The Plasma Oxylipin Signature Provides a Deep Phenotyping of Metabolic Syndrome Complementary to the Clinical Criteria

Metabolic syndrome (MetS) is a complex condition encompassing a constellation of cardiometabolic abnormalities. Oxylipins are a superfamily of lipid mediators regulating many cardiometabolic functions. Plasma oxylipin signature could provide a new clinical tool to enhance the phenotyping of MetS pathophysiology. A high-throughput validated mass spectrometry method, allowing for the quantitative profiling of over 130 oxylipins, was applied to identify and validate the oxylipin signature of MetS in two independent nested case/control studies involving 476 participants. We identified an oxylipin signature of MetS (coined OxyScore), including 23 oxylipins and having high performances in classification and replicability (cross-validated AUCROC of 89%, 95% CI: 85–93% and 78%, 95% CI: 72–85% in the Discovery and Replication studies, respectively). Correlation analysis and comparison with a classification model incorporating the MetS criteria showed that the oxylipin signature brings consistent and complementary information to the clinical criteria. Being linked with the regulation of various biological processes, the candidate oxylipins provide an integrative phenotyping of MetS regarding the activation and/or negative feedback regulation of crucial molecular pathways. This may help identify patients at higher risk of cardiometabolic diseases. The oxylipin signature of patients with metabolic syndrome enhances MetS phenotyping and may ultimately help to better stratify the risk of cardiometabolic diseases.


Introduction
Metabolic syndrome (MetS) is a serious global public health concern. It is reaching epidemic proportions, with a global prevalence estimated to be about one quarter of the world population [1]. It is a progressive and heterogeneous condition encompassing a constellation of cardiometabolic abnormalities, including central obesity, elevated blood pressure, hypertriglyceridemia, low high-density lipoprotein cholesterol (HDLc) and dysglycemia [2], which, when associated, culminate in a five-fold and three-fold increased risk of type 2 diabetes mellitus and cardiovascular diseases, respectively [3]. Mechanisms underlying MetS are complex and incompletely understood but there is epidemiological evidence showing strong associations with inflammation, oxidative stress, alteration in insulin sensitivity and thrombosis as well as endothelial, renal and hepatic dysfunctions [4].
From a clinical perspective, it is crucial to provide a more complete view of the molecular pathways involved in the onset and development of MetS to better understand, stratify and ultimately prevent the risk of cardiometabolic diseases [5]. In this field, metabolomics and lipidomics are powerful tools that can capture the complexity of MetS and inform on underlying mechanisms at the metabolite level [6]. Untargeted data-driven metabolomics has proven its utility to comprehensively characterize the metabolic changes observed in MetS [7,8]. However, the coverage of untargeted assays is often limited to the most abundant metabolites (such as building blocks of cell membrane and "fuel" metabolites involved in energy production and storage) and does not inform on signaling metabolites that are present at lower concentrations [9]. Moreover, the partial mapping of the metabolome and annotation uncertainties limit biological interpretation [7]. To circumvent this limitation while keeping a systemic approach, we targeted a specific class of lipid mediators called oxylipins (including eicosanoids), arising from the oxygenation of polyunsaturated fatty acids (PUFAs) through the coordinated action of over 50 unique and cell-specific enzymes [10]. Oxylipins are produced in abundance during inflammation and oxidative stress, two underlying processes in the pathogenesis of MetS [11]. Moreover, oxylipins are involved in the regulation of a vast array of biological processes related to cardiometabolic health, including blood clotting, endothelial permeability, blood pressure, vascular tone, adipogenesis, along with glucose homeostasis and insulin signaling [10,12,13]. Based on this, we hypothesized that comprehensive oxylipin profiling could identify relevant biomarker patterns in MetS to be used in clinics to enhance the phenotyping of MetS pathophysiology. A high-throughput validated mass spectrometry method allowing for the quantitative profiling of over 130 oxylipins [14,15] was applied to identify the oxylipin signature of MetS in a case-control study nested in the Polish branch of the Prospective Urban and Rural Epidemiological (PURE) cohort. We then replicated the study in an independent population, namely the French NutriNet-Santé cohort, allowing for external validation of the newly discovered oxylipin signature.

Baseline Characteristics of Participants
As shown in Table 1, differences between Cases (i.e. participants with ≥3 criteria of MetS, including obesity, high blood pressure, hypertriglyceridemia, low HDL-c and hyperglycemia as described in [3]) and Controls (participants with <3 criteria of MetS) were comparable for both the Discovery and the Replication studies, except for the level of education, the localization and the level of low-density lipoprotein cholesterol (LDLc) that were significantly different only between Case and Control participants in the Discovery study. Moreover, when looking at deviation from the mean of the cardiometabolic criteria (Figure 1), differences between Case and Control participants were slightly more pronounced in the Discovery study than in the Replication study. The contrast between Figure 1. Baseline cardiometabolic differences between the Control and MetS participa Discovery and Replication studies. Volcano plots showing baseline differences of MetS between Case and Control participants in the Discovery study and (b) between Case an participants in the Replication study. Differences between Case and Control are expressed ard deviation (SD) and were assessed using Wilcoxon signed-rank test followed by mul correction (BH). Dashed line represents the significant threshold (p-value BH = 0.0005). HD density lipoprotein cholesterol, TG: triglycerides, fast. Glc: fasting glucose, waist circ: wai ference, DBP and SBP: diastolic and systolic blood pressure. between Case and Control participants in the Replication study. Differences between Case and Control are expressed in standard deviation (SD) and were assessed using Wilcoxon signed-rank test followed by multiple tests correction (BH). Dashed line represents the significant threshold (p-value BH = 0.0005). HDLc: high-density lipoprotein cholesterol, TG: triglycerides, fast. Glc: fasting glucose, waist circ: waist circumference, DBP and SBP: diastolic and systolic blood pressure.

The OxyScore Includes 23 Candidate Oxylipins and Has High Performances of Classification and Replicability
Using the 29 candidate oxylipins selected in both the Discovery and the Replication studies and the participants in the Discovery study, we constructed a Least Absolute Shrinkage and Selection Operator (LASSO)-penalized conditional logistic regression model to reduce the oxylipin list to a minimal necessary set from a discriminative point of view. It selected 23 of the 29 candidate oxylipins and allowed for the definition of the OxyScore that estimates the probability of having MetS. The 23 oxylipins selected for the definition of the OxyScore are: 8-HEPE, 9(10)-Ep-stearic acid, 16-HETE, 12(13)-EpODE, 12-HETrE, 7-HDHA, 9,10-DiHOME, 5-HETE, 9-HODE, 14,15-DiHETrE, 9,10-DiHODE, 5-HETrE, 15-HODE, 13-oxo-ODE, 11(12)-EpETrE, 9-oxo-ODE, 4-HDHA, 13-HODE, 12-HODE, 7,8-DiHDPE, 9(10)-EpOME, 15-HETE and 5-HEPE. The predictive performance (i.e., specificity and sensibility) of the OxyScore was assessed using cross-validated Area Under the Receiver Operating Characteristic Curves (AUC ROC ). It reached a value of 89% (95% confidence interval (CI) 85-93%) in the Discovery study ( Figure 5a) and a value of 78% (95% CI: 72-85%) in the Replication study ( Figure 5b). In the Discovery study (i.e., training set), the OxyScore correctly classified 83% of the participants, while 72% of the participants were correctly classified in the Replication study. The Case and Control participants in the Replication study were slightly less contrasted than those in the Discovery study (see Volcano plot in Figure 1); this may have contributed to the lower precision of classification rate in the Replication study. A figure summarizing variable selection and model building/validation is available in the Supplemental Materials ( Figure S2).  For 47% of these studies, experiments were realized using human in vitro models whereas 5.4% used in vivo human experiments and 27% were realized using in vivo animal experiments. The package R "stringr" was used to establish the links between the oxylipins and the cardiometabolic functions, then the package "googleVis" was used to generate the Sankey plot. For 47% of these studies, experiments were realized using human in vitro models whereas 5.4% used in vivo human experiments and 27% were realized using in vivo animal experiments. The package R "stringr" was used to establish the links between the oxylipins and the cardiometabolic functions, then the package "googleVis" was used to generate the Sankey plot.

The OxyScore is Consistent and Complementary with the MetS-z-score and the Clinical Criteria of MetS
The OxyScore (i.e., probability of having MetS according to the selected 23 candidate oxylipins) was computed for all participants in the Discovery study and correlated with the MetS-z-score (i.e., probability of having MetS according to the five criteria of MetS: waist circumference, triglycerides (TG), HDLc, blood pressure and fasting glucose) (Figure 6). The strength of the correlation informs on the consistency and the complementarity The OxyScore (i.e., probability of having MetS according to the selected 23 candidate oxylipins) was computed for all participants in the Discovery study and correlated with the MetS-z-score (i.e., probability of having MetS according to the five criteria of MetS: waist circumference, triglycerides (TG), HDLc, blood pressure and fasting glucose) ( Figure 6). The strength of the correlation informs on the consistency and the complementarity of the information brought by the OxyScore in comparison with the MetS criteria. The correlation between the OxyScore and the MetS-z-score was strong (r = 0.7), suggesting that a large part of the information provided by the OxyScore is consistent with the information provided by the MetS-z-score. However, the OxyScore also brings some information that is not captured by the MetS-z-score; if not, we would have expected a coefficient closer to 1, despite possible technical variability. of the information brought by the OxyScore in comparison with the MetS criteria. The correlation between the OxyScore and the MetS-z-score was strong (r = 0.7), suggesting that a large part of the information provided by the OxyScore is consistent with the information provided by the MetS-z-score. However, the OxyScore also brings some information that is not captured by the MetS-z-score; if not, we would have expected a coefficient closer to 1, despite possible technical variability. Figure 6. Relationships between the OxyScore and the Met-z-score. The OxyScore (i.e., probability of having MetS according to the identified and validated oxylipin signature, see Table S2) was computed for all participants in the Discovery cross-sectional study. Spearman correlation was established between the computed OxyScore and the Met-z-score. The Spearman correlation coefficient (r) was highly significant (p < 0.001). The red line represents the linear orientation of the relation.
To go further in the assessment of the complementarity of the information provided by the oxylipin signature, we built a new LASSO model, including the five MetS criteria additionally to the candidate oxylipins for the selection step ( Table 2). All MetS criteria were selected in the new LASSO model consistently with the use of the MetS criteria as criteria of participant selection (see study design). More importantly, the new LASSO model also included most of the candidate oxylipins as only 6 candidate oxylipins (i.e., 9(10)-epoxy-stearic acid, 12-HETrE, 7-HDHA, 9,10-DiHOME, 9,10-DiHODE and 9(10)-EpOME) out of the 23 provided were not selected in the new LASSO model. This supports that most candidate oxylipins provide a characterization of MetS participants complementary to the MetS criteria.
In order to assess if the oxylipin signature was preferentially linked to one criterion of MetS, further correlation analysis and adjusted models were performed (see Supplemental tables and figures). The correlation analysis shows noticeable correlations between the OxyScore and waist circumference (r = 0.57), fasting blood glucose (r = 0.48), TG (r = 0.60) and HDLc (r = 0.51), but not with systolic and diastolic blood pressure (r = 0.27) ( Figure S3). This may suggest that among the 23 candidate oxylipins used to predict the risk of having MetS, some of them provide information shared with waist circumference, Figure 6. Relationships between the OxyScore and the Met-z-score. The OxyScore (i.e., probability of having MetS according to the identified and validated oxylipin signature, see Table S2) was computed for all participants in the Discovery cross-sectional study. Spearman correlation was established between the computed OxyScore and the Met-z-score. The Spearman correlation coefficient (r) was highly significant (p < 0.001). The red line represents the linear orientation of the relation.
To go further in the assessment of the complementarity of the information provided by the oxylipin signature, we built a new LASSO model, including the five MetS criteria additionally to the candidate oxylipins for the selection step ( Table 2). All MetS criteria were selected in the new LASSO model consistently with the use of the MetS criteria as criteria of participant selection (see study design). More importantly, the new LASSO model also included most of the candidate oxylipins as only 6 candidate oxylipins (i.e., 9(10)-epoxystearic acid, 12-HETrE, 7-HDHA, 9,10-DiHOME, 9,10-DiHODE and 9(10)-EpOME) out of the 23 provided were not selected in the new LASSO model. This supports that most candidate oxylipins provide a characterization of MetS participants complementary to the MetS criteria. Table 2. Odd ratios (OR) associated with the variable selected in the LASSO models constructed from the candidate oxylipins and/or the MetS criteria. Two new LASSO models were constructed with the participants in the Discovery study using either only the 5 criteria of MetS (i.e., waist circumference, blood pressure, fasting glucose, triglycerides and HDLc, LASSO model N • 2, "-" for not applicable for this model) or the 5 criteria of MetS and the 23 candidate oxylipins (column 3, LASSO model N • 3). Of note, the 5 MetS criteria were the criteria of selection of MetS participants. The predictive performance of these two LASSO models reached a cross-validated AUC of 93% (95% CI: 90-96%) and 95% (95% CI: 90-96%), respectively. Six candidate oxylipins (i.e., 9(10)-Ep-stearic acid, 12-HETrE, 7-HDHA, 9,10-DiHOME, 9,10-DiHODE and 9(10)-EpOME) out of the 23 provided were not selected in the LASSO model N • 3. In order to assess if the oxylipin signature was preferentially linked to one criterion of MetS, further correlation analysis and adjusted models were performed (see Supplemental tables and figures). The correlation analysis shows noticeable correlations between the OxyScore and waist circumference (r = 0.57), fasting blood glucose (r = 0.48), TG (r = 0.60) and HDLc (r = 0.51), but not with systolic and diastolic blood pressure (r = 0.27) ( Figure S3). This may suggest that among the 23 candidate oxylipins used to predict the risk of having MetS, some of them provide information shared with waist circumference, fasting blood glucose, TG or HDLc. Correlations between each candidate oxylipin and each MetS criterion ( Figure S4) were mostly below 0.50, the strongest correlations (>0.30) being observed with the MetS-z-score and with TG: 12-HODE (−0.34 and −0.33 with MetS-z-score and TG, respectively), 12,13-EpODE (0.41 with TG only) and 9(10)-epoxy-stearic acid (0.32 and 0.48 with MetS-z-score and TG, respectively). Finally, additional fitting of LASSO model was performed adjusting with each MetS criterion (one at a time), providing insights about information complementarity. Some oxylipins were no longer selected in some of the models (e.g., 12-HETrE, 7-HDHA and 9-HODE in the waist-adjusted model, see Table S3), suggesting that they were bringing similar information than the included MetS criterion (e.g., waist circumference). On the contrary, in the LASSO model adjusted with systolic or diastolic blood pressure, no oxylipin was excluded, suggesting that the differences in blood pressure between Cases and Controls were not captured by the candidate oxylipins.
In order to capture the full information brought by circulating oxylipins, we analyzed total oxylipins (i.e., free and esterified). Integrating esterified oxylipins is important when investigating cardiometabolic health as they are found in lipoproteins (especially epoxyand hydroxy-PUFA) [21,22] that are involved in many cardiometabolic processes, such as inflammation, oxidative stress or endothelial activation. Moreover, esterified oxylipins represent the major pool of circulating oxylipins. These may also arise from the secretion of blood cells (i.e., immune cells and platelets) or endothelium and they circulate unbound (free) or bound to plasma proteins, such as albumin [23,24]. Other potential vectors of circulating oxylipins in plasma include extracellular vesicles that are both carriers and producers of oxylipins [25,26].
The biological interpretation of circulating oxylipins requires consideration of their origin and biological functions. In the Discovery study, the oxylipin signature highlighted lower levels of linoleic acid (LA)-derived oxylipins (i.e., HODEs including 9-and 13-HODE and their oxo-derivatives as well as 10-and 12-HODE) in the MetS group. These 18 carbon oxylipins can arise from the LOX pathways and/or from free-radical-mediated oxidation [27] or from singlet oxygen pathways for 10-and 12-HODE [20]. In terms of biological activities, 9-and 13-HODE have been described as potent regulators of monocytes/macrophages and neutrophils, in which they stimulate lipid uptake [28,29] and induce apoptosis [30]. Concerning the inflammatory response, 13-HODE inhibits the production of the chemoattractant LTB 4 by isolated human neutrophils [31,32], while 9-HODE induces the expression of the pro-inflammatory mediators TNFα and MIP-2α in RAW264.7 macrophages [33]. Both HODEs were also showed to prevent the activation of endothelial cells [34][35][36][37]; to decrease the secretion of the fibrinolytic inhibitor plasminogen activator inhibitor-1 (PAI-1) [38] and to inhibit platelet aggregation [32,39,40]. Neither the biological activities nor the formation route of 10-and 12-HODE have been precisely described in humans but high levels of 10-and 12-HODE have been associated with insulin resistance [20]. Our results are consistent with a previous clinical trial showing lower levels of 9-and 13-HODE in very-low-density lipoproteins (VLDL) and in LDL from MetS subjects (n = 17) in comparison with healthy controls (n = 14) [41]. Another small clinical trial reported higher levels of 9-and 13-HODE in LDL of MetS participants in comparison with healthy controls [42]. However, the technical approach to quantify HODEs (i.e., thin-layer chromatography vs. liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS)) and the absence of antioxidants during sample preparation may have in-duced artificial production of HODEs [43]. Altogether, the low levels of HODEs in the oxylipin signature of the MetS participants in the Discovery study may reflect a decreased LOX activity and/or a well-controlled systemic oxidative stress that could be linked to the activation of the antioxidant systems. Further exploration is needed to confirm this hypothesis. Considering the important regulatory functions of 9-and 13-HODEs, their low levels in MetS participants may indicate that inflammation, endothelial and platelet functions were under control.
The oxylipin signature of MetS was also characterized by high levels of epoxy-PUFAs and low levels of vicinal dihydroxy-PUFAs (i.e., metabolites of epoxy-PUFAs produced by the sEH), reflecting an activation of the CYP pathway and a reduced activity of sEH. Notably, in the Discovery study, MetS participants exhibited higher levels of 12(13)-EpODE, 14(15)-EpEDE and 9(10)-epoxy-stearic acid, while 9,10-DiHOME and 7,8-DiHDPE were significantly lower. In general, epoxy-PUFAs are reported to be protective in regard to cardiometabolic disease, while activation of sEH is usually associated with metabolic stress [44]. CYP enzymes are highly expressed in the liver but they are also found in the kidney, heart, skeletal muscle, pancreas, adipose tissue, endothelium, leucocytes and platelets [45,46]. So far, the biological activities of epoxy-PUFAs have been mainly reported for those derived from AA, such as 11(12)-EpETrE. This oxylipin has important roles in the regulation of cardiometabolic health as it has been shown to improve glucose homeostasis [47][48][49][50][51][52][53][54] and endothelial function [55][56][57][58][59], to induce vasodilatation [59][60][61][62][63][64][65][66], to modulate platelet aggregation [67][68][69] and adipogenesis [50,65,70] and to inhibit leucocyte adhesion to the vascular wall [71]. The effects of 11(12)-EpETE derived from EPA and 9(10)EPOME derived from LA are also partly documented: the former was shown to inhibit platelet aggregation [69] and induce vasodilatation [72], while the latter enhanced insulin signaling in HepG2 cells [73]. To the best of our knowledge, the biological activities of 12(13)-EpODE and 14(15)-EpEDE have not been investigated so far. High levels of 9(10)-epoxy-stearic acid may induce lipid accumulation and oxidative stress in HepG2 cells [74,75]. Among the vicinal dihydroxy-PUFAs that were substantially and significantly reduced in MetS participants, 9,10-DiHOME was shown to increase coronary resistance [76] and to attenuate insulin signaling [73] but no information is available for 7,8-DiHDPE. Vicinal dihydroxy-PUFAs derived from AA (i.e., 5,6-DiHETrE and 14,15-DiHETrE) have been shown to facilitate chemoattraction of monocytes [77], to induce vasorelaxation in canine microcirculation and mouse arteries [60,61], but less potently than their corresponding epoxide, and to attenuate insulin signaling [73]. Changes in the CYP:sEH axis in favor of epoxy-PUFAs in MetS participants might seem counterintuitive, but it was already reported in a clinical trial comparing healthy volunteers to patients with coronary artery disease (CAD), associated or not with obesity [78]. The higher Epoxy-PUFAs:Diols ratios in CAD patients was suggested to be a compensatory response to the presence of advanced cardiovascular disease. Similarly, it can be hypothesized that MetS participants have activated CYP and decreased sEH activity to compensate for their cardiometabolic disturbances. In a clinical trial comparing Controls vs. MetS participants, this compensatory mechanism was not observed [79]. However, it is worth mentioning that age was a major confounding factor in this study and the significantly lower mean age of the Control participants (32 vs. 51 y) could have contributed to the higher epoxide levels independently of the absence of MetS [80].
Although it might seem counterintuitive, the oxylipin signature of the MetS participants in the Discovery study may reflect the implementation of compensatory mechanisms, including (i) the control of oxidative stress, (ii) the modulation of the CYP:sEH axis in favor of the protective epoxy-PUFAs and (iii) the production of regulatory oxylipins, such as the 15-LOX products of long-chain PUFAs to negatively regulate the 5-LOX pathway. These mechanisms could be naturally implemented or induced by medications that are used in the management of MetS. It should also be noted that the oxylipin signature could be partially linked to other environmental factors, such as diet, which is another factor known to influence MetS development.
The oxylipin signature of MetS identified in the Discovery study was highly replicable in the Replication study. To the best of our knowledge, no metabolomic or lipidomic signature of the MetS has been externally validated so far [6]. Moreover, the differences in oxylipin concentrations between Cases and Controls were mostly similar in both studies. Basically, in the Replication study, we also observed modulations in the CYP:sEH axis, indicating a higher Epoxide:Diol ratio in MetS participants as a potential compensatory mechanism for the cardiometabolic disturbances. However, in the Replication study, the levels of 9-and 13-HODE were not lower in the MetS participants but rather tended to be higher. Another difference concerns the mid-chain alcohols derived from DGLA, AA and EPA (i.e., 5-HETrE, 5-HETE and 5-HEPE) that became significantly higher, suggesting an active perturbation of cardiometabolic health associated with inflammation, vasoconstriction, endothelial dysfunction and platelet aggregation. Finally, the pathway involving the CYP ω-hydroxylase remains perturbed in the Replication study but in the opposite direction and more profoundly, with 15-HODE levels being substantially lower while 16-HETE became significantly higher. The biological function of 15-HODE is unknown but 16-HETE has been described as an anti-inflammatory mediator via the inhibition of human PMN adhesion and aggregation and of LTB 4 synthesis [95]. The high levels 9-and 13-HODE associated with the higher production of 15-and 16-HETE could reflect the implementation of compensatory mechanisms. However, contrary to what was observed in the Discovery study, it does not seem sufficient to counterbalance the activation of the detrimental 5-LOX pathway. Of note, this pathway is known to be activated with aging [96] and MetS participants in the Replication study were significantly older than those in the Discovery study. This could have contributed to this impaired control of the 5-LOX pathway and may indicate an increased risk of cardiometabolic diseases.
We acknowledge limitations in our study. A first potential limitation concerns the sample size of the studies that does not allow for stratifying the analysis, for example, to look at the results among younger and older participants. We also lacked information about the use of medication (e.g., NSAIDs) that could be interesting to integrate in future analysis. Another technical limitation concerns the lack of chiral analysis hampering to precisely determine the exact origin (i.e., enzymatic or not) of some oxylipins. Finally, it should be noted that even though having two independent cohorts was a strength, this highlighted differences between the two studies regarding the absolute concentration of several oxylipins (Table S1). However, the selection of the candidate oxylipins was independent and showed that differences between Cases and Controls within each study were consistent. This supports that the differences of absolute concentrations observed between the studies did not affect the selection of the candidate oxylipins within each study.

Discovery and Replication Cohorts
The discovery cohort was the Polish branch of the Prospective Urban and Rural Epidemiological cohort (n = 2036, aged 30-70 y), launched in 2009 [97]. For the Polish branch of the PURE cohort, the recruitment of participants took place between 2007 and 2010. During the visit, socio-demographic (i.e., education, localization) and lifestyle questionnaires (including smoking habits, physical activity), medical tests (i.e., electrocardiogram, blood pressure), anthropometric measurements (i.e., weight, height, waist and hips circumference) and biochemical tests (fasting blood glucose, total cholesterol, HDLc, LDLc and TG) were realized. Moreover, at baseline, a 134-item food frequency questionnaire (FFQ) was completed by each participant. The study was approved by the Institutional Review Board of the Wroclaw Medical University (IRB number: KB-443/2006). Independent external replication was conducted with participants from the webbased prospective French Nutrinet-Santé cohort [98] launched in 2014 and still ongoing (www.etude-nutrinet-sante.fr, accessed on 1 August 2022). The cohort included 160,000 participants (aged ≥ 18 y) but for the independent external validation the selection was made only from the participants having blood specimens (n = 19,772, aged 30-70 y). Participants of the French Nutrinet-Santé cohort are asked to complete every year questionnaires to collect information regarding socio-demography (education and localization), lifestyle (including smoking habits, physical activity), health (disease history, menopausal status, anthropometric self-assessment) and dietary habits (3 × 24 h dietary records including > 3300 food items). Blood samples were collected at baseline and were used for the biochemical tests (fasting blood glucose, total cholesterol, HDLc, LDLc and TG). These selected participants also had a clinical examination, including the assessment of blood pressure and anthropometric measurements (weight, height, waist and hip perimeters, fat mass, fat mass on body trunk and visceral fat). The study was approved by the International Research Board of the French Institute of Health and Medical research (IRB Inserm n • 0000388FWA00005831) and the "Comité National Informatique et Liberté" (CNIL n • 908450 and n • 909216).

Study Design and Selection of Participants
From the Discovery cohort (i.e., Polish branch of the PURE cohort) and the Replication cohort (i.e., French Nutrinet-Santé cohort), cross-sectional case-control studies were performed following similar designs. The outcome of the studies presented here was the MetS as defined by Alberti et al [3]. According to this definition, participants were considered as a Case (i.e., diagnosed as having MetS) if they had at least three of the following criteria: elevated waist circumference (≥94 cm for men and 80 cm for women), elevated TG (≥150 mg/dL), reduced HDLc (<40 mg/dL for men and <50 mg/dL for women), elevated blood pressure (systolic ≥ 130 and/or diastolic ≥ 85 mmHg) and elevated fasting glucose (≥100 mg/dL). Participants receiving pharmacotherapy for elevated TG, elevated blood pressure or hyperglycemia were considered as meeting the aforementioned criteria [3]. Participants having less than three of the aforementioned criteria were considered as a Control. These studies are case-control studies aiming at comparing participants with diagnosed MetS (i.e., ≥3 criteria) vs. Control participants (<3 criteria) (Figure 7).  To select the study samples, criteria of exclusion were similar between the Discovery and the Replication studies and included: participants diagnosed with cancer before and one year after blood draw, participants with cardiovascular events before blood draw (stroke, angina and heart failure) and participants with missing data for the cardiometabolic parameters (i.e., waist circumference, blood pressure, TG, HDLc, fasting glucose). For the Discovery study, Case and Control participants were matched according to their sex, age (2 y classes), smoking status (never+former vs. current) and physical activity (low vs. moderate+intense). For the Replication study, matching factors included: sex, age (2 y classes), smoking status (never/former/irregular/current), physical activity (low/moderate/intense), menopausal status (not applicable/yes/no) and season of blood draw (winter/spring/summer/fall). To avoid the selection of incomparable Control participants between the two cohorts, the Control groups of each study were balanced according to the number of cardiometabolic criteria (i.e., 0, 1 or 2). At completion of the selection and matching processes, the Discovery study included 137 Cases and 137 Controls and the Replication cross-sectional study included 101 Cases and 101 Controls (Figure 7).

Oxylipin and Fatty Acid Quantification
For each participant, fasting blood was collected into EDTA and used to prepare plasma. Briefly, for the PURE participants, blood samples were directly centrifuged and plasma was either stored at −20 • C for up to 3 days and then stored at −80 • C or directly stored at −80 • C. For the Nutrinet-Santé participants, blood samples were stored at +4 • C for up to 24 h and then centrifuged and plasma stored at −80 • C. Of note, the conditions of transitory storage usually only affect a very limited number of oxylipins [43] and the samples used in these studies had never been thawed during their storage.

Fatty Acid Profiling
Plasma samples collected in the Discovery and Replication cohorts were prepared and analyzed in two different laboratories as described in Lillington et al. [100]. Briefly, 10 µL of an internal standard (TG19) was added to 10 µL of plasma. Following the hydrolysis with 1 mL KOH (0.5 M), the derivatization with 1 mL heptane and 1 mL BF 3 -MeOH (80 • C, 1 h) and the addition of 1 mL H 2 O, the fatty acids methyl esters (FAMEs) were extracted with 1 mL H 2 O and 2 mL of heptane, dried and dissolved in 20 µL of ethyl acetate. FAMEs were separated by gas chromatography (GC) on a Clarus 600 Perkin Elmer system using a Famewax RESTEK fused silica capillary column (30 m × 0.32 mm, 0.25 µm film thickness). Oven temperature was programmed from 110 • C to 220 • C at a rate of 2 • C/min and the carrier gas was hydrogen (0.5 bar). The injector and detector temperatures were at 225 • C and 245 • C, respectively. All of the quantitative calculations were based on the chromatographic peak area relative to the internal standard. Using these methods, the number of detected and accurately quantified fatty acids was as follows for the different studies: 21 fatty acids for the Discovery study (i.e., PURE) and 19 fatty acids oxylipins for the Replication cross-sectional study (i.e., Nutrinet-Santé). Absolute concentrations are reported in Table S1.

Metadata Statistical Analysis
Metadata associated with each participant selected in the Discovery (i.e., PURE) and the Replication (i.e., Nutrinet-Santé) studies included qualitative data (i.e., sex, smoking, education, localization, season of blood draw) and quantitative data (i.e., weight, body mass index (BMI), waist and hip circumference, systolic and diastolic blood pressure, fasting blood glucose, total cholesterol, TG, HDLc and LDLc, alternative healthy eating index (AHEI) score and plasma fatty acids) recorded or assessed at baseline. The AHEI score was calculated to estimate the quality of diet of participants and based on ten components reflecting recent dietary guidelines [101]. Based on the cardiometabolic criteria, the metabolic score (MetS-z-score) was calculated as described in https://github.com/metscalc/metscalc/ (accessed on 1 August 2022) as follows: MetS-z-score = Y + a × waist − b × HDLc + c × SBP + d × log(TG) + e × glucose, with the coefficients Y, a, b, c, d and e being determined according sex, ethnicity and age (adults vs. teenager). This derived z-score is a continuous variable providing an integrative assessment of the cardiometabolic status [16]. Differences between Cases and Controls for each parameter were assessed using univariate analysis and taking into consideration the matching of participants. The non-parametric Wilcoxon signed-rank test was used for quantitative variables whereas the contingency Fisher's Exact test was used for qualitative variables. P-values were corrected for multiple testing using the false-discovery rate correction of Benjamini-Hochberg (BH). Moreover, a special emphasis was put on the cardiometabolic parameters (i.e., waist perimeter, systolic and diastolic blood pressure, fasting glucose, TG, HDLc and MetS-z-score) through Volcano plot representations. Cardiometabolic variables were mean centered and reduced allowing for comparison as the difference in variables expressed in standard deviation between Case and Control participants. Each variable was represented in the Volcano plot with the difference in standard deviation (SD) between Case and Control participants in the x-axis and the -log(p-value) of the Wilcoxon signed-rank test (BH-corrected) in the y-axis. Statistical analysis and graphical representations were generated by the R statistical computing environment (https://www.R-project.org/, accessed on 1 August 2022) using the "En-hancedVolcano" package (https://github.com/kevinblighe/EnhancedVolcano, accessed on 1 August 2022).

Pre-processing of MS Oxylipin Data
MS data generated from the Discovery and the Validation studies were pre-processed following the same protocol as described in the Supplemental Method document. This includes detailed information regarding MS data integration, normalization and imputation as well as information regarding data adjustment. This latter pre-processing step aims to reduce the impact of total oxylipin levels on data variability and required discrete intensity adjustments as described in the Supplemental Method. For the different studies, the number of quantified oxylipins was as follows: 88 oxylipins for the Discovery study (i.e., PURE) and 58 oxylipins for the Replication study (i.e., Nutrinet-Santé). Merging the oxylipin data matrix generated in the Discovery and the Replication studies identified 54 oxylipins in common. This common matrix of 54 oxylipins was used for the selection and validation of oxylipins thereafter highlighted as 'candidates' (shared upon request by the corresponding author).

Candidate Oxylipins Selection
To leverage the case-control design of our studies and the matching of the selected participants, conditional logistic regression was used to model the association of a set of oxylipins with the odds of having a MetS diagnosis. Moreover, to overcome the highdimensional setting of our datasets and to put oxylipins unlinked to the outcome modeled aside, the Elastic-net penalization [102] method was applied for the initial oxylipin selection. This method relies on a penalization parameter that controls for the strength of selection by shrinking the coefficients toward zero (setting some to exactly zero, thus, performing variable selection). Contrary to methods as the LASSO penalization that tend to select only one representative from a pool of multi-collinear variables, Elastic-net penalization preserves all variables appearing linked to the outcome. This was considered to be important for the biological interpretation since even if oxylipins are collinear they may bring complementary information in the interpretation process. These penalization models were computed using the R software with OPT2D function from the "penalized" package [103]. As penalized regression may lead to unstable solutions due to its cross-validation-based parameter determination process, bootstrap resampling was used to enhance the robustness of oxylipin selection. Elastic-net-penalized conditional logistic regressions were repeated on 350 bootstrapped samples; the oxylipins were ordered by decreasing percentage of selection across bootstraps and the oxylipin signature of metabolic syndrome was focused on the oxylipins selected in ≥80% of bootstraps ("candidate oxylipins") for the Discovery and Replication studies. A scoring of analytical robustness was established for each candidate oxylipin selected taking into account previously published results regarding (i) the stability during transitory and long-term storage [43] and (ii) the technical and interlaboratory variabilities [14]. The scoring also takes into consideration the percentage of missing data imputation. Oxylipins with putative low score of analytical robustness were set apart from the initial selection. The process of candidate oxylipin selection was independently duplicated in the Discovery and Replication studies in order to (i) assess the consistency of the selected oxylipins and (ii) generate a complete list of oxylipins including common and population-specific oxylipins.

Model Construction and Validation
The objective was to evaluate whether a discrimination model fitted on the Discovery study could be efficient on an independent study regarding the outcome of prediction. First and foremost, in order to merge the oxylipin dataset generated in the Discovery and the Replication studies, a harmonization of the quantitative data adjustment protocol was necessary (see Supplemental Methods). Then, a LASSO-penalized conditional logistic regression model was constructed using the participants in the Discovery study and including the oxylipins previously selected in the Discovery and in the Replication studies. The LASSO penalization was chosen to remove putative redundant oxylipins from the complete signature, with model classification performances assessed for both Discovery and Replication study participants using the AUC ROC which were computed using 10-fold cross-validation, with confidence intervals and cross-validated error rates being calculated. Analysis and representation were performed by the R software using the "glmnet" [104], "pROC" [105] and "selectiveInference" [106] packages. Odds ratios of each oxylipin of the optimized signature were estimated and plotted as Circos plot [107].

OxyScore Calculation
Based on the LASSO-penalized conditional logistic regression models created in the Discovery study and validated in the Replication study, the probability of being a Case (coined OxyScore) was calculated for each participant as follows: OxyScore = exp(B * X) 1+exp(B * X) where (B × X) corresponds to a combination of b 0 × x 0 , b 1 × x 1 , . . . , b p × x p (p the number of oxylipins in the model) with b n being the adjusted coefficient of the n-th oxylipin in the LASSO regression model and x n the corresponding oxylipin concentration.

OxyScore Correlation
In order to assess the consistency and the complementarity of the OxyScore with the clinical criteria of metabolic syndrome, Spearman correlations were established between the OxyScore calculated for each participant in the Discovery study and the criteria of MetS (including the MetS-z-score). Correlations were considered as significant when p < 0.05 and r ≥ 0.5 (noteworthy link) or r ≥ 0.7 (strong link).

OxyScore Adjustment
In order to see the impact of cardiometabolic criteria in the performances of the OxyScore and in the contribution of each oxylipin to the OxyScore, the LASSO-penalized conditional logistic regression model created in the Discovery study and validated in the Replication study (used for the calculation of the OxyScore) was adjusted with each cardiometabolic criterion (i.e., waist circumference, systolic and diastolic blood pressure, fasting glucose, TG, HDLc and BMI) one by one. Performance of the adjusted OxyScore was assessed as previously described (i.e., AUC ROC computed using the 10-fold crossvalidation, confidence intervals and cross-validated error rate). Odds ratios of each oxylipin of the adjusted OxyScore were estimated. The impact of adding the criteria of MetS into the LASSO model on the maintenance or exclusion of the oxylipins of the model was evaluated.

Univariate Analysis
To determine which oxylipins of the initial signature could reflect the outcome individually, Wilcoxon signed-rank test with BH multiple-tests correction were computed for the Discovery and Replication cross-sectional studies.

Conclusions
From a clinical perspective, the oxylipin signature of MetS we identified not only has excellent performances in classification and replicability but, most importantly, it provides a unique and integrative characterization of molecular pathways of MetS. This could allow for a better understanding and stratification of the risk of cardiometabolic diseases. More precisely, the originality in the oxylipin signature relies on the information it provides on several key molecular pathways of MetS, including oxidative stress, inflammation and the regulation of vascular tone, blood clotting, endothelial permeability, glucose homeostasis and adipogenesis. This level of information is not provided by the MetS clinical criteria (i.e., WC, BP, TG, HDLc, glycaemia) and it could help to distinguish patients with similar clinical diagnosis (i.e., MetS or not based on the five MetS criteria) but with a different level of risk of cardiometabolic diseases. In the Discovery and Replication studies, the oxylipin signature suggested the implementation of compensatory mechanism of the cardiometabolic disturbances in the MetS participants that could be impaired with age. The identified oxylipin signature of MetS requires further validation in various, larger-scale and well-phenotyped populations to confirm and extend the findings of this first investigation. Another perspective should be to investigate the capacity of the plasma oxylipin signature to predict the development of MetS though longitudinal studies.