation and N-glycan analysis

IgG was isolated from blood plasma utilizing 96-well protein G monolithic plates based on previously introduced methods [25]. The ‘‘in-solution’’ method was performed to release and label the N-glycans. IgG N-glycans were separated by means of hydrophilic interaction chromatography and measured on an ultraperformance liquid chromatography (UPLC) instrument (Waters Corporation, USA) into 24 glycan peaks (GPs) [26]. Detailed structures are shown in Section S2 in Appendix A. 

《2.5. Feature selection in machine learning》

2.5. Feature selection in machine learning

We used the LASSO and RFE ML algorithms to select the intersection of N-glycans. The features selected by those two methods enabled the models’ stability and minimized the risks of overfitting. The lambda was set as ‘‘1se” and the area under curve (AUC) was chosen as the performance measure for LASSO. The bagged tree was embedded in RFE, and accuracy was chosen as the performance index to optimize. Based on the glycans selected by this ML approach, a multi-step logistics regression model was used to determine the final glycans, along with calendar age and sex, for GlyCage index construction based on the Framingham ten-year cardiovascular risk.

《2.6. Statistical analysis》

2.6. Statistical analysis

Characteristics are presented as the mean (standard deviation (SD)) or number (proportions), as appropriate. Differences are compared using Student’s t-test for continuous variables and the χ2 test for categorical variables. The distribution of IgG glycans are presented using a box plot stratified by the ten-year cardiovascular risk and CVDs.

An adjusted logistics model was used to evaluate the association of the established GlyCage index with the Framingham tenyear high cardiovascular risk and CVDs in the validation stage. The calendar age, education level, marital status, and CRP were adjusted in the model. We then analyzed the effect of excessive GlyCage, defined as the distance between GlyCage and calendar age, on cardiovascular risk and diseases. A receiver operating characteristic (ROC) curve analysis was performed to compare the capacity to indicate cardiovascular health among GlyCage, calendar age, and the IgG glycosylation biological aging (GlycanAge) proposed in a previous study [27]. A comparison of the ROC curves was performed using the DeLong test. All the statistical analyses were conducted using R software, version 4.1.0 (R Foundation, Austria). The difference was considered to be statistically significant at a two-side significance level with a P value < 0.05.

《3. Results》

3. Results

《3.1. Population characteristics》

3.1. Population characteristics

A total of 1465 individuals were enrolled in the final analysis. The mean age (SD) was 59.58 (5.51) years, including 107 (7.3%) cases with a history of CVDs (Table 1). Among the 1358 individuals with no history of any CVDs, 376 cases (27.7%) had a high ten-year cardiovascular risk, as shown in Table S3 in Appendix A.

《Table 1》

Table 1 Characteristics of the study population.

SI conversion factors: to convert fasting plasma glucose to mg·dL–1, multiply by 18.0; to convert triglyceride to mg·dL–1, multiply by 88.6; to convert cholesterol to mg·dL–1, multiply by 38.66. 

SD: standard deviation; SBP: systolic blood pressure; DBP: diastolic blood pressure. 1 mmHg = 0.133 kPa.

a CVDs included angina, myocardial infarction or heart attack, TIA, stroke, experience of coronary angioplasty or stent, coronary bypass, and implant of cardiac pacemaker.

b Calculated as weight in kilograms divided by height in meters squared.

《3.2. Multi-step IgG N-glycans selection in ML》

3.2. Multi-step IgG N-glycans selection in ML

Fig. S1 in Appendix A shows the distribution of 24 GPs stratified by ten-year cardiovascular risk (Fig. S1(a)) and CVDs (Fig. S1(b)). In the primary stage, a subset of five GPs (GP6, GP7, GP8, GP13, and GP20) was selected by the LASSO (Fig. S2 in Appendix A) and RFE algorithms (Fig. 1). Ultimately, fucosylated N-glycans with bisecting GlcNAc (GP6, FA2B) and digalactosylated N-glycans with bisecting GlcNAc (GP13, A2BG2) were retained in the logistical model (P < 0.05), as shown in Table S4 in Appendix A.

《Fig. 1》

Fig. 1. Selected IgG N-glycans using LASSO and RFE algorithms.

《3.3. GlyCage index construction》

3.3. GlyCage index construction

GlyCage index was finally calculated (Table S5 in Appendix A) as follows: 

where X1 is the expression level of GP6, X2 is the expression level of GP13, and V is calendar age.

GP6 was found to be significantly higher in individuals with a high ten-year cardiovascular risk and CVDs, while GP13 was found to be lower (Figs. 2(a)–(d)). The distributions of GP6 and GP13 among the low CVDs risk, high CVDs risk, and with CVDs groups are shown in Table S6 in Appendix A. Moreover, a correlation analysis revealed that GP6 was positively correlated and GP13 was negatively correlated with calendar age (Figs. 2(e) and (f)). The correlation coefficient was 0.899 between GlyCage and calendar age (Fig. 2(g)). The GlyCage and calendar age among the low ten-year cardiovascular risk group (57.71 vs 58.28) and without-CVDs group (59.40 vs 59.40) were similar, while GlyCage was higher in the high CVDs risk group (63.84 vs 62.34) and with-CVDs group (63.19 vs 61.75), as shown in Table S7 in Appendix A. 

《Fig. 2》

Fig. 2. Correlation of GlyCage and IgG N-glycans with calendar age. (a, b) Distribution of GP6 and GP13 among 1358 individuals without CVDs stratified by ten-year cardiovascular risk. (c, d) Distribution of GP6 and GP13 among 1465 individuals stratified by CVDs. (e, f) Correlation of GP6 and GP13 with calendar age. (g) Correlation between GlyCage and calendar age.

《3.4. Association of GlyCage index with cardiovascular risk》

3.4. Association of GlyCage index with cardiovascular risk

In the adjusted logistics model, a one-year increase of GlyCage was significantly associated with a high ten-year cardiovascular risk and a high probability of existing CVDs; the ORs were 1.089 (95% confidence interval (95% CI): 1.050–1.131) and 1.068 (95% CI: 1.013–1.125), respectively, independent of calendar age (Table 2). People with a high ten-year cardiovascular risk or CVDs had an excessive GlyCage (Fig. S3 in Appendix A). Individuals were divided into three groups (excessive, equivalent, and regressive GlyCage), according to the distance between GlyCage and calendar age. People with excessive GlyCage (exceeding the calendar age by > 3 years) had the highest cardiovascular risk and probability of CVDs; the adjusted ORs were 2.220 (95% CI: 1.411–3.533) and 2.713 (95% CI: 1.253–6.406), respectively, compared with those with decreased GlyCage, as shown in Fig. 3 and Table 3.

《Table 2》

Table 2 Association of GlyCage index with high cardiovascular risk and CVDs beyond calendar age.

OR: odds ratio.

a Analyzed among individuals without CVDs (n = 1358); high risk refers to a Framingham ten-year cardiovascular risk ≥ 20%.

b Whole population (n = 1465); CVDs included angina, myocardial infarction or heart attack, TIA, stroke, experience of coronary angioplasty or stent, coronary bypass, and implant of cardiac pacemaker.

c Education level, marital status, and CRP adjusted in Model 2.

《Fig. 3》

Fig. 3. Proportion of high cardiovascular risk and CVDs stratified by the distance between GlyCage and calendar age. The distance between GlyCage and calendar age was grouped into ≤ –3 years (group 1), –2 to 2 years (group 2), and ≥ 3 years (group 3). (a) Proportion of high cardiovascular risk among 1358 individuals without CVDs; high risk refers to a ten-year cardiovascular risk ≥ 20%; the proportions of high risk were 13.6%, 27.5%, and 43.7% for groups 1–3. (b) CVDs included angina, myocardial infarction or heart attack, TIA, stroke, experience of coronary angioplasty or stent, coronary bypass, and implant of cardiac pacemaker; the proportions of CVDs were 3.1%, 7.6%, and 10.6% for groups 1–3.

《Table 3》

Table 3 Association of excessive GlyCage with cardiovascular risk and CVDs beyond calendar age.

nhCVD: number of patients with high CVDs risk; nCVD: number of patients with CVDs; ntotal: number of corresponding total patients.

The distance between GlyCage and calendar age was grouped into ≤ –3 years (regressive), –2–2 years (equivalent) and ≥ 3 years (excessive).

a Analyzed among individuals without CVDs (n = 1358) for association between excessive GlyCage and high cardiovascular risk (ten-year cardiovascular risk ≥ 20%).

b Analyzed among whole population (n = 1465) for association between excessive GlyCage and CVDs including angina, myocardial infarction or heart attack, TIA, strok, experience of coronary angioplasty or stent, coronary bypass, and implant of cardiac pacemaker.

c Adjusted for education level, marital status, and CRP level.

In addition, the AUC value of the people identified to have high ten-year cardiovascular risks was 0.726 (95% CI: 0.698–0.754) for GlyCage index, which was significantly higher than those for calendar age (0.653 (95% CI: 0.621–0.685); P < 0.001 for the DeLong test) and GlycanAge (0.686 (95% CI: 0.658–0.715); P = 0.041 for the DeLong test), as presented in Fig. 4. The AUC value of the people identified to have CVDs was 0.645 (95% CI: 0.593–0.696) for GlyCage index; this was insignificantly higher than those for calendar age (0.626 (95% CI: 0.571–0.681); P = 0.184 for the DeLong test) and GlycanAge (0.627 (95% CI: 0.575–0.679); P = 0.551 for the DeLong test).

《Fig. 4》

Fig. 4. Discrimination capacity of GlyCage, calendar age, and GlycanAge for cardiovascular risk and CVDs. (a) Capacity for discriminating high cardiovascular risk among 1358 individuals without CVDs, where high risk refers to a ten-year cardiovascular risk ≥ 20%. (b) Capacity for discriminating CVDs including angina, myocardial infarction or heart attack, TIA, stroke, experience of coronary angioplasty or stent, coronary bypass, and implant of cardiac pacemaker. GlyCage = (0.29640 * (X1 – 5.81422) – 5.88346 * (X2 – 0.22277))/0.16601 + V; GlycanAge = 139.9 + 85.1 * X1 – 5.2 X1 *X1 – 34.6 * X3 + 11.8 * X4 (male)/= 110.0 + 164.5 *X1 – 46.7 X1 *X1 – 22.4 * X3 – 1.9 * X4 (female); where X1 is the expression level of GP6, X2 is the expression level of GP13, X3 is the expression level of GP14, X4 is the expression level of GP15, and V is calendar age.

《4. Discussion》

4. Discussion

Tools for the risk assessment and prediction of CVDs have been developed using clinical features. For example, the World Health Organization group derived and calibrated risk-prediction models to estimate CVDs risk in 21 Global Burden of Disease regions [28]. The pooled cohort equation was developed among US cohorts and is central in the development of prevention guidelines for CVDs [29]. The Systemic Coronary Risk Estimation project and Prediction for Atherosclerotic Cardiovascular Disease Risk project developed tools to predict the ten-year risk of CVDs for European [30] and Chinese [31] populations, respectively. In addition, genomic data and polygenic scores were recently applied to improve the accuracy of prediction tools [32–34], and inflammatory biomarkers were applied to reflect cardiovascular aging [17].

Against this backdrop, the present study is the first to develop a cardiovascular health index, GlyCage index, to track cardiovascular risk using IgG N-glycan profiles, which can identify cardiovascular age beyond calendar age. An excessive GlyCage was found to be significantly associated with higher cardiovascular risk; thus, GlyCage index offers a practical and comprehensible index based on glycomic analysis for cardiovascular risk stratification in a real-world setting. Compared with calendar age, GlyCage index achieved a significant improvement in AUC (0.073, P < 0.001) for ten-year cardiovascular risk and an insignificant improvement (0.019, P = 0.184) for CVDs—probably due to the insufficient statistical power. In parallel, the addition of the polygenic risk score to a baseline model including age and sex only improved the AUC by 0.024 for coronary artery disease prediction, which suggests the practical application of GlyCage index as an important supplement to cardiovascular risk stratification [35].

GlyCage index has great potential to promote cardiovascular risk stratification and inform clinical decision-making for primary prevention. A large proportion of residue cardiovascular risk still exists beyond the clinical-derived risk. Polygenic risk scores reflect the innate probability of CVDs, while GlyCage index may well inform people of the real-time risk based on a simple panel of two glycans [36]. Among individuals whose guideline-based recommendations are unknown, those with an excessive GlyCage can be recommended to initiate positive lifestyles changes. We hypothesize that the combination of GlyCage index, clinical risk, and polygenic risk could largely improve the precision prevention of CVDs and reduce the disease burden, considering the relatively low efficiency of single-aspect variables and the high prevalence of unhealthy lifestyles. Individuals with both excessive GlyCage and high clinical risk may urgently need intensive prevention and management.

Through application of the LASSO and RFE ML algorithms, we selected the intersection of N-glycans among 24 IgG GPs detected via UPLC. Combining LASSO and RFE with multi-stage stepwise regression for N-glycans selection in this study made it possible to avoid the overfitting phenomenon and achieve a stable estimation for the glycan effect [37]. Identifying the most relevant IgG N-glycans played a crucial role in developing the accurate prediction tool GlyCage index for the assessment and management of cardiovascular risk. Thus, IgG N-glycan variables selected using advanced ML approaches are more likely to be effective predictors for establishing GlyCage index. On the other hand, generalization of the proposed GlyCage index requires further validation using different cohorts. The coefficients used in the construction of the GlyCage index tool by IgG N-glycans can be adjusted among different populations.

We determined that GP6 and GP13 were significantly associated with ten-year cardiovascular risk and CVDs. An individual’s cardiovascular-risk-corrected age as calculated using GlyCage is equivalent to the age of a person with the same predicted risk and population average expression levels of GP6 and GP13. In brief, the proposed GlyCage index tracks the individual-level deviation from calendar age that is attributable to cardiovascular risk.

The two selected IgG N-glycans comprising GlyCage index (i.e., GP6 and GP13) have also been reported to be associated with the risk factors of CVDs in previous studies. Menni et al. [14] found that the presence of subclinical atherosclerosis pathology was predominantly explained by an increased relative abundance of GP6. Liu et al. [38] reported a lower abundance of GP13 in patients with ischemic stroke. Wang et al. [39] found that a high abundance of GP6 and low abundance of GP13 were significantly associated with hypertension. These reports are consistent with our findings, in which a higher abundance of GP6 and lower level of GP13 were associated with ten-year cardiovascular risk and CVDs.

There are several possible underlying mechanisms linking IgG N-glycosylation and cardiovascular risk. Chronic inflammation plays a crucial role in the development of CVDs by promoting the growth of plaques, loosening plaque in the arteries, and triggering blood clots, which are the main etiologies of coronary heart disease and stroke [40,41]. IgG is the most abundant immunoglobulin in the circulation system [5]. Glycosylation is one of the most common co- and post-translational modifications of proteins [42]. As described previously, IgG N-glycans have a complex-type biantennary structure, which can decisively influence the effector function via modulating the binding affinity of IgG to activate or inhibit Fc receptors (FcRs) to induce the pro- or anti-inflammatory response [43].

Studies have shown that adding bisecting GlcNAc to fucosylated N-glycans can prevent fucosylation and then induce a proinflammatory response. This promotes the activity of antibodydependent cell-mediated cytotoxicity (ADCC) through the enhanced binding affinity of IgG to FcRIIIA on natural killer (NK) cells, macrophages, and neutrophils [44,45]. In addition, studies have reported that agalactosylation and asialylation can regulate the pro-inflammatory effect of IgG via complement system activation and ADCC [46,47]. In the current study, high levels of GP6 were observed in the high cardiovascular risk group. The presence of bisecting GlcNAc inhibits the process of fucosylation, leading to the enhancement of ADCC activity [3]. Moreover, GP6 is agalactosylated and asialylated in structure. The lack of a terminal galactose or sialic acid on GP6 has a pro-inflammatory effect by activating the complement cascade or the ADCC, respectively [48]. By extension, this effect contributes to the resultant inflammatory activity of IgG. Moreover, we observed low levels of galactosylated N-glycans (i.e., GP13) in the high cardiovascular risk group. A decrease in IgG galactosylation prevents the antiinflammatory effect of circulating IgG through the activation of the lectin-initiated complement pathway. The presence of low levels of galactosylated N-glycans being associated with CVDs risk, as observed in the current study, supports the theory that the lectin-initiated complement pathway plays a crucial role in CVDs risk. Overall, the extent of IgG N-glycosylation (i.e., agalactosylated, asialylated, and fucosylated N-glycans with a bisecting GlcNAc) is altered in individuals with high CVDs risk, and the findings of this study and those of other CVDs-related studies have demonstrated that altered IgG N-glycosylation leads to an enhancement in the level of pro-inflammatory activity [38,49]. Taken together, these observations potentially explain why the detection of changes in IgG N-glycosylation even before the emergence of symptoms could be a promising predictive biomarker of CVDs onset.

Furthermore, the production of cytokines and inflammationregulating molecules may potentially mediate the correlation between IgG N-glycosylation pattern and cardiovascular risk. An animal study demonstrated that the anti-inflammatory property of a high IgG galactosylation level is due to the activation of FcRIIB and dectin-1, which prevents the function of pro-inflammatory effectors [50]. These findings indicate that individuals with lower levels of IgG galactosylation are expected to have more proinflammatory antibodies, leading to high levels of immune activation. Moreover, it has been reported that deficient IgG galactosylation may induce the activation of NK cells, and that such alterations are associated with higher levels of interferon- (IFN-) and tumor necrosis factor-α (TNF-α), thereby enhancing a proinflammatory effect [51]. An in vivo study revealed that cytokine IFN- triggers the pathogenicity of IgG by remodeling carbohydrate structure. Other cytokines, such as interleukin (IL)-21 and IL-17A, can switch the role of IgG between pro- and antiinflammation in correlation with IgG glycan patterns [52]. Moreover, several random clinical trials (e.g., IL-6 inhibition with ziltivekimab in patients at high atherosclerotic risk (RESCUE), ASSessing the effect of Anti-IL-6 treatment in Myocardial Infarction (ASSAIL-MI), and Canakinumab Anti-Inflammatory Thrombosis Outcomes Study (CANTOS)) have targeted IL-1β or IL-6 to prevent cardiovascular risk and improve the outcome of CVDs [53–55]. In brief, the effect of an altered IgG N-glycosylation pattern on cardiovascular risk can be partially mediated by specific cytokines and molecules in the immune microenvironment, which warrants further research.

The results of this study should be interpreted in the context of the study’s limitations. First, this is a cross-sectional study without intervention; thus, a causal association between IgG N-glycans and cardiovascular risk cannot be claimed. However, we analyzed the independent effect of GlyCage index on prospective ten-year cardiovascular risk, indicating the capacity of GlyCage index to predict the onset of CVDs. The underlining mechanisms linking IgG N-glycans and CVDs warrant further research, as they may provide potential molecular targets for the prevention and intervention of CVDs. Second, the history of CVDs was self-reported, which could cause recall bias. The effect of the selected IgG N-glycans and the capacity of GlyCage index to assess and predict cardiovascular health require further validation in other populations and cohort studies. 

In summary, this study developed the GlyCage index to track cardiovascular risk beyond calendar age, thereby providing an index for assessing and predicting cardiovascular health based on glycomic analysis.

《Acknowledgments》

Acknowledgments

This study was funded by National Natural Science Foundation of China (8177120753) and China–Australia International Collaborative Grant (NHMRC APP1112767, NSFC 81561128020) to Wei Wang. Zhiyuan Wu was supported by the China Scholarship Council (201908110447). Yulu Zheng and Zheng Guo were supported by the Edith Cowan University Higher Degree by Research Scholarship (ECU-HDR ST10469322 and ST10468211).

《Authors’ contribution》

Authors’ contribution

Study conception and design: Xiuhua Guo, Wei Wang, and Zhiyuan Wu; data collection: Zheng Guo and Yulu Zheng; data analysis and interpretation: Zhiyuan Wu, Yutao Wang, Huiying Pan, Zhiwei Li, and Xia Li; manuscript writing and reviewing: Zhiyuan Wu, Zheng Guo, Yulu Zheng, Lois Balmer, and Haiping Zhang; study supervision: Wei Wang, Xiuhua Guo, Lixin Tao, and Lois Balmer.

《Compliance with ethics guidelines》

Compliance with ethics guidelines

Zhiyuan Wu, Zheng Guo, Yulu Zheng, Yutao Wang, Haiping Zhang, Huiying Pan, Zhiwei Li, Lois Balmer, Xia Li, Lixin Tao, Xiuhua Guo, and Wei Wang declare that they have no competing interests.

《Appendix A. Supplementary material》

Appendix A. Supplementary material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.eng.2022.12.004.