The experiments were performed according to the rules set by the Netherlands Law on Animal Experiments and approved by the Institutional Ethical Committee on Animal Care and Experimentation (Dierexperimenten Commissie DEC of Netherlands Organization for Applied Scientific Research, Zeist, the Netherlands) with a permit number of DEC2489.

Twenty-four female ApoE*3Leiden.CETP transgenic mice (age 6–10 weeks) were obtained from specific pathogen free (SPF) breading stock (TNO, Leiden, The Netherlands). The animals were fed a semi-synthetic modified Western-type diet (Hope Farms, The Netherlands), containing 0.2% cholesterol (Cho), 15% saturated fat and 40% sucrose as described by Nishina et al. [21], for a run-in period of 4 weeks to get a mild hypercholesterolemia (plasma Cho levels of 14–18 mmol/L) and body weight increase. Following this run-in period, mice were matched on body weight, plasma Cho and triglycerides (TG) levels (after 4 hours fasting) and divided in three groups of eight animals each (non-treated control, rimonabant- and SUB885C- treated). Preparations for intervention were given orally as admix to a Western-type diet for 4 weeks. Briefly, mice received the Western-type diet, without or with either SUB885C (SU Biomedicine, The Netherlands) at a concentration of 2% or rimonabant (Sanofi-Aventis, The Netherlands) at 10 mg/kg body weight/day. Flavoured sugar lumps were added to the diet of all groups to mask possible tastes of intervention preparations. Body weight per mouse and food intake per cage was measured at intervention day 0, 2, 3, 4, 9, 11, 14, 21 and 28. Blood was collected at the start of the intervention (week 0) and just before sacrifice of the mice (week 4) via tail vein bleeding using CB 300 LH microvettes (Sarstedt, Nümbrecht, Germany). Plasma samples were collected by centrifugation of the blood samples for 10 min at 6000 rpm at 4°C. At the end of the intervention (week 4), animals were sacrificed with rapid asphyxiation using CO₂. Livers were dissected on ice and samples were weighted and snap-frozen in liquid nitrogen. Both plasma and liver samples were stored at −80°C until use.

The 3T3-L1 preadipocyte cell line (ATCC) was used to study possible direct lipolysis or adipogenetic activity by SUB885C. The 3T3-L1 cells were cultured and differentiated from pre-adipocytes into full grown adipocytes as described by Niwano et al. [22]. Three hundred micrograms of SUB885C was extracted with 4 ml methanol. After evaporation of the methanol, the extract was dissolved in 100 µl DMSO. To study adipolysis the growth medium was removed from the cells and cells were washed twice with 1 ml HBSS per well. Then, to each well 250 µl HBSS was added containing 2% BSA and 2.5 µl of the diluted SUB885C extracts in DMSO. As a positive control, 2.5 µl of 1 mM isoproterenol in DMSO was added (final concentration 10 µM isoproterenol). For the negative controls, either 2.5 µl DMSO (DMSO control) or nothing (blank) was added. After 3 h incubation glycerol release was measured with a commercial adipolysis assay kit (Chemicon Int., Temecula, CA). For the adipose conversion assay the addition of SUB885C started at the initiation of the cell differentiation. Together with the differentiation medium, SUB885C extract was added at different dilutions. At every medium replacement, fresh extract was added as well. After nine days of differentiation, the adipose conversion of the cells was analyzed by measuring the amount of fat produced. This was done by staining the fat with Oil Red O as described by Ramírez-Zacarías et al. [23].

Plasma Cho, TG, HDL-C, lipoprotein profiles, CETP levels and activities and alanine aminotransferase (ALT) were measured at week 0 and week 4. Plasma total cholesterol (TC) and TG concentrations were determined in each animal using enzymatic kits (Roche Molecular Biochemicals, Indianapolis, Ind, USA). Pooled lipoprotein profiles were measured by fast performance liquid chromatography (FPLC) using an AKTA apparatus (Amersham Biosciences). TC, TG and phospholipid levels were measured in the fractions of freshly obtained samples. Phospholipids were determined in the FPLC fractions using a phospholipids B kit (Instruchemie Co., The Netherlands). Plasma HDL was determined by quantification of HDL-C in plasma after precipitation of apoB-containing lipoproteins. Thus, 10 µl of heparin (LEO Pharma, The Netherlands) and 10 µl of 0.2 mol/L MnCl₂ were added to 20 µl plasma, and mixtures were incubated for 20 min at room temperature and centrifuged for 15 minutes at 13000 g at 4°C. Plasma ALT was measured in pooled samples using a Boehringer Reflotron system. CETP levels were measured per mouse using RB-CETP kits from Roar Biomedical, Inc. Endogenous CETP activity was measured as described before [24]. Briefly, ³H cholesterol was equilibrated for 24 h with plasma cholesterol at 4°C followed by incubation at 37°C for 3 h. Subsequently, apoB-containing lipoproteins were precipitated by addition of heparin/MnCl₂. Lipids were extracted from the precipitate and labeled cholesteryl esters were separated from labeled unesterified cholesterol on silica columns and assayed by liquid scintillation counting.

LC-MS analysis was performed on a hybrid liquid chromatography-linear ion trap-Fourier transform ion cyclotron resonance-mass spectrometric system (LC-FTMS) consisting of a Surveyor HPLC MS pump and an autosampler (Thermo Fisher) equipped with an Ascentis Express C8 2.1×150 mm (2.7 µm particle size) column (Sigma-Aldrich Chemie B.V., Zwijndrecht, The Netherlands). The LC-MS method used has been described previously by Hu et al. [25]. With this method, seven different lipid classes including both polar lipids such as lyso-phosphatidylcholine (LPC), lyso-phosphoethanolamine (LPE), phosphatidylcholine (PC), phosphoethanolamine (PE), sphingomyelin (SPM) and non-polar lipids such as cholesterol esters (ChEs) and TG were eluted from the column ionized with electrospray ionization in the positive ion mode. The MS detection was performed in the full scan mode with a range of mass to charge ratio (m/z) 400–1500. The identification of the detected lipid peaks was performed as described previously [25] and the current accurate mass data acquired by FT and linear-ion-trap MS/MS fragmentation. For those peaks without MS/MS fragmentations, identification was based on the observed accurate m/z and the relative retention times of specific m/z peaks. Forty-eight plasma samples (week 0 and week 4) and 24 liver samples (week 4) from ApoE*3.CETP transgenic mice of three groups were prepared in duplicate and injected once according to the procedures described above. The performance of the applied lipid profiling platform was assessed through the repeated analysis of the quality control (QC) samples [27]. The QC samples, used to monitor the LC-MS response in time, were prepared by pooling aliquots of 48 plasma samples for plasma lipidomics and 24 liver samples for liver lipidomics respectively to represent the full biochemical diversity of the study samples and allow the calculation of the analytical precision for all lipids measured. The QC sample data is also used to correct systematic errors such as batch to batch response differences by a single point calibration model [28], [29]. Ten QC plasma samples and 5 QC liver samples were processed exactly in the same way as the study samples. In total 106 plasma samples including 96 study samples and 10 QC samples and 53 liver samples including 48 study samples and 5 QC samples were injected into the LC-MS system. The study samples were randomly analyzed and the QC samples were placed at regular intervals in the analysis sequence (one QC after every 10 samples). Furthermore, method performance was carefully monitored using multiple IS and duplicate analysis of QC samples. Of note, plasma and liver samples were analyzed separately.

In total, 140 plasma lipid peaks and 137 hepatic lipid peaks were identified and selected as target lipids based on retention time and m/z of peaks processed by LC-Quan 2.5 (Thermo Fisher). The peak detection algorithm was used for peak integration using the following parameters: ICIS; smoothing points, 7; window, 30 s; view width, 3 min; baseline, 40; area noise factor, 5; peak noise factor, 1045. Data for each mouse was normalized for the recovery of the IS for injection. Batch to batch differences in the data were removed by synchronizing the medians values of the QC samples per batch. Data was used only if the duplicates corresponded after visual inspection and the duplicates were averaged.

One mouse (number 3733) in the control group was excluded from data analyses, because it did not respond to the diet during the run-in period and failed the inclusion criteria for hypercholesterolemia. Univariate data analyses were done by SPSS 17.0 and the results are presented as means ± standard deviation (SD) unless indicated otherwise. Statistical differences of biochemical parameters and lipidomics regulation at the end of the intervention of the three groups were analyzed parametrically by one-way analysis of variance (ANOVA). Data were log transformed if homogeneity of variance assumption was rejected. The Dunnett post hoc method was used to identify which treatment was significantly different from the control. One-tailed and two-tailed tests were used for biochemistry parameters and the lipidomics data respectively, with p values <0.05 as statistical significance. To correct for false positives, the multiple test correction (MTC) of Benjamini and Hochberg false discovery rate (FDR) analysis was applied to adjust p values derived from the univariate results of the lipidomics data. To avoid the possibility that a few high-intensity variables dominate the final results [30], plasma and liver lipidomics datasets were autoscaled per variable, meaning that first the means of each variable were subtracted, and then all variables were divided by their SD. To find treatment dependent lipidomics grouping effects, both lipidomics datasets were analyzed by Principal Component Analysis (PCA) [31] in MATLAB (version 7.7.0471, the Mathworks) with the PLS toolbox (version 5.0.3, Eigenvector Research, InC.). PCA was used to find patterns in the data such as clusters of mice (scores) of non-treated controls and undergoing the treatments; and to identify which lipids contributed most to these clusters (loadings). Partial least squares discriminant analysis (PLS-DA) [32] was further applied to identify the specific metabolites which contribute most to discriminate between the subtypes; and PLS-DA models were validated using double cross validation (DCV) [33]. Week 4 lipidomics were used for data analysis because (1) the study interest is in the effect of SUB885C in the end of the intervention and all baseline parameters are homogeneous among the three groups; (2) liver lipidomics results are only available at week 4. Finally data matrices of 15 mice (7 controls, 8 SUB885C) ×140 lipids for plasma lipidomics and 15 mice ×137 lipids for liver lipidomics were used for multivariate data analysis.

Compared to the controls, SUB885C did not have an effect on food intake in ApoE*3Leiden.CETP mice until day 20, and did not affect mean body weight (Figure 1). After day 21, an increase in food intake was observed with SUB885C. According to published results in the same experiment [34], mice treated with rimonabant showed a reduced food intake with a pronounced dip at day 2. A significant reduction of mean body weight of approximately 9% remained visible until the end of the intervention [34].

SUB885C treatment of ApoE*3Leiden.CETP mice caused a significant decrease in plasma Cho by 49% (8±1 mM versus 15±2 mM, p<0.001) after 4 weeks as compared to the controls (Figure 2A). During the 4 week intervention period plasma TG levels in both the SUB885C treated group and in control mice were significantly reduced by 67% (1.4±0.5 mM versus 4±2 mM, p<0.01) and 46% (2±1 mM versus 4±2 mM, p<0.01), respectively (data not shown). At week 4, TG levels in the SUB885C group tended to be lower compared to the control mice (Figure 2A), but this difference did not reach significance (1.4±0.5 mM versus 2±1 mM, p = 0.06). Plasma HDL-C in the SUB885C group was significantly increased by 39% (1.1±0.3 mmol/L versus 0.8±0.3 mmol/L, p<0.05) as compared to that of the controls in week 4. Lipoprotein fraction analyses showed that SUB885C treatment caused a 3-fold decrease of very low density lipoprotein-cholestol (VLDL-C) and 2.5 fold of VLDL-TGs and VLDL-phospholipids, as compared to those of the control group (Figure 2 B–D). In ApoE*3Leiden.CETP mice, the reduction of plasma TC was caused by the decrease of (V)LDL-C; while the increase of plasma HDL-C was in line with the obvious increase in HDL-C fraction.

Based on previous published results [34], rimonabant treatment of ApoE*3Leiden.CETP mice induced a significant decrease of plasma Cho with 24% (p<0.05) when compared with that of the controls. However, there were no significant changes of plasma TG and HDL-C at the end of the intervention. No adverse signs were observed during the study and ALT levels of all three groups were in the reference range.

Four weeks of SUB885C intervention (Figure 2A) of ApoE*3Leiden.CETP mice caused a significant decrease in CETP level by 31% (22±2 µg/mL versus 31±4 µg/mL, p<0.001) and CETP activity by 74% (24±11 µg/mL versus 91±27 µg/mL, p<0.001) as compared to the controls. However, rimonabant did not significantly affect the CETP level or the CETP activity in the same experimental setting [34].

Two Principal Components (PCs) were selected for plasma and liver lipidomics. These PCs described 67% and 57% of the total variance of the plasma- and liver-lipidomics, respectively (Figure 3A and B). In both PCA biplots, a separation was observed between control- and SUB885C treated-mice, indicating major changes of measured plasma and liver lipids between two groups. The loadings in the biplots (lipid species represented by colored symbols in Figure 3A and B) indicated that the lipids separating the SUB885C treated-mice from the control mice were TG, ChE and SPM. PLS-DA with DCV was used to further investigate the main discriminating lipids between SUB885C and control groups. The DCV error rates in both plasma and liver were 20%, indicating that 3 out of 15 mice were misclassified. Among 30 main discriminating lipids between the control and SUB885C treated mice, ChEs, SPMs, PCs, TGs contributed most for the separation in plasma while SPMs, PCs, TGs contributed most in liver (Table S1).

Quantitative differences in liver and plasma lipids between the SUB885C treatment group and untreated controls were statistically further analyzed by ANOVA with a two-tailed Dunnett post hoc method. SUB885C caused a significant reduction of plasma lipid classes of SPM, ChE, and TG by −28% (p<0.001), −46% (p<0.001), and −60% (p<0.01) compared to the controls, respectively. It caused a 12% reduction (p<0.05) of total liver SPM, without significant influence on total liver TG or ChE lipids.

It was found that 86 (61%) plasma and 22 (16%) liver lipids out of 140 and 137 lipids respectively were significantly changed after the 4-week SUB885C intervention as compared to the controls. Fifteen lipids changed significantly under SUB885C intervention in both plasma and liver, including LPC (18∶2), PC (36∶5), PE (34∶2), SPM (22∶1), SPM (24∶1), ChE (18∶2), ChE (20∶4), TG (50∶1), TG (54∶0), ether TG (TG-O) (50∶0), TG-O (50∶1), TG-O (50∶2), TG-O (52∶1),TG-O (52∶2) and TG-O (58∶2) (Table S2 and S3). The correlation analysis, data with normal distribution by Pearson's and without by Spearman's, was performed to evaluate whether there is a relation between the concentrations of these lipids in plasma and liver. Only TG-O (50∶0) showed a positive correlation between its concentration of plasma and liver; the others showed no correlation (data not shown).

After MTC analysis, 70 out of 86 plasma lipids and 3 out of 22 hepatic lipids remained significant (p = 0.00−0.02, p values marked in bold) by SUB885C intervention (Table S2 and S3). All three significantly changed hepatic lipids and 68 out of 70 (97%) significantly changed plasma lipids were down regulated by SUB885C treatment. In plasma, SUB885C caused a significant reduction of 80% neutral lipids including ChE and TG, 50% SPM, 17% LPC and 14% PE. Three out of 5 significantly changed plasma PC (14%) by SUB885C was down-regulated (Table 1). In liver, SUB885C caused a significantly reduction of 17% ChE and 3% TG, but no obvious changes in phospholipids. Thus SUB885C induced lipid changes were more prominent in plasma (50%, 70 out of 140, changed lipids) than in liver (2%, 3 out of 137, changed lipids). It affected the lipid classes of TG, ChE and SPM the most, which was in line with PCA results.

Compared with the results of rimonabant induced lipid changes from the previous publication [34], twenty four plasma lipids including LPC (18∶2), PC (38∶2), SPM (16∶0), ChE (18∶1), TG (46∶0), TG (46∶1), TG (48∶0), TG (48∶1), TG (48∶2), TG (50∶0), TG (50∶1), TG (52∶2), TG (52∶6), TG (54∶2), TG (54∶3), TG (54∶4) TG (56∶3), TG (56∶4), TG (56∶7), TG (58∶3), TG (58∶4), TG (58∶5), TG (58∶6), TG (58∶7) were significantly changed by both SUB885C and rimonabant, all of which were down regulated except for SPM (16∶0) that SUB885C induced a down-regulation while rimonabant a up-regulation. After MTC, only LPC (18∶2) and PC (38∶2) remained significant in both SUB885C and rimonabant groups; while all lipids except for SPM (16∶0), TG (56∶3) and TG (58∶5) still remained significant under SUB885C intervention. Two liver lipids, PE (36∶3) and TG (50∶1), were significantly changed by SUB885C and rimonabant. The former was down-regulated by both treatments while the latter was down-regulated by SUB885C but up-regulated by rimonabant. Both lipids did not hold significance after MTC in SUB885C and rimonabant treatment.

To investigate SUB885C's effect on adipolysis, the release of glycerol in the medium was measured after incubation of 3T3-L1 adipocytes with SUB885C. The positive control isoproterenol, a non-selective agonist of the β-adrenergic receptors, is known to increase the hydrolysis of TG [22], [35] and to stimulate adipolysis in 3T3-L1 adipocytes [22] (Figure 4A). When at a 500× dilution, SUB885C extract caused a 22% higher glycerol release as compared to isoproterenol at a concentration of 20 µM. When SUB885C extract was 2500× diluted, glycerol release was 20% lower than 20 µM isoproterenol, yet still higher than DMSO and Blank.

The effect of SUB885C on lipid accumulation in 3T3-L1 adipocytes was investigated during their differentiation process. As positive control the cytokine tumor necrosis factor-α (TNF-α) was used, which interferes with adipocyte differentiation [22]. SUB885C was found to dose-dependently inhibit in vitro adipogenesis (Figure 4B). At 5000× dilution there was 30% lipid inhibition in the cells as compared to the control (0.1% DMSO). At 1000× dilution there was 45% lipid inhibition and at 500× dilution there was 100% lipid inhibition in the cells. This was similar for TNF-α at a concentration of 2 ng/ml.

The appearance of 3T3-L1 cells cultured for 9 days in the presence of SUB885C extract was shown in Figure 4C. The number of nuclei did not differ much among the 0.1% DMSO control, 500× and 1000× diluted SUB885C extracts. With 500× dilution (top row) there was a decrease in the amount of nuclei, indicating toxicity induced cell loss, which was in line with the phase contrast picture of microscope with the 500× dilution (bottom row). The other dilutions did not show a clear effect on the cell morphology. TNF-α, as a control, did not show any effect on nuclei numbers and a densely packed cell layer. Inhibition of lipid accumulation in 3T3-L1 cells after exposure to SUB885C extract could be clearly seen at 1000× and 500× dilutions.