Identification of candidate hub genes and signatures in gastric adenocarcinoma: the role of CDRT15P1 in tumor progression

Gastric cancer (GC) is among the most prevalent malignant tumors globally, ranking fifth in newly diagnosed cancers and fourth in cancer-related mortality^[1]. Most patients with GC are not identified until the disease has progressed to an advanced stage, which is associated with unfavorable clinical outcomes^[2]. Although studies have shown a steady decline in the incidence of GC in recent years^[3], research output related to GC is expected to remain substantial in the foreseeable future. Previous studies have suggested that the development of GC involves multiple biological mechanisms, including DNA damage repair and chronic inflammation of the gastric epithelium^[4−8]. GC represents the final outcome of long-term pathological processes involving the accumulation of multiple genetic and epigenetic alterations. Although numerous genes have been reported to be involved in GC-related progression and clinical outcomes^[9,10], effective molecular targets for predicting GC remain to be identified.

Weighted correlation network analysis (WGCNA) is a widely used bioinformatics approach for exploring gene expression patterns across different samples; it groups genes with comparable expression profiles into modules, enables the analysis of relationships between modules and specific traits or phenotypes, and facilitates the identification of intramodular hub genes^[11,12]. Least absolute shrinkage and selection operator (LASSO) regression is a regularization-based estimation method that extends the least squares approach, aiming to achieve an optimal balance between a model's simplicity and predictive accuracy. In this study, hub genes associated with stomach adenocarcinoma (STAD) were identified through integrated bioinformatics analyses combined with machine learning algorithms and LASSO regression.

CMT1A replication region transcript 15 (CDRT15) was initially described by Inoue et al. in 2001 during investigations of the 1.4-Mb genomic region associated with Charcot–Marie–Tooth disease Type 1A (CMT1A) duplication and hereditary neuropathy with liability to pressure palsy (HNPP) deletion^[13]. According to current genome annotations (GRCh38/hg38), CDRT15 is located on chromosome 17 (17:14,235,673–14,236,862) and spans 1,190 nucleotides. The encoded product is referred to as the CDRT15 protein. Previous studies have reported elevated expression of CDRT15 in cholangiocarcinoma (CCA)^[14], where higher expression levels were associated with unfavorable clinical outcomes and alterations in immune cell infiltration patterns. These findings suggest that CDRT15 may have potential clinical relevance as a biomarker in CCA. However, its involvement in GC has not been clearly characterized. Subsequent research has indicated that CDRT15-related transcripts may participate in various biological processes across different physiological and pathological contexts, particularly in malignant diseases. In light of these observations, we examined the expression pattern of CDRT15 pseudogene 1(CDRT15P1) in GC using The Cancer Genome Atlas (TCGA) datasets and evaluated its relationship with clinicopathological features. In addition, in vitro functional assays were conducted to assess the potential association between the expression of CDRT15P1 and phenotypic behaviors of GC cells.

Transcriptomic profiles of STAD and the corresponding adjacent non-tumor tissues were retrieved from TCGA repository. Only samples with complete expression information were included, comprising 32 normal tissue specimens and 375 GC tissue specimens. The adjacent normal tissues were derived from the gastric mucosa of the same patients and were collected from noncancerous regions more than 2 cm away from the tumor margin. To minimize technical variability across samples, transcript abundance values were transformed using log₂ (transcripts per million [TPM] + 1) prior to downstream analyses^[15].

Construction of the gene co-expression network

To investigate coordinated transcriptional patterns associated with STAD, a weighted gene co-expression framework was established using normalized RNA sequencing data. Genes exhibiting greater variability across samples were preferentially selected to enhance the network's robustness. Specifically, genes ranked within the upper 50% of variance and presenting a median absolute deviation (MAD) greater than 0.01 were retained. A range of candidate soft-thresholding parameters was examined, and the value that achieved an approximate scale-free topology while preserving adequate network connectivity was selected for adjacency calculation. Pairwise gene correlations were converted into connection strengths and further transformed into topological overlap measures to characterize shared neighborhood similarity. Hierarchical clustering based on topological overlap dissimilarity was subsequently performed to group genes with similar expression patterns. Modules were identified through dynamic branch segmentation, and modules displaying high similarity were consolidated according to a predefined merging threshold.

Identification of candidate hub genes

To reduce redundancy among genes and improve the model's interpretability, a penalized regression approach was adopted for variable selection. LASSO regression was implemented using the "glmnet" package in R. Model tuning was conducted through 10-fold cross-validation, and the penalty coefficient corresponding to the minimum cross-validation error was selected. Genes retained at this optimal parameter were considered to be potential hub candidates.

Differential expression analysis

Differential gene expression between tumor and adjacent normal tissues was evaluated using R statistical software. After normalization, genes meeting the criteria of an absolute log₂ fold change greater than 1.5 and an adjusted p-value below 0.05 were regarded as significantly dysregulated.

Construction of regulatory networks of transcription factors and miRNAs

We identified the pre- and post-transcriptional gene regulatory factors of microRNA (miRNA) and the transcriptional factors (TFs) by analyzing the interaction networks among the shared hub genes' encoded proteins and miRNAs and TFs. The interaction network of TFs and shared differentially expressed hub genes (hub-DEGs) was constructed using the JASPAR^[16] TF database, and the TarBase V8.0^[17] and miRTarBase^[18] miRNAs databases were utilized to construct the miRNA–hub-DEG interaction network.

Immune infiltration estimation

To characterize the immune landscape of STAD samples, enrichment scores representing immune cell populations were computed at the individual sample level using a single-sample gene set enrichment framework. Associations between hub genes' expression and immune infiltration scores were assessed by correlation analysis^[19].

Cell culture conditions

The human gastric carcinoma cell lines AGS and MKN-45 were obtained from the Cell Resource Center of the Chinese Academy of Medical Sciences (Beijing, China). AGS cells were maintained in Dulbecco's modified Eagle medium (DMEM), whereas MKN-45 cells were cultured in Roswell Park Memorial Institute (RPMI)-1640 medium. Both media were supplemented with 10% fetal bovine serum and standard concentrations of penicillin (100 U/mL) and streptomycin (100 μg/mL). Cells were incubated at 37 °C under 5% CO₂ in a humidified atmosphere. The culture medium was replaced every 48 h to maintain optimal growth conditions. All in vitro experiments were independently repeated three times, with three technical replicates per experiment.

Gene silencing procedure

To assess the biological role of CDRT15P1, transient knockdown experiments were performed using small interfering RNA (siRNA). Transfections were carried out according to the manufacturer's protocol provided with the siRNA transfection kit. Following transfection, cells were incubated for 48 h before the subsequent analyses. Knockdown efficiency was confirmed by quantitative real-time polymerase chain reaction (qRT-PCR). Cells transfected with a nontargeting siRNA sequence served as the negative controls. The siRNA sequences and qRT-PCR primer sequences are listed in Supplementary Table S1.

Quantitative real-time polymerase chain reaction

Total RNA was isolated using TRIzol reagent following standard extraction procedures. One microgram of total RNA was reverse-transcribed into complementary DNA using a commercial synthesis kit. Real-time polymerase chain reaction (PCR) amplification was conducted using SYBR Green chemistry. Relative gene expression levels were calculated using the 2^−ΔΔCt method, with appropriate internal controls applied for normalization.

Protein expression analysis

Whole-cell protein lysates were prepared using Radio Immunoprecipitation Assay (RIPA) buffer supplemented with protease inhibitors. Protein concentrations were quantified using a bicinchoninic acid (BCA) assay. Equal amounts of protein were separated via sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene difluoride (PVDF) membranes. After blocking, the membranes were incubated overnight at 4 °C with primary antibodies targeting Phosphatidylinositol 3-kinase (PI3K), phosphorylated PI3K, AKT serine/threonine kinase (also known as Protein Kinase B, PKB), phosphorylated AKT, and β-actin. Subsequently, membranes were exposed to horseradish peroxidase-conjugated secondary antibodies. Immunoreactive bands were detected using enhanced chemiluminescence reagents.

Assessment of cell proliferation

Cell viability was evaluated using the cell counting kit-8 (CCK-8) assay. Transfected cells were seeded into 96-well plates at a density of 3 × 10³ cells per well. At 24, 48, 72, and 96 h post-seeding, CCK-8 reagent was added according to the manufacturer's instructions. Optical density values were measured at 450 nm to estimate proliferative capacity.

Migration analysis

To examine migratory behavior, wound-healing assays were performed. Cells were cultured in six-well plates until reaching near-complete confluence. A sterile pipette tip was used to generate a linear scratch across the monolayer. After washing away detached cells, cultures were maintained in serum-free medium. Images were captured at predefined time points to monitor the wound closure dynamics.

Invasion assay

Cell invasive potential was evaluated using transwell chambers coated with Matrigel. Transfected cells were suspended in serum-reduced medium and placed in the upper chamber. The lower chamber contained the medium supplemented with 10% fetal bovine serum as a chemoattractant. Following incubation, noninvading cells were removed from the upper surface of the membrane. Cells that traversed the membrane were fixed, stained with crystal violet, and quantified microscopically.

Apoptosis analysis

Apoptotic populations were analyzed using Annexin V-fluorescein isothiocyanate (FITC)/propidium iodide dual staining. Cells were harvested 48 h after transfection, washed with phosphate-buffered saline, and resuspended in binding buffer. Staining reagents were applied sequentially according to the manufacturer's instructions. Flow cytometry was used to distinguish the early and late apoptotic fractions.

Statistical considerations

All analyses were conducted using R software (version 3.6.1) and GraphPad Prism (version 9.5.1). Comparisons between two groups were performed using Student's t-test. For multiple-group comparisons, one-way analysis of variance (ANOVA) was applied. The Benjamini–Hochberg method was applied for multiple testing correction in the enrichment analysis and all correlation analyses to control the false discovery rate (FDR). Statistical significance was defined as a two-sided p-value less than 0.05.

Differential expression analysis between tumor and adjacent normal gastric tissues yielded 20,525 genes meeting the predefined screening criteria (adjusted p < 0.05 and |log₂ fold change| > 1.5). Among these, 10,501 genes showed increased expression, whereas 10,024 exhibited decreased expression in STAD samples. The expression distribution profile is summarized in Fig. 1a.

Figure 1. 

Differential expression and co-expression network analysis in STAD. (a) Distribution of differentially expressed genes (DEGs) between tumor and normal gastric tissues visualized by a volcano plot. (b) Determination of the soft-thresholding power (β = 6) based on the scale-free topology fit and mean connectivity analysis. (c) Correlation matrix illustrating the relationships between gene modules and STAD clinical traits. (d) Topological overlap matrix heatmap showing gene connectivity patterns. (e) Hierarchical clustering dendrogram with module color assignment. (f)–(i) Correlation analysis between module membership and gene significance for the magenta, green, cyan, and purple modules.

To further investigate coordinated transcriptional behavior, a co-expression network was constructed using the filtered dataset. When multiple probes mapped to the same gene symbol, the probe with the highest average expression value was selected to represent that gene. Genes demonstrating relatively higher variability across samples (top 50% based on MAD and MAD > 0.01) were retained for network modeling.

Evaluation of the candidate soft-thresholding parameters indicated that β = 6 provided a satisfactory approximation of scale-free topology (R² = 0.85; Fig. 1b). This parameter was therefore used to compute pairwise gene connectivity, which was subsequently converted into topological overlap measures to better capture shared interaction patterns. Modules were defined using hierarchical clustering, with a minimum size of 30 genes. Modules exhibiting high similarity (correlation > 0.75) were merged to reduce redundancy.

Correlation analysis between module eigengenes and clinical traits identified several modules strongly associated with STAD status (Fig. 1c). The magenta (r = −0.86, p = 2E−47), green (r = −0.74, p = 6E−29), cyan (r = −0.69, p = 1E−23), and purple (r = −0.69, p = 1E−23) modules demonstrated the most pronounced relationships and were therefore prioritized for downstream investigation.

The clustering dendrogram and corresponding module color annotations are displayed in Fig. 1e. The heatmap in Fig. 1d illustrates gene–gene connectivity patterns within the network. Furthermore, associations between gene significance and module membership were examined within the selected modules, revealing strong correlations in the magenta, green, and cyan modules, and a moderate association in the purple module (Fig. 1f–i).

Screening of hub genes

To further refine candidate genes derived from the overlapping differentially expressed gene (DEG) set, LASSO regression analysis was performed to minimize model overfitting and identify the most informative variables (Fig. 2a, b). The optimal penalty parameter (λ) was determined by 10-fold cross-validation. At λ = 0.01, five genes—CDRT15P1, LYSMD2, STRC, KCTD19, and ETV2—were retained for subsequent analyses.

Figure 2. 

Identification and prognostic evaluation of hub genes. (a), (b) Construction of the LASSO regression model and selection of the optimal λ value using cross-validation. (c), (d) Comparison of gene expression levels between tumor and normal tissues. (e) Kaplan–Meier survival analysis stratified by the expression levels of selected hub genes in STAD patients.

Univariate Cox proportional hazards regression was then conducted to evaluate their prognostic significance. Four genes were identified as risk-associated factors, whereas one gene exhibited a protective effect (Fig. 2c, d).

For survival analysis, patients were stratified into high- and low-expression groups according to optimal cut-off values. The corresponding thresholds were 0.002 for CDRT15P1, 0.001 for LYSMD2, 0.001 for STRC, 0.002 for KCTD19, and 0.001 for ETV2.

Kaplan–Meier survival curves demonstrated that elevated expression of CDRT15P1 (p = 0.002) and KCTD19 (p = 0.029) was significantly associated with poorer overall survival, whereas higher ETV2 expression (p = 0.009) was associated with favorable survival outcomes (Fig. 2e). No statistically significant differences were observed for the remaining genes.

Functional enrichment analysis of common DEGs in normal and STAD tissues

To characterize the biological functions potentially associated with the overlapping DEGs identified in STAD, pathway enrichment analysis was conducted using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. The enrichment profile indicated that metabolic pathways constituted the predominant functional category among these genes. In particular, the pathway related to metabolism of cofactors and vitamins exhibited the highest level of statistical significance (Fig. 3). In addition to metabolic processes, genes involved in proteasome-mediated protein folding, sorting, and degradation were also enriched (Fig. 3). These findings suggest that alterations in metabolic regulation and protein processing may be linked to the molecular characteristics of STAD, although further experimental validation is required to clarify their precise biological implications.

Figure 3. 

Functional enrichment analysis of overlapping DEGs. Pathway enrichment was performed using publicly available KEGG annotations. Pathway diagrams were generated by the authors on the basis of KEGG database information without direct reproduction of original KEGG images.

Correlation analysis between hub gene expression and tumor-related biological pathways

We next examined the associations between hub genes' expression levels and tumor-related biological features. Elevated CDRT15P1 expression showed positive correlations with proliferation (R = 0.16, p = 0.00084), stemness (R = 0.20, p = 6.3e−05), and differentiation (R = 0.18, p = 2e−04) scores (Fig. 4a). These correlations suggest a potential relationship between CDRT15P1's expression and phenotypic characteristics associated with tumor progression, although the strength of these associations was modest. A comparable pattern was observed for KCTD19, which displayed positive correlations with angiogenesis (R = 0.15, p = 0.0016) and apoptosis (R = 0.13, p = 0.0075) (Fig. 4c). In contrast, ETV2's expression demonstrated a different correlation profile. ETV2 was positively associated with quiescence (R = 0.19, p = 7.6e−05) but showed negative correlations with angiogenesis (R = −0.28, p = 1.1e−08), differentiation (R = −0.27, p = 1.9e−08), and epithelial–mesenchymal transition (EMT) (R = −0.24, p = 1e−06) (Fig. 4b). In addition, ETV2's expression was negatively correlated with invasion (R = −0.18, p = 0.00031) and metastasis (R = −0.22, p = 5.6e−06), suggesting that higher ETV2 levels may be associated with comparatively less aggressive phenotypic features. Consistent with these observations, the expression of ETV2 was lower in STAD tissues compared with normal tissues (Fig. 2e), and reduced ETV2 expression was associated with improved overall survival in STAD patients (Fig. 2e). By contrast, the opposite trends were observed for CDRT15P1 and KCTD19. Furthermore, both CDRT15P1 and KCTD19 showed positive correlations with DNA damage repair signatures. These associations raise the possibility that their expression levels may be linked to cellular responses to radiotherapy or chemotherapy; however, this hypothesis requires additional experimental validation.

Figure 4. 

Correlation between hub genes' expression and tumor-related biological features. (a) CDRT15P1; (b) ETV2; (c) KCTD19; (d) LYSMD2; (e) STRC. Correlation coefficients and corresponding p-values are indicated in each panel.

Association between hub genes and immune cell infiltration

Given the established role of immune cell dynamics in tumor development, we further evaluated the relationship between hub genes' expression and estimated immune cell infiltration in STAD samples. Analysis indicated that the expression of CDRT15P1 exhibited modest correlations with variations in immune cells' abundance, particularly involving T lymphocytes, B lymphocytes, macrophages, and mast cells (Fig. 5a).

Figure 5. 

Association between hub genes' expression and immune cell infiltration in STAD. (a) CDRT15P1; (b) ETV2; (c) KCTD19; (d) LYSMD2; (e) STRC. Immune infiltration scores were estimated using single sample Gene Set Enrichment Analysis (ssGSEA).

A similar tendency was observed for KCTD19, where expression levels were also related to patterns of T cell and B cell infiltration (Fig. 5c).

In contrast, the correlation pattern associated with ETV2 differed, as its expression showed the opposite tendency with respect to T cell and B cell infiltration (Fig. 5b).

Overall, higher CDRT15P1 expression appeared to correspond to slightly reduced immune cell infiltration across several immune subsets. However, the magnitude of these correlations was relatively limited, and the findings primarily reflect general trends rather than strong quantitative effects.

Construction of transcriptional regulatory networks

To further explore potential upstream regulatory relationships of the identified hub genes, TF and microRNA interaction analyses were performed.

On the basis of predictive databases, regulatory interactions were mapped and organized into a gene-centered network, which was subsequently displayed using Cytoscape (Fig. 6). The resulting structure included a four-node gene–miRNA interaction framework. Within the TF–hub gene interaction analysis, CDRT15P1, LYSMD2, and KCTD19 were associated with a relatively larger number of predicted TFs compared with the remaining genes. Among these TFs, SCRT1 and ZFX appeared more frequently across multiple hub genes. The gene–miRNA interaction mapping indicated that CDRT15P1, KCTD19, and LYSMD2 were connected to several predicted miRNAs. Specifically, hsa-miR-26b-5p, hsa-let-7g-5p, hsa-miR-148b-3p, hsa-let-7i-5p, hsa-miR-26a-5p, and hsa-miR-34a-5p were predicted to interact with both CDRT15P1 and KCTD19. In addition, hsa-miR-182-5p, hsa-miR-20b-5p, and hsa-miR-129-2-3p were linked to LYSMD2 and KCTD19.

Figure 6. 

Predicted transcription factor and miRNA interaction network of STAD-related hub genes. Green nodes represent transcription factors and blue nodes represent miRNAs.

These analyses provide a preliminary overview of potential regulatory connections, although experimental validation would be required to confirm these predicted interactions.

CDRT15P1 affects the proliferation, migration, invasion, and apoptosis of GC cells in vitro

Since CDRT15P1 is located within the CMT1A duplication region on chromosome 17p12, and its gene name is derived from this region, we performed an intersection analysis between the DEGs in STAD and the CMT1A-related gene set (Fig. 7a) to evaluate the overall differential expression characteristics of this gene family in GC. The results showed that CDRT15P1 is one of the most significant DEGs within the CMT1A-related gene set. To further investigate the potential biological role of CDRT15P1 in GC cells, loss-of-function experiments were performed using siRNA-mediated knockdown in AGS and MKN-45 cell lines. qRT-PCR confirmed that CDRT15P1's expression was effectively reduced following siRNA transfection in both cell lines (Fig. 7b). Cell viability was subsequently evaluated using CCK-8 assays at multiple time points. Compared with cells transfected with negative control siRNA, CDRT15P1-silenced cells exhibited decreased proliferation rates in both AGS and MKN-45 cells (Fig. 7c). Consistent with these findings, colony formation capacity was also reduced after CDRT15P1 knockdown (Fig. 7d). To assess cell motility, wound-healing assays were conducted. Cells with reduced CDRT15P1 expression demonstrated slower wound closure relative to the control cells in both cell lines (Fig. 7e). In parallel, transwell assays indicated that silencing CDRT15P1 was associated with a reduction in invasive cell numbers (Fig. 7f). Apoptotic cell populations were analyzed by flow cytometry using Annexin V-FITC/propidium iodide staining. An increase in both early and late apoptotic fractions was observed in CDRT15P1-knockdown cells compared with the controls (Fig. 7g). To explore potential downstream signaling alterations, the PI3K/AKT pathway was examined by Western blotting. Reduced phosphorylation levels of PI3K and AKT were detected in CDRT15P1-silenced cells, whereas total protein levels showed no marked changes (Fig. 7h). Collectively, these results suggest that decreased CDRT15P1 expression is associated with reduced proliferative and migratory capacities and increased apoptotic tendency in GC cells. However, the precise molecular mechanisms underlying these observations require further investigation.

Figure 7. 

Functional assessment of CDRT15P1 in GC cells. (a) Venn diagram showing the overlap between DEGs and CMT1A-related gene sets. (b) Relative mRNA expression of CDRT15P1 following siRNA transfection. (c) Cell proliferation assessed by the CCK-8 assay. (d) Colony formation assay. (e) Wound-healing assay evaluating migratory capacity. (f) Transwell invasion assay. (g) Apoptosis analysis by Annexin V-FITC/propidium iodide staining. (h) Western blot analysis of PI3K/AKT pathway proteins following CDRT15P1 knockdown. Data are presented as the mean ± SD. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.

In the past few decades, the incidence rate of GC has declined steadily worldwide^[20], mainly because of the successful prevention and treatment of Helicobacter pylori infection. The clinical therapy of GC includes surgery, adjuvant chemotherapy, neoadjuvant chemotherapy, and immunotherapy^[21−25]. Given that conventional treatment strategies often fail, many cancer-related molecules have been characterized with the aim of developing novel anticancer therapies, including targeted antibodies and vaccines^[26]. With the rise of immunotherapy, four main strategies are currently considered: Immune checkpoint inhibitors (ICI), tumor vaccines, adoptive immunotherapy (ACT), and nonspecific immunomodulators^[27]. Despite many advances in the therapeutic field, the prognosis and quality of life of GC patients have hardly improved in the past decade^[28].

The 17p12 chromosomal region contains a 1.4-Mb segment flanked by two approximately 24-kb low-copy repeat sequences (CMT1A–REPs). Structural rearrangements within this region have been reported to give rise to inherited peripheral neuropathies, including CMT1A and HNPP. CDRT15P1 is one of the genes in this genomic region. In the study of Tianyang Yu et al., the results showed that the elevated expression of CDRT15P1 was associated with a higher incidence rate and poorer prognosis of cholangiocarcinoma. At present, there are no reports on whether CDRT15P1 has a role in the occurrence and development of GC. We conducted biological information analysis on the TCGA database between normal and tumor tissues of the stomach, and the results showed that CDRT15P1 expression was increased in tumor tissues. We analyzed the expression level of CDRT15P1 and its relationship with GC-related cellular function. The results showed that increased expression of CDRT15P1 was associated with elevated levels of proliferation-related factors, stemness factors, and differentiation factors in GC cells, which suggested a higher risk of metastasis and invasion.

CDRT15P1 expression showed a negative correlation with the infiltration levels of multiple immune cell types in STAD, including T cells, B cells, macrophages, and mast cells. Concurrently, in vitro functional assays revealed that CDRT15P1 knockdown significantly reduced PI3K/AKT phosphorylation levels. These two independent observations suggest a potential link between CDRT15P1 expression, PI3K/AKT pathway activity, and the tumor immune microenvironment. Notably, emerging evidence has linked PI3K/AKT signaling to immune infiltration patterns across various solid tumors. In hepatocellular carcinoma, FAT10 upregulates PD-L1 expression by activating the PI3K/AKT/mammalian targed of rapamycin (mTOR) pathway, thereby suppressing CD8+ T cell infiltration and promoting immune evasion^[29]. Similarly, in the GC microenvironment, interleukin (IL)-4 induces metabolic reprogramming and M2 polarization of macrophages by upregulating the PI3K/AKT/mTOR signaling pathway, leading to CD8+ T cell dysfunction and resistance to anti-PD-1 therapy^[30]. Furthermore, SERPINE1 in gastric cancer activates the PI3K/AKT pathway through autocrine mechanisms to promote anoikis resistance, while also promoting M2 macrophage polarization and suppressing CD8+ T cell infiltration and function via paracrine binding to LRP1^[31]. In breast cancer, inhibition of the PI3K/AKT pathway has been shown to downregulate PD-L1 expression in tumor cells, promote macrophage polarization toward the M1 phenotype, enhance CD8+ T cells' function, and increase T and B cells' tumor infiltration^[32]. The findings of these studies, in conjunction with our observations, support the reasonable hypothesis that CDRT15P1 may contribute to shaping an immunosuppressive microenvironment in GC, including inhibiting T cell infiltration and promoting M2 polarization of macrophages, at least in part through activation of the PI3K/AKT pathway. This could partially explain the association between high CDRT15P1 expression and poor clinical outcomes. Nevertheless, it is important to acknowledge that our study currently provides only indirect, correlational evidence. Whether the CDRT15P1/PI3K/AKT axis directly influences immune cell infiltration and through which molecular mediators this occurs (e.g., PD-L1, chemokines) are questions that require direct validation using immune cell–tumor cell co-culture models or in vivo xenograft models.

Among the five candidate hub genes, CDRT15P1 was the only one showing significantly upregulated expression in STAD, a strong association with poor prognosis, and positive correlations with multiple malignant phenotypes (proliferation, stemness, differentiation). Therefore, it was prioritized for further functional investigation. We conducted functional experiments on GC cells in vitro, including examinations of proliferation, migration, invasion, and apoptosis. The results showed that increased CDRT15P1 expression levels promoted the proliferation, migration, and invasion of GC cells, while inhibiting apoptosis. Collectively, these findings suggest that CDRT15P1 may contribute to GC's progression. Through KEGG enrichment analysis, we found that elevated CDRT15P1 expression may be associated with alterations in cofactor metabolism pathways. Numerous studies have reported the involvement of cofactor metabolism in GC's progression^[33]. Therefore, it is necessary to find the cofactors involved in the regulation of GC by CDRT15P1.

However, there are limitations to this study. The small number of GC patients in TCGA and other databases may result in significant bias. Moreover, this study did not collect tissues from specific clinical cases for experimentation. The biological function and mechanism of CDRT15P1 in GC require further experimental verification. In addition, more robust validation needs to be done to confirm the role of CDRT15P1 in the prognosis of GC through in vivo or in vitro studies. A major limitation of this study is the lack of in vivo validation. Though our bioinformatics and in vitro findings suggest a potential role for CDRT15P1 in gastric adenocarcinoma, animal model experiments are needed to further confirm its functional significance. We therefore plan to conduct rigorous in vivo studies in the future. Whether the PI3K/AKT pathway mediates all observed phenotypes requires further validation through pathway rescue experiments.

In summary, our results demonstrated that CDRT15P1 may facilitate GC's progression and tumor malignancy. Therefore, CDRT15P1 is not only a promising prognostic biomarker but also a potential molecular therapeutic target for GC.