Lipidomics Reveals Associations of Phospholipids With Obesity and Insulin Resistance in Young Adults

Abstract

Context:

Obesity and related diseases have become a global public health burden. Identifying biomarkers will lead to a better understanding of the underlying mechanisms associated with obesity and the pathways leading to insulin resistance (IR) and diabetes.

Objective:

This study aimed to identify the lipidomic biomarkers associated with obesity and IR using plasma samples from a population-based cohort of young adults.

Design and Setting:

The Western Australian Pregnancy Cohort (Raine) study enrolled 2900 pregnant women from 1989 to 1991. The 20-year follow-up was conducted between March 2010 and April 2012.

Participants and Samples:

Plasma samples from 1176 subjects aged 20 years were analyzed using mass spectrometry-based metabolomics.

Main Outcome Measures:

Associations of analytes with markers of obesity and IR including body mass index, waist circumference, homeostasis model assessment (HOMA-IR), and insulin were examined. Analyses were stratified by body mass index and adjusted for lifestyle and other factors.

Results:

Waist circumference was positively associated with seven sphingomyelins and five diacylphosphatidylcholines and negatively associated with two lysophosphatidylcholines. HOMA-IR was negatively associated with two diacylphosphatidylcholines and positively with one lysophosphatidylcholine and one diacylphosphatidylcholine. No significant association was found in the obese/overweight group of the HOMA-IR model. In the normal-weight group, one lysophosphatidylcholine was increased.

Conclusion:

A possible discriminative effect of sphingomyelins, particularly those with two double bonds, and lysophosphatidylcholines was identified between subjects with normal weight and obesity independent of low-density lipoprotein cholesterol and high-density lipoprotein cholesterol concentrations. Our results suggest weight status-dependent mechanisms for the development of IR with lysophosphatidylcholine C14:0 as a key metabolite in nonobese IR.

Obesity has become a global public health burden (1). As a worldwide growing epidemic, obesity raises the need for better tools to monitor this disease adequately and enable better strategies for early prevention and detection of underlying metabolomic changes, especially because obesity is associated with insulin resistance (IR) and type 2 diabetes mellitus (T2DM) (2).

Inflammation of adipose tissue is a key factor in the pathogenesis of IR, but the underlying mechanisms for this inflammation and why not all people with obesity develop IR remain unknown (3). Evidence suggests that dyslipidemia plays a central role in the pathogenesis of IR and T2DM (4), but the underlying metabolic processes are complex and interconnected.

Studies showing correlations between levels of lysophosphatidylcholines (LPC), sphingomyelins (SM), nonesterified fatty acids (NEFAs), and phosphatidylcholines (PC) in the blood plasma of patients with obesity are in accordance with the hypothesis of lipotoxicity. This hypothesis states that an increased supply of fat through nutrition leads to an excess of lipid storage in tissues other than adipocytes, such as hepatocytes, leading to IR (5). Therefore, it may be of particular interest to identify differences between lipid profiles of obese and normal-weight individuals and determine whether these profiles associate with the risk of developing IR or T2DM (6, 7).

Mass spectrometry coupled with liquid or gas chromatography is a powerful tool for the identification of possible lipid biomarkers (8). The aim of this study was to identify lipidomic markers associated with obesity, IR, or diabetes in a population cohort of more than 1000 young adults at 20 years of age. These young adults are of special interest because they are the optimal subjects to study IR and obesity due to an early onset of those diseases, combined with the possibility for lifestyle intervention in this early stage. We specifically focused on the effects of polar lipid species PC, LPC, SM, and NEFAs, in relation to obesity, IR, and clinical lipid measurements, independent of low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein cholesterol (HDL-C) concentrations.

Materials and Methods

The Western Australian Pregnancy Cohort (Raine) Study is a prospective longitudinal cohort study that enrolled 2900 pregnant women from 1989 to 1991 with the purpose to examine the effects of ultrasound imaging on the fetus (9). The 2868 live births were evaluated and followed up serially to 20 years of age. The 20-year follow-up, which occurred between March 2010 and April 2012, included 87% of the active participants. Ethics approval at the 20-year follow-up was obtained from the University of Western Australia Human Research Ethics Committee. Informed and written consent was obtained from the participants.

During the 20-year follow-up, calibrated measurements of height, weight, body mass index (BMI), waist, hip, and arm circumference were measured by trained research assistants. Height and weight were measured by electronic chair scales and stadiometer. Waist and hip measurements were obtained using steel tape measures and skinfold thickness by calipers (Holtain). A phlebotomist visited the home of each participant early in the morning, and venous blood samples were taken from an antecubital vein after an overnight fast. Samples were stored at −80°C until analyzed. Serum insulin, glucose, lipids, and liver function tests were analyzed using standardized protocols in the PathWest Laboratory at Royal Perth Hospital. HDL-C was determined on a heparin manganese supernatant, and LDL-C was calculated by using the Friedewald formula (10).

When the BMI status is reported, the underlying borderlines were those of the World Health Organization, in which a value greater than 29.99 kg/m² means obesity, 25.0–29.99 kg/m² overweight, 18.5–24.99 kg/m² normal weight, and less than 18.5 kg/m² underweight (11). We used homeostasis model assessment index of insulin resistance (HOMA-IR) values as a steady variable to measure IR. In general, higher values are associated with IR. HOMA-IR is defined by Insulin (microinternational units per milliliter) × glucose (millimoles per liter)/22.5 (12).

Potential confounding variables

Analyses were adjusted for variables from the 20-year questionnaire, which may confound the relationship between metabolite concentrations and waist circumference (WC), BMI, HOMA-IR, and insulin values. Ethnicity was classified as Caucasian if both parents were Caucasian, or as non-Caucasian if one or both parents were of another ethnicity because numbers of other ethnic groups besides Caucasian were too low for separate statistical analysis (Supplemental Table 1).

Smoking behavior was measured as currently smoking cigarettes (yes or no).

Alcohol consumption was assessed by asking for the frequency of drinking any alcohol. We calculated a dichotomous variable for ever vs never consumption of alcohol in the last month.

To adjust for dietary intake, dietary patterns were calculated as previously described (13).

Dietary misreporting was calculated by using the Goldberg equation (14), and the final variable was categorical with three values: underreporting, overreporting, and plausible reporting.

Physical activity was assessed as more than 10 minutes of moderate or vigorous physical activity in the last 7 days. We created a categorical variable with less than one time, one to three times, and four or more times as categories according to previous reports on the Raine study (15).

The variable for sedentary behavior was created based on the hours spent in front of a screen (watching TV, playing videogames, and socializing and nonsocializing activities on the Internet). The resulting variable includes the amount of hours spent in front of any screen per day from 0 to 4.

A three-level sex variable (females using hormonal contraceptives, females not using hormonal contraceptives, and males) was used to evaluate the effects of sex and hormonal contraceptive use as previously described (16).

Plasma analysis

The plasma EDTA samples of 1176 subjects were labeled and packed in 100-μL tubes at the Royal Perth Hospital Research Unit in Perth, Western Australia. Samples were transported on dry ice to the Division of Metabolic and Nutritional Medicine of the Dr von Hauner Children's Hospital in Munich, Germany, and stored at −80°C until being analyzed.

Polar lipids

Flow-injection mass spectrometry was used to analyze polar lipids. Plasma (10 μL) or standard solution (10 μL) were diluted with 500 μL methanol, containing internal standards for different lipid groups and ammonium-acetate. After centrifugation, a 96-deep well plate was prefilled with 700 μL methanol, and 200 μL of the centrifuged supernatant was carefully pipetted with a multipipette into the plate and then used for flow-injection mass spectrometry analysis. Samples were analyzed with a triple-quadrupole mass spectrometer (QTRAP4000; Sciex) with an electrospray ionization source, which was used in both positive and negative modes. The mass spectrometry was coupled to a liquid chromatography system (Agilent). Tandem mass spectrometry analysis was performed in Multiple Reaction Monitoring mode. Analyst 1.5.1 software, followed by in-house handling with the statistical program R (R Project for Statistical Computing, http://www.r-project.org/) was used to postprocess the data.

The analysis comprised acylcarnitines, diacylphosphatidylcholines (PCaa), acylalkylphosphatidylcholines, SM, lysophophatidylcholine, lysophosphatidylethanolamine, and the sum of hexoses (80% glucose). The analytical technique is not capable of determining the position of the double bonds and the distribution of carbon atoms between fatty acid side chains. The polar lipids are described using the nomenclature CX:Y, where X is the length of the carbon chain, Y is the number of double bonds, OH in the formula means the molecule contains a hydroxyl-group. The letter a means that the acyl chain is bound via an ester bond to the backbone, and e means an ether bond.

Nonesterified fatty acids

Analysis for NEFAs was performed as previously reported (17). Briefly, 20 μL plasma was mixed with 200 μL isopropanol (containing 2 mg per 100 mL ¹³C-labeled palmitic acid) in a 96-deep well plate. After centrifugation the supernatant was transferred into a 96-well plate for liquid chromatography and tandem mass spectrometry analysis. An ultraperformance liquid chromatography diphenyl column (Pursuit UPS Diphenyl, 1.9 μm, 100 mm, 3.0 mm; Varian) was used for chromatographic separation at 40°C with an Agilent 1200 SL series HPLC system. The injection volume was set to 10 μL with an eluent flow rate of 700 mL/min. A hybrid triple-quadrupole MS (4000 QTRAP; AB Sciex) operating in negative electrospray ionization mode was coupled to the HPLC system for the identification of NEFAs. Fatty acids are separated according to chain length and the number of double bonds but not according to the position of double bonds. NEFAs are mentioned using the nomenclature described above as CX:Y. Metabolite concentrations are reported in micromoles per liter of plasma.

Statistics

Statistical analyses were performed using the software R (3.1.3). The quality control criterion was defined as inter- and intrabatch coefficient of variance of 30%.

For outlier analysis, boxplots with whiskers of 1.5-fold the interquartile range were created and exploratively analyzed. Outliers of metabolites were defined as measured values outside the whiskers. Samples were excluded if more than 30% of the measured analytes were defined as outliers. To test for associations of analytes of the 20-year follow-up with indicators of obesity and IR, we regressed BMI, WC, HOMA-IR, and insulin on each of the analytes separately. We checked the assumptions of the linear models, among them the normality of the residuals, with diagnostic plots, and performed sensitivity analyses using log-transformed outcomes. The only prominent outcome was homeostasis model assessment, which we log transformed in all subsequent analyses.

The first step was to test for possible confounding by HDL-C, LDL-C, smoking, alcohol consumption, dietary patterns, physical and sedentary behavior, and biological sex, which were found to be associated with body composition and metabolite concentration in previous studies (18–20) (Table 1). This was done by a multivariate model with a potential confounder and BMI, WC, HOMA-IR, and insulin values as the outcome. An analysis of association between metabolites and confounder was done accordingly (Supplemental Table 2).

Table 1.

Associations Between BMI, WC, Insulin, and HOMA-IR and Predefined Confounder Variables

Variables	BMI			HOMA-IR			WC			Insulin
	P Value	β	CI	P Value	β	CI	P Value	β	CI	P Value	β	CI
HDL-C	<.001	−5.178	−7.213, −3.142	<.001	−.35	−0.657 to 0.043	<.001	−12.58	−17.73, −7.429	.001	−1.475	−2.823, −0.127
LDL-C	<.001	1.37	0.512, 2.227	.326	.074	−0.055, 0.204	<.001	3.239	1.068, 5.41	.294	.333	−0.235, 0.901
Sex	<.001	−1.588	−2.928, −0.248	.001	−.222	−0.424, −0.02	1.00	1.452	−1.938, 4.842	<.001	−1.091	−1.978, −0.203
Ethnicity	1.00	.562	−0.98, 2.104	1.00	−.006	−0.238, 0.227	.212	2.419	−1.467, 6.304	1.00	−.051	−1.072, 0.97
Western dietary pattern	<.001	1.138	0.16, 2.116	.005	.14	−0.007, 0.287	<.001	3.135	0.663, 5.606	.008	.596	−0.052, 1.244
Healthy dietary pattern	.006	.636	−0.04, 1.312	1.00	−.039	−0.141, 0.063	.272	1.016	−0.69, 2.722	1.00	−.159	−0.607, 0.289
Dietary misreporting	<.001	−2.757	−4.18, −1.333	<.001	−.311	−0.525, −0.096	<.001	−6.601	−10.197, −3.004	<.001	−1.39	−2.333, −0.448
Physical activity	.001	.869	0.051, 1.686	1.00	−.01	−0.133, 0.113	.043	1.602	−0.464, 3.668	1.00	−.041	−0.583, 0.5
Sedentary behavior	1.00	.259	−0.434, 0.951	1.00	.037	−0.067, 0.142	.219	1.081	−0.664, 2.826	1.00	.183	−0.276, 0.641
Smoking	.398	.899	−0.728, 2.526	.689	.119	−0.126, 0.364	.885	1.862	−2.238, 5.961	.96	.478	−0.6, 1.555
Alcohol consumption	1.00	−0.358	−2.292, 1.576	1.00	−.006	−0.297, 0.286	1.00	−.831	−5.704, 4.043	1.00	−.065	−1.345, 1.216

Standardized β, Bonferroni-corrected confidence intervals, and P values of multiple linear regression models are from the 20-year follow-up of the Raine Study.

At first, sex and metabolite concentration interaction was added to the models because other studies found differences in metabolite concentrations between males and females (21, 22). We also tested for interactions between metabolites and physical activity, HOMA-IR (model with WC as outcome), and BMI (model with HOMA-IR as outcome). Subsequent results excluded the interaction terms and adjusted for sex, physical activity, BMI (model with HOMA-IR as outcome), and HOMA-IR (model with WC as outcome) because the interactions were not significant.

The standardized concentrations were applied to the final adjusted regression model to make the different metabolite concentrations comparable. Standardization of the coefficients was performed by standardizing the metabolite concentrations (xi) to their mean and standard deviation (sdi), using the following equation: |$\frac{xi-mean}{sdi}$|⁠.

We focused on models including WC for the association of metabolites with obesity because WC is more closely related to obesity-related health risks than BMI (23). Using BMI instead yielded similar results (Supplemental Figure 1). Additionally, in the final models, the effects of LDL-C and HDL-C were excluded by adjustment because this led to almost the same results as the more complex method of residualizing (Supplemental Figure 2). Adding triglyceride concentrations to the model did not change the tendencies in metabolite concentrations (Supplemental Figure 3).

Bonferroni correction was used to correct for multiple hypothesis testing. The level of significance was divided by the number of analytes, and P values were considered statistically significant if they were less than the adjusted significance level of α = .000286 (⁠|$\frac{0.05}{175\left( number\,of\,analytes \right)}$|⁠). Bonferroni-adjusted confidence intervals (CI) were calculated accordingly by changing the default confidence level to 1 ± |$\frac{\alpha }{175\left( number\,of\,analytes \right)}$|⁠.

Overviews and trends of the results for all metabolites were depicted in a figure by plotting the −log₁₀ (P) values on the y-axis and the metabolites on the x-axis, further referred to as a Manhattan plot. The direction below or above 0 was used to make statements on whether the concentration of the metabolites was higher or lower in one of the groups, depending on the outcome.

Results

One hundred seventy-five different lipid metabolites (32 NEFAs, 14 LPCa, three lysophosphatidylethanolamines, 43 PCaas, 37 acylalkylphosphatidylcholines, 47 SMs) of 1011 subjects were included in the analysis. The characteristics of the study population are provided in Table 2. Standardized estimates, Bonferroni CIs, and Bonferroni P values can be seen in Supplemental Tables 3 and 4.

Table 2.

Characteristics of the Study Subjects at 20 Years Divided by Sex

	Female	Male
n, %	464 (45.85)	547 (54.05)
Ethnicity, n/%
Caucasian	390 (84.05)	449 (84.05)
Not Caucasian	66 (14.22)	84 (15.36)
NA	8 (1.72)	14 (2.56)
BMI status, n/%
Obese	57 (12.28)	55 (10.05)
Overweight	85 (18.32)	133 (24.31)
Normal weight	302 (65.09)	350 (63.99)
Underweight	20 (4.31)	9 (1.65)
BMI, kg/m2, mean ± SD
Obese	35.46 ± 4.52	33.39 ± 3.39
Overweight	26.81 ± 1.42	26.75 ± 1.33
Normal weight	21.70 ± 1.71	22.11 ± 1.68
Underweight	17.39 ± 0.91	17.73 ± 0.63
Smoking, n/%
Yes	56 (12.07)	67 (12.25)
No	368 (79.31)	355 (64.90)
Not reported	40 (8.62)	125 (22.85)
Alcohol, n/%
Yes	385 (82.97)	378 (69.10)
No	36 (7.76)	43 (7.86)
Not reported	43 (9.27)	40 (7.31)
Physical activity (last 7 d), n/%
Less than once	94 (20.26)	37 (6.76)
1–3 times	194 (41.81)	135 (24.68)
4 or more	133 (28.66)	244 (44.61)
NA	43 (9.27)	131 (23.95)
Sedentary behavior (hours per day), n/%
0	1 (0.22)	/
1	69 (14.87)	38 (6.95)
2	193 (41.59)	203 (37.11)
3	120 (25.86)	141 (25.78)
4	42 (9.05)	41 (7.50)
NA	39 (8.41)	124 (22.67)
WC, cm, mean ± SD	76.71 ± 12.27	82.43 ± 11.18
HOMA-IR, mean ± SD	0.93 ± 0.67	0.81 ± 0.62
Glucose, mmol/L, mean ± SD	4.84 ± 0.37	5.03 ± 0.42
Insulin mU/L, mean ± SD	4.26 ± 2.97	3.58 ± 2.68
Calculated LDL, mmol/L, mean ± SD	2.55 ± 0.63	2.43 ± 0.67
HDL, mmol/L, mean ± SD	1.42 ± 0.31	1.23 ± 0.26
Western dietary pattern z-score, mean ± SD	−0.34 ± 0.72	0.41 ± 0.93
Healthy dietary pattern z-score, mean ± SD	0.004 ± 0.82	0.03 ± 0.96
Dietary misreporting, n/%
Underreporting	190 (40.95)	128 (23.40)
Plausible reporting	190 (40.95)	224 (40.95)
Overreporting	19 (4.09)	33 (6.03)
NA	65 (14.01)	163 (29.80)

Abbreviation: NA, not applicable. Sample size/percentages and means ± standard deviations are reported.

Obesity

The regression analysis showed significant associations between WC and SM, LPC, and PC concentrations (Figure 1). There was no significant sex and metabolite concentration interaction. The stratified analysis of the data is shown in Supplemental Figure 4.

Figure 1.

Manhattan plot of the analytes of the multiple regression model for WC to show metabolite trends.

Light dashed line corresponds to the significance level of α = .05. Dark dashed line corresponds to the Bonferroni-corrected significance level of α = .05/175 (number of analytes). Points are −log10 P values of the regression model. Dependent variable is WC; independent variable is the respective analyte; adjustment is HDL-C, LDL-C, HOMA values, smoking, alcohol consumption, dietary patterns, physical and sedentary behavior, and biological sex. Metabolite names given only for significant values. All β-coefficients, CIs, P values, and Bonferroni corrected P values can be seen in Supplemental Table 3.

The analytes that were significantly positively associated with WC values are shown in Figure 2. These were NEFA C17:1, PCaa C40:6, PCaa C40:5, PCaa C38:5, PCaa C38:4, PCaa C38:3, SMa C43:3, SMa C42:4, SMa C42:3, SMa C40:2, SMa C36:3, SMa C36:2, SMa C36:0, SMa C34:3, SMa C34:2, SMa C33:2 and SMa C32:2. The metabolites LPCa C18:2 and LPCa C18:1 were significantly decreased.

Figure 2.

Significant analytes of the multiple linear regression model with WC as outcome and metabolite concentrations as predictor, adjusted for sex, HOMA values, LDL-C, HDL-C, dietary patterns, dietary misreporting, smoking and drinking behavior, physical activity, and sedentary behavior in the 20-year follow-up of the Western Australian Pregnancy Cohort (Raine) Study.

Standardized estimates, Bonferroni-corrected CIs, and P values are reported.

Although NEFA concentrations were not significantly elevated, there was a trend toward an increase of most of the species (Figure 1).

Insulin resistance

Log-HOMA-IR values were used as a continuous variable to examine associations with lipid analytes. The additional model used insulin values (Supplemental Figure 5). The increased metabolites in subjects with high HOMA-IR values were PCaa C32:2 and LPCa C14:0, whereas PCaa C43:6 and PCaa C44:12 were decreased (Table 3).

Table 3.

Significant Analytes of the Multiple Linear Regression Model With HOMA-IR as Outcome and Metabolite Concentrations as Predictor, Adjusted for Sex, BMI, LDL-C, HDL-C, Dietary Patterns, Dietary Misreporting, Smoking and Drinking Behavior, Physical Activity, and Sedentary Behavior at 20 Years

Analyte	Standardized Estimate	Bonferroni CI	Bonferroni P Value
PCaa C43:6	−0.09	−0.18 to 0.01	.01
PCaa C44:12	−0.09	−0.18 to 0.01	.01
LPCa C14:0	0.08	0.001, 0.16	.05

Standardized estimates, Bonferroni corrected confidence intervals, and P values are reported.

The associations between insulin and analytes showed the same trend (Supplemental Figure 5).

The HOMA-IR model without BMI as confounder showed similar metabolite trends as the WC model for differences in obesity, meaning that elevated SM values and SMa C32:2 are the most significant analytes with higher BMI (Supplemental Table 5).

Our data suggest increased SM concentrations are indicative of only obesity but not for IR. To explore the link between obesity and IR, we stratified the data into subjects with overweight/obesity and normal weight. This analysis showed a significant association between LPC C14:0 (SE 0.11; Bonferroni CI 0.02, 0.2; P correlation 0.001) concentration and log-HOMA in the normal-weight group. In the obese/overweight group, no metabolites were significantly associated with log-HOMA concentrations, but there was a trend toward decreased LPC concentrations.

Discussion

Concentrations of NEFA C17:1, PCaa C40:6, PCaa C40:5, PCaa C38:5, PCaa C38:4, PCaa C38:3, SMa C43:3, SMa C42:4, SMa C42:3, SMa C40:2, SMa C36:3, SMa C36:2, SMa C36:0, SMa C34:3, SMa C34:2, SMa C33:2, and SMa C32:2 were positively and LPCa C18:2 and LPCa C18:1 were negatively associated with WC in study participants, after adjusting for HOMA-IR, ethnicity, sex, dietary patterns and misreporting, smoking, alcohol consumption, physical activity, sedentary behavior, and LDL-C and HDL-C concentrations. HOMA-IR values were positively associated with LPC C14:0 and PCaa C32:3 and negatively with PCaa C44:12 and PCaa C43:6, with the same covariates but BMI instead of HOMA-IR.

The main source of phospholipids are lipoproteins, secreted by the liver. Although many studies focus on the effect of HDL-C or LDL-C on metabolic diseases, our study is among the few metabolomics studies that account for their possible effect of the polar lipid compartments (24, 25).

HDL has a lower SM to PC ratio than LDL, with PC being the first and SM being the second most abundant lipid subclass in lipoproteins (26). Triacylgylcerols, glycerophospholipids, and sphingolipids are packed in very low-density lipoprotein, which can cross the cell border into the bloodstream. By the release of triacylglycerols, very low-density lipoproteins are converted to LDL.

Obesity

Increased concentration of SM species associated with high WC suggests an enhanced SM biosynthesis. During SM biosynthesis, the phosphocholine head group of PC is bound to a ceramide molecule, an intermediate in sphingolipid metabolism, via ceramide-choline phosphotransferase (27). The major flux through this pathway occurs through ceramidase to generate sphingosine and subsequently sphingosine 1 phosphate (28). Our finding of higher LDL compared with HDL with higher WC values (Supplemental Figure 6) suggests greater amounts of LDL-derived SM, reflected in our findings that higher SM levels are associated with obesity.

This could also be an explanation for the decreased levels of LPC C18:1 and C18:2 in subjects with higher WC. Barber et al (29) reported decreased LPC C18:1 and C18:2 with obesity and suggested that obesity-related factors, such as diet and adiposity, rather than IR and diabetes per se, may make the major contribution to the changes in the LPC profile. LPCs in human plasma are mainly derived by the action of lecithin cholesterol acyltransferase (LCAT). LCAT removes fatty acids from the sn-1 or -2 position of PC; the former are then transferred to cholesterol (30). LCAT is extracted from the liver into the bloodstream and bound to the surface of HDL. LCAT is negatively affected by higher levels of SM (31). Subjects with higher BMI had higher levels of LDL, lower levels of HDL, and therefore a higher level of SM. A decrease in LCAT activity could be a reason for higher PC levels, which are no longer used for the esterification of cholesterol as well as the lower LPC levels and subsequent the concentration of HDL. In the process to form HDL, LCAT activity is the final step to esterify the cholesterol. Otherwise, reverse cholesterol transport is disturbed.

A possible biological explanation for the elevated SM species C34:2 and C36:2 could be an increase in the activity of the δ9 desaturase stearoyl-CoA desaturase 1 (SCD1). This desaturase is specific for palmitic acid (C16:0) and stearic acid (C18:0) and is a key enzyme in the synthesis of palmitoleic acid (C16:1) and oleic acid (C18:1), respectively. The monounsaturated fatty acids resulting from the SCD1 activity are major substrates in the synthesis of complex lipids (32). Because the sphingosine backbone of SM consists of a C18:1 molecule, the species C34:2 and C36:2 likely also contain palmitoleic and oleic acids, respectively (33). Studies in rats have found positive associations between SCD1 activity and obesity (34). This is reflected in our analysis of the ratio of NEFA C16:1 to C16:0, which is a marker for elevated SCD1 activity. The ratio was significantly positive associated with WC, supporting our hypothesis. Although the ratio of NEFA C18:1 to C18:0, another marker for the SCD1 activity, was increased, it was not significantly associated (Supplemental Table 6).

Insulin resistance

In normal-weight subjects, increased LPC C14:0 was significantly associated with high HOMA-IR, whereas there was no significant association with obesity.

A study in infants found elevated LPC C14:0 levels predictive of later obesity at early school age (35). Some evidence exists that myristic acid augments the release of insulin in pancreatic β-cells (36).

The lipotoxicity hypothesis is a possible explanation for the development of IR in obesity. This hypothesis states that an oversupply of fats, which exceed the capacity of adipocytes, leads to storage in other tissues. In response, these cells produce bioactive lipids that reduce insulin sensitivity and fat flow into the cell.

Higher NEFAs in adipose tissue can lead to IR and consequently more NEFA release. Elevated levels of NEFAs are then taken up by other insulin-sensitive tissues including liver or muscle cells, resulting in constant NEFA blood levels during early stages of IR/T2DM. Our finding of few elevated NEFAs may be due to the young age of the Raine participants (37).

Epigenetic programming

To look for possible associations of early-life influences (possibly mediated through epigenetic mechanisms) on the metabolite concentration at 20 years of age, we adjusted the models for maternal prepregnancy BMI, maternal birth weight, birth weight of the child, mode of delivery, and weight change in the first year of life (Supplemental Figures 7 and 8) (38). The results suggest that there is no change in the trend of the metabolite concentrations by including those variables, but LPC concentrations were no longer significantly associated with WC after adding weight change in the first year of life. This could suggest genetic effects on the body composition of the child and should be further analyzed.

Strengths and limitations

The strength of our study is the metabolomics/lipidomics approach targeting polar lipids with a sample size of more than 1000 subjects, from a general population with a relatively high prevalence of overweight and obesity. Thus, our statistical power is exceptionally high. To our knowledge, this is the first time a statistical approach that accounts for the effects of LDL and HDL and then also examines the residual relation of lipidomics in obesity and IR has been applied. Additionally, our analyses accounted for possible sex differences, ethnicity, and a number of potential lifestyle confounders.

Limitation is the cross-sectional design, which does not allow for differentiation between causes and consequences. The study's primary aim is hypothesis generation because most of the pathways are not very well understood at the moment. Also, although the number of obese and overweight is relatively high at approximately 33%, higher rates of obese subjects would lead to more valid conclusions.

There are also limitations to the HOMA model as a measure for insulin resistance, with the gold standard being clamp techniques. Nevertheless, it is frequently used in population-based studies (39).

Conclusion

Even though the pathways are not very well understood to date, the results presented here suggest that SM, PC, and LPC are associated with obesity and IR independent of LDL-C and HDL-C. The underlying mechanism for the elevated SM and PC levels in human plasma could be explained by the hypothesis of lipotoxicity.

In conclusion, we have shown plasma SM and particularly SMa C32:2 associates predominantly with obesity, and plasma LPCa C14:0 associates with IR, in young adults. In normal-weight individuals, there was a significant association with LPC C14:0, whereas there was no significant association between HOMA-IR and metabolite concentrations in the obese/overweight group. Future studies should focus on metabolites involved in sphingolipid metabolism, and their relationship to a diet rich in saturated fats in individuals with obesity and normal weight and those with IR to examine the effects of sphingolipids on diseases associated with the metabolic syndrome and examine saturated, monounsaturated, and polyunsaturated SM in relation to IR, obesity, or overweight.

It is also of major interest for future research to analyze the influence of different lifestyle factors on metabolites for a better understanding of their interplay and influence on diseases like obesity and IR.

In addition, further studies are required to examine possible different diagnostic approaches for IR in normal-weight/underweight compared with overweight/obese patients. Also, longitudinal studies are necessary to determine causality in the hypotheses we have generated here.

Acknowledgments

We are grateful to Stephanie Winterstetter (Division of Metabolic and Nutritional Medicine, Dr von Hauner Children's Hospital, University of Munich, Munich, Germany), who analyzed the blood plasma samples and to Gina L. Ambrossini (Nutrition and Health Research Group, School of Population Health, University of Western Australia, Perth, Australia), who calculated the dietary patterns and misreporting. Also, we gratefully acknowledge the Raine Study participants and their families and the Raine Study Team for cohort coordination and data collection.

Supplemental information is available at the JCEM web site.

This manuscript does not necessarily reflect the views of the Commission of the European Communities and in no way anticipates the future policy in this area. The supporters of this work had no role in the study design, the data collection and analysis, the decision to publish, or the preparation of the manuscript.

The data presented are part of the PhD thesis accomplished by Sebastian Rauschert at the Medical Faculty of the Ludwig-Maximilians-University of Munich.

Author contributions included the following: S.R. wrote the paper and performed the statistical analysis and data interpretation. C.H. performed the quality control analysis and contributed to the statistical analysis and data interpretation. O.U. performed the liquid chromatography and tandem mass spectrometry analysis and contributed to the data interpretation. F.K. contributed to the statistical analysis. B.K. and W.H.O. conceived the study. T.A.M. was responsible for the collection, storage, and shipment of the plasma to Germany for the metabolomics analysis. T.A.M., L.J.B., and W.H.O. were responsible for all the data collection. All coauthors have contributed to the content and read and approved the final manuscript.

This work was supported by the National Health and Medical Research Council Project Agreement 1022134 and European Union Collaborative Project Agreement 1037966. Further support was provided by the European Union's seventh Framework Programme Grant FP7/2007–2013, the EarlyNutrition Project under Grant Agreement 289346, and the European Research Council Advanced Grant ERC-2012-AdG, number 322605 META-GROWTH. Core funding for the Western Australian Pregnancy Cohort (Raine) Study is provided by the University of Western Australia; the Faculty of Medicine, Dentistry, and Health Sciences at the University of Western Australia; the Telethon Kids Institute (formerly known as the Telethon Institute for Child Health Research); the Women and Infants Research Foundation; Curtin University of Technology; Edith Cowan University; and the Raine Medical Research Foundation.

Disclosure Summary: The authors have nothing to disclose.

C.H. and W.H.O. contributed equally to this work.

Abbreviations

 
• BMI

  body mass index

 
• CI

  confidence interval

 
• IR

  insulin resistance

 
• HDL-C

  high-density lipoprotein-cholesterol

 
• HOMA-IR

  homeostasis model assessment index of insulin resistance

 
• LCAT

  lecithin cholesterol acyltransferase

 
• LDL-C

  low-density lipoprotein-cholesterol

 
• LPC

  lysophosphatidylcholine

 
• NEFA

  nonesterified fatty acid

 
• PC

  phosphatidylcholine

 
• PCaa

  diacylphosphatidylcholines

 
• SCD1

  stearoyl-CoA desaturase 1

 
• SM

  sphingomyelin

 
• T2DM

  type 2 diabetes mellitus

 
• WC

  waist circumference.