Cell lines and culture conditions

The human non-tumorigenic epithelial cell lines MCF10A and MCF12A were cultured in DMEM/F12 medium (Thermo Fisher Scientific, Waltham, US) with 5% Horse Serum (Thermo Fisher Scientific), 20 ng/mL EGF (Sigma-Aldrich, St. Louis, US), 10 μg/mL insulin (Sigma-Aldrich), 100 ng/mL cholera toxin (Sigma-Aldrich) and 0.5 mg/mL hydrocortisone (Sigma-Aldrich). The human breast cancer cell line MCF7 (RRID: CVCL_0031) was cultured in MEM medium (Thermo Fisher Scientific) with 10% fetal bovine serum (FBS, Thermo Fisher Scientific). The human breast cancer cell lines T47D (RRID: CVCL_0553) and BT474 (RRID: CVCL_0179) were cultured in RPMI medium (Thermo Fisher Scientific) with 10% FBS. The human breast cancer cell line SKBR3 (RRID: CVCL_0033) was cultured in McCoys 5A medium (Thermo Fisher Scientific) with 10% FBS. The human breast cancer cell lines MDA-MB-231 (RRID: CVCL_0062), MDA-MB-468 (RRID: CVCL_0419) and Hs578T (RRID: CVCL_0332) were cultured in DMEM (Thermo Fisher Scientific) with 10% FBS. All cell lines listed above were purchased from the American Type Culture Collection (ATCC, Manassas, US). Human breast cancer cell lines SUM149PT (RRID: CVCL_3422) and SUM159PT (RRID: CVCL_5423) were purchased from BioIVT (Detroit, MI, US) and cultured in Ham’s F12 medium (Thermo Fisher Scientific) with 10% FBS. Cells were incubated at 37 °C with 5% CO2 and a humidified atmosphere. All cell lines were annually tested to be free from mycoplasma, and STR profiling was performed to verify cell line identity.

siRNA silencing

On-target plus Human PNPLA8 siRNA-SMARTpool and negative scrambled siRNA were purchased from Dharmacon (Cambridge, UK). The target sequences of PNPLA8 are listed below: GAGAAGGGCUGUUGCUAAU, UCAGUAACUUGAUGGAUUU, GACCUGAAACAUCGAUUUA and GAGUCUCAUUUGUCCAAUA.

LC–MS/MS lipidomic analysis

Cell lines were cultured under standard condition as described above for 48 h. Approximately 1 × 107 cells growing in the log phase with a confluence of 80% were trypsinized and pelleted. 200 µL ddH2O were added to pellets, followed by sonication with short bursts for homogenization. 5 µL of homogenate was used for BCA protein assay for normalization, and the remaining sample was extracted using a modified Bligh and Dyer procedure to obtain a crude lipid fraction [22]. Cell homogenate samples were gently mixed in a glass tube with ddH2O, followed by extraction with methanol/dichloromethane containing twelve internal standards. Following incubation on ice for 30 min and centrifugation (10 min, 3000 g, 4 °C) for phase separation, the organic phase containing lipids was collected and stored at -20 °C. Prior to analysis, 800 μL aliquot of the organic layer was dried and re-suspended in 150 µl of running solvent (dichloromethane:methanol (1:1) containing 5 mM ammonium acetate). Lipid analysis was conducted in MS/MSALL electrospray ion positive mode on a TripleTOF 5600 (AB Sciex, Redwood City, CA) quadrupole-time of-flight mass spectrometer (Q-TOF) coupled to a high-performance liquid chromatograph (Shimazu, Canby, OR) using a LC-20AD pump and SIL-20AC XR autosampler. The mass spectrometer was operated at a mass resolution of 30,000 for TOF MS scan and 15,000 for product ion scan (MS/MS) in high sensitivity mode, and the instrument was automatically calibrated after every ten-sample injection using an APCI positive calibration solution delivered through an automatic calibration delivery system (AB SCIEX). Details of the mass spectrometry method were previously published [23] and are detailed in the Supplemental Information. MultiQuant software and LipidView database (version 1.3, AB SCIEX, Concord, Ontario, Canada) were used for identification and annotation of lipid species. For relative quantification, lipid peak intensities were normalized using their corresponding internal standards. Each sample was run in duplicate and averaged normalized intensities of each lipid were used for statistical analysis. Detailed Triple TOF MS/MSALL comprehensive lipidomic analysis protocols are described in supplemental methods. All chemicals and solvents used in this study are summarized in Additional file 1: Table S1.

Dual-phase extraction and 1H magnetic resonance spectroscopy

Cell pellets (at least 1 × 107 cells) were harvested and ground over liquid nitrogen. Water-soluble metabolites of cells were extracted using the dual-phase extraction method (methanol:chloroform:water = 1:1:1) as previously described [24]. The aqueous fractions were lyophilized and re-dissolved in D2O containing 0.24 × 10 −6 mol 3-(trimethylsilyl)propionic-2,2,3,3,-d4 acid (TSP, Sigma-Aldrich) as chemical shift and concentration reference for metabolite quantification. Fully relaxed 1H high-resolution (HR) magnetic resonance spectroscopy (MRS) was performed using a Bruker Avance-III 750 MHz spectrometer equipped with a 5-mm TXI probe. Water-suppressed spectra were acquired using a 1D NOESY pulse sequence with a relaxation delay of 10 s, 256 scans, 8 dummy scans, receiver gain 40.3, and mixing-time of 80 ms. Water-soluble metabolites were quantified using TopSpin software (Bruker BioSpin Corp., Billerica, MA) as previously described [25].

RNA extraction and quantitative RT-PCR

RNA extractions were performed using the RNeasy Mini Kit (QIAGEN, Hilden, Germany) following the manufacturer’s protocol. 500 ng of RNA was reversely transcribed using iScript cDNA Synthesis Kit (Bio-Rad, Hercules, US). Quantitative PCR (qPCR) analysis was performed with the CFX Connect Real-Time PCR System (Bio-Rad) using the IQ SYBR Green Supermix (Bio-Rad). The housekeeping gene actin beta (ACTB) was used as an internal control. The relative fold changes in gene expression were calculated using the 2−△△Ct method. Primer sequences are provided in Additional file 1: Table S2.

Western blots

Cell pellets (at least 1 × 106 cells) were harvested and suspended in RIPA lysis buffer (Sigma-Aldrich) with Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific). Samples were homogenized by sonication with short burst and centrifuged at 13,000 rpm for 20 min at 4℃. Supernatants were collected, and protein concentration was measured using a BCA Protein Assay Kit (Thermo Fisher Scientific). Proteins (30 μg) were analyzed by SDS-PAGE and transferred onto PVDF membranes (EMD Millipore, Darmstadt, Germany). Following blocking with 5% non-fat milk (EMD Millipore) in Phosphate Buffered Saline (PBS) at room temperature for 1 h, membranes were incubated with primary antibodies overnight at 4℃. Then membranes were incubated with horseradish peroxidase-conjugated secondary antibodies at room temperature for 1 h. Visualization was performed using the Pierce™ ECL Plus Western Blotting Substrate (Thermo Fisher Scientific) and ChemiDoc MP Imaging System (Bio-Rad). The density of the bands was analyzed by Image Laboratory software version 4.0.1 (Bio-Rad). GAPDH was used for normalization. Full uncropped images are shown in Additional files 1: Fig. S4–7. Antibody information is provided in supplemental materials. All antibodies used in this study are summarized in Additional file 1: Table S1.

Immunohistochemistry of TMAs

Two human breast cancer tissue microarray (TMA) slides (catalogue #BR1921C; Biomax, Rockville, MD and catalog #BC081120f; US Biomax, Rockville, MD) were used. All human tissues on the TMAs were collected under HIPAA approved protocols by the supplier, US Biomax, Inc. (https://www.biomax.us/FAQs). Paraffin-embedded TMAs were deparaffinized and blocked for endogenous peroxidase activity with 0.3% hydrogen peroxide in methanol for 15 min. After rehydration, antigen retrieval was performed in 10 mM citrate buffer (pH 6.0, > 90 °C) for 20 min. Tissues were blocked for non-specific antibody binding with 5% Bovine Serum Albumin (BSA) in PBS for 30 min at room temperature, followed by overnight primary antibody incubation for PNPLA8 (Rabbit, 1:200, HPA020083, RRID: AB_1851849, Sigma) at 4 °C. Next, TMAs were incubated with biotin-conjugated secondary antibody for 30 min (Goat-anti-Rabbit, RRID: AB_3073814, Vectastain, Vectorlabs, Burlingame, CA), followed by incubation with avidin-streptavidin complex (Vectastain, Vectorlabs, Burlingame, CA) for 30 min. Sections were developed with 3,3′-diaminobenzidine (DAB; Dako, Glostrup, Denmark) and counterstained with hematoxylin. After dehydration, slides were mounted using Entellan (Merck Millipore, Burlington, Massachusetts, USA). The stains were digitized with an Aperio CS2 scanner (Leica Microsystems) using a 20 × magnification. Images were scanned using Aperio ImageScope (Version 12.3.3, Leica, Microsystems). The pixel H-score of PNPLA8-positive staining for each tissue core on the TMAs was calculated by using QuPath (Version 0.3.2, University of Edinburgh, UK). Briefly, tumor cells and stromal cells were classified by nuclear/cell area ratio. The classifier was further trained by manual cell annotations by a pathologist. H-scores of PNPLA8 staining of tumor cells were analyzed by the calculation of mean intensity of cytoplasm DAB.

Cell viability

Cells (3 × 103) were seeded onto a 96-well plate in 100 μL culture medium and incubated for 72 h at 37 °C. Then 10 μL of tetrazolium salt WST-1 (4-[3-(4-Iodophenyl)-2-(4-nitro-phenyl)-2H-5-tetrazolio]-1,3-benzene sulfonate) (Roche) was added into each well and incubated at 37 °C for 4 h. The absorbance was measured on an Epoch Microplate Spectrophotometer (BioTek, Winooski, US) at 450 nm.

Cell migration

Cells (1 × 105) were seeded on the upper chambers of an 8.0 μm pore size Transwell plate from Corning (#3422, Corning, USA). The lower chambers were supplemented with 600 μl DMEM and 10% FBS. After 24 h, invaded cells were fixed with methanol, stained with 0.1% crystal violet (Sigma-Aldrich, USA), and counted under a microscope (Olympus, USA).

Measurement of reactive oxygen species (ROS)

Intracellular ROS levels were determined by using DCFDA-Cellular ROS Assay Kit (ab113851, Abcam) according to the manufacturer’s protocol. Briefly, cells (5 × 105) were harvested and incubated with 1 μM DCFDA for 30 min at 37 °C. Then, the fluorescence was detected by Cytek Aurora flow cytometry (Cytek Biosciences, US) with excitation/emission at 485 nm/535 nm.

Measurement of mitochondrial superoxide

Mitochondrial superoxide levels were determined by using MitoSOX Green mitochondrial superoxide indicator (M36005, Thermo Fisher Scientific) according to the manufacturer’s protocol. Briefly, cells (5 × 105) were harvested and incubated with 1 μM MitoSOX Green for 30 min at 37 °C. Then, the fluorescence was detected by Cytek Aurora flow cytometry (Cytek Biosciences, US) with excitation/emission at 488 nm/510 nm.

Eicosanoid extraction

Cell homogenate (200 uL in ddH2O) was spiked with 50 uL of eicosanoids heavy isotope internal standards mixture (10 ng/mL) followed by addition of ddH2O (750 uL) to make a total volume of 1 mL cell suspension. Eicosanoids from cell suspension (1 mL) were extracted using solid phase extraction (SPE) as described by Wang et al. with minor changes [26]. Briefly, Strata TM-X 33 µm polymeric reversed phase SPE columns (cat # 8B-S100-UBJ; Phenomenex, CA, US) were preconditioned with 3.0 mL of 100% methanol, followed by 3.0 mL of water. The samples (1 mL) were then loaded into the SPE columns and washed with 2 mL of ddH2O followed by 2.0 mL of 10% methanol to elute polar and semi-polar metabolites. Finally, eicosanoids were eluted from SPE with 1.5 mL of 100% methanol. The collected extracts were completely dried under a nitrogen evaporator (Organomation, MA, USA) and stored at − 80 ̊C until analysis.

LC–MS/MS analysis of eicosanoids

LC–MS/MS analysis of eicosanoids was performed as described earlier [26] with some optimizations. Briefly, eicosanoids were separated on a C18 reverse-phase column 2.6 µm, 100 × 2.1 mm (Phenomenex, Torrance, CA, USA) employing a binary mobile phase gradient program (Eluate-A: ACN/water/acetic acid (60/40/0.02, v/v), and eluate-B: ACN/IPA (50/50, v/v)) using an Ultrafast Liquid Chromatography (UFLC) system (Shimadzu, Nakagyo-ku, Kyoto, Japan). The gradient elution for 13 min was as follows: 0.1–90% B (0.01–9.0 min); hold 90% B for 2 min (9.0–11.0 min); 90–0.1% B (11.00–13.00 min) at a constant flow rate of 0.4 mL/min. Eluted eicosanoids were introduced into hybrid quadrupole ion trap (API4000 QTRAP LC–MS/MS, AB Sciex, ON, Canada) mass spectrometer where individual eicosanoids were ionized in electrospray ionization (ESI) negative mode and acquired under multiple reaction monitoring (MRM) mode for quantification. Mass spectrometer source and analyzer parameters were optimized to get good signal for all eicosanoid species. Eicosanoid standard cocktail (Cayman Chemicals, MI, USA and Avanti Polar Lipids, AL, USA) was used to construct nine-point calibration curves (1, 10, 25, 50, 75, 100, 150, 200, and 500 ng/mL) by plotting the graph between area under the curve (AUC) response to the standards concentrations. These calibration curves were employed to the measured AUC for analytes in the extracted samples to calculate measured quantities in each sample. Instrument control and data acquisition were performed using Analyst (version 1.4.2, SCIEX Inc. Thornhill, Ontario, Canada), and data analysis was completed using MultiQuant software (version 2.0, SCIEX, Thornhill, ON, Canada).

Prostaglandin E2 ELISA assay

Prostaglandin E2 levels of cell culture medium were determined by using a PGE2 ELISA Kit (ab133055, Abcam). Briefly, cells (2 × 104) suspended in 200 μL culture medium were plated in 96-well plates. After 24 h, cell culture media were collected and centrifuged at 2000 rpm for 5 min. Then, the supernatants were collected for subsequent PGE2 analysis according to the manufacturer’s protocol. After incubation, para-Nitrophenylphosphate (pNpp) substrate was added to the ELISA plate, and samples were analyzed using an Epoch Microplate Spectrophotometer (BioTek, Winooski, US) at 405 nm.

Volcano-plot analysis

The average intensities of individual lipids in the immortal human mammary epithelial cell lines MCF10A and MCF12A were calculated as controls. The average intensities of individual lipids in the TNBC cell lines MDA-MB-231, SUM159PT, Hs578T, MDA-MB-468 and SUM149PT were calculated as TNBC comparison group. Volcano plots of fold changes of individual lipids in the TNBC group compared with controls were conducted by using the SIMCA software version 14.1 (Umetrics, Sweden).

S-plot analysis

Immortal human mammary epithelial cell lines MCF10A and MCF12A were assigned to control group. TNBC cell lines MDA-MB-231, SUM159PT, Hs578T, MDA-MB-468 and SUM149PT were assigned to TNBC group. The intensities of individual lipids of the control and the TNBC group were uploaded to the SIMCA software version 14.1 (Umetrics, Sweden) to conduct S-plot analysis. S-plot visualizes the covariance and the correlation structure between individual lipids and the predictive score of the predictive component. The confidence of individual lipids as a discriminant of variance increases with increasing numerical values on the y-axis and the size of the contribution increases with increasing numerical values on the x-axis. Individual lipids that were most up- or down-regulated in the TNBC group were selected at the cutoff value p(corr) ≥ 0.7 for upregulation (labeled as red dots) and ≤ -0.7 for downregulation (labeled as blue dots).

Correlation analysis of protein expression and lipid levels

The densities of the Western Blot bands were analyzed by Image Laboratory software version 4.0.1 (Bio-Rad) and normalized by GAPDH intensity of each sample, which represented relative protein expression levels. Pearson’s correlation analyses of protein expression and lipid levels were conducted by using GraphPad Prism 5 software (La Jolla, CA, USA).

Statistical analysis

Principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA) were conducted by using SIMCA software version 14.1 (Umetrics, Sweden). MetaboAnalyst 5.0 (http://www.metaboanalyst.ca/) was used for performing clustering analysis and heatmap analysis. Venny 2.1 (https://bioinfogp.cnb.csic.es/tools/venny/index.html) was used for preparing Venn-diagrams. Gene set enrichment analysis (GSEA) was performed using GSEA software (version 4.2, San Diego, CA, USA) to analyze the correlation of PNPLA8 to hallmark gene sets according to the data of the TCGA breast cancer dataset. The Kaplan–Meier Plotter (http://www.kmplot.com/) breast cancer mRNA gene chip platform was used for generating Kaplan–Meier curves based on the specified gene expression levels for the reported patient survival rates. The probes selected for the analysis of each gene are indicated in their corresponding figures. Bar graph analyses were conducted by using GraphPad Prism 5 software (La Jolla, CA, USA). Differences between groups were evaluated using an unpaired two-tailed student’s t test. p values < 0.05 were considered significant.