We first downloaded the RNA-seq expression data for EOCRC (<50 years old) from The Cancer Genome Atlas (TCGA) database (76 tumor and 7 paired non-tumor samples). We next downloaded the RNA-seq expression data for LOCRC (≥50 years old), which included 566 tumor and 44 paired non-tumor samples. We screened for DEGs using the “edgeR” package in R, with the screening criteria being |log2FC| > 2.5, P < 0.01. The clinical information incorporated in the analysis were N and M stages. After removing samples with missing clinical information and duplicate samples, 66 EOCRC tumor samples remained for subsequent analysis, while 478 LOCRC samples remained.

WGCNA reveals the functional links among genes by constructing gene co-expression networks to group genes with comparable expression profiles into common modules. We used the “WGCNA” package in R for analysis, which identified modules and genes associated with EOCRC and LOCRC metastasis.

We used the “clusterProfiler” package in R for functional analysis to confirm the potential roles of the identified genes. Disease Ontology (DO) and Gene Ontology (GO) analyses were conducted for genes associated with EOCRC metastasis. Additionally, we performed the Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis to study the enrichment of gene signaling pathways. P < 0.05 was considered to be statistically significant.

The genes associated with EOCRC metastasis were analyzed with the “glmnet” and “kernlab” packages in R to identify feature variables by LASSO regression and SVM-RFE machine learning algorithms, respectively. We also used the online tool Venny 2.1.0 to conduct an intersecting analysis of the results to select feature genes only related to EOCRC metastasis.

The ROC curve was constructed with the “pROC” package in R to evaluate the potential of feature genes as biomarkers and their association with disease occurrence and progression.

The “GSVA” package in R was utilized to evaluate the association between immune cell infiltration and the feature genes related to EOCRC metastasis. We obtained distribution patterns of the feature genes related to EOCRC metastasis for 28 types of immune cells using the ssGSEA algorithm. The immune cells related to the feature genes of EOCRC metastasis were screened using P < 0.05.

The “corrplot” package in R was used for immune checkpoint analysis to understand the immune regulatory mechanisms and gene interactions.

LOVO cells (EallBio, CA) were maintained in F-12K complete medium and HCT116 cells (EallBio, CA) were maintained in McCoy’s 5a complete medium. Cells were cultivated in a humidified incubator with an atmosphere containing 5% CO₂ at 37°C. Once the cells reached 40% confluence, they were transfected with LipofectamineTM 2000 (Invitrogen, United States) and siRNA targeting PMP2 with a negative control siRNA to reduce PMP2 expression. After transfection for 6 h, fresh culture medium was added to replace the old medium, the cells were cultured for 24 h, and the cells were collected for invasion and migration experiments.

TRIzol (Invitrogen, United States) was used to extract RNA from clinical tissues collected from the General Hospital of the Northern Theater Command. The concentration of RNA was determined using the NanoDrop 2000 (Thermo Fisher Scientific, United States). RNA was reverse transcribed into complementary DNA using the Prime-Script RT reagent kit (TaKaRa, Japan). Subsequently, qRT-PCR was done using Ex-Taq (TaKaRa, Japan) and expression levels were determined with the 2^−ΔΔCT method. The primer sequences are given below: PMP2 forward, 5′-GACGATTACATGAAAGCTCTGGG-3′; PMP2 reverse, 5′-TGCACTTGATTCAGTGATCCTCT-3′; GAPDH forward: 5′-GGAGCGAGATCCCTCCAAAAT-3′; GAPDH reverse: 5′-GGCTGTTGTCATACTTCTCATGG-3′.

To determine the migration of LOVO and HCT116 cells, 1 × 10⁵ cells were suspended in medium containing 2% FBS and plated in the upper chamber, while the lower chamber medium contained 20% FBS. After incubation for 24 h, cells were fixed with methanol, stained with crystal violet, and counted under an inverted microscope. In the invasion assay, 1 × 10⁵ cells were suspended in medium with 2% FBS and plated in the upper chamber coated with Matrigel (Corning, NY, United States), while the lower chamber medium contained 20% FBS. After incubation for 24 h, cells were fixed with methanol, stained with crystal violet, and counted under an inverted microscope. Each experiment was repeated at least three times.

The EOCRC and adjacent normal tissue slices were subjected to dewaxing, hydration, and rinsed with pure water. Antigen retrieval was done by placing the sections into a citrate buffer solution and heating them in a microwave for 10 min before cooling them to room temperature. The endogenous peroxidases were inactivated by treating sections with a 3% hydrogen peroxide solution for 10 min. The sections were then incubated with anti-PMP2 primary antibodies (1:100; Proteintech Group, China) overnight at 4°C. Subsequently, the sections were incubated with secondary antibodies at 37°C for 30 min. After using the DAB solution for color development, the slices were stained with hematoxylin, dehydrated and sealed with resin. Finally, the sections were observed and photographed under a microscope (Nikon, Japan).

In this study, 49 EOCRC specimens and adjacent normal tissue were collected from patients who underwent surgery at the General Hospital of the Northern Theater Command. None of the patients received neoadjuvant therapy before surgery. We obtained informed consent from each patient who participated in the study. The Ethics Committee of the Northern Theater General Hospital approved the current work (Y (2022) 063).

Statistical analysis was performed using R software (version 4.1.3). The two-tailed Student’s t-test was done to evaluate differences between groups, with statistical significance at P < 0.05.

The methodologies employed in this study have been shown as a flowchart in Figure 1.

The EOCRC and LOCRC gene expression matrices and clinical data were downloaded from the TCGA database. We selected DEGs using the criteria |log2FC| > 2.5, P < 0.01. We identified 1,411 DEGs in EOCRC, and 3,434 DEGs in LOCRC (Figures 2A,B). The association between these DEGs and clinical traits was assessed by individual WGCNA for EOCRC and LOCRC, with the soft thresholds set at 4 and 7, respectively, ensuring the biological significance of network construction (Figures 2C,D). The M and N were selected as clinical traits in the analysis to identify gene modules related to metastasis. We found a positive correlation between the cyan module and M in EOCRC, with a correlation coefficient of 0.25. In LOCRC, the blue and gray modules were positively correlated with N, and a positive correlation emerged between the yellow module and M, with correlation coefficients of 0.12, 0.12, and 0.11, respectively. The modules significantly related to M and N based on these findings were selected for further analysis (Figures 2E,F).

A Venn diagram was employed to identify 44 genes specifically related to EOCRC metastasis (Figure 3A). We performed functional analysis on these genes to delve deeper into their biological roles and mechanisms of action. The DO analysis results suggested these genes are primarily linked to Lewy body dementia and mood disorders (Figure 3B). Biological processes (BP) were mainly related to synaptic organization and regulation of membrane potential, Cellular components (CC) were associated with neuronal cell bodies and postsynaptic membranes and Molecular Functions (MF) were mostly linked with ion channel activity (Figure 3C). KEGG analysis highlighted that these genes predominantly participated in neuroactive ligand-receptor interaction and calcium signaling pathways (Figure 3D).

We utilized two methods to highlight feature genes related to EOCRC metastasis. While the SVM-RFE technique identified 19 genes, the LASSO regression analysis filtered out five genes. (Figures 4A,B). We then performed an intersecting analysis of the results from both approaches using a Venn diagram and ultimately selected four feature genes: LINC02268, AC092652.1, GRIK1, and PMP2 (Figure 4C). We plotted their ROC curves and calculated the area under the curve (AUC) values to be 0.995, 0.995, 1.000, and 0.997, respectively. These high AUC values indicate the high accuracy of these genes as disease feature genes to be effective diagnostic biomarkers potentially (Figure 4D).

We further analyzed the differences in immune infiltration between EOCRC and control samples. Significant differences were seen in activated B cells, activated CD8 T cells, effector memory CD4 T cells, effector memory CD8 T cells, gamma delta T cells, immature B cells, memory B cells, T follicular helper cells, Type 1 T helper cells, CD56bright natural killer cells, eosinophils, macrophages, mast cells, monocytes, natural killer cells, natural killer T cells, and neutrophils (Figure 5A). These results indicated significant alterations in composition and functional status of immune cells in the EOCRC tumor microenvironment, suggesting the possible role of immune cells in EOCRC occurrence and progression. We utilized the ssGSEA algorithm to assess the relationship between the four feature genes and immune cells in EOCRC. we found that the four feature genes were closely related to activated B cells, eosinophils, immature B cells, mast cells, and neutrophils, suggesting their putative role in disease progression by immune microenvironment modulation (Figure 5B). Additionally, immune checkpoint analysis documented a positive correlation between the four feature genes and immune-related genes (Figure 5C).

We selected PMP2 for subsequent experiments as there are no current reports on the association between PMP2 and CRC. We obtained 49 paired EOCRC and neighboring normal tissues from the Northern Theater Command General Hospital. The qRT-PCR demonstrated that PMP2 levels were markedly lower in EOCRC tissues compared to normal samples (Figure 6A). Subsequently, we used IHC to analyze PMP2 expression levels in three pairs of EOCRC and adjacent normal samples and found that PMP2 is underexpressed in EOCRC (Figure 6B). Subsequently, three different siRNAs were used to knock down PMP2 in HCT116 cells to verify the siRNA knockdown efficiency. qRT-PCR results showed significantly decreased PMP2 expression in HCT116 cells transfected with siPMP2#1/2/3 compared to the siNC group (Figure 6C). Additionally, the lowered PMP2 expression significantly enhanced the migratory and invasive abilities of HCT116 and LOVO cells than the siNC group (Figures 6D,E). These results indicate that PMP2 knockdown significantly enhanced the migratory and invasive abilities of HCT116 and LOVO cells. To summarize, our findings demonstrate that PMP2 plays a role in inhibiting tumor metastasis during EOCRC tumorigenesis.