As shown in Figure 2, the mono- and poly- unsaturated fatty acids, TNF-α and their combinations were found to be non-toxic. However, the saturated fatty acid palmitate was toxic to the cells and TNF-α increased this toxicity. Because the release of LDH is a late event in the process of cell death, we also measured the activation of caspase-3. Exposure to palmitate caused a significant elevation in the caspase-3 activity (Figure 3). Exposure to TNF-α significantly increased the caspase-3 activation only in the palmitate-treated cells, but not in the cells treated with control medium or oleate. These results indicated significant interactive effects of palmitate and TNF-α on the process of cell death. Finally, we also found that cells exposed to palmitate were also TUNEL positive (Figure 4), further corroborating with the evidence of initiation of apoptosis in these cells. Because palmitate was found to be the primary factor in the toxicity and the effects of TNF-α were only secondary to that of palmitate, we further investigated the mechanisms of toxicity of this FFA. To evaluate the metabolic changes responsible for mediating the toxicity, the rates of uptake/release of 27 metabolites, of glucose, fatty acid and amino acid metabolism, were measured and shown in additional data file 1. FDA was applied to identify the metabolic changes responsible for separating the phenotypic response (cytotoxic versus nontoxic as defined by the level of lactate dehydrogenase (LDH) release). As shown in Figure 5, the palmitate-treated samples were separated from the rest in a space defined by the first FDA projection (T1) with the palmitate samples taking on negative values while the rest of the samples taking on positive values on the Y axis. Furthermore, as shown in Figure 6 the production of BOH and acetoacetate, with the most negative coefficients, separated the palmitate samples in the direction of toxicity, while the uptake of cysteine, ornithine and phenylalanine, with the most positive coefficients, separated the samples away from the toxicity. In addition, the palmitate-treated samples were also separated from the rest of the sample in the space defined by the second FDA projection (T2) with the palmitate samples having smaller values than the rest of the samples as shown in Figure 7. Correspondingly, we found that TG and ornithine with the most positive coefficients in this projection separated the palmitate samples away from the toxicity as shown in Figure 8. Acetoacetate and BOH are the products of fatty acid oxidation. Increased fatty acid oxidation is known to increase oxidative stress and cell death 2728. TG accumulation has been found to protect cells from palmitate induced lipoxicity 29. Thus, FDA analysis confirmed that fatty acid oxidation is positively associated with the palmitate-induced cytotoxicity, while accumulation of TG is negatively associated to cytotoxicity.

37 gene sets involved in a variety of cellular processes and organelles which are known to play important roles in the changes caused by FFA and TNF-α were evaluated to identify which of these processes were associated with palmitate-induced cytotoxicity. The gene sets evaluated included those of fatty acid metabolism, cell death, TNF-α signaling, etc., (see Table 1 for a full list). 14 of the 37 gene sets were significantly enriched with nominal p values less than 0.05, and shown in Table 2. The enriched pathways included electron transport chain (ETC), fatty acid metabolism, glycolysis, oxidative phosphorylation, ROS, the pentose phosphate pathway (PPP), cell death, fatty acid beta-oxidation, TCA cycle, fructose metabolism, glutathione, and the ERK 1, ERK 2, MAP kinase pathways. This suggested that the toxicity may be associated with changes in oxidative stress related pathways such as ROS and glutathione, as well as energy generating processes of glycolysis, oxidative phosphorylation and ETC. Importantly, the gene set of sphingolipid (ceramide) metabolism was not selected, indicating that ceramide metabolism may not play an important role in the observed toxicity. Notably, most of the enriched gene sets belong to the mitochondria, for example fatty acid beta-oxidation, electron transport chain and oxidative phosphorylation, depicting a central role of mitochondria in the cytotoxicity. The gene set of mitochondria, which included 229 genes, was also significantly altered in the palmitate-treated cells, with a nominal p value of 0.00189.

To identify the genes that regulate the metabolic functions most closely associated with cytotoxicity, MBPLS models were developed to predict the metabolic processes based upon the expression data of the gene sets identified by GSEA. Ketogenesis and TG accumulation were identified to be positively and negatively related to LDH release in step 2, respectively. Therefore, two MBPLS models were developed to model BOH and TG, respectively. The MBPLS models contained 14 blocks each, corresponding to the 14 enriched gene sets identified by the GSEA. The importance of individual genes within these functional groups was identified by evaluating the regression coefficients of the genes. In particular, the genes with high positive regression coefficients to ketogenesis and TG accumulation are discussed below as the predicted roles of these genes were evaluated using inhibitors or RNA interference (RNAi).

Increased production of ketone bodies, such as BOH and acetoacetate, has been shown to be associated with oxidative stress 2728 and an important role of oxidative stress has been identified in the palmitate toxicity 18. Therefore, genes with large (positive or negative) regression coefficients to ketone body release could suggest pathways and genes relevant to the cytotoxicity. The genes with the largest positive regression coefficients to B-OH are listed in Table 3. The complete list of regression coefficients can be found in the additional data file 2. Genes relevant to ROS production, fatty acid metabolism, and detoxification of lipid peroxidation products were found to have high regression coefficients. GSTM5 was most positively related to the BOH (and in turn, the cytotoxicity), while other isoforms of GST, e.g., GSTM4 and GSTM1 (listed in additional file 2), had negative regression coefficients. The basal level of GSTM5 in the liver is much lower than those of GSTM1 and GSTM4 30. Therefore, in spite of a large positive coefficient of GSTM5, the combined negative effect of GSTM1 and GSTM4 on the cytotoxicity is expected to dominate. Similarly, aldehyde dehydrogenase 1 (ALDH1A1) was found to have a positive relation to cytotoxicity, while ALDH1A3 (listed in additional file 2) had a negative regression coefficient. We found that the basal levels of ALDH1A3 were about 10^-5-times lower than that of ALDH1A1 (Table 4). Therefore, the cytotoxic effects of ALDH1A1 would be greater than the cytoprotective effects of ALDH1A3. Indeed, over-expression of ALDH1A3 had no significant effect on the cytotoxicity or the caspase-3 activation by palmitate (not shown). Two NADH dehydrogenases had highly positive regression coefficients, indicating a positive role of these genes in the cytotoxicity. Their combined coefficient was greater than any other gene, indicating that these genes may play very important roles in the toxicity. NADH dehydrogenase complex, or complex I of the electron transport chain, is a main site of superoxide production 3031. In the course of electron transfer some of the activated oxygen is released as superoxides or H₂O₂ 31. Their selection suggested involvement of ROS in the toxicity. Indeed, elevated ROS production was observed in cells treated with palmitate (see section 4 b ii).

Intracellular TG was identified to have significantly negative association to the toxicity. In agreement with this, the channeling of palmitate to TG has been shown to protect the cells from its toxic effects 29. Thus, genes with high positive regression coefficients to intracellular TG accumulation may be potential targets for cytoprotection. Such genes are listed in Table 5. A complete list of regression coefficients is available in the additional data file 3. Endothelial membrane protein 3 (EMP3) was the gene with greatest positive coefficient, indicating that a reduction in the expression of this gene would be cytotoxic and vice-versa. The function of EMP3 is not clearly understood, it is known to be involved in cell growth, differentiation, and apoptosis 3233. Many genes involved in glutathione metabolism had positive regression coefficients. Examples of genes related to glutathione that were selected included glutamate-cysteine ligase, catalytic subunit (GCL-c), glutathione S-transferase M4 (GSTM4), microsomal glutathione S-transferase 1 (MGST1), glutathione S-transferase A2 (GSTA2) and glutathione reductase (GSR). The selection of these genes indicated a cytoprotective role of these genes, and corroborated with the role of oxidative stress in the toxicity. Finally, some genes involved in regulation of glycolysis such as hexokinase 1 (HK1) and glucokinase (hexokinase 4) regulatory protein (GCKR) had large positive regression coefficients, indicating that upregulation of glycolysis may be cytoprotective. Exposure to FFAs is associated with a reduction in cellular energetics, e.g., a decrease in glycolytic enzymes 3435 and a reduction in ATP synthesis due to mitochondrial uncoupling 36. This reduction in cellular ATP levels may play a role in the toxicity 37 or exacerbate it. Under these conditions, increasing glycolysis may provide an alternative route for ATP synthesis and reduce the toxicity.

Hepatocytes are resistant to the cytotoxic effects of TNF-α due to rapid upregulation of cytoprotective genes, mediated in part by the activation of NF-κB by TNF-α. It is for this reason that most of the previous studies on the cell death caused by TNF-α employ a secondary insult, such as transcriptional inhibition or glutathione depletion 415. The observation of the dependence of TNF-α toxicity on that of palmitate suggests that the toxicity of palmitate can act as a secondary/additional insult. Because the saturated free fatty acid (palmitate) was found to have the greatest toxicity of all the treatments and the toxicity of TNF-α depended on the effect of palmitate, the subsequent validations were conducted for the palmitate conditions.

HepG2/C3A cells and Fetal Bovine Serum were purchased from American Type Culture Collection (ATCC, Manassas, VA). Dulbecco's modified Eagle's medium with high glucose and no pyruvate (DMEM), Penicillin-Streptomycin (P/S), phosphate buffered saline (PBS, pH 7.4) and trizol reagent were purchased from Invitrogen (Carlsbad, CA). Fatty acid free bovine serum albumin (BSA) was purchased from MP Biomedicals (Chillicothe, OH). Sodium salts of all the fatty acids (palmitate, oleate and linoleate) were purchased from Sigma Aldrich chemical company (St. Louis, MO). 6-carboxy-2',7'-dichlorodihydrofluorescein diacetate, di(acetoxymethyl ester) (DCFDA dye) was obtained from Molecular Probes (Eugene, OR). Recombinant human TNF-α was from Peprotech (Rocky Hill, NJ).

One million HepG2/C3A cells were seeded into each well of a 6-well culture plate. The cells were cultured in 2 ml of medium containing DMEM supplemented with 10% fetal bovine serum (FBS) and 2% Penicillin-streptomycin (P/S). Cells were incubated at 37°C and in 10% CO₂ atmosphere. After the cells reached confluence, the medium was replaced with 2 ml of the chosen medium, either HepG2, or the FFA medium containing 0.7 mM palmitate, oleate or linoleate; or the FFA-TNF-α medium. The fatty acids chosen are the most prevalent in the class of saturated (palmitate), monounsaturated (oleate) and polyunsaturated (linoleate) fatty acids in the plasma. The concentration of fatty acids chosen (0.7 mM) is commonly found in conditions of obesity. The FFAs were dissolved in 4% fatty acid-free BSA. Therefore, in addition to the HepG2 medium control, 4% fatty acid-free BSA in HepG2 medium was used as another control. TNF-α was added from a 100ug/ml stock in deionized water to make the desired final concentrations of either 20 or 100 ng/ml.

The cytotoxicity of the treatments was measured as the fraction of lactate dehydrogenase (LDH) released into the medium. LDH is a cytosolic enzyme which leaks out into the medium on exposure to toxic chemicals. Thus, the fraction of LDH released into the medium provides an accurate estimate of the cytotoxicity. For the LDH measurements, the cells were cultured in different media for 24 h and the supernatant was collected. Cells were washed with phosphate buffered saline (PBS) and kept in 1% triton-X-100 in PBS for another 24 h at 37°C. Cell lysate was then collected, vortexed for 15 seconds and centrifuged at 7000 rpm for 5 minutes. Cytotoxicity detection kit (Roche Applied Science, Indianapolis, IN) was used to measure the LDH release. LDH released was normalized to the total LDH (released + lyzed). The LDH released into the medium was normalized to total LDH, as shown in the equation below

% L D H r e l e a s e = L D H ( m e d i u m ) L D H ( t o t a l ) × 100 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqGGLaqjcqWGmbatcqWGebarcqWGibascqWGYbGCcqWGLbqzcqWGSbaBcqWGLbqzcqWGHbqycqWGZbWCcqWGLbqzcqGH9aqpdaWcaaqaaiabdYeamjabdseaejabdIeaijabcIcaOiabd2gaTjabdwgaLjabdsgaKjabdMgaPjabdwha1jabd2gaTjabcMcaPaqaaiabdYeamjabdseaejabdIeaijabcIcaOiabdsha0jabd+gaVjabdsha0jabdggaHjabdYgaSjabcMcaPaaacqGHxdaTcqaIXaqmcqaIWaamcqaIWaamaaa@596A@

Cells were treated with different FFA in the presence and absence of TNF-α. Caspase-3 activity was measured using a commercially available fluorescence-based assay kit from Biomol, according to manufacturer's protocol. Data were normalized to protein measurements from parallel experiments. For TUNEL staining, cells were cultured in Labtek Chamber Slide System. TUNEL staining was performed using the Dead-End fluorimetric TUNEL system from Promega Biosciences, according to manufacturer's protocol and imaged under a fluorescence microscope (Leica).

The net uptake or production of a species was calculated by the difference in the concentration of the specie in the medium, before and after the treatment. The concentrations of metabolites were measured using enzymatic assays or HPLC. The amino acids were measured by HPLC using the AccQTag method (Waters) according to the manufacturer's protocol. Briefly, the media collected after treatments were de-proteinized by adding 4X acetonitrile (by volume) and keeping on ice for 1 h, after which the samples were centrifuged and the supernatant was derivatized using the reagent in the AccQTag kit as per the instructions given. The samples were analyzed with HPLC on a Waters 2690 separations module, using an AccQTag column (Waters) and fluorescence detection (Waters detector). Intracellular triglycerides were measured by lysing the cells with 1% triton-X-100 for 24 h and measuring the triglycerides by an enzymatic assay kit from Sigma. Glucose, lactate and glycerol were measured using enzymatic analysis kits from Sigma. Free fatty acid half micro kit (Roche Diagnostics) was used to find the concentration of free fatty acids in the medium. Beta-hydroxybutyrate (BOH) released was measured using a kit from Stanbio. Acetoacetate was measured using enzymatic assay 42.

FDA identifies the projection axis that maximizes the ratio of the between-group and the within-group variations. Details on the FDA algorithm can be found in 23. FDA was applied to identify which among the 27 measured metabolic uptake/production contributed to the separation of the different phenotypes (cytotoxic versus nontoxic). Because there were 2 classes (toxic and non-toxic), a single discriminating vector is sufficient to separate them 43. The relative importance of the metabolites was identified by the projection of the metabolites in the new discriminant vector space.

Cells were cultured in 10 cm tissue culture plates until confluence and then exposed to different treatments for 24 h. RNA was isolated with the Trizol reagent. The gene expression profiles were obtained with the cDNA microarrays. The microarray analyses were conducted at the Van Andel Institute, Grand Rapids, MI. The protocols are available online at 44. There were two biological replicates for each condition and each replicate was measured with the Cy3 and Cy5 dyes (i.e. there were two technical replicates for each biological replicate). The microarray data has been deposited at the GEO website 45, with a query number of GSE5441.

Confluent cells were treated with the control medium or 0.7 mM palmitate for different time periods. Positive control cells were treated with 2 mM H₂O₂ for 1 h at 37°C. Measurement of ROS was performed by flow cytometry, using 6-carboxy-2',7'-dichlorodihydrofluorescein diacetate, di(acetoxymethyl ester) (C-2938, Molecular Probes, Eugene, OR) dye. A 5 mM stock solution of the dye was prepared in DMSO, and diluted to the 5 umol/L in DMEM. After the desired treatment times, the cells were exposed to the dye for 30 min, following which the cells were washed and imaged under a fluorescence microscope. Quantitative measurements were made by reading the fluorescence of the cells in a microplate fluorimeter with excitation at 488 nm and emission at 527 nm.

The cells were treated with 0.7 mM palmitate in the presence or absence of 20 μM Fumonisin B1 (FB1) for 24 h, following which the lipids were extracted by the method of Bligh and Dyer 46. The C-16 ceramide in the samples were then quantified with LC-MS as per a previously published protocol 47.

GSEA integrates a priori knowledge of a gene's functional role with the expression data to detect concerted expression changes in a set of genes responsible for producing a phenotype. The software GSEA-P, from 48, was used for the GSEA analysis. Thirty seven gene sets were selected from the molecular signature database, MsigDB 25 functional gene group c2, as shown in Table 1. These sets included 10 metabolic pathways, 26 signal pathways and 1 cellular component. An enrichment score of a gene set S characterizes whether the set of genes randomly distributed across the list or falls mainly at the bottom or top of the list. The null hypothesis that a gene set S randomly distributes across the ranked gene list was tested with Kolmogorov-Smirnov test, with the statistical significance value estimated through 1000 random permutations of the phenotype label. The gene sets with a high significance of enrichment are considered important in separating the distinct phenotypes.

MBPLS is a hierarchical multivariate analysis method 4950, where the variables are divided into different blocks based upon a priori knowledge, for example according to different stages of a process 49 or different metabolic pathways in a cell 26. For details of the MBPLS algorithm, refer 26. In the MBPLS model, the different gene sets identified by GSEA formed the blocks of the MBPLS. This ensured that the blocks were separated according to the functional role of the genes. Block scores were extracted from each block to predict the metabolic uptake/productions. This facilitated the identification of an important block (gene set) to a metabolic flux, and identified the important genes within the block. Important genes sets were identified by evaluating the weights of each block and importance of individual genes were then further identified by evaluating the regression coefficients of the genes within the block. Two latent variables were selected to be extracted from each block based upon the prediction accuracy of the MBPLS model. The N-way toolbox 51 was applied to conduct the MBPLS modeling.

Silencer^® Validated siRNAs targeting human EMP3 and ALDH1A1 mRNA were purchased from Ambion (Austin, TX). The synthesized oligonucleotides for siRNA of EMP3 are 5'-GUCCCUGAAUCUCUGGUACtt-3' and 5'-GUACCAGAGAUUCAGGGACtc-3', and the synthesized oligonucleotides for siRNA of ALDH1A1 are 5'-GGAACAGUGUGGGUGAAUUtt-3' and 5'-AAUUCACCCACACUGUUCCtg-3'. Reverse transfection of siRNA was performed. In general, siRNA and the transfection reagent, Lipofectamine RNAiMAX (Invitrogen, Carlsbad, CA), were diluted in Opti-MEM (Invitrogen) devoid of any serum and antibiotics and added to the 6-well plates. After 20 minutes, the same numbers of HepG2 cells in antibiotic free media were plated in each well and incubated at 37°C. After 24 h of transfection, the HepG2 cells were cultured for another 24 hours in regular media with other additives, for example palmitate. Then cells were harvested, washed twice with phosphate-buffered saline and lysed.

The ALDH1A3 plasmid, pCMV6-XL4-hALDH1A3, was purchased from Origene (Rockville, MD). Transient transfection was performed according to the Lipofectamine 2000 (Invitrogen) method. In general, the HepG2 cells were seeded in 6-well plates and cultured until reaching 80% confluency. Before transfection, the cells were washed twice with phosphate buffered saline, and medium was replaced with 2 ml of Opti-MEM medium. 1 μg/well of pCMV6-XL4-hALDH1A3 was then mixed with 5ul/well of Lipofectamine 2000 in Opti-MEM, and 20 minutes later, the mixture was added to the wells. After 6–12 hours of transfection, the cells were then cultured in regular media, treated with other additives like palmitate or TNF-α, and harvested after the treatment.

Total RNA was extracted from cells with an RNeasy mini kit and depleted of contaminating DNA with RNase-free DNase (Qiagen, Valencia, CA). Equal amounts of total RNA (1 μg) were reverse-transcribed using an iScript cDNA synthesis kit (Bio-RAD). The first-strand cDNA was used as a template. The primers used for quantitative RT-PCR analyses of human EMP3 (5'-GTGGTCTCAGCCCTTCACA-3' and 5'-ACGTGCAGTCGTACCAGAGA-3'), human ALDH1A1 (5'-AGCCTTCACAGGATCAACAGA-3' and 5'-GTCGGCATCAGCTAACACAA-3'), human ALDH1A3 (5'-GCCCTTTATCTCGGCTCTCT-3' and 5'-CGGTGAAGGCGATCTTGT-3') and human GAPDH (5'-AACTTTGGTATCGTGGAAGGA-3' and 5'-CAGTAGAGGCAGGGATGATGT-3') were synthesized by Operon Biotechnologies, Inc. (Huntsville, AL). RT-PCR was performed in 25-μl reactions using 1/10 of the cDNA produced by reverse transcription, 0.2 μM each primer, 1 X SYBR green supermix from Bio-RAD, and an annealing temperature of 57°C for 40 cycles. Each sample was assayed in three independent RT reactions and triplicate reactions each and normalized to GAPDH expression. Negative controls included the absence of enzyme in the RT reaction and absence of template during PCR. The cycle threshold (C_T) values corresponding to the PCR cycle number at which fluorescence emission in real time reaches a threshold above the base-line emission were determined using MyIQ™ Real-Time PCR Detection System.