Multi‐Omics and Computational Pharmacology Approach With Experimental Validation Reveals the Antiproliferative Activity of Sophoricoside Against Pancreatic Cancer

ABSTRACT

Pancreatic cancer has a dismal prognosis and limited therapeutic options, highlighting an urgent need for effective treatments. Sophoricoside (SOP), a natural isoflavone glycoside, has exhibited anticancer activities in multiple malignancies, including lung cancer, glioblastoma, and hepatocellular carcinoma. We combined cellular assays, network pharmacology, machine learning, and multi‐omics to investigate SOP's effects. SOP‐inhibited proliferation of MIA PaCa‐2, SW1990, and PANC‐1 cells dose‐dependently. Network pharmacology revealed 85 overlapping targets enriched in MAPK, apoptosis, and PD‐L1/PD‐1 pathways. Machine learning and differential expression identified PTPN1 as the core target. PTPN1 was markedly upregulated in pancreatic adenocarcinoma, and its high expression correlated with poor survival and immune infiltration. Functional enrichment linked PTPN1 to TGF‐β, VEGF, and metabolic reprogramming. Molecular docking suggested a possible binding mode between SOP and PTPN1, involving four predicted hydrogen bonds. SOP reduced PTPN1 mRNA, and PTPN1 knockdown phenocopied SOP's antiproliferative effect with no additivity upon combination. Collectively, this first report demonstrates that SOP restrains pancreatic cancer cell proliferation, with PTPN1 identified as a key functionally required downstream mediator based on integrative computational and functional evidence. This work offers an integrated strategy for mechanistic exploration and highlights PTPN1 as a promising therapeutic biomarker and target for pancreatic cancer.

An integrated pipeline combining network pharmacology, machine learning, and multi‑omics reveals that sophoricoside suppresses pancreatic cancer cell proliferation through PTPN1, a downstream effector. Elevated PTPN1 correlates with poor prognosis and immune infiltration, positioning it as a promising therapeutic node and prognostic biomarker for this malignancy.

Abbreviations

AUC

Area under the receiver operating characteristic curve

CCK‐8

Cell Counting Kit‐8

DEG

Differentially expressed gene

DT

Decision tree

EdU

5‐Ethynyl‐2'‐deoxyuridine

FDR

False discovery rate

GBM

Gradient boosting machine

GEO

Gene expression omnibus

GLM

Generalized linear model

GO

Gene ontology

GSEA

Gene set enrichment analysis

GSVA

Gene set variation analysis

HPA

Human Protein Atlas

IC₅₀

Half‐maximal inhibitory concentration

IF

Immunofluorescence

KEGG

Kyoto Encyclopedia of Genes and Genomes

KNN

k‐Nearest neighbors

LASSO

Least absolute shrinkage and selection operator

MD

Molecular docking

NES

Normalized enrichment score

NNET

Neural network

PPI

Protein–protein interaction

PTPN1

Protein tyrosine phosphatase non‐receptor type 1

RF

Random forest

ROC

Receiver operating characteristic

RT‐qPCR

Quantitative real‐time reverse transcription PCR

SD

Standard deviation

siRNA

Small interfering ribonucleic acid

SOP

Sophoricoside

ssGSEA

Single‐sample gene set enrichment analysis

SVM

Support vector machine

TCGA

The cancer genome atlas

TGF‐β

Transforming growth factor‐β

VEGF

Vascular endothelial growth factor

1. Introduction

Clinically, pancreatic cancer carries an exceptionally dismal prognosis. Its long‐term outcomes remain among the poorest of all malignancies [1, 2]. At the current time, the overall 5‐year survival rate of pancreatic cancer hovers near 5%. By 2030, however, this disease is expected to claim the second‐highest number of cancer‐related deaths globally [3, 4]. Despite decades of continuous efforts to refine treatment strategies, patient outcomes have shown little meaningful improvement [5, 6]. The limitations of current therapeutic modalities are becoming increasingly evident. Surgery, the only potentially curative option, is feasible in only 10%–20% of patients [7]. Radiotherapy, commonly employed as a locoregional treatment, requires further investigation to clarify its optimal application across different disease stages [8]. In the context of chemotherapy, single‐agent regimens offer only modest efficacy [7, 9]. Altogether, the above hurdles reveal a critical requirement for the identification of novel, highly active pharmacological options targeting pancreatic neoplasms.

The isoflavone glycoside known as SOP originates from the dry fruit of Sophora japonica L., a plant with a history of use in traditional healing systems [10]. Previous studies have demonstrated its protective effects in a range of pathological conditions, including autoimmune‐mediated liver injury [11], acute lung injury [12], atopic dermatitis [13], nerve damage [14], and cardiac hypertrophy [15]. In the context of oncology, SOP has shown promising activity against several cancer types, such as lung cancer [16], glioblastoma [17], and hepatocellular carcinoma [18, 19]. Nevertheless, the possible function of SOP in pancreatic cancer has not yet been investigated, and its molecular targets and mechanistic pathways remain to be elucidated.

To our knowledge, this study establishes a comprehensive and multidimensional framework to explore the potential function of SOP in pancreatic cancer. Through a combination of biological experiments, integrative computational pharmacology, and functional enrichment analyses, we preliminarily characterized the antiproliferative effects of SOP and the associated mechanisms. Further mechanistic insights were gained by leveraging transcriptomic data, differential expression analysis, machine learning algorithms, prognostic evaluation, immune infiltration assessment, immunohistochemical profiling, and enrichment analyses including GSEA, GSVA, and ssGSEA, which collectively facilitated the identification of key targets underlying the antiproliferative activity of SOP. Molecular docking was subsequently employed to predict the binding affinity between SOP and its identified key target, with preliminary biological assays suggesting that SOP inhibits pancreatic cancer proliferation through this target. Collectively, these findings provide fresh perspectives concerning the growth‑suppressing actions of SOP within pancreatic tumors and the underlying mechanisms, thereby establishing a foundation for further development of targeted therapeutic strategies.

2. Materials and Methods

2.1. Cell Culture and Reagents

Human‐derived pancreatic cancer cell lines—specifically MIA PaCa‑2 (RRID: CVCL_0428), SW1990 (RRID: CVCL_1723), and PANC‑1 (RRID: CVCL_0480)—were acquired from the Shanghai Cell Bank (Chinese Academy of Sciences). These cells were maintained in DMEM that had been fortified with 10% FBS and 1% PS (penicillin–streptomycin) (Procell, PM150210B), and incubated at 37°C under a humidified atmosphere with 5% CO₂. Sophoricoside (SOP) (GLPBIO, GC31781) was dissolved in dimethyl sulfoxide (DMSO; Solarbio, D8371) for cell treatment.

2.2. IC₅₀ Determination

Seeding was conducted at 8000 cells per well, with 200 µL culture media added to each well of the 96‑well microplates. Following overnight incubation, the cells were treated with fresh medium containing serially diluted sophoricoside (0, 5, 10, 20, 40, 80, and 200 µM). A solvent control group with 0.1% DMSO was included. Cells were incubated with serial concentrations of sophoricoside for 48 h. After treatment, each well received CCK‑8 reagent (Abmole, M4839), after which the microplates were kept at 37°C for a 1‑h incubation period [20, 21]. Absorbance was recorded at 450 nm via a microplate spectrophotometer. Each condition was tested in triplicate across three independent biological replicates (n = 3, where n represents independent biological replicates, each with three technical replicate wells). Determination of the half‑maximal inhibitory concentration (IC₅₀) was carried out with GraphPad Prism 10.0 (GraphPad Software, La Jolla, CA, USA).

2.3. Morphological Observation and Quantitative Cell Counting

At a density of 1000 cells per well, MIA PaCa‑2 along with SW1990 were plated into 24‑well plates and incubated overnight to allow attachment. Cells were treated with 30 µM sophoricoside or an equal volume of 0.1% DMSO for 48 h. Cell morphological changes were observed and photographed under an inverted microscope (Nikon, Ti2). The total number of viable cells in each field was counted quantitatively for statistical comparison using ImageJ software.

2.4. Colony Formation Assay

MIA PaCa‑2 and SW1990 cells were plated into six‑well dishes at a seeding of 100 cells in each well. Before plating, cells were thoroughly resuspended by repeated pipetting and serially diluted; random microscopic inspections confirmed that the cells were predominantly in single‐cell suspension without significant clumping. After seeding, plates were allowed to stand at room temperature for 30 min before transfer to the incubator to minimize cell drifting and aggregation. Overnight incubation facilitated attachment. Once adhered, the cultures received SOP or DMSO for 2 weeks of incubation. During this period, the medium was replaced every 2 days by gentle aspiration of the upper old medium (half‐volume exchange) to avoid disturbing early adherent micro‐colonies; fresh SOP was replenished to maintain the final drug concentration in the treatment groups, while the control groups received an equal volume of medium containing the corresponding concentration of DMSO (0.1%). All procedures were performed under sterile and light‐protected conditions. After fixation using 4% paraformaldehyde (Sigma‐Aldrich, F8775) and staining with 0.1% crystal violet (Beyotime, C0121) for 10 min under ambient conditions, the colonies were enumerated via ImageJ software [22]. A colony was defined as a cluster containing ≥50 cells; smaller aggregates and scattered single cells were excluded from the count. ImageJ was used only for outlining colony contours and assisting area verification; the final counts were manually validated. Plating efficiency was verified by the concurrent control group and calculated as: (number of colonies in control group/number of cells seeded) × 100%.

2.5. EdU Proliferation Assay

EdU staining [23] was conducted to assess cell proliferation activity of MIA PaCa‐2 and SW1990 cell lines. Briefly, cells were seeded in 24‐well plates and treated with 30 µM sophoricoside or 0.1% DMSO for 48 h. EdU reagent (GLPBIO, GK10046) was added to the culture medium 2 h before the end of treatment according to the manufacturer's instructions. After fixation and permeabilization, cell nuclei were counterstained with DAPI. Fluorescence images were captured under a laser scanning confocal microscope, and the ratio of EdU‐positive proliferative cells was analyzed. At least three randomly selected fields per well were captured, and the experiment was performed in three independent biological replicates (n = 3).

2.6. Target Identification of Sophoricoside

The structure of sophoricoside (PubChem CID: 5321398) was retrieved from the PubChem (https://pubchem.ncbi.nlm.nih.gov/) and subsequently submitted to multiple target prediction platforms, including STITCH [24], TTD [25], SuperPred [26], TargetNet [27], SwissTargetPrediction [28], and SEA, to predict its potential targets. All candidate targets obtained from different databases were integrated and deduplicated with Microsoft Excel to eliminate redundant entries. We next analyzed the candidate targets through PPI networks using the STRING database [29]. Construction and rendering of the resulting network were performed with Cytoscape (version 3.9.1) [30]. A network diagram illustrating the interactions between sophoricoside and its putative targets was constructed accordingly.

2.7. Screening of Pancreatic Cancer‐Associated Targets

Pancreatic cancer‐related targets were retrieved from the OMIM (https://www.omim.org/), GeneCards (https://www.genecards.org/), and Comparative Toxicogenomics Database (CTD) [31] via the keyword “pancreatic cancer.” Using the Venn diagram tool available at https://www.xiantaozi.com/, we determined the intersection between SOP's predicted targets and genes linked to pancreatic cancer.

2.8. GO and KEGG Functional Enrichment Analysis

Using R software (version 4.3.2) and the clusterProfiler package, we performed functional enrichment analysis on the genes common to SOP targets and pancreatic cancer‑associated genes, examining both Gene Ontology (GO) terms and KEGG pathways. The Benjamini–Hochberg procedure was used to adjust for multiple comparisons, and a q‑value < 0.05 was set as the significance threshold.

2.9. GEO Data Download and Differential Expression Analysis

From the GEO repository, we obtained the transcriptomic dataset identified as GSE183795. Samples were grouped according to the experimental design into tumor and normal groups. Group comparisons were carried out via the limma package, which implements an empirical Bayes moderation strategy, and genes were considered DEGs when their |log₂FC| surpassed 1 and their adjusted p‑value (using the BH correction) fell under 0.05.

2.10. Machine Learning

For this analysis, we utilized R software together with the caret, randomForest, and gbm packages to run eight distinct classifiers: random forest (RF), support vector machine (SVM), least absolute shrinkage and selection operator (LASSO), gradient boosting machine (GBM), neural network (NNET), k‑nearest neighbors (KNN), decision tree (DT), and generalized linear model (GLM). We randomly partitioned the GSE183795 dataset into training and test cohorts at a 7:3 allocation. Model hyperparameters were optimized using 10‐fold cross‐validation combined with grid search. Predictive performance and error distribution characteristics of the models were evaluated using residual cumulative distribution plots and residual histograms.

2.11. Multi‐Omics Analysis of Key SOP Targets

Multi‐omics analyses of the key targets of SOP were performed using the Xiantao Academic platform (https://www.xiantaozi.com/) based on TCGA data. The analyses encompassed chromosomal localization, differential expression profiling, prognostic value assessment, and immune infiltration characterization of the target genes.

TCGA transcriptomic data were derived from the XENA database (https://xenabrowser.net/datapages/) as standardized datasets uniformly processed by the Toil pipeline [32], representing the official archived stable release of TCGA‐PAAD (Pancreatic Adenocarcinoma). The sample composition comprised 179 TCGA tumor samples, 4 TCGA adjacent normal tissue samples, and 167 GTEx normal pancreatic tissue samples. Data were provided in TPM format and output after log₂ (value + 1) transformation, with no additional filtering applied to the raw data. Differential expression analysis between groups was performed using the Wilcoxon rank‐sum test. Immune infiltration was evaluated using two independent algorithms, namely CIBERSORT and ssGSEA (implemented via GSVA). The association between PTPN1 expression and immune cell infiltration was assessed by Spearman's correlation analysis, with adjusted p‐values calculated using the Benjamini–Hochberg method.

Immunohistochemical images illustrating the expression of the key targets in pancreatic cancer and normal tissues were obtained from the Human Protein Atlas (HPA) database (https://www.proteinatlas.org/).

Proteomic analyses were performed using the publicly available pancreatic cancer cohort data from the Clinical Proteomic Tumor Analysis Consortium (CPTAC), accessed and visualized via the CPTAC analysis module of the UALCAN database (https://ualcan.path.uab.edu/cgi‐bin/CPTAC‐Result.pl?genenam = PTPN1&ctype = PAAD). These analyses included comparisons of PTPN1 protein expression between 74 normal pancreatic tissues and 137 primary tumor tissues, as well as stratified comparisons by pathological grade (Grades 1–4) and TNM stage (Stages 1–4), based on the publicly released CPTAC pancreatic adenocarcinoma proteomic dataset. The platform default outputs log₂ protein abundance quantitative values (Z‐values) after median‐normalized intensity processing. All analyses were conducted using the pre‐standardized data modules embedded in the respective platforms [33].

2.12. Gene Set Enrichment Analysis (GSEA) of Key Targets

GSEA was performed by R packages including limma, clusterProfiler, and org.Hs.eg.db. Specimens derived from the GSE183795 cohort were assigned to high‑ or low‑expression classes, with the split determined by each key target's median abundance. All genes were ranked in descending order by log₂ fold change (log₂FC). Enrichment analyses of KEGG pathways and GO gene sets were conducted using the GSEA function in the clusterProfiler package. For each gene set, normalized enrichment scores (NES) were extracted; pathways with NES > 0 were defined as up‐regulated, and those with NES < 0 as down‐regulated. The top three most significantly enriched pathways in each category were selected for visualization.

2.13. GSVA Analysis

Gene set variation analysis, abbreviated as GSVA, was performed using the GSVA package, with parameters set as kcdf = “Gaussian,” maxDiff = TRUE, and absRanking = FALSE. Pathway scores for KEGG and GO gene sets were normalized to a range of 0–1. Samples were classified into high and low expression groups by the respective gene‑specific median expression levels. Group differences in normalized GSVA scores were assessed by two‑sided Student's t‐tests, and the Benjamini–Hochberg method was applied for FDR adjustment, with FDR < 0.05 deemed significant.

2.14. ssGSEA Analysis

Single‑sample gene set enrichment analysis (ssGSEA) was performed with the R package GSVA, and enrichment scores for KEGG and GO gene sets were subsequently calculated for each sample from the expression matrix. Samples were grouped into control and disease cohorts by the respective suffixes in their sample identifiers. Between‑group differences in enrichment estimates were contrasted using the Wilcoxon rank‑sum procedure, with significance thresholds designated as *p < 0.05, **p < 0.01, and ***p < 0.001.

2.15. Molecular Docking

The 3D crystal structures of target proteins were retrieved from the UniProt database (https://www.uniprot.org/). Prior to docking, all water molecules and native ligands were removed to prevent interference with the initial docking conformation and to focus on the predicted interaction with SOP. Polar hydrogen atoms were added, and Kollman charges were assigned using AutoDock Tools (version 1.5.6). The three‐dimensional structure of SOP was obtained from the PubChem database. Gasteiger charges were calculated, non‐polar hydrogen atoms were merged, and rotatable bonds were defined using AutoDock Tools. A global blind docking strategy was employed without prespecifying a binding site. The docking grid box was configured to encompass the entire protein structure to allow comprehensive exploration of all possible binding modes.

2.16. RT‐qPCR Analysis

Total cellular RNA was obtained with TRIzol (Invitrogen, USA). For reverse transcription, the PrimeScript RT Reagent Kit (TaKaRa, Japan) was employed, and subsequent qPCR was performed using the SYBR Premix Ex Taq Kit (TaKaRa, Japan) following the manufacturer's recommendations. The oligonucleotide primers for PTPN1 were forward 5′‐CCCTTAAATGCCGCACCCTA‐3′ and reverse 5′‐CCCACGACCCGACTTCTAAC‐3′. Relative transcript levels were computed by the 2⁻ΔΔCt method [34, 35, 36].

2.17. SiRNA Transfection

GenePharma (Shanghai, China) synthesized the siRNAs targeting PTPN1 for this study. After seeding into six‑well plates, the cells received transfection mixtures containing 50 nM siRNA along with Lipofectamine 3000 (Thermo Fisher Scientific, USA), following the manufacturer's instructions; a nontargeting scrambled duplex served as the negative control. The sequences of the PTPN1 siRNAs are as follows: siRNA*1 (sense, 5'–UCGGAUUAAACUACAUCAAGATT–3'; antisense, 3'–TTAGCCUAAUUUGAUGUAGUUCU–5), siRNA*2 (Sense, 5'–CAGGGAGUUCUUCCCAAAUCATT–3′; antisense, 3'–TTGUCCCUCAAGAAGGGUUUAGU–5′), and siRNA*3 (Sense, 5'–AAGCAUGAGUCAAGACACUGATT–3'; antisense, 3'–TTUUCGUACUCAGUUCUGUGACU–5') [37]. Cells were gathered at three time intervals—12, 24, and 48 h—after the transfection procedure, and knockdown efficiency was evaluated by RT‐qPCR, with a silencing efficiency greater than 70% considered effective for further experiments.

2.18. Immunofluorescence Staining for PTPN1 Knockdown Efficiency Verification

To verify the transfection efficiency of PTPN1‐targeted siRNA, immunofluorescence staining was performed [38]. MIA PaCa‐2 cells were first transfected with negative control siRNA or PTPN1 siRNA. After 48 h of transfection, the transfected cells were seeded onto cell slides, followed by cell fixation, permeabilization, and blocking. The slides were incubated with primary anti‐PTPN1 antibody (Catalog# M1511‐7, Huabio, China) overnight at 4°C, followed by fluorophore‐conjugated secondary antibody (Fluoro 647‑conjugated AffiniPure Goat Anti‑Rabbit IgG (H+L), Catalog# BA1150, Boster Biological Technology, Pleasanton, CA, USA) incubation in the dark. Nuclei were stained with DAPI, and fluorescence signals were observed and photographed under a confocal microscope to evaluate the reduction of intracellular PTPN1 protein expression after siRNA interference. All images were acquired under identical instrument settings, including laser power, exposure time, gain, and scanning resolution, which were kept consistent across all experimental and control groups. For each coverslip, at least three randomly selected non‐overlapping fields were captured, avoiding areas with cell clumps, debris, or morphologically abnormal cells. Quantitative analysis was performed using ImageJ (Fiji) software. Briefly, after standardized background subtraction (rolling ball radius = 50 pixels), the integrated density of PTPN1 fluorescence was measured for individual intact cells and normalized to the DAPI nuclear signal of the same field (ratio of PTPN1 fluorescence intensity/DAPI fluorescence intensity). The normalized values were used to represent relative PTPN1 protein expression levels.

2.19. Statistical Analysis

All statistical analyses were performed using GraphPad Prism 10.0 (GraphPad Software, La Jolla, CA, USA) or R software (version 4.3.2). Data are presented as the mean ± standard deviation (SD). Prior to statistical analysis, normality was assessed using the Shapiro–Wilk test, and homogeneity of variances was assessed using Levene's test. When data met both assumptions (normal distribution and equal variance), parametric tests were applied as described below; otherwise, non‑parametric tests (Mann–Whitney U test for two groups, or Kruskal–Wallis test for multiple groups) were used. In the present study, all cellular assay data passed both normality and equal variance tests, and thus parametric tests were used. Comparisons between two groups were conducted using unpaired Student's t‑test, while comparisons among multiple groups were performed using one‑way analysis of variance (ANOVA) followed by Tukey's post hoc test. All experiments involving statistical analyses were independently repeated at least three times. In all figures and figure legends, “n = 3” denotes three independent biological replicates, unless otherwise stated. Technical replicates (multiple wells or fields from the same biological replicate) were used only to calculate the intra‑experimental mean and were not treated as independent sample units for statistical inference. A p‑value < 0.05 was considered statistically significant. Significance levels are denoted as *p < 0.05, **p < 0.01, and ***p < 0.001.

3. Results

3.1. SOP Inhibits Pancreatic Cancer Proliferation: IC₅₀ Determination, Morphological Observation, and Colony Formation Validation

To investigate whether SOP suppresses pancreatic cancer growth, a series of biological assays were performed. The structure of SOP, in chemical terms, appears in Figure 1A. CCK‐8 results (Figure 1B–D) showed that SOP suppressed the proliferation of MIA PaCa‐2, SW1990, and PANC‐1 cells in a dose‐dependent manner. We adopted 48 h treatment data to calculate IC₅₀ instead of 24 h, since 24 h incubation only induced weak growth inhibition that failed to support reliable curve fitting, whereas 48 h treatment yielded stable inhibitory responses. The calculated IC₅₀ values were 22.77 µM for MIA PaCa‐2, 36.17 µM for SW1990, and 38.85 µM for PANC‐1, demonstrating MIA PaCa‐2 cells were most sensitive to SOP intervention.

FIGURE 1.

(A) Chemical structure of sophoricoside (SOP). (B–D) Cell viability of MIA PaCa‐2 (B), SW1990 (C), and PANC‐1 (D) cells treated with increasing concentrations of SOP for 48 h. IC₅₀ values are indicated. Data are mean ± SD from three independent biological replicates (n = 3).

Based on the lowest IC₅₀ value, MIA PaCa‐2 cells were selected as the primary cell model for subsequent validation, and parallel experiments were further performed in SW1990 cells to ensure result reliability. Morphological examination under a microscope revealed that treatment with 30 µM SOP for 48 h resulted in reduced cell growth and increased floating cells compared with the control group (Figure 2A,B). Quantitative cell counting further verified the obvious reduction in viable cell number after SOP intervention in both MIA PaCa‐2 and SW1990 cells (Figure 2C–F). In addition, a 14‐day colony formation assay demonstrated that 30 µM SOP significantly suppressed the clonogenic capacity of MIA PaCa‐2 cells and SW1990 cells (Figure 3A,B). Moreover, EdU proliferation staining confirmed that SOP remarkably reduced the proportion of S‐phase proliferative cells in the two pancreatic cancer cell lines, further solidifying the antiproliferative activity of SOP (Figure 3C–F).

FIGURE 2.

A and B) Representative morphological images of MIA PaCa‐2 and SW1990 cells following 48 h treatment with 30 µM SOP or vehicle control. Scale bar = 100 µm. (C–F) Quantitative statistical analysis of viable cell numbers in MIA PaCa‐2 and SW1990 cells with or without SOP intervention. Data are mean ± SD from three independent biological replicates (n = 3). Data significance: *p < 0.05, **p < 0.01, ***p < 0.001; NS, not significant.

FIGURE 3.

(A and B) Representative colony formation images and quantitative analysis showing the impaired clonogenic ability of MIA PaCa‐2 (A) and SW1990 (B) cells after 14‐day treatment with 30 µM SOP. (C–F) EdU proliferation staining and statistical analysis of EdU‐positive cell proportions in MIA PaCa‐2 (C and D) and SW1990 (E and F) cells with or without SOP treatment, indicating reduced S‐phase cell proliferation upon SOP intervention. Data are mean ± SD from three independent biological replicates (n = 3). Colony counts represent clusters containing ≥50 cells. Data significance: *p < 0.05, **p < 0.01, ***p < 0.001; NS, not significant.

3.2. Network Pharmacology Identified Candidate Targets and Mechanisms of SOP in Inhibiting Pancreatic Cancer

To identify the potential targets and underlying mechanisms by which SOP inhibits pancreatic cancer, network pharmacology analysis was performed. Potential targets of SOP and pancreatic cancer were predicted using multiple databases. A total of 3933 pancreatic cancer‐associated targets (relevance score > 10) were obtained, and 179 SOP‐related targets were identified. Venn diagram analysis revealed 85 overlapping targets between SOP and pancreatic cancer (Figure 4A), which formed a highly interconnected PPI network (Figure 4B).

FIGURE 4.

(A) Venn diagram showing 85 overlapping targets between SOP and pancreatic cancer. (B) PPI network of the 85 targets. (C–E) GO enrichment analysis (BP, CC, MF). (F) KEGG pathway enrichment analysis.

To gain insight into the possible biological processes and signaling cascades involving the 85 common targets, we performed enrichment analyses via the GO and KEGG databases. GO BP analysis revealed enrichment in terms such as negative regulation of response to external stimulus, positive regulation of the MAPK cascade, regulation of the ERK1 and ERK2 cascade, and response to xenobiotic stimulus (Figure 4C). GO CC analysis identified enrichment in terms including cytoplasmic vesicle lumen, basolateral plasma membrane, and secretory granule lumen (Figure 4D), while GO MF analysis showed enrichment in protein tyrosine kinase activity, hydro‐lyase activity, carbonate dehydratase activity, and growth factor receptor binding (Figure 4E). Finally, the KEGG enrichment assay showed that several malignancy‑associated pathways—namely, apoptosis, the PD‑L1/PD‑1 immune checkpoint axis, and nitrogen metabolism—were prominently overrepresented (Figure 4F).

3.3. Identification of Key Targets of SOP in Pancreatic Cancer via Machine Learning and Differential Expression Analysis

To identify key targets mediating the inhibitory effect of SOP on pancreatic cancer, machine learning algorithms and differential expression analysis were performed based on the GEO dataset GSE183795. Using the training dataset, we assessed the prediction accuracy of the eight classifiers—namely, RF, SVM, GLM, GBM, KNN, NNET, LASS, and DT. Model performance was assessed by reverse cumulative distribution of residuals (Figure 5A), precision–recall curves (Figure 5B), residual histograms (Figure 5C), and ROC curves (Figure 5D). The area under the ROC curve (AUC) values were 0.875 for RF, 0.868 for SVM, 0.732 for GLM, 0.839 for GBM, 0.736 for KNN, 0.808 for NNET, 0.873 for LASSO, and 0.685 for DT, indicating that RF and SVM exhibited superior predictive performance.

FIGURE 5.

(A–D) Performance evaluation of eight machine learning models. (E) Volcano plot of DEGs in GSE183795. (F) Venn diagram showing intersection of RF feature genes, upregulated DEGs, and 85 overlapping targets, yielding PTPN1.

Subsequent comparison of tumor and non‑tumor specimens from GSE183795 yielded 896 differentially expressed transcripts—622 up‑regulated and 274 down‑regulated—as depicted in the accompanying volcano plot (Figure 5E).

To identify the most core target of SOP in pancreatic cancer, the feature genes from the best‐performing machine learning model (RF) were intersected with the upregulated DEGs and the 85 SOP–pancreatic cancer overlapping targets identified by network pharmacology. This intersection yielded a single key target, protein tyrosine phosphatase non‐receptor type 1 (PTPN1) (Figure 5F).

3.4. Multi‐Omics Analysis Reveals the Expression and Biological Role of PTPN1 in Pancreatic Cancer

To further elucidate the expression pattern and biological function of PTPN1, a key target mediating the antipancreatic cancer effect of SOP, a multi‐omics analysis was performed. Chromosomal localization analysis indicated that PTPN1 is located on human chromosome 20 (Figure 6A). Notably, PTPN1 transcript levels were markedly upregulated in pancreatic tumor samples relative to their normal counterparts (Figure 6B). Notably, higher PTPN1 expression was linked to reduced overall survival in pancreatic cancer patients (Figure 6C).

FIGURE 6.

(A) Chromosomal localization of PTPN1. (B) PTPN1 mRNA expression in tumor versus normal tissues. (C) Overall survival analysis. (D–F) Immune cell infiltration analysis. *p < 0.05, **p < 0.01, ***p < 0.001.

In view of the established relationship between immune dysfunction and cancer advancement, our next step was to explore the correlation between PTPN1 levels and the degree of immune cell presence within tumors. The proportions of infiltrating immune cells were compared between groups with high and low PTPN1 expression (Figure 6D). Correlation analysis revealed that PTPN1 expression was positively correlated with multiple immune cell types (Figure 6E). Furthermore, significant differences in immune cell infiltration between the high‑ and low‑PTPN1 expression groups were observed across multiple immune cell subsets (Figure 6F).

To corroborate the protein‐level findings, the HPA database provided immunostaining data that demonstrated elevated PTPN1 expression within pancreatic cancer tissues versus healthy pancreatic samples (Figure 7A). This finding was further supported by analysis of CPTAC data, which revealed a statistically significant increase in PTPN1 protein expression in pancreatic cancer tissues compared with normal controls (Figure 7B). Moreover, PTPN1 protein expression levels were positively correlated with higher tumor grade (Figure 7C) and advanced tumor stage (Figure 7D).

FIGURE 7.

(A) Immunohistochemical images from HPA database. (B) CPTAC analysis of PTPN1 protein expression in tumor versus normal tissues. (C and D) PTPN1 protein expression by tumor grade (C) and stage (D). *p < 0.05, ***p < 0.001.

3.5. Functional Enrichment Analysis of PTPN1 in Pancreatic Cancer via GSEA, GSVA, and ssGSEA

To gain insight into the molecular mechanisms mediated by PTPN1—a primary target of SOP in pancreatic cancer—multidimensional functional enrichment analyses were carried out. GSEA based on PTPN1 expression levels in pancreatic cancer transcriptomic data revealed that upregulated pathways in the high‐expression group (GO‐UP) were enriched in terms such as cellular response to peptide hormone, microtubule organizing center, and regulation of cellular senescence (Figure 8A), whereas downregulated pathways (GO‐DOWN) were associated with ARF protein signal transduction, neurotransmitter metabolic process, and protein self‐association (Figure 8B). For KEGG pathways, upregulated terms (KEGG‐UP) included fructose and mannose metabolism, the TGF‐β signaling pathway, and the VEGF signaling pathway (Figure 8C), while downregulated terms (KEGG‐DOWN) were enriched in unsaturated fatty acid biosynthesis, butanoate metabolism, and vasopressin‐regulated water reabsorption (Figure 8D).

FIGURE 8.

(A and B) GO GSEA (A: upregulated, B: downregulated). (C and D) KEGG GSEA (C: upregulated, D: downregulated). FDR < 0.05 was considered significant.

GSVA further supported these findings. GO‐GSVA results showed enrichment in processes such as vesicle‐mediated cholesterol transport, mitochondrial crista, modification‐dependent protein binding, and Ras protein signal transduction (Figure 9A). KEGG‐GSVA analysis identified enrichment in taurine and hypotaurine metabolism, the insulin signaling pathway, O‐glycan biosynthesis, and adherens junctions (Figure 9B).

FIGURE 9.

(A) GO GSVA results. (B) KEGG GSVA results. FDR < 0.05 was considered significant.

ssGSEA was subsequently performed to compare pathway activities between tumor and normal samples. GO‐ssGSEA revealed that pathways including interleukin‐17‐mediated signaling, regulation of exocytosis, and regulation of extracellular exosome secretion were upregulated in tumor tissues (Figure 10A). KEGG‐ssGSEA indicated that fatty acid metabolism, the adipocytokine signaling pathway, and peroxisome pathways were also enriched in tumor samples (Figure 10B).

FIGURE 10.

(A) GO ssGSEA results. (B) KEGG ssGSEA results. ****p < 0.001.

3.6. SOP Interacts With PTPN1 and Suppresses its mRNA Expression

To investigate whether SOP exerts its antiproliferative effects in pancreatic cancer cells through PTPN1, molecular docking, RT‐qPCR, and CCK‐8 assays were performed. Molecular docking analysis predicted a possible binding mode between SOP and the PTPN1 protein (Figure 11A), involving four predicted hydrogen bonds with residues LYS131, GLN127, ASP137, and GLY93 (Figure 11B,C). Following treatment of MIA PaCa‐2 cells with SOP for 48 h, RT‐qPCR analysis demonstrated that SOP significantly suppressed PTPN1 mRNA expression (Figure 12A). Moreover, upon successful knockdown of PTPN1 expression (Figure 12B–D), CCK‐8 assays conducted at 12, 24, and 36 h showed that both SOP treatment and PTPN1 knockdown significantly inhibited MIA PaCa‐2 cell proliferation. Notably, combined treatment with SOP and PTPN1 siRNA did not result in a further reduction in cell viability compared with PTPN1 knockdown alone (Figure 12E).

FIGURE 11.

(A) Overall docking pose. (B) Detailed view of hydrogen bonds. (C) Schematic diagram of interactions with residues LYS131, GLN127, ASP137, and GLY93.

FIGURE 12.

(A) PTPN1 mRNA expression after SOP treatment. (B) Knockdown efficiency of PTPN1 siRNAs. (C) Immunofluorescence staining showing the protein‐level knockdown efficiency of PTPN1 siRNA (representative images). (D) Quantitative analysis of PTPN1 fluorescence intensity normalized to DAPI signal. (E) Cell viability after SOP treatment and/or PTPN1 knockdown. Data are mean ± SD from three independent biological replicates (n = 3). Data significance: *p < 0.05, **p < 0.01, ***p < 0.001; NS, not significant.

4. Discussion

In this study, we established a comprehensive and multidimensional research framework integrating experimental biology, network pharmacology, machine learning, and multi‐omics analysis to investigate the antiproliferative effects of SOP in pancreatic cancer and to elucidate its underlying molecular mechanisms. Our results suggest that SOP exerts a significant inhibitory effect on pancreatic cancer cell proliferation, and MIA PaCa‑2 cells were the most sensitive (IC₅₀ = 22.77 µM). Through an integrative computational approach combining network pharmacology, machine learning algorithms, and differential expression analysis, we identified PTPN1 as the key functionally required mediator of SOP's antipancreatic cancer activity. Through multi‑omics validation, we found that PTPN1 is upregulated in pancreatic cancer and correlates significantly with adverse clinical outcomes and immune cell infiltration. Functional enrichment analyses further indicated that PTPN1 is involved in multiple oncogenic pathways, including MAPK signaling, TGF‐β signaling, and metabolic reprogramming. Finally, molecular docking suggested a possible interaction between SOP and PTPN1, and functional experiments demonstrated that SOP suppresses PTPN1 mRNA expression, collectively supporting the functional involvement of PTPN1 in SOP's antiproliferative activity.

Our initial experimental findings indicated that SOP inhibits the proliferation of three pancreatic cancer cell lines in a dose‑dependent manner, with MIA PaCa‑2 cells exhibiting the highest sensitivity. Morphological observation, colony formation assays, and EdU further demonstrated that SOP treatment induces growth inhibition and reduces clonogenic capacity. These findings are consistent with previous reports demonstrating the anticancer activity of SOP in other tumor types. Consistent with our observations, SOP has been shown to exhibit antiproliferative effects in lung cancer [16], glioblastoma [17], and hepatocellular carcinoma [19], with reported experimental concentration gradients spanning 0–300, 0–40, and 0–600 µM across these models. The IC₅₀ values observed in our study (22.77–38.85 µM) are comparable to the low‐dose effective range documented in other cancer cell lines, where SOP demonstrated dose‐dependent growth inhibition at micromolar concentrations. Notably, the differential sensitivity among the three pancreatic cancer cell lines suggests that genetic background or specific molecular characteristics may influence cellular responsiveness to SOP, a phenomenon that warrants further investigation.

Network pharmacology analysis identified 85 overlapping targets between SOP and pancreatic cancer, which were enriched in pathways critical for tumor progression, including MAPK cascade regulation, ERK1/ERK2 signaling, and the PD‐L1/PD‐1 checkpoint pathway. These findings are consistent with the growing recognition that natural products often exert therapeutic effects through multi‐target mechanisms rather than single‐target interactions. The enrichment of protein tyrosine kinase activity and growth factor receptor binding among the molecular functions of SOP targets is particularly noteworthy, as these pathways are well‐established drivers of pancreatic cancer pathogenesis [39, 40]. The identification of apoptosis and nitrogen metabolism among the enriched KEGG pathways further supports the potential of SOP to modulate apoptosis‐related signaling and metabolic reprogramming in pancreatic cancer cells.

A distinctive feature of our study is the integration of machine learning algorithms to refine target identification. Among the eight models assessed, RF and SVM showed the highest predictive accuracy (AUC = 0.875 and 0.868, respectively). The intersection of RF feature genes, upregulated differentially expressed genes, and network pharmacology‐derived targets converged on a single candidate gene, PTPN1.

Our multi‐omics analysis revealed that PTPN1 is significantly upregulated in pancreatic cancer tissues at both the transcriptional and translational levels, and high PTPN1 expression correlates with poor overall survival. These findings are consistent with recent pan‐cancer analyses demonstrating that PTPN1 is frequently overexpressed in various malignancies and associated with unfavorable clinical outcomes [41]. PTPN1, a member of the PTP superfamily, regulates key signaling pathways involved in cell proliferation, differentiation, and survival [42]. Previous studies have identified PTPN1 as an oncogene in breast cancer, where its overexpression promotes tumor progression and correlates with immune infiltration [41, 43]. In the present study, the functional connection between SOP and PTPN1 was established through siRNA‐mediated knockdown experiments, which showed that PTPN1 silencing phenocopied the antiproliferative effect of SOP, with no additive effect upon combination. These data suggest that PTPN1 is functionally required for SOP's effect, although they do not distinguish between direct targeting and indirect pathway modulation. Our immune infiltration analysis extends these observations to pancreatic cancer, revealing significant correlations between PTPN1 expression and multiple immune cell types. While this trend hints that PTPN1 may drive pancreatic cancer progression via both cell‐intrinsic pathways and crosstalk with the tumor immune microenvironment, bulk RNA‐seq immune deconvolution only generates indirect predictive data and cannot establish causal immune regulatory relationships. Subsequent clinical and in vivo validation experiments are needed to confirm this immunomodulatory phenotype.

The functional enrichment analyses using GSEA, GSVA, and ssGSEA provided complementary insights into the biological processes associated with PTPN1 in pancreatic cancer. The upregulation of MAPK cascade [44], TGF‐β signaling [45], and VEGF signaling pathways [46] in PTPN1‐high tumors aligns with established roles of these pathways in pancreatic cancer progression. The enrichment of metabolic pathways, including fructose and mannose metabolism, fatty acid metabolism, and the adipocytokine signaling pathway, further highlights the potential involvement of PTPN1 in metabolic reprogramming, a hallmark of cancer. Notably, the enrichment of the interleukin‐17‐mediated signaling pathway and regulation of exosome secretion in tumor tissues suggests that PTPN1 may influence intercellular communication within the tumor microenvironment. These findings are consistent with recent studies demonstrating that PTPN1 expression correlates with immune checkpoint gene expression and immune cell infiltration in various cancers.

Molecular docking simulations in this study suggest that SOP may form hydrogen bonds with PTP1B (encoded by PTPN1) at residues LYS131, GLN127, ASP137, and GLY93, offering preliminary structural references for their potential intermolecular interaction. To assess the rationality of this predicted binding mode, we compared our computational results with previously published studies focusing on flavonoid‐PTP1B interactions [47, 48, 49, 50, 51, 52].

Lys120 is a well‐documented conserved lysine residue within the catalytic pocket of PTP1B that enables hydrogen bonding with the hydroxyl groups of flavonoids [47]. LYS131, the residue predicted to interact with SOP in our model, is located in the functionally equivalent region of the catalytic pocket and shares similar physicochemical properties, leading to consistent hydrogen‐bonding tendencies between the two binding patterns. In contrast, prenylated flavonoids such as gancaonin Q preferentially bind to the allosteric pocket of PTP1B to exert inhibitory effects, showing distinct binding preferences compared with the predicted binding site of SOP [48, 49]. Reported molecular docking scores of flavonoid‐PTP1B complexes generally range from −6 kcal/mol to −8 kcal/mol [47, 50]. The docking score obtained for the SOP‐PTP1B complex in this study falls within this documented range, indirectly providing preliminary computational support for the plausibility of this predicted conformation. In vitro functional assays have verified that isoflavones and C‑glycosylated flavones can inhibit PTP1B activity to varying degrees. Structure–activity relationship analyses further indicate that glycosylation modification does not completely abolish inhibitory potency, which is consistent with the possibility that glycosylated SOP may interact with PTP1B [50, 52].

Taken together, the predicted catalytic‐pocket interaction mode of SOP is broadly consistent with the interaction profiles of flavonoid glycosides and PTP1B described in existing literature. Importantly, our functional assays demonstrated that SOP treatment significantly suppressed PTPN1 mRNA expression in MIA PaCa‑2 cells, and that PTPN1 knockdown phenocopied the antiproliferative effects of SOP. Furthermore, no additive effect was observed when SOP treatment was combined with PTPN1 siRNA, suggesting that PTPN1 serves as a functionally relevant downstream mediator of SOP's antiproliferative activity in pancreatic cancer cells. This integrated approach—combining in silico prediction with functional validation—aligns with the recently recommended “phytochemical–biochemical–computational” strategy for PTPN1 inhibitor discovery.

Several inherent limitations of this study should be acknowledged. First, all functional and mechanistic validations were confined to in vitro pancreatic cancer cell models; in vivo antitumor activity and pharmacokinetic profiling of SOP in subcutaneous xenograft models remain to be established. Moreover, the absence of a positive chemotherapy control group in the cellular functional assays precludes quantitative comparison of the antiproliferative potency of SOP with that of first‑line pancreatic cancer chemotherapeutic agents. Second, although integrative multi‑omics screening and functional knockdown experiments have identified PTPN1 as a functionally relevant downstream effector molecule, the evidence supporting PTPN1 as a direct mediator remains indirect. Specifically, this study lacks in vitro protein–small molecule binding assays (e.g., CETSA, MST, SPR, or DARTS) to confirm direct physical interaction between SOP and PTP1B, as well as PTPN1 overexpression rescue experiments and direct cell death detection assays (e.g., Annexin V/PI staining, LDH release, or caspase activation). Consequently, we are unable to determine whether the observed growth inhibition results from cytostatic or cytotoxic effects. Third, while the multi‑omics screening points to PTPN1 as a prioritized candidate, SOP may act through a complex multi‑target network, and the cross‑regulatory relationships among other potential targets remain to be elucidated. In addition, the immune infiltration analyses in this study were based on bulk RNA‑seq data using deconvolution algorithms, which can only provide indirect, predictive evidence and cannot establish causal relationships.

From a translational perspective, natural flavonoid glycosides generally exhibit low oral bioavailability and modest in vivo efficacy [53, 54, 55], meaning current in vitro evidence alone is insufficient to support SOP as a standalone monotherapy. Instead, SOP is more promising as an adjuvant agent combined with conventional chemotherapy to achieve synergistic antitumor effects. In future work, we will add positive drug control groups, construct xenograft tumor models, and carry out clinical tissue immune staining and single‐cell transcriptomic profiling to systematically verify our bioinformatic immune predictions, fully evaluating the single‐drug efficacy, combinatorial therapeutic value and immunoregulatory capacity of SOP.

5. Conclusion

This study provides the first comprehensive evidence that sophoricoside inhibits pancreatic cancer proliferation, with PTPN1 identified as a key functionally relevant downstream mediator based on integrative computational and functional evidence. By integrating experimental validation with computational pharmacology, machine learning, and multi‐omics analysis, we have established a robust framework for elucidating the mechanisms of natural products in cancer therapy. Combined computational prediction and cellular functional data suggest that PTPN1 may act as a functionally relevant mediator underlying the antiproliferative effect of SOP in pancreatic cancer cells, implying the potential therapeutic relevance of PTP1B.

Author Contributions

Peng Lin: conceptualization, methodology, software, formal analysis, investigation, data curation, writing – original draft, visualization. Wei Cheng: validation, resources, writing – review and editing. Xin Qi: resources. Jing Li: resources, supervision, project administration, funding acquisition. All authors read and approved the final manuscript.

Funding

The funding for our study was provided by various sources, including the Hainan Provincial Joint Project of Sanya Yazhou Bay Science and Technology City (2021CXLH0012), the State Key Program of National Natural Science of China (82030074), major basic research projects from the Shandong Provincial Natural Science Foundation (ZR2021ZD28), the National Natural Science Foundation of China (82273847), and the Shandong Provincial Natural Science Foundation (ZR2023MH116).

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The datasets generated and analyzed during the current study are available from public repositories. The gene expression datasets GSE183795 were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/).