The Moli-sani cohort was randomly recruited in the Molise region (Italy) from city hall registries, as previously described (27, 28). The Moli-sani study complies with the Declaration of Helsinki and was approved by the ethics committee of the Catholic University of Rome, Italy. In total, 24, 325 subjects were enrolled between March 2005 and April 2010. All participants provided written informed consent. For the current study, TG was measured in a total of 22,866 subjects, due to a lack of sufficient plasma volume for the remaining participants (Figure 1). Additionally, subjects with incomplete baseline questionnaires were excluded, as well as subjects using anticoagulants, leaving 22,546 observations for statistical analysis. At the time of enrollment in the study, none of the subjects received direct oral anticoagulants.

Venous blood samples were previously obtained by venipuncture between 07:00 am and 09:00 am from participants who had fasted overnight and had refrained from smoking for at least 6 h (27). Citrated plasma samples for this study were stored in straws containing the sample code and barcode in liquid nitrogen in a dedicated biobank (http://www.neuromed.it/biobanking-centre/) for 8.2 years (IQR: 7.2 to 9.2 years). Platelet poor plasma was prepared by centrifugating twice at 2,821 g for 10 min. The samples were express-shipped in 3 shipments on dry ice to Synapse Research Institute, Maastricht, the Netherlands on 27-10-2016, 08-05-2017 and 23-06-2017, where they were immediately stored at−80°C. Levels of labile coagulation factors (FV and FVIII) were determined in a subset of 144 randomly chosen samples from the first shipment to confirm plasma sample quality. FV and FVIII levels were measured on the STA-R device using STA deficient FV and STA deficient FVIII reagents, according to the manufacturer's specifications (Diagnostica Stago, Asnières, France). All coagulation factor levels measured were in line with the reference ranges, established by the manufacturer.

TG was determined in platelet-poor plasma (PPP) using the Calibrated Automated Thrombinography (CAT) method (29). Due to the number of samples in the cohort, every condition was measured in a single well, for each sample. To standardize the measurement procedure as much as possible, commercial reagents from one batch were used to measure TG for all samples. Commercial calibrator, PPP Reagent Low, PPP Reagent [with and without thrombomodulin (TM)] and FluCa kits were purchased from Diagnostica Stago (Asnières, France). All reagents were stored at 4°C, according to the manufacturer's recommendation. Analyses were performed in 96-well round bottom microplates (Immulon 2 HB plate, Fisher Scientific, Roskilde, Denmark). Each well contained 20 μL PPP Reagent Low, PPP Reagent or calibrator and 80 μL plasma. The samples were preincubated for 10 min and analysis was performed at 37°C in a fluorescence plate reader (Fluoroskan Ascent, Diagnostica Stago, Asnières, France). TG curves and calibration curves were measured immediately after the automatic dispensing of 20 μL FluCa. The TG curve was monitored for 50 min, or shorter if all curves reached the baseline level earlier, to ensure that the TG curves could be calculated in a reliable way. TG curves were measured and analyzed by the Thrombinoscope software (version 5.0), as specified by the manufacturer (Diagnostica Stago, Asnières, France). TG parameters lagtime (time until the first trace of thrombin is formed), peak height (maximum amount of active thrombin present), time-to-peak (TTP, time until the maximum amount of active thrombin is present), area-under-the-curve (endogenous thrombin potential, ETP), and maximum slope of thrombin formation (velocity index, VI) were quantified automatically by the software.

Blood glucose, serum lipids and C-reactive protein (CRP) were measured within 3 h from blood sampling by enzymatic reaction methods using an automatic analyzer (ILab 350, Milano, Italy). LDL-C level was calculated according to the Friedewald formula (30). Height and weight were measured at baseline, and body mass index (BMI) was calculated as kg/m². For specific analyses, BMI was grouped into three categories as normal (<25 kg/m²), overweight (25–30 kg/m²) or obese (>30 kg/m²), according to WHO guidelines (6). A status of “current smoker” was defined as smoking 1 or more cigarettes per day on a regular basis. Waist and hip circumferences were measured according to the WHO guidelines (31) and used to calculate the waist-to-hip (WH) ratio.

The normality of continuous variables was assessed graphically in histograms and normality Q-Q plots (data not shown). Distributions of serum triglycerides, blood glucose, and serum CRP concentrations were skewed and are therefore presented as median and interquartile range (IQR). Values for continuous non-skewed variables are presented as means ± SD. Categorical variables are presented as frequencies. Comparisons of continuous variables between two groups or more than two groups were performed using the Student's t-test or Mann-Whitney U test and analysis of variance F test, respectively. Associations between continuous variables were tested using the Spearman's rank correlation coefficient.

The effect of blood lipid levels (total cholesterol, triglycerides, LDL-C, HDL-C) and BMI on TG parameters was evaluated by linear regression analyses. Crude linear regression models stratified by gender were generated with the TG parameters ETP/Peak/LT/TTP/VI as main outcomes and BMI or the different types of lipid parameters (total cholesterol, LDL-C, HDL-C, triglycerides) as exposures. To build the multivariate models, the following strategy was applied: (1) simple univariate regression (UVR) analysis was used to identify independent/predictor variables associated with the dependent/outcome variable (ETP, peak, VI) at the level P < 0.05; (2) all the variables identified in the univariate analysis were inserted in a multiple regression (MR) analysis. For all models, the normality of residuals and homoscedasticity were evaluated by plotting standardized residuals against the predicted values. In case of violation of one or both of these conditions, the variable was not included in the model. The presence of collinearity among independent variables was tested by the variance inflation factor (VIF). A VIF of ≥10 indicates that the variable under consideration is almost a perfect linear combination of the independent variables already in the equation, in which case the variable is not added to the regression equation.

Further associations between TG parameters and general laboratory results and patient characteristics were performed by dividing the TG parameters into quartiles and performing ANOVA analysis on the groups to detect significant differences between high and low parameter values.

A two-sided P-value <0.05 was considered statistically significant for all analyses. IBM Statistical Package for Social Sciences (SPSS) version 25 software was used for all statistical analyses (SPSS Incorporated, Chicago, USA). Figures were prepared using GraphPad Prism version 5.00 (GraphPad Software, San Diego, USA).

We first determined baseline general characteristics for the subsection of the Moli-sani cohort in which we studied TG (Table 1). Our study population consisted of slightly more women (n = 11,766; 52.2%) than men (n = 10,780; 47.8%). On average, male subjects were older, more frequently smoking and had a slightly higher BMI than female subjects. Men had lower total cholesterol, LDL-C as well as HDL-C levels than women, although these differences were small. In contrast, male subjects had substantially higher triglyceride levels than female subjects. Blood glucose was higher in men, whereas CRP levels were slightly higher in women.

TG [as well as the risk for bleeding and thrombosis (32, 33)] is known to be influenced by age (34) and sex (35). Table 2 presents TG parameters measured in platelet poor plasma with the two types of TF concentrations, stratified by sex. For both the PPP reagent low and PPP reagent, women had significantly higher TG ETP, peak and VI, in combination with a shorter lagtime and TTP, compared to men. The inhibition of the ETP by thrombomodulin (TM) was significantly reduced in women (ETP_TM− = 1,764 ± 436 nM·min vs. ETP_TM+ = 1,426 ± 536 nM·min) compared to men (ETP_TM− = 1,692 ± 418 nM·min vs. ETP_TM+ = 1,346 ± 518 nM·min). This difference was slightly larger when considering only women taking oral contraceptives (n = 934), with an ETP inhibition of 12.11% (IQR 9%) and peak inhibition of 0.66% (IQR 4%), compared to male subjects (p <0.001) (data not shown).

The relationship of the ETP with age was also explored. Both in PPP reagent low and PPP reagent-induced TG, subjects in the lowest quartile of the ETP values were significantly (p <0.001) younger (mean age 54.5 ± 11.1 and 54.4 ± 11.1 years) compared to the individuals in the highest quartile of the ETP values (mean age 57.5 ± 12.7 and 57.4 ± 12.7 years), respectively. Additionally, the inhibition of the ETP by the anticoagulant actions of TM was inversely associated with age: subjects in the lowest quartile of ETP inhibition by TM were significantly older [median age 58 (IQR 21)] than the subjects whose ETP responded more to the inhibitory effect of TM [median age 53 (IQR 16)]. The other TG parameters were not significantly associated with age in the current Moli-sani cohort.

In both women and men, overweight (BMI 25–30 kg/m²) and obesity (BMI>30 kg/m²) were associated with an increased TG, characterized by increased ETP, peak and VI and prolonged lagtime and TTP (Supplementary Table 1). Inhibition of the ETP by the anticoagulant actions of TM was on average lower in obese men compared to men in the other BMI categories, while no significant differences were observed for TM-dependent inhibition in women.

These differences were further confirmed in linear regression models, where higher BMI was significantly associated with higher ETP and VI. The BMI of subjects in the upper quartile of the ETP was on average 8% higher than that of subjects in the lowest ETP quartile (p < 0.001). The same trend was observed for VI, where BMI was 4% higher in individuals in the upper quartile of VI distribution, compared to those in the lowest quartile (p < 0.001). Finally, a high BMI was associated with a more coagulable state of the blood, as measured by TG.

Since high BMI is related to altered blood lipid composition, we next studied the relationship between lipid profiles and TG in our study population. Blood lipid levels were divided into two groups according to the National Cholesterol Education Program (NCEP) guidelines (36), see Supplementary Table 2. Older individuals had significantly higher (≥240 mg/dL and ≥190 mg/dL) total cholesterol and LDL-C levels, respectively. Whereas, subjects with high triglyceride levels (≥200 mg/dL) were more frequently men, the groups with high total cholesterol (≥240 mg/dL), high LDL-C (≥190 mg/dL) and high HDL-C (≥60 mg/dL) contained more women. BMI, proportion of smokers, WH-ratio, CRP and glucose levels were all significantly higher in individuals with lipid levels above the cutoff threshold, except for HDL-C ≥60 mg/dL, which contained on average fewer smokers and more subjects with lower BMI, WH-ratio, CRP and glucose levels. Similarly, TG parameters were generally significantly increased in individuals with lipid levels above the cutoff, except for HDL-C, for which most differences in TG parameters between the two lipid level categories were non-significant (except for the time-dependent parameters).

The relation between lipids and TG was further explored in an association analysis, based on quartiles of the TG parameters ETP and peak (Table 3). LDL-C and HDL-C showed different associations with TG: individuals in the lowest quartile for ETP and peak had the lowest plasma LDL-C levels, whereas the HDL-C content did not differ between the quartiles. LDL-C levels were 9% higher in subjects in the highest compared to the lowest quartile of peak TG (p < 0.001), and 14% higher in those in the highest compared to the lowest quartile of the ETP (p < 0.001).

Finally, linear regression analyses were performed to investigate to what degree lipid levels contribute to the (normally distributed) TG parameters ETP, peak and VI (Table 3). In simple UVR, most lipid parameters were significantly associated with each of the TG parameters. Exceptions were HDL-C levels, which were not significantly associated with TG peak height in PPP reagent low-induced TG and with ETP in PPP reagent-induced TG, and LDL-C which was not associated with the VI in PPP reagent-induced TG. Triglyceride levels significantly contributed to PPP reagent low and PPP reagent-induced TG parameters. Total cholesterol had limited added predictive value for TG parameters (also due to collinearity) and was hence not included in most of the final MR models. ETP and peak were mainly affected by triglycerides and LDL-C, whereas VI was also influenced by HDL-C for both TG conditions. Overall, R² values of the final models were low, indicating that only a small proportion of the variance in TG parameters could be explained by lipid levels.

Many factors are known to influence TG, besides the blood lipid profile. Table 1 summarizes the most important variables that affect TG parameters. To adjust for these variables, multiple linear regression analysis was performed for all TG variables to investigate whether, and to what degree the lipid profile influences TG results, adjusting for other differences related to the lipid profile between the subjects (Table 4).

Total cholesterol levels are associated with a slightly prolonged lag time and time-to-peak, an elevated ETP and a reduced velocity index (Table 4). Triglycerides are positively associated with the lag time and time-to-peak, whereas the ETP and peak height significantly increase with increased triglyceride levels. LDL-C is posivitly associated with a prolonged lag time and time-to-peak, and an increased ETP, whereas HDL-C is associated with a reduced lag time and time-to-peak, and a lower TG peak height.

Additionally, as expected, age, sex and BMI were important influencers of TG parameters, especially, ETP, time-to-peak and velocity index. Smoking was associated with a significantly shorter lag time, and higher levels of acute phase reactant C-reactive protein were associated with a higher TG peak height and velocity index.