Adipocyte-Induced FABP4 Expression in Ovarian Cancer Cells Promotes Metastasis and Mediates Carboplatin Resistance

Peritoneal metastasis is a characteristic feature of epithelial ovarian cancers (1), and the primary site of metastasis is the omentum, a hormonally active fat pad of 20 × 15 × 3 cm size that protects the small and large bowel (2). Although the origin of ovarian cancer is not certain, there is increasing evidence that transformed cells originating from the distal end of the fallopian tube or the surface epithelium of the ovary undergo an epithelial-to-mesenchymal transition, detach, and readily spread throughout the abdominal cavity (1, 2). We have previously shown that ovarian cancer cells preferentially home to the omentum, guided by secreted factors such as IL6, IL8, and MCP1 (3). In addition, cancer cells with an active ERBB3–NGR1 axis are capable of colonizing the omentum via intravasation from the primary site (4). Irrespective of the mode of transmission, omental colonization marks an essential step in disease progression and in certain cases might even contribute to secondary metastasis (5).

Adipocytes, the major cell type of the omentum, play an essential role in maintaining its tissue architecture and endocrine functions. Once cancer cells colonize the omentum; they initiate lipolytic signals in omental adipocytes, resulting in the release of long-chain fatty acids (3). Subsequently, the cancer cells take up the adipocyte-derived lipids through the CD36 receptor (6). It is unclear which lipid species are affected by the cancer cell–adipocyte interactions. In addition, the effect that these lipids have on cancer cell lipidome and the mechanism by which these lipids are metabolized are still not known.

FABPs are a family of intracellular lipid chaperones (7) that coordinate cellular lipid responses in cells by binding to and redistributing intracellular fatty acids (FA). FABP4 is primarily expressed in adipocytes and macrophages and contributes to obesity-induced insulin resistance (8–10). Moreover, circulating FABP4 has also been found to regulate systemic glucose levels in mice (11) and is a marker of preclinical metabolic syndrome in humans (12). Hence, it is evident that FABP4 is inherently linked to both cellular and systemic metabolic processes.

Cancer cells engage in specific bidirectional communication with the surrounding tumor stroma, which affects their metabolism in the tumor microenvironment (13). Although the metabolic consequences of fibroblast/cancer cell interactions have been studied (14–16), less is known about how the interaction of cancer cells with adipocytes affects global metabolism in cancer cells. This knowledge is important for understanding ovarian cancer metastasis because cancer cells in all patients with advanced or recurrent disease invade adipose-rich tissues such as the omentum, the bowel mesentery, and the fat appendages (“appendices epiploicae”) along the large bowel.

We have characterized the altered proteome and metabolome of ovarian cancer cells following adipocyte coculture and report that FABP4-regulated metabolic changes are essential for cancer cells to adapt to lipid-rich microenvironments. Furthermore, targeting FABP4 in cancer cells reduced omental metastasis and sensitized tumor cells to platinum chemotherapy.

SKOV3ip1 and HeyA8 cells were a gift from Dr. Gordon Mills, and 293T cells were received from Dr. L. Godley (University of Chicago). OVCAR5, OVCAR8, and CAOV3 cells were purchased from ATCC. These cell lines were grown in DMEM supplemented with 10% fetal bovine serum (FBS), 100 U/mL penicillin, 100 μg/mL streptomycin, and 1% nonessential amino acids and vitamins (NEAA and Vit). PE01 and PE04 cells (obtained from Scott Kauffman, Mayo Clinic, Rochester, MN) were maintained in DMEM and were grown according to recommended culture conditions. ID8 cells (obtained from Dr. Katherine Roby, University of Kansas Medical Center, Kansas City, KS) were maintained as described in ref. 3. The cell lines were regularly tested for Mycoplasma and genotyped (IDEXX Bioresearch short tandem repeat marker profiling) to determine authenticity.

A xenograft experiment (Fig. 5C) was conducted using female immunocompromised athymic nude mice (Envigo), and syngeneic experiments were carried out using female C57/Bl6 mice (Envigo). OVCAR8-derived control clustered regularly interspaced short palindromic repeats (CRISPR) and FABP4 CRISPR cells were injected intraperitoneally (5 million cells/mouse) and tumor burden assessed after 44 days (10 mice/group). ID8 cells were injected intraperitoneally (5 million cells/mouse) or intrabursally (1 million cells/mouse), and tumor burden was assessed after 66 or 88 days, respectively. Carboplatin treatments (administered intraperitoneally) were carried out once a week for 4 weeks at a concentration of 30 mg/kg, and FABP4 inhibitor (BMS309403) was delivered orally 6 days a week for 4 weeks at a concentration of 20 mg/kg. BMS309403 compound was dissolved using the vehicle (5% ethanol, 10% 1-methyl-2-pyrrolidone, 5% cremophor EL, and 80% water) for oral gavage. Tumor burden was determined by counting metastatic lesions, omental weight, and total metastatic tumor weight.

Freshly isolated human omental adipocytes were cocultured with SKOV3ip1 cells for 4 and 18 hours (1:5 PCV, adipocytes: SFM). For the 4-hour time point, adipocytes were isolated from 7 donors. For the 18-hour time point, adipocytes from 6 donors were used for coculture with SKOV3ip1. After coculture, adipocytes were removed, and cancer cells were washed with phosphate-buffered saline (PBS) several times to remove any remaining adipocytes. Cancer cells were then trypsinized, and an equal number of cells were frozen for metabolomic analysis at Metabolon. Metabolites were extracted as per established protocols at Metabolon, and metabolite identification was carried out using a Thermo-Finnigan Trace DSQ fast-scanning single-quadrupole gas chromatography/mass spectrometer (GC/MS) and Waters ACQUITY UPLC and a Thermo-Finnigan LTQ liquid chromatography/mass spectrometry (LC/MS) setup. A paired t test was carried out to determine statistical significance and altered metabolites between experimental groups with a P < 0.05 and q < 0.05 were considered significant. Metabolomics on FABP4 shRNA cells was carried out by coculturing cells with omental adipocytes isolated from 7 donors. Samples were isolated as described above, and untargeted analysis was carried out using both gas chromatography (GC)/MS and LC/MS platforms at the West Coast Metabolomics Center (17, 18). Statistically significant metabolites (P < 0.05) altered between experimental groups were identified using repeated-measures t test, and the biochemical similarities between them were depicted as a network map using Cytoscape 2.0.

After coculture with adipocytes for 4 hours (as described above), cancer cells were trypsinized and frozen. Sample preparation and MS were carried out as described by us (19, 20). Shotgun proteomics was carried out using a Q Exactive (Thermo Fisher Scientific) connected to EASY-nLC 1000 HPLC system (Thermo Fisher Scientific). Columns (75-mm inner diameter, 50-cm length) were in-house packed with 1.9-μm C18 particles (Dr. Maisch, GmbH, Germany). Peptides were separated over a 250-minute gradient from 2% to 60% (5 minutes to 5%, 180 minutes to 25%, 45 minutes to 35%, 20 minutes to 60%) in buffer B (80% acetonitrile, 0.5% formic acid) at 200 nL/minute. The 5 most intense precursors were subjected to high-energy collisional dissociation fragmentation at a normalized collision energy of 25, an isolation window of 2.2 Th, and a resolution of 17,500 at m/z 200. For survey scans, ion injections times were set to 20 ms (target value 3E6) and 120 ms (target value 1E5) for MS/MS scans. Raw data were analyzed using MaxQuant software (version 1.5.0.38) and statistical analysis was performed using Perseus (20, 21). A t test was carried out to determine statistical significance and a P value of < 0.05 and false discovery rate q < 0.05 was considered significant.

Microarray analysis was performed at the University of Chicago Functional Genomics Core Facility as described previously (6), using Illumina HumanHT-12_V4 Expression BeadChip microarray (Illumina). Total RNA was isolated from SKOV3ip1 cells transduced with control shRNA and FABP4 shRNA, using an RNeasy kit (Qiagen), after column DNase treatment (RNase Free DNase set; Qiagen). RNA quality and integrity were analyzed using Agilent 2100 Bioanalyzer (Agilent Technologies, Inc.), and biotinylated cRNA was generated using Illumina Total Prep RNA Amplification Kit (Ambion, Inc.). Following hybridization, arrays were scanned on an Illumina's BeadStation 500GX Genetic Analysis System, and data were extracted using GenomeStudio Software (v2011.1, Illumina). Three replicates were analyzed per experimental groups. Differentially expressed genes (DEG), with a q value of < 0.05 and fold-change value > 1.5, were used for Ingenuity Pathway Analysis (IPA) analysis to identify altered canonical pathways and explore the impact of gene networks on cellular processes.

Lipid peroxidation was estimated by quantitating malondialdehyde levels in cancer cells using the Thiobarbituric Acid assay kit (Cayman; ref. 6). Cancer cells cocultured with adipocytes (1:5 PCV adipocytes: SFM) were sonicated and boiled for 10 minutes following the addition of sodium dodecyl sulfate (SDS) and TBARS color reagent. Samples were transferred onto a black-walled 96-well plate, and fluorescence was read at excitation 530 nm and emission 550 nm using Spectra max i3 (Molecular Probes).

Cells were treated with adipocyte-conditioned media (Adi CM) for 24 hours, following which, they were stained using BODIPY 493/503 (5 μg/mL). Cells were subsequently fixed using 4% paraformaldehyde and imaged using a Zeiss 510 confocal microscope (6).

Fatty acid β-oxidation in cancer cells was determined as described previously (3, 6). Briefly, cancer cells were cocultured with adipocytes (1:4 packed cell volume, adipocyte: media) for 24 hours. Adipocytes were removed, and cancer cells were washed with PBS. The cancer cells were then incubated with bovine serum albumin-conjugated [³H]-palmitate for 3 hours, following which, supernatant was collected and ³H₂O eluted out using Dowex-1 × 8 ion exchange resin (Thermo Fisher Scientific). Samples were then mixed with 5-mL scintillation fluid (Scintiverse, Thermo Fisher), and radioactivity was determined as counts per minute.

Cellular glycolytic rate and mitochondrial oxidation of glucose were determined as extracellular acidification rate (ECAR) and oxygen consumption rate (OCR), respectively, using Seahorse XF^e analyzer (Agilent; ref. 16). Cancer cells (with stable knockdown or overexpression of FABP4) were plated in 96-well Seahorse plates (10,000 cells/well) and allowed to attach. For ECAR analysis, media were removed from cells and replaced with DMEM (D5030, Sigma-Aldrich) supplemented with glutamine (2 mmol/L), pH adjusted to 7.4. After incubation of the cells in a non-CO₂ incubator for 1 hour, ECAR was determined following sequential injection of glucose (10 mmol/L, Sigma-Aldrich), oligomycin (2 μmol/L, Sigma-Aldrich), and 2-deoxyglucose (50 mmol/L, Sigma-Aldrich). For OCR analysis, media were replaced with Seahorse DMEM base media (Agilent), supplemented with sodium pyruvate (1 mmol/L), glucose (25 mmol/L, Sigma-Aldrich), glutamine (2 mmol/L), and pH adjusted to 7.4. OCR was measured after sequential injection with oligomycin (1 μmol/L, Sigma-Aldrich), FCCP (1 μmol/L, Sigma-Aldrich), Antimycin A (1 μmol/L, Sigma-Aldrich), and Rotenone (1 μmol/L, Sigma-Aldrich). Background corrected rate data were plotted to determine ECAR and OCR changes associated with FABP4.

Genomic DNA (5 μg) isolated using Wizard Genomic DNA isolation kit (Promega) was spotted on positively charged nylon membrane (Amersham, GE Life Sciences) using a 96-well dot-blot apparatus (Bio-Rad; refs. 20, 22). DNA was cross-linked using Stratagene stratalinker 1800 and standard immunoblotting carried out using 1:1,000 dilution of anti 5-hydroxymethylcytosine antibody (A-1018, Epigentek). Total DNA was stained by using 0.04% methylene blue in 0.5 M acetic acid solution and image using G: Box XT4 (Syngene).

Omental biopsies were made using 10-mm punch biopsies (0.3 g), 4 per treatment group (DMSO or BMS309403) and placed in ultralow attachment 24-well plates (23). One million GFP-labeled HeyA8 and SKOv3ip1 cells were then added to each biopsy, and the growth of cancer cells was imaged using Nikon eclipse Ti2 microscope using a 4× objective after 72 hours culture. Cells were released from the tissues using Trypsin, and fluorescence intensity of total cells isolate was measured using Spectra max i3 (Molecular Devices).

All patient samples were collected at the University of Chicago Medical Center following approval from the Institutional Review Board, and after written informed consent from the patients. All animal experiments were carried out in accordance with approved protocols by the Institutional Animal Care and Use Committee, at the University of Chicago.

GraphPad Prism 7 software was used to calculate statistical significance, unless otherwise stated. Data are presented as mean ± SEM, and a two-tailed Student t test was used to determine statistical significance, and a value less than 0.05 was considered significant. Perseus software (21) was used to identify proteins with significantly altered expression. A t test was carried out to determine statistical significance, and a P value of < 0.05 and false discovery rate q < 0.05 was considered significant. For metabolomic analyses, a paired t test was carried out, and metabolites with a P value of < 0.5 were considered significantly changed.

We carried out global untargeted metabolomics to understand the effect of human primary omental adipocytes (HPA) on the ovarian cancer cell metabolome. The ovarian cancer cells were cocultured with HPA for 4 and 18 hours to determine early and late metabolic alterations, respectively. Cancer cells were separated from adipocytes, and the extracted metabolites were analyzed for alterations in the metabolome using GC and LC/MS. Several metabolite classes, including lipids, amino acids, carbohydrates, and cofactors were altered in cancer cells cocultured with HPA (Supplementary Fig. S1A and S1B). Cancer cells were initially found to accumulate amino acids (such as serine, asparagine, and histamine), but these levels were reduced with longer coculture (18 hours). There was also a consistent reduction in the amount of metabolic intermediates of branched-chain amino acids with adipocyte coculture. Interestingly, metabolomics analysis revealed reduced glycogen synthesis (reduced glucose-1-phosphate) consistent with what we reported with CAF coculture (16) and increased utilization of the nonoxidative pentose phosphate pathway as indicated by initial increase (4 hours) and later consumption of sedoheptulose-7-phosphate (18 hours; Supplementary Fig. S1A and S1B).

The transfer of lipids from adipocytes to cancer cells is emerging as a hallmark of tumors that thrive in a lipid-rich tumor environment (3, 24–27). However, the specific alterations of different lipid species are unknown. Our study of the lipidome in the presence of HPA showed that cancer cells accumulate several species of short- and long-chain FA, with linoleate (18:2) being the most abundant (Fig. 1A) and consistently elevated species at both 4- and 18-hour time points (Supplementary Fig. S1A and S1B). The cocultured cancer cells also exhibited an oxidative stress signature characterized by increases in hydroxy species of palmitate, stearate, cholesterol, and oxylipins such as 13-HODE and 9-HODE (Fig. 1B). Consistent with these findings, adipocyte coculture of ovarian cancer cells induced intracellular reactive oxygen species (ROS) and the lipid peroxidation product malondialdehyde in a time-dependent manner (Fig. 1C and D). This increase in lipid peroxidation products observed in cell culture was confirmed in human omental metastases from 4 patients with high-grade serous ovarian cancer (HGSOC). Cancer cells adjacent to omental adipocytes showed increased expression of anti-4-hydroxynonenal adducts, which is a surrogate of lipid peroxidation (Fig. 1E).

To identify the proteins regulating the metabolic changes seen in cancer cells upon adipocyte coculture, we carried out unbiased mass spectrometry–based proteomic profiling with identical culture conditions. Because intracellular lipid changes were detected at an early time point, we cocultured SKOV3ip1 cells with adipocytes for 4 hours and profiled the proteome. Using a single-shot label-free strategy (19), we were able to identify more than 6,700 proteins per sample with a similar number of proteins identified in cancer cells alone or when cocultured with adipocytes (Supplementary Fig. S2A). After 4-hour coculture, 16 proteins were significantly altered in cancer cells (Supplementary Table 1). Among them, lipid metabolism proteins, FABP4, CD36, and alcohol dehydrogenase (ADH) 1 proteins were increased upon coculture (Fig. 2A). ADH1B regulates alcohol metabolism and is of prognostic significance for residual tumor disease in ovarian cancer (28). CD36, a fatty acid receptor (29), and FABP4, a lipid chaperone protein, are both regulated by PPARγ (30) and are important for fatty acid uptake, compartmentalization, and metabolism (31). Interestingly, using the The Cancer Genome Atlas (TCGA) ovarian cancer data set (32), we identified a subgroup of patients expressing all 3 proteins with a strong correlation (Supplementary Fig. S2B and S2C). Although these proteins are expressed in the normal omental tissue (Supplementary Fig. S2D), immunohistochemistry on sections of human omental HGSOC metastasis samples showed elevated expression of ADH1B, CD36, and FABP4 in the epithelial tumor compartment (Fig. 2B). In addition, using a publicly available HGSOC data set (33), we found significant upregulation of CD36, FABP4, and ADH1b proteins in peritoneal metastatic tumors when compared with the primary tumor (Supplementary Fig. S2E).

Given that the principal finding of the global metabolomics analysis was lipidomic changes, we focused further on CD36 and FABP4. Validating our proteomic data by Western blotting showed increased expression of CD36 and FABP4 proteins in SKOV3ip1 cells upon adipocyte coculture (Fig. 2C). We also found increased expression of FABP4 with adipocyte coculture in a panel of cell lines representative of HGSOC (Supplementary Fig. S2F). Moreover, knockdown of CD36 was found to reduce adipocyte-induced FABP4 expression (Fig. 2D), whereas knockdown of FABP4 did not affect adipocyte-mediated CD36 expression (Fig. 2E). Our data thus suggest a functional and/or regulatory link between these 2 proteins, with CD36 being an upstream regulator of FABP4.

Having recently shown that CD36 plays a role in adipocyte mediated lipid uptake in cancer cells (6), we expanded these findings by examining the role of the lipid chaperone protein FABP4 in intracellular lipid metabolism (7, 31, 34). Stable knockdown of FABP4 in ovarian cancer cells cocultured with adipocytes followed by untargeted metabolomics revealed major changes in the cancer cell lipidome (Fig. 3A). Loss of FABP4 reduced the capacity of cancer cells to accumulate neutral lipids such as triacylglycerol and cholesterol esters as manifested by a reduction in lipid droplets (Supplementary Fig. S3A–S3C; refs. 7, 25, 31, 34). There was also a marked reduction in various ceramides species and increases in glucosylceramide and sphingomyelin species (Supplementary Fig. S3D–S3F). Moreover, phosphatidylcholine species also increased in cells with the loss of FABP4 (Supplementary Fig. S3G). Because neutral lipids are oxidized by cancer cells (3, 26), we measured fatty acid oxidation and found that loss of FABP4 reduced adipocyte-induced but not constitutive β-oxidation in cancer cells (Fig. 3B). FABP4 inhibition also blocked adipocyte-induced ROS generation (Fig. 3C) and lipid peroxidation (Fig. 3D).

Because β-oxidation and ROS are known drivers of cell proliferation, we next analyzed their role in adipocyte-induced cell proliferation. Using N-acetyl cysteine (a ROS chelator) and etomoxir (an inhibitor of CPT1 and β-oxidation), we found that both β-oxidation and ROS contribute to adipocyte-induced cell proliferation (Fig. 3E).

The metabolomic data (from Fig. 3A) indicated that the TCA cycle metabolites regulated by FABP4 in the presence of adipocytes are citrate, isocitrate, α-ketoglutarate, and malate (Fig. 4A). No significant change was detected in intracellular glutamine/glutamate levels (Fig. 4A), thereby pointing to a role for FABP4 in glucose oxidation. Indeed, FABP4 overexpression (Supplementary Fig. S4A) increased mitochondrial oxidation with no change observed in glycolytic rate (Fig. 4B and C). In the opposite experiment, FABP4 inhibition reduced mitochondrial oxidation with a concomitant increase in glycolytic rate (Fig. 4D and E). FABP4 inhibition led to an increase in glycolysis-derived ATP production and a decrease in mitochondrial ATP production (Fig. 4F). The total ATP production, however, remained unchanged (Fig. 4F). This suggests for a compensatory increase in glycolysis when FABP4 is inhibited.

Because α-ketoglutarate was the most strongly induced TCA cycle (2.7-fold) intermediate in FABP4 knockdown cells, we focused on determining its effect on cancer cells. α-Ketoglutarate is a cosubstrate for ten-eleven translocase (TET) enzymes, which oxidizes DNA 5-methylcytosines to 5-hydroxymethylcytosines (5-hmc; ref. 35), thereby regulating DNA methylation and reducing cell proliferation (36, 37). Using dimethyl-2oxoglutarate (dm-2OG) a cell-permeable analogue of α-ketoglutarate, we observed increased DNA 5-hmc levels, accompanied by G₁ arrest (Supplementary Fig. S4B and S4C). Consistent with these results, FABP4 knockdown SKOV3ip1 cells cocultured with adipocytes exhibited increased levels of 5-hmc DNA modifications with a greater number of cells in the G₁ phase of the cell cycle (Fig. 4G and H). Similarly, inhibition of FABP4 using a small-molecule inhibitor (BMS309403) led to an increase in 5-hmc levels in CAOV3 cells cocultured with adipocytes (Supplementary Fig. S4D), whereas no increase was seen in these cells with adipocyte coculture alone.

To determine the effect of FABP4 inhibition on gene-expression and identify cellular processes regulated by FABP4, we carried out microarray analyses on RNA isolated from stably transduced shRNA FABP4 cells. Inhibition of FABP4 changed the expression of 189 genes, of these 116 were upregulated and 73 genes were downregulated. The top 20 genes with altered expression are listed (Supplementary Table S2). The adipocyte-induced increase in IL6 and decrease in HES2 mRNA levels in cancer cells (6) were abrogated upon FABP4 inhibition (Supplementary Table S2), suggesting that FABP4 is the upstream regulator of multiple adipocyte-induced genes in ovarian cancer cells. IPA showed that knockdown of FABP4 led to inhibition of several tumorigenic pathways including proliferative, migration, and invasion regulatory networks, and subsequently reduced signaling through protumorigenic pathways (Fig. 5A; Supplementary Fig. S5A; Supplementary Table S3). Consistent with the IPA analysis, shFABP4 cells showed reduced colony-forming capacity in clonogenic assays (Fig. 5B).

The growth-sustaining effect of FABP4 led to the loss of shRNA knockdown in vivo, making long-term in vivo studies challenging. Therefore, to pursue in vivo studies, we generated CRISPR-mediated knockout of FABP4 in serous OVCAR8 cells (38). Using single-guide RNA targeting FABP4 exon 3, we generated FABP4 knockout clones with heterozygous deletion of FABP4, leading to significant loss of FABP4 protein (Supplementary Fig. S5B and S5C). Animals injected intraperitoneally with FABP4 CRISPR cell clones developed significantly fewer metastatic implants (n = 51.8 vs. n = 194.7) and smaller tumors (102 mg vs. 491 mg) as compared with control CRISPR cells (Fig. 5C). Notably, there was a significant reduction of tumor burden in the omentum (129 mg vs. 239 mg; Supplementary Fig. S5D), and the overall size of the omental tumor mass was significantly decreased (Supplementary Fig. S5E). Consistent with the shRNA-mediated knockdown of FABP4 (Fig. 4H), we observed increases of 5-hmc levels in the DNA of FABP4 CRISPR cells treated with adipocyte-conditioned media (Supplementary Fig. S5F). Corroborating the in vitro data, omental tumors generated by FABP4 CRISPR cells had stronger 5-hmc staining (Fig. 5D).

Given the effects of FABP4 inhibition using shRNA and CRISPR on ovarian cancer proliferation and metastasis, we sought to determine whether we could reproduce these results with a small-molecule inhibitor. For this purpose, we used BMS309403, an orally active small-molecule tool compound against FABP4 that interacts with the FABP4 lipid binding pocket and inhibits atherosclerosis and type 2 diabetes in rodent models (39, 40).

Because molecular inhibition of FABP4 leads to cell-cycle arrest, first we sought to determine whether BMS309403 affected the cell-cycle. Flow cytometry analysis of propidium iodide–stained cancer cells showed that BMS309403 treatment also resulted in a greater number of cells in the G₁ phase, with a concomitant reduction of cells in the S-phase (Fig. 6A). The addition of BMS309403 to full thickness human omental explants ex vivo, significantly reduced the colonization of GFP-labeled HeyA8 and SKOV3ip1 cells (Fig. 6B). Next, we performed orthotopic injection of mouse ID8 ovarian cancer cells into the bursa of immunocompetent mice and allowed tumors to establish for 60 days, followed by oral gavage with BMS309403 for 28 days. In this intervention study, there was a significant reduction in metastasis with BMS309403 treatment, as evidenced by fewer (n = 98 vs. n = 169) and smaller (252 mg vs. 483 mg) metastatic nodules (Fig. 6C), but no difference was observed in primary ovarian tumor burden (Supplementary Fig. S6). Histologic examination revealed that the normal omental tissue architecture was partially restored following treatment with BMS309403. This was manifested both by decreased tumor volume and by interspersed islands of histologically normal-appearing adipose tissue. In contrast, these islands were completely lacking in the vehicle-treated groups because the tumor had completely replaced the omentum (Fig. 6D).

Because single-agent BMS treatment reduced ovarian cancer metastasis, we explored the efficacy of BMS309403 in combination with carboplatin, the most effective chemotherapy drug against epithelial ovarian cancer (2). PE01 and PE04 are carboplatin-sensitive and resistant cell line pairs isolated from the same patient before and after recurrent disease following platinum-based chemotherapy (41). The sensitivity of the resistant subclone (PE04) to carboplatin was considerably increased with the addition of the BMS309403 compound (49.8 vs. 87.0 μmol/L; Fig. 7A). The specific contribution of FABP4 toward platinum response was illustrated by the fact that FABP4 shRNA-transduced cells had a much lower IC₅₀ for carboplatin than the control shRNA-transduced cells (26.2 vs. 69.5 μmol/L; Fig. 7B). This was confirmed in 2 additional ovarian cancer cell lines that also had reduced survival with both treatments (Fig. 7C).

Next, we examined the metastatic tumor burden in a syngeneic xenograft mouse model treated with either a single-agent drug (carboplatin or BMS) or with the drugs in combination. We injected mouse ID8 ovarian cancer cells intraperitoneally and allowed 40 days for tumor growth, and then initiated drug treatment for 26 days (“intervention study”; ref. 42). BMS309403 and carboplatin together had a more profound effect than that observed with single-agent treatments (Fig. 7D). The combination treatment resulted in a reduction in both the number and weight of metastatic nodules (Fig. 7D). Furthermore, combination treatment restored to a greater extent the normal omental tissue architecture and resulted in markedly smaller tumors (Fig. 7E).

High-grade serous cancer primarily metastasizes to adipose-rich areas such as the omentum, the bowel mesentery, and appendices epiploicae on the colon. During metastasis, adipocytes in the microenvironment are recruited by cancer cells and transform into cancer-associated adipocytes. This term, which conveys an evolving symbiotic relationship between adipocytes and cancer cells, was coined by the Muller laboratory in 2010 (43, 44). Over the last decade, we and others (3, 25, 45) have shown that when cancer cells from different organs are cocultured with adipocytes, lipids are transferred from the adipocytes to the cancer cells. These lipids are then stored as long-chain FA in lipid droplets and used for energy production through β-oxidation.

In our comprehensive analysis of changes in the lipid composition of cancer cells cultured with primary human visceral adipocytes, we detected over 315 individual, significantly altered metabolites using both gas and liquid chromatography MS platforms. This systematic metabolomics analysis showed that the unsaturated long-chain fatty acid linoleate (18:2) was the most abundant fatty acid in cancer cells following adipocyte coculture. Linoleic acid is a precursor for arachidonic acid and prostaglandins, which regulate inflammation, and is associated with an increased risk of developing ovarian cancer (46). In contrast, breast cancer cells cocultured with a differentiated murine preadipocyte cell line show strong induction of saturated FA (palmitate16:0, stearate 18:0; ref. 25). One possible explanation for this difference is that the predominant oncogene mutation (e.g., myc, ras, and p53), which will vary by tumor type, will affect adipocyte–cancer cell metabolism. Second, the metabolism of visceral, subcutaneous, and bone marrow adipocytes differs significantly (26, 47) and will affect their interaction with cancer cells. Because visceral adipocytes are in close proximity to the portal circulation, they are particularly affected by the daily nutrient uptake.

As well as a significant increase in linoleic acid in ovarian cancer cells cocultured with adipocytes, we detected an increase in levels of oxidized lipids (13-HODE and 9-HODE), which are stereoisomer metabolites of linoleic acid. Coculture also strongly induced ROS and lipid peroxidation in the cancer cells. These 4 convergent experimental findings suggest that adipocyte-derived lipids not only provide energy to cancer cells but also serve as proinflammatory mediators and signaling lipids, which by inducing ROS in the cancer cells makes them more tumorigenic. Indeed, prostate cancer cells cocultured with mature adipocytes have a higher expression of NADPH oxidases and ROS (48), supporting the concept that cancer cells are activated in the presence of adipocytes.

In order to discover which proteins are upregulated in cancer cells by their interaction with adipocytes, we performed proteomic analysis, detecting close to 7,000 different proteins. Several proteins involved in lipid uptake and lipid transport were significantly upregulated, including CD36, ADH1A/B/C, and FABP4. Interestingly, the interrogation of 2 large data sets of primary HGSOC (TCGA, ref. 32; Tothill, ref. 33) by Tucker and colleagues (28) determined that ADH1B and FABP4 were the 2 most important genes associated with a significant risk of residual tumors in HGSOC after a R0 resection (surgery intended to remove all visible disease). In their analysis of ovarian cancer cells in the Cancer Cell Line Encyclopedia (49), they did not find overexpression of ADH1 and FABP4 (28), indicating that these genes are regulated by the tumor microenvironment. Indeed, our data suggest that ADH1 and FABP4 were induced in tumor cells by adipocytes, and make the cancer cells more invasive and metastatic. Because the TCGA and Tothill data sets report gene expression in whole tumor tissue, the upregulation of ADH1B and FABP4 seen here probably reflects the condition of tumor cells residing in adipocyte rich stroma. We previously found that FABP4 was upregulated at the leading edge of ovarian cancer cells invading and replacing omental adipose tissue (3). Likewise, ADH1B also promotes metastasis, including the invasion and adhesion of HGSOC (50). Alcohol dehydrogenases metabolize and remove toxic lipid peroxidation products, so there may be another mechanism in place that protects the cancer cells from the proapoptotic effects of ROS.

In view of the strong evidence that FABP4 is important in HGSOC biology, we set out to determine how inhibition of FABP4 affected global cancer cell metabolism and whether FABP4 inhibition might have therapeutic benefit. The stable knockdown of FABP4 in the cancer cells affected the entire lipidome, reducing β-oxidation, ROS generation, and lipid peroxidation, thereby impacting key tumorigenic pathways such as proliferation and cytokine release. The deletion of FABP4 with CRISPR also confirmed an inhibitory effect on tumor growth in vivo. This is consistent with the study by Gharpure and colleagues, showing that treatment with nanoliposomes encapsulating FABP4 siRNA impairs tumor growth and metastasis (34).

These studies (3, 28, 34, 51) and the results presented here focus on the role of FABP4 in cancer cells where it regulates proliferation and possibly quiescence. It is yet to be determined if inhibition of FABP4 reduces homing of ovarian cancer cells to the omentum or slows down the growth or metastatic implants.

However, FABP4 also plays an important role in the cells of the tumor microenvironment, contributing to the protumorigenic role of adipocytes, macrophages (52), and endothelial cells (53).

Indeed, intraperitoneal injection of high FABP4 expressing ovarian cancer cells into FABP4^−/− mice resulted in minimal tumor growth and only a few metastases (3). The overexpression of FABP4 in tumor-associated macrophages promotes breast cancer growth (52), and blocking FABP4 in endothelial cells reduced angiogenic activity and tumor growth (53). Therefore, testing an orally available FABP4 inhibitor (BMS309403), which could target both tumor and stromal cells in the tumor organ, was the next logical step if we were to translate biological insights into concrete treatments. All our results with BMS309403 (39), which was developed as a treatment for metabolic syndrome, blocked early metastasis and very efficiently reduced established metastasis and established tumor burden but did not significantly reduce primary tumor growth. Combining the FABP4 inhibitor with carboplatin, one of the standard treatments for ovarian cancer, almost completely inhibited metastasis and reduced tumor weight by over 75%. Although BMS309403 is primarily an experimental tool compound effective in the micromolar range, other FABP4 inhibitors are in development. For example, the FDA approved antibiotic levofloxacin inhibits FABP4 activity efficiently (IC₅₀ = 5.83 μmol/L) through binding to the FA binding pocket (BMS induces an allosteric regulation of FABP4 and upregulation of PPARγ; ref. 54).

Our data also provide a novel link between lipid metabolism and DNA methylation. The metabolomic study showed that FABP4 knockdown cells cocultured with adipocytes accumulate a large amount of α-ketoglutarate, a regulator of TET proteins. TETs are DNA demethylating enzymes that convert 5-mc to 5-hmc (35, 55, 56). The increased α-ketoglutarate levels in cells with FABP4 knockdown resulted in increased 5-hmc levels and more cells blocked in G₁, an effect that could be mimicked with the addition of a cell-permeable α-ketoglutarate analogue (57). Consistent with the in vitro data, we also observed increased 5-hmc staining in FABP4 knockdown tumors in vivo. Although specific gene-expression changes regulated by 5-hmc marks are cellular context dependent, increased DNA 5-hmc levels bear a strong correlation to reduced cell proliferation (36, 58), explaining why 5-hmc levels are reduced in multiple cancers (58, 59). Thus, one mechanism through which blocking FABP4 reduces tumor aggressiveness is the regulation of methylation.

In summary, we describe the proteo-metabolomic changes that occur in cancer cells with adipocyte coculture and provide support for targeting FABP4 in combination with platinum chemotherapy as a treatment for HGSOC.

K.M. Nieman and E. Lengyel have ownership interest in patent US8748470B2. E. Lengyel reports receiving a commercial research grant from AbbVie. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Mukherjee, K.M. Nieman, E. Lengyel

Development of methodology: A. Mukherjee, K.M. Nieman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Mukherjee, H.A. Daifotis, J.F. Fahrmann, R.R. Lastra, I.L. Romero, O. Fiehn

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Mukherjee, K.M. Nieman, J.F. Fahrmann, I.L. Romero, E. Lengyel

Writing, review, and/or revision of the manuscript: A. Mukherjee, H.A. Daifotis, K.M. Nieman, J.F. Fahrmann, E. Lengyel

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Mukherjee, C.-Y. Chiang, I.L. Romero, E. Lengyel

Study supervision: A. Mukherjee, O. Fiehn, E. Lengyel

We thank Drs. Fabian Coscia and Matthias Mann (Max Plank Institute of Biochemistry, Martinsried, Bavaria) for help with the proteomic analysis, Dr. A.F. Haney (University of Chicago) for collecting omental biopsies, and G. Isenberg for editing this manuscript. We really appreciate the help provided by C. Zhang and J. Andrade at the Center for Research Informatics (University of Chicago) in analyzing microarray data. We thank Bristol-Myers Squibb for providing us with the FABP4 inhibitor BMS309403. This study was supported by an NCI grant (R01CA169604) to E. Lengyel. Finally, we would like to thank all the patients, donors, and resident physicians in the department of Obstetrics and Gynecology at the University of Chicago for their support and commitment toward tissue collection and banking. This project was supported by a grant from the NCI (5R01CA169604), awarded to Ernst Lengyel.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.