2.1 Data acquisition from online datasets

In this study, the selection of the cancer cohort varied depending on the specific analyses being performed. For the pan-cancer analysis, we included all cancer types available in The Cancer Genome Atlas (TCGA) dataset. For more detailed analysis and presentation, we selected high-incidence cancers such as bladder urothelial carcinoma (BLCA), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), pancreatic adenocarcinoma (PAAD), skin cutaneous melanoma (SKCM), stomach adenocarcinoma (STAD), breast invasive carcinoma (BRCA), kidney renal clear cell carcinoma (KIRC), liver hepatocellular carcinoma (LIHC), lung carcinoma (LUCA, including adenocarcinoma and squamous cell carcinoma cohorts), Prostate adenocarcinoma (PRAD), uterine corpus endometrial carcinoma (UCEC), and head and neck squamous cell carcinoma (HNSC). These cancer types were selected based on their prevalence and clinical significance. All transcriptomes (gene expression RNAseq – IlluminaHiSeq, TCGA.cohort.sampleMap/HiSeqV2) and clinical data for cancer patients in TCGA were obtained from University of California, Santa Cruz Xena in September 2021 (http://xena.ucsc.edu). These datasets provided gene-level transcription estimates presented as log₂(x+1) transformed RNA-seq by expectation-maximization normalized counts. Additionally, robust multi-array average (RMA) normalized expression data for cancer cell lines were sourced from the Cancer Cell Line Encyclopedia (CCLE, https://sites.broadinstitute.org/ccle/) [24] and the Genomics of Drug Sensitivity in Cancer (GDSC, https://www.cancerrxgene.org) [25] respectively. The half-maximal inhibitory concentration (IC₅₀) values of cancer cell lines treated with CAY10566, SB505124, and LY2109761 were obtained from GDSC. CAY10566 is a selective inhibitor of SCD, while SB505124 and LY2109761 are selective inhibitors of transforming growth factor-beta (TGF-β) receptor kinase. Microarray data were sourced from datasets GSE13507 [26], GSE49644 [27], GSE24202 [28], and GSE41485 [29], available from Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/). Proteomes and clinical data (TCGA-PANCAN32-L4) from patients in The Cancer Proteome Atlas (TCPA) were acquired in September 2021 (https://tcpaportal.org) [30], and Level 4 data were used for batch analysis.

2.2 In-silico analysis of the online dataset

High and low expression levels of FADS2 and SCD were determined using the median transcriptional expression values of these genes in each cancer cohort from the TCGA database, with the median value serving as the cutoff point. Patients were classified into four groups based on their combined expression levels of FADS2 and SCD: FADS2^highSCD^low, FADS2^lowSCD^high, FADS2^highSCD^high, and FADS2^lowSCD^low. Pan-cancer expression correlation analyses were performed using Spearman's rank correlation coefficient, with data obtained from GEPIA2 (http://gepia2.cancer-pku.cn/#index) [31]. Differentially expressed genes (DEGs) were generated through NetworkAnalyst (https://www.networkanalyst.ca) [32] or GEO2R (https://www.ncbi.nlm.nih.gov/geo/geo2r/), following the grouping of patients. Enrichment analyses were conducted with Metascape (https://metascape.org/) [33]. Enrichment analyses were also performed using Gene Set Enrichment Analysis (GESA, https://www.gsea-msigdb.org/) [34] and visualized by Sangerbox (http://sangerbox.com) [35]. EMT scores provide a quantitative measure of the EMT status of cancer patients in TCGA (Creighton or Byers EMT score) [36] and cell lines in CCLE and GDSC (Tan EMT score) [37], ranging from epithelial to mesenchymal states. An EMT score less than 0 (with more negative values indicating a stronger epithelial-like phenotype) classified the tumor sample as epithelial, while a score greater than 0 (with more positive values indicating a stronger mesenchymal-like phenotype) classified the tumor sample as mesenchymal. This classification enabled us further grouping and analysis, allowing us to investigate the relationship between EMT status and various biological and clinical parameters. The transcription factor prediction analysis for the FADS2 promoter (hg38_refGene_NM_004265, range = chr11:61826300-61828299) was executed in JASPAR (https://jaspar.elixir.no) [38] with a relative score > 0.8. Protein-protein interaction assessment was conducted using the STRING database (https://string-db.org) [39] and modeled using Cytoscape [40].

2.3 Collection of patient tissues and clinical data

A total of 216 paraffin-embedded BLCA tissues were obtained from patients who underwent radical cystectomy at Peking University First Hospital (Beijing, China) between Jan 2007 and Dec 2012 (PKU-BLCA-IHC, Supplementary Table S1). All patients or their legal guardians signed an informed consent form. Comprehensive clinical and pathological information was available for all patients, and paraffin-embedded tissue specimens remained suitable for analysis. Of the 216 patients, 21 patients were lost to follow-up for overall survival information, and 22 patients were lost to follow-up for metastasis information. The median follow-up period was 35 months (range: 1-85). For isomer-resolved lipidomics, 27 paired snap-frozen BLCA and adjacent noncancerous tissues were obtained from patients who underwent radical cystectomy at Peking University First Hospital between April 2018 and May 2020 (PKU-BLCA-MS, Supplementary Table S1). Due to limited sample availability, only 23 of the 27 paired samples were included in the FA profiling analysis (from FA 12:0 to FA 24:1), and sufficient tissue for protein extraction to evaluate EMT marker expression was available for only 16 patients. EMT states of patients in PKU-BLCA-MS were defined by E-cadherin and vimentin (VIM) protein levels via Western blotting. All pathological information for the samples underwent verification by at least two pathologists. Tumor stage classification followed the 2017 UICC TNM classification of malignant tumors, and histological grade assessment adhered to the World Health Organization classification of 1973. These studies were conducted in accordance with approved guidelines, and all experimental protocols received approval from the ethics committee at Peking University First Hospital. Human BLCA (GFB-BLCA) tissue microarray containing 94 samples (Supplementary Table S1) and BRCA (GFB-BRCA) tissue microarray containing 194 samples (Supplementary Table S2) were purchased from Guilin Fanpu Biotech (Guilin, Guangxi, China), and the ethical approval for these samples was obtained through the Guilin Fanpu Biotech Ethics Committee. Six cases in BLCA tissue microarray did not have histological grade information. The interpretation criteria for estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER-2) in BRCA followed the clinical guidelines for the diagnosis and treatment of breast cancer from the Chinese Society of Clinical Oncology 2024 [41]. In the absence of in situ hybridization results for HER-2⁺⁺ cases, only HER-2⁺⁺⁺ cases were considered as positive.

2.4 Immunohistochemistry (IHC) staining and analysis

Paraffin-embedded tissues were cut into 5 µm sections and mounted on slides. After deparaffinization, antigen retrieval was performed using microwave treatment in ethylenediaminetetraacetic acid antigen retrieval buffer (Servicebio, Wuhan, Hubei, China). Endogenous peroxidase was blocked with hydrogen peroxide, and nonspecific binding was prevented with goat serum (Servicebio) blocking. Primary antibodies (listed in Supplementary Table S3) were applied overnight at 4°C, followed by incubation with secondary antibodies and subsequent diaminobenzidine staining (Servicebio). Nuclear counterstaining was performed using hematoxylin, and the slides were dehydrated before mounting (Servicebio). Each slide was scanned and graded based on the staining intensity of cytoplasmic FADS2 and SCD expression. The intensity was categorized into four levels: – (negative), + (weak), ++ (moderate) or +++ (strong).

2.5 Fatty acid and isomer profiling in human tissues

Total lipid extraction was performed following previously reported methods [17]. Each BLCA sample was weighed (approximately 20 mg), placed in a 10 mL centrifuge tube, mixed with 0.5 mL water and 0.5 mL methanol, and supplemented with internal standard [D4] FA 18:0 (60 nmol). The sample was homogenized by a handheld homogenizer (Jingxin Technology, Shanghai, China) at 40,000 Hz for 6 min. Subsequently, 0.5 mL methanol, 0.5 mL water, and 2 mL chloroform were added for liquid-liquid extraction. The mixture was vortexed for 3 min twice. Phase separation was achieved by centrifugation at 16,000 ×g for 10 min at 4°C. The bottom layer, containing the total lipid, was collected and transferred to a 10 mL glass vial. The extraction procedure was repeated once. The organic layers from the two extractions were combined and dried under nitrogen flow. The dried extracts were redissolved into 400 µL methanol and stored at -80°C before analysis.

Next, 125 µL of the total lipid extract in methanol was placed in a 1.5 mL tube and mixed with 125 µL 15% KOH aqueous solution for saponification. After saponification for 30 min at 37°C, 500 µL 1 mol/L HCl was added for acidification. The total FAs were extracted twice by 750 µL hexane. The collected organic layer was then dried under nitrogen and redissolved in 250 µL methanol for further derivatization.

The total fatty acids were subjected to the N-[4-(aminomethyl)phenyl]pyridinium (AMPP) derivatization, following the vendor's instructions (AMP+ MaxSpec Kit, Cayman Chemical, Ann Arbor, MI, USA). The solution containing 25 µL of total fatty acids was dried by nitrogen flow, redissolved in 10 µL 4:1 acetonitrile/dimethylacetamide solvent, 10 µL 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide solution (640 mmol/L in water), 5 µL N-hydroxybenzotriazole solution (20 mmol/L in acetonitrile / dimethylacetamide, 99:1, v/v), and 15 µL AMPP solution (20 mmol/L in acetonitrile). The mixture was incubated at 60°C for 30 min. After derivatization, 600 µL water was added. AMPP derivatized sample was extracted twice with 600 µL methyl tert-butyl ether, then dried under nitrogen stream and resuspended in 250 µL methanol for mass spectrometry analysis.

The solution containing total fatty acids (75 µL) was dried under nitrogen flow and redissolved in 150 µL 2-acetylpiridine solution (10 mmol/L in acetonitrile/dichloroform, 95:5, v/v). The solution was the subjected 254 to 254 nm UV-irradiation for 15 s using flow microreactor. The resulting reaction solution (50 µL) was washed by 600 µL 10 mmol/L HCl solution to remove the excess reagent. The Paternò-Büchi derivatized sample was extracted twice by 600 µL isooctane, dried under nitrogen flow, and resuspended in 50 µL methanol before mass spectrometry analyses.

Reversed-phase liquid chromatography-tandem mass spectrometry (RPLC-MS/MS) analyses were conducted on a Shimadzu LC-20AD system (Nakagyo-ku, Kyoto, Japan) hyphenated with an SCIEX X500R QTOF mass spectrometer (Framingham, MA, USA). Precursor ion scan was collected on a SCIEX QTRAP 4500 mass spectrometer. A C18 column (150 mm × 3.0 mm, 2.7 µm, Sigma-Aldrich, Burlington, MA, USA) was used for gradient elution. The mobile phase A contained water/acetonitrile (40:60, v/v, added with 20 mmol/L ammonium formate), while mobile phase B contained isopropanol / acetonitrile (40:60, v/v, added with 0.2% formic acid). The flow rate was set at 0.45 mL/min. Mass spectrometry parameters were optimized as follows: electrospray ionization voltage, 4500 V; curtain gas, 241 kPa interface heater temperature, 450°C; nebulizing gas 1 and gas 2, 206.84 kPa; declustering potential, 100 V, collision-induced dissociation energy for MS/MS, 20 eV; and CID energy for multiple reaction monitoring or product ion scan, 50 eV. The relative quantitation strategy was performed following reported methods [42]. Multiple reaction monitoring was used to quantify both saturated and unsaturated FAs, tracking the transition from the AMPP-modified FA [^AMPPFA]⁺ precursor ion to fragment ion at m/z 183.1, representing a characteristic fragment peak of AMPP modified FA. The total ion counts were normalized to the internal standard ([D4] FA 18:0) for relative quantitation at chain composition level. For relative quantitation of composition of C = C location isomers, the C = C-related distinct fragments, from the 2-acetylpiridine Paternò-Büchi derivatization products of unsaturated FA were utilized. The position of C = C is defined by the n (ω)- nomenclature, counting from the methyl terminus. The proportion distribution of certain isomer is calculated based on the summed ion abundances from each isomer.

2.6 Cell lines and cell culture

Authenticated human bladder carcinoma cell lines T24 (HTB-4; CL-0227) and SW780 (CRL-2169; CL-0449); human breast carcinoma cell lines MDA-MB231 (HTB-26; CL-0150) and MCF7 (HTB-22; CL-0149); human pancreatic adenocarcinoma cell lines PANC1 (CRL-1469; CL-0184) and BXPC3 (CRL-1687; CL-0042); and the highly transfectable derivative of the 293 cell line HEK-293T (CL-0005) cells were procured from the American Type Culture Collection (Manassas, VA, USA) and Pricella Biotechnology (Wuhan, Hubei, China), respectively. The murine bladder cancer cell line MB49 (iCell-m030) was obtained from Cellverse (Shanghai, China). Cells purchased in Pricella and Cellverse were verified by short tandem repeat genotyping. Cells were cultured in the recommended base media according to vendors' guidelines. T24 cells were maintained in McCoy's 5A (Gibco, Waltham, MA, USA), MDA-MB231, PANC1, HEK-293T, and MB49 cells in Dulbecco's Modified Eagle Medium (DMEM, Gibco), and SW780, BXPC3, and MCF7 cells in RPMI 1640 (Gibco). MCF7 cells received 0.01 mg/mL human recombinant insulin (Gibco). All media were supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% penicillin-streptomycin (Gibco), with incubation at 37°C in a 5% CO₂ environment.

For consistency in comparative analyses, all cells were unified in DMEM or RPMI 1640. To induce EMT in cancer cell lines, SW780 cells were treated with either 10 ng/mL transforming growth factor-β1 (TGF-β1) (PeproTech, Cranbury, NJ, USA) or 1× StemXVivo EMT Inducing Media Supplement (R&D Systems, Minneapolis, NE, USA) for 48 h. EMT was similarly induced MCF7 and BXPC3 cells using 1× StemXVivo EMT Inducing Media Supplement for 48 h. To evaluate the inhibition of FADS2-related expression and function, we replaced the standard 10% fetal bovine serum with 10% charcoal stripped fetal bovine serum (referred to as De-lipid, Gibco). Cells were treated with this De-lipid serum in combination with different compounds for 48 h. The treatments included 10 µmol/L SC26196 (Cayman Chemical) and 50 µmol/L FA 16:1 (n-10) (Matreya, Cayman Chemical). Tumor cells were seeded into six-well plates and allowed to adhere fully. Once attached, the culture medium was replaced with medium containing the appropriate drug-containing or specialized media, including XAV-939 (Cayman Chemical), Bortezomib (Medchemexpress, Monmouth, NJ, USA), SB431542 (Cayman Chemical), Rapamycin (Cayman Chemical), A769662 (Medchemexpress), De-lipid media (Gibco), and glucose-free media (Gibco). Cells were incubated with these treatments for 48 h in a cell culture incubator before subsequent experiments.

2.7 Cell migration assay

Transwell migration assays were conducted using 24-well transwell chambers with 8 µm pore-sized membranes (Corning, NY, USA). A total of 2 × 10⁵ cells, subject to various treatments, were plated in the upper chamber with medium lacking FBS, while the lower chamber contained medium supplemented with 20% FBS. Incubation times varied based on the migration ability of the cells: 8 h for comparison in EMT-paired cell lines, 6 h for T24 cells, 8 h for MDA-MB231 cells, and 24 h for SW780 cells. After incubation, the transwell membranes were fixed and stained with propidium iodide (Life Technologies, Carlsbad, CA, USA). Migrated cells were visualized and quantified using an FV3000 confocal microscope (Leica, Wetzlar, Germany) and analyzed using ImageJ software.

2.8 Cell adhesion assay

Collagen Type I-coated 96-well plates (Gibco) were used for the adhesion assay. After serum deprivation in DMEM for 8 h, cells were washed twice with phosphate-buffered saline (PBS) and then plated onto collagen-coated wells at a density of 2×10⁴ cells per well. Cells were allowed to adhere to the collagen-coated surface for 20 min, followed by three PBS washes to remove non-adherent cells. Both washed and non-washed wells were subsequently incubated with complete medium for 4 h. Following this incubation, all cell counts were performed 1 h post-treatment using 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium reagent (Abcam, Cambridge, UK). All procedures adhered to the manufacturer's instructions, and the adhesive cell percentage was determined by the ratio of washed to non-washed cells.

2.9 RNA extraction and reverse transcription-quantitative PCR (RT-qPCR)

Total RNA extraction from cells was performed using RNeasy Kits (Qiagen, Hilden, Germany). Reverse transcription was performed using the iScript cDNA Synthesis Kit (Bio-rad, Hercules, CA, USA) with the following conditions: 25°C for 5 min, 46°C for 20 min, and 95°C for 1 min. Quantitative PCR was performed using PowerUp SYBR Green Master Mix (Applied Biosystems, Waltham, MA, USA) at StepOne Plus Real-Time PCR System. The reaction conditions for the qPCR were: Uracil-DNA glycosylase activation at 95°C for 2 min, polymerase activation at 95°C for 2 min, followed by 40 cycles of denaturation at 95°C for 15 seconds and annealing/extension at 60°C for 1 min. Primers' sequences and product lengths were listed in Supplementary Table S4. Data analysis employed the ∆∆Ct method, with expression normalized to ribosomal protein lateral stalk subunit P0 (RPLP0).

2.10 FA 16:1 (n-10) and FA 16:1 (n-7) measurement in cell samples

The extraction and detection protocols were modified from a previous study [18]. Cells were extracted using a modified Bligh-Dyer extraction. To each extracted sample, 0.4 mL of ice-cold 65% methanol was added, followed by 1 min of vortexing. The samples were placed on ice, spiked with 1 µg of C17:0 margaric acid as an internal standard, and then pulse vortexed to mix. Next, 0.25 mL of chloroform was added, and the mixture was vortexed for an additional 5 min. The samples were centrifuged at 20,000 ×g for 5 min at 4°C. The bottom layer was collected for fatty acid analysis. The samples were dried using a rotary evaporation device at room temperature for 2 h. The samples were then derivatized for gas chromatography-mass spectrometry (GC-MS) analysis. Each sample received 0.5 mL of 14% boron trifluoride solution (Sigma-Aldrich) and incubated at 60°C for 30 min. Then 0.25 mL of water and 1 mL of hexane were added, followed by mixing. The samples were dried with approximately 0.2 g of anhydrous sodium sulfate. The hexane layer was then collected and dried using a stream of nitrogen. The final derivatized sample was reconstituted in 0.1 mL of hexane for GC-MS analysis.

The fatty acid methyl ester (FAME) composition for each sample was analyzed using a Thermo Fisher Triplus RSH autosampler and Trace 1310 gas chromatography system, coupled to a Thermo Fisher TSQ 8000 mass spectrometer. An Agilent Select FAME GC column (50 m × 0.25 mm, 0.2 µm film thickness) was used for the analysis (Agilent Technologies, Lexington, MA, USA). The GC carrier gas was helium with a linear flow rate of 1.0 mL/min. The programmed GC temperature gradient was as follows: The oven was held at 80°C, ramped to 175°C at a rate of 13°C/min with a 5-min hold, then ramped to 245°C at a rate of 4°C/min with a 2-min hold. The total run time was 38.3 min. The GC inlet was set to 250°C, and samples were injected in split-less mode. The MS transfer line and ion source were both set to 250°C. MS data were collected in selected ion monitoring (SIM) mode. All data were analyzed with Thermo Fisher Chromeleon software. A standard mixture of 37-Component FAME Mix (Sigma-Aldrich), FA 16:1 (n-10), and FA 16:1 (n-7) (Sigma-Aldrich) were used to confirm spectra and column retention times.

2.11 Lentivirus production and stable cell line development

The plasmids and lentiviruses used in this study were synthesized by Gentarget (San Diego, CA, USA) and Syngenbio (Beijing, China). To generate lentiviral vectors targeting FADS2, short hairpin RNA (shRNA) sequences specific for FADS2 (shFADS2 #1, TRCN0000064755, 5'-CCACGGCAAGAACTCAAAGATCTCGAGATCTTTGAGTTCTTGCCGTGG-3'; shFADS2 #2, TRCN0000064757, 5'-CCACCTGTCTGTCTACAGAAACTCGAGTTTCTGTAGACAGACAGGTGG-3') [18], and a non-targeting negative control (shNC, 5'-GTCTCCACGCGCAGTACATTTCGAGAAATGTACTGCGCGTGGAGAC-3') were synthesized as DNA oligonucleotides. These sequences were then sub-cloned into the pLKO.1 shRNA lentivector under the U6 promoter. For FADS2 overexpression, the human FADS2 coding sequence (NM_004265.4) was subcloned into the pLVX expression lentivector. This insert was designed to be expressed with a C-terminal S-tag under a constitutive cytomegalovirus promoter. Both the shRNA and overexpression lentivectors were constructed with a puromycin resistance marker. Lentiviral particles were produced by transfecting HEK-293T cells and quantified before infecting cancer cells at a multiplicity of infection of 10, followed by selection with puromycin.

2.12 Protein extraction and Western blotting

The total cellular protein was extracted using cell lysis buffer (Invitrogen, Waltham, MA, USA) with 1× protease & phosphatase inhibitor cocktail (Thermo Fisher, Waltham, MA, USA), and 1× dithiothreitol (Invitrogen). Protein concentrations were quantified via a bicinchoninic acid assay (Thermo Fisher). The lysates and prestained protein ladder (Thermo Fisher) were then separated on a Bis-Tris plus gel (Invitrogen) and electroblotted onto polyvinylidene difluoride membranes. After incubation with the primary and secondary antibodies (Supplementary Table S3) in sequence, the bands were visualized using enhanced chemiluminescence (SuperSignal West Pico PLUS, Thermo Fisher). Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) served as a loading control for total and cytoplasm protein, while Na,K-ATPase was utilized as a control for membrane protein.

2.13 Cell viability assay

A total of 1.5 × 10⁵ cancer cells were seeded into a well of a 6-well plate and treated with various inhibitors (10 µmol/L A939572, CAY10566, or SC26196, Cayman Chemical) for 48 h. Cells were fixed and stained with a 0.5% crystal violet solution in methanol for 30 min. After washing with PBS and air-drying, the plates were scanned using a Perfection V370 scanner (Epson, Los Alamitos, CA, USA). Crystal violet was solubilized with 10% SDS, and absorbance at 580 nm was measured by SpectraMax i3x Microplate Detection Platform (Molecular Devices, San Jose, CA, USA). The absorbance values directly correlated with viable cell numbers. Relative viability was quantified after normalization to the untreated control.

2.14 SC26196-loaded nanoparticle (NP-SC26196) synthesis and quantification

The cationic lipid 76-O17Se was synthesized, purified, and characterized following our previously reported procedure [43].

The typical procedures for encapsulating drug molecules into lipid nanoparticles are as follows [44]: 1 mg of 76-O17Se was dissolved in 100 µL of dimethyl sulfoxide (DMSO). 80 µL of 10 mmol/L SC26196 (MW 423.55 Da; 339 µg in total) stock solution in DMSO was mixed with 76-O17Se DMSO solution. Resulting in a total volume of 180 µL. Then 420 µL of distilled water was added into DMSO solution at room temperature, with continuous stirring using a magnetic stirrer bar. The SC26196-loaded lipid nanoparticle solution was then transferred into a dialysis cassette (Thermo Scientific Slide-A-Lyzer Dialysis Cassette, MWCO = 3500 Da) and dialyzed against distilled water for 24 h to remove the DMSO and free SC26196. Water outside was changed every 4 h. After dialysis, SC26196-loaded lipid nanoparticle solution was stored at 4°C for further use.

The hydrodynamic size and size distribution of both blank and drug-loaded 76-O17Se nanoparticles were examined by dynamic light scattering [45]. The average hydrodynamic diameters of the blank and SC26196-loaded 76-O17Se lipid nanoparticles were 252.96 ± 11.76 nm and 248.63 ± 13.01 nm, respectively. The polydispersity index values of the blank and drug-loaded nanoparticles were 0.295 ± 0.03 and 0.303 ± 0.019, respectively.

The amount of lipid nanoparticle encapsulated SC26196 was measured by spectrophotometry. A series of SC262196 drug solutions in DMSO/H₂O (9:1, v/v) were prepared (2.11, 4.23, 10.6, 21.2, 31.7, and 42.3 µg/mL), and their UV-Vis absorption curves were recorded by the Spectra Max microplate reader (Molecular Devices). Absorbance at 310 nm was plotted against SC26196 concentration to generate the standard curve. The SC26196-loaded lipid nanoparticles solution was diluted in 9 equivalent volumes of DMSO, and its absorbance at 310 nm was measured. The drug loading content [loaded drug / (loaded drug + lipid) × 100%] was determined to be 77.5 / (77.5 + 1000) = 7.19% w/w, and the drug loading efficacy (loaded drug / total drug × 100%) was determined to be 77.5 / 339 = 22.86%.

2.15 IC₅₀ measurement of nanoparticles

To determine the IC₅₀ for the nanoparticles, cells were seeded at a density of 5 × 10³ cells per well in a 96-well plate. After 24 h, the cells were exposed to NP-SC26196 or a control nanoparticle at concentrations ranging from 0.625 µmol/L to 10 µmol/L. Following 48 h of treatment, cell numbers were assessed using the 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium reagent (Abcam).

2.16 16 siRNA transfection

All siRNAs were synthesized by Syngenbio, and the detailed sequences were provided in Supplementary Table S5. For siRNA transfection experiments, 3.5 × 10⁵ cells were seeded into each well of a 6-well plate. After the cells adhered sufficiently, transfection was performed using Lipofectamine RNAiMAX (Thermo Fisher), according to the manufacturer's protocol. In the gene knockdown group, three siRNA sequences targeting the same gene were combined for the transfection. The final transfection mixture contained 10 pmol of siRNA diluted in 150 µL of Opti-MEM(Gibco), combined with 4 µL of Lipofectamine diluted in 150 µL of Opti-MEM. This mixture was incubated at room temperature for 15 min before being added to the cells. The cells were then incubated at 37°C in a 5% CO₂ incubator for 24 h. After that, the transfection medium was replaced with complete medium, and the cells were cultured for an additional 48 h before being harvested for subsequent experiments.

2.17 Hyperspectral SRS imaging

For hyperspectral SRS imaging, cells were cultivated on glass-bottom dishes (Cellvis, Mountain View, CA, USA) under various treatment conditions. Prior to imaging, the cells were fixed with 10% neutral buffered formalin (Sigma-Aldrich) for 30 min and washed three times with PBS. Hyperspectral SRS imaging employed a spectral focusing approach, where the Raman shift was tuned by controlling the temporal delay between two chirped femtosecond pulses. The pump beam and Stokes beam were tuned to 798 nm and 1,040 nm, respectively, to cover the C-H vibration region. The Stokes beam was modulated by an acousto-optic modulator (AOM, 1205-C, Isomet, Manassas, VA, USA) at 2.2 MHz. Both beams were chirped by six 12.7 cm long SF57 glass rods and directed to a laser-scanning microscope. A 60× water immersion objective lens (NA = 1.2, UPlanApo/IR, Olympus, Center Valley, PA, USA) focused the light on the sample, and an oil condenser (NA = 1.4, U-AAC, Olympus) collected the signal.

To acquire a hyperspectral SRS image, a stack of 70 images at different pump-Stokes temporal delays was captured. The temporal delay was controlled by an automatic stage moving forward with a step size of 10 µm, corresponding to approximately 5 cm⁻¹ Raman shift. Standard chemicals with known Raman peaks in the C-H region (2,800-3,050 cm⁻¹), including DMSO and glyceryl trioleate, were used to calibrate the Raman shift to the temporal delay. Image analysis was performed using ImageJ, and no cell damage was observed during the imaging process. Phasor analysis in ImageJ was applied to separate different chemicals in the same window based on their spectral profiles.

2.18 Raman spectromicroscopy

For Raman imaging verification, cancer cells were seeded into glass bottom dishes and treated with 50 µmol/L of fatty acids. The cells were incubated for 48 h before obtaining the Raman spectra from individual lipid droplets (LDs). Individual LDs within the cells were analyzed using a confocal Raman microscope (LabRAM HR Evolution, Horiba, Irvine, CA, USA) equipped with a 40× water immersion objective (Apo LWD, 1.15 N.A., Nikon, Melville, NY, USA). The samples were excited by a 532 nm laser. The Raman intensity of the entire spectrum was normalized to the region between 1,400 cm⁻¹ and 1,500 cm⁻¹, corresponding to the -CH₂ bending vibration region.

2.19 Phospholipid mapping in cell sample

Lipids were extracted using a modified Bligh-Dyer extraction method. The dried lipid extracts were diluted into 200 µL injection solvent (acetonitrile/methanol/ammonium acetate, 3:6.65:0.35, v/v/v) to prepare a stock solution. The stock solution was further diluted 50× into injection solvent spiked with 0.1 ng/µL of Equisplash Lipidomics (Avanti Polar Lipids, Alabaster, AL, USA) for sample injection. Lipid analysis was performed by flow-injection mass spectrometry (without chromatographic separation) using 8 µL of the diluted lipid extract stock solution, delivered via a micro-autosampler (G1377A, Agilent Technologies) to the electrospray ionization source of an Agilent 6410 triple quadrupole mass spectrometer (Agilent Technologies). The capillary pump was connected to the autosampler and operated at a flow rate of 7 µL/min with a pressure of 100 bar. Capillary voltage on the instrument was 5 kV and the gas flow 5.1 L/min at 300°C. The multiple-reaction monitoring-profiling methods and instrumentation used were recently described [46, 47].

2.20 Lipidomics analysis in cell sample

Lipid extraction was performed by adding 0.75 mL of methanol to a 100 µL cell sample, followed by 2.5 mL of methyl tert-butyl ether and 10 µL of SPLASH internal standard (Avanti Polar Lipids). The mixture was incubated for 1 h at room temperature. Phase separation was induced by adding 0.625 mL of MS-grade water. After centrifugation at 1,000 ×g for 10 min, the upper organic phase was collected, re-extracted with 1 mL of tert-butyl ether/methanol/water (10:3:2.5, v/v/v), and combined with the initial organic phase. The organic phases were dried and redissolved in 100 µL of isopropanol for storage. Samples were analyzed by ultra-high-performance liquid chromatography coupled with tandem mass spectrometry (UHPLC-MS/MS) using a Vanquish UHPLC system coupled to an Orbitrap Q Exactive™ HF mass spectrometer (Thermo Fisher) at Novogene Co., Ltd (Beijing, China). Data analysis included peak alignment, peak picking, and quantitation using LipidSearch (Thermo Fisher Scientific). Partial least squares discriminant analysis and univariate analysis (t-test) were performed at metaX [48]. Lipids with a |log₂FC| greater than 1, P value less than 0.05, and variable importance in the projection greater than 1 were considered to be differential metabolites.

2.21 Cell membrane fluidity assay

Membrane fluidity was assessed using MarkerGene Membrane Fluidity Kit (Abcam), following the manufacturer's protocol. Briefly, 5 × 10³ cells per well were seeded in 96-well plates. After 24 to 48 h, cells were rinsed with PBS and incubated with 10 µmol/L of pyrenedecanoic acid in a Perfusion Buffer containing 0.08% pluronic F127 for 20 min at 25°C in the dark. Unincorporated pyrenedecanoic acid was removed by washing the cells with serum-free media twice. The incorporated PDA was quantified by measuring fluorescence signals at both 400 nm (monomer) and 460 nm (excimer) with excitation at 360 nm using the SpectraMax i3x Microplate Detection Platform (Molecular Devices). The membrane fluidity was expressed as a ratio of excimer fluorescence to monomer fluorescence, with a higher ratio indicating increased membrane fluidity.

2.22 Membrane protein separation and extraction

Membrane and cytoplasm proteins were separated and extracted using the Mem-PER Plus (Thermo Scientific). First, 5 × 10⁶ cells were harvested by centrifugation at 300 ×g for 5 min. The cell pellet was then washed with 3 mL of Cell Wash Solution and centrifuged at 300 ×g for 5 min. After discarding supernatants, the cell pellet was resuspended in 1.5 mL of Cell Wash Solution and transferred to a 2 mL centrifuge tube. Further centrifugations and washes were performed, followed by the addition of 0.75 mL of Permeabilization Buffer to the cell pellet. After a brief vortex and a 10-min incubation at 4°C, the separation of cytosolic proteins was achieved through centrifugation. The resulting supernatant was carefully transferred to a new tube. Solubilization Buffer (0.5 mL) was added to the pellet, followed by resuspension via pipetting, and incubated for 30 min at 4°C with constant mixing. Finally, samples were centrifuged at 16,000 ×g for 15 min at 4°C. The supernatant containing solubilized membrane and membrane-associated proteins was transferred to a new tube.

2.23 Popliteal lymph node metastatic mouse model with FADS2 inhibition

All animal procedures in this part were ethically approved by the Boston University Institutional Animal Care and Use Committee under protocol PROTO201800533. Mice were housed in a controlled environment with a 12 h light/dark cycle, with temperature maintained at 22 ± 2°C and relative humidity at 50 ± 10%. Mice had free access to standard rodent chow and water ad libitum. Mice were monitored daily for signs of distress, including inability to eat or drink, severe respiratory distress, or large tumor size impairing movement. Any mouse exhibiting these signs was euthanized to prevent unnecessary suffering.

The mouse popliteal lymph node metastatic model was used to study the development of lymph node metastasis. Cancer cells were injected into the hock, and metastasis was evaluated by examining the popliteal lymph nodes for tumor metastasis. The popliteal lymph nodes were collected for analysis because cancer cells naturally spread from the injection site at the hock to these lymph nodes through the lymphatic system, providing a way to monitor the progression of metastasis. Six-week-old male NU/J nude mice (Jackson Laboratory, Bar Harbor, ME, USA) were used for the study. Luciferase-labeled T24 cells, transfected with lentivirus carrying either negative control shRNA (T24-Luc shNC) or shFADS2 targeting shRNA (T24-Luc shFADS2 #2), were inoculated into the hock of the mice (5 × 10⁶ cells per mouse). After six weeks of tumor development, tumor lesions were visualized and quantified using the IVIS Spectrum system (PerkinElmer, Waltham, MA, USA) following anesthesia with isoflurane and intraperitoneal injection of D-luciferin sodium salt (Biovision, Milpitas, CA, USA).

Mice were euthanized using an overdose of isoflurane (5%) inhalation followed by cervical dislocation to ensure death. The euthanasia procedures were performed in accordance with institutional animal care and use guidelines and were approved by the Institutional Animal Care and Use Committee at Boston University. Popliteal lymph nodes were collected and subjected to hematoxylin and eosin (HE) staining. After enucleating the popliteal lymph nodes, photographs were taken, and the longest diameter of each lymph node was measured. To estimate the lymph node volume, we approximated each lymph node as a spherical body using the formula for the volume of a sphere.

2.24 Lung metastasis mouse model with FADS2 inhibition

All animal procedures in this part were conducted in accordance with the Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of Peking University First Hospital. The information about the conditions for housing, humane endpoint, and euthanasia method for the mice were described above.

The mouse tail vein injection model was employed to investigate lung metastasis development. Six-week-old B-NDG mice (Biocytogen, Beijing, China) were divided into two groups, each consisting of six mice. Cells (T24-Luc shNC or T24-Luc shFADS2 #2, 1 × 10⁶ cells per mouse) were injected into the lateral tail vein, and tumor lesions were visualized and quantified as previously described. After six weeks of tumor development, lung metastases were collected and subjected to HE staining.

2.25 Lung metastasis mouse model with SCD inhibition

All animal procedures in this part were conducted in adherence to the Guidelines for the Care and Use of Laboratory Animals and received approval from the Institutional Animal Care and Use Committee of Peking University First Hospital. Luciferase-labeled MB49 cells (5 × 10⁵ per animal) were injected into the lateral tail vein of six-week-old BALB/c mice (Biocytogen). The mice were then divided into two groups: a control group receiving intraperitoneal vehicle (10% DMSO + 90% corn oil) and a treatment group receiving intraperitoneal A939572 (25 mg/kg; 10% DMSO + 90% corn oil) for two weeks, administered once daily.

Tumor lesions were visualized and quantified as previously described. After an additional week (a total of three weeks), the animals were sacrificed, and lungs were harvested and snap-frozen for subsequent HE staining and SRS imaging. Note: One mouse in the A939572 group accidentally perished during model development.

2.26 Animal tissue imaging with picosecond SRS

Frozen tissue slices were washed thoroughly with PBS to eliminate optimal cutting temperature compound residue and subsequently fixed in 10% formalin for 15 min. Picosecond SRS imaging was conducted using a picoEmerald system (Applied Physics & Electronics, Berlin, Germany), featuring synchronized pump (tunable wavelength 700-990 nm and 80 MHz repetition rate) and Stokes beams (fixed wavelength at 1,031 nm and 2-ps pulse width) collinearly coupled into an upright microscope (BX51WI, Olympus) equipped with a 60× 1.2 NA water objective (Olympus). SRS images were acquired at 2,850 cm-¹ with a pixel dwell time of 10 µs. The power at the sample was 46 mW and 73 mW for the pump and Stokes beams, respectively, with no observed photodamage.

SRS data were collected from about three tissue slices per group, with 6-7 field views imaged per slice. Quantitative analysis of lipid area was performed by ImageJ. Lipid and cell areas were calculated by adjusting the threshold value and analyzing particles with the percentage area of lipids relative to cells determined in each sample.

2.27 Dual-luciferase reporter assay

The plasmids used in the dual-luciferase reporter assay were synthesized by Syngenbio. In the dual-luciferase reporter assay, the FADS2 wild-type promoter (hg38_refGene_NM_004265, range = chr11:61826300-61828299) and a mutant version (with three potential binding sites deleted: 5'-AGGGTTCAAAGCCCT-3' / 5'-ATAGCTGAAAGGCAG-3' / 5'-AACCTTCAAAGCCTC-3') were subcloned into the Firefly plasmid (pGL4.10). The coding sequences of human lymphoid enhancer-binding factor 1 (LEF1, NM_016269) were subcloned into a constitutive expression plasmid (pcDNA3.1). Various plasmid combinations, including FADS2 promoters, the LEF1 expression plasmid, the Renilla luciferase plasmid (pRL-TK), and corresponding control plasmids, were transiently transfected into HEK-293T cells using Lipofectamine 3000 (Invitrogen). For transfection, HEK-293T cells were seeded in a 24-well plate at a density of 8 × 10⁴ cells per well in 500 µL of complete growth medium. On the day of transfection, DNA-lipid complexes were prepared by mixing 25 µL of Opti-MEM I medium with 1.5 µL of Lipofectamine 3000 reagent and 500 ng of DNA with 1 µL of P3000 reagent. After a 15-min incubation, the complex was added to the cells.

Following a 48-hour transfection period, luciferase activities were assayed using the Dual Luciferase Reporter Gene Assay Kit (Beyotime, Haimen, Jiangsu, China). Cells were lysed using the reporter gene cell lysis buffer. The Firefly luciferase assay reagent and Renilla luciferase assay buffer were thawed to room temperature, and the Renilla luciferase assay working solution was prepared by diluting the substrate (100×) in the buffer at a 1:100 ratio. Luminescence was measured using the iMark microplate absorbance reader (Bio-Rad) with a 2-second interval and a 10-second measurement time.

For each sample, 20-100 µL of cell lysate was used consistently across samples. To measure Firefly luciferase activity, 100 µL of Firefly luciferase assay reagent was added to the lysate, mixed, and the relative light unit was measured. Then, 100 µL of Renilla luciferase assay working solution was added to the same tube, mixed, and the relative light unit was measured. The Firefly luciferase values were normalized to the Renilla luciferase values to compare the activation levels of the target reporter gene across samples.

2.28 RNA-seq of cancer cells in vitro

The experiments were conducted in biological triplicates. In the OE-FADS2 group, cells were seeded in a 6-well plate for 48 h before harvesting, and sequencing was performed by Genewiz (South Plainfield, NJ, USA). In the A939572 groups, cells were treated with the inhibitor or DMSO for 48 h before harvesting, and subsequent sequencing was conducted by Novogene Co., Ltd.

Generally, total RNA was extracted, and RNA concentration and integrity were assessed using the RNA Nano 6000 Assay Kit with the Bioanalyzer 2100 system (Agilent). mRNA was purified from total RNA using poly-T oligo-attached magnetic beads. After fragmentation, the first strand of cDNA was synthesized using random hexamer primers, followed by second strand cDNA synthesis. mRNA-seq libraries were prepared sequentially following the manufacturer's protocol. The libraries were quantified using Qubit (Invitrogen) and real-time, and size distribution was analyzed using the Bioanalyzer. Quantified libraries were pooled and sequenced on Illumina platforms, generating 150 bp paired end reads.

Sequencing reads were aligned to the human genome (GRCh38). featureCounts [49] were employed to count the reads numbers mapped to each gene. mRNA expression values were transformed into fragments per kilobase of transcript per million mapped reads. Differential expression analyses between the defined groups were conducted using the R package DESeq2. DEGs were included in the further enrichment analysis based on their log₂ fold change (log₂FC) values. Genes with an |log₂FC| greater than 1 and P_adjusted value less than 0.05 were included in further analysis. In cases where the number of genes was excessively high, the threshold was adjusted to an |log₂FC| greater than 1.2. Specific adjustments and criteria are detailed in the relevant figure legends.

2.29 Immunofluorescence assay

Cells were cultured on glass coverslips for 24 h and then categorized into different groups. Following the treatments, the coverslips were fixed with 4% paraformaldehyde for 30 min at room temperature and permeabilized with 0.2% Triton. The cells on the coverslips were subsequently probed with the primary antibody and the corresponding fluorescent secondary antibody (Supplementary Table S3). The coverslips were then mounted using antifade mounting medium with 4′,6-diamidino-2-phenylindole (DAPI, Vector Laboratories, Newark, CA, USA). All slides were examined using an FV3000 confocal microscope (Leica, Germany).

2.30 Oxylipin profiling in cell sample

This part of the experiment was analyzed by Profleader (Shanghai, China). The cell suspension sample (T24 shFADS2 #2 and shNC) underwent a processing step involving 5 cycles of ultrasonication (1 min) with intervals of 1 min in an ice-water bath. The mixture was then incubated at -20°C for 30 min, followed by centrifugation at 4°C and 15,000 ×g for 15 min. The resulting supernatant was dried under nitrogen and redissolved in 1 mL of 10% methanol, supplemented with 10 µL of internal standard solution. This mixture was used for subsequent extraction using a Strata-X SPE column (Phenomenex, Torrance, CA, USA). The columns were activated with methanol, equilibrated with water, and the sample was loaded and washed with 10% methanol to remove impurities. Finally, elution was performed with 1 mL methanol. The eluates were dried under nitrogen gas and reconstituted in 20 µL of methanol prior to mass spectrometry analysis. A quality control sample was obtained by isometrically pooling all prepared samples.

All oxylipins (oxidized fatty acids) and deuterated standards as internal standards were procured from Cayman Chemical. The standard solution in ethanol was mixed, dried, and redissolved in a 1/2 concentration of internal standards in sample preparation to obtain a stock solution for the calibration curve, with concentrations of 1 or 2 µg/mL in methanol. Calibration curve samples were prepared by serially diluting the stock solution of the calibration curve with a 1/2 concentration of internal standards in sample preparation. The final concentrations in the calibration curve samples ranged from 0.002 to 1 µg/mL or from 0.004 to 2 µg/mL.

UHPLC-MS/MS was conducted using an Agilent 1290 Infinity II UHPLC system coupled to a 6470A Triple Quadrupole mass spectrometry (Agilent). Samples were injected into a Waters Acquity BEH C18 column (100 mm × 2.1 mm, 1.7 µm) at a flow rate of 0.3 mL/min. The mobile phase consisted of 0.05% acetic acid and acetonitrile/isopropanol (50/50, v/v). Chromatographic separation was achieved using a gradient elution program. Eluted analytes were ionized in an electrospray ionization source in negative mode. The source drying gas and sheath gas temperatures were set to 300°C and 350°C, respectively. The flow rates of the source drying gas and sheath gas were 5 and 11 L/min, respectively. The nebulizer pressure was 275.79 kPa, and the capillary voltage was 2000 V. Dynamic multiple reaction monitoring was employed for data acquisition.

The raw data were processed using MassHunter Workstation Software (Agilent) with default parameters, supplemented by manual inspection to ensure qualitative and quantitative accuracy for each compound. Peak areas of each compound were integrated, and nine-point calibration curves were constructed by plotting the peak area ratio of each compound to the internal standard against the concentration of each compound. The concentrations of oxidized fatty acids in the prepared samples were automatically quantified and output for the quantitative calculation of tube cell samples, subsequently normalized to the protein level in each sample.

2.31 Isotope metabolite labeling with Femtosecond SRS imaging

To visualize and quantify fatty acid uptake, 50 µmol/L of deuterated palmitic acid (PA-D_31, Cambridge Isotope Laboratories, Tewksbure, MA, USA) was supplemented in the complete medium, and cells were cultured for 6 h. SRS imaging was performed using a femtosecond SRS microscope. The ultrafast laser system, with dual output at 80 MHz (InSight DeepSee, Spectra-Physics, Milpitas, CA, USA) provided pump and Stokes beams. The pump and Stokes beams were set to 802 nm and 1,045 nm, respectively, resonant with the C-H vibration band at 2,899 cm⁻¹. For the C-D vibration band at 2,168 cm⁻¹, the pump beam was tuned to 852 nm. The Stokes beam was modulated by an acoustooptic modulator at 2.2 MHz. Both beams were linearly polarized.

A motorized translation stage was used to scan the temporal delay between the two beams. These beams were then directed into a home-built laser-scanning microscope. A 60× water immersion objective lens (NA = 1:2, UPlanApo/IR, Olympus) focused the light into the sample, and an oil condenser (NA = 1:4, UAAC, Olympus) collected the signal. Stimulated Raman loss and pump-probe signals were detected by a photodiode and extracted with a digital lock-in amplifier (Zurich Instrument, Zurich, Switzerland). The power of the pump beam (802 nm and 852 nm) and the Stokes beam at the specimen were maintained at ∼30 mW and 100 mW, respectively. Images were acquired at a 10 µs pixel dwell time. No cell damage was observed during the imaging procedure. To quantify the SRS intensity at the C-D region, the total intensity was measured and normalized by the number of cells in the corresponding image using ImageJ.

2.32 TGF-β2 enzyme-linked immunosorbent assay

Total cellular protein was extracted and quantified as mentioned above. TGF-β2 level was measured using the Human TGF-beta 2 Quantikine enzyme-linked immunosorbent assay Kit (R&D Systems) following the manufacturer's protocol. The assay involves the activation of latent TGF-β2 to its immunoreactive form using acid and neutralization solutions. After activation, the samples underwent a series of steps: incubation with assay diluent, addition of standards or samples, conjugate incubation, substrate solution addition, and stop solution application.

Optical density was measured at 450 nm using the SpectraMax i3x Microplate Detection Platform (Molecular Devices), and results were calculated by averaging duplicate readings, subtracting the average standard optical density, and using a standard curve. Concentrations for activated samples were normalized by protein level.

2.33 Statistical analysis

Statistical analysis was performed using GraphPad Prism. Data comparisons between groups were conducted using an unpaired two-tailed t-test, nonparametric test, or Chi-square test, depending on the nature of the data. For survival analysis, the log-rank (Mantel-Cox) test was employed. For overall survival, the endpoint was defined as the time from the date of surgery to death from any cause. Patients still alive at the time of analysis were censored at the date of the last follow-up. For metastasis-free survival, the endpoint was defined as the time from the date of surgery to the occurrence of the first distant metastasis or death, whichever occurred first. Patients without any metastasis or death events were censored at the last follow-up date. A P-value < 0.05 was considered statistically significant.

3.1 Clinical evidence for desaturation reprogramming from SCD to FADS2 in cancer EMT and progression

To address the coordination of SCD and FADS2 during cancer development, we analyzed cancer cohorts from TCGA. Positive correlations between SCD and FADS2 expression were observed across various cancer types (Supplementary Figure S1A), indicating potential synergistic effects. Patients from five TCGA cohorts were classified according to the median expression levels of SCD and FADS2 (Figure 1A, Supplementary Figure S1B). Notably, the intersections of DEGs were obtained only in upregulated genes of FADS2^highSCD^low and FADS2^lowSCD^high patients (Figure 1B, Supplementary Figure S1C-E and Supplementary Tables S6-S7). Functional enrichment analysis revealed significant enrichment of pathways related to EMT and cancer metastasis among the consensus upregulated genes in FADS2^highSCD^low patients (Figure 1B and Supplementary Table S8). GSEA further confirmed that EMT-related gene sets were enriched in FADS2^highSCD^low cancer patients across multiple cancer types, particularly those originating in hollow organs (Figure 1C). The EMT states across cancer patients from the TCGA were assessed separately using Creighton and Byers EMT scores [36]. Higher EMT scores, indicating a more mesenchymal-like phenotype, were consistently observed in FADS2^highSCD^low patients compared to FADS2^lowSCD^high patients in both scoring systems (Figure 1D). EMT markers and pathways [50] were differentially expressed between these groups. FADS2^highSCD^low patients exhibited upregulated mesenchymal and EMT-promoting signatures, while epithelial signatures were downregulated (Supplementary Figure S1F). Using the Creighton and Byers EMT scores, cancer patients were further stratified into epithelial-like (EMT score < 0) and mesenchymal-like (EMT score > 0) subtypes. A significant decrease in the SCD:FADS2 ratio was observed in mesenchymal-like cancer patient groups across various cancer types (Figure 1E), highlighting the strong association between SCD-to-FADS2 desaturation reprogramming and the EMT state.

It is well established that the dynamic EMT in cancer is closely linked to metastatic potential, which significantly impacts patient outcomes [20]. High FADS2 expression was associated with advanced clinicopathologic features and unfavorable prognosis in BLCA patients (Supplementary Figure S2A-C). IHC analysis of BLCA patients (Supplementary Table S1) confirmed that higher FADS2 protein levels were associated with advanced pathological stage and higher histologic grade (Figure 1F-G, Supplementary Table S9). Furthermore, elevated FADS2 expression was significantly associated with worse overall survival and metastasis-free survival in BLCA patients (Figure 1H). Additionally, SCD in BLCA did not show significant associations with clinicopathologic features at either the transcriptomic and proteomic levels (Figure 1G, Supplementary Figure S2A-D, and Supplementary Table S10). In BRCA, opposite expression patterns of FADS2 and SCD were observed in various clinicopathologic and molecular subtypes. FADS2 upregulation was prominent in patients with triple-negative breast cancer (TNBC) and basal-like cancers, which are known to exhibit more aggressive behavior and poorer prognosis, whereas SCD expression was increased in non-triple-negative (nTNBC) and luminal patients (Supplementary Figure S2E-F). IHC validation revealed higher FADS2 expression and lower SCD expression in TNBC compared to nTNBC (Figure 1I-J, Supplementary Figure S2G, Supplementary Table S11). Notably, the association between FADS2 and SCD was significantly weaker in TNBC than in nTNBC, with the correlation coefficient decreasing from 0.466 to 0.260 (Figure 1K). This decrease in correlation was also validated by IHC, with the correlation coefficient dropping from 0.385 to 0.351 (Supplementary Figure S2H). These findings suggested metabolic reprogramming between FADS2 and SCD during cancer progression and metastasis.

Using isomer-resolved lipidomics [16], we further identified and quantified the FAs desaturated by FADS2 and SCD in BLCA tissues. A range of FAs from FA 12:0 to FA 24:1 was profiled in paired paracancer and cancer tissues. Notably, FA 18:1 and FA 18:3 were reduced in cancer tissues, while FA 20:3 was increased. Only FA 16:1 levels were decreased in advanced pathologic stages and histologic grades (Supplementary Figure S3). Subsequently, the C = C bond isomers at FA 16:1, FA 18:1 and FA 18:3 were further detected (Supplementary Figure S4A-B). Notably, the proportions of the FADS2 metabolites, FA 16:1 (sapienic acid, n-10) and FA 18:3 (γ-Linoleic acid, n-12), were significantly increased in carcinoma tissues, while SCD metabolites, FA 16:1 (palmitoleic acid, n-7) and FA 18:1 (oleic acid, n-9), were decreased (Figure 1L-M and Supplementary Figure S4C). Additionally, levels of metabolic derivatives from SCD metabolites, such as FA 16:1 (n-9) and FA 18:1 (vaccenic acid, n-7), were elevated, indicating the enhanced transformation of SCD metabolites in cancer. Isomer analyses revealed that only FADS2 metabolites, FA 16:1 (n-10) and FA 18:1 (n-10), were selectively increased in advanced and mesenchymal subgroups (Figure 1N-O and Supplementary Figure S4D-E). Collectively, the isomer analyses validated the reprogramming between FADS2 and SCD during cancer progression, with a marked promotion of FADS2 in cancer.

3.2 SCD-to-FADS2 reprogramming interplay in cancer EMT

Given multiple lines of evidence indicating a metabolic reprogramming from SCD to FADS2 in cancer tissues, we sought to validate these observations in vitro. Tan EMT scores [37] were utilized to characterize the EMT states of different cancer cells from the CCLE and GDSC databases. Tan EMT states were defined from epithelial (EMT score = -1.0) to mesenchymal states (EMT score = +1.0) (Supplementary Figure S5A). FADS2^highSCD^low cell lines exhibited higher EMT scores than did FADS2^lowSCD^high cell lines (Figure 2A). Consistent with TCGA results, FADS2 expression was upregulated in mesenchymal-like (Tan EMT score > 0) cells and was positively correlated with Tan EMT score, while SCD showed the opposite trend (Figure 2B and Supplementary Figure S5B). The expression patterns of the ELOVL fatty acid elongases (ELOVL) of PUFAs, ELOVL2, ELOVL4 and ELOVL5 [51], also resembled that of FADS2 (Supplementary Figure S5B), suggesting a dominant role of PUFA metabolism in EMT. In addition, decreased SCD:FADS2 ratios were observed in cell lines with advanced features (Figure 2C).

We further validated the SCD-to-FADS2 reprogramming in vitro using three EMT-paired cancer cell lines from various cancer types (Supplementary Figure S5C-E). All three mesenchymal cells displayed downregulated SCD:FADS2 ratio compared to their epithelial counterparts (Figure 2D, Supplementary Figure S5F-G). Monounsaturated metabolites FA 16:1 (n-10) and FA 16:1 (n-7) were separated and quantified by GC-MS, confirming elevated FA 16:1 (n-10):(n-7) ratios in mesenchymal cancer cells (Supplementary Figure S6A). The FA 16:1 (n-10):(n-7) ratio of 0.25 in FBS was used to exclude the potential exogenous contributions (Supplementary Figure S6B). Mesenchymal cells exhibited FA 16:1 (n-10):(n-7) ratios greater than 0.25, whereas epithelial cells showed lower ratios (Figure 2E and Supplementary Figure S6C). Similar assessments were also performed in TGF-β1-induced SW780 cells. TGF-β1 effectively induced partial EMT in SW780 cells (Supplementary Figure S7A-B), followed by increased migration ability (Supplementary Figure S7C). Similar to VIM expression, FADS2 was upregulated by TGF-β1 (Figure 2F-G). Meanwhile, SCD expression was significantly mitigated by TGF-β1, particularly in RPMI 1640 (Figure 2G-H), resulting in an increased FA 16:1 (n-10):(n-7) ratio (Figure 2I). Although TGF-β is the potent and widely used EMT inducer [52], it alone was insufficient to induce EMT in all cell types, such as MCF7 [53, 54] and BXPC3 cells [55, 56]. To drive EMT in all three epithelial cell lines, a robust EMT-inducing mixture was utilized [57]. Similar results were obtained for SW780 (Supplementary Figure S7D). In MCF7 and BXPC3 cells, EMT was activated, with MCF7 showing decreased SCD expression and BXPC3 showing upregulated FADS2 (Supplementary Figure S7E-F). To further validate the expression pattern of SCD-to-FADS2 reprogramming in EMT, previous datasets of induced EMT models were examined. In GSE49644 [27], EMT was induced by TGF-β in three lung cancer cell lines with various EMT scores (Supplementary Figure S7G-H). In all three cell lines, SCD was downregulated, but FADS2 expression increased only in H358 cells, which were defined as the epithelial cells with the lowest EMT scores (Supplementary Figure S7I-J). In GSE24202 [28], EMT was triggered in human mammary epithelial (HMLE) cells by genetically altering EMT regulators (Supplementary Figure S7K). Genetic manipulation of EMT regulators in HMLE cells showed increased FADS2 expression with the downregulation of CDH1 and the upregulation of transforming growth factor-β1 (TGFB1), twist family BHLH transcription factor 1 (TWIST1), and snail family transcriptional repressor 1 (SNAI1), while SCD downregulation was observed only with CDH1 repression (Supplementary Figure S7L-N). These findings confirm the SCD-to-FADS2 reprogramming during EMT, though the simultaneous upregulation of FADS2 and downregulation of SCD may not be universally observed in every EMT model. The regulation of FADS2 and SCD during EMT is highly context-dependent, influenced by basal EMT states and signaling networks.

To assess the functional consequences of FADS2 and SCD interplay, we evaluated the sensitivity profiles of epithelial and mesenchymal cancer cells to SCD and FADS2 inhibitors. Augmented FADS2-driven desaturation can bypass the SCD pathway, resulting in low sensitivity towards SCD inhibition [18]. Compared with epithelial cells, mesenchymal cells exhibited significantly lower sensitivity upon treatment with SCD inhibitors A939572 and CAY10566 (Supplementary Figure S8A-B). Similar results were observed in the GDSC dataset, where mesenchymal cancer cells had higher IC₅₀ values for CAY10566 (Supplementary Figure S8C). Importantly, TGF-β1 treatment effectively reversed the SCD-mediated inhibition of SW780 cell proliferation (Figure 2J). Due to the lower toxicity and solubility of the FADS2 inhibitor SC26196, it was challenging to achieve inhibition solely with SC26196 (Supplementary Figure S8A-B). Therefore, we synthesized SC26196-loaded nanoparticles (NP-SC26196) to improve the delivery and efficiency (Supplementary Figure S8D-F). NP-SC26196 effectively inhibited cancer cell proliferation through FADS2 inhibition. As expected, T24, EMT-induced SW780, and BXPC3 cells exhibited lower IC₅₀ values for NP-SC26196 (Figure 2K and Supplementary Figure S8G-K), demonstrating their increased dependency on FADS2. These results demonstrated the functional reprogramming from SCD to FADS2 in EMT.

Genetic and pharmacologic approaches were then used to explore the effect of FADS2 and SCD on EMT. To minimize the compensatory effect of exogenous FAs, De-lipid was applied [58]. Silencing FADS2 with shRNA effectively reduced mesenchymal markers and increased CDH1 expression (Figure 2L and Supplementary Figure S9A-B). VIM repression induced by De-lipid treatment was enhanced by SC26196 and alleviated by the addition of FA 16:1 (n-10) in T24 cells (Supplementary Figure S9C). Notably, the EMT-inducing effect of TGF-β1 was significantly mitigated by the knockdown of FADS2 expression in SW780 cells (Figure 2M-N and Supplementary Figure S9D). EMT activation was induced by FADS2 overexpression in both epithelial and mesenchymal cancer cells (Figure 2O-P and Supplementary Figure S9E-G). Moreover, FADS2 overexpression in SW780 cells robustly restructured the cell to a mesenchymal-like morphology (Figure 2Q). Both SCD inhibitors also partially altered EMT marker expression in the three epithelial cancer cell lines (Figure 2R and Supplementary Figure S9H-I), similar to observations in previous dataset GSE41485 (Supplementary Figure S9J). Furthermore, knocking down SCD in SW780 cells similarly upregulated mesenchymal marker expression (Figure 2S). Interestingly, both genetic and pharmacologic inhibition of SCD resulted in the upregulation of FADS2 expression (Figure 2T). Concurrently, under De-lipid culture conditions, FADS2 knockdown led to an increase in SCD expression (Figure 2U). This reciprocal regulation confirmed a compensatory mechanism between SCD and FADS2, highlighting the dynamic interplay between these two desaturases in cancer cells. Overall, the comprehensive analysis of EMT induction experiments and previous datasets confirmed the significance and the promotion effect of the SCD-to-FADS2 reprogramming in facilitating EMT.

3.3 SCD inhibition promoted lipid accumulation, while increased FADS2 manipulated lipid unsaturation during cancer EMT

Lipid desaturation is the primary and fundamental function of SCD and FADS2 [59]. To investigate lipid reprogramming between epithelial and mesenchymal cancer cell lines, we employed hyperspectral SRS microscopy to image the intracellular lipids in individual cells (Supplementary Figure S10A). By tuning the Raman shift frame by frame in the C-H vibration region from 2,800 to 3,050 cm⁻¹, 70 images of individual cells were recorded. Coupled with spectral phasor analysis [60], the hyperspectral SRS images were segmented into specific cellular components corresponding to whole cells and lipids based on spectral properties (Figure 3A-B). A comparison of SRS images of different cancer cells revealed that the area of visualized lipids was larger in mesenchymal cells compared to epithelial cells (Figure 3C-D), while TGF-β1 treatment further increased lipid accumulation in the mesenchymal state (Figure 3E-F), indicating that lipid accumulation is enhanced during EMT. SCD inhibitors significantly enhanced lipid accumulation and induced mesenchymal-like morphology in two epithelial cancer cell lines (Figure 3G-H). However, FADS2 overexpression and knockdown did not significantly alter the lipid accumulation (Figure 3E-F and Supplementary Figure S10B). To evaluate the composition of LDs, SRS spectra were normalized to 2,930 cm⁻¹ which contributed to the vibration of C-H bonds in lipids [61]. Lipids in mesenchymal cells showed a greater proportion of unsaturated bonds, evidenced by a stronger signal at 3,008 cm⁻¹ (Figure 3I), which was predominantly derived from = C-H bonds in unsaturated lipids [61]. The unsaturation ratio (I₃₀₀₈/I₂₉₃₀) was significantly greater in mesenchymal cells (Figure 3J). FADS2 suppression by SC26196 or shFADS2 decreased the unsaturation ratio (Figure 3K), while FADS2 overexpression and TGF-β1 treatment increased it (Figure 3L), indicating that mesenchymal cancer cells contain increased levels of FADS2-generated unsaturated lipids.

To confirm the observation outlined above, confocal Raman microspectroscopy was employed to analyze individual LDs in mesenchymal cells. All the spectra were normalized to the area of 1,400 cm⁻¹ to 1,500 cm⁻¹, with the -CH₂ bending vibration peak at 1,445 cm⁻¹. Mesenchymal cells showed characteristic peaks of unsaturated lipids (Supplementary Figure S10C), including those at 1,265 cm⁻¹ (bending = C-H), 1,654 cm⁻¹ (C = C), and 3,008 cm⁻¹ (= C-H) [61]. To quantify the degree of unsaturation in these mesenchymal cells, we established a calibration curve based on the Raman spectral measurements of FAs with varying numbers of C = C bonds. The height ratio between 1,445 cm⁻¹ and 1,654 cm⁻¹ was used to quantify unsaturation, showing a linear correlation with the number of C = C bonds (Supplementary Figure S10D-E). Lipid droplets in mesenchymal cancer cells contained high levels of unsaturated lipids (Figure 3M), including polyunsaturated lipids, confirming the role of FADS2 in mesenchymal cells. Because FA 16:1 (n-10) is the unique and direct metabolite of FADS2, we further acquired the full spectra of pure FA 16:1 (n-10) and FA 16:1 (n-9) (Supplementary Figure S10F). After normalization, the main difference between these two FAs was found in the range of 1,000 to 1,140 cm⁻¹ (Supplementary Figure S10G-H), attributed to the vibration of the C-C stretch [61]. Raman spectra of LDs from T24 cells treated with FA 16:1 (n-10) or FA 16:1 (n-7) revealed similar patterns (Supplementary Figure S10I), with a significantly higher intensity ratio of I_1060-1073 / I_1073-1086 in FA 16:1 (n-10) -treated T24 cells (Supplementary Figure S10J). This ratio was also greater in MDA-MB231 cells (Figure 3N), with increased FADS2 expression (Supplementary Figure S10K) and a higher FA 16:1 (n-10):(n-7) ratio (Figure 2E). These results confirmed that elevated FADS2 activity contributes to increased lipid unsaturation in mesenchymal cancer cells. Phospholipids are critical derivatives of FAs. To validate the distinct desaturation that occurs during EMT, we profiled major phospholipids in both epithelial and mesenchymal cells. Compared to epithelial cells, mesenchymal and TGF-β1-induced SW780 cells exhibited an increase in phospholipids with high level of unsaturation (with C = C number ≥3) and a decrease in phospholipids with a low level of unsaturation (C = C number <3) across the four species of phospholipids (Figure 3O). Notably, phospholipids carrying FA 16:1 (n-10) or FA 16:1 (n-7) exhibited significant differences between mesenchymal and epithelial cells (Supplementary Figure S11). This increase in phospholipid unsaturation in mesenchymal and TGF-β1-induced SW780 cells likely resulted from the upregulation of highly unsaturated FAs, predominantly catalyzed by FADS2.

To further elucidate the specific components of lipid accumulation following SCD inhibition, we conducted a lipidomic analysis on SW780 cells treated with A939762. The lipidomic results indicated that triglycerides (TGs) constituted the majority of the differential lipid metabolites, with a significant upregulation of many TGs (Figure 3P), confirming the formation of LDs observed in the SRS imaging. Many of the differential lipid metabolites were phospholipids, with significantly increased unsaturation levels of four types of phospholipids after SCD inhibition, likely due to the compensatory FADS2 upregulation (Figure 3Q). Interestingly, further analysis of TG unsaturation revealed no significant changes (Figure 3R), suggesting that the FAs synthesizing TGs might originate from exogenous uptake. Together, these findings indicated that SCD inhibition and FADS2 enhancement contribute to the lipid adaptation during cancer EMT.

3.4 SCD-to-FADS2 reprogramming contributed to cancer metastasis both in vitro and in vivo

The saturation of incorporated FAs contributes to membrane fluidity, and highly unsaturated FAs and their phospholipids play a role in cell property reconstruction and adaptive development [62, 63]. Both T24 and TGF-β1-induced S780 cells exhibited increased membrane fluidity (Figure 4A-B). FADS2, which contains a C-terminal membrane-spanning desaturase portion [64], was found primarily in the cytoplasm, with partial localization in the membrane (Figure 4C). Notably, membrane-bound FADS2 was significantly upregulated in mesenchymal cells (Figure 4D), potentially contributing to increased membrane desaturation and fluidity. Our findings also showed that FADS2 downregulation led to impaired membrane fluidity, particularly following De-lipid treatment (Figure 4E). Increased membrane fluidity serves as a typical cellular signature of cancer EMT and contributes to high metastatic potential [20]. Additionally, FA reprogramming has been implicated in cancer metastasis development [1]. Subsequently, we validated the correlation between SCD-to-FADS2 reprogramming and metastasis.

First, we investigated the migration ability of cancer cells in response to FADS2 manipulation. The FADS2 metabolite FA 16:1 (n-10) enhanced the migration ability of mesenchymal cells (Figure 4F), while NP-SC26196, a FADS2 inhibitor, significantly suppressed cell migration (Figure 4G). Similarly, FADS2 silencing in mesenchymal cells following De-lipid media pretreatment reduced migration (Figure 4H), and FADS2 inhibition effectively reversed the TGF-β-induced migration in SW780 cells (Figure 4I). Moreover, overexpressing FADS2 in both epithelial and mesenchymal cells significantly promoted cell migration (Figure 4J). To further validate the role of FADS2 in cancer metastasis, we employed mouse models of popliteal lymph node metastasis and lung metastasis. In the popliteal lymph node metastasis model, luciferase-labeled T24-Luc cells expressing shFADS2 #2 or the control vector were inoculated into the hock of nude mice. After six weeks, the luminescence signals of the primary xenografts in the shFADS2 group were significantly decreased (Figure 4K), indicating impaired tumor growth. The harvested popliteal lymph nodes in the shFADS2 group exhibited significantly smaller volumes and fewer invading cancer cells compared to the control group (Figure 4L). In the lung metastasis model, a significant reduction in lung metastasis was observed in the FADS2-silenced group (Figure 4M). HE staining of lung sections revealed smaller sizes of the metastatic lesions in the FADS2-knockdown group (Figure 4N) with the tumor/lung area being 16.2% in the shFADS2 compared to 50.7% in the shNC, further confirming the stimulating effect of FADS2 in promoting cancer metastasis both in vitro and in vivo.

Inconsistent findings have been reported regarding the effect of SCD inhibition on cancer metastasis [65, 66]. To exclude the potential effect of SCD inhibition on primary xenografts, we generated a tail vein-lung metastatic mouse model using the murine BLCA cell line MB49, which has low metastatic potential [67]. Mice were treated with either a control vehicle or the SCD inhibitor A939572 daily for two weeks. While MB49 cells exhibited low invasive capability of lung metastasis, SCD inhibition effectively promoted an increase in lung metastases (Figure 4O-P). In line with the in vitro findings, SCD inhibition resulted in upregulated FADS2 expression in lung metastatic lesions (Figure 4Q) and increased lipid accumulation in lung metastatic lesions (Figure 4R). These results suggested that SCD inhibition enhances extravasation capability and lipid adaptation during metastasis development. Collectively, our findings provided compelling evidence supporting the role of FADS2 in promoting the metastatic potential of cancer cells, both in vitro and in vivo. Additionally, SCD inhibition appeared to exert a prometastatic effect in the context of lung metastasis, indicating a complex interplay between SCD and FADS2 in cancer metastasis.

3.5 The FADS2/β-catenin self-reinforcing feedback loop facilitated EMT

Having established that FADS2 upregulation promotes lipid desaturation during cancer EMT and metastasis, we sought to elucidate the interaction network of FADS2 in this process. In-silico transcription factor prediction analyses on the FADS2 promoter identified several EMT-inducing transcription factors, including TWIST1, zinc finger E-BOX binding homeobox 1 (ZEB1), lymphoid enhancer binding factor 1 (LEF1), and snail family transcriptional repressor 1/2 (SNAI1/2), along with other EMT-promoting pathways such as TGF-β/Smad, Wnt/β-catenin, NF-κB, and Hedgehog signaling [50] (Figure 5A, Supplementary Figure S12A and Supplementary Table S12). Notably, EMT-TFs (TWIST1, SNAI1) and TGFB1 were shown to significantly upregulate FADS2 expression based on the above analysis (Supplementary Figure S7M). Correlation analysis across different TCGA cohorts confirmed positive correlations between FADS2 and EMT-TFs in various cancer types, with the highest overall correlation observed between FADS2 and LEF1 (Figure 5B). To explore the regulation of FADS2 by LEF1, we conducted dual-luciferase assays and found that LEF1 significantly increased luciferase activity in cells with the wild-type FADS2 promoter, but not in the mutant promoter lacking LEF1-binding sites (Figure 5C-D), indicating that FADS2 expression was positively regulated by LEF1. Moreover, the inhibition of EMT-promoting pathways using specific inhibitors [68-70] respectively downregulated FADS2 expression in various cancer cells (Figure 5E-G). Collectively, these findings suggested that FADS2 is upregulated by various EMT-inducing pathways during cancer EMT.

To further elucidate the mechanism linking FADS2 to EMT, we performed RNA-seq of a stable cell line overexpressing FADS2 and identified multiple DEGs, including several associated with EMT (Figure 5H and Supplementary Table S13). Protein-protein interaction analysis of the upregulated DEGs (log₂FC > 1.2, n = 60) identified transforming growth factor-β2 (TGFB2) as a hub gene with significant relevance to EMT (Supplementary Figure S12B and Supplementary Table S14). GSEA revealed significant upregulation of EMT, TGF-β signaling, and canonical Wnt/β-catenin signaling in the FADS2 overexpressing group (Figure 5I and Supplementary Table S15), while the oxidative phosphorylation (OXPHOS) pathway, a negatively correlated metabolic signature in EMT [71], was suppressed (Supplementary Table S15). Pathway enrichment analysis of upregulated DEGs (log₂FC > 1) further confirmed significant enrichment of EMT- and cancer metastasis-associated pathways (Figure 5J and Supplementary Table S16), demonstrating that FADS2 overexpression promoted EMT in cancer. Interestingly, our transcriptomic analysis revealed that upregulating FADS2 slightly downregulated SCD expression (Supplementary Table S13), which was further validated using Western blotting (Supplementary Figure S12C). Guided by the transcriptomic data, RT-qPCR validation demonstrated that overexpressing FADS2 increased TGFB1 and Thrombospondin 1 (THBS1) expression (Supplementary Figure S12D). Given that TGF-β1 could downregulate SCD, we proposed that the mechanism by which FADS2 overexpression downregulates SCD involves activation of the TGF-β pathway.

Since FADS2 functions as a fatty acid desaturase rather than a transcription factor, we hypothesized that the promotion of EMT by FADS2 is mediated by its fatty acid metabolites and their derivatives. Oxylipins, vital signaling regulators, are functional derivatives of PUFA [72, 73]. To explore this possibility, we profiled various oxylipins in FADS2-knockdown cells and found that only a few metabolites oxygenated by cyclooxygenase significantly decreased after FADS2 inhibition, while several oxylipins catalyzed by lipoxygenase were upregulated. Among these, the levels of prostaglandin E2 (PGE2) and 13,14-dihydro-15-keto PGD2 were notably reduced (Figure 5K and Supplementary Figure S12E). Correspondingly, COX-2 (PTGS2) was the only typically increased oxygenase after FADS2 overexpression (Supplementary Figure S12F). Both PGE2 and UFAs have been reported to stabilize β-catenin [74, 75], leading to the induction of EMT directly or sequentially [50]. The role of Wnt/β-catenin signaling, highlighted in GSEA, prompted an investigation into the interaction between FADS2 and β-catenin. De-lipid culture led to a significant reduction in β-catenin levels. This effect was effectively diminished by exogenous supplementation with FA 16:1 (n-10) or PGE2, while further decreases were observed with genetic inhibition of FADS2 (Figure 5L and Supplementary Figure S12G). Although SCD-catalyzed UFAs can also stabilize β-catenin [74], we observed stronger associations between β-catenin activation signatures and FADS2 in comparison with SCD (Supplementary Figure S12H). According to the RNA-seq, FADS2 upregulation led to an increase in key regulators of the Wnt/β-catenin pathway, notably WNT5B and TCF7L1 (Figure 5M and Supplementary Figure S12B). Silencing FADS2 effectively mitigated the activation of Wnt/β-catenin induced by TGF-β (Figure 5N). Remarkably, treatment with the β-catenin inhibitor XAV-939 completely annulled the Wnt/β-catenin-inducing effect of FADS2 overexpression (Figure 5O). These findings collectively underscore the presence of a novel self-reinforcing FADS2-UFA/PGE2-β-catenin feedback loop during EMT development.

3.6 SCD inhibition enhanced EMT plasticity and FA uptake by activating AMPK

To further investigate the mechanism behind SCD downregulation during cancer EMT, we performed pathway enrichment analysis on the upregulated DEGs (log₂FC > 0.6) obtained from the comparison between FADS2^lowSCD^high patients and FADS2^highSCD^low in the TCGA-BLCA cohort (Figure 6A and Supplementary Table S17). Key pathways identified included epithelial cell differentiation and several FA synthesis pathways, such as sterol regulatory element binding transcription factor (SREBF) and mammalian target of rapamycin complex 1 (MTORC1) pathways. Since SCD plays a critical role as a desaturase in FA synthesis [1] and is a downstream target of SREBF and MTORC1 [2], the enrichment of key regulators and enzymes involved in FA synthesis, including fatty acid synthase (FASN), was observed in FADS2^lowSCD^high patients in TCGA-BLCA, ESCA, STAD, COAD cohorts (Figure 6B). In mesenchymal-featured cancer cells and EMT-induced models, a negative correlation was detected between the expression levels of genes involved in FA synthesis (e.g., SREBF1 and FASN) and the extent of EMT (Figure 6C-D and Supplementary Figure S13A-D). Notably, FASN expression was increased in the FADS2^lowSCD^high group across multiple cancer types. FASN, a key enzyme in palmitate synthesis, was notably downregulated during TGF-β1-induced EMT [76]. SCD exhibited a stronger correlation with SREBF1/FASN expression than FADS2 across different TCGA cancer cohorts (Supplementary Figure S13E). To explore FA synthesis modulation, we used an MTORC1 inhibitor (Rapamycin) and glucose-free media to attenuate FA synthesis (Supplementary Figure S13F). Strikingly, SCD expression dramatically decreased with FA synthesis suppression, while FADS2 expression remained unchanged in SW780 cells (Figure 6E). In line with previous studies [77, 78], both Rapamycin and glucose deprivation enhanced mesenchymal marker expression in SW780 cells (Figure 6E and Supplementary Figure S13G). In MCF7 cells, glucose deprivation triggered alterations in partial EMT markers, while Rapamycin's ineffectiveness was likely due to its simultaneous suppression of FADS2 (Supplementary Figure S13H). Taken together, these results indicated that FA synthesis is reprogrammed during cancer EMT and that SCD expression is positively reliant on FA synthesis.

Mammalian cells obtain FAs either through de novo synthesis or by uptake from the exogenous environment. Alongside FA synthesis reduction, increased expression of key FA transport genes was observed in FADS2^highSCD^low patients (Figure 6B), mesenchymal cancer cells (Figure 6C), and EMT-induced models (Figure 6D and Supplementary Figure S13A-D). To visualize and assess FA uptake in epithelial and mesenchymal cancer cells, we performed SRS imaging of deuterated FAs. By tuning the beating frequency to be resonant with C-H and C-D stretching vibrations at 2,899 cm⁻¹ and 2,168 cm⁻¹ respectively, the signals from C-H biomolecules and deuterated FA derivatives were acquired. As palmitate is the sole metabolite of FASN, deuterated palmitic acid (PA-D₃₁) was thus used to investigate the adaptive FA uptake. The C-D signal from PA-D₃₁ was significantly stronger in mesenchymal cancer cells (Figure 6F-G), indicating enhanced FA uptake capability in these cells. Collectively, these results demonstrated that SCD downregulation results from FA acquisition reprogramming, transitioning from FA synthesis to increased reliance on FA uptake during cancer EMT.

To further elucidate the role of SCD inhibition in promoting EMT, we performed RNA-seq on SW780 and MCF7 cells treated with the SCD inhibitor A939572 (Supplementary Tables S18-S19). Consistent with previous findings [65], enrichment analyses revealed pathways associated with NF-κB, P53, and cell proliferation, confirming the anticancer effect of the SCD inhibitor. Additionally, cancer metastasis-related pathways, such as blood vessel development, tissue morphogenesis, cell migration, and EMT, were also observed following SCD inhibition (Figure 6H, Supplementary Figure S13I, and Supplementary Tables S20-S21). Interestingly, at 1 µmol/L A939572, the number of DEGs was significantly lower compared to the 10 µmol/L A939572 treatment group (Supplementary Figure S13J and Supplementary Table S22). While the lower concentration induced apoptosis-related pathways, it did not significantly enhance EMT pathways, as seen with 10 µmol/L treatment (Supplementary Figure S13K and Supplementary Table S23). These results collectively indicated a “grow or go” phenotypic reprogramming [79] in response to 10 µmol/L A939572 treatment. Lipid metabolic pathways, including the repressed MTORC1 signaling pathway and upregulated AMPK signaling pathway, were also identified (Figure 6H, Supplementary Figure S13I, and Supplementary Tables S20-S21). The interplay between mTOR and AMPK is well known as a pivotal sensor and regulator of FA reprogramming, encompassing both FA synthesis and uptake [80]. Compared with Rapamycin and A769662 (a potent AMPK activator). SCD inhibition significantly increased AMPKα phosphorylation and decreased S6 ribosomal protein (S6) phosphorylation (Figure 6I). Silencing SCD also activated AMPKα phosphorylation (Supplementary Figure S13L). A negative correlation between SCD and AMPK phosphorylation was found among cancer patients in the TCPA cohort, while a positive correlation was found with p70 S6 kinase (P70S6K) phosphorylation (Figure 6J). AMPK activation has been shown to facilitate FA uptake [81] and TGF-β-induced EMT [82]. Consistent with this, AMPK activation via A769662 effectively induced partial EMT and upregulated FA uptake-associated genes (Figure 6K). Moreover, these findings were corroborated by SRS imaging which demonstrated enhanced PA-D₃₁ uptake, LD accumulation, and mesenchymal morphology in SCD inhibitor-treated epithelial cancer cells, particularly in SW780 (Figure 6L-M). To survive the potentially lethal effects induced by SCD inhibition, cancer cells increase their EMT capacity and uptake FA through AMPK activation.

3.7 TGF-β2 mediated the interplay between the SCD-to-FADS2 reprogramming and EMT

To gain deeper insights into the reprogramming from SCD to FADS2, we identified a total of 32 consistently upregulated DEGs (log₂FC > 1, P_adjusted < 0.05) in the SW780 sequencing data (Figure 7A). Among them, TGFB2 was identified as a hub gene, with enrichment in pathways related to EMT and metastasis (Figure 7A-B and Supplementary Tables S24-S25). Given that TGF-β2 functions as a potent cytokine that manipulates FA metabolism and promotes EMT [83, 84], we further explored its role in SCD-to-FADS2 reprogramming. FADS2 upregulation effectively increased TGF-β2 expression at both mRNA and protein levels, while FADS2 knockdown significantly reduced TGF-β2 and blocked TGF-β1-induced TGF-β2 autocrine signaling (Figure 7C-D). The role of TGF-β2 in EMT induction was also verified (Figure 7E). Treatment with the TGF-β receptor inhibitor SB431542 fully suppressed the EMT-inducing effect of FADS2 overexpression, as indicated by changes in mesenchymal-like morphology, migratory capability, and mesenchymal marker expression (Figure 7F-H). These findings underscored the role of TGF-β2 as a robust EMT inducer downstream of FADS2. Furthermore, we demonstrated that XAV-939 decreased TGF-β2 expression in both epithelial and mesenchymal cancer cells (Figure 7I) and completely suppressed FADS2's capacity to promote TGFB2 upregulation (Figure 7J), suggesting that FADS2 and its metabolites promote TGF-β2 expression through β-catenin. Importantly, TGF-β2 stimulation increased FADS2 expression (Figure 7K), while inhibition of β-catenin or TGF-β decreased FADS2 levels (Figure 5E-F). Correlation analysis across different TCGA cohorts also validated positive correlations between FADS2 and TGFB2 (Figure 7L). Collectively, these results demonstrated that the extended FADS2/β-catenin/TGF-β2 feedback loop functions in cancer EMT.

Integrating the transcriptomic results from the SCD inhibitor-treated SW780 and MCF7, TGF-β2 was identified as the only significantly upregulated TGF-β cytokine (Supplementary Tables S18-S19). Both SCD inhibitors significantly upregulated TGFB2 expression, as did the AMPK activator A769662 (Figure 7M). Similarly, SCD knockdown also increased TGFB2 in SW780 (Figure 7N). To further clarify the mechanism by which SCD inhibition upregulates FADS2 expression, we investigated the role of TGF-β. Before applying the SCD inhibitor, we knocked down both TGF-β1 and TGF-β2 in separate experiments. Consistent with the sequencing results, we found that only TGF-β2 was significantly upregulated following SCD inhibition in both cell lines, while TGF-β1 was significantly downregulated in MCF7 cells (Figure 7O). Knocking down both TGF-β1 and TGF-β2 effectively suppressed the SCD inhibition-induced upregulation of mesenchymal markers. In SW780 cells, knocking down TGF-β1 and TGF-β2 counteracted the upregulation of TGF-β2 and FADS2 induced by the SCD inhibitor. Notably, in MCF7 cells, only TGF-β2 knockdown was able to counteract the SCD inhibition-induced upregulation of TGF-β2 and FADS2 (Figure 7O). These data suggest that the upregulation of FADS2 expression by SCD inhibitors occurs through TGF-β2. Similarly, the upregulation of TGFB2, CDH2, and FADS2 expression resulting from SCD repression was also counteracted by the TGF-β inhibitor SB431542 (Figure 7P). Given the verified role of TGF-β2 in FA uptake (Figure 7Q), we found that SB431542 could effectively mitigate the increase in PA-D₃₁ uptake induced by SCD inhibition (Figure 7R). Interestingly, SCD downregulation was not observed after TGF-β2 treatment (Figure 7K). Together, these results demonstrated that SCD inhibition fosters lipid reprogramming and engages in a complex interplay with cancer EMT by activating AMPK and upregulating TGF-β2.

Finally, we verified that the levels of TGF-β2 expression were significantly greater in FADS2^highSCD^low patients than in FADS2^lowSCD^high patients across various cancer cohorts in TCGA (Figure 7S). These findings were subsequently validated at the cancer cell level in the CCLE (Figure 7T). Notably, the IC₅₀ of TGF-β receptor inhibitors was lower in FADS2^highSCD^low cancer cells (Figure 7U), suggesting that TGF-β becomes increasingly dependent during the reprogramming from SCD dependence to FADS2 dependence. The aforementioned results demonstrated a novel UFA metabolic reprogramming during cancer EMT and progression that transitions from SCD to FADS2 dependence. Such a reprogramming responsively supports the cancer EMT phenotype, lipid adaptation, and metastatic potential by augmenting TGF-β2 transcription (Figure 8).

AUTHOR CONTRIBUTIONS

Z.C., Y.G., and F.C., designed and performed experiments. Z.C. contributed to bioinformatics, cell line studies, animal experiments, and SRS imaging. Y.G. collected and analyzed clinical samples, performed the animal experiments. F.C. contributed to cell line studies and SRS imaging. H.J.L. and J.Q. conducted animal experiments. H.J.L. and Y.Z. contributed to SRS imaging. J.Z. and W.Z. analyzed fatty acid isomers. Y.L. synthesized nanoparticles. Q.X. provided the platform and technical support for nanoparticle synthesis. Y.X. offered support for fatty acid metabolite analysis. L.Z. supplied clinical samples and supported cell line and animal experiments. J.X.C. and Z.C. conceived and supervised the project. Z.C. and J.X.C. wrote the first draft. All authors edited and approved the final manuscript.