Serum level of HDL particles are independently associated with long-term prognosis in patients with coronary artery disease: The GENES study

HDL-Cholesterol (HDL-C) is not an accurate surrogate marker to measure the cardioprotective functions of HDL in coronary artery diseases (CAD) patients. Hence, measurement of other HDL-related parameters may have prognostic superiority over HDL-C. In this work, we examined the predictive value of HDL particles profile for long-term mortality in CAD patients and to compare its informative value to that of HDL-C and apoA-I. HDL particles profiles were measured by nuclear magnetic resonance (NMR) spectroscopy in 214 male participants with stable CAD (45–74 years). Median follow up was 12.5 years with a 36.4% mortality rate. Cardiovascular mortality accounted for 64.5%. Mean concentrations of total HDL particles (HDL-P), small-sized HDL (SHDL-P) and apoA-I were lower in deceased than in surviving patients whereas no difference was observed according to HDL-C and large HDL particles. All NMR-HDL measures were correlated between themselves and with other HDL markers (HDL-C, apoA-I and LpA-I). In a multivariate model adjusted for cardiovascular risk factors and bioclinical variables, HDL-P and SHDL-P displayed the strongest inverse association with all-cause and cardiovascular mortality. Weaker associations were recorded for apoA-I. Based on our results, we conclude that HDL particle profile measured by NMR spectroscopy should be considered to better stratify risk in population at high risk or in the setting of pharmacotherapy.

HDL-Cholesterol (HDL-C) has been repeatedly inversely related to cardiovascular risk in all epidemiological studies. However, pharmacological trials aimed at increasing HDL-C have failed to demonstrate a beneficial effect on clinical outcomes 1 . Also, some genetic variants associated to increased HDL-C have not been found associated to a decreased cardiovascular risk 2 but those HDL randomization studies are questionable because they disregarded the complexity of lipoprotein metabolism by excluding from their analyses important genes that not exclusively regulate HDL-C levels but also those of other lipoproteins 3,4 . This has led to the concept that a single measurement of HDL-C does not necessarily reflect the functional properties of HDL particles and their effects against atherosclerosis. Indeed, HDL particles are heterogeneous in size and biochemical composition, and HDL subpopulations might have different functional properties 5 . NMR-spectroscopy has been recently proposed as a tool to quantify HDL particles and HDL subpopulations 5 . This technology enables to measure the total concentration of HDL particles and their size distribution. Numerous recent studies have shown that the atheroprotective properties of HDL are supported by small and medium-sized HDL particles 6 , which were inversely related to cardiovascular risk in various clinical settings 7,8 .

Results
Characteristics of CAD patients according to vital status. The present cohort was constituted of 214 CAD male patients. After inclusion, the vital patients' status was yearly assessed. The median follow-up period was 12.5 years (mean: 10.7 years). During follow-up, 78 deaths had been recorded giving a death rate of 36.4% and a mean annual rate of 3.4%. Cardiovascular mortality accounted for the majority of deaths recorded (64.1%, n = 50) and cancers accounted for 16.7% (n = 13).
Comparison of patients' data when they were included in the cohort is given in Table 1. Results are presented distinguishing two groups: the "alive group" (patients living at the end of the follow-up) and the "deceased group" (patients who died during follow-up).
Hypertension was diagnosed or treated in 65.6% patients (45.5% under treatment) with no difference between the deceased and alive groups (p = 0.95). Dyslipidemia was diagnosed or treated in 71.3% (65.4% treated) of the alive group, and in 50.6% (43.6% upon treatment) of the deceased group and the difference was statistically significant (p = 0.002). Among patients treated for dyslipidemia, 87.8% were treated with statins, 16.2% with fibrates, 4% with both statins and fibrates and none were treated with niacin or other lipid-lowering therapy. Diabetes was diagnosed in 18.7% (16.9% treated) of the alive group and in 41.8% (37.2% upon treatment) of the deceased group and the difference was statistically significant (p = 0.001).
Patients having deceased during the follow-up period had a longer duration of CAD, a decreased left ventricle ejection volume (LVEF), a higher heart rate and a more severe angiographic lesion score (Gensini). Regarding cardiovascular risk factors, smoking habits and treatment for diabetes were more frequent in the deceased group, whereas lipid-lowering therapy was less frequent. Among lipoprotein parameters, only apoA-I, a major HDL marker, was significantly lower in further deceased patients. Hs-CRP, an inflammatory marker, was lower in surviving than in deceasing patients. NT-proBNP levels were significantly higher in further deceased patients but no difference was observed in the concentration of hs-TnT between the two groups, indicating that deceased patients suffered from a more severe myocardial dysfunction rather than from a more extended myocardial necrosis.
The distribution of prior cardiovascular events in the studied population of stable CAD patients is shown in Supplementary Table 1. The majority is represented by myocardial infarction (MI, 52.3%), followed by revascularization procedures (37.9%) and then by other ischemic heart disease (IHD, 9.8%, e.g. stable angina). Logically, among patients deceased during follow-up, past history of MI and other IHD were more frequent than in alive patients.
HDL particles according to vital status. HDL particles' profile was determined by NMR spectroscopy, enabling to distinguish large HDL (LHDL-P, 8.8-13 nm) and small-sized HDL (SHDL-P, 7.3-8.7 nm) particles. The latter accounted for about ∼85% of total HDL particles (HDL-P). HDL-P was ∼10% lower in deceased than in surviving patients (24.6 μmol/L [SD, 6.0] vs. 27.5 μmol/L [SD, 4.9], p = 0.001, Table 2). This difference was entirely due to a decreased number of SHDL-P, whereas number LHDL-P was not different according to the vital status ( Table 2). The average size of total HDL particles (HDL size) was found higher in deceased patients (8.94 nm versus 8.82 nm, p = 0.014).

correlations between HDL particles measures and clinical and biological parameters.
Correlations were investigated between markers of HDL particles and other clinical or biological parameters in the study population (Table 3). All NMR-HDL measures were correlated between themselves and with other HDL markers: HDL-C, apoA-I and lipoprotein A-I (LpA-I). Logically, the average HDL size was correlated positively with the number of LHDL-P, and negatively, with the number of SHDL-P. Triglycerides were associated positively with HDL-P and SHDL-P but negatively with LHDL-P, mean HDL size and HDL-C. No association was observed with apoA-I. These correlations might reflect the remodeling of HDL lipids induced by the cholesterol ester transfer protein (CETP) acting between HDL particles and triglyceride-rich lipoproteins. Alcohol consumption positively correlated with HDL-C and HDL-P, and more specifically with SHDL-P but not with apoA-I. Inflammation, as documented by plasma hs-CRP, was inversely associated with apo A-I, HDL-P and SHDL-P, but not with LHDL-P. HDL-P, SHDL-P and apoA-I were negatively associated with NT-proBNP and hs-TnT.
The severity of coronary lesions, as illustrated by the Gensini score, was inversely related to HDL-P. Strong positive associations were observed between LVEF and both HDL-P and SHDL-P. ApoA-I levels correlated with LVEF but not significantly with the Gensini score. For comparison, no relationship was recorded between HDL-C and either LVEF or the Gensini score. Thus, although HDL markers were strongly correlated between themselves, total and subclasses of HDL particles displayed specific association with clinical variables. total and cardiovascular mortality according to tertiles of HDL markers. Each one HDL marker was considered according to tertiles of its distribution in the whole study population (Table 4). Death rates during follow-up were determined across the different tertiles and associations were determined after adjustment on classical risk factors including age, smoking and treatments for hypertension, diabetes and dyslipidemia. Similar associations were observed without adjustment. The strongest association to total and cardiovascular mortality was observed for HDL-P distribution. A 45% reduction in death rates was recorded in tertiles 2 and 3, as  Table 1. Clinical and biological characteristics in coronary artery disease patients when they were first included in the GENES cohort. Data are expressed in mean (SD) or %. BMI, body mass index; hs-CRP, high-sensitivity C-reactive protein; eGFR, estimated glomerular filtration rate; hs-TnT, high-sensitive cardiac troponin T; NT-ProBNP, N-terminal pro-brain natriuretic peptide; LVEF, left ventricular ejection fraction; CAD, coronary artery disease. a "high" physical activity during 20 min at least twice a week versus "low" physical activity once a week or less. b Hypertension, systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg or treatment; Dyslipidemia, total cholesterol ≥2.50 g/L or treatment; Diabetes, fasting blood glucose ≥7. 8  . Taking into account fibrate and/or statin use in statistical analyses had no effect on the association of HDL-P and SHDL-P particles with mortality (not shown). By contrast, no association between LHDL-P and mortality was observed, except for an almost significant positive trend (p = 0.07) between LHDL-P and death rates. Concordantly, death rates were significantly different across HDL size distribution, an increase in particles size being associated with highest death rates. Considering classical HDL markers, HDL-C tertiles did not display different death rates, yet a 1 SD increase in HDL-C was associated with a significantly lower HR for total mortality (0.78, p = 0.03). ApoA-I distribution was associated to total and cardiovascular mortality; each 1 standard deviation increase of apoA-I was associated to a ∼31% risk reduction (HR = 0.69 [95% CI, 0.54-0.88] and 0.67 [95% CI, 0.49-0.91], respectively). A further multivariate analysis was conducted using a forward stepwise selection of adjustment variables (Fig. 1). Regarding total mortality, the variables retained in the model were age, smoking, treatment for dyslipidemia, renal function (eGFR), LVEF, duration of CAD and Gensini score, the 3 last parameters reflecting  Table 3. Correlation between NMR HDL measure and biological and biochemical parameters and other cardiovascular risk factors in coronary artery disease patients. Spearman rank correlation coefficients (95% confidence interval). *p < 0.05, **p < 0.01, ***p < 0.001. a Score from no physical activity to at least 20 minutes of vigorous physical activity at least twice a week. BMI, body mass index; hs-CRP, high-sensitivity C-reactive protein; eGFR, estimated glomerular filtration rate; hs-TnT, high-sensitive cardiac troponin T; NT-ProBNP, N-terminal pro-brain natriuretic peptide; LVEF, left ventricular ejection fraction.
Associations between HDL markers and mortality are illustrated in the survival curves established during the whole follow-up period for the different tertiles (Fig. 2). Death rates were regular during the whole time course of follow-up. For both HDL-P and SHDL-P, patients in the first tertile had a poorer survival than patients in tertiles 2 and 3. This was particularly evident for early events: of the 17 deaths recorded during the 0-3 year follow-up, 11 (65%) were in the lowest tertile of HDL-P as compared to 8 out of 20 (40%) for the deaths recorded in the 9-12 year period. A comparable trend was observed for apoA-I distribution yet survival differences during the whole period did not reach statistical significance.

Discussion
In the present study, serum levels of total HDL particles (HDL-P) and of small-sized HDL particles (SHDL-P) were inversely related to all-cause as well as to specific cardiovascular mortality in CAD patients. Every 1-SD increase of HDL particle number was associated to a 41% decrease in cardiovascular mortality, after multiple  www.nature.com/scientificreports www.nature.com/scientificreports/ adjustments on cardiovascular risk factors and on clinical markers of heart condition, including LVEF, duration of CAD and Gensini score. Among other HDL markers, apoA-I was also inversely related, though to a lesser extent, to total and cardiovascular mortality, and HDL-C was found weakly associated to all-cause mortality. Conversely, large HDL particles (LHDL-P) were not associated with total and cardiovascular mortality. However, higher death rates were recorded as average HDL particle size increased.
To date, the inverse and independent association between HDL-P / MS-HDL-P and cardiovascular risk has been extensively documented with respect to primary prevention, either in individuals without baseline CAD 7,15-18 and in those with pre-clinical atherosclerosis, as documented by carotid intima-media thickness 7,19 , or coronary calcifications 20 . More recently, in a large study carried out in high-risk individuals undergoing coronary catheterization for suspicion of CAD, followed-up during 8 years, HDL-P and MS-HDL-P were independent predictors of all-cause mortality 8 .
With respect to secondary prevention, only a few studies have evaluated the relationship of NMR-derived HDL particle subclasses with cardiovascular disease outcomes in CAD patients. In a prospective nested case-control study of 364 men with new CAD events (non-fatal myocardial infarction or cardiac death) during a 5.1 year follow-up paired to 697 age-matched control, total HDL-P and small HDL-P (7.3-8.2 nm) were strong, independent, predictors of recurrent coronary events, whereas levels of HDL-C were not 21 . Similarly, inverse association between HDL-P and coronary events has also been reported in a large mixed-gender cohort of CAD patients during a 5.3 years of follow up 22 . In patients suffering from acute heart failure, concentrations of both HDL-P and small HDL-P (7.3-8.7 nm) were inversely related to short-term (3-month) mortality, after multiple adjustments on confounding variables, including NT-proBNP, a classical marker of heart failure 23 .
The present study brings some additional insights on this relation between NMR-derived HDL particles subclasses and long-term prognosis in CAD patients, with angiographically documented coronary lesions. Association of HDL markers to all-cause and cardiovascular mortality was assessed after adjustment on a large variety of confounders: life-style parameters, clinical and biological variables documenting cardiovascular risk Figure 1. Relative risk of all-cause and cardiovascular mortality as a function of HDL-C, apoA-I, HDL-P, SHDL-P and LHDL-P. Graphic represents hazard ratios (dots) and corresponding 95% confidence interval (95% CI) for risk of all-cause and cardiovascular mortality per 1 standard deviation increase of HDL-C, apoA-I, HDL-P, SHDL-P or LHDL-P. Among the 78 deceased patients analyzed, the number of CV death was 50. Model 1: adjusted for age, smoking, treatment for dyslipidemia, eGFR, LVEF, duration of CAD and Gensini score. Model 2: adjusted for treatment for dyslipidemia, LVEF, duration of CAD and Gensini score. (2020) 10:8138 | https://doi.org/10.1038/s41598-020-65100-2 www.nature.com/scientificreports www.nature.com/scientificreports/ factors, inflammatory status, renal function and heart condition. In addition, this study has enabled to compare the predictive value of apoA-I versus HDL-P measurements, which showed comparable associations, although Hazard ratios were better with HDL-P levels.
HDL particles, and most particularly small-sized HDL, may act against atherosclerosis through different mechanisms. Small HDL behave as the best acceptors of ABCA1-mediated cholesterol efflux from macrophages, leading subsequently to the mobilization of intracellular cholesterol to the plasma membrane 24,25 . Small and dense HDL particles also protect LDL from oxidation. HDL particles act through removing phospholipid hydroperoxides from LDL and by inactivating oxidized lipids by specific enzymes like paraoxonase-1 (PON-1) and PAF-acetylhydrolase 26,27 . Moreover small protein-rich HDL exert anti-inflammatory properties by depressing expression of VCAM-1 at the surface of endothelial cells 28 . On these cells, HDL particles appear to be cytoprotective by inhibiting apoptosis induced by oxidized LDL, and small HDL 3 would be the most effective in this function 29 . Altogether those observations suggest that the proteome associated to small HDL particles support various biological activities, which impair atherosclerosis development.
Moreover, HDL particles may exert beneficial effects on myocardial functions. Indeed, in different experimental contexts, it was demonstrated that HDL particles protect against ischemia reperfusion injury 30 , leading to a reduction in infarct size. HDL may also improve myocardial function by reducing ventricular remodelling following infarction 31 . In isolated cardiomyocytes, HDL particles were shown to prevent apoptosis through an www.nature.com/scientificreports www.nature.com/scientificreports/ AMP-kinase dependent mechanism 32 . These experimental observations on a direct impact of HDL on myocardial functions might translate into clinical impacts. In support of this concept is the positive correlation observed here between HDL-P, small HDL-P and the left ventricular ejection fraction, concordant with the negative association between HDL-P, small HDL-P and NT-proBNP observed here and previously reported 23 , although potential confounders might interfere with these associations.
In this study, concentrations of large HDL particles were not associated to mortality. However, higher death rates were recorded as HDL size increased (p < 0.01); following multiple adjustments, association to total mortality for the upper tertile of HDL size was close to statistical significance (p = 0.06). Similar observations regarding all-cause mortality in individuals who are at high cardiovascular risk have been previously reported 8 . Large HDL-P might be less effective than SHDL-P regarding various atheroprotective functions, like cholesterol efflux, anti-oxidative and anti-inflammatory properties, and cytoprotective effects on endothelium 6,33 . Moreover, accumulation of large HDL might reflect a defect in HDL catabolism, and particularly in HDL liver uptake, which constitutes the last step of reverse cholesterol transport 34 . Similarly, we did not observe any association of HDL-C with mortality. This is concordant with the lack of association between LHDL-P and mortality, since HDL-C mainly reflects cholesterol associated with large, lipid rich, HDL particles.
ApoA-I was inversely related to mortality: for each 1-SD increase of apoA-I, a 31% and 33% decrease in all-cause and cardiovascular mortality was recorded, respectively. So far, apoA-I has been little used in epidemiological studies. However, calibration on reference international standards has made the immunoassay of apoA-I robust and comparable between studies. Furthermore, apoA-I measurement is much less influenced than HDL-C by intravascular enzymes and lipid transfer proteins, which participate in HDL remodelling. Thus, apoA-I measurement may improve assessment of cardiovascular risk 35 . Association to mortality was somewhat weaker for apoA-I than for HDL-P or SHDL-P. This might be explained by the fact that the apoA-I content per particle varies on average from 2 to 4, between small HDL 3 and large HDL 2 36 , so that large HDL particles are somewhat overrepresented in apoA-I quantification.
A number of limitations of the present study must be noted. First, the small size of the study is a limitation. Indeed, the first 214 consecutive CAD patients from the GENES cohort where included in the present study, which represent 25% of the whole cohort (n = 834). Accordingly, mortality rate was higher in included patients than in the whole cohort (36.4% versus 29.1%, data not shown), probably because of the longer duration of follow-up for those first included patients. Clinical and biological characteristics were comparable, between included and non-included patients, except for dyslipidemia treatment, somewhat less frequent in included patients (57.5% vs 65.8%, p = 0.04, Supplementary Table 2). This difference might result from difference in statin administration practice between the beginning (year 2001) and the end (2004) of the GENES study. Second, this study was designed only with men, which has the advantage of recording a larger number of events than in a mixed all-gender cohort, particularly in south-western France where the incidence rates are particularly low for women under 65 years of age 37 but limited the translatability of our results to women. Third, despite adjustments on established CVD risk factors the possibility remains that other important confounding variables with effects on HDL parameters were not measured or considered in our analyses. Finally, NMR-based HDL classification in large and small HDL particles provides limited insights into the biology of these complex particles, now known to include more than 10 subspecies, for which proteome and lipidome compositions affect functional properties 38 . In particular the pre-beta HDL which are the principal acceptors for cholesterol effluxed from the arterial wall 39 are not accurately quantified by this method. Thus, comparison with other methods of analysis of HDL subfractions 40 would be relevant to confirm our findings and to improve our understanding of HDL functions.
In conclusion, the present study demonstrated that the concentration of total HDL particles and small-sized HDL particles may serve as a better prediction tool than HDL-C and apoA-I to assess long term prognosis in coronary patients. Studies on HDL metabolism had progressively led to the schematic view of an interconversion cycle of HDL particles in the plasma compartment, driven by cell cholesterol efflux, enzymes like LCAT, lipases and lipid transfer proteins 41,42 . More recently the concept has emerged that HDL particles of different geometry and chemical composition have distinct metabolic fate and display specific functional properties 5,6,43 . This supports the idea that HDL functionality might be more precisely assessed by the quantification of specific HDL particles with high atheroprotective effects. In the future, quantification of HDL particle concentration and HDL subclasses along with HDL functional measurement could be clinically useful in CVD risk assessment.

Methods
Study participants. The "Geńetique et Environnement en Europe du Sud" (GENES) study is a case-control study designed to assess the role of genetic, biological and environmental determinants in the occurrence of CAD 44 . Written informed consent was obtained from all participants, and the study protocol was conducted in accordance with the Helsinki Declaration and approved by the local ethics committee (Comité consultatif de protection des personnes dans la recherche biomédicale (CCPPRB), Toulouse/Sud-Ouest, file #1-99-48, Feb 2000). All research was performed in accordance with procedures and regulations at the Toulouse University Hospital. A blood sample collection has been constituted (declared as DC-2008-463 #1 to the Ministry of Research and to the Regional Health Authority). As previously described, cases were stable male CAD patients living in the Toulouse area (South-west France), aged 45-74 and prospectively recruited from 2001 to 2004 after admission to the Cardiology department, Toulouse University Hospital, for cardiovascular examination and referred for evaluation and management of their CAD 45 . Patients included in the present study presented a stable CAD that was defined by a previous history of acute coronary syndrome, a previous history of coronary artery revascularization, a stable angina or a documented myocardial ischemia, as assessed by presence of coronary stenosis of ≥50% of luminal narrowing at the coronary angiography. Patients who had presented an acute coronary episode in the seven days prior to the recruitment were not included in the study, because they were considered unstable. In the present analysis, we only took into account the first 214 consecutive patients (i.e. those with CAD) in whom NMR-HDL profile was measured and complete data were available for all the subjects. The minimal sample size (n = 186), increased by 15% to take into account missing data, was calculated to detect a HR of 0.8 (for one SD increase of HDL-C) with a 80% power at a 0.05 significance level. The sample size was adjusted for an anticipated event rate of ∼35% corresponding to a median follow-up of 12.5 years.

Assessment of the vital status.
Vital status was obtained for each participant through the national database ("RNIPP"), which records, every year, all deaths occurring in the French population (http://cesp.vjf.inserm. fr/svcd). For each patients, vital status was assessed yearly from the year of recruitment until December 31, 2014, with a median follow up of 12.5 years. All dates and causes of death were obtained for participants who died during the follow-up. Main and associated causes of deaths were provided by the French National Institute of Health Research (CépiDc-INSERM), which systematically collects and codes (using the International Classification of Diseases coding system) data recorded on death certificates. Death from a cardiovascular cause during follow-up was assessed by a committee of four medical doctors, every time cardiovascular disease was reported as the main cause of death, or when it was mentioned as an associated cause, if the main cause was a plausible complication of cardiovascular disease. Authorizations to use these data were obtained in accordance with French law (Commission nationale de l'informatique et des libertés (CNIL): authorization 355152v1, September 3, 2008).
Biological measurement. Blood was collected after an overnight fast. Serum sample aliquots were subsequently stored at −80 °C until biological analyses. The following biomarkers were assayed with enzymatic reagents on automated analyzers (Hitachi 912 and Cobas 8000, Roche Diagnostics, Meylan, France): serum total cholesterol, HDL-C, triglycerides, fasting glucose, creatinin. eGFR was calculated using the abbreviated Modification of Diet in Renal Disease (MDRD) Study equation 46 . ApoA-I, high-sensitive C-Reactive protein (hs-CRP), N-terminal pro-brain natriuretic peptide (NT-proBNP) and high-sensitive cardiac troponin T (hs-TnT) were determined on the same analyzer by immunoturbidimetry assays (Roche Diagnostics). Lipoprotein A-I, which refers to lipoproteins containing apoA-I but not apoA-II, were measured by an electroimmunodiffusion technique 47 using the HYDRAGEL LPAI PARTICLES Kit (Sebia, Issy-les-Moulineaux, France).

Data collection.
Age, environmental characteristics and information on cardiovascular risk factors were collected through standardized face-to-face interviews, performed by a single physician. Past medical history was collected and checked in the patients' medical files. Presence of dyslipidemia, diabetes mellitus or hypertension was assessed from the subjects' current treatments. Dyslipidemia was defined as treatment with drugs or fasting serum total cholesterol ≥2.40 g/L. Hypertension was defined as treatment with drugs or systolic blood pressure ≥160 mmHg or diastolic blood pressure ≥95 mmHg. Diabetes was defined as treatment with drugs or fasting blood glucose ≥7.8 mmol/L. Smoking status was classified as current smokers, past smokers having quit for more than 3 years and patients having never smoked. Among current smokers, cigarette consumption was estimated with the pack-year quantification and recorded as the average number of cigarettes per day. Alcohol consumption was assessed using a typical week pattern. The total amount of pure alcohol consumption was calculated as the sum of different types of drinks and was expressed as grams per day. Physical activity was investigated through a standardized questionnaire 48 and categorized into three levels as: no physical activity, moderate physical activity during 20 minutes no more than once a week, and high physical activity during 20 minutes, at least twice a week. Blood pressure and resting heart rate were measured with an automatic sphygmomanometer (OMRON 705 CP). Measurements were performed after a minimum of 5 minutes rest; average values from two different measurements were recorded for further analysis.

Assessment of CAD severity and extension and estimation of cardiac function. All stable CAD
patients enrolled had a coronary angiography even without evidence of active chronic ischemia. Coronary artery stenoses of ≥50% luminal narrowing were considered significant. Extent of coronary artery disease lesions was assessed by calculating the Gensini Score, based on data from coronary angiography [49][50][51] . Left Ventricular Ejection Fraction (LVEF) was assessed by contrast ventriculography using an isotopic method, and/or by echocardiography.
HDL measurement by nuclear Magnetic Resonance (nMR) spectroscopy. HDL particle concentration and size were measured by NMR spectroscopy using the AXINON lipoFIT-S100 test system (Numares AG, Regensburg, Germany) as previously described 11,12,23 . Serum samples for NMR spectroscopy were stored at −80 °C and kept unthawed until the day of NMR measurements. Serum (630 μL) gently mixed with 70 μL of an additives solution containing reference substances, NaN3 and D2O, and 600 μL of the mixture were transferred into 5 mm NMR tubes with barcode-labeled caps. Briefly, 1 H NMR spectra were recorded at a temperature of 310 K on a shielded 600 MHz Avance III HD NMR spectrometer (Bruker Biospin) with a 5 mm triple resonance TXI probe head including deuterium lock channel, a z-gradient coil and automatic frequency tuning and matching. Prior to each analytical run, calibration was performed using a calibration sample comprising an aqueous solution of various calibration substances with different molecular masses, 0.01% (w/v) NaN3, 10% (v/v) D 2 O as a locking substance and 1% glycerol to adjust viscosity. Two identical control samples were measured directly after calibration and at the end of each run. Each spectrum was referenced, normalized and subjected to a set of quality checks including checks of baseline properties, noise level, shift, width, and symmetry properties of quality control signals. Lipoprotein analysis was conducted via deconvolution of the broad methyl group signal at about 0.9-0.8 ppm. In this process, lipoprotein subclasses are reflected by a fixed number of pre-defined bell-shaped (e.g. Gaussian or Lorentzian) base functions, each of which has a constant position and defined width. The concentrations of lipoprotein particle subclasses as well as the average particle size were calculated based on the integrals attributable to specific base functions. Fit quality was checked by calculating the residual deviation between fit and spectrum intensity. In this study, the concentrations of large-sized HDL particles (LHDL-P), small-sized (2020) 10:8138 | https://doi.org/10.1038/s41598-020-65100-2 www.nature.com/scientificreports www.nature.com/scientificreports/ HDL-particle (SHDL-P) and total HDL particles (HDL-P, reported in μmol/L) as well as the average HDL particle size (HDL size, reported in nm) are used. The two measured HDL subclasses had the following estimated diameter ranges: LHDL-P, 8.8-13 nm; SHDL-P, 7.3-8.7 nm. Statistical analyses. Continuous variables are displayed as means and standard deviations (SD). Categorical variables are presented as proportions. We first described and compared characteristics of participants according to vital status. Categorical variables were compared between groups using the χ 2 -test (or Fisher's exact test when necessary). Student's t-test was used to compare the distribution of continuous data. A Wilcoxon Mann-Whitney's test (or logarithmic transformation of the variable when necessary) was performed when distribution departed from normality, or when homoscedasticity was rejected. Spearman rank correlations were used to test the associations of NMR-HDL parameters and HDL-C with cardiovascular risk factors, severity, extension and estimation of cardiac function of the disease.
Cumulative survival of patients were determined by the Kaplan-Meier method and compared, using the Log-rank test for the individual endpoints of all-cause mortality. The relation between baseline variables and mortality was assessed using Cox proportional hazards regression analysis. We tested the proportionality assumption using cumulative sums of martingale-based residuals. We performed regression analyses with polynomial models (quadratic and cubic) to examine for possible non-linear relations between continuous variables and mortality. Cox regression analyses were performed first without any adjustment for co-variables and, second, with adjustment on classical cardiovascular risk factors (age, smoking, treatments for dyslipidemia, hypertension and diabetes). In addition, a forward stepwise selection was used to create the multivariate models. Variables with an entry criterion of p ≤ 0.20 were used as candidate covariates in a multivariate cox regression model. Variables were maintained in the model with a retention criterion of p ≤ 0.10 for all-cause of death analyses and p ≤ 0.05 for cardiovascular mortality. The same covariates were identified for all HDL markers studied. The following variables were introduced into the model as covariates when considering all cause of death: age, smoking, treatment for dyslipidemia, eGFR, LVEF, duration of CAD and Gensini score. For cardiovascular mortality the variables retained were treatment for dyslipidemia, LVEF, duration of CAD and Gensini score. All statistical analyses were carried out using the SAS statistical software package 9.4 (SAS Institute, Cary, NC). Analyses were two-tailed and p < 0.05 was considered to be significant.