Triglyceride profiling in adipose tissues from obese insulin sensitive, insulin resistant and type 2 diabetes mellitus individuals

Participants’ recruitment criteria were described previously [12]. Briefly, 64 consented obese individuals undergoing bariatric surgery at AlEmadi hosptial (Doha, Qatar) were recruited. Protocols were approved by Institutional Review Board of ADLQ (X2017000224). Blood was taken prior to operation and 1–5 g of abdominal SC and OM adipose tissues biopsies were collected during the surgery and stored at − 80 °C until use. Plasma cholesterol, fasting glucose and liver function enzymes were measured by COBAS INTEGRA (Roche Diagnostics, Basil). IL-6, leptin and insulin were determined using commercially available ELISA. Insulin resistance was computed by homeostatic model assessment (HOMA-IR) [13] using 30th percentile (HOMAIR = 2.4) as a threshold point. Accordingly, subjects were dichotomized into IS (HOMA-IR < 2.4, n = 18, 3 males and 15 females) and IR (HOMA-IR > 2.4, n = 35, 9 males and 26 females). Eleven participants were clinically diagnosed with T2DM (8 males and 3 females).

Stromal vascular fraction (SVF) cells were obtained by collagenase digestion of adipose tissues as described previously [12]. Cell pellets were re-suspended in stromal medium containing Dulbecco’s modified Eagle’s Medium-F12 (DMEM-F12) supplemented with 10% fetal bovine serum (FBS) and Penicillin/Streptomycin, then maintained at 37 °C with 5% CO₂ until confluence. To induce differentiation, early passaged stromal vascular fraction (SVF)-derived preadipocytes (passages 1–3) were grown at 2 × 10⁴/cm² in stromal medium overnight, then incubated in differentiation medium (DMEM-F12, 3% FBS, 33 μM biotin, 17 μM d-pantothenate, 1 μM dexamethasone, 250 μM of methylisobutylxanthine, 0.1 μM human insulin, 5 μM of Peroxisome proliferator-activated receptor gamma PPARγ agonist, rosiglitazone) for 7 days, followed by 12 days in maintenance medium containing the same components as the differentiation medium but omitting methylisobutylxanthine and rosiglitazone. Differentiation potential (adipogenic capacity) was determined as a percentage of lipidtox positive stained cells to total number of stained nuclei (DAPI).

Human SC and OM adipose tissue specimens from IS, IR and T2DM individuals were extracted using the Bligh and Dyer Method [14]. Homogenization of tissue was carried out in the gentle MACS Dissociator (Miltenyi Biotech, Germany) with one volume of PBS for every gram of tissue. Following tissue homogenization, 1 mL of each sample solution was transferred into a separate 15 mL Falcon tube, and 3 mL of 3:1 ratio of Chloroform:MeOH were added into each tube. One microliter of PBS was added and samples were centrifuged at 3000 RPM for 20 min at room temperature. The organic layer (bottom layer) was carefully transferred into new 15 mL Falcon tubes and evaporated to dryness under a stream of high purity nitrogen. Samples were then reconstituted with 1:1:1 mixture of hexane, isopropanol, acetonitrile. Subsequently the extracts were analysed using data dependent full scan MS and MS/MS acquisition using the Thermo LTQ VelosPro dual cell linear ion trap mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA).

Separation of TAGs was carried out using non-aqueous reverse phase UHPLC separation (NARP), on a Dionex Ultimate 3000 UHPLC system, using acetonitrile w/0.1% formic acid (eluent A), and isopropanol w/10 mM ammonium formate (eluent B) as the mobile phase. The column was a Phenomenex UHPLC C30 core shell, 150 mm × 2.1 mm and 2.7 µm particle size (Phenomenex Torrance CA, USA). Gradient conditions started with 5% B held for 2 min, then raised to 50% B at 30 min, held for 10 min and then reduced to 5% B at 45 min and held for further 5 min.

MS analyses were conducted using the Thermo LTQ VelosPro dual cell linear ion trap mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA), acquiring both full scan MS and subsequent data dependent full scan MS/MS product ion spectra with wide band activation. Target parent ions were automatically selected from an inclusion list. The low resolution full scan analysis provides molecular parent masses (M⁺NH₄⁺). These parent ion full scan MS/MS analysis provided further elucidation of possible structures represented in each lipid (fatty acid composition). Relative abundances of each identified TAG were estimated from the height values for each extracted ion current profile for parent masses of each compound (M⁺NH₄⁺).

The above UHPLC conditions (NARP) provide separations of TAGs by their equivalent carbon number (ECN). The ECN is calculated, from the total number of non-glycerol carbons in the TAG minus twice the number of the double bonds in the molecule (ECN = CN − 2DB). NARP eluted the TAGs from lower to higher ECN with increasing percent B in the eluent. NARP-HPLC is commonly used for TAG separation because it works on both the chain and absolute height or area counts for each identified TAG. As some of the TAG may not show baseline resolution, the height counts were chosen to better represent the TAG.

All statistical analyses were carried out using R version 3.2.1 and SIMCA 13.0.1 software (Umetrics, Sweden). Variables with skewed distributions were log transformed or taken the square root of as appropriate to ensure normality. An initial PCA was conducted to identify components that explain large proportion of the TAG variance. A repeated measures linear model incorporating confounders: gender, age, BMI, PC1 and PC2 (derived from earlier principle component analysis, PCA) and covariates: tissue and diabetic status (IS, IR, T2DM) was used to assess the differences in each TAG between the two tissues and amongst the insulin/diabetes groups. The model was based on repeated measures statistics since a TAG measurement from an individual was taken from two separate tissues: SC and OM. The model allows the individual inherent variation to be taken out of the total variance. Such enhanced modelling of the error structure increases the model’s ability to detect significance of covariate effects. Nonetheless, we have repeated the analysis using the standard linear model and confirmed the superiority of the repeated measures linear model counterpart. The linear model was sometimes used when fitting the repeated measures model was not possible due to missing data. False discovery rate (FDR) multiple testing correction was also performed on the differentially expressed TAG species identified between adipose tissues from IS, IR and T2DM individuals. Fatty acid enrichment amongst diabetes/tissue significant TAGs was assessed using the one tailed Wilcoxon sum of the ranks test on the list of metabolites that differed significantly between IS, IR and T2DM after correcting for covariates including gender, age, BMI, PC1 and PC2. The analysis was based on assessing the likelihood of randomly observing a given fatty acid that often amongst highly ranked TAGs along the list of all TAGs ordered by p value as follows: For each of the following contrasts: subcutaneous versus omental, IR versus IS, IR versus T2DM and IS versus T2DM, TAGs were ranked by their p values and a given fatty acid mapped to the ranks of TAGs within which it is found. The analysis proceeds by assessing the likelihood of obtaining the observed sum of fatty acid identified ranks by chance. If the fatty acid is observed amongst the significant TAG at the top of the list, the sum of the ranks would be too small to be explained by chance alone; hence the null hypothesis is rejected in favor of enrichment. Enrichment hits failed to remain significant after FDR multiple testing correction but data was reported because of agreement with literature as elaborated in “Discussion” section. A similar test was used to assess enrichment in constituent fatty acid saturation levels.

Sixty-four (44 females and 20 males) obese and morbidly obese (BMI = 43.1 ± 7.5 kg/m²) participants were recruited from amongst patients undergoing weight reduction surgery. Participants exhibited hyperleptinemia and hyperinsulinemia and were dichotomized into IS and IR groups based on their HOMA-IR index and into T2DM based on their medical records. Compared to BMI-matched IS and IR subjects, T2DM individuals were older and had higher circulating levels of TAG and lower leptin (Table 1). Compared to females, males had higher mean arterial blood pressure (MAP) (93.7 vs 84.7, p > 0.01) and lower HDL (1.1 vs 1.5, p = 0.05) and leptin (42.0 vs 67.3, p > 0.01) (Additional file 1: Table S1). IS males had lower HOMA-IR than their age and BMI-match IS females, whereas IR males had higher HOMA-IR than their age, but not BMI, matched females (Additional file 1: Table S1). Compared to obese subjects (n = 26), the morbidly obese participants (n = 46) had significantly higher BMI, SBP, IL-6, FPG and HOMA-IR (Additional file 1: Table S2).

Using a non-targeted approach, a comprehensive parent mass list of 120 identified TAGs was created, of which 76 TAG species were identified (Appendix, Table 6) based on their molecular weights and peak heights. A linear model was used to assess depot-specific TAG associations after correcting for participant diabetes group, gender, PC1 and PC2 (refer to “methods” section). Analysis revealed 7 TAGs that were significantly different between SC and OM tissues. C53:5, C51:3, C50:4, C59:1, C54:6 and C50:2 were higher in OM than SC. C38:1 was higher in the SC compared to OM tissues. The full scan MS/MS analysis revealed the fatty acid composition for each identified TAG (Table 2).

A linear model was used to assess TAG associations with participant groups after correcting for possible confounders (refer to “Methods” section). A number of TAGs were significantly decreased in T2DM compared to IS and/or IR including C46:4, C48:5, C48:4, C38:1, C50:3 and C40:2 whereas a number of TAGs were increased in T2DM compared to the other two groups including C56:3, C56:4, C56:7 and C58:7. No significant differences in TAGs between IS and IR groups was detected. Table 3 summarizes the list of differentially expressed TAGs with their fatty acids compositions. When looking at gender versus group (IS, IR and T2DM) interaction, there were no FDR significant interaction effects. However, when considering BMI versus group interaction, two TAG species showed FDR significant interaction effects including C40:2 and C53:4. Whereas the former (C40:1) shows more pronounced decrease in T2DM compared to IS in low BMI than in high BMI, the latter (C53:4) shows a more pronounced increase in low BMI than in high BMI (Additional file 1: Table S3).

An orthogonal partial least square discriminate analysis (OPLS-DA) comparing subjects from IS, IR and T2DM revealed two significant class-discriminatory components (R2X = 0.18, R2Y = 1, R2Q2 = 0.27, CV-ANOVA p = 0.0001) (Fig. 1). The score plot in Fig. 1a indicates an x-axis differentiating the T2DM group from IS and IR; the latter two groups being rather separated along the y-axis. The corresponding loading score, shown in Fig. 1b, features similar TAG/group associations to those obtained with the linear model (shown in Table 3). Specifically, lower amounts of C38:1, C46:4, C48:5 and C48:4 as opposed to higher levels of C58:7, C56:4, C56:4 and C56:7 in the T2DM group (also circled in red, Fig. 1b).

In order to study the possible enrichment/depletion of certain fatty acid constituents of TAGs in association with insulin sensitivity or diabetes, an enrichment analysis based on the Wilcoxon sum of the ranks test was conducted (refer to “Methods” section). The results of the analysis are presented in Table 4 and further illustrated on Fig. 2. Overall, C12:0 appears to be associated with TAGs least abundant in T2DM in both tissues whereas C18:3 is found in both depleted and enriched TAGs in T2DM (both sides of the x-axis in Fig. 2). This could be justified by a potentially induced flow of C18:3 in certain recipient TAGs at the expense of other TAGs with diabetes. Further supporting this are the observed negative correlations between depleted and enriched C18:3 carrying TAGs (Fig. 3). Interestingly, many of the C12:0 and C18:3 containing TAGs, including TAG21, TAG22, TAG75 and TAG9, were previously identified as significantly changing in level with diabetes by the linear model (Table 3).

Furthermore, C18:3 was also found to have a strong tissue signature featuring frequently amongst highly ranked TAGs from statistical analysis of TAG association with tissue type (data not shown). However, C18:3 does not feature amongst the TAGs found to significantly differ between tissues by the linear model (Table 2). This indicates that the collective tissue changes in C18:3 containing TAGs were rather subtle.

The Wilcoxon sum of the ranks analysis was also used to look for enrichment of fatty acid saturation level amongst the highly ranked significant TAGs from comparison of tissue/diabetes-insulin sensitivity groups but no significance was detected.

A step-wise regression was performed to identify the best TAG predict of various traits including age, BMI, systemic TAG, total cholesterol, IL-6 and HOMA-IR, SC and OM adipogenic capacity reported previously [12, 15] and shown in Additional file 2: Figure S1. Table 5 lists TAG species identified with significant (p = 0.0001) association with various metabolic traits and shows their importance and fatty acid compositions.

Our data identified few TAGs that were differently expressed between SC and OM tissues. One TAG that was higher in SC compared to OM was C38:1, which contains C4:0 (butyric acid). The latter was shown before to inhibit lipolysis and increase insulin sensitivity in primary rat adipocytes [20], perhaps contributing to the greater association of insulin resistance with OM mass compared to SC mass [21]. A previous study in obese men has shown increased C50:0, C59:2, C58:2, C60:3, C64:4, C51:0 and C65:1 fatty acids in OM compared to SC adipose tissues [6]. Changes in lipid composition between the two depots were attributed to differences in adipocyte differentiation, metabolism of the lipid droplet, and extent of beta-oxidation [6]. Differences between the two studies may reflect ethnic and/or diet differences between our Asian and the other study’s Caucasian population. Variations in fatty acid composition between SC and OM fat depots confirm the specific metabolism of each depot, as selective lipolytic and lipogenic mechanisms may function in each tissue depot. Indeed, studies have shown that desaturase enzymes, regulating the number of saturated fatty acids, exhibit a depot-specific profile [22] in close association with insulin resistance [23].

Systemic levels of fatty acid increase with obesity and T2DM, perhaps as a result of insulin resistance of adipose tissue and subsequent increased lipolysis; although in some obese individuals, fatty acid release from adipose tissues is reduced per kg fat in order to normalize plasma non-esterified fatty acid concentrations [24]. In our study, significant differences in levels of TAG composition were detected between IS, IR and T2DM. A number of candidates were either increased or decreased with risk of insulin resistance and T2DM, despite the predominant view of TAGs as an adverse risk factor for diabetes. Using the Wilcoxon sum of ranks statistics, fatty acids frequently occurring in highly-ranked TAGs along the list of TAGs ordered by p value from diabetes association analysis were revealed. Two fatty acids were identified: C12:0 and C18:3. The strength of this enrichment analysis approach is that, unlike the Fishers’ exact test, no arbitrary significance cut-off is applied on the list of TAGs. However, a possible weakness relates to the fact that since the TAGs are ordered by p value, no account is given to the direction of change and therefore one may not speak of depletion or increase in fatty acid TAG level but rather a dynamic in metabolic activity involving the fatty acid in association with the phenotype of interest. This was observed with C18:3, and a negative correlation was noted between C18:3 host TAGs found increased and others decreased with diabetes, effectively suggesting a metabolic link between the two sets of TAGs. Our findings confirm previous studies that showed significant correlations of specific fatty acids with insulin sensitivity. These include a cross-sectional analysis of adipose tissue biopsies from elderly obese men, which identified positive correlations between levels of C12:0, C18:2 and C18:3 and insulin sensitivity [25]. Our data also confirmed the association of C18:3 with metabolic status as shown previously in two groups of obese individuals who underwent weight loss surgery [26]. Furthermore, subjects in the most insulin-sensitive quintile showed a significantly higher percentage of circulating C18:2 (pre-cursor of C18:3) than the remaining subjects [27], further confirming our data. Functionally, previous work implicated C18:2 in the modulation of insulin signaling in rat skeletal muscle [28]. Therefore, our findings confirm previous results with regard to the association of C12:0 with insulin sensitivity [25], perhaps through triggering Glut4 translocation [27]. Our data also revealed reduction in C18:3 with T2DM incidents. This also confirms previous findings showing a negative correlation of C18:3 and its precursor with insulin resistance and positive association with insulin sensitivity [27].

Further, our data highlight a panel of TAGs that were associated with mediators of metabolic disease in obese individuals. Increased age was associated with accumulation of C56:1 that is composed of saturated fatty acids C20:0 and C18:0 and mono-unsaturated C18:1, whereas age was negatively correlated with C54:8 that is composed of unsaturated fatty acids C18:2 and C18:3. Although participants had comparable BMI, the small increase in BMI was positively correlated with three unsaturated TAGs (C57:1, C48:1 and C54:5). Whereas circulating TAGs were associated with accumulation of C52:1 and C54:1 in the adipose tissue, total cholesterol was positively correlated with C40:2. The negative correlation between IL-6 and C38:1, C42:1 and C56:1 may suggest an anti-inflammatory effect of fatty acids that constitute these TAGs, in particular C10:0 that was shown previously to exert an anti-inflammatory properties [29]. HOMA-IR was also negatively correlated with C44:2 and C56:7, both containing C18:2 shown previously to negatively correlate with insulin resistance [25].

Several TAGs were highly correlated with SC or OM adipogenic capacity. Previous studies have shown that the greater adipogenic capacity of SC and OM preadipocytes taken from IS obese individuals compared to IR and T2DM counterparts is partially mediated by lower IL-6 secretion and oxidative stress [12, 15, 30]. Secretion of interleukin IL-6 is significantly decreased after treatment with C18:2, C22:6 and C16:0 via inhibition of nuclear factor kappa B (NF-κB) and subsequent activation of the master regulator of adipogenesis, PPARγ [31]. Our data revealed positive correlations of C56:4 and C57:4, containing C18:2, C16:0, with SC adipogenic capacity. OM adipogenic capacity was associated with C49:1, C38:0 and C56:2, containing C16:0, C18:1 and C14:0. These fatty acids were shown previously to induce adipocyte differentiation in rodents [32‐36] and potentially play a similar role in human preadipocytes.

One main limitation of this study is the relatively low number of participants, especially in the T2DM group. Additionally, the difference in gender distribution between IS and IR groups (predominantly females) and T2DM (predominantly males) group may have introduced bias in the study design that may have influenced the results. Despite these factors, clear TAG and fatty acid signatures were identified after correcting for potential confounders such as gender and BMI. Another limitation of the current work is its focus on association of TAGs with insulin resistance and risk of T2DM without an absolute quantitation of any specific analyte. Incorporating isotope-labeled standards would allow absolute quantitation and improve the precision of measurements. Finally, differences in TAG composition in adipose tissues among the studied groups may have been influenced by their diet. Indeed, previous studies have shown that the process of fatty acid and TAG deposition in rat adipose tissue depends on the composition of the diet [37]. Dietary linoleic acid content was shown to influence the distribution of TAG species in rat adipose tissue, particularly di- and trilinoleoyl containing TAG as a result of linoleic acid intake [38]. Other studies have shown that the composition of TAG in rat epididymal, subcutaneous and perirenal adipose tissues was broadly reflecting dietary oils such as isomeric octadecenoic acids from coriander oil and high oleic sunflower oil [39]. Taken all these limitations into account, confirmation in different populations is warranted to validate these findings.