Blood plasma lipidomic signature of epicardial fat in healthy obese women
Author contributions: FPJM SR VG conceived and designed the experiments, FPJM SDQ VG performed the experiments, FPJM IM MS analyzed the data, all authors were involved in the writing and had final approval of the submitted and published versions.
Disclosure: The authors declare no conflict of interest.
Abstract
Objectives
A lipidomic approach was employed in a clinically well-defined cohort of healthy obese women to explore blood lipidome phenotype ascribed to body fat deposition, with emphasis on epicardial adipose tissue (EAT).
Methods
The present investigation delivered a lipidomics signature of epicardial adiposity under healthy clinical conditions using a cohort of 40 obese females (age: 25–45 years, BMI: 28–40 kg/m2) not showing any metabolic disease traits. Lipidomics analysis of blood plasma was employed in combination with in vivo quantitation of mediastinal fat depots by computerized tomography.
Results
All cardiac fat depots correlated to indicators of hepatic dysfunctions (ALAT and ASAT), which describe physiological connections between hepatic and cardiac steatosis. Plasma lipidomics encompassed overall levels of lipid classes, fatty acid profiles, and individual lipid species. EAT and visceral fat associated with diacylglycerols (DAG), triglycerides, and distinct phospholipid and sphingolipid species. A pattern of DAG and phosphoglycerols was specific to EAT.
Conclusions
Human blood plasma lipidomics appears to be a promising clinical and potentially diagnostic readout for patient stratification and monitoring. Association of blood lipidomics signature to regio-specific mediastinal and visceral adiposity under healthy clinical conditions may help provide more biological insights into obese patient stratification for cardiovascular disease risks.
Introduction
Metabolic syndrome encompasses multifactorial metabolic abnormalities including visceral obesity, glucose intolerance, hypertension, hyperuricaemia, dyslipidemia, and nonalcoholic fatty liver disease, all of which are associated with excess cardiovascular deaths (1). Although insulin resistance still remains a key mechanism underlying this pathophysiology, many studies further investigate the more complex etiology. In obese subjects, the inability of adipose tissue to expand or to store fat results in lipid overflow to other organs under conditions of excess caloric intake and physical inactivity (2). Metabolically deleterious and life-threatening forms of obesity associate with accumulation of fat in the visceral adipose tissue and ectopic fat (2, 3). In particular, higher visceral adiposity (VA) associates with a higher probability of developing atherosclerotic cardiovascular diseases mediated via complex metabolic signaling that interplays with the endocrine and immune systems (4, 5).
VA includes a complex topographical fat deposition, namely mesenteric and epicardial adipose tissues (EAT), as well as peripheric depots around other organs (6). Despite a direct relationship between the amount of EAT, VA, and general body adiposity, EAT is considered as a more specific cardiovascular risk indicator, and a determinant of coronary artery disease (e.g., coronary atherosclerosis, coronary artery atherothrombosis) (7) and hypertension (8, 9). EAT metabolism is uniquely regulated due to high rates of fatty acid incorporation, insulin-induced lipogenesis or fatty acid breakdown, and low expression of lipoprotein lipase, stearoyl-CoA desaturase and acetyl-CoA carboxylase-α (7). However, the relationship between VA and cardiac steatosis remains complex. While the accumulation of triglycerides (TG) in and around the myocardium of obese subjects was linked to free fatty acid exposure, generalized fat excess, and peripheral vascular resistance, myocardial TG content may present a separate entity that is influenced by factors beyond VA (2, 10). A recent study confirmed the occurrence of a connection between insulin resistance, cardiac ectopic fat deposition, and cardiac dysfunction in morbid obesity (11). Increasing interest for general public health points towards EAT as a major predictor for metabolic health in overweight and obese children (12). Moreover, obesogenic and diabetogenic conditions may lead to very different adiposity-associated cardiometabolic risks (13). In this context, metabolic contribution from EAT and VA to metabolic syndrome strongly depends on ethnic background and other disease states, which require further studies in well-controlled cohorts (14).
Omics-rooted molecular phenotyping and systems biology developments are expected to change clinical practice and enhance understanding of the interactions between genetics, environmental factors, and individual health and disease trajectory (15). Recent applications have associated specific metabolite and lipid profiles to body fat distribution (16, 17), which may in turn generate new mechanistic knowledge of complex physiological processes if validated. Towards this end, comprehensive assessment of the lipidome will help further expand current knowledge on molecular lipid species associated with these diseases (18-20).
Recently, in a clinically well-defined cohort of healthy obese women, we have explored the metabolic signatures ascribed to VA and body composition metabolic phenotypes generated by dual energy X-ray absorptiometry (DXA) and computed tomography (CT) (21). The integration of these visceral fat estimates with blood and urine metabolic profiles described a distinct amino acid, diacyl, and ether phospholipid phenotype in women with higher visceral fat (21). In the present study, we sought to further investigate the plasma lipidome with regards to specific visceral adipose depots. This work takes advantage of modern imaging technologies based on CT to generate accurate regio-specific quantification of visceral fat depots, including EAT (22). Therefore, we applied a shotgun lipidomics approach to link circulating levels of blood plasma lipids to body fat deposition, with emphasis on VA and EAT. The present study sheds new light on the underpinning lipidome variation in relation with visceral and epicardial adiposity in healthy obese women.
Methods
Ethics statement
This work was approved by the Ethical Committee of Lausanne University School Medicine (Lausanne, Switzerland). All participants gave written informed consent in French or in English as described in the consent procedure of the study protocol approved by the Ethics committee. The clinical study is registered at clinicaltrials.gov with the identifier NCT01726647.
Participants and experimental design
The observational study was conducted on 40 healthy obese Caucasian women at the out patient obesity clinic of the University Hospital of Lausanne (CHUV), Switzerland (21). Briefly, the participants had a BMI between 28 and 40 and ages between 25 and 45 years old, and they showed no metabolic disease traits. Additional exclusion criteria were diabetes, pregnancy, antibiotic therapy within 1 month prior to the beginning of the study, any therapy (contraception apart) within the run-in period of 1 week before the visit day, and eating disorders. Subjects having recently undergone a weight loss of more than 3 kg during the last 3 months were also excluded. In the current cohort, no subjects suffered from hypertension, glucose intolerance, polycystic ovary syndrome, thyroid dysfunction, or adrenal disorders.
Clinical, anthropometric, and body composition measurements
Clinical and anthropometric measures (body weight, height, waist, and hip) were conducted using standard clinical practices, following previously reported procedures (21) (Tables 1 and 2). Full-body scans were performed to determine both abdominal fat distribution and total body composition. Total-body scans were made on a GE Lunar iDXA system (software version: enCORE version 12.10.113) with scan mode automatically determined by the device and using the previously reported procedure (21). Similarly, quantification of mediastinal fat was obtained using semi-automatic segmentation method. First, mediastinal fat was detected using a dedicated interactive thresholding method. The volume of the segmented region of interest was calculated automatically. Then, the pericardium was segmented manually to delineate the epicardial fat (intrapericardial) and the extrapericardial fat from the total mediastinal fat. The volume of the epicardial fat was then calculated automatically. High-resolution cardiac fat deposition was generated from 28 out of the 40 subjects.
| Factor | Mean ± SD (Min-Max) |
|---|---|
| Intrapericardial fat volume (cm3) | 77.1 ± 33.0 (25.4-144.2) |
| Mediastinal total fat volume (cm3) | 136.0 ± 55.5 (44.6-231.7) |
| Mediastinal extrapericardial fat volume (epicardial fat, cm3) | 58.9 ± 26.1 (19.2-101.8) |
| Ratio epicardial/total mediastinal | 0.4 ± 0.1 (0.3-0.7) |
| Ratio epicardial/intrapericardial | 0.8 ± 0.3 (0.4-2.2) |
| Abdominal fat volume (ml) | 20123.4 ± 4443.7 (11338-33643) |
| Subcutaneous fat volume (ml) | 15733.5 ± 3571.4 (9263-25006) |
| Intraperitoneal fat volume (ml) | 4451.6 ± 1392.2 (2075-8637) |
| Log10 IPVF | 3.6 ± 0.1 (3.3-3.9) |
| Log10 ratio 1 (VAT/SAT) | −0.7 ± 0.1 (−0.9--0.5) |
| Log10 ratio 2 (VAT/abdominal fat) | −0.6 ± 0.1 (−0.8--0.3) |
| Abdominal total vol (ml) | 34735.0 ± 5794.5 (24857-53498) |
| Android/gynoid fat ratio | 0.5 ± 0.1 (0.3-0.8) |
| Total fat mass (g) | 44525.0 ± 8743.0 (31389.6-68339) |
- IPVF, intraperitoneal fat; SAT, subcutaneous fat tissue; VAT, visceral adipose tissue.
| Factor | Mean ± SD (Min–Max) |
|---|---|
| Age (years) | 35.6 ± 5.1 (25-46) |
| BMI (kg/m2) | 35.5 ± 3.7 (29-43.7) |
| Hip (cm) | 124.9 ± 7.6 (110-144) |
| Waist (cm) | 103.5 ± 11.0 (87.6-127.5) |
| Waist/hip ratio | 0.8 ± 0.1 (0.7-1) |
| Na (mmol/l) | 140.7 ± 1.4 (138-143) |
| K (mmol/l) | 4.0 ± 0.2 (3.5-4.4) |
| Glucose (mmol/l) | 5.2 ± 0.5 (4.3-6.3) |
| Creatinine (mmol/l) | 66.5 ± 9.2 (50-93) |
| Cholesterol (mmol/l) | 5.5 ± 0.9 (3.8-7) |
| HDL-C (mmol/l) | 1.4 ± 0.3 (0.9-2.6) |
| LDL-C (mmol/l) | 3.5 ± 0.8 (2.2-4.9) |
| TG (mmol/l) | 1.5 ± 1.2 (0.4-8) |
| Urates (µmol/l) | 282.2 ± 57.4 (172-421) |
| ASAT (U/l) | 22.8 ± 5.5 (13-40) |
| ALAT (U/l) | 22.1 ± 9.2 (10-56) |
| ALAT/ASAT ratio | 1.0 ± 0.3 (0.5-1.7) |
| MAP (mmHg) | 62.3 ± 18.6 (27-104) |
| GGT (U/l) | 20.9 ± 9.2 (9-47) |
| Calorimetry (kcal/24 h) | 1423.3 ± 174.7 (1100-1970) |
| Insulin (µI/ml) | 22.6 ± 7.3 (0.6-39.8) |
| HOMA-IR | 5.3 ± 1.8 (0.1-9.6) |
| OGTT insulin AUC | 315.4 ± 103.0 (111.6-521) |
| OGTT glucose AUC | 450.9 ± 70.2 (333-617) |
- ALAT, alanine aminotransferase; ASAT, aspartate aminotransferase; BMI, body mass index; GGT, gamma-glutamyl transpeptidase; HDL-C, high-density lipoprotein cholesterol; HOMA-IR, homeostasis model assessment of insulin resistance; LDL-C, low-density lipoprotein cholesterol; MAP, mean arterial blood pressure; OGTT, oral glucose tolerance test; TG, triglycerides.
Shotgun lipidomics analysis
The methodology is fully described in the Supporting Information. Briefly, a high-throughput, fully automated liquid/liquid extraction method utilizing a Hamilton Microlabstar robot (Hamilton, Bonaduz, Switzerland) was developed in house for lipidomics extraction with minor modifications from previous methods (23). A shotgun lipidomics approach was developed in house based on previously published works (24, 25). Analysis was carried out on an LTQ Orbitrap Velos MS (Thermo Fisher Scientific, Reinach, Switzerland) system coupled to a Nanomate nanoinfusion ion source (Advion Bioscience, Harlow, Essex, UK). The entire set of analyzed lipid species is summarized in (see Supporting Information, Tables S3 and S4). Individual lipid species were identified by the LipidXplorer software tool following the published protocol (25). Lipid species were annotated as recently described in a shorthand notation (26). Lipid class abbreviation was based on: number of carbon atoms in the radylside chains: number of double bonds in the radyl side chains. Total lipid profile for each lipid class was obtained by summing individual species, while total saturated fatty acid, monounsaturated fatty acid, and polyunsaturated fatty acid species within each class of lipids were obtained by summing individual species containing 0, 1, or ≥2 fatty acid unsaturations.
Chemometrics
Chemometric analysis was performed using the software package SIMCA-P+ (version 13.0, Umetrics AB, Umeå, Sweden), own and thirdly developed routines programmed in R. Multivariate regression models were used to identify influential lipidomics variables. A modification of partial least squares regression (PLSR) that removes all information orthogonal to the response variable during the fitting process was employed. This variant, orthogonal projection to latent structures (OPLS) (27), provides sparser models (better interpretable) with the same degree of fit as PLSR (Table 3). To highlight the weight of individual variables in the model, variable importance in projection (VIP) was used, with a value above 1 used as a threshold by convention. Confirmatory nonparametric modeling was performed by random forests (RF), using the R implementation in the package “randomForest” (28). The method builds an ensemble of classification and regression trees (CARTs) and has a similar predictive performance to some of the best “state of the art” algorithms. Its embedded dual randomization scheme enables RF to show a very low bias predictor, both in classification and regression. Monitoring of the fitting process is performed by tracking the evolution of the “out of bag” (OOB) prediction error. The mean decreases in accuracy/node impurity after variable permutation show the influence of the variables in the model.
| OPLS models | Visceral fat | HOMA-IR | EAT |
|---|---|---|---|
| Lipid composition | R2X = 0.33, R2Y = 0.90, Q2Y = 0.64 | R2X = 0.33, R2Y = 0.89, Q2Y = 0.43 | R2X = 0.29, R2Y = 0.86, Q2Y = 0.58 |
| Lipid class proportion | R2X = 0.37, R2Y = 0.92, Q2Y = 0.67 | R2X = 0.35, R2Y = 0.92, Q2Y = 0.57 | R2X = 0.36, R2Y = 0.85, Q2Y = 0.59 |
| Lipid species | R2X = 0.16, R2Y = 0.87, Q2Y = 0.59 | R2X = 0.32, R2Y = 0.55, Q2Y = 0.21 | R2X = 0.32, R2Y = 0.81, Q2Y = 0.59 |
- The degree of fit of these models is measured using R2X (lipidomics data) and R2Y (clinical readout) parameters that account for the total variance captured by the model in the selected conditions. Moreover, Q2Y parameter reflects the total variance explained by the model on test samples through internal validation cycles.
Results
Relationships between measures of body composition, visceral fat, epicardial fat, and clinical parameters
We previously reported relationships between different visceral fat estimates and other body composition and clinical parameters in 40 subjects that were also assessed in the present complementary study (21). We report here in Tables 1 and 2 body composition, anthropometric, and cardiac fat depot parameters. The analysis of the covariance between body composition and clinical data was performed by OPLS modeling using 2 predictive components and zero orthogonal for each array. The models described 62% of total variance in CT and iDXA, and 20% of total variance in clinical data. The analysis of the projection of model parameters (correlation-scaled loadings) for CT and iDXA data showed a close relationship between epicardial, mediastinal, and intrapericardial fats (Figure 1). Observed values for adjusted R2 of a linear fit between indicators were: 0.85 for epicardial/mediastinal fats, 0.57 for epicardial/intrapericardial fats, and 0.90 for intrapericardial/mediastinal fats. Epicardial, mediastinal, and intrapericardial fat depots show positive associations with abdominal volume, abdominal fat volume, subcutaneous fat volume, total fat mass, trunk fat mass, and android fat mass. Table S1 summarizes adjusted R2 and associated P-values. The cardiac fat depots also showed specific relationships with region-specific body composition. While intrapericardial fat has a relevant correlation with android lean mass (adjusted R2: 0.30, P-value: 0.0014), mediastinal fat shows correlation with arms fat mass (adjusted R2: 0.26, P-value: 0.0034). Epicardial fat pattern presents features shared by both in mediastinal and intrapericardial fat.
Correlation-scaled OPLS loadings for CT and iDXA array. Correlation structure around target fat deposition indicators. Results corresponding to: (A) mediastinal fat, (B) epicardial fat, and (C) intrapericardial fat. Proximity area between parameters (blue circle) defined arbitrarily at a radius of 0.4 units.
Further analysis of cardiac fat and clinical data showed close relationships between mediastinal fat and ALAT/ASAT ratio and some anthropometric parameters such waist, BMI, hip, and waist/hip ratio (Figure 2a). The strongest positive correlation with hip showed an adjusted R2 of 0.3349 (P-value: 0.0014). Other species linked by the model structure to this feature were ALAT/ASAT ratio (adjusted R2: 0.31, P-value: 0.0013), BMI (adjusted R2: 0.30, P-value: 0.0016) and waist (adjusted R2: 0.33, P-value: 0.0026). Epicardial fat showed a positive correlation with hip (adjusted R2: 0.36, P-value: 9.827e-4) (Figure 2b). Similarly to mediastinal fat, variables such BMI (adjusted R2: 0.34, P-value: 6.563e-4), waist (adjusted R2: 0.33, P-value: 0.0023), and ASAT/ALAT ratio (adjusted R2: 0.22, P-value: 0.0073) were also highlighted.
Overlay of correlation-scaled OPLS loadings for both arrays (X, iDXA, and Y, clinical data). Only target parameters on CT and iDXA measures are shown to improve data visualization. Results corresponding to: (A) mediastinal fat, (B) epicardial fat, and (C) intrapericardial fat. Proximity area between parameters (blue circle) defined arbitrarily at a radius of 0.4 units.
Intrapericardial fat showed a slightly more complex profile of associations (Figure 2c). The strongest association was with ALAT/ASAT ratio, showing an adjusted R2 of 0.31 (P-value: 0.0013). Variables highlighted by the 2 previous body composition indicators were also present: BMI (adjusted R2: 0.20, P-value: 0.010), hip (adjusted R2: 0.24, P-value: 0.007), and waist (adjusted R2: 0.26, P-value: 0.0075). However, additional relationships were highlighted, including TG (adjusted R2: 0.17, P-value: 0.016), insulin (adjusted R2: 0.14, P-value: 0.035), HOMA-IR (adjusted R2: 0.13, P-value: 0.039), MAP (adjusted R2: 0.11, P-value: 0.047), and calorimetry (adjusted R2: 0.17, P-value: 0.016).
Epicardial and visceral adiposity and HOMA-IR link to specific blood plasma lipid composition remodeling
The blood plasma lipidome was characterized using an in-house developed shotgun lipidomics approach based on previously published work (24), which enabled us to quantify 252 individual lipid species over 14 different classes. The analysis encompasses triacylglycerol TG (n = 50), diacylglycerol (DAG) (n = 18), sphingomyelin SM (n = 24), ceramide Cer (n = 17), phosphatidylinositol PI (n = 12), lysophosphatidylinositol LPI (n = 2), phosphatidyglycerol PG (n = 6), phosphatidylethanolamine PE (n = 23), ether phosphatidylethanolamine PE-O (n = 15), lysophosphatidylethanolamine LPE (n = 4), phosphatidylcholine PC (n = 45), ether phosphatidylcholine PC-O (n = 29), lysophosphatidylcholine LPC (n = 6), and phosphatidic acid PA (n = 1).
On the basis of this compositional analysis of blood plasma lipids, additional parameters were generated for further analyses, namely (i) total lipid profile: relative percentage of each lipid class in total blood; (ii) lipid composition: concentration of total, saturated fatty acids (SFAs), monounsaturated fatty acids (MUFAs), and polyunsaturated fatty acids (PUFAs) species within each class of lipids; and (iii) lipid species level. Multivariate analysis was conducted to assess the relationships between plasma lipid data and overall visceral adiposity, epicardial fat, and HOMA-IR (Tables 3 and 4).
| VIP values for influential metabolites | OPLS-model on visceral fat | OPLS-model on HOMA-IR | OPLS-model on EAT | |
|---|---|---|---|---|
| Ceramide | Cer 41:2 | 1.77 | 0.37 | 1.76 |
| DAG | DAG 30:1 | 0.30 | 0.29 | 1.53 |
| DAG 36:3 | 1.62 | 1.50 | 0.98 | |
| DAG 36:4 | 1.88 | 0.14 | 1.60 | |
| DAG 42:8 | 1.88 | 0.97 | 1.10 | |
| LPC | LPC 20:3 | 1.95 | 0.67 | 0.42 |
| LPE | LPE 16:0 | 0.99 | 1.53 | 0.58 |
| LPI | LPI 16:0 | 0.36 | 1.68 | 0.96 |
| PC | PC 14:0/20:3 | 1.98 | 1.37 | 1.82 |
| PC 14:0/22:4 | 2.83 | 1.98 | 1.05 | |
| PC 16:0/20:2 | 2.01 | 1.05 | 2.00 | |
| PC 16:1/16:1 | 2.14 | 0.43 | 1.70 | |
| PC 18:0/22:5 | 0.52 | 0.69 | 1.41 | |
| PC-O | PC-O 42:5 | 1.73 | 0.98 | 0.50 |
| PC-O 28:4 | 1.04 | 0.69 | 1.25 | |
| PC-O 30:2 | 0.92 | 0.56 | 0.23 | |
| PE | PE 16:0/22:4 | 1.76 | 1.89 | 1.74 |
| PE 18:1/22:6 | 1.43 | 1.69 | 1.09 | |
| PE 18:1/22:5 | 2.23 | 0.47 | 0.09 | |
| PE-O | PE-O 32:5 | 1.51 | 0.95 | 1.58 |
| PE-O 36:5 | 1.99 | 1.54 | 1.12 | |
| PG | PG 20:3/20:3 | 1.21 | 1.19 | 2.05 |
| PG 22:5/18:1 | 1.21 | 1.19 | 2.05 | |
| PI | PI 18:0/16:1 | 2.60 | 1.27 | 1.81 |
| SM | SM 40:3 | 1.94 | 1.74 | 1.43 |
| SM 43:2 | 1.88 | 1.88 | 1.46 | |
| TAG | TAG 48:4 | 1.63 | 0.91 | 0.90 |
| TAG 49:2 | 1.56 | 0.89 | 0.78 | |
| TAG 50:4 | 2.00 | 0.89 | 1.76 | |
| TAG 51:2 | 1.67 | 0.77 | 1.75 | |
| TAG 51:3 | 1.80 | 0.82 | 1.81 | |
| TAG 52:1 | 0.66 | 0.34 | 1.53 | |
| TAG 53:3 | 1.37 | 1.80 | 0.14 | |
| TAG 54:1 | 1.68 | 0.97 | 0.56 | |
| TAG 54:5 | 0.57 | 0.71 | 1.58 | |
| TAG 56:5 | 2.41 | 1.65 | 1.88 | |
| TAG 56:6 | 2.48 | 1.59 | 1.87 | |
| TAG 56:7 | 2.42 | 1.05 | 1.72 | |
| TAG 56:8 | 1.80 | 0.57 | 1.58 | |
- Variable importance in projection (VIP) above 1.5 are shown in bold to highlight the weight of individual variables in the models.
VAT was associated with an increase in DAG species (relative concentration, VIP =1.36 and lipid percentage, VIP=1.19), mainly composed of SFAs (VIP = 1.37); and a decrease in PI species mainly composed of MUFAs (VIP = 1.44). The metabolic signature of VAT was mainly dominated by changes in phosphocholines, phosphoethanolamines, triacylglycerols, and DAG (VIP > 1.5). VAT was associated with increased concentrations in PC 14:0/20:3, PC 14:0/22:4, PC-O 42:5, LPC 20:3, PE 16:0/22:4, PE-O 32:5, PE-O 36:5, SM 40:3, SM 43:2, TAG 48:4, DAG 42:8. VAT was associated with decreased concentrations in PC 16:1/16:1, PE 18:1/22:5, PI 18:0/16:1, Cer 41:2, TAG 49:2, TAG 50:4, TAG 51:2, TAG 51:3, TAG 54:1, TAG 56:5, TAG 56:6, TAG 56:7, TAG 56:8, DAG 36:3, and DAG 36:4.
HOMA-IR was associated with greater LPE species (relative concentration, VIP =1.13), mainly related to an increased concentration of LPE containing SFAs concomitant to a decrease in LPE containing MUFAs and PUFAs. HOMA-IR was also correlated with lower levels of ceramides (VIP = 1.21), mainly containing MUFAs (VIP = 1.21). The metabolic signature for HOMA-IR was composed of main changes (VIP > 1.5) in specific phospholipids, including increased concentrations in PC 14:0/22:4, PE 16:0/22:4, LPE 16:0, PE-O 36:5, SM 40:3, SM 43:2, and decreased concentrations in PE 18:1/22:6, LPI 16:0, TAG 53:3, TAG 56:5, TAG 56:6, and DAG 36:3.
Epicardial fat was associated with a lower proportion of LPC (relative concentration, VIP = 1.19). EAT was also associated with higher concentration of TAG species (VIP = 1.14), mainly composed of SFAs (VIP = 1.43), and lower concentrations of TAG and DAG containing PUFAs (VIP < 1). The metabolic signature of EAT was mainly dominated by the following lipid species (VIP > 1.5). An increase in PG 20:3/20:3, PG 22:5/18:1, PC 16:0/20:2, PC 14:0/20:3, PE 16:0/22:4, PC 18:0/22:5, PE-O 32:5, TAG 52:1, DAG 42:8, and a decrease in PC 16:1/16:1, PI 18:0/16:1, Cer 41:2, TAG 50:4, TAG 51:2, TAG 51:3, TAG 54:5, TAG 56:5, TAG 56:6, TAG 56:7, TAG 56:8, DAG 30:1, and DAG 36:4. In particular, PG 20:3/20:3 and PG 22:5/18:1 are isobaric lipid species, which suggest a change in fatty acid composition of the different PG 40:6 lipid species.
RF were used for confirmatory analyses on highlighted variables (Figure 3, Table S2). Epicardic adiposity (EAT) was associated to DAG 42:8, PE-O 32:5, and PC 16:1/16:1. The two first metabolites were positively and exclusively correlated to EAT, the last one being negatively correlated and common with VAT. After outlier removal, confirmatory test by individual linear fitting confirmed the relationship for DAG 42:8 (adj.R2:0.36; P-value: 6e-4) and PC 16:1/16:1 adj.R2: 0.35, P-value: 0.0011). RF modeling for HOMA-IR was weak and associated 11 metabolites. Univariate confirmatory tests filtered LPC 22:6 (adj.R2:0.30; P-value: 0.011), PE 16:0/22:4 (adj.R2:0.14; P-value: 0.065) and SM 43:2 (adj.R2:0.37; P-value: 0.017) with positive correlations. The RF model for VAT showed the best predictive performance (coefficient of determination in OOB samples of 0.72) with 5 variables. Univariate confirmatory tests filtered: PC 14:0/22:4 (adj.R2:0.38; P-value: 1.371e-5), PC 16:1/16:1 (adj.R2:0.21; P-value: 0.0020), PE 18:1/22:5 (adj.R2:0.23; P-value: 0.0012), and PI 18:0/16:1 (adj.R2:0.32; P-value: 9.275e-5). Correlation with PC 14:0/22:4 was positive, and negative with PC 16:1/16:1, PE 18:1/22:5, and PI 18:0/16:1.
Summary of relevant variables for predicting epicardial fat, insulin resistance, and visceral adiposity. The table lists metabolite importance and robustness in predicting epicardial fat, visceral fat, and HOMA-IR as assessed by random forest analysis. Several generations of models (104 in total) calculated the overall variable weights by averaging their mean decrease of mean standard error (MSE) in OOB samples. These weights were used as general indicators of the variable importance, and final variable selection was performed according to these values. Threshold for variable selection was set by progressive elimination of variables until the validation error of the OOB samples reached a minimum. The value of this threshold was calculated individually for each model/parameter set (EAT: 1.7, HOMA-IR: 0.62, VA: 4.4). The figure highlights similitudes and differences between the influential variables for each model.
Discussion
In the present study, we have expanded our previous investigation on metabolic signatures of VA by integrating topographical cardiac fat deposition with systemic lipidomic information. The study presents several potential applications, including the use of CT and DXA measures to provide robust biological readouts to compare the association between different estimates of cardiac steatosis, VA, and region-specific body composition in a single ethnic group under well-defined conditions and healthy medical criteria. We observed a positive correlation between overall VA estimates and the different cardiac fat depots, which were also dependent on BMI, abdominal, and subcutaneous fat depots. All cardiac fat depots were strongly related to indicators of hepatic steatosis (ALAT and ASAT), which corroborates the strong physiological connections between hepatic and cardiac steatosis. In adults, a close relationship was previously found between visceral fat, EAT, metabolic syndrome, insulin resistance, and NAFLD markers (11, 29, 30), as well as accumulated fat in the epicardial area and abnormal energy metabolism in the left ventricle (31). Moreover, in nondiabetic obese men, all components of cardiac steatosis have been correlated with subcutaneous fat tissue, VA, and hepatic TG content (2). In this population, myocardial TG content, epicardial fat, pericardial fat, VA, and hepatic TG content correlated with BMI, HDL, blood TG, and HOMA-IR. In our cohort of healthy obese women, only intrapericardial fat depots showed some associative trends with glucose and insulin metabolism at fasting state and plasma TG. It had previously been reported that glycaemia, insulin, gender adiponectin, and cardiac workload were associated with heart adiposity (7, 32). Interestingly, once glucose tolerance becomes impaired, the evolution of cardiac steatosis becomes more pronounced in women (32). Since our study population suffered neither from glucose intolerance nor diabetes, this may explain the absence of relationships in the present study with epicardial fat depots. Moreover, visceral adiposity, insulin, and noninsulin-mediated glucose uptake in females are also dependent on endogenous androgens (33), which was harmonized in our cohort, and future studies should explore the inference with cardiac metabolic specificities.
Modeling of relationships between visceral fat depots and the blood plasma lipidome highlighted a strong association between VA and DAG (mainly composed of SFAs), and between EAT and TAG (mainly composed of SFAs). The consistent and significant increase in SFAs highlighted a well-known feature of metabolic syndrome and CVD, marked by an increased concentration of SFAs in serum (34). Stimulated production of SFAs was suggested to originate from de novo lipogenesis and to be 5 times higher in subjects phenotyped with IR and fatty liver compared to healthy subjects (35). In the present cohort, hepatic markers of steatosis correlate strongly with VA and, especially, EAT. Despite the fact that fatty liver was not a condition included in the phenotyping of the population by imaging, our observations may describe a functional relationship between circulating SFAs in TAG and DAG on the one hand and regio-specific VA and EAT on the other hand in healthy obese women.
Since VA also correlated with glucose and insulin parameters in the present cohort, insulin related features may have a different influence on the plasma lipidome. In particular, evidence supports the role of accumulated hepatic DAG in the development of hepatic IR, mediated via the activation of PKCε (36). Analysis of DAG species revealed an increase in saturated and monounsaturated species in steatotic versus normal human specimens (37), in particular with several-fold increases in relatively short chain (30-36 carbon atoms) species containing 0, 1, 2, and 3 double bonds (37). Interestingly, at molecular level, the blood plasma signature of EAT and VAT reflected a depletion in relatively short chain (30-36 carbon atoms) species containing PUFAs, which more likely results from altered hepatic metabolism. Furthermore, elevated DAG in the liver may contribute to alter TAG, PC, and PE levels, a feature specific to NAFLD (37). In the present study, we observed a strong correlation between SFAs in plasma TAGs with EAT, while reduced plasma concentrations of TAG containing long chain PUFAs were consistently associated with VA, EAT, and HOMA-IR. These data suggest an effect that may be driven by the loss of glucose and insulin homeostatic equilibrium, as reported recently (38). In particular, the lipidomic analysis of blood plasma from the Framingham Heart Study showed that lipids, mainly TAGs, of lower carbon number (e.g., <52) and double bond content (e.g., <2) were associated with increased risk of diabetes, whereas lipids of higher carbon number (e.g., ≥56) and double bond content (e.g., >2) were associated with decreased risk (38).
Furthermore, we found EAT to be also specifically correlated with elevated levels of specific phosphatidylglycerol species, namely PG 20:3/20:3 and PG 22:5/18:1, which may relate to specific PG and cardiolipin metabolism. Cardiolipin has recently been found to be deficient in the heart at the earliest stages of diabetes, possibly due to a lipid-digesting enzyme that becomes more active in diabetic heart muscle (39). An earlier lipidomic analysis highlighted how cardiolipin and its direct metabolic precursors, phosphatidylglycerol, were depleted in target diabetic myocardium (40). More importantly, the lipidomics approach showed that the earliest stages of diabetes were accompanied by a profound remodeling of cardiolipins (40). The changes in PG species could also be related to changes in hepatic and heart steatosis, since PG species were shown to increase with fatty liver disease (37). EAT was also associated with an overall lower concentration of PE-O lipid species. Ether lipids represent 18% of the total pool of phospholipids and are mainly composed by plasmalogens (39). Although little is known about their systemic metabolism, plasmalogens have been implicated in the protection of cellular functions against oxidative damage, and their diminished levels have been reported in several diseases (39), e.g., diabetes mellitus, vascular diseases, and obesity (19). Therefore, their depletion and their association with EAT may suggest a metabolic adaption to cope with oxidative stress related to altered TAG and DAG metabolism.
In conclusion, the present investigation delivered a systemic lipidomics signature associated with regio-specific mediastinal and visceral adiposity under healthy clinical conditions. We have extended here our previously published work, by reporting a metabolic signature in blood plasma of healthy obese women related to EAT, as per a specific pattern of DAG and phosphoglycerols.
Acknowledgments
We thank Dr. L. Favre at CHUV, Dr. M. Beaumont, S. Oguey-Araymon, A. Blondel Lubrano, and C. Nielsen Moennoz at Nestlé Clinical Development Unit, and Dr. O. Cominetti at NIHS for study design, sample provision, and scientific discussion. We thank Dr. L. Fay at Nestlé Research Center for managerial support. We would like to acknowledge the collaboration with GE GRC and in particular Dr. D. Ergun, Dr. M. Rothney, and Dr. F. Ginty for DXA data generation, analysis, and interpretation.
