Patient cohort study

We established a cohort of 98 patients with hepatocellular carcinoma (HCC) who underwent hepatectomy at the Organ Transplantation Center of China Medical University Hospital (CMUH), Taiwan, between 2011 and 2013 (Supplementary Figure S1). Tumor specimens were collected, preprocessed, aliquoted, and stored at − 80 °C for subsequent protein, RNA, and lipid extraction; some of the specimens were embedded in paraffin for immunohistochemical (IHC) or histological examination. The cohort’s data were monitored for up to 2000 days. Basic demographic data and survival hazard ratios are presented in Supplementary Figure S1A and S1B. Patient chart data, including follow-up information, were linked to primary HCC tumor specimen data, including lipidomic profiling data and transcriptomic profiling data, for which the patient number of each specimen type is shown in a Venn diagram plot (Supplementary Figure S1C). This study was approved by the Institutional Review Board of CMUH (IRB# DMR100-IRB-088).

Sample preparation for lipidome and transcriptome analysis

All samples were prepared following the manufacturer’s instructions to perform lipidomic analysis and transcriptome analysis. For lipidome analysis, the liver tissue from the HCC cohort was resuspended in D-PBS (Ca2+/Mg2+-free) for homogenization. Then, homogenates were prepared at a concentration of 5 mg/ml, and 300 µL aliquots were used for analysis.. Treated cells were resuspended in D-PBS (Ca2+/Mg2+-free) at a concentration of 3000 cells per μl, and then at least 300 μl of the cell suspension was prepared for analysis. For transcriptome analysis, the HCC cell lines (Tong and Huh7) were passaged three consecutive times. At each passage, 1 × 106 cells were seeded into new culture vessels. After reaching the appropriate confluency, cells were harvested and subjected to RNA isolation. Purified RNA was quantified at OD260 nm using an ND-1000 spectrophotometer (Nanodrop Technology, USA), and RNA quality was assessed using a Bioanalyzer 2100 (Agilent Technologies; add location, e.g., Santa Clara, CA, USA) with an RNA 6000 LabChip kit (Agilent Technologies).

Lipidome profiling

Lipids were extracted using a two-step chloroform/methanol procedure [20]. Samples were spiked with internal lipid standard mixture containing: cardiolipin 14:0/14:0/14:0/14:0 (CL), ceramide 18:1;2/17:0 (Cer), diacylglycerol 17:0/17:0 (DAG), hexosylceramide 18:1;2/12:0 (HexCer), lyso-phosphatidate 17:0 (LPA), lyso-phosphatidylcholine 12:0 (LPC), lyso-phosphatidylethanolamine 17:1 (LPE), lyso-phosphatidylglycerol 17:1 (LPG), lyso-phosphatidylinositol 17:1 (LPI), lyso-phosphatidylserine 17:1 (LPS), phosphatidate 17:0/17:0 (PA), phosphatidylcholine 17:0/17:0 (PC), phosphatidylethanolamine 17:0/17:0 (PE), phosphatidylglycerol 17:0/17:0 (PG), phosphatidylinositol 16:0/16:0 (PI), phosphatidylserine 17:0/17:0 (PS), cholesterol ester 20:0 (CE), sphingomyelin 18:1;2/12:0;0 (SM) and triacylglycerol 17:0/17:0/17:0 (TAG). After extraction, the organic phase was transferred to an infusion plate and dried in a speed vacuum concentrator. The first-step dry extract was re-suspended in 7.5 mM ammonium acetate in chloroform/methanol/propanol (1:2:4, V:V:V), and 2nd step dry extract in 33% ethanol solution of methylamine in chloroform/methanol (0.003:5:1; V:V:V). All liquid handling steps were performed using the Hamilton Robotics STARlet robotic platform with the Anti-Droplet Control feature for organic solvents pipetting.

Mass spectrometry-based lipidome analysis was performed by Lipotype GmbH (Dresden, Germany). Samples were analyzed by direct infusion on a QExactive mass spectrometer (Thermo Scientific) equipped with a TriVersa NanoMate ion source (Advion Biosciences). Samples were analyzed in both positive and negative ion modes with a resolution of Rm/z = 200 = 280,000 for MS and Rm/z = 200 = 17,500 for MSMS experiments, in a single acquisition. MSMS was triggered by an inclusion list encompassing corresponding MS mass ranges scanned in 1 Da increments. Both MS and MSMS data were combined to monitor CE, DAG, and TAG ions as ammonium adducts; PC, PC O–, as acetate adducts; and CL, PA, PE, PE O–, PG, PI, and PS as deprotonated anions. MS only was used to monitor LPA, LPE, LPE O–, LPI, and LPS as deprotonated anions; Cer, HexCer, SM, LPC, and LPC O– as acetate adducts. Lipid identification was performed using LipidXplorer software. Only lipid identifications with a signal-to-noise ratio > 5, and a signal intensity fivefold higher than in corresponding blank samples, were considered for further data analysis.

Lipidomics data analysis

We use our previously developed tools [21,22,23] to perform lipidomics data analysis. In sum, we applied a data filtering process to exclude lipid species with a missing value rate exceeding 70%. For the remaining missing values or values below the limit of detection, we imputed them using half of the minimum detected value for that lipid species across all samples. To address the skewed distribution of lipid abundance measurements, we further performed a log 10 transformation before conducting differential lipid expression analyses.

To visualize the samples by reducing the dimensionality of the processed lipidomics data, the t-SNE (t-Distributed Stochastic Neighbor Embedding) algorithm was used with specific hyperparameters: output dimensionality = 2, perplexity = 15, and iterations = 3000. Subsequently, we applied K-means clustering, an unsupervised learning method, to segregate and label two groups of patients based on the transformed t-SNE data. To identify differentially expressed lipids, we employed two-tailed Student’s t-tests with the Benjamini–Hochberg correction method to calculate fold changes and adjusted p-values between the two groups. Unless otherwise specified, lipid species were considered significant if their adjusted p-values were < 0.05 and their absolute log2 fold change was ≥ 1.

The results of differentially expressed lipid species were utilized for enrichment analysis of various lipid characteristics. Over-representation analysis with Fisher's exact test was applied to identify whether specific categories of lipid characteristics have significantly more or fewer up-regulated or down-regulated lipid species than expected by chance. The cut-off criterion for determining significantly enriched or depleted categories in lipid characteristics was set at a p-value < 0.05.

Trans-Omics data analysis

We conducted a Spearman correlation analysis to evaluate the associations between lipid abundances and gene expression. We reported the resulting p-values and correlation coefficients to assess the strength of these associations. An association was considered significant if it met the following criteria: an absolute correlation coefficient greater than 0.4 and a p-value less than 0.05. We further analyzed the overlapping genes correlated with ether-lipids using functional enrichment analysis to investigate the biological functions of ether-lipids.

Survival analysis

We calculated the median value of each lipid species and used it to stratify patients into two groups. Lipidomics data and clinical records, encompassing demographics, liver function parameters, and tumor characteristics, were transformed into binary variables. We constructed a univariate Cox proportional hazards (Cox-PH) model and a log-rank test to investigate survival-related clinical variables, lipid species, and lipid characteristics. This analysis allowed us to calculate log-rank p-values, hazard ratios (HR), and their corresponding 95% confidence intervals. Significant survival-relevant factors were identified based on a log-rank p-value < 0.05. Additionally, for survival analysis, we calculated the median value of each lipid and used it to stratify patients into two groups. Additionally, we conducted enrichment analysis for lipid species in relation to patient prognosis. We employed the hazard ratio (HR) to determine whether specific lipid species were beneficial or detrimental to patient survival.

Transcriptome profiling

All RNA sample preparation procedures were carried out according to the Illumina's official protocol. Illumina's TruSeq Stranded Total RNA Library Prep Gold Kit (Cat. 20020598) was used for library construction, followed by AMPure XP beads (Beckman Coulter, USA) size selection. The sequence was determined using Illumina's sequencing-by-synthesis (SBS) technology (Illumina, USA) NovaSeq 6000. Sequencing data (FASTQ reads) were generated using Welgene Biotech's pipeline based on Illumina's basecalling program bcl2fastq v2.20. Transcriptomic data are publicly available in Gene Expression Omnibus (GEO) at GSE242315.

Transcriptomics data analysis

We utilized our previously developed pipelines to analyze the RNA-seq data [24]. In sum, Raw sequencing reads in FASTQ format underwent initial quality assessment using FastQC (v0.11.5) (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Subsequently, adapter sequences and low-quality bases were removed using Trimmomatic (v0.39) [25], employing standard parameters for quality trimming and minimum read length filtering. The resulting high-quality, processed reads were aligned to the GRCh38 reference genome from the Ensembl database using the HISAT2 aligner (v2.2.1) [26]. Alignments were processed and sorted before quantification. Transcript assembly and gene-level quantification were performed using StringTie (v2.1.5) [26], guided by the reference genome annotation (Homo_sapiens.GRCh38.97.gtf downloaded from the Ensembl official website). This involved merging transcript information across all samples to create a consensus annotation before calculating final gene expression estimates for each sample. Gene-level read counts derived from the StringTie output were then used for differential expression analysis using the DESeq2 package (v1.30.1) [27] in the R statistical environment (v4.0.0). Genes were considered statistically significant if they met criteria: mean number of transcripts per million was > 1, adjusted p-value < 0.05 (Benjamini–Hochberg correction), and absolute log2 fold change > 1.

Functional enrichment analysis elucidates the biological roles and pathways associated with the identified significantly differentially expressed genes (DEGs). The set of significant DEGs was tested for over-representation in established functional categories, including Gene Ontology (GO) and pathway databases such as KEGG, Wikipathway, and Reactome. P-values were adjusted for multiple comparisons using the Benjamini–Hochberg procedure to control the False Discovery Rate (FDR). Functional categories and pathways with an adjusted p-value below 0.05 were considered significantly enriched.

Cell culture

Human HCC cell lines (Tong and Huh7) were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Gibco, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS; Gibco, Carlsbad, CA, USA) and 1% penicillin/streptomycin (30-002-CI; Corning, New York, USA) in a 5% CO2 atmosphere at 37 °C. The Huh7 cell lines were purchased from the Bioresource Collection and Research Center (Hsinchu, Taiwan) and the American Type Culture Collection (Manassas, VA, USA). The Tong cells were generously provided by YS Jou (Academia Sinica, Taiwan).

Immunoblot and quantitative analyses

For protein extraction from the human liver tissues and HCC cells, samples were lysed in radioimmunoprecipitation assay (RIPA) buffer (100 mM Tris, 5 mM EDTA, 5% NP40; pH 8.0) containing protease inhibitors and kept on ice for 30 min. Lysates were then centrifuged at 14,000 rpm for 25 min to isolate total protein [28]. These proteins were mixed with 6X sample buffer, subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis, and subsequently transferred to a polyvinylidene fluoride membrane (Millipore, MA, USA). The membrane was soaked in 5% nonfat milk for 1 h to block nonspecific binding, after which it was incubated with primary antibodies overnight. The following day, the membrane was washed thrice in Tris-buffered saline with Tween 20 (TBST; 1% Triton) for 10 min each and then incubated with secondary antibodies for 1 h. After another three washes in TBST and one in TBS, ECL reagent was applied to the membrane. Protein signals were detected using a Chemidoc XRS + (Bio-Rad) equipped with a charge-coupled device (CCD) camera. All captured images were analyzed using Image J [29].

RNA isolation and qRT-PCR

All samples from the HCC cells and human liver tissues were lysed with 1 mL TRIzol (15,596,026; Invitrogen, Carlsbad, CA, USA). Phenol–chloroform (pH 6.7/8.0; 0883; VWR International, Pennsylvania, USA) was subsequently added for phase separation, and RNA-rich layers were isolated through centrifugation. Soluble RNA was precipitated using 2-propanol, and the salt was washed off using 75% ethanol. The RNA was then dissolved in RNase-free water. For cDNA synthesis, 1 µg of total RNA was subjected to reverse transcription conducted using the PrimeScript RT reagent kit (TAKARA Bio Inc., Kyoto, Japan) in accordance with the manufacturer’s instructions. The subsequent cDNA was analyzed using a real-time detection system (Bio-Rad Laboratories, Inc., California, USA) and the KAPA SYBR FAST One-Step qRT-PCR Kit (Kapa Biosystems, Inc., Wilmington, MA, USA) in accordance with manufacturers’ instructions. mRNA expression levels were then detected using an AZURE CIELO real-time PCR system (AZURE CIELO, Azure Biosystems, Inc., Dublin, CA, USA). Gene expression was normalized to the housekeeping genes β-actin and GAPDH and was quantified using the 2−ΔΔCt method.

IHC staining and scoring

Tissue Sects. (2 μm) were stained with primary antibodies, followed by amplification using an ABC kit (Vector Laboratories, Inc., Burlingame, CA, USA) for enhanced signal visualization. The SQSTM1/p62 primary antibody was employed. The staining intensity of SQSTM1/p62 was quantified using ImageJ software [30, 31].

Wound healing assay

For ether-lipid treatments, the human HCC cell lines (Tong, and Huh7) were pretreated with 10 nM of ether-linked phosphatidylcholine (PC O–; 878112; Avanti Research, Alabaster, USA) and ether-linked phosphatidylethanolamine (PE O–; 878130; Avanti Research, Alabaster, USA) for 2 days. After treatment, the cells were seeded onto Culture-Insert two-well plates (80209; ibidi GmbH, Martinsried, Germany) until they reached 90% confluence. After 24 h, the Culture-Insert plates were removed, and images were captured at 0 and 24 h using a bright-field microscope.

For tranilast treatment, the HCC cells were pretreated with PC O–, combined with tranilast (200 µM; T0318, Sigma-Aldrich, Burlington, MA, USA) for 2 days. After treatment, the cells were seeded onto Culture-Insert two-well plates until they reached 90% confluence. After 24 h, the Culture-Insert plates were removed, and images were captured at 0 and 24 h using a bright-field microscope.

For the β-carotene (C4582; Sigma-Aldrich, Burlington, MA, USA) and retinoic acid (R4643; Sigma-Aldrich, Burlington, MA, USA) treatments, the human HCC cell lines were seeded onto Culture-Insert two-well plates until they reached 90% confluence. After a 24 h wait, the Culture-Insert plates were removed. Subsequently, the cells were treated with β-carotene and retinoic acid (20 µM) for another 24 h. Images were taken at both the start (0 h) and end (24 h) of this treatment by using a bright-field microscope. For combined treatments, the ether-lipid-treated cells were seeded onto Culture-Insert two-well plates. After 24 h, the Culture-Insert plates were removed, and the cells were maintained in β-carotene or retinoic acid-containing culture medium, and then images were captured at 0 and 24 h using a bright-field microscope.

For fenofibrate treatment, HCC cells were pretreated with fenofibrate (10 µM; F6020, Sigma-Aldrich, Burlington, MA, USA) for 72 h. The cells were then seeded onto Culture-Insert two well until they reached 90% confluence. After 24 h, the insert well was removed, and the cells were maintained in a culture medium, and then images were taken at 0 and 8 h. All images were analyzed using the “Wound_healing.ijm” macro in ImageJ/Fiji [32].

Cell invasion assay

For treatment with ether-linked lipids, the HCC cells were seeded at a density of 1 × 105 cells/well in the upper chamber of 24-well culture inserts filled with FBS-free DMEM medium. These inserts were coated with Matrigel (BD Bioscience, Franklin Lakes, New Jersey, USA) for 4 h. In the lower chamber, culture medium containing PC O– (10 nM) and PE O– (10 nM) was added. After a 24 h incubation, the Matrigel was removed, and the membrane was fixed with 4% formaldehyde for 30 min. The membrane was then stained with 0.05% crystal violet in 60% ethyl alcohol for 30 min, washed twice with PBS, excised, and placed onto microscope slides. Imaging was conducted using a fluorescence microscope (Nikon, ECLIPSE 80i). For the treatments and co-treatments with β-carotene (20 μM), retinoic acid (20 μM), and tranilast (200 μM), a similar protocol was followed, with the lower chamber of the 24-well culture inserts containing β-carotene, retinoic acid, or tranilast. Cell number quantification was conducted using the Image J analysis method [32, 33].

Immunofluorescence and quantitative analysis

Tong cells were seeded onto four-well glass chamber slides (PEZGS0416, Millipore, Burington Massachusetts, USA) and treated accordingly. Cells were rinsed with PBS and fixed with 2% PFA (Macron) for 15 min at room temperature. After three PBS washes (5 min each), cells were permeabilized with 0.2% Triton X-100 (Bio Basic Inc., Markham, Ontario Canada) in PBS for 30 min. They were then washed, blocked with 1% BSA in PBST for 1 h, and incubated overnight at 4 °C with primary antibodies (1:100). After washing, cells were stained with fluorescent secondary antibody (1:200; Abcam plc, Cambridge, UK) and DAPI (1 µg/mL; Sigma-Aldrich, Burlington, MA, USA) for 1 h at room temperature. Cells were washed again, stained with 5 μg/mL Bodipy 493/503 (Invitrogen, Carlsbad, CA, USA) for 15 min, washed, and finally mounted with mounting media (Invitrogen, Carlsbad, CA, USA). Imaging was conducted using a Leica SP8 microscope (63x/1.4NA lens) or an ANDOR dragonfly high-speed confocal system (63x/1.4NA lens), capturing at least five images per well for each experiment. For immunofluorescence (IF) staining, the following antibodies were used: anti-ADFP antibody (ab108323; Abcam plc, Cambridge, UK), anti-ADRP antibody (sc-377429; Santa Cruz Biotechnology, Inc., California, USA), anti-SQSTM1/p62 antibody (ab91526; Abcam plc, Cambridge, UK), anti-LC3-B antibody (ab192890; Abcam plc, Cambridge, UK), anti-LC3 A/B antibody (Abc929; Sigma-Aldrich, Burlington, MA, USA), anti-ALDH3A2 antibody (PA5-120435; Invitrogen, Carlsbad, CA, USA), and anti-ALDH3A2 antibody (sc373921; Santa Cruz Biotechnology, Inc., California, USA).

p62 puncta number and size were measured using Fiji/ImageJ. Colocalization analysis was performed by comparing green and red channels in the same field, generating a scatterplot and applying thresholds to identify overlapping areas. The size, diameter, and number of colocalized puncta were quantified. For each experiment, over 1000 cells from 3–5 images were analyzed. Results are shown as mean ± standard deviation, and statistical significance was determined using Student’s t-test with a p-value > 95% considered meaningful.

Luciferase reporter assay

The human HCC cells were seeded onto a 24-well plate at a density of 105 cells per well. The cells were transfected with 0.5 µg of PPRE X3-TK-luc (#1015; addgene, Watertown, USA) plasmid and pRL-TK using Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) in accordance with the manufacturer’s instructions. After 24 h of incubation, the cells were maintained in culture medium containing 20 µM PPAR ligand (either β-carotene or retinoic acid) for an additional 48 h. After 48 h, both firefly and Renilla luciferase activities were measured using the Dual-Luciferase Reporter Assay System (Promega Corporation, Madison, WI, USA) in accordance with the manufacturer’s instructions and then detected through FLUOROSKAN FL (Thermo Fisher Scientific Inc., Waltham, MA, USA) [34, 35].

Three-dimensional invasion assay

The method was modified from published paper [36]. In brief, Preparation of cell suspension (0.5–1 × 104 cells/ml) and dispense 200 µL of cell suspension to 96-well round bottom plates. Incubate the plate in incubator for four days to form tumor spheroid. Four days later, place the plate on ice and remove 100 μl of medium. Then dispense 100 μl of BMM (containing 10 ng/ml of EGF) into the well. Transfer the plate to an incubator to make the BMM solidify. One hour later, add 100 μl/well of complete growth medium containing 30 μM of Fenofibrate (3 × the desired final concentration). After treatment, take the images at different time point (0, 24 h, 48 h). All the images were quantified by using Image J.

Isolation of G-actin and F-actin

G-actin and F-actin were isolated using a previously described method [37, 38], with some modifications. In brief, the human HCC cells were cultured in a 60-mm dish and treated with PC O– and co-treated with tranilast for 48 h. Subsequently, the cells were washed with PBS and then lysed with 0.5 mL of F-actin stabilizing buffer (comprising 50 mM PIPES, pH 6.9, 50 mM NaCl, five mM MgCl2, five mM EGTA, 5% glycerol, 0.1% Triton X-100, 0.1% Tween-20, 0.1% NP-40, 0.1% 2-mercaptoethanol, one mM ATP, and a protease inhibitor cocktail) for 15 min on ice. The cells were then scraped off and centrifuged at 16,000 × g for 75 min. The supernatant containing the G-actin fraction was carefully collected, and the pellet containing the F-actin fraction was resuspended in 0.5 mL of chilled water supplemented with one mM cytochalasin D (BML-T109; Enzo Biochem, Inc., NY, USA) and allowed to incubate for one h on ice. Equal volumes of the G-actin and F-actin fractions were then subjected to Western blot analysis.

Statistics

All the results and analyses in the in vitro experiments were performed in triplicate and shown as mean ± SD. Statistical analysis was performed with Prism version 6 (GraphPad Software, San Diego, CA, United States). An unpaired Student’s t-test was used to compare two groups to determine significant differences (p < 0.05). Additional material and methods are described in the supplementary section (Supplementary Table S1- Table S3).