Androgen-induced gene 1 correlates with extracellular vesicle-associated cholesterol metabolism and hepatic metastasis of pancreatic cancer

Abstract

BACKGROUND

Pancreatic ductal adenocarcinoma (PDAC) liver metastasis is a major cause of cancer-related mortality and is associated with profound metabolic and immune disorders. However, the tumor-intrinsic mechanisms that drive metabolic reprogramming and shape the immunosuppressive hepatic niche remain poorly understood.

AIM

To examine the potential role of androgen-induced gene 1 (AIG1) as a candidate regulator linking cholesterol metabolism to immune remodeling during PDAC liver metastasis.

METHODS

We integrated single-cell RNA sequencing data from 199 human samples across 12 cohorts to identify metastasis-associated PDAC tumor cell-intrinsic gene, whose expression was subsequently validated by immunohistochemistry. With AIG1 pinpointed as a candidate gene, functional studies were performed using a clustered regularly interspaced short palindromic repeats/Cas9-engineered AIG1-knockdown mouse pancreatic cancer cell line in a portal vein injection liver metastasis model. Immune landscape alterations were revealed by flow cytometry and immunofluorescence staining. Tumor cell metabolic changes were evaluated by bulk RNA sequencing and cholesterol measurement. We performed isolation, characterization and cholesterol content detection of small extracellular vesicles (sEVs) derived from tumor cells. We co-cultured bone marrow-derived macrophages with tumor cell-conditioned media to assess macrophage polarization by quantitative real-time polymerase chain reaction.

RESULTS

AIG1 was enriched in basal-like tumor cells and further amplified during PDAC liver metastasis. This metastasis-associated upregulation was more pronounced in male patients, as validated by paired clinical specimens. AIG1 knockdown suppressed metastatic growth in vivo and reshaped the immune microenvironment, characterized by reduced monocyte-derived macrophage infiltration (with decreased programmed death-ligand 1 expression), decreased regulatory T cells and exhausted CD8⁺ T cells, and increased Kupffer cell proportions. Bulk RNA sequencing revealed that cholesterol metabolism was the only significantly downregulated pathway upon AIG1 loss, accompanied by elevated intracellular cholesterol and upregulation of inflammatory cytokines. Furthermore, we observed that AIG1 knockdown led to reduced sEV particle number and decreased cholesterol content within sEVs. Conditioned medium from AIG1-deficient cells polarized bone marrow-derived macrophages toward a more inflammatory phenotype.

CONCLUSION

Our findings suggest that AIG1 is related to sEV-associated cholesterol metabolic reprogramming and immunosuppressive niche formation in PDAC liver metastasis, and may represent a candidate for further therapeutic exploration.

Key Words: Androgen-induced gene 1; Pancreatic ductal adenocarcinoma; Liver metastasis; Small extracellular vesicle; Cholesterol metabolism; Immune remodeling

Core Tip: Active cholesterol metabolism is a hallmark of pancreatic cancer malignancy. Here, we show that tumor-intrinsic androgen-induced gene 1 knockdown disrupts this metabolic program, remodels the hepatic niche away from immunosuppression, and inhibits metastatic growth. Our findings position androgen-induced gene 1 at the intersection of metabolic and immune regulation, warranting further investigation.

INTRODUCTION

Globally, pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal malignant tumors, with a 5-year overall survival rate of 13% in the United States[1] and 8.5% in China[2], a statistic that has shown only marginal improvement over the past decades[3]. The dismal prognosis of PDAC is largely attributable to its strong early spread tendency, the highly fibrotic and stroma-rich tumor microenvironment, and limited responsiveness to immunotherapy[4]. The liver is the most common metastatic site of PDAC. More than half of patients have liver metastasis at the time of diagnosis, which is the leading cause of cancer-related death[5].

The process of liver metastasis is a dynamic interaction between disseminated tumor cells and the liver microenvironment. The liver is an organ with unique immune properties, rich in resident macrophages (Kupffer cells), tolerogenic antigen-presenting cells, and special sinusoidal structures. This unique architecture and cellular composition foster a predominantly tolerogenic immune environment[6]. In PDAC liver metastasis, this tolerogenic environment is further amplified to suppress anti-tumor immune surveillance by recruiting and expanding immunosuppressive cell populations, including monocyte-derived macrophages, exhausted CD8⁺ T cells (Tex), and regulatory T cells (Treg)[7].

Recent single-cell and spatial profiling studies highlighted extensive diversity of immune cell states within PDAC metastases, revealing profound functional remodeling of neutrophils, T cell compartments, myeloid-derived suppressor cells, cancer-associated fibroblasts, and lipid-associated macrophages compared with primary tumors[8-11]. However, despite a growing appreciation of the immunosuppressive ecosystem in PDAC liver metastasis, the upstream tumor cell-intrinsic regulators that initiate this immune remodeling remain poorly defined (only prognosis-associated genes were found by comparison between metastatic and primary tumors[11,12]), and how metastatic cells actively instruct the hepatic microenvironment to favor immune evasion is still unclear.

Metabolic reprogramming is a defining feature of cancer metastasis[13]. In PDAC, metabolic abnormalities not only enable tumor cells to meet their biosynthetic needs under nutrient poor and hypoxic conditions, but also interact with the tumor microenvironment and even influence the immune response[14]. Growing evidence highlights the central roles of lipid and cholesterol metabolism in PDAC progression. Surplus lipid droplets stored in organelles provided a crucial resource for the energy-demanding process of PDAC metastasis[15]. Targeting sterol regulatory element-binding protein 1 disturbed lipid metabolism, decelerated tumor metastasis, and enhanced the efficacy of immunotherapy in PDAC[16,17].

Cholesterol plays a key role in preserving membrane integrity and fluidity, as well as regulating cell signaling[18,19]. While enhanced cholesterol flux represents a hallmark of cancer, positioning cholesterol biosynthesis as a promising target for therapeutic intervention[20]. In pancreatic cancer, sterol O-acyltransferase 1-mediated conversion of free cholesterol (FC) into cholesterol esters disrupts cholesterol-dependent negative feedback, thereby sustaining mevalonate pathway activity and metabolic dependency[21]. In addition, low-density lipoprotein cholesterol has been shown to drive proliferation, invasion, and migration through activation of signal transducer and activator of transcription 3 signaling[22]. Inhibition of cholesterol esterification by sterol O-acyltransferase 1 inhibitor effectively suppresses PDAC metastasis[23] and affects the composition of CD8⁺ T cells in the tumor microenvironment[24]. Consistently, combined treatment with simvastatin and anti-programmed death-1 (PD1) therapy exhibits enhanced anti-tumor efficacy in pancreatic cancer xenograft models[25].

Collectively, these studies indicate that aggressive PDAC tumor behavior and its immunosuppressive microenvironment are closely associated with enhanced lipid metabolic reprogramming, with cholesterol metabolism emerging as a prominent component[26]. However, the tumor cell-intrinsic factors that coordinate cholesterol metabolic states with immune niche formation during PDAC liver metastasis remain poorly understood.

Androgen-induced gene 1 (AIG1) is a relatively understudied gene reported to be involved in lipid-associated metabolic processes, including fatty acid esters of hydroxy fatty acids hydrolysis[27]. A few studies have shown that AIG1 contributes to cancer-related processes[28] via its interaction with Golgi soluble N-ethylmaleimide-sensitive factor attachment protein receptor complex member 1 complex member 1 in T cell lymphoma and nuclear factor 1/B in salivary adenoid cystic carcinoma[29]. Yet its functional role in pancreatic cancer progression and metastatic immune modulation has not been explored.

In this study, we integrated public large-scale human single-cell transcriptomic analyses with in vivo functional experiments to investigate tumor cell-intrinsic regulators of immune microenvironment remodeling during pancreatic cancer liver metastasis. We focused on AIG1, a lipid metabolism-associated gene that is upregulated in metastatic tumor cells. By combining AIG1 silencing with immune profiling and lipid metabolism assays, we aimed to examine how AIG1-driven alterations in tumor cell lipid metabolism are associated with the formation of an immunosuppressive niche during liver metastasis.

MATERIALS AND METHODS

Public single-cell RNA-sequencing analysis

Single-cell RNA sequencing (scRNA-seq) FASTQ files were downloaded from twelve 10X Genomics datasets and preprocessed independently with Seurat version 5.2. Cells were excluded if they had < 200 expressed genes, > 25% mitochondrial reads, or total UMIs in the top 5%. All qualifying cells were merged into a single object, then normalization and scaling were carried out via NormalizeData and ScaleData functions. Dimensionality reduction of the dataset was achieved through principal component analysis. To correct batch effects across samples, Harmony was employed within the principal component analysis space, with each sample treated as a distinct batch. Cell-cell nearest neighbors were identified using FindNeighbors, followed by graph-based clustering using FindClusters. We performed uniform manifold approximation and projection (UMAP) using RunUMAP on the corrected principal components to visualize cell clusters. Based on the expression of well-established marker genes, we subsequently annotated cell types. For PDAC tumor cells, fine clustering was performed by independently normalizing, scaling, integrating, and re-clustering the subset to define fine-grained labels.

Differential expression analysis between experimental groups was conducted using the Wilcoxon rank-sum test, implemented via Seurat’s FindMarkers function. Significantly differentially expressed genes (DEGs) were identified using thresholds of an absolute log₂ fold change greater than 1 and a Benjamini-Hochberg adjusted P-value below 0.05. Gene expression distributions were visualized using violin plots, gene expression patterns across the UMAP space were displayed using FeaturePlot, and dot plots for cell type annotation were generated using DotPlot, all implemented in the Seurat R package. Volcano plots were generated to visualize and summarize differential expression results. Basal and Classical subtype scores were calculated at the single-cell level using gene signatures derived from the Moffitt PDAC subtype classification[30]. Using the UCell R package, signature enrichment scores were calculated for each cell based on the ranking of gene expression levels within individual cells. For comparisons involving more than two groups, non-parametric Kruskal-Wallis tests were applied. All the above analyses and graphical visualizations were performed using R (version 4.5.0).

Human sample collection

We retrospectively collected formalin-fixed, paraffin-embedded specimens of primary PDAC and matched liver metastases from patients treated at Beijing Chao-Yang Hospital (Beijing, China). Paired primary and metastatic samples were obtained from the same patient. The use of samples was approved by the Institutional Review Board of Beijing Chao-Yang Hospital (Approval No. 2026-1-22-2).

Cell culture and gene silencing

The KPC-1 cell line, which has been previously established[31], was derived from pancreatic tumors of p48^creTrp53^LSL-R172HKras^LSL-G12D mice on a C57BL/6J background and engineered to stably express firefly luciferase. Cells were maintained in Dulbecco’s Modified Eagle medium containing 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin, under standard culture conditions (37 °C, humidified 5% CO₂ atmosphere).

A single-guide RNA (sgRNA) targeting mouse AIG1 (target sequence: 5’-AGTCTTCTGACCAGAGGCAG-3’) and a non-targeting control sgRNA (sgControl) were designed and cloned into the lentiviral clustered regularly interspaced short palindromic repeats (CRISPR)/Cas9 vector pLenti-U6-sgRNA-SFFV-Cas9-2A-Puro by Applied Biological Materials Inc. (Richmond, Canada).

Lentiviral particles were produced by transfecting Lenti-X 293T cells with the sgRNA/Cas9 lentiviral plasmid and standard packaging plasmids, using polyethylenimine as the transfection reagent. The viral supernatant was collected 2 days after transfection, cleared by centrifugation, and mixed 1:1 with fresh complete medium. Polybrene was added to this mixture at a final dilution of 1:1000 to enhance transduction. KPC-1 cells were then incubated with the virus-polybrene mixture. We replaced the virus-containing medium with fresh complete medium 24 hours post-infection, and started puromycin selection (2 μg/mL, empirically determined for KPC-1 cells) 24 hours afterwards. The medium was refreshed every other day for about a week to establish stable polyclonal AIG1-knockdown (AIG1-KD) and sgControl cell lines.

Nested polymerase chain reaction

Genomic DNA was extracted from the putative AIG1-KD cells. To confirm successful CRISPR/Cas9 editing at the intended locus, nested polymerase chain reaction (PCR) was performed to amplify the genomic region surrounding the sgRNA target. The second-round PCR products were then purified and submitted for Sanger sequencing. We used SnapGene software to align the resulting chromatograms, where overlapping peaks downstream of the sgRNA target site indicated the presence of mixed alleles from CRISPR-induced insertions or deletions. Primer sequences and PCR conditions are listed in the Supplementary material.

Western blot

Equal numbers of KPC-1-wild type, sgControl, and AIG1-KD cells were lysed with radio-immunoprecipitation assay reagent containing protease inhibitors. The resulting lysates were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and immobilized onto polyvinylidene difluoride membranes via electroblotting. The membranes were horizontally cut based on molecular weights to allow parallel probing of AIG1 (approximately 30 kDa) and β-actin (approximately 42 kDa) within the same lane. After blocking, membrane strips were incubated overnight with anti-AIG1 (polyclonal, 14468-1-AP; 1:2000; Proteintech, IL, United States) and anti-β-actin (rabbit monoclonal antibody, AC026; 1:100000; ABclonal, MA, United States) at 4 °C. Horseradish peroxidase-conjugated donkey anti-rabbit IgG (H+L) was used as secondary antibody (AS038; 1:10000; ABclonal, MA, United States), 1.5 hours at room temperature. We quantified band intensities with ImageJ software and normalized AIG1 expression levels to β-actin. All experiments included three independent biological replicates.

Cell proliferation, migration and invasion

Cell proliferation was assessed using the Cell Counting Kit-8 (CCK-8) assay (TargetMol Chemicals Inc., MA, United States). Briefly, cells were seeded into 96-well plates at a density of 5 × 10³ cells per well following the manufacturer’s protocol. A total of four 96-well plates were prepared to measure cell proliferation on day 0, day 1, day 2, and day 3, respectively. Each plate containing only culture medium (without cells) was set as a blank control. After adding the CCK-8 reagent, the plates were incubated at 37 °C for 1 hour, and the absorbance at 450 nm was measured using a microplate reader.

For the migration assay, a wound-healing scratch test was performed. Cells were cultured in 6-well plates until reaching approximately 70% confluence, after which a linear scratch was created. Wound closure was photographed at 0 hour, 12 hours, 24 hours, and 36 hours, and the migration rate was calculated based on the change in scratch area.

We performed invasion and migration assays using Transwell chambers (3422; Corning, NY, United States) coated with or without Matrigel (354230; Corning, NY, United States). Cells suspended in serum-free medium were seeded into the upper chambers at 2 × 10³ cells per well, while complete medium was added to the lower chamber. After incubation at 37 °C in 5% CO₂ for 24 hours, cells that had migrated or invaded through the membrane were fixed, stained with crystal violet, washed, and airdried at room temperature. Stained cells were visualized under a microscope, and images were analyzed using ImageJ software.

Liver metastasis model

AIG1-KD and sgControl cells were cultured to 70%-80% confluence, harvested, and resuspended in sterile phosphate-buffered saline (PBS) for portal vein injection (1.5 × 10⁶ cells in 50 μL PBS per mouse). A total of 10 C57BL/6J male mice (n = 5 per group) purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd. (Beijing, China) were anesthetized with isoflurane to minimize suffering, and the portal vein was exposed via a midline abdominal incision. The cell suspension was slowly injected into the portal vein using a fine needle. Gentle pressure was applied to the injection site, and the abdominal cavity was closed in layers. Postoperatively, all mice received appropriate analgesia and were monitored daily for signs of pain or distress; no severe discomfort was observed. Two weeks after tumor cell injection, we assessed liver metastasis burden by in vivo bioluminescence imaging. Mice were subsequently euthanized, and liver tissues were collected for downstream analyses. The animal study was approved by the Institutional Animal Care and Use Committee of Beijing Chao-Yang Hospital (Approval No. 26-2005). All animal experiments were conducted in compliance with institutional requirements and local legislation.

In vivo bioluminescence imaging

Because the injected tumor cells stably expressed luciferase, tumor burden could be monitored non-invasively via bioluminescent signal. We administered D-luciferin potassium salt via mouse orbital venous plexus, enabling its systemic distribution through the circulation. The injected luciferin salt acts as a substrate and reacts with the tumor-native luciferase. Bioluminescent signals were then captured using an IVIS Spectrum system (PerkinElmer, MA, United States). We quantified bioluminescence images using LivingImage software (version 4.4). Signal intensity was reported as radiance in units of photons second/cm²/steradian.

Tissue processing and flow cytometry

We collected tumor and adjacent liver tissues, adjusting the sampling proportions according to each mouse’s tumor burden. Tissues were then chopped up and enzymatically digested in PBS containing 3% FBS, DNase Ⅰ (0.2 mg/mL; Sigma-Aldrich, MO, United States), dispase (2.4 mg/mL; Thermo Fisher Scientific, MA, United States), and collagenase D (1 mg/mL; Sigma-Aldrich, MO, United States). After that, we filtered the cell suspension through a 70 μm strainer. For enrichment of non-parenchymal immune cells, we first removed hepatocytes by centrifuging the single-cell suspension at 50 × g for 3 minutes. The supernatant was then spun at 400 × g for 5 minutes, and the resulting pellet was resuspended in fluorescence-activated cell sorting buffer following red blood cell lysis.

Two equal aliquots of the single-cell suspension were stained separately with two antibody panels: A myeloid/Tex panel and a natural killer (NK)/Treg panel. For the NK/Treg panel, cells were first incubated with surface antibodies in fluorescence-activated cell sorting buffer for 30 minutes on ice, followed by a 1-minute incubation with 4’,6-diamidino-2-phenylindole (DAPI; 1:20000) on ice. After washing, cells were fixed and permeabilized (2% paraformaldehyde for 1 minute, then 1 × permeabilization buffer for 10 minutes) for intracellular staining with appropriate antibodies for 30 minutes at room temperature. For the myeloid/Tex panel, cells underwent surface staining only. After staining, cells were incubated with DAPI (1:20000) for 4 minutes on ice to exclude dead cells.

Flow cytometry was performed on an Attune NxT instrument (Thermo Fisher Scientific, MA, United States), and the acquired data were analyzed via FlowJo software (version 10.9.0). Immune cell subsets were identified through sequential gating, beginning with single, live (DAPI^-), CD45⁺ cells, followed by branch-wise identification using lineage- and subset-specific markers: Kupffer cells, CD11b^low F4/80⁺ T cell immunoglobulin and mucin domain-containing protein 4 (TIM4)⁺; monocyte-derived macrophages, F4/80⁺ Ly6G^- CD11b⁺ TIM4^-; neutrophils, CD11b⁺ Ly6G⁺; conventional dendritic cells, CD11c⁺ MHC-Ⅱ⁺; Tex, CD8⁺ PD-1⁺ Tim³⁺; NK cells, NK1.1⁺ T cell receptor (TCR) β^-; NKT cells, NK1.1⁺ TCRβ⁺; Tregs, NK1.1^- TCRβ⁺ forkhead box protein P3⁺; γδ T cells, NK1.1^- TCRβ^- TCRγ/δ⁺; B cells, NK1.1^- TCRβ^- CD19⁺. Due to fluorophore channel limitations, CD4 was not included in the panel; thus, our gating strategy may result in a slight overestimation of Treg numbers. Gating strategies are detailed in Supplementary Figure 1, and antibody panels are listed in the Supplementary material.

Immunofluorescence staining

Tumor-bearing liver tissues, including both metastatic tumor regions and adjacent non-tumorous parenchyma, were carefully segmented and fixed overnight in phosphate-buffered periodate-lysine-paraformaldehyde fixative. After fixation, tissues were cryoprotected by immersion in 30% sucrose at 4 °C until they sank, embedded in optimal cutting temperature compound, and stored at -80 °C for long-term preservation. Cryosections were cut at a thickness of 15 μm and mounted onto glass slides for subsequent immunofluorescence staining.

Cryosections were brought to room temperature and washed with PBS containing 0.3% Tween-20 before blocking with 3% bovine serum albumin for 1 hour. We stained two panels separately: F4/80 co-stained with Tim-4, and F4/80 co-stained with programmed death-ligand 1 (PD-L1). Both panels were stained with DAPI (1:10000) for nuclei. Detailed experimental procedures, antibody usage and staining protocols are described in the Supplementary material. We mounted the slides with an anti-fade medium and imaged them using a Zeiss LSM 980 with Airyscan 2. Acquired images were quantitatively analyzed via Imaris software (Bitplane, Switzerland).

Immunohistochemistry

Human paraffin-embedded tissue blocks were obtained from archived clinical specimens. Tumor-bearing mouse liver tissues (containing both metastatic regions and adjacent parenchyma) were dissected and processed following standard histological procedures: Overnight fixation in 4% paraformaldehyde at 4 °C, dehydration through graded ethanol, and paraffin embedding. Paraffin blocks from both human and mouse samples were sectioned to 5 μm thick.

After baking and deparaffinizing, we performed heat-induced antigen retrieval using citrate-based buffer (P0081; Beyotime Biotechnology, Shanghai, China). The sections were then blocked and incubated overnight at 4 °C. For human samples, sections were stained with anti-AIG1 polyclonal antibody (14468-1-AP, 1:100; Proteintech, IL, United States). And for mouse samples, sections were stained with anti-Ki67 monoclonal antibody (A20018, 1:500; ABclonal, MA, United States). Horseradish peroxidase-conjugated donkey anti-rabbit IgG (heavy chain + light chain; AS038; 1:100; ABclonal, MA, United States) was used as secondary antibody, 1 hour, room temperature. We developed the signal with 3,3’-diaminobenzidine, revealing brown staining, counterstained the nuclei with hematoxylin, and then dehydrated, cleared, and mounted the sections. Whole-slide images were captured on an Olympus VS200 microscope and quantitatively analyzed using ImageJ software.

Two experienced pathologists, who were blinded to the clinical information, independently evaluated the immunohistochemical staining. We calculated the AIG1 immunohistochemistry (IHC) score by multiplying positive tumor area scores and staining intensity. The percentage of positive tumor area was graded as 1 (0%-1%), 2 (1%-5%), 3 (5%-10%), and 4 (> 10%). Staining intensity was scored as 0, 1, 2, and 3, indicating negative, weak, moderate, and strong staining, respectively. For each specimen, three representative fields at 200 × magnification were randomly selected, and the mean score was used for statistical analysis.

Bulk RNA sequencing analysis

We extracted total RNA from AIG1-KD and sgControl pancreatic cancer cells (n = 3 biological replicates per group) via TRIzol reagent. RNA concentration and integrity were measured to ensure sample quality. We constructed strand-specific RNA-sequencing libraries using poly(A) enrichment and ran them on an Illumina NovaSeq 6000 (Berry Genomics Co., Ltd., Beijing, China) to obtain 150-bp paired-end reads. Each sample generated roughly 35-45 million raw reads. The resulting FASTQ data were then quality controlled and filtered. Clean reads were aligned to the mouse reference genome (GRCm39/mm39), and gene-level read counts were obtained for downstream analyses.

We analyzed differential gene expression between AIG1-KD and sgControl cells with the edgeR package. DEGs were defined as those with an adjusted P-value [false discovery rate (FDR)] < 0.05 and an absolute log₂ fold change > 1. Global transcriptional alterations were visualized using a volcano plot. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was performed on the list of DEGs, and the top 10 significantly enriched pathways for up-regulated and down-regulated genes were visualized separately via the ggplot2 package in R (version 4.5.0).

Bone marrow-derived macrophage culture and Tumor cell-conditioned medium treatment of bone marrow-derived macrophages

We generated mouse bone marrow-derived macrophages (BMDMs) using the following protocol. Firstly, femurs were harvested from C57BL/6J mice, and bone marrow was flushed out using PBS. After red blood cell lysis, the remaining bone marrow cells were seeded onto 10 cm dishes and cultured for 7 days in Dulbecco’s Modified Eagle medium supplemented with 10% FBS, 20 ng/mL macrophage colony-stimulating factor (M-CSF, PeproTech, 315-02, 50 ug), and 1% penicillin-streptomycin to allow differentiation into naive macrophages. Three individual C57BL/6J mice were humanely euthanized, each serving as a biological replicate.

Tumor cell-conditioned medium (CM) was collected from AIG1-KD and sgControl pancreatic cancer cells. To generate CM, cells were incubated in medium containing 1% FBS for 16 hours. The supernatant was then harvested, cleared by centrifugation, and passed through a 0.22-μm filter. BMDMs were treated with CM from sgControl or AIG1-KD tumor cells, or cultured in medium containing 1% FBS as control, for 16 hours.

Quantitative real-time PCR

Total RNA was extracted from AIG1-KD and sgControl pancreatic cancer cells, as well as from BMDMs after co-culture with tumor cell-conditioned media (n = 3 biological replicates per group), using a phenol-chloroform-based RNA extraction kit (G3013; Wuhan Servicebio Technology Co., Ltd., Hubei Province, China) with a chloroform substitute (G3014; Wuhan Servicebio Technology Co., Ltd., Hubei Province, China). After testing the RNA for purity and concentration, we reverse transcribed the RNA into cDNA and performed real-time quantitative PCR. For detailed procedures, kits and primer sequences, please see the Supplementary material. For tumor cells, the expression levels of C-X3-C motif chemokine ligand 1 (CX3CL1), interferon beta 1 (IFNB1), C-X-C motif chemokine ligand 10 (CXCL10), interleukin (IL)-6, low density lipoprotein receptor-related protein 2 (LRP2), apolipoprotein E (APOE), 3-hydroxy-3-methylglutaryl-CoA reductase, 24-dehydrocholesterol reductase, ATP-binding cassette subfamily A member 1 and sterol regulatory element binding transcription factor 2 were analyzed. For BMDMs, interferon regulatory factor 1 (IRF1), CXCL10, tumor necrosis factor (TNF), IL-10, arginase 1, and CD274 were quantified. Glyceraldehyde-3-phosphate dehydrogenase was used as the internal control for all assays. Relative gene expression was calculated using the 2^-ΔΔCt method.

Cellular cholesterol measurement

Total cholesterol (TC) and FC levels were measured in AIG1-KD and sgControl pancreatic cancer cells (n = 3 biological replicates per group) using commercial assay kits according to the manufacturers’ instructions. TC was quantified using a TC Assay Kit (BYSH-0909W; Byabscience Biotechnology, Jiangsu Province, China), while FC was measured using a FC Assay Kit (BYSH-0911W; Byabscience Biotechnology, Jiangsu Province, China). For normalization, protein concentrations were quantified with a Bicinchoninic Acid Protein Assay Kit (BYSH-0418F; Byabscience Biotechnology, Jiangsu Province, China). Cholesterol levels were calculated relative to total protein and presented as μg/mg protein.

Cholesterol efflux assay

Cholesterol efflux was measured using a commercial cholesterol efflux assay kit (ab196985; Abcam, Cambridgeshire, United Kingdom) according to the manufacturer’s instructions. Murine pancreatic cancer cells were seeded in 96-well plates and incubated with the fluorescent labeling reagent for 1 hour in the dark, followed by overnight incubation with equilibration buffer. Then cells were treated with cholesterol acceptor (serum depleted of low-density lipoprotein/very-low-density lipoprotein) for 6 hours in cell incubator. For both cell types, an additional group without cholesterol acceptor was included as a negative control. The fluorescence intensity of the supernatant and cell lysates was measured via a fluorescent microplate reader (excitation wavelength/emission wavelength = 482/515 nm). Cholesterol efflux was calculated as the percentage of fluorescence intensity in the supernatant relative to the total fluorescence (supernatant plus cell lysate). The net cholesterol efflux was then obtained by subtracting the efflux percentage of the negative control from that of each experimental group.

Small extracellular vesicles isolation, characterization, and cholesterol measurement

We isolated small extracellular vesicles (sEVs) from cell culture supernatants using an exosome purification kit (Ome-01E; Beijing Omiget Pharmaceutical Technology Co., Ltd., Beijing, China). In detail, cells were cultured in medium supplemented with extracellular vesicles-free fetal bovine serum for 48 hours. The collected supernatant was centrifuged at 10000 × g for 10 minutes at 4 °C to remove debris, followed by filtration through a syringe filter. The filtered supernatant was then incubated with pre-washed magnetic beads along with exosome adsorption buffer and exosome wash buffer on a rotator at 4 °C for 40 minutes. After magnetic separation and washing, sEVs were eluted using exosome elution buffer, filtered through a syringe filter, and collected as a highly purified sEV solution.

sEVs were characterized using three methods: Transmission electron microscopy, nanoparticle tracking analysis (NTA), and western blotting. For NTA, cell confluence at harvest, volume of conditioned media, isolation procedures, and dilution ratios were kept consistent between groups. Cellular protein content was quantified by bicinchoninic acid assay to normalize particle counts. The ZETA VIEW (PMX 120; Particle Metrix, Germany) was used to automatically measure the average size and concentration of the sEV. Western blotting was performed as described in section 2.5. Primary antibodies against CD9 (13174, 1:1000; Cell Signaling Technology, Inc., MA, United States) and TSG101 (72312, 1:1000; Cell Signaling Technology, Inc., MA, United States) were used, followed by horseradish peroxidase-conjugated donkey anti-rabbit IgG (heavy chain + light chain; AS038, 1:10000; ABclonal, MA, United States).

Cholesterol content in isolated sEVs was measured using an Amplex Red Assay Kit (Beyotime Biotechnology, Shanghai, China), and protein concentration was determined using a Bicinchoninic Acid Protein Assay Kit (Omp-03; Beijing Omiget Pharmaceutical Technology Co., Ltd., Beijing, China). Cholesterol levels were normalized to sEV total protein and presented as μg/μg sEV protein.

Statistical analysis

Data are presented as mean ± SD. Statistical analyses for all experiments other than single-cell and bulk RNA sequencing were performed using GraphPad Prism software (version 10.1.2). The normality of data distribution was assessed using the Shapiro-Wilk test. For comparisons between two independent normally distributed groups, two-tailed unpaired t-tests were applied. For paired non-normally distributed data, the Wilcoxon signed-rank test was performed. For comparisons among more than two groups of normally distributed data, one-way ANOVA was performed, followed by Tukey’s multiple comparisons test. For non-normally distributed data, the Mann-Whitney U test or the Kruskal-Wallis test was applied, followed by Dunn’s test for pairwise comparisons with adjustment for multiple comparisons. P < 0.05 was considered statistically significant. The statistical methods of this study were reviewed by Si-Qi Cheng from Capital Medical University.

RESULTS

AIG1 is enriched in basal-like tumor cells and further amplified during PDAC liver metastasis

To characterize tumor cell-intrinsic transcriptional alterations associated with PDAC liver metastasis, we integrated scRNA-seq data from 199 human samples across 12 different studies, including tissue from 172 primary tumors, 21 liver metastatic lesions, and 6 healthy pancreatic from donors. After pre-processing and applying stringent quality control, 628212 high quality cells remained, with 543365 from primary tumor, 52292 from liver metastatic, and 32555 from healthy donors. The inclusion of a large number of samples from multiple published data sets allowed us to perform a robust, cross-cohort analysis (Supplementary Table 1). Using unsupervised cluster analysis, we identified 14 major cell populations across all samples. These clusters were then manually annotated based on classical marker genes to identify cell types including epithelial tumor cells, immune cells, and stromal compartments, which were visualized by a dot plot (Supplementary Figure 2A). Their distribution across disease states (healthy donors, primary tumor, and liver metastatic) is shown in Figure 1A. To further investigate tumor cell heterogeneity associated with liver metastasis, PDAC cells were extracted from the primary tumor and liver metastases, and further sub-cluster analysis was performed to identify 15 tumor cell subsets (Figure 1B).

Figure 1 Human single-cell RNA sequencing analysis identifies basal-associated androgen-induced gene 1 upregulation during pancreatic ductal adenocarcinoma liver metastasis. A: Global cellular landscape of human pancreatic tissues across disease states. Uniform manifold approximation and projection (UMAP) visualization of (n = 6) healthy donor pancreata, (n = 172) primary pancreatic tumors, and (n = 21) metastatic biopsies from the liver. The “pancreatic ductal adenocarcinoma (PDAC)” cluster represents PDAC tumor cells; B: Subclustering of PDAC tumor cells from primary tumor and liver metastatic. UMAP analysis resolved the cells into 15 distinct clusters; C: Single-cell scoring of classical and basal transcriptional programs in PDAC tumor cells. Violin plots (left) show the distribution of classical and basal signature scores across individual tumor cell subpopulations. FeaturePlots (right) visualize the spatial distribution of signature scores across the tumor cell UMAP embedding; D: UMAP visualization of PDAC tumor cells stratified by disease state and renamed by molecular subtypes; E: FeaturePlot showing increased AIG1 expression and broad distribution in tumor cells from liver metastases; F: Violin plot showing elevated AIG1 expression in tumor cells from liver metastases. Compared using Wilcoxon rank-sum test, with adjusted P  < 0.0001; G: Enhanced AIG1 expression in basal-like tumor cells during liver metastasis. Compared using Wilcoxon rank-sum test, with adjusted P  < 0.0001. PDAC: Pancreatic ductal adenocarcinoma; EC: Endothelial cells; TNK: T and natural killer; NK: Natural killer; AIG1: Androgen-induced gene 1.

Using Moffitt subtype gene signatures, we calculated per-cell basal and classical signature scores and examined their distribution across PDAC tumor cell subpopulations. Subpopulations 11, 2 and 5 exhibited the highest scores for the classical signature, whereas subpopulations 10, 4, and 3 showed the highest basal signature scores. Subpopulations 0, 7, 6 and 9 scored positively for both signatures and were therefore classified as ‘mixed’ cell types. Subpopulations 1, 8, 12, 13, and 14 exhibited relatively low or negative scores for both signatures and were designated as ‘others’. The distribution of classical and basal signature scores across individual tumor cell subpopulations is shown in Figure 1C. The gene lists used for classical and basal signature scoring, as well as the corresponding signature scores for each tumor cell subpopulation, are provided in the Supplementary material. Dot plots used as the reference for clustering are shown in Supplementary Figure 2B and C. UMAP visualization of PDAC tumor cells stratified by disease state (primary tumor vs liver metastatic) is shown in Figure 1D. In agreement with prior studies[30,32,33], tumor cells from liver metastases exhibited a marked reduction in classical-like transcriptional programs, consistent with the notion that the classical subtype confers lower malignant potential.

DEGs analysis between liver metastatic and primary tumor cells was visualized by a volcano plot (Supplementary Figure 2D). AIG1 was present among the DEGs and showed a highly significant adjusted P-value approaching zero. FeaturePlot visualization revealed higher AIG1 expression in tumor cells from liver metastatic compared with primary tumor (Figure 1E), consistent with violin plot analysis (P < 0.0001; Figure 1F). At the UMAP level, AIG1 expression was broadly distributed across tumor cell subpopulations without apparent restriction to a specific molecular subtype. To further clarify subtype-associated expression patterns, AIG1 expression was examined across classical, basal, mixed, and other tumor cell subpopulations in primary tumor and liver metastatic separately. Violin plot analysis showed that, in both disease contexts, AIG1 expression was relatively higher in basal-like tumor cells compared with other subpopulations (Supplementary Figure 2E and F). Importantly, when basal-like tumor cells were analyzed independently, AIG1 expression was significantly higher in liver metastatic than in primary tumor (P < 0.0001; Figure 1G), indicating that basal-associated AIG1 expression is further enhanced during liver metastasis.

Potential sex-biased upregulation of AIG1 in human PDAC liver metastases

The nomenclature of AIG1 led us to investigate whether its expression in PDAC liver metastases varies by sex. To this end, we reanalyzed the human scRNA-seq dataset containing liver metastasis samples and selected cases with clearly annotated sex information with no prior neoadjuvant treatment, resulting in nine eligible liver metastatic samples (5 males and 4 females); see details in Supplementary Table 2. The raw data were reprocessed, followed by unsupervised clustering and cell-type annotation using canonical marker genes. Tumor cells, stromal compartments, immune cells, and hepatocytes were identified and renamed accordingly. The re-annotated UMAP is shown in Supplementary Figure 3A, with representative marker expression patterns displayed as a dot plot in Supplementary Figure 3B. Stratified by sex, UMAP embeddings revealed a comparable cellular landscape in male and female samples (Figure 2A). Within tumor cells, AIG1 expression was significantly higher in male liver metastases than in females (P < 0.0001; Figure 2B). Although the magnitude of difference was modest, it suggests a potential sex-associated bias in AIG1 expression at the metastatic site.

Figure 2 Sex-associated expression pattern of androgen-induced gene 1 in human pancreatic ductal adenocarcinoma liver metastases. A: Uniform manifold approximation and projection visualization of liver metastatic samples stratified by sex (male, n = 5; female, n = 4) from the public single-cell RNA sequencing dataset; B: Violin plot showing androgen-induced gene 1 (AIG1) expression levels in tumor cells derived from male and female liver metastases. Compared using Wilcoxon rank-sum test, with adjusted P  < 0.0001; C: Representative immunohistochemical staining of AIG1 in paired primary pancreatic tumors and matched liver metastatic lesions from patients (n = 24; 13 males and 11 females). Semi-quantification of AIG1 immunohistochemical scores is shown on the right. Comparisons between primary and metastatic tumors were performed using Wilcoxon signed-rank test, and comparisons between male and female metastatic samples were performed using Mann-Whitney U test. ^aP < 0.01; ^bP < 0.0001. Scale bar = 50 μm. PDAC: Pancreatic ductal adenocarcinoma; EC: Endothelial cells; TNK: T and natural killer; NK: Natural killer; AIG1: Androgen-induced gene 1; IHC: Immunohistochemical.

We next sought to validate these findings using an independent set of paired clinical specimens from our institution. The cohort consisted of 24 patients (13 males and 11 females), and each case included primary pancreatic tumor and corresponding liver metastatic tissue. Immunohistochemical staining revealed stronger AIG1 immunoreactivity in metastatic lesions compared with their matched primary tumors (Figure 2C). Specifically, metastatic tumors showed a deeper brown chromogenic signal as well as a higher proportion of cytoplasmic positive staining. Semi-quantitative analysis confirmed that AIG1 IHC score was significantly elevated in liver metastases than in primary tumors (P < 0.0001). We further stratified metastatic lesions by sex and observed that male metastatic tumors displayed higher AIG1 IHC score compared with female counterparts, with statistical significance (P = 0.0025; Figure 2C). Due to the small sample sizes of both the public single-cell sequencing data and our own clinical cohort, we can only describe this as a potential sex-associated pattern.

AIG1 knockdown suppresses PDAC liver metastatic growth in vivo

After knowing the male bias reported above, we selected male mouse pancreatic cancer cells to construct a liver metastasis model for subsequent studies. The pancreatic tumor-derived KPC-1 cell line from male C57BL/6J mice was used for AIG1 knockdown experiments. We confirmed the presence of the Y chromosome in KPC-1 cells by PCR. The results of agarose gel electrophoresis are shown in Supplementary Figure 3C.

To suppress AIG1 expression, CRISPR/Cas9-mediated gene editing was performed using sgRNA targeting the mouse AIG1 locus. A non-targeting sgRNA construct was used to generate sgControl cells. After plasmid transfection, lentivirus packaging and puromycin selection, stable cell lines of AIG1-KD and sgControl were successfully established. Genomic DNA was extracted from AIG1-KD cells, and the target locus was amplified by nested PCR to verify the efficiency of genome editing. Sanger sequencing of the PCR-amplified region revealed overlapping chromatographic peaks starting from the predicted CRISPR/Cas9 cleavage site, indicating the presence of insertion/deletion (indel) mutations (Figure 3A). In silico off-target prediction was performed using CRISPOR. The sgRNA used in this study was classified as a high-specificity guide sequence and showed no predicted off-target sites with 0-1 mismatches. Predicted off-target sites with 2-3 mismatches were subsequently ranked according to CFD scores. None of the top 10 genes met the criteria for significant differential expression (|log₂FC| > 0.5 and adjusted P < 0.05) in bulk RNA sequencing, indicating minimal functional off-target effects. The list of the top 10 predicted off-target genes is provided in Supplementary Table 3.

Figure 3 Androgen-induced gene 1 knockdown suppresses pancreatic ductal adenocarcinoma liver metastatic growth in vivo. A: Sanger sequencing chromatograms of the androgen-induced gene 1 (AIG1) target region. Overlapping peaks downstream of the single-guide RNA target site (highlighted in yellow) indicate successful clustered regularly interspaced short palindromic repeats/Cas9-mediated indel mutations; B: Western blot analysis of AIG1 expression in KPC-1-wild type, non-targeting control single-guide RNA (sgControl) and AIG1 knockdown (AIG1-KD) cells. Quantification of AIG1 protein levels normalized to β-actin from three independent biological replicates is presented on the right. Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparisons test. ^aP < 0.01; C: In vivo bioluminescence imaging of liver metastases two weeks after portal vein injection of sgControl or AIG1-KD cells into male C57BL/6J mice (n = 5 per group). Images from all individual mice are shown. Quantification of average radiance (× 10⁷ photons/second/cm²/steradian) is shown beneath. Each dot represents an individual mouse. Comparisons between AIG1-KD group and sgControl group were performed using unpaired t-test, ^bP < 0.0001; D: Gross morphology of livers harvested two weeks after portal vein injection of sgControl or AIG1-KD cells (n = 5 per group). Metastatic lesions are outlined with red dashed lines. Images from all individual mice are shown. Scale bar = 1 cm. AIG1: Androgen-induced gene 1; AIG1-KD: Androgen-induced gene 1 knockdown; wt: Wild type; sgControl: Non-targeting control single-guide RNA.

Western blot analysis further confirmed that AIG1 protein expression was significantly reduced in AIG1-KD cells compared with sgControl cells (Figure 3B). Quantitative analysis of gray values from three independent biological replicates showed that the knockdown was statistically significant (P = 0.0091).

Next, we established mouse liver metastasis models by injecting AIG1-KD or sgControl cells into male C57BL/6J mice via the portal vein (n = 5 per group). In vivo bioluminescence imaging at 2 weeks post-injection showed that mice receiving AIG1-KD cells had a significantly lower tumor burden than the control group (P < 0.0001; Figure 3C). Corroborating this, macroscopic observation of the harvested liver at the same time points revealed smaller metastatic nodules in AIG1-KD mice compared with sgControl mice (Figure 3D). Images of all individual mice are shown to provide a comprehensive visual comparison.

To determine whether AIG1 directly affects the intrinsic malignant behavior of tumor cells, we performed in vitro proliferation, migration, and invasion assays. The CCK-8 assay showed a slight but non-significant decline in proliferation in AIG1-KD cells, and Transwell migration and invasion assays also revealed no significant changes between groups (Supplementary Figure 4A and B). The wound-healing assay showed a modest reduction in migration only at 36 hours (P = 0.0154), while no differences were observed at 12 hours or 24 hours (Supplementary Figure 4C). Overall, AIG1 depletion alone may be insufficient to substantially alter tumor cell phenotypes in vitro.

AIG1 knockdown remodels the immune microenvironment in metastatic livers

In order to analyze the immune pattern in the metastatic microenvironment, we prepared the tumor tissue and adjacent liver tissue into single cell suspension, removed the liver cells, and analyzed the composition of immune cells by flow cytometry. We first gated out debris using forward and side scatter parameters. After removing doublets and excluding dead cells (DAPI⁺), we selected CD45⁺ immune cells for downstream analysis.

Results of the myeloid compartment revealed a striking phenotypic shift. The AIG1-KD group exhibited a significant expansion of CD11^blow F4/80⁺ TIM4⁺ Kupffer cells among CD45⁺ cells, accompanied by a concurrent reduction in F4/80⁺ Ly6G^- CD11b⁺ TIM4^- monocyte-derived macrophages and CD11b⁺ Ly6G⁺ neutrophils (Figure 4A). This remodeling of the innate immune landscape extended to the lymphoid compartment. The proportion of Treg (NK1.1^- TCRβ⁺ forkhead box protein P3⁺) within total TCRβ⁺ T cells was significantly reduced in the AIG1-KD group. Exhausted T cells (Tex; CD8⁺ PD-1⁺ Tim³⁺) in total CD8⁺ T cells also decreased. Representative flow cytometry dot plots are presented in Figure 4B. Consistent with the reduction in Tex, functional assessment showed that the mean fluorescence intensity (MFI) of PD-1 on CD8⁺ T cells was markedly downregulated following AIG1 depletion (P = 0.0001; Figure 4C). However, the proportions of conventional dendritic cells, CD8⁺ T cell, NK T cell, NK cell, B cell, and γδ T cells (within CD45⁺ cluster) were unaffected by AIG1 knockdown - as was the interferon-γ MFI in NK cells (Figure 4D). Collectively, our findings suggest that AIG1 knockdown alters the myeloid and lymphoid landscape and is associated with reduced Tex and Treg proportions within the metastatic liver microenvironment.

Figure 4 Androgen-induced gene 1 knockdown remodels the immune landscape in metastatic livers. A: Flow cytometry dot plots showing Kupffer cells and monocyte-derived macrophages from non-targeting control single-guide RNA and androgen-induced gene 1 knockdown group; B: Flow cytometry dot plots showing neutrophils, regulatory T cells, and exhausted CD8⁺ T cells from on-targeting control single-guide RNA and androgen-induced gene 1 knockdown group; C: Quantification of programmed death-1 mean fluorescence intensity in CD8⁺ T cells. Representative histograms are shown below; D: Quantification of immune cell proportions, including Kupffer cells (P = 0.0002), monocyte-derived macrophages (P = 0.0012), neutrophils (P = 0.0101), regulatory T cells (percentage of T cell receptor β⁺ T cells; P = 0.0091), exhausted CD8⁺ T cells (P = 0.0005), and other indicated immune populations within the CD45⁺ compartment. Data are presented as mean ± SD. Each dot represents an individual mouse (n = 5 per group). Statistical significance was determined using unpaired t-test. ^aP < 0.05; ^bP < 0.01; ^cP < 0.001. sgControl: Non-targeting control single-guide RNA; AIG1-KD: Androgen-induced gene 1 knockdown; TIM4: T cell immunoglobulin and mucin domain-containing protein 4; FOXP3: Forkhead box P3; PD-1: Programmed death-1; Treg: Regulatory T cells; Tex: Exhausted CD8⁺ T cells; cDCs: Conventional dendritic cells; NKT: Natural killer T cells; NK: Natural killer cells; IFN: Interferon; MFI: Mean fluorescence intensity.

Spatial validation of immune remodeling and tumor proliferation in metastatic livers

Immunofluorescence staining of liver sections from tumor-bearing mice was performed to visualize macrophage distribution at the tumor-liver junction. Co-staining of F4/80 with TIM4 distinguished Kupffer cells from monocyte-derived macrophages (Figure 5A). Tumor areas were identified by increased nuclear density (DAPI staining). As shown in the figure, Tim4⁺ Kupffer cells were largely excluded from tumor regions and confined to the adjacent liver parenchyma, consistent with their known spatial distribution[31]. In contrast, F4/80⁺ Tim4^- cells, regarded as monocyte-derived macrophages, were predominantly localized within tumor areas. Quantitative analysis confirmed that AIG1-KD livers harbored significantly more Kupffer cells in the non-tumorous parenchyma (P = 0.0394), while intratumoral monocyte-derived macrophages were markedly reduced (P = 0.0138; Figure 5B). These spatial findings were consistent with our flow cytometry results.

Figure 5 Spatial validation of immune remodeling and tumor proliferation in metastatic livers. A: Co-staining of F4/80 and T cell immunoglobulin and mucin domain-containing protein 4 in tumor-bearing liver sections. Tumor regions were outlined with red dashed lines according to increased nuclear density (4’,6-diamidino-2-phenylindole); B: Quantification of Kupffer cell density in adjacent liver tissue and monocyte-derived macrophage density within tumor regions (cells/mm²). Data represent n = 3 mice per group; C: Co-staining of F4/80 and programmed death-ligand 1in tumor regions; D: Quantification of programmed death-ligand 1 mean fluorescence intensity in F4/80⁺ macrophages within tumor areas. Each dot represents an individual F4/80⁺ cell. The total number of analyzed cells per group is indicated on the X-axis. Data represent n = 3 mice per group; E: Ki67 immunohistochemical staining of metastatic liver sections. Data represent n = 5 mice per group. For panels B, D, and E, quantitative analyses were performed by randomly selecting three non-overlapping fields per section. For panels B and E, mean value was calculated for each mouse. Statistical significance was determined using a two-tailed unpaired t-test or Mann-Whitney U test as appropriate. ^aP < 0.05; ^bP < 0.01; ^cP < 0.0001. TIM4: T cell immunoglobulin and mucin domain-containing protein 4; DAPI: 4’,6-diamidino-2-phenylindole; AIG1-KD: Androgen-induced gene 1 knockdown; sgControl: Non-targeting control single-guide RNA; PD-L1: Programmed death-ligand 1; IHC: Immunohistochemistry.

Next, we co-stained F4/80 and PD-L1 to further assess the immunoregulatory features of monocyte-derived macrophages. Within tumor regions, PD-L1 signal intensity in F4/80⁺ cells decreased in the AIG1-KD group compared with sgControl (Figure 5C), as quantified by significantly lower PD-L1 MFI (P < 0.0001; Figure 5D). Thus, AIG1 knockdown not only reduced monocyte-derived macrophage infiltration but also decreased its PD-L1 expression.

Ki67 IHC staining also revealed a significant signal intensity reduction in tumors derived from AIG1-KD cells (P = 0.0050; Figure 5E). IHC Ki67 index (%), defined as the percentage of Ki67-positive nuclei among total tumor cell nuclei, was markedly reduced in the AIG1-KD group, consistent with the lower tumor burden observed in vivo. Together, these spatial analyses corroborate our flow cytometry findings, indicating that AIG1-KD reshapes the immune landscape and is associated with reduced intratumoral proliferation.

AIG1 knockdown rewires cholesterol metabolism in tumor cells

To explore how AIG1 knockdown reprograms the immune composition, we performed bulk RNA-sequencing on AIG1-KD and sgControl cells. Transcriptomic profiling revealed a distinct signature associated with AIG1 loss (Figure 6A). Genes involved in inflammatory and chemotactic responses - including IRF1, CXCL10, IFNB1, IL-6, and CX3CL1 - were markedly upregulated in AIG1-KD cells, whereas lipid handling and cholesterol transport genes such as APOE, apolipoprotein A1, apolipoprotein B, LRP2, and ATP-binding cassette subfamily G member 8 were significantly downregulated. Notably, several genes with established tumor-suppressive functions, including selenium binding protein 1[34,35], Ras association domain family member 10[36,37], cell wall biogenesis 43[38,39], and aldehyde dehydrogenase 1 family member A2[40,41], were upregulated in AIG1-KD cells, whereas genes related to tumorigenesis and malignant progression, such as dynein axonemal heavy chain 11[42,43], V-set and immunoglobulin domain containing 2[44], and chondrolectin[45], were downregulated.

Figure 6 Androgen-induced gene 1 knockdown rewires cholesterol metabolism in tumor cells. A: Volcano plot of differentially expressed genes identified by bulk RNA sequencing comparing androgen-induced gene 1 knockdown (AIG1-KD) and non-targeting control single-guide RNA (sgControl) cells. Genes not meeting both criteria are labeled as “none”. Selected upregulated and downregulated genes relevant to inflammatory signaling, cholesterol metabolism, tumor-suppression and tumor-promotion are highlighted; B: Kyoto Encyclopedia of Genes and Genomes enrichment analysis of downregulated genes. The Q-value shown in the plot represents the false discovery rate-adjusted P-value. Cholesterol metabolism was the only significantly enriched pathway among downregulated genes (false discovery rate-adjusted P < 0.05); C: Heatmap showing the relative expression of cholesterol metabolism-related genes between groups based on bulk RNA-sequencing analysis; D: Quantitative real-time polymerase chain reaction validation of cholesterol metabolism-related genes in sgControl and AIG1-KD cells; E: Total cholesterol, free cholesterol, and free cholesterol/total cholesterol ratio measured in sgControl and AIG1-KD cells; F: Comparison of cholesterol efflux between the two groups. Data are presented as mean ± SD of three independent experiments. Statistical analysis was performed using unpaired t-tests. ^aP < 0.05; ^bP < 0.01; ^cP < 0.001. KEGG: Kyoto Encyclopedia of Genes and Genomes; CAM: Cell adhesion molecule; sgControl: Non-targeting control single-guide RNA; AIG1-KD: Androgen-induced gene 1 knockdown; HMGCR: 3-hydroxy-3-methylglutaryl-coenzyme A reductase; DHCR24: 24-dehydrocholesterol reductase; LRP2: Low-density lipoprotein receptor-related protein 2; ABCA1: ATP-binding cassette subfamily A member 1; APOE: Apolipoprotein E; SREBF2: Sterol regulatory element binding transcription factor 2; FC: Free cholesterol; TC: Total cholesterol.

KEGG analysis of upregulated genes revealed significant enrichment in inflammatory and cytokine-related pathways, including TNF, IL-17, and phosphatidylinositol 3-kinase-protein kinase B signaling, as well as cytokine-cytokine receptor interaction (Supplementary Figure 5A). A subset of upregulated genes was validated by quantitative real-time PCR (Supplementary Figure 5B).

In contrast, analysis of downregulated genes identified cholesterol metabolism as the only significant pathway (FDR-adjusted P < 0.05), highlighting the specific interference of AIG1 deletion in cholesterol regulation (Figure 6B). Genes involved in cholesterol biosynthesis, uptake, esterification, and transcriptional regulation were downregulated in AIG1-KD cells, whereas some cholesterol efflux-related genes showed increased expression (Figure 6C). A subset of downregulated genes was validated by quantitative real-time PCR (Figure 6D).

Accordingly, biochemical assays revealed cholesterol accumulation in AIG1-KD cells, with increased levels of both TC and FC compared with sgControl cells (P = 0.0091, 0.0045, respectively; Figure 6E). The FC/TC ratio remained unchanged, indicating that the elevation in cholesterol was proportional and did not involve a shift in its esterification status. Furthermore, cholesterol efflux assays demonstrated that AIG1-KD cells had significantly lower efflux capacity compared with control cells (P = 0.0150; Figure 6F), indicating impaired efflux function. Notably, the upregulation of efflux-related genes in bulk RNA-sequencing might represent a compensatory attempt in response to this impairment.

AIG1 loss reduces sEV-mediated cholesterol release and alters macrophage polarization

We next investigated how AIG1 knockdown affects cholesterol efflux in tumor cells. Notably, genes involved in vesicle biogenesis were downregulated in AIG1-KD cells (Figure 7A), prompting us to examine whether sEVs are involved. sEVs isolated from mouse pancreatic cancer cells exhibited typical cup-shaped morphology with a diameter of approximately 50-200 nm under electron microscopy (Figure 7B) and expressed the canonical sEV markers CD9 and TSG101 by Western blotting (Figure 7C). NTA was performed using equal volumes of CM collected from AIG1-KD and sgControl cells at comparable confluence. Particle numbers were reduced in AIG1-KD-derived medium, and the difference remained statistically significant after normalization to cellular protein content (P = 0.0143; Figure 7D). Furthermore, cholesterol content within sEVs was also lower in the AIG1-KD group, with statistical significance after normalization to sEV protein levels (P = 0.0122; Figure 7E).

Figure 7 Androgen-induced gene 1 loss reduces small extracellular vesicle-mediated cholesterol release and alters macrophage polarization. A: Heatmap showing the relative expression of vesicle biogenesis genes between groups based on bulk RNA sequencing analysis; B: Representative transmission electron microscopy images of small extracellular vesicles (sEVs) isolated from mouse pancreatic cancer cells. Scale bars: 200 nm (left) and 100 nm (right). sEVs exhibit typical cup-shaped morphology; C: Western blot validation of sEV markers. Isolated vesicles expressed the characteristic sEV markers CD9 and TSG101; D: Nanoparticle tracking analysis of sEVs isolated from equal volumes of conditioned medium collected at comparable cell confluence. Particle diameter ranged from 50 nm to 200 nm. Total particle counts were lower in the androgen-induced gene 1 knockdown group than in the control group. Particle counts were normalized to cellular protein content; E: Measurement of cholesterol content in sEVs normalized to sEV protein levels; F: Quantitative real-time polymerase chain reaction analysis of bone marrow-derived macrophages cultured for 16 hours with conditioned medium from non-targeting control single-guide RNA or androgen-induced gene 1 knockdown cells. The control group refers to bone marrow-derived macrophages cultured in 1% fetal bovine serum medium without tumor-derived conditioned medium. All quantitative real-time polymerase chain reaction and cholesterol measurements were conducted using three independent biological replicates. Statistical significance was determined using a two-tailed unpaired t-test or one-way ANOVA followed by Tukey’s multiple comparisons test. ^aP < 0.05; ^bP < 0.01; ^cP < 0.001. sgControl: Non-targeting control single-guide RNA; AIG1-KD: Androgen-induced gene 1 knockdown; sEV: Small extracellular vesicle; IRF1: Interferon regulatory factor 1; CXCL10: C-X-C motif chemokine ligand 10; TNF: Tumor necrosis factor; IL: Interleukin; ARG1: Arginase 1; CM: Conditioned medium.

Given these alterations in sEV release and cholesterol content, we next asked whether the CM from AIG1-KD cells, which contains both sEVs and soluble factors, could influence the polarization of BMDMs. Macrophages exposed to AIG1-KD CM exhibited a pro-inflammatory transcriptional profile, marked by elevated IRF1, CXCL10, and TNF, and reduced immunoregulatory molecules IL-10, arginase-1, and CD274 (Figure 7F). This polarization pattern, distinct from unstimulated controls, indicates that the AIG1-KD secretome actively shapes macrophage phenotype. Thus, AIG1 loss in PDAC tumor cells triggers concurrent metabolic and inflammatory reprogramming, which in turn remodels macrophages and T cells in the metastatic microenvironment.

DISCUSSION

Through integrated analysis of human single-cell data, a murine portal vein metastasis model, immunological and transcriptomic profiling, we found that AIG1 knockdown in pancreatic tumor cells attenuates liver metastatic progression and influences the immune microenvironment, which may be associated with changes in cholesterol metabolism and inflammatory factors.

The in vitro assays showed limited effects of AIG1 knockdown on tumor cell proliferation, migration, and invasion. In contrast, the in vivo model exhibited markedly reduced liver metastatic burden accompanied by substantial immune remodeling. These findings suggest that AIG1-associated tumor progression may involve interactions between tumor cells and the hepatic immune microenvironment, rather than being fully explained by tumor cell-autonomous proliferative or invasive capacity.

Cholesterol metabolism was the sole KEGG pathway significantly downregulated upon AIG1 depletion, with reduced expression of lipoprotein handling and cholesterol transport genes. Representative examples included LRP2[46], APOE[47], apolipoprotein A1[48], and ATP-binding cassette subfamily G member 8[49] - all of which have been reported to promote tumor progression, likely through their roles in cholesterol metabolic programming. Accordingly, AIG1-KD cells showed elevated cholesterol level with an unchanged FC/TC ratio and reduced cholesterol efflux. This transcriptional suppression, together with cholesterol accumulation, supports a blunted cholesterol metabolic program following AIG1 loss.

Notably, aggressive tumors rely on active cholesterol metabolism to construct an immunosuppressive microenvironment[26,50]. Especially for tumor cells themselves, emerging evidence points to a tumor-intrinsic cholesterol signaling axis that shapes the macrophage phenotype. For example, tumor cell proprotein convertase subtilisin/kexin type 9 upregulation increases microenvironmental cholesterol, inducing macrophages toward an M2-like phenotype - a shift that can be reversed by proprotein convertase subtilisin/kexin type 9 inhibition, which lowers cholesterol levels, restores macrophages toward a pro-inflammatory state, and suppresses tumor growth[51].

In line with our findings, recent studies have reported that tumor cells produce cholesterol-containing sEVs. Moreover, they further demonstrated that these sEVs can be internalized by myeloid cells and induce immunosuppression. Yang et al[52] used mouse B16 and MC38 tumor cells to show that cancer cell-intrinsic X-box binding protein 1 drives the production of cholesterol-containing sEVs. These sEVs are taken up by myeloid-derived suppressor cells via macropinocytosis, thereby promoting myeloid-derived suppressor cell expansion and enhancing Tcell suppression. Likewise, acyl-CoA cholesterol acyltransferase 2 promotes cholesterol efflux via sEVs in Panc02 tumor cells, polarizing tumor-associated macrophages toward an immunosuppressive M2 phenotype[53]. These reports provide external support for the missing mechanistic link in our study, raising the possibility that cholesterol-containing sEVs derived from pancreatic cancer cells may affect macrophage phenotypes and sustain an immunosuppressive niche.

Our data also revealed that AIG1 depletion was accompanied by increased expression of inflammatory cytokines and chemokines in tumor cells, including IFNB1, CXCL10, IL-6, and CX3CL1. The induction of interferon-β is particularly significant, given its well-established role in enhancing antigen presentation and priming cytotoxic CD8⁺ T cell responses. This type I interferon signature was accompanied by increased CXCL10, a C-X-C motif chemokine receptor 3 ligand known to facilitate effector T cell recruitment into tumors and bolster anti-tumor immunity[54]. While IL-6 has pleiotropic functions, its upregulation here may contribute to the complex inflammatory network that shapes myeloid and T cell dynamics within the tumor microenvironment[55]. Of note, the increase in CX3CL1, which engages C-X3-C motif chemokine receptor 1⁺ monocyte/macrophage populations, could have important implications in regulating hepatic macrophage recruitment and phenotype[56], and may therefore influence the balance between resident Kupffer cells and infiltrating monocyte-derived macrophages in the metastatic liver niche.

A key question emerging from our data is whether the observed alterations in myeloid phenotypes are primarily driven by cholesterol-containing sEVs, cytokine-mediated paracrine effects, or a coordinated interplay between metabolic and inflammatory programs. Given the intimate crosstalk between cholesterol metabolism and inflammatory pathway in cancer cells, it is plausible that AIG1 regulates a broader immunomodulatory axis integrating both metabolic and cytokine outputs. Future studies should clarify the relative roles of cholesterol transfer and inflammatory mediators in shaping the myeloid landscape.

The clinical relevance of AIG1 extends beyond our murine model. In human scRNA-seq results, AIG1 is enriched in basal-like tumor cells and further amplified in liver metastases. This points to AIG1 helping drive the aggressive phenotype, rather than being an incidental consequence of the tumor’s transcriptional program. Basal-like PDAC, unlike the classical subtype with its retained epithelial features (e.g., GATA binding protein 6 expression and mucin production)[30], is consistently associated with enhanced invasiveness, therapeutic resistance, and poor prognosis[32,57]. These tumors exhibit transcriptional programs linked to epithelial-mesenchymal transition[57], inflammatory signaling[58], and a metabolic adaptability that enables dynamic switching between oxidative phosphorylation and glycolysis[59]. Given their heightened metabolic demands and adaptive stress responses, AIG1 may support this state by sustaining the metabolic programs required for metastatic success.

Beyond subtype-specific enrichment, we noticed something else: AIG1 expression in liver metastases was potentially higher in males than females - confirmed by IHC in our own cohort. That’s interesting because AIG1 was originally identified as an androgen-inducible gene, raising the possibility that hormonal signaling might modulate its expression in metastatic PDAC. Sex differences in pancreatic cancer are well documented: Males face higher incidence rates and worse outcomes[60,61]. This extends to metastatic patterns, with males showing a greater propensity for liver metastases - a phenomenon mechanistically attributed to the upregulation of tissue inhibitor of metalloproteinases 1 in primary tumors of males, which consequently elevates systemic tissue inhibitor of metalloproteinases 1 and primes a pre-metastatic hepatic niche[62]. Since sex hormones are increasingly recognized as modulators of tumor metabolism and immune responses[63,64], our finding provides a compelling context for future investigation.

There are several limitations in this study that need to be addressed. First, our in vivo experiments were performed using only male mice, as the KPC-derived tumor cells were of male origin. Whether AIG1 contributes to sex-specific alterations in metastatic PDAC requires future studies that incorporate both sexes and directly explore the role of androgen receptor signaling pathways. Second, although our CRISPR/Cas9-mediated AIG1 knockdown was supported by sequencing validation, protein-level reduction, and in silico off-target analysis, a formal rescue experiment (re-expressing sgRNA-resistant AIG1 in AIG1-KD cells) remains necessary to further confirm the specificity of the observed phenotypes. Third, while we observed changes in the abundance of several immune populations following AIG1 loss, we did not perform direct functional assays, such as phagocytosis or T cell cytotoxicity tests, to confirm their functional status. Fourth, we recognize that direct evidence linking AIG1-mediated cholesterol metabolism to immune cell interaction is still lacking. Functional intervention experiments, such as cholesterol supplementation or rescue assays, are needed to determine whether the observed immune and metastatic phenotypes can be reversed. Moreover, it remains unclear which components in the CM - sEVs, cholesterol/Lipid metabolites, or inflammatory cytokines - are responsible for BMDM polarization. In future studies, we plan to incubate fluorescently labeled sEVs with BMDMs to directly observe internalization, followed by measuring changes in BMDM cholesterol content and polarization status. In parallel, cytokine levels in the CM will be quantified by enzyme-linked immunosorbent assay to distinguish lipid-mediated effects from cytokine-mediated effects. These investigations will be pursued in our future work to establish a complete mechanistic chain.

CONCLUSION

Using human single-cell transcriptomics and functional in vivo models, we show that AIG1 is related to sEV-associated cholesterol metabolic reprogramming and immunosuppressive niche formation in PDAC liver metastasis, and may represent a candidate for further therapeutic exploration.

ACKNOWLEDGEMENTS

The authors would like to acknowledge Li Li, Kun Luo, Hui-Min Zhang and Han Xiao for skillful technical and emotional assistance.