We acquired and processed data from the TCGA database (https://portal.gdc.cancer.gov), specifically focusing on the TCGA-COADREAD project related to colorectal cancer. RNA sequencing (RNA-seq) data were analyzed using the STAR pipeline to extract FPKM-format data for stage II colorectal cancer. This data was then matched with the corresponding clinical information.

This was a retrospective study using archived pathological tissue samples from 118 patients who had undergone radical resection for colorectal cancer at Shaanxi Provincial People’s Hospital between January 2017 and December 2018. All samples were residual tissues obtained after the completion of clinical diagnosis. The inclusion criteria were as follows: 1) postoperative histopathological diagnosis confirming colorectal cancer; 2) no preoperative radiotherapy, chemotherapy, or other systemic tumor-targeted therapy; and 3) no history of other malignant tumors. The study protocol was approved by the Ethics Committee of Shaanxi Provincial People’s Hospital (Approval No. 2023-R057), which granted a waiver for individual patient consent due to the retrospective nature of the research and the use of fully anonymized patient data.

We conducted a comprehensive collection and documentation of all patients’ clinical and pathological characteristics. This included various parameters, such as gender, age at diagnosis, tumor location, and clinical staging. For TNM staging, we evaluated the size of the primary tumor and depth of invasion (T), the status of regional lymph node metastasis (N), and the presence of distant metastasis (M). We also classified histological grading as well-differentiated, moderately differentiated, or poorly differentiated.

We used the PV8000 immunohistochemistry kit and the DAB staining kit (Zhongshan Jinqiao Biotechnology) for immunohistochemical detection. The primary antibody was Rabbit anti-Claudin 18 polyclonal antibody (21126-1-AP, Proteintech), and experiments followed the manufacturer’s protocol. The specific procedural steps included the following: Firstly, tissue sections were subjected to deparaffinization and rehydration. Next, we blocked endogenous peroxidase activity using 3% hydrogen peroxide (H₂O₂). Following this, we performed microwave-mediated antigen retrieval with a citrate buffer. After antigen retrieval, tissue sections were incubated overnight at 4°C with the Claudin 18 antibody, followed by DAB staining to visualize the immunoreactivity. Subsequently, sections were counterstained with hematoxylin and dehydrated. Finally, the specimens were mounted using neutral resin. Immunohistochemical reactions resulting in brown-yellow granules were considered positive. We quantified the relative expression levels of the proteins based on staining intensity and the percentage of positively stained area. The scoring criteria for staining intensity were as follows: no staining (0 points), weak staining (1 point), moderate staining (2 points), and strong staining (3 points). The scoring criteria for the stained area were: ≤10% staining (0 points), 11%-20% (1 point), 21%-50% (2 points), 51%-70% (3 points), and >70% (4 points). The final score was the product of the two scores. A total score of ≥6 indicated high expression, while a score of <6 indicated low expression. All IHC procedures adhered to strict quality control standards, and the results were independently evaluated by two experienced pathologists to ensure accuracy and reproducibility of the findings.

We used the DESeq2 package in R to perform differential expression analysis on the raw counts matrix. This analysis compared stage II CRC with other stages and examined tumors with and without PNI. We followed standard protocols for the analysis, including data normalization using the Variance Stabilizing Transformations (VST) method from DESeq2. We applied stringent filtering criteria, requiring a log2 fold change (FC)| > 1 and a p < 0.05 to identify differentially expressed genes (DEGs) between the two groups. Subsequent analyses examined both unique and shared DEGs across the different groups. We used the ggplot2 and VennDiagram packages to create clear graphical representations of the results.

The Surv cutpoint function was utilized to calculate Kaplan-Meier curves for overall survival (OS), with differences assessed using the log-rank test. Genes associated with overall survival in stage II colorectal cancer were identified, and those with p>0.05 were excluded to obtain a set of prognostic-related genes. Univariate and multivariate Cox regression analyses were performed to determine the independent prognostic factors. A nomogram was constructed based on these prognostic factors to provide a comprehensive assessment of patient survival probabilities. The effectiveness of the nomogram model was evaluated through receiver operating characteristic (ROC) curves, area under the curve (AUC) values, calibration curves, and decision curve analysis (DCA).

To test the proportional hazards assumption and fit survival regression, the survival package was employed, and the results were visualized using the survminer and ggplot2 packages. During the univariate analysis, the optimal cut-off values were determined using the surv_cutpoint function from the survminer package. The R package “rms” was utilized to generate Cox regression models, the nomogram, and calibration plots. Survival probabilities were predicted using the R package “pec.” ROC curves and DCA curves were plotted using the R packages “timeROC” and “ggDCA,” respectively. Statistical significance was defined as a p-value of less than 0.05.

Statistical analysis of the collected data was performed using software such as Microsoft Excel, GraphPad Prism 7.0, and SPSS 23.0. This study used the chi-square test to evaluate the significance of differences between the two groups based on contingency tables. Additionally, we used the χ² test and Fisher’s exact test to analyze correlations among clinical pathological parameters. All tests were performed using a two-sided approach with a significance level set at α=0.05.

We screened differential genes from the TCGA database using thresholds of log2 fold change (FC) | > 1 and p < 0.05. This process identified 370 DEGs related to stage II CRC and 161 DEGs associated with PNI. The results were visualized with a volcano plot, which clearly displayed both upregulated and downregulated genes. Next, we conducted a Venn diagram analysis to identify unique and shared genes between the two groups, revealing 33 commonly expressed genes that were significantly altered ( Figure 1 ).

We downloaded and organized RNA sequencing data and corresponding clinical information from the TCGA-COADREAD project. The clinical data for 208 patients with stage II CRC were complete and used in the survival analysis. We applied a threshold of p < 0.05 to identify five genes significantly associated with overall survival: CLDN18, FTCD, MSH4, MUC6, and PLG. Among the identified genes, high expression of PLG was associated with longer overall survival. In contrast, higher expression of the other genes (CLDN18, FTCD, MSH4, and MUC6) correlated with shorter overall survival ( Figure 2 ).

Among the five prognostic genes identified from the Kaplan-Meier analysis, we first excluded PLG, as our study aimed to build a risk model based on adverse prognostic factors (oncogenes), whereas high expression of PLG was associated with better survival and thus considered a protective factor. For the remaining four oncogenes (CLDN18, FTCD, MSH4, and MUC6), we proceeded with univariate Cox regression analysis. To select the most robust candidates for the subsequent multivariate analysis and nomogram construction, we applied a stricter significance threshold for inclusion (P < 0.01). As shown in the univariate analysis ( Figure 3A ), only CLDN18 (P = 0.003) and FTCD (P = 0.004) met this criterion, while MSH4 (P = 0.019) and MUC6 (P = 0.041) did not. Therefore, only CLDN18 and FTCD were carried forward into the final model for multivariate analysis, which confirmed them as independent adverse prognostic factors ( Figure 3B ).

Given that predicting the prognosis of CRC patients using a single biomarker or clinical characteristic is often difficult, we developed a comprehensive prognostic nomogram model that integrates CLDN18 and FTCD with clinical factors like age and gender. This nomogram is designed to assess the probability of OS in CRC patients. In the nomogram, the length of each variable’s line indicates its contribution to the prognosis. The analysis showed that CLDN18 and FTCD significantly impacted the OS of patients with stage II CRC compared to other parameters ( Figure 3C ).

We evaluated the prognostic model’s performance using calibration curves, AUC values, and DCA. First, we plotted calibration curves for the nomogram. These curves showed that the model’s predicted probabilities of overall survival at 1, 3, and 5 years closely matched the actual outcomes for stage II colorectal cancer patients, indicating high accuracy ( Figures 4A–C ). Next, we generated ROC curves to assess the predictive sensitivity and specificity of the nomogram model. The AUC values for overall survival were 0.726 at 1 year, 0.679 at 3 years, and 0.786 at 5 years ( Figure 4D ). Moreover, the AUC values for the nomogram model in predicting 1-year, 3-year, and 5-year survival were higher than those for age, CLDN18, and FTCD. This highlights the superior predictive performance of the nomogram model ( Figures 4E, F ).

We performed an immunohistochemical analysis to evaluate Claudin 18 protein expression levels in CRC tissues. The results showed that Claudin 18 protein primarily localized to the cell membrane and cytoplasm ( Figure 5 ). In colorectal cancer tissues with PNI, the positive expression rate of Claudin 18 was 56.8% (50/88), while it was 33.3% (10/30) in tissues without PNI. Statistical analysis indicated a significant difference in Claudin 18 protein expression levels between the two groups (P < 0.05, ( Table 1 ). We further analyzed the correlation between Claudin 18 protein expression in colorectal cancer tissues and clinical pathological parameters. The results revealed that Claudin 18 protein expression levels were significantly associated with patient age and lymphatic invasion status (P < 0.05, Table 2 ).