N-Glycosylation Profiles of Immunoglobulin G and Future Cardiovascular Events

Abstract

Background:

Posttranslational glycosylation of immunoglobulin G (IgG) can modulate its inflammatory capacity through structural variations. We examined the association of baseline IgG N-glycans and an IgG glycan score with incident cardiovascular disease (CVD).

Methods:

IgG N-glycans were measured in two nested CVD case-control studies: Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER, NCT00239681, primary prevention; discovery; Npairs=162); and Treating to New Targets (TNT, NCT00327691, secondary prevention; validation; Npairs=397). Using conditional logistic regression, we investigated the association of future CVD with baseline IgG N-glycans and a glycan score adjusting for clinical risk factors (statin treatment, age, sex, race, lipids, hypertension, and smoking) in JUPITER. Significant associations were validated in TNT, using a similar model further adjusted for diabetes. Using the least absolute shrinkage and selection operator regression, an IgG glycan score was derived in JUPITER as a linear combination of selected IgG N-glycans.

Results:

Six IgG N-glycans were associated with CVD in both studies: an agalactosylated glycan (IgG-GP4) was positively associated, while three digalactosylated glycans (IgG-GPs12, 13, 14) and two monosialylated glycans (IgG-GPs18, 20) were negatively associated with CVD after multiple testing correction (overall [FDR]<0.05). Four LASSO-selected IgG N-glycans comprised the IgG glycan score, which was associated with CVD in JUPITER (adjusted hazard ratio [HR] per glycan score standard deviation [SD]: 2.08, 95%CI=1.52–2.84) and validated in TNT (adjusted HR per SD=1.20, 95%CI=1.03–1.39). The area under the curve (AUC) changed from 0.693 for the model without the score to 0.728 with the score in JUPITER (PLRT=1.1×10–6) and from 0.635 to 0.637 in TNT (PLRT=0.017).

Conclusions:

An IgG N-glycan profile was associated with incident CVD in two populations (primary and secondary prevention), involving an agalactosylated glycan associated with increased risk of CVD, while several digalactosylated and sialylated IgG glycans associated with decreased risk. An IgG glycan score was positively associated with future CVD.

Graphical Abstract

Introduction

The value of posttranslational modifications of proteins on cardiovascular disease (CVD) has received increasing attention^1–4 due to the growing body of evidence showing their involvement in human diseases, such as chronic inflammation,⁵ cancers,⁶ diabetes,⁷ and hypertension.⁸ Posttranslational modification refers to the transformation or modification of specific amino acid residues after biosynthesis, which can influence the activity or expression level of proteins.⁹ Glycosylation is one of the most important posttranslational modifications ¹⁰ and refers to the enzymatic attachment of complex carbohydrates, i.e., glycans, to specific amino acids in proteins.

Glycans are sugar moieties critical for physiological and pathological cellular functions with crucial structures for cell-to-cell and cell-to-extracellular environment interactions, including folding, localization, signaling, and stability.^5,11,12 The elucidation of the structure and activity of glycans is of such importance that the 2022 Nobel Prize in chemistry was awarded for glycan-based science.¹³ A common attachment of glycans to proteins is to the amide nitrogen of an asparagine residue (i.e., N-linked). Among the most abundant circulating N-glycoproteins,^1,11 N-glycans represent an integral part of the Immunoglobulin G (IgG) molecule and impact structural stability, conformation, half-life, aggregation, and effector functions of IgG through their structural variation.^14,15 These IgG properties depend on the presence, absence, and composition of the attached glycans¹⁴ (Supplemental Figure S1 and Supplemental Table S1). For example, IgGs lacking glycans have a reduced level or a complete loss of function due to their interactions with specific cellular activating or inhibitory immunoglobulin‑γ Fc receptors.^16,17

Glycans have a role in atherosclerosis,¹⁸ a chronic progressive inflammatory disease of the artery walls.^19,20 At the same time, alterations in IgG glycosylation patterns can modulate IgG receptor binding, directly impacting inflammatory properties.^3,21 Since inflammation is both a cause and an aggravating factor in cardiometabolic disorders and CVD, as well as a mediator of a worse prognosis, certain IgG glycosylation profiles may characterize an at-risk phenotype for CVD development. Therefore, we hypothesized an association between IgG glycan traits and CVD occurrence. Understanding how differences in the IgG N-glycosylation profile relate to CVD events can lead to new insights about potential residual risk factors and novel prevention approaches that leverage glycobiological mechanisms and pathways.

Since there is a lack of prospective studies evaluating this relationship, the goal of this study was to identify an IgG N-glycosylation profile associated with future CVD events. Specifically, we aimed to 1) examine the association of baseline individual IgG N-glycans and derived IgG N-glycosylation traits with future CVD events and 2) investigate whether an IgG glycan score is associated with CVD risk and improve model performance in two different nested case-control studies.

Methods

Data Availability

Because the data collected for this study was nested within two randomized clinical trials, requests to access the dataset from qualified researchers trained in human subject confidentiality protocols should be sent to the Steering Committees of the parent trials.

Study populations

The present study used two nested case-control studies from primary and secondary prevention for a discovery-validation study design. The discovery population, a primary prevention cohort (N=162 case-control pairs with sufficient plasma sample volume for the IgG glycan analysis, matched for sex and age [±2 years]), was selected based on incident CVD cases from the Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin²² (JUPITER; ClinicalTrials.gov: NCT00239681). Briefly, JUPITER was a prospective (median follow-up 1.9 years), randomized, double-blind, and placebo-controlled trial that tested 20 mg rosuvastatin vs. placebo for primary CVD prevention in 17,802 participants with elevated high-sensitivity C-reactive protein (hs-CRP 2 mg/L or higher) and average to low levels of low-density lipoprotein cholesterol (LDL-C < 130 mg/dl). Participants provided written informed consent at the time of enrollment. Institutional review board approval was obtained from Mass General Brigham (Boston, MA). The first and senior authors had full access to all data in the study and took responsibility for their integrity and data analysis.

Significant associations from JUPITER were then validated in a sub-study of the Treating to New Targets²³ (TNT; ClinicalTrials.gov: NCT00327691) trial, a prospective (median follow-up of 4.9 years), randomized, multicenter study comparing the efficacy of high-dose (80 mg) vs. usual-dose atorvastatin (10 mg) for the secondary prevention of cardiovascular events in 10,001 randomized patients with clinically evident coronary heart disease. A total of 397 nested CVD case-control pairs of participants from the main trial were matched for low- or high-dose statin therapy and a disease risk score (comprised of 17 cardiovascular clinical and biomarker risk factors ²⁴). All patients gave written informed consent, and the study was approved by the local research ethics committee or institutional review board at each center and by the Mass General Brigham institutional review board.

IgG N-glycan profiling and glycan data preparation

Paired case-control plasma samples from each study were placed in neighboring wells in random order throughout 96-well plates. Additionally, for quality control (QC) and to avoid experimental biases, each plate contained 5 wells with duplicated samples from the same plate, 5 wells with duplicates from other plates, and 5 wells with aliquots of standard plasma sample (pooled plasma from healthy volunteers) to further control the repeatability of the procedure. Laboratory personnel were blinded to case status.

IgG was isolated from individual plasma samples using CIM^® r-Protein G LLD 0.2 mL Monolithic 96-well Plate (2 μm channels) (catalog no. 120.1012–2, Sartorius BIA Separations),²⁵ while N-glycans were released by peptide:N-glycosidase F (PNGase F, catalog no. V4831, Promega) and labelled with a fluorescent dye, 2-aminobenzamide (catalog no. A89804, Sigma-Aldrich), as previously described in detail.^26,27 Prepared samples were sent to the processing laboratory where they were stored at −20˚C until ultrahigh-performance liquid chromatography analysis (UPLC) was performed on a Waters ACQUITY UPLC H-Class instrument (Supplemental Methods). All chromatograms were separated in the same manner into 24 IgG glycan peaks (IgG-GPs), and the amount of glycans in each peak was expressed as the percentage of the total integrated area (Supplemental Figure S1 and Supplemental Table S1). Baseline plasma samples were analyzed for all case-control pairs. Participants with 1-year follow-up measurements (128 pairs from JUPITER and 363 pairs from TNT) were used for sensitivity analysis.

Batch effects were evaluated by analyzing inter-plate variability (coefficient of variation [CV] Supplemental Table S2) of standard samples and visually inspection of scatterplots of the samples with standards before and after batch correction. We used an empirical Bayes method (ComBat; R-package sva²⁸) for batch effects correction of percent normalized and log-transformed data due to the right-skewness of IgG-GP distributions. This correction model includes each sample’s position and plate as explanatory variables, and the output is a dataset from which estimated batch effects were removed.

In addition to 24 directly measured IgG-GPs, 8 IgG N-glycosylation traits were derived using the exponential of normalized and batch-corrected glycan measurements:⁷ agalactosylation (G0), monogalactosylation (G1), digalactosylation (G2), asialylation (S0), monosialylation (S1), disialylation (S2), bisecting N-acetylglucosamine (GlcNAc) (B), and core fucosylation (CF). Derived traits were calculated as sums of relative areas of glycans with specific common structural features and denoted the percentage of glycans with said features in the total IgG N-glycome.⁷

Cardiovascular disease outcomes

JUPITER cases were defined as incident myocardial infarction, stroke, coronary revascularization, unstable angina requiring hospitalization, or death. In TNT, CVD was defined as nonfatal non–procedure-related myocardial infarction, resuscitation after cardiac arrest, fatal or nonfatal stroke, and coronary heart disease death. All events were prospectively ascertained and confirmed by medical review by the respective clinical trial endpoint committees.^22,23

Clinical and biomarker risk factors

Baseline questionnaires were used to collect sex, age, race/ethnicity, use of non-randomized supplements or medications, smoking, and other relevant aspects of health history. Body weight and height were measured during physical examination by the study personnel, and standard fasting lipid panels were obtained in a central laboratory as part of each clinical trial. LDL-C concentrations were calculated by the Friedewald equation when triglycerides were <400 mg/dL and measured by ultracentrifugation when ≥400 mg/dL.^24,29,30 In JUPITER, hs-CRP was measured using a high-sensitivity assay (Behring Nephelometer).³⁰ In TNT, hs-CRP was measured at Quest Diagnostics (San Juan Capistrano, CA) using a nephelometric method with latex particles coated with CRP monoclonal antibodies. In JUPITER, we also used nuclear magnetic resonance spectroscopy (NMR) to measure a glycoprotein acetylation biomarker (GlycA) on acute phase reactants. GlycA signals were quantified at LipoScience Inc (Raleigh, NC) from plasma nuclear magnetic resonance spectra obtained from the automated NMR Profiler system.³¹ We also measured secretory phospholipase A2 (sPLA2) at Quest Diagnostics Nichols Institute (San Juan Capistrano, CA) with a commercially available enzyme immunoassay (Cayman assay, Cayman Chemical Co. Ann Arbor MI) based on a double-antibody sandwich technique that is specific for sPLA2-IIA.³² Concentrations of lipoprotein-associated phospholipase A2 (LpPLA2) mass were determined by a latex particle–enhanced turbidimetric immunoassay for LpPLA2 (PLAC^™ test, diaDexus) run on the Roche P-modular analyzer. LpPLA2 activity was measured in a research-use automated enzyme assay system, run on the Roche P-modular analyzer with a colorimetric substrate that is converted upon hydrolysis by the phospholipase enzyme (CAM, diaDexus).³³

Statistical analyses

Outliers for each IgG-GP (< or >6 SD from the mean) were examined and truncated to mean ± 6SD.² Individual IgG N-glycans and derived IgG glycosylation traits risk associations with CVD were standardized to mean = 0 and scaled to SD = 1 to allow for comparison of the effect estimates. Throughout this study, we used conditional logistic regression models stratified for matching factors (clogit; R package survival³⁴). In the discovery cohort (JUPITER), matched for caliper age (±2years) and sex, the minimally adjusted model (model 1) included age (continuous variable) and statin randomization assignment (statin therapy or placebo). Model 2 was further adjusted for race, LDL-C, high-density lipoprotein cholesterol (HDL-C), hypertension, and current smoking. Model 3 was adjusted for the variables in model 2 plus body mass index (BMI) and hs-CRP. Using the same models, we exploratorily examined if there was effect modification by age (above and below the median), sex, or aspirin use. In the validation cohort (TNT), CVD cases and controls were matched on a disease risk score ²⁴ and statin randomization arm (low vs. high statin dose). Regression model 1 included age, sex. Model 2 was adjusted for the same variables in model 1 plus race, LDL-C, HDL-C, hypertension, current smoking, and diabetes. Model 3 additionally included BMI and hs-CRP.

Since we used risk set sampling for CVD case-control selection in both studies, the results were reported as hazard ratios (HR) and 95% confidence intervals (95% CIs).³⁵ All regressions were controlled for multiple testing using a two-stage procedure proposed by Benjamini & Yekutieli,³⁶ setting false discovery rate (FDR) levels at 0.2 for each stage. Briefly, this approach guarantees an overall FDR correction < 0.05 by the multiplication of the FDR levels from each stage. Then, IgG-GPs or derived IgG N-glycosylation traits significantly associated with CVD in JUPITER according to this criterion (FDR < 0.2) were carried over for validation in TNT. Interaction analysis multiple comparisons were adjusted using Bonferroni correction.

To gain physiological insights about these glycans, we performed sex- and age-adjusted partial Spearman correlation analysis (partial_Spearman; R package PResiduals³⁷) with LDL-C, HDL-C, total cholesterol (TC), triglycerides, glucose, hs-CRP, and BMI in both sub-studies, and with GlycA, sPLA2, LpPLA2 mass, and LpPLA2 activity, available in JUPITER. Furthermore, to examine the reproducibility of IgG-N-glycans, we additionally performed partial Spearman correlation analysis between baseline and year-1 glycan readings, adjusting for age and sex. All correlations analyses were restricted to non-cases participants from the placebo group in JUPITER and non-cases from low-dose statin in TNT, to avoid results driven by underlying relationships with future CVD or statin treatment.

IgG glycan score

An IgG glycan score was derived in JUPITER using a linear combination of standardized IgG N-glycans associated with CVD risk. First, for variable selection and dimension reduction, we used the least absolute shrinkage and selection operator (LASSO) approach adapted for case-control studies which fits a conditional-logistic regression model that regressed the CVD outcome on all GPs simultaneously, applying a penalty on the magnitude of regression coefficients to achieve sparse variable selection³⁸ (R-package clogitL1³⁹). GPs with non-zero LASSO regression coefficients were selected if they were significantly associated with CVD in a conditional regression model adjusted for age, statin randomization, race, LDL-C, HDL-C, hypertension, and current smoking. The optimal tuning parameter for the penalty term was selected via cross-validation. Significantly associated GPs (p<0.05) were included simultaneously in a new model adjusted for age and statin randomization to obtain the coefficients for the IgG glycan score. The glycan score formula learned in JUPITER was applied to TNT. Glycan scores were calculated in both JUPITER and TNT populations and their distributions were centered to mean = 0 and SD = 1. The relationship with CVD was examined using the same three conditional logistic regression models previously described for JUPITER and TNT.

Because case-control design does not output absolute risk,⁴⁰ we used a parametric approach to calculate the area under the receiver operating characteristic curve (AUC) for the IgG glycan score. Briefly, using a basic prediction model estimated in the parent trial and adjusted for age, statin randomization, race, LDL-C, HDL-C, hypertension, current smoking, and diabetes, we used a parametric equation to infer AUC for a model additionally adjusted for glycan score. Assumptions were checked using the Shapiro-Wilk’s test and Quantile-Quantile Plots, and non-normal variables were transformed using Yeo-Johnson’s Power Transformation ⁴¹ (R-package VGAM ⁴²). For consistent estimates of 95% confidence intervals in TNT, we performed a bootstrap resampling procedure with 200 bootstrap replicates. To compare results in both sub-studies and avoid over-optimism, AUC in JUPITER sub-study was obtained using 10-fold cross-validation.

All statistical analyses were performed using the R software (v 4.1.0).⁴³ Please see the Major Resources Table in the Supplemental Materials.

Results

Participant characteristics

Table 1 displays baseline characteristics of JUPITER (discovery cohort) and TNT (validation cohort) participants according to CVD case-control status. Within each sub-study, cases and controls were generally well-balanced except for race, BMI, and smoking in JUPITER, and hs-CRP in TNT.

Table 1.

Baseline characteristics of participants in the discovery primary prevention cohort (JUPITER) and the validation secondary prevention cohort (TNT).

Characteristic	JUPITER*
	Overall (N = 324)	Controls (N = 162)	Cases (N = 162)	p-value
Age (years) †	70.0 [64.8, 75.0]	70.0 [64.3, 75.0]	70.0 [65.0, 75.8]	0.947
Women, N (%)	78 (24.1)	39 (24.1)	39 (24.1)	1.000
Race or ethnic group				0.026
White, N (%)	273 (84.3)	143 (88.3)	130 (80.2)
African American, N (%)	27 (8.3)	7 (4.3)	20 (12.3)
Other, N (%)	24 (7.4)	12 (7.4)	12 (7.4)
BMI (kg/m2) †	27.6 [24.9, 30.9]	28.41 [25.6, 31.3]	26.94 [23.9, 30.0]	0.002
Hypertension, N (%)	190 (58.6)	97 (59.9)	93 (57.4)	0.735
Current smoker, N (%)	56 (17.3)	19 (11.7)	37 (22.8)	0.012
HDL-C (mg/dL) †	48 [40, 60]	48 [40, 60]	49 [41, 61]	0.872
LDL-C (mg/dL) †	110 [94, 120]	109 [97, 120]	111 [92, 120]	0.672
TC (mg/dL) †	187 [166, 201]	188 [166, 201]	185 [166, 200]	0.284
Triglycerides (mg/dL) †	114 [82, 166]	114 [81, 173]	114 [82, 165]	0.538
Glucose (mg/dL) †	95 [88.5, 102]	95 [88, 102]	95 [89, 102]	0.652
hs-CRP (mg/L) †	4.4 [2.8, 8.0]	4.4 [2.9, 7.7]	4.4 [2.8, 8.2]	0.865
GlycA (μmol/L) †	406 [362, 459.5]	399 [361.5, 445]	418 [363.5, 471]	0.056
Lp(a) (nmol/L) †	27 [12.5, 58]	25 [12, 45]	32 [13, 88]	0.066
sPLA2 (ng/mL) †	4.4 [2.8, 7.6]	3.7 [2.6, 7.4]	4.7 [2.9, 7.7]	0.166
LpPLA2 mass (ng/L) †	306.2 [254.5, 372.7]	295.7 [247.7, 356.8]	320.3 [255.9, 383.0]	0.149
LpPLA2 activity (nmol/min/mL) †	197.1 [170.1, 232.7]	204.6 [173.9, 232.6]	194.7 [168.0, 233.7]	0.481
Characteristic	TNT‡
	Overall (N = 734)	Controls (N = 397)	Cases (N = 397)	p-value
Age (years) †	64.5 [57.2, 70.0]	63.8 [56.9, 70.0]	64.7 [57.9, 69.7]	0.545
Women, N (%)	123 (16.8)	53 (14.4)	70 (19.1)	0.114
Race or ethnic group				0.916
White, N (%)	673 (91.7)	337 (91.8)	336 (91.5)
Black, N (%)	28 (3.8)	13 (3.5)	15 (4.1)
Other, N (%)	33 (4.5)	17 (4.6)	16 (4.4)
BMI (kg/m2) †	28.4 [25.8, 32.0]	28.2 [25.6, 31.2]	28.6 [25.9, 32.3]	0.070
Diabetes, N (%)	168 (22.9)	86 (23.4)	82 (22.3)	0.792
Hypertension, N (%)	505 (68.8)	251 (68.4)	254 (69.2)	0.873
Current smoker, N (%)	123 (16.8)	59 (16.1)	64 (17.4)	0.693
Disease risk score ||	0.6 [0.2, 1.1]	0.6 [0.2, 1.1]	0.6 [0.2, 1.1]	0.995
HDL-C (mg/dL) †	44 [38, 50]	43 [38, 50]	44 [38, 51]	0.938
LDL-C (mg/dL) †	98 [87, 111]	98 [86, 111]	98 [88, 110]	0.840
TC (mg/dL) †	175 [158, 191]	176 [158, 192]	174 [160, 190]	0.754
Triglycerides (mg/dL) †	140 [107, 199]	140 [108, 199]	140 [107, 200]	0.692
Glucose (mg/dL) †	102 [93, 118]	102 [94, 119]	102 [93, 118]	0.778
hs-CRP (mg/L) †	0.8 [0.3, 2.2]	0.7 [0.2, 2.1]	0.9 [0.4, 2.4]	0.036

^*JUPITER case-control matching was for age and sex.

^†Median [IQR].

^‡TNT case-control matching was for high vs. low intensity statin randomization and disease risk score.

^||Calculated based on age, sex, smoking, hypertension, diabetes, BMI, myocardial infarction, angina, cerebrovascular disease, peripheral vascular disease, congestive heart failure, coronary revascularization, apolipoprotein B, use of calcium channel blockers, use of aspirin, and baseline blood urea nitrogen.²⁴

Abbreviations: BMI – body mass index; LDL-C– low density lipoprotein cholesterol; HDL-C– high density lipoprotein cholesterol; TC – total cholesterol; hs-CRP – high-sensitivity C-reactive protein. P-values were estimated by Mann-Whitney U test for continuous variables and X² test for categorical variables.

Associations of individual IgG-GPs with CVD outcomes

Figure 1 displays HRs and 95% CI for the association of CVD events with each IgG-GP, adjusted for model 2 covariates. Nine individual GPs were associated with CVD events in JUPITER (first stage FDR < 0.2), which were carried over for validation in TNT. Significant results in validation (second stage FDR < 0.2; overall FDR <0.05) were detected for IgG-GP4, IgG-GP12, IgG-GP13, IgG-GP14, IgG-GP18, and IgG-GP20 and are highlighted in the figure. Supplemental Table S3 displays detailed regression results per IgG N-glycan SD for all three models and shows that model 3 resulted in very similar point estimates and p-values as in model 2 for all IgG-GPs. We detected no significant associations for BMI and hs-CRP in the CVD prediction model. Moreover, an additional analysis further adjusting models 2 and 3 for fasting glucose in both discovery and validation did not change the results or improve model fit (Likelihood Ratio test p-values [p_LRT] > 0.1). All analyses were corrected for multiple testing using overall two-stage FDR<0.05.

Figure 1. IgG-GPs 4, 12, 13, 14, 18, 20 were significantly associated with CVD in both JUPITER and TNT.

The graph displays conditional logistic regression HRs and 95%CIs for CVD events per SD increment in each IgG-GP. Significant results for model 2 in JUPITER (adjusted for age, statin randomization, race, LDL-C, HDL-C, smoking, hypertension) and validated in TNT (the same covariates as in JUPITER plus diabetes) are highlighted in dark turquoise and orange, respectively. The top panel shows the most abundant IgG N-glycan structure in each GP. Analyses were adjusted for multiple comparisons at the two-stage overall FDR level of 0.05 (FDR < 0.2 in each stage).

Supplemental Figure S2 displays the baseline distribution of IgG-GPs relative abundance in cases and controls of each sub-study and Supplemental Table S4 shows IgG N-glycan percent normalized means (SDs). Sensitivity analysis in 1-year follow-up glycan measurements from participants with CVD incidence after 1 year from randomization showed reproducible results compared to baseline predictions in both cohorts (Supplemental Figure S3). Moreover, age- and sex-adjusted partial Spearman correlation analysis of baseline and year-1 glycan readings revealed moderate to high correlations between each IgG-GP with concordant results between the two studies (Spearman rho 0.38 – 0.97 in JUPITER and 0.35 – 0.95 in TNT) (Supplemental Table S5).

Associations of Derived IgG N-glycosylation traits with CVD risk

We examined the association of eight derived IgG N-glycan traits with CVD risk. Derived traits were calculated as sums of relative areas of glycans with specific common structural features and denote the percentage of glycans with said features in the total IgG N-glycome.⁷ Figure 2 displays HRs and 95%CIs for each derived IgG N-glycan trait and CVD, adjusted for model 2 covariates. The results were quite consistent with the observed associations for individual IgG N-Glycans: specifically, agalactosylation and asialylation traits were FDR-adjusted positively associated with CVD risk, while digalactosylation, and monosialylation were inversely associated. Supplemental Table S6 displays the results for all models.

Figure 2. Derived IgG N-glycosylation traits agalactosylation, digalactosylation, asialylation, and monosialylation were significantly associated with CVD in both JUPITER and TNT.

Left panel: Conditional logistic regression HRs and 95%CIs for CVD risk per SD increment in derived IgG N-glycosylation traits. Significant results for model 2 in JUPITER (adjusted for age, statin randomization, race, LDL-C, HDL-C, smoking, hypertension) and validated in TNT (the same covariates as in JUPITER plus diabetes) are highlighted in dark turquoise and orange, respectively. Analyses were adjusted for multiple comparisons at the two-stage overall FDR level of 0.05 (FDR < 0.2 in each stage). *Borderline significant (FDR = 0.20). Right panel: Table for each derived IgG N-glycosylation trait calculation. Total GP: sum of all IgG-GPs.

Association of an IgG glycan score with CVD risk

The IgG glycan score was derived in the discovery cohort, JUPITER. First, seven IgG-GPs were LASSO selected and entered into a conditional logistic regression model to test their association with CVD risk. This model yielded four IgG-GPs significantly associated with CVD risk (p < 0.05): IgG-GP9, IgG-GP12, IgG-GP19, and IgG-GP20 (Figure 3). Then, a new multivariable adjusted conditional logistic regression model including only the four significant GPs plus age and sex was fitted to generate the β-coefficients used as weights for IgG glycan score for CVD risk, calculated as:

Figure 3. LASSO regression selected 7 IgG-GPs out of which 4 were significantly associated with CVD and comprised the IgG glycan score.

Flowchart for JUPITER IgG-GPs selection for the IgG glycan score. *Adjusted for age, statin randomization, race, LDL-C, HDL-C, smoking, and hypertension. ** The IgG glycan score was a linear combination of these 4 IgG-GPs using the β-coefficients from an adjusted model controlled for age and intervention arm, defined as: −0.595*IgG-GP9 −0.272*IgG-GP12 +0.218*IgG-GP19 −0.264*IgG-GP20.

$$-0.595*\text{IgG-GP9}-0.272*\text{IgG-GP12}+0.218*\text{IgG-GP19}-0.264*\text{IgG-GP20}$$ (Equation 1)

In JUPITER, the IgG glycan score was associated with over two-fold higher CVD risk per SD increment in the score (HR= 2.08, 95%CI= 1.52–2.84, adjusted for age and sex (matching variables), statin randomization, race, LDL-C, HDL-C, smoking, hypertension; Table 2). Using a similar model, but additionally adjusting for diabetes, the score was significantly validated in TNT (HR = 1.20 per SD; 95%CI =1.03 – 1.39; p = 0.018). Due to data collection for the original clinical trials, an identical composite CVD outcome variable across the two studies for CVD events would not be possible. However, we performed sensitivity analysis excluding all-cause death from JUPITER cases (N = 26) to comprise a more similar CVD outcome definition. The result was attenuated and closer to the point estimate found in TNT, but remained highly significant (excluding all-cause death: HR = 1.70; 95%CI=1.24–2.34; p = 0.001; including all-cause death: HR = 2.08; 95%CI= 1.52–2.84; p= 4.5e⁻⁶). Additionally, we provide β-coefficients for score calculation that were derived using glycan values directly provided by the laboratory (non-standardized) (Supplemental Results).

Table 2.

Associations of IgG glycan score and it’s four components included in a multivariable model with CVD events in JUPITER and validation in TNT.

		JUPITER			TNT
IgG-GP	Glycan structure*	Model 1 HR (95% CI); p-value	Model 2 HR (95% CI); p-value	Model 3 HR (95% CI); p-value	Model 1 HR (95% CI); p-value	Model 2 HR (95% CI); p-value	Model 3 HR (95% CI); p-value
IgG glycan score	(Equation 1)	2.14 (1.58–2.89); 9.9e−7	2.08 (1.52–2.84); 4.5e−6	2.03 (1.48–2.8); 1.2e−5	1.20 (1.03–1.39); 0.016	1.20 (1.03–1.39); 0.018	1.20 (1.03–1.4); 0.019
IgG-GP9		0.55 (0.42–0.73); 3.0e−5	0.58 (0.43–0.77); 1.9e−4	0.59 (0.44–0.8); 5.2e−4	0.86 (0.74–1.01); 0.066	0.87 (0.74–1.01); 0.076	0.88 (0.75–1.03); 0.114
IgG-GP12		0.76 (0.59–0.98); 0.032	0.74 (0.57–0.96); 0.025	0.74 (0.56–0.96); 0.024	0.86 (0.74–1.00); 0.053	0.85 (0.73–0.99); 0.04	0.84 (0.72–0.99); 0.039
IgG-GP19		1.24 (0.96–1.61); 0.098	1.20 (0.92–1.56); 0.177	1.23 (0.93–1.62); 0.148	1.01 (0.87–1.17); 0.924	1.01 (0.87–1.17); 0.939	1.04 (0.89–1.22); 0.582
IgG-GP20	(Structure not determined)	0.77 (0.59–1.00); 0.049	0.76 (0.57–1.00); 0.049	0.78 (0.58–1.03); 0.082	0.88 (0.76–1.03); 0.121	0.89 (0.76–1.04); 0.153	0.88 (0.75–1.04); 0.141

HRs (95% CIs) for single IgG-GPs are from multivariable adjusted conditional logistic regression models. The IgG glycan score for CVD risk is the linear combination of the selected N-glycans, weighted by the regression coefficient from the mutually adjusted conditional logistic regression model and centered to mean = 0 and scaled to SD = 1. In JUPITER, model 1 was adjusted for age and statin randomization, model 2 was additionally adjusted for race, LDL-C, HDL-C, smoking, prevalent hypertension, and model 3 was model 2 plus BMI and hs-CRP. Sex was a matching criterion for case-control selection. In TNT, model 1 was adjusted for age and sex, model 2 was additionally adjusted for race, LDL-C, HDL-C, hypertension, current smoking, and diabetes, and model 3 was model 2 plus BMI and hs-CRP. Treatment and disease risk score were matching criteria for case-control selection in TNT.

^*red triangle – Fucose; blue square – N-acetylglucosamine; green circle – mannose; yellow circle – galactose; purple diamond – N-acetylneuraminic acid.

The AUC in TNT changed from 0.635 (bootstrapped 95%CI = 0.622 – 0.653) for a model with standard risk factors (age, sex, statin randomization, race, LDL-C, HDL-C, smoking, hypertension, and diabetes) to 0.637 (bootstrapped 95%CI = 0.627 – 0.663) for a model additionally adjusted for the IgG glycan score (P_LRT = 0.017). Similar models (without diabetes since the JUPITER trial excluded patients with diabetes) were used for a 10-fold cross validation in JUPITER. The mean AUC changed from 0.693 (range of 0.691 – 0.696) to 0.728 (range of 0.718 – 0.750) for models without and with the score, respectively (p_LRT = 1.1×10⁻⁶).

Interaction analyses

Considering IgG-GPs individual associations with CVD events in JUPITER, we detected nominal significant effect modification by sex for IgG-GP20 in model 3 (p = 0.014). Subgroup analysis by sex under the same model showed decreased risk for women (HR = 0.27, 95%CI = 0.11–0.69, p = 0.006) and non-significant association in men (HR = 0.91, 95%CI = 0.68 −1.21, p = 0.511). The results of all three models can be found in Supplemental Figure S4. No significant effect modification by age (above and below the median) was detected for the association between individual IgG-GPs and CVD risk (P > 0.05). Additional interaction analysis with aspirin was performed and no significant interaction was detected for CVD outcome (Bonferroni adjusted p-value > 0.05).

As for derived IgG N-glycosylation traits and IgG glycan score, no significant effect modification by sex, age or aspirin use was detected (P > 0.05).

Spearman correlations of IgG-GPs, IgG glycan score, and derived IgG N-glycosylation traits with clinical biomarkers

Supplemental Figure S5 displays Spearman correlations of IgG glycan score, IgG-GPs, and IgG derived traits with cardiometabolic traits adjusted for age and sex in controls in JUPITER and TNT cohorts. Focusing on CVD-associated GPs and IgG glycan components, we highlight significant and consistent results across studies in terms of sign for IgG-GP4 (positive correlation with hs-CRP and negative correlation with HDL-C), IgG-GP14 (positive correlation with HDL-C), and IgG-GP20 (positive correlation with LDL-C, HDL-C, and TC).

Discussion

The present findings show significant and reproducible associations of IgG N-glycosylation profiles with risk of future CVD events. We detected an IgG N-glycan signature predominantly related to galactosylation and sialylation of IgG that may play a role in CVD risk. This study is the first to demonstrate the potential relevance of these glycans to predict incident CVD events, with a robust discovery-validation study design and utilizing two clinically distinct contemporary populations (primary and secondary prevention). Furthermore, we derived an IgG glycan score that significantly improved CVD model prediction performance independent of standard risk factors and statin therapy, with replicable results in a different secondary prevention population.

A core-fucosylated agalactosylated glycan (FA2; IgG-GP4) was the only IgG N-glycan to show a positive association with CVD in both cohorts. It was also the only one to significantly correlate with hs-CRP in both sub-studies. There is evidence that terminal GlcNAc residues in agalactosylated glycans are exposed and accessible to mannose-binding⁴⁴ since it binds to GlcNAc, glucose and fucose, but not to galactose.⁴⁵ This mannose-binding lectin activation occurs in a pro-inflammatory manner triggering the lectin complement pathway.^44,46 Therefore, current results suggest a pro-inflammatory activity induced by the lack of terminal galactose residues with a potential role in CVD pathophysiology. We also found positive correlations of IgG-GP4 with triglycerides and BMI that are consistent with previous studies.^21,47,48

Galactosylation of IgG acts as a modulator of its inflammatory activity ⁴⁹ and represents an interface between physiological and pathological processes. Moreover, galactosylation is also essential for subsequent sialic acid additions by glycosyltransferases.¹¹ We observed a predominant protective effect of galactosylated and sialylated IgG glycans that potentially have a pivotal role in CVD prevention. Current longitudinal results based on clinical CVD events corroborate findings from a previous cross-sectional study that reported negative associations of a digalactosylated (IgG-GP14; FA2G2) and a monosialylated IgG N-glycan (IgG-GP18; FA2G2S1) with the 10-year atherosclerotic CVD risk score.² This potential atheroprotective property of IgG-GP14 and IgG-GP18 is supported by their consistent positive correlation with HDL-C across both cohorts in the current study. In line, IgG-GP14 and IgG-GP18 were previously found to be cross-sectionally inversely associated with dyslipidemia in comparison to healthy controls⁴⁸, and also associated with lower incidence and progression of complications of diabetes.⁵⁰

As for sialylation, although it has been consistently associated with IgG anti-inflammatory activity,^16,51 terminal sialic acids appear to mainly serve as a switch between IgG pro- and anti-inflammatory activity in cases of homeostasis disturbance.¹¹ In the current study, besides inverse association with CVD, we detected positive correlations for the monosialylated IgG-GP20 with HDL-C, LDL-C, and total cholesterol. We also observed inverse correlation of this glycan with triglycerides and hs-CRP. Previous studies have observed positive associations with total-C in patients with dyslipidemia and pregnant women compared to controls.^48,52 In a cross-sectional study, this glycan has been inversely associated with combined diabetes and hypertension,⁵³ and a Mendelian randomization analysis showed negative causal effect estimation for diabetes.⁵⁴ Collectively, these findings indicate that IgG-GP20 is involved in lipid and glucose metabolism and/or uptake and a favorable cardiometabolic profile.

It is established that IgG-mediated inflammation is crucially dependent on its fragment crystallizable (Fc) interaction with cellular Fc receptors on innate immune effector cells such as mast cells, macrophages, and neutrophils.¹⁷ Cell activation is regulated by the simultaneous expression of activating and inhibitory Fc receptors. Galactosylation of IgG Fc is described to be necessary for the efficient initiation of the anti-inflammatory signaling cascade through binding to FcγRIIB, the only inhibitory Fc receptor for IgG.¹⁷ FcγRIIB is the only IgG Fc receptor expressed by B cells and negatively regulates B cell receptor activation by immune-complexed antigens.⁵⁵ The mechanisms through which sialylation acts are yet to be fully elucidated, but terminal sialic acid on N-linked glycans reduces receptor binding, reducing the pathogenicity of IgG antibodies.¹⁷

Since glycans can regulate IgG antibody-mediated inflammatory activity, both significant and null correlations with hs-CRP might be biologically and physiologically meaningful. As already discussed, IgG-GP4 (CVD positively associated) was the only glycan to correlate with hs-CRP significantly and positively in both studies, reiterating the relationship between inflammation and IgG agalactosylation.^7,21 On the other hand, several IgG-GPs showed non-significant correlation with hs-CRP, and most of them also had no significant association with CVD (Supplemental Figure S5). One interpretation of these findings can be that IgG N-glycans may play a primary role in the inflammatory pathway of CVD. This hypothesis can be supported by a study of obese mice supplemented with N-acetylmannosamine (a precursor for sialic acid) that showed both decreased pro-inflammatory potential of IgG and lower risk of hypertension development.⁵⁶

Noteworthy, a core-fucosylated monogalactosylated glycan (IgG-GP9; FA2[3]G1) and a core-fucosylated agalctosylated glycan with bisecting GlcNAc (IgG-GP6; FA2B) showed the most significant associations with CVD in JUPITER. Nonetheless the results did not replicate in TNT, possibly due to TNT having fewer women than JUPITER. Previously, IgG-GP9 has shown a significantly negative association with incident CVD in women but not in men from a European cohort,⁷ and IgG-GP6 was cross-sectionally positively associated with the 10-year atherosclerotic CVD risk score in a study that only included women.²

The current results on individual N-glycans were concordant with derived traits, showing a predominance of galactosylation and sialylation as IgG N-glycosylation traits that may have important roles in future CVD events. Interestingly, this predominance seems to vary according to the disease. For instance, for type 2 diabetes risk, IgG galactosylation and bisecting GlcNAc levels were the most expressive patterns,⁷ whereas IgG bisecting GlcNAc and fucose were the most significant for incident hypertension.⁸

The association of the IgG glycan score with CVD risk was validated in an independent study with different risk factor profiles. The magnitude of association of the IgG glycan score with CVD in the validation cohort was similar to the risk found for high LDL-C or apolipoprotein B levels in a metanalysis⁵⁷. Moreover, the changes in AUC by adding the IgG glycan score to standard risk factors models were similar to or greater than changes in AUC observed in prior studies that compared models with and without non-standard biomarkers for coronary heart disease or CVD mortality prediction, such as hs-CRP, apolipoprotein B, lipoprotein(a), intima-media thickness, and flow-mediated dilation.^58–68

Correlations between IgG N-glycans and clinical biomarkers were concordant in terms of sign and magnitude across studies for most IgG-GPs, and several were significant in both cohorts (Supplemental Figure S5). Notably, less significant correlations observed in JUPITER than in TNT can be attributable to the sample size (N = 162 vs. N = 397, respectively) and characteristics of the populations (primary vs. secondary prevention). Furthermore, IgG glycans with no significant correlation in either sample may also be relevant by suggesting the involvement of distinct pathways.

Our longitudinal study is the first to derive and replicate N-glycan associations with CVD using two completely independent cohorts with a uniquely derived and validated IgG glycan risk score. JUPITER was a CVD primary prevention study of individuals with high levels of hs-CRP and low LDL-C, whereas TNT studied a secondary prevention coronary artery disease population and both populations were well treated with statins. Furthermore, we highlight that the biomarker discovery associated with CVD in the validation cohort, TNT, was independent of the covariates included in the models, and an extensive set of potential residual risk factors were incorporated in the disease risk score,²⁴ used as one of the case-control matching criteria.

The current CVD case-control sampling depended on sufficient plasma availability for glycan profiling; nonetheless, the samples were representative of the parent trial populations. Another limitation is an underrepresentation of ethnicities other than white, which may restrict the generalizability of the findings. Research on IgG glycobiology and its relationship with cardiometabolic risk in other populations, especially in diverse ethnicities, is warranted.

In summary, our results revealed that baseline IgG N-glycosylation profiles are associated with the risk of future CVD, whether considered as individual IgG N-glycans, grouped (derived traits), or weighted (glycan score) levels. Current findings indicate that an IgG N-glycan biosignature related to galactosylation and sialylation may play a role in CVD risk by regulating the pro- or anti-inflammatory responses of IgG. We also detected an IgG glycan score that was positively associated with future CVD events and improved model prediction performance. Therefore, IgG N-glycans might be not only novel biomarkers for cardiometabolic health, but also potential new drug targets. The glycome represents a promising and underappreciated class that can be used for risk stratification, CVD prevention, diagnostics, and treatment purposes. Looking into the future, IgG glycome warrants further research in other population samples with a view to applications in cardiology and public health practice.

Novelty and Significance.

What is known?

• Glycans are involved in a variety of biological pathways that may be relevant for atherosclerosis and cardiovascular events.

• Alterations in Immunoglobulins G (IgGs) glycosylation patterns can modulate receptor binding, directly impacting inflammatory properties.

What new information does this article contribute?

• IgG N-glycosylation profiles reflecting galactosylation and sialylation patterns, were consistently associated with the risk of future CVD and may play a role in CVD risk by regulating the pro- or anti-inflammatory responses of IgG.

• An IgG glycan score was positively and consistently associated with incident CVD with up to 2-fold increased risk of CVD independent of known CVD risk factors and improved model performance.

Glycans are sugar moieties critical for physiological and pathological cellular functions, including folding, localization, signaling, and stability. Understanding the relationship between the glycome, in particular IgG N-glycans, and incident CVD can contribute new insights to the fields of glycobiology and cardiovascular prediction risk, making glycans promising novel biomarkers for cardiovascular disease. In this study, we detected an IgG N-glycan biosignature related to galactosylation and sialylation that potentially plays a role in CVD risk by regulating the pro- or anti-inflammatory responses of IgG. The glycome represents a promising and underappreciated class and should be further investigated for CVD prevention, diagnostics, and potential therapies.

Non-standard Abbreviations and Acronyms

CVD

Cardiovascular disease

B

Bisecting N-acetylglucosamine

BMI

Body mass index

CF

Core fucosylation

FDR

False Discovery Rate

G0

Agalactosylation

G1

Monogalactosylation

G2

Digalactosylation

GlycA

Glycoprotein acetylation

HDL-C

High-density lipoprotein cholesterol

HR

Hazard ratio

hs-CRP

High-sensitivity C-reactive protein

IgG

Immunoglobulin class G

IgG-GPs

Immunoglobulin class G glycan peaks

JUPITER

Justification for the Use of statins in Prevention: an Intervention Trial Evaluating Rosuvastatin

LASSO

Least absolute shrinkage and selection operator

LDL-C

Low-density lipoprotein cholesterol

LpPLA2

Lipoprotein-associated phospholipase A2

LRT

Likelihood ratio test

S0

Asialylation

S1

Monosialylation

S2

Disialylation

sPLA2

Secretory phospholipase A2

TC

Total cholesterol

TNT

Treating to New Targets