Large-scale lipidomics profiling reveals characteristic lipid signatures associated with an increased cardiovascular risk

One-thousand and fifty-seven (1057) patients were enrolled in this prospective, consecutive study (Table 1). All patients underwent catheter angiography within 24 h after hospital admission and were treated for symptomatic CAD. Blood sampling was performed according to a standardized protocol from peripheral venipuncture with patients fasting overnight for 12 h. Platelet isolation and preparation for mass spectrometry analysis was performed as previously described [15, 16].

All patients completed a standardized questionnaire on smoking status, previous dietary factors and medication history. Further, medication on admission was assessed and patients received antiplatelet therapy and statin treatment according to current CAD guidelines and disease severity. Cardiovascular risk factors were assessed according to the Framingham score and outlined in the supplementary methods section. A clinical follow-up of cardiovascular events was performed over a 3-year period after hospital discharge and relied on a review of medical records including documentation of hospital or general practitioners. Alongside major bleeding events, the composite cardiovascular endpoint included cardiac death, myocardial infarction and ischaemic stroke. Bleeding events were categorized according to ISTH score as defined in the supplementary methods section. The study was approved by the local ethics committee of Tübingen (270/2011B01) and all patients gave written informed consent. The experiments were performed in accordance with the highest ethical standards and comply with the Declaration of Helsinki.

Preparation of platelets and analysis by mass spectrometry was performed as previously described [15] and summarized in the supplementary methods section.

Analysis of platelet function was performed using whole blood impedance aggregometry (Multiplate) after stimulation with collagen, arachidonic acid, and ADP as previously described [9].

Patients baseline characteristics data and pre-processed lipidomics data were analysed using JMP^® Version 16.2 (SAS Institute, Cary, North Carolina, USA). Normally distributed data were analysed using Student’s t test, non-normally distributed data were compared using the Mann–Whitney U test. Mean values are presented as mean ± standard deviation (SD) or as median and interquartile range (IQR) where applicable. Categorical parameters were compared using Chi-square. Orthogonal partial least square discriminant analysis (OPLS-DA) and principal component analysis (PCA) of all identified lipids were performed after normalization and boxplots of the first principal components were displayed with Mann–Whitney U test. Concentrations of platelet lipids were analysed by Student’s t test and to account for the multiple hypothesis testing, a false discovery rate (FDR) controlling procedure was further adopted to correct significance levels (p < 0.05) for FDR ≤ 5%. Therefore, Benjamini–Hochberg procedure (BH) was implemented for testing of significant alterations between groups. For subgroup comparison, patients were grouped into quartiles and dichotomized according to mean platelet concentrations, where applicable. Significantly altered lipids were further analysed by Cox proportional hazards models to assess the relationship of distinct lipids with the cardiovascular risk. Therefore, hazards ratios were tested for a 1-standard deviation (SD) higher lipid concentration by analysing the significance level of the correlation coefficient between Kaplan–Meier transformed survival time and scaled Schoenfeld residuals. Adjustment for age, gender and statin medication was performed. L₁-regularized Cox regression with a least absolute shrinkage and selection operator (LASSO) algorithm was implemented for lipid selection and optimal λ was chosen by tenfold cross-validation and repeated with variant partitions to account for the potential variance of cross-validation. Likewise, adjustment for age, gender and statin medication was implemented. Further assessment of the association of selected lipids with an increased cardiovascular risk was performed by calculating Harrell’s concordance index (C index). Estimators of incremental predictive values and the referring confidence intervals were computed with cross-validation and percentile bootstrap analysis. PCA, OPLS-DA, Cox proportional hazards models, LASSO and C index were computed using different software packages in RStudio (RStudio Inc., Boston, USA) and are further elucidated in the supplementary method section. Correlation data are based on Pearson’s product-moment correlation coefficient (r) and Spearman’s rank correlation coefficient (R). Graphic output was performed with different software packages including RStudio and JMP.

Previously, we characterized changes of the platelet lipidome in patients with coronary artery disease (CAD) and showed that significant alterations occur in patients with acute coronary syndrome (ACS) when compared to chronic coronary syndrome (CCS) or healthy controls [12, 15]. In the present study, we prospectively analysed the platelet lipidome in a large cohort of patients with symptomatic CAD utilizing an untargeted UHPLC–ESI-QTOF-MS/MS approach to elucidate associations of distinct lipids with an increased cardiovascular risk. Baseline demographic, clinical and laboratory characteristics of the 1057 patients are shown in Table 1. Over a median follow-up period of 36 months of 1057 patients (2987 person-years), 72 individuals (6.8%) experienced a major cardiovascular (cardiac death, ischaemic stroke, myocardial infarction) or bleeding event corresponding to an incidence rate of 24.1 (95% CI 18.9–30.4) per 1000 person-years (Table 2).

After pre-processing and structural annotation of the lipids detected by MS, we could verify 767 lipids from isolated circulating platelets (Fig. 1A). The identified platelet lipidome of this study comprised lipids attributable to 40 subclasses, which can be summarized into six main lipid categories: glycerophospholipids (55.8% of verified lipids), glycerolipids (15.9%), fatty acyls (9.2%), sphingolipids (17.1%), sterol (1.6%) and prenol (0.4%) lipids. Further structural analysis of the platelet lipid composition highlighted a predominance of lipids with carbon chain length between 25 and 45 with mainly poly-unsaturated fatty acid (PUFA) sidechain and less monounsaturated (MUFA) or saturated (SFA) sidechains (Fig. 1B).

To elucidate characteristic changes in the platelet lipidome associated with cardiovascular events, we compared the mean of relative concentrations of distinct lipids and their referring lipid subclasses between patients with and without adverse cardiovascular events during the follow-up period. We found that 18 lipids were significantly (p < 0.05, FDR < 5%) upregulated in patients with adverse cardiovascular events compared to patients without events (Fig. 2A). Significantly upregulated lipids mostly comprised lipids belonging to the category of glycerophospholipids [lysophosphatidylethanolamines (LPEs, n = 13; LPE 0:0/18:1, LPE 0:0/18:2, LPE 0:0/20:3, LPE 0:0/22:4, LPE 18:1/0:0, LPE 18:2/0:0, LPE 20:1/0:0, LPE 20:3/0:0, LPE 20:4, LPE 20:5, LPE 22:4/0:0, LPE 22:5, LPE 22:6), lysophosphatidylserine (LPS, n = 1; LPS 0:0/18:1), phosphocholine (PC, n = 3; PC 18:1–20:1, PCO-16:0–20:1, PCO-36:2) and phosphoethanolamine (PE, n = 1; PE 16:0–18:1)] (Fig. 2A). Interestingly, LPE concentrations were significantly downregulated in patients with statin treatment in contrast to statin-naïve patients (Supplementary Fig. S7).

A different platelet lipidome signature was found when patients were categorized into bleeding versus no bleeding within the follow-up period. Nine platelet lipids were significantly (p < 0.05, FDR < 5%) increased in patients with a major bleeding event during the follow-up when compared to patients free from bleeding (Fig. 2B). Among upregulated lipids, different lipid groups comprised mainly fatty acyls [(acyl carnitines (CAR, n = 7, CAR 10:0, CAR 12:0, CAR 14:0, CAR 14:1, CAR 16:0, CAR 16:1, CAR 18:2), fatty acids (FA, n = 2, FA 18:1, FA 18:2)] and one phosphocholine (PC 18:0–22:5) (Fig. 2B). Here, it was striking that upregulated platelet acylcarnitines in patients with bleeding events were associated with an increased bleeding risk according to PRECISE-DAPT score. The established score of predictive bleeding risk was significantly higher in patients with adverse events. We found that the number of patients with high CAR concentration was significantly increased in patients with high (61.9%) when compared to moderate (48.5%), low (46%) or very low (39.1%) PRECISE-DAPT score (Supplementary Fig. S8).

Metabolic pathway analysis of altered lipids revealed enrichment of fatty acyl and phospholipids metabolism in patients with adverse CV events when compared to patients without events during the follow-up (Supplementary Fig. S5).

To evaluate, whether critical changes in the total lipid composition of the platelet lipidome occur between patients with recurring adverse events and patients without CVD events, we performed principal component analysis (PCA) including all 767 detected lipids of this study. Clustering of patients enrolled into the 3-year follow-up showed a distinct separation of patients with adverse cardiovascular and bleeding events when compared to patients free from recurring adverse events (Fig. 3). Supervised orthogonal partial least square discriminant analysis (OPLS-DA) revealed a clear separation of patients with ischaemic events in contrast to bleeding events (Supplementary Fig. S1). Further, a distinct separation of patients with ischaemic or bleeding events in contrast to patients with an event-free follow-up is depicted in Supplementary Figs. S2 and S3. This indicates that the platelet lipidome signature in patients with adverse ischaemic and bleeding events significantly differs from patients without cardiovascular events. Remarkably, concentrations of poly-unsaturated lysoglycerophospholipids (PUFA-LPLs) were increased in patients with recurrent adverse events, whereas glycerolipids and especially triglycerides were decreased by trend in contrast to patients without CVD events (Fig. 4A). Likewise, besides mono- and poly-unsaturated fatty acyls, PUFA-PLs were increased in platelets of patients with major bleeding events. Interestingly, physiochemical attributes of distinct lipid classes significantly differed among patients with recurring cardiovascular events when compared to patients without any incident CVD during the follow-up (Fig. 4B). Side chain length (R = 0.760, p < 0.001) and degree of saturation (R = 0.873, p < 0.0001) of LPEs were significantly associated with a higher abundance in patients with recurrent cardiovascular adverse events when compared to patients without incidence of CVD events. Similar trends were observed for TG side chain length (R = 0.320, p = 0.002) and number of double bonds (R = 0.300, p = 0.003). In contrast, an inverse correlation of mean ratios between patients with recurrent cardiovascular events compared to the event-free group was shown for PC side chain length (R = − 0.391, p < 0.0001) and number of double bonds (R = − 0.392, p < 0.001) (Fig. 4C). Likewise, side chain length (R = 0.778, p < 0.0001) and degree of saturation (R = 0.807, p < 0.0001) of LPEs were associated with major bleeding events during the follow-up period. Similar results were achieved for TG side chain length (R = 0.312, p = 0.002) and double bonds (R = 0.292, p = 0.004). In contrast, FA side chain length (R = − 0.622, p = 0.041) and number of double bonds (R = − 0.635, p = 0.036) were inversely associated with major bleeding events (Fig. 4D).

To further elucidate the impact of changes in the platelet lipidome in patients with adverse cardiovascular events, we employed hazard models for assessing the ability of significantly altered lipids to indicate an enhanced risk for adverse cardiovascular events. Therefore, Cox proportional hazard models were calculated with 14 lipids found to be highly increased in patients with adverse cardiovascular events after adjustment for age, gender and statin treatment (Fig. 5). Upregulated lipids of patients with recurrent CVD events shared a hazard ration > 1, indicating an increased cardiovascular risk. Furthermore, lysophospholipids (LPE/LPS) were significantly associated with an increase in CV risk. Thus, LPE 20:1/0:0 (HR 1.18, 95% CI 1.07–1.29), LPE 0:0/18:1 (HR 1.16, 95% CI 1.04–1.28), LPE 0:0/20:3 (HR 1.14, 95% CI 1.06–1.23), LPE 20:3/0:0 (HR 1.14, 95% CI 1.06–1.23), LPE 18:1/0:0 (HR 1.14, 95% CI 1.06–1.23), LPE 0:0/18:1 (HR 1.14, 95% CI 1.06–1.23), LPE 20:4 (HR 1.14, 95% CI 1.05–1.23), LPE 18:2/0:0 (HR 1.14, 95% CI 1.05–1.23), LPE 0:0/22:4 (HR 1.13, 95% CI 1.05–1.22), LPE 0:0/18:2 (HR 1.13, 95% CI 1.05–1.22), LPE 22:4/0:0 (HR 1.13, 95% CI 1.05–1.22), LPE 22:6 (HR 1.13, 95% CI 1.05–1.22), LPE 20:5 (HR 1.13, 95% CI 1.05–1.22) and LPE 22:5 (HR 1.13, 95% CI 1.05–1.22) independently predicted an enhanced risk for adverse cardiovascular events in this study (Fig. 5A). Correspondingly, upregulated lipids of patients with major bleeding events indicated an increased risk of incident major haemorrhage. FA 18:1 (HR 1.16, 95% CI 1.00–1.40) and FA 18:2 (HR 1.16, 95% CI 1.00–1.40), both independently predicted an enhanced risk for major bleeding events during the follow-up period (Fig. 5B).

In conclusion, it becomes evident that distinct lipid clusters including LPE and fatty acids might be suitable to predict the cardiovascular risk in patients with CAD.

In order to identify new biomarkers of the platelet lipidome in patients with CAD, being eligible to sufficiently determine the risk of adverse cardiovascular events, we applied a penalized model including all lipids. Thus, overfitting of lipids is minimized and thus variable selection and regularization is optimized leading to an enhanced prediction accuracy and interpretability in the setting of omics data and generation of new potential biomarkers.

Therefore, cross-validated LASSO selection algorithm was performed and adjusted for age, gender and statin therapy. In the L₁-regularized Cox regression analysis nine lipids shared nonzero coefficients and thus were related to incident CVD (Table 3A). Selected lipids mostly comprised characteristic glycerophospholipids (PC 34:2;O, LPE 20:3/0:0, PI 18-1:20:4, PE O 16-1-18:2, PI 16:0–20:4), as well as two fatty acyls (FA 18:2;2O, FA18-1) and triglycerides (TG 14:0–16:0–18:1, TG 16:0–14:1–18:1). It was further noticeable, that these lipids shared at least one of the following characteristics: (i) FA 18:1 or FA 18:2 constituent; (ii) lipid oxidation or alkenyl substituent; (iii) FA 20:3 or 20:4 fatty acid side chain.

In this study, we performed a cross-validated approach of increasing the predictive value to discriminate the risk of adverse events in patients with CAD. Thus, we calculated concordance indices of conventional cardiovascular risk factors belonging to the Framingham risk score to discriminate the 3-year CV risk and added stepwise models including selected lipids of LASSO analysis (Table 3B). The addition of three lipids with highest ß-coefficients PC 34:2; O, LPE 20:3/0:0 and PI 16:0–20:4 to the baseline model (C index 0.647, 95% CI 0.612–0.741) including conventional risk measures (including age, gender, diabetes mellitus, smoking status, arterial hypertension, total cholesterol, and HDL cholesterol) increased the C index by 0.006 (C index 0.650, 95% CI 0.623–0.754). However, an even stronger elevation of C index by 0.095 was achieved by adding six lipids including FA 18:2; 2O, TG 16:0–14:1–18:1 and PE O 16-1-18:2 (C index 0.739, 95% CI 0.703–0.816). Ultimately, the total cluster of nine lipids with addition of PI 18:1:20:4, FA18:1 and TG 14:0–16:0–18:1 only showed a marginal augmentation of C index by 0.001 (C index 0.740, 95% CI 0.710–0.820). To further elucidate the weight of adding new lipid biomarkers to established lipids risk measures, we replaced total cholesterol (TC) and high-density lipoprotein (HDL) with platelet lipid models. Here we found that replacing three lipids did not increase predictive value (C index 0.634, 95% CI 0.606–0.731), but six lipids were outperforming established measures by amplifying C index by 0.094 (C index 0.738, 95% CI 0.698–0.810). In addition, replacement of HDL and TC by all selected lipids caused the highest enhancement of C index by 0.097 (C index 0.741, 95% CI 0.695–0.812). Thus, addition of platelet lipid signatures to established cardiovascular risk factors might significantly augment the 3-year risk discrimination in patients with CAD.

Platelet hyperreactivity is associated with CAD and predicts outcome in CVD patients [3, 9, 18, 19]. Only recently, we found that changes in the platelet lipidome have a significant impact on platelet function [15]. To elucidate whether described changes of the platelet lipidome in patients with adverse cardiovascular events are associated with modulation of platelet function, we performed ex vivo function assays at the time of blood sampling. Patients were divided into quartiles according to the relative concentration levels of distinct lipids found to be significantly upregulated in patients with recurring adverse events when compared to those “event-free” patients. Collagen-induced platelet aggregation was significantly increased in patients with high levels of LPE 0:0/18:1 (p = 0.025), LPE 0:0/18:2 (p = 0.005), LPE 18:1/0:0 (p = 0.011), LPE 18:2/0:0 (p < 0.0001), LPE 0:0/20:3 (p = 0.023) and LPE 22:4/0:0 (p = 0.027) (Fig. 6A). In contrast, high concentrations of CAR 5:0 were associated with lowered arachidonic acid- (p = 0.007) and ADP-mediated (p = 0.006) platelet aggregation. Likewise, collagen-induced platelet aggregation was decreased in patients with high concentrations of CAR 8:0 (p = 0.001) and CAR 16:0 (p = 0.040) (Fig. 6A).

Further correlation analysis revealed that CAR 5:0 (r = − 0.123, p = 0.021), CAR 8:0 (r = − 0.175, p = 0.001) and CAR 16:0 (r = − 0.108, p = 0.043) concentrations were inversely associated with collagen-induced platelet aggregation (Fig. 6B). In contrast, collagen-mediated platelet aggregation significantly correlated with increasing concentrations of LPE 0:0/18:1 (r = 0.164, p = 0.002), LPE 0:0/18:2 (r = 0.154, p = 0.004), LPE 0:0/20:3 (r = 0.156, p = 0.003), LPE 0:0/22:4 (r = 0.176, p = 0.001), LPE 18:1/0:0 (r = 0.119, p = 0.025), LPE 18:2/0:0 (r = 0.173, p = 0.001), LPE 20:3/0:0 (r = 0.148, p = 0.006), LPE 20:4 (r = 0.172, p = 0.001), LPE 20:5 (r = 0.127, p = 0.017), LPE 22:4/0:0 (r = 0.166, p = 0.001), LPE 22:5 (r = 0.157, p = 0.003) and LPE 22:6 (r = 0.145, p = 0.007) (Fig. 6B). Thus, high levels of lysophospholipids seem to promote platelet activation, whereas acylcarnitines lower platelet hyperreactivity in patients with CVD (Fig. 7A, B).