Stemness signatures reflect prognostic disturbances in gastric cancer

Abstract

BACKGROUND

Tumors characterized by high cellular stemness often have unfavorable clinical outcomes, primarily due to their heightened potential for metastasis and resistance to chemotherapy. Among the model genes, the clinical relevance and prognostic significance of Niemann-Pick type C2 (NPC2) in gastric cancer (GC) remained largely unexplored.

AIM

To identify stemness-associated genes in GC.

METHODS

In this study, epithelial cells were categorized as either tumor or normal epithelial cells using the infer copy number variation method. Stemness scores were calculated for both cell types. The hierarchical Weighted Gene Co-expression Network Analysis identified two gene modules with the strongest association with stemness. Prognostically significant stemness-related genes were pinpointed using univariate Cox regression based on The Cancer Genome Atlas dataset. A predictive model related to stemness was constructed using Least Absolute Shrinkage and Selection Operator regression followed by multivariate Cox analysis.

RESULTS

Functional roles of NPC2 were validated using single-cell and bulk RNA sequencing data. Further experimental validation revealed that elevated NPC2 expression promoted tumor cell stemness, invasiveness, migratory ability, and resistance to standard chemotherapeutic agents. Importantly, high NPC2 expression correlated with poorer overall survival in GC patients.

CONCLUSION

In summary, the proposed model offers prognostic insights that outperform traditional clinical staging and may inform more tailored therapeutic approaches for gastric cancer management.

Key Words: Gastric cancer; Gastric cancer stemness sensitive; Cancer stem cells; Gastric cancer stem cells; Niemann-Pick protein C2

Core Tip: The results of this study indicate that the proposed gastric cancer (GC) stemness sensitivity model offers prognostic value that exceeds that of conventional clinicopathological staging, therefore providing more precise guidance for clinical decision-making and post-treatment management. The strong positive association between Niemann-Pick type C2 (NPC2) expression and GC progression highlights NPC2 as a promising prognostic biomarker. However, the functional role of NPC2 in GC remains insufficiently characterized. In this study, cellular experiments demonstrated that elevated expression of NPC2 enhances tumor cell stemness, invasiveness, migratory capacity, and resistance to radiotherapy and standard chemotherapeutic agents.

INTRODUCTION

Gastric cancer (GC) is a common malignant tumor in the digestive tract, ranking fifth in incidence among malignant tumors and third as a cause of cancer-related mortality[1,2]. Current treatment options for GC include surgery, chemotherapy, and targeted therapies. Despite improvements in diagnosis and treatment that have led to a gradual decrease in the incidence of GC, the prognosis for patients with advanced stages of the disease remains bleak. Recurrence, metastasis, and chemotherapy resistance contribute to low overall survival rates, with the median survival of patients being less than one year[3]. Thus, identifying new diagnostic and therapeutic targets is critical for improving the prognosis of patients with advanced GC, highlighting the necessity for developing new prognostic models for this tumor. The presence of cancer stem cells (CSCs) and the process of epithelial-mesenchymal transition (EMT) are primary factors in tumor metastasis and recurrence. Cell stemness, defined as the ability for self-renewal and differentiation from cells of origin, has gained attention due to the discovery of CSCs in recent years. CSCs possess the capacity for self-renewal, indefinite proliferation, and differentiation, significantly contributing to tumor initiation, primary tumor growth, recurrence, and the spread of tumor cells. These cells are fundamentally responsible for tumor recurrence, metastasis, and resistance to treatments[4]. To date, the presence of CSCs has been demonstrated in a variety of malignant tumors, including breast cancer, leukemia, glioblastoma, hepatoma, and GC. Previous studies have shown that tumors with high cell stemness are more likely to develop metastasis and exhibit chemotherapy resistance, ultimately leading to poor prognosis[5,6].

Niemann-Pick type C2 (NPC2) is an evolutionarily conserved protein widely distributed across various tissues and organs, primarily localized within late endosomes and lysosomes where it facilitates cholesterol transport[7,8]. Accumulating evidences have indicated that NPC2 is closely related to tumor occurrence and development and plays an important role in tumor cell differentiation, proliferation and tumor microenvironment regulation[9-12]. Recent studies have established a notable correlation between NPC2 expression and various human cancers, including GC. Studies have demonstrated a robust correlation between NPC2 expression and tumor presence, with notable overexpression observed in GC, breast cancer, colon cancer, and lung cancer[13,14]. Yao et al[15] have demonstrated that NPC2 could serve as a prognostic biomarker for GC, potentially enhancing early diagnosis, treatment strategies, and personalized care. However, the clinical implications and prognostic value of NPC2 expression in GC remain unclear, with limited reports on the NPC2-GC relationship. The present study aimed to explore the clinical relevance of the interaction between NPC2 and GC, aiming to improve the diagnostic, therapeutic, and prognostic approaches in managing GC.

This study provides novel therapeutic targets for treating GC and lays the groundwork for future drug development. It explores the prognostic and oncogenic potential of significant molecular markers, enhancing the ability to predict the outcomes of GC. By examining these markers, this research not only contributes to a deeper understanding of the pathology of this disease but also opens avenues for developing targeted therapies that could potentially improve survival rates and reduce the recurrence of this aggressive cancer.

MATERIALS AND METHODS

Data collection and procession

This study obtained single-cell RNA sequencing (scRNA-seq) data for GC, along with bRNA-seq data and validation datasets. The datasets analyzed in this research are publicly available in the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) and TCGA, https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga repositories, with accession IDs GSE183904[16], GSE84437[17], and TCGA-STAD. Dimensionality reduction of the scRNA-seq data was performed using Principal Component Analysis and Uniform Manifold Approximation and Projection (UMAP), with quality control measures applied through the 'Seurat' R package[18]. Cell annotation was carried out using marker genes identified in previous studies[19].

Stemness, CNV, function, drug reaction, and immune analysis

Epithelial cells were extracted from the scRNA-seq dataset, and their expression data were used to evaluate stemness and CNV using the 'Cytotrace'[20] and 'inferCNV'[21] R packages, respectively. Gene lists for functional enrichment analysis were sourced from the Molecular Signatures Database, with analyses performed using the 'clusterProfiler' R package. Drug sensitivity information was retrieved from the Genomics of Drug Sensitivity in Cancer database, and cell sensitivity predictions were made using the 'beyondcell' R package. Immune-related analysis of the bulk RNA-seq data was conducted using the 'IOBR' R package.

Analysis and model establishment of hdWGCNA

Stemness-related gene modules were identified through hdWGCNA[22]. The relationship between these modules and stemness indices was assessed to determine the most relevant modules. Prognostically significant modules were then selected using univariate Cox regression analysis. Key genes within these modules were refined using LASSO regression. The optimal genes identified by LASSO were further examined through multivariate Cox regression to build prognostic models. The performance of these models was evaluated using Kaplan-Meier (KM) and receiver operating characteristic (ROC) curves. All analyses were conducted with R packages such as 'survminer', 'survival', and 'glmSparseNet' for prognosis analysis.

Analysis of clinical features

Initially, the impact of clinical features and the developed model on prognosis was evaluated using univariate and multivariate Cox regression analyses. Afterward, nomograms were constructed by incorporating the model and clinical features with P-values less than 0.05 in the univariate Cox analysis, utilizing the 'rms' R package. The predictive accuracy of both the clinical features and the nomograms for prognosis at 1, 3, and 5 years was then assessed through ROC curves.

Tumor spheroid formation experiment

The cells were treated with trypsin and seeded into 6-well plates at a density of approximately 20000 cells per well, using Dulbecco's Modified Eagle Medium (DMEM)-F12 supplemented with 20 ng/mL B27 and 20 ng/mL epidermal growth factor, but without FBS. The medium was refreshed every two days, retaining the previous medium. After 14 days of culture, spheroid formation was assessed using a microscope.

Migration and invasion test

For invasion assays, Transwell chambers were pre-coated with 50 μL of Matrigel matrix (diluted 1:8 in serum-free medium) and allowed to polymerize at 37 °C for 4 hours. Migration assays were performed using uncoated chambers as parallel controls. The transfected AGS and HGC27 cell lines were treated with 0.25% trypsin, centrifuged, and resuspended in serum-free medium at a concentration of 2 × 10⁴ cells/200 μL, then seeded into the upper chambers and equilibrated at 37 °C for 30 minutes before assembly. The lower chambers contained 500 μL of DMEM supplemented with 20% FBS as a chemoattractant. After 24 hours of incubation for migration or 48 hours for invasion, non-migrated cells in the upper chamber were removed using a cotton swab. Cells that had migrated or invaded through the membrane were fixed with methanol and stained with 0.1% crystal violet. These cells were then examined and photographed using an inverted microscope to quantify the extent of cell invasion.

Cell counting kit-8 assay

After transfection, AGS and HGC27 cells at 90% confluence were treated with 0.25% trypsin to generate cell suspensions at a concentration of 2 × 10⁴ cells/mL in DMEM. The viability of each cell group was assessed on days 1, 2, 3, 4, and 5 using the cell counting kit-8 (CCK8) reagent. Cellular activity was measured using a microplate reader at a wavelength of 450 nm.

Clonogenicity assay

Single-cell suspensions were prepared from each cell group and seeded into culture dishes at approximately 500 cells per dish. The dishes were gently shaken to ensure uniform cell distribution. Cells were incubated at 37 °C with 5% CO₂ for 14 days to allow colony formation. After incubation, cells were washed with phosphate-buffered saline (PBS; 1 × concentration, pH 7.4), fixed with 4% paraformaldehyde, and stained with 0.1% crystal violet for 15 minutes. Colonies were counted under a microscope. Colonies that met predefined criteria (≥ 50 contiguous cells with intact cell-cell contacts) were quantified using ImageJ software (size threshold: 200 μm²), with manual verification of ≥ 20% randomly selected fields. Edge colonies (within 5 mm of the dish periphery) and amorphous aggregates were excluded from the analysis.

Immunohistochemistry

The GC tissues with prognoses of less than 5 years, more than 5 years, and adjacent non-cancerous tissues were fixed in formalin and embedded in paraffin. All tissue samples were obtained from the Department of Pathology at Nanchong Central Hospital. This study was approved by the Ethics Committee of Nanchong Central Hospital (approval number 2024 No. 113). We confirm that human GC tissue samples were used in compliance with relevant ethical guidelines and regulations. Informed consent was obtained from all subjects and/or their legal guardians for using their tissue samples in this research. Each group included six samples. Tissue sections were cut to a thickness of 4 μm for analysis. Hematoxylin and eosin staining was performed to assess pathological features. For immunohistochemistry (IHC), the sections were incubated overnight at 4 °C with primary antibodies targeting human NPC2. After washing three times with PBS, the sections were incubated with a biotinylated secondary antibody (Vectastain ABC Kit, Vector Laboratories, CA) for 30 minutes at room temperature. Pathological evaluations were performed by two pathologists, blinded to the patient's clinical details, under a light microscope. For IHC analysis, lighting conditions were standardized before capturing images to ensure consistency in data quality. ImageJ software was used to measure the area and integrated optical density (IOD) of all positively stained cells, and the Average Optical Density (AOD) was calculated using the formula: AOD = IOD/Area.

Western blot

Protein lysates were extracted from cultured cells using ice-cold RIPA buffer containing protease and phosphatase inhibitors. Protein concentrations were determined using the BCA assay with a microplate reader. Following denaturation at 95 °C for 5 minutes, 20 μg of protein per sample was loaded onto 12% SDS-PAGE gels and separated under dual voltage conditions alongside pre-stained molecular weight markers. Electrophoresed proteins were transferred to PVDF membranes via wet transfer (200 mA, 0.5 hours) in a methanol-containing transfer buffer. Membranes were blocked with 5% BSA in TBST and then incubated sequentially with primary antibodies [anti-NPC2 (1:1000) and anti-glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (1:10000)] at 4 °C overnight. Afterward, horseradish peroxidase-conjugated secondary antibodies (1:5000) were applied for 1 hour at room temperature. Chemiluminescent signals were developed using enhanced chemiluminescence substrate, captured at multiple exposures on a chemiluminescence imaging system, and quantified by densitometric analysis with background subtraction. NPC2 expression levels were normalized to GAPDH for each biological replicate.

Reverse transcription polymerase chain reaction

Total RNA was extracted from cell lysates using TRIzol^™ reagent according to the manufacturer's protocol. One microgram of total RNA was reverse-transcribed using PrimeScript^™ RT Master Mix in a 20 μL reaction volume under thermal cycling conditions. Quantitative polymerase chain reaction (PCR) amplification was performed on a LightCycler^® 480 System using TB Green^™ Premix Ex Taq^™ II with the following cycling profile. Relative mRNA expression levels were analyzed using the 2^-△△Ct method, with GAPDH as the internal control. Target genes were amplified using the following primer pairs (5′→3′): SOX2 (F: GCTACCTCATGAAGAAGGCAAC, R: TCTGCGAGCTGGTCATGGAG); OCT4 (F: GGGAGATTGATAACTGGTGTGTT, R: GTGTATATCCCAGGGTGATCCTC); CD133 (F: CAGGTAACGGAGCATTGGC, R: CAGAGTACAACGCCAACCAC); NANOG (F: CAAAGGCAAACAACCCACTT, R: TCTGCTGGAGGCTGAGGTAT); and GAPDH (F: GGAGCGAGATCCCTCCAAAAT, R: GGCTGTTGTCATACTTCTCATGG).

Statistical analysis

All bioinformatics analyses were performed using R (Version 4.4.0), with P values less than 0.05 considered statistically significant. Statistical evaluations were conducted using the t test. The migration and invasion assays, CCK8 assay, clonogenicity test, and IHC assay were carried out using GraphPad Prism 8. Statistical analysis was performed using the t test for these tests, and results with P values below 0.05 were regarded as statistically significant.

RESULTS

Single-cell analysis of GC samples

The study was initiated by analyzing the scRNA-seq data from the GEO database, which contained samples of both normal and tumor tissues (Figure 1B). Using established markers from previous studies (Figure 1C), we categorized the cells into three primary groups: Epithelial cells, stromal cells, and immune cells (Figure 1A). We isolated the epithelial cells from this dataset and assessed their CNV scores, using immune cells as a reference (Figure 1D). This analysis revealed that tumor cells frequently show significant CNV, with significant gains or losses in specific genomic regions (Figure 1E). We then further divided the epithelial cell group into malignant tumor cells and non-malignant epithelial cells (Figure 1F). The Cytotrace algorithm analysis showed that tumor cells demonstrated higher stemness characteristics than their non-malignant counterparts (Figure 1G). This distinction highlights the potential to identify critical biological differences that could guide the development of targeted therapeutic strategies.

Figure 1 Single-cell analysis of gastric cancer dataset. A: Uniform manifold approximation and projection (UMAP) visualization of identified cell populations within the gastric cancer (GC) dataset; B: UMAP representation showing the origin of tissues in the GC dataset; C: Dot plot illustrating the expression of marker genes used to identify different cell types; D: Copy number variation (CNV) scores for epithelial cell clusters, as predicted by inferCNV; E: Heatmap depicting CNV across epithelial cells; F: UMAP plot highlighting malignant cells as identified by inferCNV; G: Stemness scores across cells were determined using the Cytotrace algorithm.

hdWGCNA screening for the most relevant modules for stemness in GC

To identify the most significant stemness-related genes in GC, we analyzed epithelial cells using hdWGCNA. The analysis began by determining the optimal soft threshold value, which was calculated to be 8, enabling precise module clustering (Figure 2A). Using UMAP for dimensionality reduction, nine distinct gene modules were identified (Figure 2B). Key representative genes from these modules are shown in (Figure 2C). To determine which modules were most strongly associated with stemness characteristics, we evaluated their correlations with Cytotrace scores and the distinction between cancerous and adjacent non-cancerous tissues (Figure 2D). This analysis revealed that the turquoise and pink modules correlated highly with these parameters. Further analysis of these two modules included calculating the core gene interactions within the turquoise (Figure 2E) and pink (Figure 2F) modules. These interactions suggest that the turquoise and pink modules, in particular, may be critical for understanding the stemness properties of GC and could serve as valuable targets for future studies on the disease's biology and therapeutic strategies.

Figure 2 Development of optimal stemness modules via hierarchical weighted gene co-expression network analysis. A: Selection of the best soft-thresholding power; B: Uniform manifold approximation and projection plot displaying the dimensionality reduction of nine identified modules; C: Representative genes from each module; D: Correlation analysis between stemness scores and modules; E: Gene-gene interaction network within the turquoise module; F: Gene-gene interaction network within the pink module.

Construction of stemness model and functional enrichment analysis of independent risk factor NPC2

Univariate Cox regression analysis performed on genes identified through module analysis within the TCGA-STAD dataset revealed 28 genes significantly associated with prognosis (Figure 3A). Further differential analysis of single-cell data showed marked differences between all genes in tumor cells compared to epithelial cells (Supplementary Figure 1A) and between cancerous and adjacent non-cancerous tissues (Supplementary Figure 1B). In the bulk RNA-seq data from TCGA, all genes, except for CD99, GEM, and EPHB3, demonstrated significant differences in the paraneoplastic group, and these trends mirrored those observed in the single-cell differential analysis (Supplementary Figure 1C). KM analysis confirmed that the remaining 25 genes were significantly correlated with survival outcomes (Supplementary Figure 2, P value < 0.05). To further refine the gene set, LASSO regression was applied to optimize the gene composition (Figure 3B). Multivariate Cox regression analysis identified NPC2 as an independent prognostic factor, showing the highest hazard ratio (HR) (Figure 3C). A gastric cancer stemness sensitivity (GCSS) model was constructed using multivariate Cox regression (Figure 3D). The GCSS score, based on gene expression levels in the model, provides a prognosis indicator, with higher scores correlating with poorer outcomes. Patients were classified into high and low GCSS score groups based on the median score. Interestingly, NPC2 expression was significantly higher in the high GCSS score group compared to the low score group (Figure 3D). To explore the relationship between high NPC2 expression and poor prognosis, tumor cells from the single-cell dataset were clustered into 21 groups (Figure 3E), which were then divided into high and low NPC2 expression categories (Figure 3F). The differentially expressed genes (DEG) between these groups were analyzed for enrichment in Gene Ontology (GO), (Figure 3G), Kyoto Encyclopedia of Genes and Genomes (KEGG) (Figure 3H), and various pathways (Figure 3I). The low NPC2 expression group was predominantly enriched in pathways related to ribosomal function. Park et al[23] showed that SMYD5 methylation of rpL40 is linked to increased ribosomal activity in GC. Similarly, Wang et al[24] reported reduced METTL5 protein levels in GC tissues compared to adjacent intestinal metaplasia and normal tissues, suggesting its potential as a prognostic biomarker in GC. In another analysis, bulk RNA-seq data were divided into high and low NPC2 expression groups based on median values, and DEGs were identified and displayed in a heatmap (Figure 3J). These DEGs were found to be enriched in the regulation of T cell activation pathways according to the Gene Ontology database (Figure 3K).

Figure 3 Development of the gastric cancer stemness sensitivity model and functional enrichment analysis of Niemann-Pick type C2. A: Univariate Cox regression analysis of module genes; B: Selection of optimal genes for prognosis prediction using LASSO regression; C: Multivariate Cox regression analysis of selected model genes; D: Comparison of gastric cancer stemness sensitivity (GCSS) scores, survival times, and model gene expression between high and low GCSS groups; E: Uniform manifold approximation and projection (UMAP) visualization of clustered tumor cells from scRNA-seq data; F: UMAP plots showing groups with high and low Niemann-Pick type C2 (NPC2) expression; G-I: Enrichment analyses for the NPC2 high-expression group across Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and other pathway datasets; J: Heatmap of differentially expressed genes between NPC2-high and -low groups from bulk RNA-seq; K and L: Enrichment analysis of differentially expressed genes in GO and KEGG.

Baumeister et al[25] highlighted that GC tissues harbor a significant number of regulatory T cells showing high cytotoxic T-lymphocyte-associated protein 4 expression, which competes with CD28 molecules on T cells for B7 molecule binding, inhibiting T cell activation and proliferation. The adhesion molecule pathway was also prominently featured in the KEGG database (Figure 3L). Epithelial cell adhesion molecules, frequently used as markers in CSCs, have been documented to show higher expression in GC compared to control groups, with overexpression linked to increased tumor size, lymph node metastasis, and poorer prognosis in GC[26].

NPC2 promoted proliferation, migration, invasion, and an increase in stemness of GC cells

To investigate the impact of NPC2 on the biological behavior of GC cells, three siRNA sequences targeting NPC2 were designed. Following transfection, Western blot analysis was performed to assess NPC2 protein levels. Among the siRNAs, Si-2 significantly reduced NPC2 expression (Figure 4A) and was selected for further experiments. As expected, NPC2 knockdown decreased the proliferation of AGS and HGC27 cell lines, as demonstrated by the data (Figure 4B). A clonogenicity assay was carried out to evaluate the long-term effects of NPC2 suppression on cell renewal and proliferation. This assay revealed a marked reduction in colony formation by AGS and HGC27 cells following NPC2 knockdown (Figure 4C), indicating that NPC2 is critical in promoting in vitro proliferation of GC cells.

Figure 4 Impact of Niemann-Pick type C2 on the biological characteristics of gastric cancer cells. A: Western Blot analysis demonstrates that siRNA sequence Si-2 substantially reduced Niemann-Pick type C2 (NPC2) protein levels; B: Reduction in cell proliferation observed in AGS and HGC27 cell lines following NPC2 knockdown; C: Decrease in colony formation by AGS and HGC27 cells after NPC2 knockdown, indicating reduced clonogenic potential; D: Transwell invasion assays demonstrating that siNPC2-treated AGS and HGC27 cells exhibited reduction in migrated and invaded cell numbers compared to the NC. Scale bar: 200 μm; E: Spheroidization experiments showing a decrease in spheroid diameter and impaired spheroidization ability in AGS and HGC-27 cells following NPC2 knockdown; F: Reverse transcription-quantitative polymerase chain reaction profiling of stemness markers (SOX2, OCT4, CD133, NANOG) showing significant transcriptional downregulation in NPC2-depleted cells; G: Immunohistochemistry (IHC) analysis of NPC2 expression in gastric cancer (GC) tissues and adjacent non-cancerous tissues. A statistical representation of IHC results, summarizing NPC2 expression across different patient groups; H and I: NPC2-associated drug resistance in GC treatment was analyzed using Beyondcell.

To further investigate the functional impact of NPC2 on malignant progression, we next evaluated the effects of NPC2 knockdown on metastatic potential using Transwell assays. Remarkably, NPC2-depleted AGS and HGC27 cells exhibited reduction in migratory and invasive capacity compared to NC (Figure 4D). Additionally, a spheroidization assay was conducted, which showed that the spheroids formed by AGS and HGC-27 cells were smaller in diameter after NPC2 knockdown, and their spheroidization capacity was diminished (Figure 4E). Consistent with the functional observations in spheroid formation, quantitative reverse transcription PCR analysis revealed concomitant downregulation of stemness-associated transcription factors in NPC2-knockdown cells. Specifically, AGS and HGC-27 cells subjected to NPC2 knockdown exhibited significant reductions in mRNA expression levels of SOX2, OCT4, CD133, and NANOG compared to NC (Figure 4F). This suggests that NPC2 not only supports cell proliferation but also impacts other aspects of GCici cell pathology, such as the ability to form spheroids, which is indicative of enhanced tumorigenty and stemness. Collectively, these data highlight the critical role of NPC2 in promoting proliferation and stemness in GC cells, contributing to the aggressive progression of the disease. IHC analysis further revealed that NPC2 expression was significantly higher in GC tissues from patients with a survival rate of less than 5 years compared to those with a survival rate of over 5 years. Moreover, NPC2 expression was particularly elevated in GC tissues from patients with a survival duration greater than 5 years, in contrast to adjacent non-cancerous tissues (Figure 4G). Moreover, a significant reduction in NPC2 expression was observed in adjacent non-cancerous tissues (Figure 4G). Statistical analysis of the IHC data (Figure 4G) suggested that NPC2 expression may be a crucial determinant influencing the survival rates of GC patients, positioning it as a potential prognostic biomarker and indicating its possible impact on disease progression. Based on scRNA-seq data, a drug sensitivity analysis performed on tumor cells using the Beyond Cell tool revealed that NPC2 could play a role in drug sensitivity in GC cells. Specifically, NPC2 was found to potentially enhance sensitivity to several commonly used chemotherapeutic and radiotherapeutic agents, including cisplatin, fluorouracil, doxorubicin, and capecitabine (Figure 4H and I). These results suggest that NPC2 may be a key factor influencing the sensitivity of GC patients to radiotherapy and chemotherapy, potentially affecting the effectiveness of these treatments. This underscores the importance of personalized therapies to address the genetic heterogeneity observed in GC.

Validation of the GCSS model efficiency, immunoassay, and functional analysis

In the validation cohort (TCGA-STAD), the GCSS model was established using multivariate Cox regression, and KM analysis demonstrated its strong prognostic significance (Figure 5A). ROC analysis showed that the area under the curve (AUC) values for the GCSS model effectively predicted survival at 1, 3, and 5 years, with AUCs of 0.70, 0.67, and 0.71, respectively (Figure 5B), therefore confirming the model's reliability. In an additional validation set (GSE84437), KM analysis further supported the model's ability to differentiate between poor and favorable prognoses (Figure 5C). At the same time, the AUC curve validated its high predictive accuracy for 1, 3, and 5-year survival outcomes (Figure 5D). Differences in the immune microenvironment between high and low GCSS score groups were also observed (Figure 5E). Immune infiltration was analyzed using eight algorithms: CIBERSORT, EPIC, ESTIMATE, IPS, MCPCOUNTER, QUANTISEQ, TIMER, and XCELL (Figure 1A). The results revealed significantly higher infiltration of cancer-associated fibroblasts (CAFs) in the high GCSS group, with findings from EPIC and stromal scores from ESTIMATE. Fibroblast counts from MCPCOUNTER and smooth muscle assessments from XCELL. Previous studies have emphasized the role of CAFs in GC progression. Specifically, CAFs have been shown to provide GC cells with the signaling protein WNT5A and its receptor ROR2, which sustains the activation of the metastasis receptor ROR2 in tumor cells, influencing cellular polarization and migration[27].

Figure 5 Evaluation and analysis of the gastric cancer stemness sensitivity model. A and B: Kaplan-Meier (KM) survival and receiver operating characteristic (ROC) curves for prognosis prediction in The Cancer Genome Atlas training set; C and D: KM survival and ROC curves for prognosis prediction in the GSE84437 validation set; E: Differences in the immune microenvironment between high and low gastric cancer stemness sensitivity (GCSS) score groups; F-I: Correlation analysis between GCSS scores and functional pathways, with enrichment shown in GO, HALLMARK, KEGG, and REACTOME databases. ^aP < 0.05, ^bP < 0.01, ^cP < 0.001. TCGA: The Cancer Genome Atlas; GCSS: Gastric cancer stemness sensitivity; KEGG: Encyclopedia of Genes and Genomes.

Moreover, CAF-secreted miR-522 has been shown to inhibit iron efflux in GC cells and contribute to the development of acquired chemotherapy resistance[28]. CAFs in GC also promote chemotherapy resistance through the expression of NRP2[29]. Further analysis of the associations between pathway enrichment and the GCSS score was performed across four databases: GO (Figure 5F), HALLMARK (Figure 5G), KEGG (Figure 5H), and REACTOME (Figure 5I). The results revealed significant correlations between the GCSS score and several pathways, including Wnt and transforming growth factor (TGF) signaling. Song et al[30] demonstrated that in spherical GC cells with strong self-renewal capabilities and the expression of CSC-related markers, inhibition of the Sonic Hedgehog pathway reduced the self-renewal properties of GC stem cells (GCSCs) and enhanced their sensitivity to chemotherapeutic agents. Mao et al[31] identified that the classical Wnt pathway is critical in maintaining stemness, proliferation, and EMT in GCSCs. Fan et al[32] found that the upregulation of miR-501-5p in GC cells activated the Wnt/β-catenin signaling pathway, promoting stem cell-like characteristics. Furthermore, several studies have highlighted the role of the TGF pathway in GC cell stemness[33-35]. Peng et al[36] reported that SRY-related HMG-box 4 (SOX4), a target of TGF-beta (TGF-β) signaling, mediates TGF-β-induced EMT and stem cell traits in GC cells, thus revealing a novel role of the TGF-β/SOX4 axis in regulating the malignant behavior of GC[36].

The GCSS model and analysis of clinical features

A heatmap was generated to examine the relationships between model genes and clinical factors such as tumor stage (T, M, N), race, gender, and GCSS level (Figure 6A), which revealed no significant correlations among these clinical variables. In the univariate Cox regression analysis, the GCSS score, acting as a risk factor, revealed the highest HR (Figure 6B). In the subsequent multivariate Cox regression analysis, the GCSS score remained the most significant independent risk factor with a P-value less than 0.05, demonstrating improved performance in survival prediction when compared to traditional clinical staging (Figure 6C). A nomogram was constructed by incorporating variables with p-values below 0.05 from the univariate Cox analysis (Figure 6D). The predictive accuracy of the nomogram, along with individual clinical factors like stage (Figure 6E), T (Figure 6F), M (Figure 6G), and N (Figure 6H), for forecasting 1-, 3-, and 5-year survival rates was assessed using ROC curves (Figure 6I). While the clinical factors alone did not predict survival as effectively as the GCSS model, the nomogram combined these factors. It provided an enhanced prognosis prediction and demonstrated greater clinical applicability than the GCSS model alone.

Figure 6 Integration and evaluation of clinical features with the gastric cancer stemness sensitivity model. A: Heatmap displaying clinical features, GCSS gene expression, and gastric cancer stemness sensitivity (GCSS) levels; B: Forest plot from univariate Cox regression analysis of clinical features, GCSS gene, and GCSS score; C: Forest plot from multivariate Cox regression analysis detailing the impact of clinical features, GCSS gene, and GCSS score; D: Construction of a nomogram incorporating pathological stage, stages T, M, N, and GCSS score for prognosis prediction; E-I: Receiver operating characteristic curves assessing the predictive accuracy of the pathological stage (E), stage T (F), stage M (G), stage N (H), and nomogram score (I).

DISCUSSION

GC is a common and life-threatening malignancy that continues to pose a significant public health challenge. In cases of early-stage GC, endoscopic resection is the preferred treatment option. For patients with operable but non-early GC, surgical resection is typically recommended. Furthermore, perioperative or adjuvant chemotherapy has been shown to improve survival outcomes in individuals with stage IB or more advanced disease, with platinum and fluorouracil-based regimens serving as the standard first-line therapy for advanced GC. Despite significant progress in diagnostic techniques and therapeutic strategies, recurrence and metastasis remain frequent in advanced stages, contributing to the poor prognosis[37,38]. Existing treatments often fall short for patients with advanced GC, with metastasis and recurrence being the principal factors contributing to the dire outcomes associated with this cancer[39]. The poor prognosis of GC is primarily attributed to disease recurrence and metastasis. Emerging evidence indicates that CSCs play pivotal roles in tumor initiation, recurrence, metastasis, and resistance to radiotherapy and chemotherapy across multiple cancer types[40,41]. Therefore, gaining deeper insights into the mechanisms by which CSCs sustain their stemness and developing effective strategies to target GCSCs may significantly improve the prognosis of GC. Recent studies have underscored the pivotal role of CSCs in driving tumor invasion, metastasis, recurrence, and resistance to chemotherapy[42,43]. In the context of GC, Takaishi et al[44] were the first to report the existence and pathological significance of GCSCs, a finding that has since been corroborated by multiple studies identifying GCSCs or stem cell-like populations in GC cell lines and tumor tissues[45-48].

In this study, we initiated our investigation with a comprehensive bioinformatics approach, incorporating immune infiltration assessments and functional enrichment analyses to uncover factors underlying the poor prognosis associated with the GCSS model. Comparative analyses demonstrated that the GCSS model outperformed traditional clinical staging parameters (T, M, N) in prognostic prediction. Building on this, we constructed a nomogram integrating clinical variables and the GCSS score, further enhancing predictive precision.

The relevance of NPC2, a core gene within the GCSS model, was validated using both single-cell and bulk RNA transcriptomic data. In line with our findings, Yao et al[15] analyzed TCGA datasets. They reported that NPC2 expression significantly correlates with key clinicopathological features in GC patients, suggesting its utility as a prognostic biomarker. NPC2 may influence GC initiation and progression via multiple biological and signaling pathways. Enrichment analyses across GO, KEGG, and Pathway databases highlighted that NPC2 is potentially involved in ribosomal activity, T cell activation regulation, and adhesion molecule-associated pathways. Targeting CSC-related pathways may offer a novel therapeutic avenue for GC management[49]. Moreover, CAFs, the predominant stromal component in the tumor microenvironment, possess distinct tumor-promoting properties[50-52].

Like other stromal components, CAFs surrounding primary tumor sites secrete a range of signaling molecules and factors that modulate tumor phenotypes, influencing multiple stages of tumor progression[53,54]. In our study, the GCSS model—developed through multivariate Cox regression, leveraged KM analysis to identify key prognostic determinants in GC. Importantly, patients in the high GCSS score group demonstrated significantly elevated levels of CAF infiltration. To further explore underlying biological mechanisms, we assessed correlations between GCSS scores and enriched signaling pathways using the GO, HALLMARK, KEGG, and REACTOME databases. The GCSS score is strongly associated with numerous pathways linked to CD8+ T cells and Wnt signaling. A variety of signaling pathways have been implicated in sustaining the self-renewal properties of GCSCs, such as PI3K/Akt[55], STAT3[56], TGF-β[57,58], Wnt[59,60], Hedgehog[61-63], and Notch signaling[64]. Furthermore, the enrichment of immune-related pathways may be influenced by the anti-tumor immune responses initiated by tumor-associated antigens and immune modulation driven by malignant epithelial cells during GC progression. These observations are consistent with the current understanding of how anti-tumor immunity can be undermined during the cancer immune cycle[65-68].

Bioinformatics analysis revealed a strong correlation between NPC2 and GC; however, these computational findings require further validation through both in vitro and in vivo experiments. The IHC analysis of GC tissues confirmed that NPC2 expression was significantly elevated in tumor tissues compared to adjacent non-cancerous counterparts, regardless of whether patient survival exceeded or fell below five years. Beyond GC, altered NPC2 expression has been reported in several other malignancies, including lung, breast, and liver cancers, where it appears to influence cellular proliferation. Furthermore, NPC2 may modulate apoptotic signaling pathways, which are implicated in forming papillary structures in cancers such as papillary thyroid carcinoma and pulmonary adenocarcinoma[9-12]. Further functional assays, including spheroid formation, migration, invasion, and CCK8 proliferation tests, were performed using GC cell lines AGS and HGC-27. These experiments revealed that silencing NPC2 significantly reduced GC cells' stemness migratory, invasive, and proliferative capacities.

Moreover, NPC2 contributes to resistance against several standard radiotherapy and chemotherapy agents in GC. These results underscore the pivotal role of NPC2 in GC progression, affecting multiple malignant phenotypes. As such, NPC2 emerges as a promising biomarker for prognosis and a potential therapeutic target, offering new avenues for improving the diagnosis and treatment of GC.

CONCLUSION

In conclusion, this study demonstrates that the GCSS model provides improved prognostic value compared to traditional clinicopathological staging in GC, offering improved direction for clinical decision-making and patient management. Future research should prioritize identifying key molecular targets across multiple signaling pathways to more effectively disrupt the self-renewal mechanisms of GCSCs, aiming for their complete eradication. NPC2 has emerged as a promising prognostic marker among these targets due to its strong association with GC progression. Although its precise role in GC remains fully elucidated, the integration of bioinformatics analysis and experimental validation in this study highlights NPC2 as a potentially novel and clinically significant biomarker in GC.