Single-cell transcriptomic profiles were sourced from the GEO repository (Accession: GSE197177), comprising three clinical cohorts: primary pancreatic carcinoma specimens (GSM5910784, GSM5910787), hepatic metastatic lesions (GSM5910785, GSM5910788), and non-malignant pancreatic tissues (GSM5910786). Spatial transcriptomic datasets (GSM7498811, GSM7498813) were concurrently extracted for tumor microenvironmental architecture analysis. Clinical annotation-matched RNA-seq data for 178 pancreatic ductal adenocarcinoma cases were procured from TCGA (National Cancer Institute Genomic Data Commons). To address technical variability across platforms, cross-dataset normalization was conducted via the “limma”(3.60.3) (17) and “sva”(3.52.0) (18) computational frameworks in R.

The GSE132257 scRNA-seq dataset was curated and interrogated through the “Seurat”(5.1.0) computational framework (19). Initial quality control retained 28,912 high-confidence cells from a raw dataset of 35,487 cells after applying thresholds: mitochondrial gene content <10% (percent.mt < 10), unique gene counts between 200 and 2,500 (nFeature_RNA > 200 & nFeature_RNA < 2500). Following multi-algorithmic dimensional compression—implementing principal component analysis (PCA) for feature extraction with subsequent visualization via t-SNE and UMAP projection (20) —we identified high-variance transcripts (top 2000 genes) for cellular clustering and phenotypic characterization (21). Five complementary scoring paradigms (AUCell, UCell, singscore, GSEA, AddModuleScore) were systematically implemented for pathway activity quantification (22). Comparative module score analyses across defined cellular clusters were conducted using nonparametric statistical approaches.

Our analytical pipeline incorporated “monocle”(2.32.0) (23–25) and “Seurat” (5.1.0) frameworks (20) for fibroblast-specific transcriptomic interrogation. After quality filtering and stromal cell population isolation, we constructed CellDataSet architectures to calculate transcriptional variability metrics (26). High-confidence genes meeting dual thresholds (expression magnitude & dispersion variance) underwent DDRTree-based manifold learning (27) for nonlinear dimensional reduction. Quasitemporal ordering algorithms reconstructed cellular ontogeny trajectories (28), with trajectory heatmaps and state projection plots delineating fibroblast phenotypic evolution within tumor niches (29). Intercellular signaling dynamics were decoded through CellCall/CellChat platforms (30, 31), enabling systematic identification of ligand-receptor interplay across microenvironmental compartments.

Spatial metabolomic analysis of pancreatic tumor architecture was performed using integrated computational workflows. After stringent quality control—including removal of mitochondrial transcripts and selection of ribosomal protein-coding genes—raw sequencing data were normalized using SCTransform. Dimensionality reduction was conducted via PCA, followed by UMAP for cluster identification (32). SpatialFeaturePlot algorithms mapped transcriptomic gradients across tissue microdomains (33), while “scMetabolism”(0.2.1) quantified pathway activation states through enzyme-centric scoring. DotPlot visualizations highlighted metabolic heterogeneity across cellular compartments (34), complemented by spatiotemporally resolved heatmaps delineating pathway-associated gene distributions within tumor topographies.

Our analytical cascade implemented a multi-stage computational pipeline for spatial transcriptomic feature extraction and pathway enrichment. Following rigorous preprocessing protocols (gene filtration, read depth normalization, logarithmic conversion, and expression standardization) (35, 36), principal component analysis (PCA) facilitated feature space compression and discriminative gene identification. Topological cell neighborhood mapping through K-nearest neighbors (KNN) algorithms preceded community detection via Louvain-Leiden clustering paradigms (37). Correlation-driven gene prioritization employing Spearman’s rank metrics identified statistically significant (p<0.05) positive/negative transcriptional regulators, subsequently subjected to pathway enrichment interrogation via Metascape’s bioinformatics suite (38, 39).

We implemented a dual-algorithm spatial omics pipeline (“RCTD”/”MISTy”) to decode multicellular interactomes in pancreatic malignancies (40, 41). Initial scRNA-seq curation involved systematic cell type classification to generate reference cellular atlases. Spatial transcriptomic inputs underwent topological parsing through RNA localization mapping and marker colocalization profiling. The “RCTD” computational framework (41) enabled probabilistic cell type deconvolution by integrating single-cell references with spatial resolution data. For microenvironmental crosstalk mapping, the “MISTy” platform (40) quantified intercellular signaling across defined spatial domains (intra-tumoral, juxta-tumoral, para-tumoral), generating multiplexed interaction matrices and spatially resolved ligand-receptor activation heatmaps.

Our analytical framework integrated scRNA-seq and spatial transcriptomic (stRNA-seq) datasets for cellular topology reconstruction and temporal dynamics inference. Fibroblast subpopulations were computationally isolated from annotated single-cell repositories and stratified into TSPAN4+/- cohorts based on transmembrane protein expression thresholds. Transcriptomic profiles underwent format standardization via “SingleCellExperiment” object conversion, followed by rigorous preprocessing (count normalization, low-abundance ribosomal/mitochondrial transcript filtration). The “SPOTlight”(1.8.0) (42) deconvolution engine enabled cellular spatial mapping through integrative analysis of single-cell and spatial resolution data, employing randomized subsampling (n=100 per cell type) to optimize computational tractability. High-confidence markers (AUC>0.7) were prioritized for spatial pattern validation using Seurat’s visualization modules (25), while “monocle” (2.32.0) (23, 24) pseudotemporal ordination algorithms reconstructed cellular transition trajectories.

Our analytical pipeline implemented the “spacexr” (v2.2.1) computational framework (41) for spatial deconvolution coupled with multi-modal prognostic modeling. Stromal fibroblasts were computationally segregated from scRNA-seq repositories and stratified into TSPAN4-expressing versus null subpopulations. Reference spatial atlases were constructed using “spacexr”‘s probabilistic modeling architecture, enabling high-resolution mapping of fibroblast spatial topographies within tumor microdomains. Complementarily, bulk transcriptomic infiltration profiling and survival prognostication were executed through TSPAN4+ fibroblast marker-derived signatures. Pathway dysregulation was quantified via “GSVA” (1.52.3) enrichment scoring (43), with parallel evaluations of immunotherapy responsiveness and survival correlations conducted through multivariate Cox regression modeling.

Human pancreatic adenocarcinoma cell lines (SW1990, PANC-1) were maintained in DMEM culture medium containing 10% heat-inactivated FBS and antibiotic-antimycotic solution (100 U/mL penicillin, 100 µg/mL streptomycin). Gene silencing experiments employed TSPAN4-specific siRNA (si-TSPAN4) or scrambled control siRNA (si-NC) delivered via GA-DNA Transfection Reagent (GeneAdv Co., Suzhou), following the manufacturer’s protocol. Post-transfection cellular models were subjected to a 48-hour incubation period prior to downstream functional assays.

Total RNA was isolated employing TRIzol reagent (Thermo Fisher Scientific), followed by reverse transcription into cDNA using PrimeScript RT Master Mix (Takara Bio Inc.). Amplification reactions were conducted using the Applied Biosystems 7500 platform with SYBR Green chemistry (Takara Bio Inc.), utilizing GAPDH as an endogenous control. TSPAN4 mRNA levels were determined through comparative threshold cycle analysis, with normalized expression levels computed via the ΔΔCt (2^−ΔΔCt) algorithm.

Proliferative dynamics were quantified using the CCK-8 assay system (Dojindo Molecular Technologies). SW1990 and PANC-1 cellular models were plated in 96-well microtiter plates (1.5×10³ cells/well) and subjected to temporal monitoring at 24-hour intervals (24-96h) post-transfection. Following each incubation epoch, CCK-8 chromogenic solution (10μL/well) was introduced, with subsequent spectrophotometric quantification (450nm wavelength) performed using a multi-mode microplate reader to determine relative proliferative indices.

The clonogenic capacity of SW1990 and PANC-1 cellular models was evaluated through 6-well plate assays (500 cells/well). Following 14-day incubation under standard culture conditions, clonogenic units were chemically immobilized with 4% paraformaldehyde and chromatically labeled with 0.1% crystal violet solution. Quantitative analysis of colony formation efficiency was performed through digital image processing using ImageJ’s automated enumeration algorithms.

Cellular migratory potential was analyzed through a standardized scratch assay. SW1990 and PANC-1 cell monolayers were established in 6-well culture plates until achieving 90% confluency. Mechanical wound induction was performed using sterile 200 μL pipette tips, followed by sequential image acquisition at baseline (0 h) and 48 h post-wounding using an Olympus IX73 inverted phase-contrast microscope system. Wound closure kinetics were computationally quantified via ImageJ’s MRI Wound Healing Tool plugin for normalized metric derivation.

Cellular invasiveness was evaluated using Matrigel-coated Transwell systems (8μm pore, Corning Inc.). Post-transfection cellular cohorts (si-TSPAN4 vs. scrambled siRNA controls) were suspended in serum-free medium (2×10⁴ cells/insert) within upper chambers, while lower compartments contained chemoattractant-enriched medium (10% FBS). Following 24-hour chemotactic induction at 37°C/5% CO₂, non-invasive cellular fractions were mechanically cleared from upper membranes. Transmigrated cell populations were chemically immobilized (4% paraformaldehyde), chromatically labeled (0.1% crystal violet), and quantified through microscopic enumeration (5 randomized high-power fields). Invasion indices were calculated by normalizing experimental group transmigration counts against control baselines.

Programmed cell death quantification was performed using the Annexin V-FITC/PI Apoptosis Detection System (BD Biosciences). Post-transfection cellular cohorts (si-TSPAN4 vs. scrambled siRNA controls) were harvested at 48h, subjected to dual PBS rinses, and dual-stained with Annexin V-FITC/PI fluorophores. Cellular fluorescence profiles were acquired via a BD FACSCanto II analytical cytometer, with apoptotic indices (encompassing both initial and terminal phases) computationally derived through FlowJo™ v10.8.1 multiparametric analysis suites.

Statistical analyses were performed using R software (version 4.4.1). Data from the GEO and TCGA databases underwent quality control and batch effect correction. Various computational approaches were employed for dimensionality reduction, data visualization, gene set scoring, pseudotime trajectory inference, intercellular communication analysis, spatial transcriptomics, and prognostic evaluation. Experimental validation was conducted to confirm the findings. Statistical significance was defined as p-values and false discovery rate (FDR) q-values < 0.05. These methodologies ensured rigorous data processing and analysis, providing insights into the initiation, progression, and metastasis of pancreatic cancer, as well as potential therapeutic targets.

Through t-SNE-optimized transcriptomic cartography, we deconstructed cellular landscapes across three clinical cohorts: non-malignant controls (CT), metastatic carcinoma (MC), and pancreatic adenocarcinoma (PC). Lineage-specific stratification partitioned cell populations via canonical biomarkers, identifying 11 functionally discrete compartments including T cell effectors, ductal epithelia, β-islet clusters, myeloid phagocytes, acinar units, macrophage subtypes, plasmacytoid secretors, stromal fibroblasts, mast cell derivatives, vascular networks, and B lymphocytes ( Figure 2A ). Quantitative intergroup comparisons unveiled profound divergence in cellular stoichiometry and spatial patterning among CT, MC, and PC specimens ( Figure 2B ).

Multi-algorithmic pathway activation profiling (AUCell/UCell/singscore/ssGSEA/AddModuleScore) revealed preferential upregulation of prometastatic gene modules within fibroblastic compartments, as evidenced by concordant multi-metric scoring matrices ( Figure 2C violin plots). Transcriptional gradient mapping confirmed fibroblast dominance in metastasis-related pathway activation across all analytical platforms. t-SNE projection analysis ( Figure 2D ) demonstrated compartment-specific spatial segregation of major lineages, with MC/PC stromal fibroblasts and vascular endothelia exhibiting heightened activation states versus CT counterparts.

Differential expression validation identified fibroblast-selective overexpression of TSPAN4/ITGA5 ( Figure 2E ). UMAP-based spatial transcriptomics ( Figure 2F ) resolved distinct molecular geographies: TSPAN4 showed niche-restricted expression in fibroblastic zones and ductal interfaces, contrasting with ITGA5’s pan-stromal/endothelial distribution. These polarized expression topographies implicate complementary roles in tumor-stromal signaling - TSPAN4 as a focal signaling node versus ITGA5 as a ubiquitous adhesion mediator.

Migrasome-enriched fibroblast subsets underwent comprehensive characterization through pseudotemporal trajectory reconstruction and intercellular signaling network resolution. Initial trajectory modeling ( Figure 3A ) mapped the temporal activation of migrasome-related transcriptional programs, pinpointing TSPAN4/ITGA5 as key early-stage mediators during malignant transformation. Phenotypic state transitions along the developmental continuum were visualized through trajectory topology mapping ( Figure 3B ), revealing progressive lineage diversification. UMAP cluster progression analysis ( Figure 3C ) documented dynamic fibroblast subset expansion from 17 to 19 distinct phenotypes during disease progression.

Transcriptome-based stratification segregated fibroblasts into migrasome-abundant (MigrasomeHigh-Fibro) and -depleted (MigrasomeLow-Fibro) subgroups. Fibroblasts were classified as MigrasomeHigh-Fibro if their expression of migrasome-associated genes (TSPAN4, ITGA5) exceeded the 75th percentile of all fibroblasts, and as MigrasomeLow-Fibro if below the 25th percentile. Ligand-receptor flux quantification via chordal network mapping ( Figure 3D ) revealed MigrasomeHigh-Fibro as principal signal transducers, exhibiting preferential communication with tumor-associated macrophages, vascular networks, and ductal interfaces ( Figure 3E ). Pathway activation stratification ( Figure 3F ) demonstrated subgroup-specific signaling polarization, with MigrasomeHigh-Fibro displaying marked hyperactivity in PERIOSTIN, non-canonical WNT, COMPLEMENT, FGF, TGFβ, and ANGPTL pathways. Mechanistic dissection ( Figure 3G ) established MigrasomeHigh-Fibro as central PERIOSTIN network hubs, coordinating pro-metastatic signaling through PI3K/Akt axis activation while concurrently regulating COMPLEMENT/FGF/TGFβ/ANGPTL cascades. CellCall-derived multilayer Sankey network visualization ( Figure 3H ) decoded the hierarchical architecture of fibroblast-dominated signaling circuits, revealing topological specialization in niche-specific regulatory networks.

Spatial omics mapping unveiled niche-restricted overexpression of migrasome regulators TSPAN4/ITGA5 within specialized tumor microdomains, prompting systematic interrogation of their metabolic network engagement. UMAP-driven manifold learning resolved 14 transcriptionally discrete cellular modules ( Figure 4A ), with spatial deconvolution algorithms mapping their histoanatomical zonation ( Figure 4B ). Transcriptional gradient quantification ( Figure 4C ) identified clusters 4/13/14 as primary reservoirs of TSPAN4/ITGA5 co-expression, demonstrating exceptional cellular penetrance (>80% detection frequency). Metabolic flux profiling ( Figure 4D ) revealed coordinated activation of bioenergetic networks - particularly oxidative phosphorylation and lipid handling machinery - within migrasome-enriched compartments.

Cross-cohort validation delineated 13 conserved cellular ecotypes ( Figure 4E ), with spatial zonation cartography ( Figure 4F ) confirming microniche conservation across specimens. Cluster 13 emerged as the predominant migrasome signaling hub ( Figure 4G ), exhibiting multi-omic metabolic reprogramming signatures ( Figure 4H ) through integration of glutaminolysis intermediates and nucleotide biosynthesis precursors. These data position migrasome-active cellular coalitions as metabolic master regulators, synchronizing tumor microenvironmental rewiring via spatiotemporal control of nutrient allocation and anabolic programming.

Temporal trajectory reconstruction of spatial transcriptomic signatures decoded molecular hierarchies governing ECM remodeling, signaling axis activation, migratory programming, and immune modulation across tumor progression. Cluster 5→0 temporal progression ( Figure 5A ) identified divergent transcriptional patterning across 30 co-directional and inverse regulatory modules ( Figure 5B ). Functional enrichment clustering of co-directional modules revealed predominant ECM catabolic pathway activation ( Figure 5C ), establishing mechanistic linkage to desmoplastic remodeling, invasive niche formation, and PD-L1-mediated immune evasion. Counter-gradient modules exhibited inverse correlation with MAPK signaling cascades ( Figure 5D )—central regulators of proliferative signaling and survival pathways.

Cluster 5→1 trajectory analysis ( Figure 5E ) uncovered integrin signalosome assemblies as dominant co-directional networks ( Figure 5G ), while inverse modules governed endosomal trafficking and receptor recycling ( Figure 5H ). Cluster 8→5 temporal mapping ( Figure 5I ) exposed interferon-responsive apoptotic regulators as synchronized modules ( Figure 5K ), opposed by counter-regulated programs controlling ECM scaffolding, oxidative stress buffering, and differentiation arrest ( Figure 5L ).

Cross-cohort validation confirmed conserved temporal logic. Cluster 6→10 progression ( Figure 5M ) highlighted co-directional activation of ECM proteolysis, FAK signaling, and mesenchymal motility ( Figure 5O ), contrasted with suppressed leukocyte recruitment and β2-integrin adhesion ( Figure 5P ). Cluster 7→10 deconvolution ( Figure 5Q ) demonstrated synchronized αvβ3 integrin mechanosensing, miR-509-3p regulatory hubs, and IL-mediated paracrine signaling ( Figure 5S ), juxtaposed against attenuated antigen presentation and complement surveillance ( Figure 5T ).

Dynamic spatial profiling of tumor specimens identified fibroblast subpopulations with migrasome accumulation displaying peritumoral localization patterns, coordinating immune-suppressive microenvironment formation through adaptive molecular remodeling. Zonal architecture analysis ( Figure 6A ) revealed concentric spatial patterning of high-migrasome fibroblasts along tumor-stroma boundaries. Regulatory network decomposition ( Figure 6B ) delineated counteractive associations between migrasome-enriched fibroblasts and cytotoxic immune cells, particularly demonstrating heightened interface propensity scores for T cell and macrophage populations ( Figure 6C ). Three-dimensional spatial quantification across distinct anatomical regions – tumor parenchyma (intra), peritumoral stroma (juxta-5μm), and invasive periphery (para-15μm) – ( Figure 6D ) validated migrasome-rich fibroblast dominance at para-15μm invasion zones.

Core tumor microenvironment evaluation ( Figure 6E ) revealed operational T cell-epithelial communication networks, implying immune escape mechanisms through direct intercellular signaling. Stromal compartments within juxta-5μm zones ( Figure 6F ) displayed low-migrasome fibroblast-endothelial assemblies functionally associated with angiogenesis promotion and premetastatic conditioning. Para-15μm invasive territories ( Figure 6G ) manifested migrasome-dense fibroblast-induced immune silencing, establishing self-contained signaling nodes with diminished leukocyte infiltration.

Multiregional verification analyses ( Figures 6H, I ) corroborated this spatial architecture, identifying migrasome-enriched fibroblast-mediated immune suppression mechanisms involving STAT3-dependent checkpoint activation and chemokine signaling attenuation. Weighted significance analysis ( Figure 6J ) coupled with cellular distribution mapping ( Figure 6K ) enhanced spatial mechanistic interpretation. Intratumoral migrasome-rich fibroblasts ( Figure 6L ) activated B regulatory pathways via dual IL-10/TGFβ signaling, whereas juxta-5μm low-migrasome counterparts ( Figure 6M ) facilitated stromal-tumor cooperativity through MMP9/VEGF-A axis activation. Para-15μm migrasome-abundant fibroblasts ( Figure 6N ) executed L1CAM-integrin mechanical signaling with epithelial cells, potentiating Wnt/β-catenin-mediated proliferative cascades.

Comprehensive multi-platform profiling identified TSPAN4-positive fibroblasts as critical network hubs within tumor microecosystems, exhibiting mutualistic functional partnerships with innate immune components (macrophages, monocytes) and stromal elements (vascular networks). Graph theory-based centrality assessments ( Figure 7A ) highlighted these fibroblasts as primary conductors of cell-cell communication, whereas boundary signaling evaluations ( Figure 7B ) characterized their dual role as stromal-immune mediators via selective interactions with endothelial and epithelial interfaces. Reciprocal regulatory axes emerged between adaptive immune clusters, where Th2-skewed T cell/plasma cell interaction metrics ( Figure 7C ) implied IgE-dependent modulation of antitumor immunity.

Network topology mapping ( Figure 7D ) delineated TSPAN4⁺ fibroblasts as core organizers, generating spoke-like connectivity frameworks bridging CD14⁺ myeloid progenitors, M2-polarized macrophages, and CD31⁺ vascular units. Spatial interdependency analyses ( Figure 7E ) uncovered microdomain colocalization patterns between TSPAN4⁺ fibroblasts and CX3CR1⁺ monocytic derivatives at tumor invasion zones. Temporal progression modeling ( Figure 7F ) traced phenotypic diversification trajectories from resting (Cluster 1) to activated fibroblast subtypes (Clusters 4-5), with stepwise TSPAN4 upregulation ( Figure 7G ) paralleling extracellular matrix remodeling dynamics and SOX9-mediated malignant reprogramming.

TSPAN4-expressing stromal cells were identified as pivotal regulators of immune suppression and desmoplastic remodeling, operating through multifaceted ligand-receptor (LR) networks involving myeloid (macrophages) and vascular (endothelial) lineages. Spatial colocalization analysis across pancreatic ductal adenocarcinoma microenvironments ( Figure 8A ) unveiled compartment-specific cellular cooperativity. Systematic LR network mapping ( Figure 8B ) prioritized high-avidity COL1A2-ITGB1 complexes as key biomechanical signaling units ( Figure 8C ), with thermodynamic spatial profiling ( Figure 8D ) highlighting their concentration at stromal-vascular junctions. Directional communication profiling ( Figure 8E ) revealed asymmetrical signaling dominance from TSPAN4⁺ fibroblasts to macrophages (via CXCL12-CXCR4 pathways) and endothelial cells (through VEGFC-VEGFR3 cascades).

Graph-based architecture modeling ( Figure 8F ) established TSPAN4⁺ fibroblasts as integrative nodes coordinating fibrogenic (TGFβ1-LTBP1) and angiogenic (ANGPT2-TIE1) signaling hierarchies. Unified LR pathway mapping ( Figure 8G ) exposed tri-modal signaling networks linking fibroblast-secreted proteases (MMP2/MMP9) to macrophage phagocytic regulation (CD47-SIRPα) and vascular barrier restructuring (JAM3-ITGAV). Hierarchical network resolution ( Figure 8H ) characterized TSPAN4⁺ fibroblasts as stromal control units, harmonizing immune inhibitory (IL10-IL10R) and matrix-condensing (LOXL2-EGFR) mechanisms via spatial orchestration of myeloid and vascular components.

Analysis of the TCGA-PAAD transcriptome identified TSPAN4-positive fibroblasts as critical regulators of immune checkpoint balance. Computational evaluation using the TIDE algorithm ( Figure 9A ) revealed an inverse relationship between TSPAN4⁺ fibroblast prevalence and therapeutic efficacy, signifying stroma-driven immune suppression. Multidimensional covariance analysis ( Figure 9B ) uncovered synchronized transcriptional patterns linking these fibroblasts to CD8⁺ T cell dysfunction (p=7.51×10⁻⁶), with lineage specification profiling ( Figure 9C ) confirming their canonical stromal origin (p=1.96×10⁻³⁷). Energy-based correlation networks ( Figure 9D ) mapped TSPAN4’s functional integration with immune escape components (CD27, CTLA4, PDCD1, TNFRSF; r=0.5–0.9), further substantiated by dual-axis interaction analysis ( Figure 9E ) demonstrating concurrent associations with both activating (CD28) and suppressive (TIM3) immune receptors.

Regulatory network topology ( Figure 9F ) characterized TSPAN4’s transcriptional interplay with matrix regulators (TYRBP9, TSPAN9) and stromal adaptability markers (VIM, VASN). Immune compartment decoding ( Figure 9G ) revealed counterintuitive relationships: robust FOXP3⁺ regulatory T cell (p=0.003) and immature B cell recruitment contrasted with diminished antigen-presenting cell interactions. Transcriptional covariance networks ( Figure 9H ) further confirmed TSPAN4’s integration with protein homeostasis (UBTQ1) and mesenchymal transition (WTIP) pathways. Notably, TSPAN4 displayed inverse correlation with genomic instability (R=-0.16, p=0.046) ( Figure 9I ), implying its involvement in shaping immunologically silent microenvironments via tumor mutational burden regulation.

Gene-specific suppression of TSPAN4 was confirmed through quantitative polymerase chain reaction (qPCR), with siRNA-mediated silencing achieving substantial mRNA downregulation in SW1990 (p < 0.001) and PANC-1 (p < 0.0001) cell lines relative to scramble controls ( Figure 10A ). Functional characterization uncovered potent growth-inhibitory effects, as evidenced by CCK-8 proliferation assays showing progressive suppression of cellular viability over a 96-hour observation period (SW1990: p < 0.01; PANC-1: p < 0.001) ( Figure 10B ). Colony formation analyses further validated these observations, revealing near-total elimination of clonogenic capacity in both models (SW1990: p < 0.0001; PANC-1: p < 0.001), consistent with disrupted tumorigenic self-renewal mechanisms ( Figure 10C ). In parallel assessments of metastatic behavior, TSPAN4-depleted cells exhibited statistically significant migratory impairment at 48 hours post-transfection, demonstrating 82% invasion reduction in SW1990 (p < 0.0001) and 63% attenuation in PANC-1 (p < 0.05), thereby establishing TSPAN4’s functional necessity for metastatic competence ( Figure 10D ).

Genetic silencing of TSPAN4 significantly impaired invasive behavior in pancreatic cancer cell models (SW1990/PANC-1). Quantitative assessment of cellular invasion demonstrated 4.7-fold (SW1990: p < 0.0001) and 3.9-fold (PANC-1: p < 0.0001) suppression of transmigration capacity, establishing TSPAN4’s mechanistic involvement in preparing metastatic microenvironments ( Figure 11A ). Cytometric analysis uncovered apoptosis-promoting reprogramming, with TSPAN4-depleted cells showing dramatic increases in programmed cell death: SW1990 apoptotic rates surged from 0.08% (control) to 11.56% (p < 0.001), while PANC-1 apoptosis rose from 2.09% to 12.87% (p < 0.01), confirming caspase-mediated death pathway activation ( Figure 11B ).