COMP promotes the progression of colorectal cancer by regulating epithelial mesenchymal transition

Abstract

Background

Epithelial-mesenchymal transition (EMT) plays a crucial role in the progression and metastasis of colorectal cancer (CRC). This study investigates the molecular mechanisms of EMT and its prognostic biomarkers in CRC.

Methods

Multi-omics bioinformatics analyses were conducted using CRC transcriptomic datasets from GEO and TCGA. EMT-related differentially expressed genes (EMT-DEGs) were identified and subjected to pathway enrichment and machine learning-based prognostic modeling. COMP was selected as a hub gene for further validation. Single-cell RNA sequencing (scRNA-seq) data were also analyzed to determine the cell-type-specific expression pattern of COMP and EMT-DEGs. Clinical CRC tissue samples were analyzed via RT-qPCR, Western blot, and histology. Functional assays in HT-29 cells assessed the effects of COMP knockdown on EMT markers, proliferation, apoptosis, invasion, and migration.

Results

EMT was significantly enriched in CRC, with 36 EMT-DEGs identified. These DEGs were enriched in pathways such as ECM-receptor interaction, focal adhesion, and the PI3K-Akt signaling pathway. Among the constructed machine learning models, the random survival forest (RSF) model demonstrated the strongest ability to predict CRC prognosis. This model stratified CRC patients into high-risk and low-risk groups, with poorer prognosis observed in the high-risk group. Cox regression forest plots and Kaplan-Meier survival analysis identified COMP as a top EMT-related prognostic gene enriched in pathways including ECM-receptor interaction and PI3K-Akt signaling. High COMP expression correlated with poor patient prognosis and EMT marker dysregulation in metastatic CRC tissues. In vitro, COMP knockdown significantly reduced mesenchymal markers, restored E-cadherin, and inhibited proliferation and invasion of CRC cells.

Conclusion

EMT plays a vital role in CRC progression and metastasis, with COMP identified as a key prognostic biomarker and potential therapeutic target. This study provides new insights into the molecular mechanisms and intervention strategies for CRC metastasis.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-025-15000-3.

Introduction

Colorectal cancer (CRC) is one of the most common malignant tumors worldwide, with increasing incidence and mortality rates each year [1, 2]. Its clinical characteristics mainly include locally invasive growth and distant metastasis [3]. Despite advances in surgical techniques, chemotherapy, and radiotherapy, the prognosis for patients with advanced or metastatic CRC remains poor, largely due to aggressive invasion and distant metastasis [4]. The 5-year survival rate for advanced CRC patients remains below 15%, with distant metastasis being the primary cause of poor outcomes [5]. Therefore, elucidating the molecular mechanisms underlying CRC progression and metastasis and identifying novel diagnostic, prognostic, and therapeutic targets are essential to improving patient outcomes.

Epithelial-mesenchymal transition (EMT) is a dynamic biological process whereby epithelial cells acquire mesenchymal properties, leading to enhanced motility, invasiveness, and resistance to apoptosis [6, 7]. EMT has been widely recognized as a driving force in cancer metastasis, including in CRC, where its activation is associated with poor prognosis and increased metastatic potential [8]. During EMT, tumor cells gain stronger migratory and invasive abilities, enabling them to breach the basement membrane and form metastatic foci [9]. While EMT is originally described as a developmental process in normal epithelial cells, increasing evidence has shown that carcinoma cells can reactivate this program to facilitate metastasis. This phenomenon, referred to as “pathological EMT,” is a hallmark of epithelial tumor progression and is triggered by a complex tumor microenvironment (TME) [8]. Studies have shown that EMT activation is particularly crucial in CRC metastasis and is closely associated with poor prognosis [10]. Thus, uncovering the specific mechanisms of EMT in CRC is essential for understanding tumor invasiveness and developing effective targeted therapies.

The extracellular matrix (ECM) plays a critical role in regulating EMT through biochemical and biomechanical signaling [11]. Cartilage Oligomeric Matrix Protein (COMP), a non-collagenous ECM glycoprotein initially identified in cartilage, has been increasingly implicated in tumorigenesis [12, 13]. Recent studies have shown that COMP is overexpressed in several solid tumors and may contribute to tumor progression by remodeling the ECM and promoting EMT [14]. However, the precise role of COMP in CRC, particularly in the context of EMT and metastasis, remains poorly understood.

In the era of high-throughput sequencing and systems biology, integrative bioinformatics approaches have enabled the identification of EMT-related genes with prognostic and therapeutic significance [15]. Bioinformatics analyses based on public datasets, combined with machine learning methods, offer opportunities to identify key genes and molecular signaling pathways closely related to tumor biology [16].

In this study, we aimed to systematically investigate the potential role of EMT in CRC prognosis by integrating bioinformatics analyses, machine learning model construction, and experimental validation. Through this research, we seek to provide a scientific basis for personalized treatment strategies for CRC patients, explore the potential application of COMP as a therapeutic target, and propose new strategies to mitigate CRC metastasis and improve patient survival.

Materials and methods

Data collection and processing

The GSE39582 dataset [17] contains gene expression profiles of 566 CRC tissues and 19 intestinal tissues obtained using an array-based approach on the GPL570 platform. Background correction and quantile normalization were performed using the limma package. The GSE131418 dataset [18] includes gene expression profiles of 333 primary CRC tumors and 184 CRC metastases, obtained using an array-based approach on the GPL15048 platform. Background correction and quantile normalization were also performed using the Limma package. The TCGA dataset comprises gene expression profiles of 287 CRC tissues and 41 intestinal tissues. Gene expression values were normalized using the DESeq2 package. The single-cell RNA sequencing (scRNA-seq) dataset GSE200997 contains 16 CRC tissues with corresponding 8 adjacent normal tissue samples. The clinicopathological characteristics of the patients in these datasets were provided in the Table S1-S3.

Differential expression and enrichment analysis

Gene Set Enrichment Analysis (GSEA) was performed to rank the gene expression data across all samples and assess the enrichment of EMT-related gene sets in CRC. The EMT gene set was derived from the MSigDB database (Hallmark EMT Pathway). Single-sample gene set variation analysis (GSVA) was employed to calculate EMT enrichment scores for each sample. These scores were used to evaluate differences in EMT activity between CRC and control samples. Differential expression analysis was conducted using the limma package to compare CRC samples with normal controls, as well as metastatic samples with primary tumors. Differentially expressed genes (DEGs) were identified with thresholds set at |log2(FC)| > 1 and P < 0.05. The EMT-related DEGs were further subjected to KEGG pathway enrichment analysis using the ClusterProfiler package. A significance threshold of P < 0.05 was applied to identify enriched pathways.

Screening of prognostic genes

To construct prognostic models for CRC patients based on EMT-related differentially expressed genes (EMT-DEGs), 10 Machine learning algorithms were utilized to build 35 machine learning models [19]. The GSE39582 dataset was used as the training dataset, while the TCGA dataset served as the validation dataset. The area under the curve (AUC) was calculated for all models to evaluate their predictive performance.

We applied the Random Survival Forest (RSF) algorithm using the randomForestSRC package in R. The RSF model was trained on the GSE39582 cohort and validated using the TCGA-COAD cohort. The model was constructed with the following parameters: Number of trees (ntree): 1000; Node size (nodesize): 15; Splitting rule: log-rank splitting. Feature importance was assessed using permutation-based variable importance (VIMP). To assess model stability and generalizability, we performed 10-fold cross-validation within the GSE39582 training set. The mean concordance index (C-index) and time-dependent AUC values were calculated across folds to evaluate predictive accuracy. The final RSF risk scores were calculated for each patient and used to stratify CRC patients into high-risk and low-risk groups.

Prognostic differences between the groups were assessed using Kaplan-Meier survival analysis and time-dependent ROC curves. Additionally, the prognostic and diagnostic value of the model genes was evaluated using the Cox proportional hazards model and Kaplan-Meier survival analysis.

ScRNA-seq analysis

The data of GSE200997 were processed using the Seurat (v4.3.0) package in R. After standard quality control (removal of cells with < 200 or > 6000 genes, and > 20% mitochondrial gene expression), gene expression was normalized using the LogNormalize method. Highly variable genes were identified using the FindVariableFeatures function. Principal component analysis (PCA) was then performed on the scaled data, and the top 20 PCs were used for clustering. Cells were clustered using FindNeighbors and FindClusters (resolution = 0.5), followed by Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction. Cell types were annotated based on canonical markers and verified using the SingleR package. The expression levels of COMP and EMT-DEGs were visualized across annotated cell clusters using feature plots.

Sample collection

Tumor tissue samples were collected from 10 primary CRC patients and 10 metastatic CRC patients at the Fifth Affiliated Hospital of Xinjiang Medical University. The study protocol was reviewed and approved by the Ethics Review Committee of the Fifth Affiliated Hospital of Xinjiang Medical University (Approval No. XYDWFYLSk-2024-149). All patients provided written informed consent, and the sample collection process strictly adhered to the principles of the Declaration of Helsinki. Inclusion Criteria: Diagnosis of CRC confirmed by histopathology, no history of other malignant tumors, no prior targeted therapy or systemic chemotherapy before sample collection, tumor tissue samples with ≥ 70% tumor cell content, confirmed by two independent pathologists, for metastatic CRC patients, histopathological confirmation of CRC with distant metastasis. Exclusion Criteria: Patients with severe infectious or autoimmune diseases, patients with severe dysfunction of the liver, kidneys, heart, or lungs, or those whose overall health conditions were unsuitable for surgery, samples severely contaminated by necrosis, fibrosis, or other non-tumor components. The clinicopathological characteristics of the patients were provided in the Table S4.

Hematoxylin and Eosin (HE) staining

The collected tissue samples were immediately fixed in 4% neutral formalin solution for 48 h. After fixation, the tissues were dehydrated sequentially in a graded ethanol series (70%, 80%, 95%, and 100%), cleared with xylene, and embedded in Liquid paraffin. The paraffin-embedded tissues were sectioned into slices 4 μm thick using a microtome and mounted on slides. The sections were deparaffinized and rehydrated before being stained with Hematoxylin. Subsequently, the sections were stained with 0.5%−1% eosin solution for 1 min. The staining results were observed under a light microscope to evaluate the staining quality.

Cell culture and transfection

To determine the basal expression levels of COMP, five commonly used human colorectal cancer cell lines (HCT116, SW480, LOVO, DLD-1, and HT-29; Procell, China). Cells were cultured in DMEM medium (Thermo) supplemented with 10% fetal bovine serum (FBS) (Gibco) and 1% penicillin-streptomycin (Procell). The cells were Maintained at 37 °C in a humidified incubator with 5% CO₂. Total RNA and protein were extracted for RT-qPCR and Western blotting, respectively. To induce EMT, HT-29 cells were seeded in 6-well plates at a density of 3 × 10⁵ cells per well and cultured to approximately 60–70% confluence. The culture medium was then replaced with serum-free DMEM containing recombinant human TGF-β1 (MCE, USA) at a final concentration of 10 ng/mL. Cells were incubated with TGF-β1 for 48 h.

The siRNA sequences targeting the COMP gene were designed and synthesized by GenePharma (Shanghai, China), along with negative control siRNA (si-NC). COMP overexpression plasmid (pcDNA3.1-COMP) or empty vector (pcDNA3.1-NC) were purchased from GenePharma. Transfection of siRNA or plasmids was performed using Lipofectamine™ 3000 reagent (Thermo), and the cells were cultured for an additional 24 h post-transfection. The efficiency of siRNA transfection was validated using RT-qPCR and Western blot analysis.

Cell proliferation and apoptosis

The proliferation of HT-29 cells was assessed using the Cell Counting Kit-8 (CCK-8) reagent (Beyotime, China). After transfection, HT-29 cells were digested and resuspended. Cells were seeded into a 96-well plate at a density of 5 × 10³ cells/well and incubated for 12 h. At 24 h and 48 h post-transfection, 10 µL of CCK-8 reagent was added to each well, and the optical density (OD) was measured at 450 nm using a microplate reader.

Cell apoptosis was analyzed using the TUNEL (Terminal deoxynucleotidyl transferase dUTP Nick End Labeling) assay kit (Beyotime). Transfected HT-29 cells were seeded into 6-well plates at a density of 1 × 10⁵ cells/well. At 48 h post-transfection, cells were fixed with 4% paraformaldehyde for 15 min. After fixation, cells were permeabilized with 0.1% Triton X-100 in PBS for 5 min at room temperature. 50 µL of TUNEL reaction mixture was added to each well following the kit instructions, and cells were incubated in the dark for 1 h. DAPI (1 µg/mL) was used to stain the nuclei at room temperature in the dark for 5 min. Images of stained cells were captured using a fluorescence microscope.

Transwell

Transwell chambers were used to assess the migration and invasion abilities of HT-29 cells under conditions without Matrigel (migration assay) and with Matrigel coating (invasion assay). For each Transwell insert, 200 µL of cell suspension (5 × 10⁴ cells/mL in serum-free medium) was added to the upper chamber. The lower chamber was filled with 600 µL of complete DMEM medium containing 10% FBS as a chemoattractant. The Transwell chambers were incubated at 37 °C with 5% CO₂ for 24 h (migration assay) or 48 h (invasion assay). After incubation, cells that did not migrate or invade were gently removed from the upper surface of the membrane using a cotton swab. The cells adhering to the lower surface of the membrane were fixed with 4% paraformaldehyde for 20 min. The fixed cells on the lower membrane were observed under an inverted microscope, and images were captured from five randomly selected fields per insert. The number of migrated or invaded cells was counted and recorded.

In vivo mouse model of CRC liver metastasis

The shRNA sequences targeting the COMP gene were designed and synthesized by GenePharma (Shanghai, China), along with negative control shRNA (sh-NC). Transfection of shRNA or plasmids was performed using Lipofectamine™ 3000 reagent (Thermo), and the cells were cultured for an additional 24 h post-transfection. Female C57BL/6 mice (6 weeks old) were purchased from the Laboratory Animal Center of Xinjiang Medical University. Animal procedures were approved by the Institutional Animal Care and Use Committee of Xinjiang Medical University (No. IACUC20240615-1). All mice were randomly assigned to five groups (n = 6/group): CRC, CRC + sh-NC, CRC + sh-COMP, CRC + OE-NC, CRC + OE-COMP. Mice were anaesthetised with 1–2% isoflurane (RWD Life Science, Shenzhen, China) delivered in oxygen at a flow rate of 1 L/min. HT-29 cells stably expressing luciferase and transfected with the corresponding constructs were injected into the colon under anesthesia. Liver metastases were monitored weekly using in vivo bioluminescence imaging (IVIS Lumina) after intraperitoneal injection of D-luciferin (150 mg/kg). After 8 weeks, mice were deeply anesthetized with 3–4% isoflurane until complete loss of reflexes was observed, then cervical dislocation was performed to ensure death, and liver tissues were collected.

Reverse transcription quantitative PCR (RT-qPCR)

Total RNA was extracted from tissue and cell samples using TRIzol reagent (Invitrogen). RNA concentration and purity were assessed using a NanoDrop spectrophotometer, ensuring a 260/280 ratio between 1.8 and 2.0. Complementary DNA (cDNA) was synthesized from the extracted RNA using a cDNA synthesis kit (Takara). Quantitative reverse transcription PCR (RT-qPCR) was performed using SYBR Green reagent (Invitrogen). The sequences of the specific primers are listed in Table S5. The expression of the target genes was normalized to β-actin as an internal control. Relative gene expression levels were calculated using the 2^−ΔΔCt method.

Western blotting

Total protein was extracted from tissues and cells using RIPA lysis buffer containing protease and phosphatase inhibitors. Samples were lysed on ice for 30 min. Protein concentrations were determined using a BCA Protein Assay Kit (Beyotime). A total of 30 µg of protein was separated by gel electrophoresis and transferred onto a PVDF membrane. The membrane was blocked with 5% skim milk at room temperature for 2 h. The blocked membrane was incubated with primary antibodies (ABclonal, China) at 4 °C overnight. After washing, the membrane was incubated with HRP-conjugated secondary antibodies at room temperature for 1 h. The membrane was exposed to a chemiluminescence reagent mixture for 2 min, and the signals were detected using an imaging system. β-actin was used as the internal control protein, and the grayscale intensity of the protein bands was analyzed using ImageJ software (NIH, USA).

Dual-luciferase reporter assay

The promoter regions of human EpCAM were cloned into the pGL3-Basic luciferase reporter vector (Promega). HT-29 cells were co-transfected with pGL3-promoter constructs and either pcDNA3.1-COMP or pcDNA3.1-NC, along with the Renilla luciferase control vector (pRL-TK) using Lipofectamine™ 3000. Luciferase activity was measured using the Dual-Luciferase Reporter Assay System (Promega) 48 h post-transfection. Relative luciferase activity was normalized to Renilla.

Statistical analysis

All data were analyzed using R software or GraphPad Prism. Kaplan-Meier survival analysis was performed, and the Log-rank test was used to evaluate the significance of survival differences. The mean (Mean) and standard deviation (SD) were calculated for each dataset. Statistical comparisons were conducted using one-way analysis of variance (ANOVA) or Student’s t-test, as appropriate. A significance level of P < 0.05 was considered statistically significant.

Results

Enrichment of EMT in CRC

Analysis of gene expression data from the GSE39582 and TCGA datasets revealed that GSEA results indicated significant enrichment of EMT in CRC, suggesting that EMT might play a critical role in the initiation and progression of CRC (Fig. 1A, B). The GSVA scores showed that EMT activity was significantly higher in CRC samples compared to the control group (Fig. 1C, D). Differential expression analysis identified 2409 DEGs between CRC and control samples in the GSE39582 dataset (Fig. 2A), 11,446 DEGs between metastatic and primary CRC samples in the GSE131418 dataset (Fig. 2B), and 4810 DEGs between CRC and control samples in the TCGA dataset (Fig. 2C). Intersection analysis with EMT-related genes identified 36 EMT-DEGs (Fig. 2D). KEGG pathway enrichment analysis revealed that these EMT-DEGs were significantly enriched in several key signaling pathways, including ECM-receptor interaction, Focal adhesion, and the PI3K-Akt signaling pathway (Fig. 2E).

Fig. 1.

Enrichment analysis of EMT in CRC. GSEA results indicate significant enrichment of epithelial-mesenchymal transition (EMT)-related gene sets in CRC samples compared to control samples in GSE39582 (A) and TCGA (B). NES, normalized enrichment score; NP, normalized P value. C GSVA analysis of EMT activity scores across normal and CRC in GSE39582 dataset. n = 585. D. GSVA analysis of EMT activity scores across normal and CRC in TCGA. n = 328. **P < 0.01, ***P < 0.001

Fig. 2.

Differentially expressed genes and pathway enrichment analysis. A Volcano plot showing the distribution of differentially expressed genes between colorectal cancer and control samples in GSE39582. B Volcano plot showing the distribution of differentially expressed genes between metastatic CRC and primary CRC in GSE131418. C Volcano plot showing the distribution of differentially expressed genes between colorectal cancer and control samples in TCGA. D The intersection between differentially expressed genes and EMT-related genes. E KEGG pathway enrichment analysis of the 36 EMT-DEGs

Machine learning screening for EMT-related prognostic genes

The 35 constructed machine learning models were used to train and evaluate gene expression data from CRC patients to identify key genes associated with prognosis. Among all models, the RSF model demonstrated the strongest prognostic predictive capability, with the highest AUC value (Figure S1A). The RSF model included 22 EMT-DEGs (Figure S1B).

Based on the RSF model scores, CRC patients were stratified into high-risk and low-risk groups in the GSE39582 (Fig. 3A) and TCGA datasets (Fig. 3B). Kaplan-Meier survival curve analysis showed that patients in the high-risk group had significantly worse survival outcomes compared to those in the low-risk group (P < 0.001, Fig. 3C, E). The time-dependent ROC curves demonstrated the prognostic diagnostic performance of the RSF model scores, with AUC > 0.9 in GSE39582 (Fig. 3D) and AUC > 0.6 in TCGA (Fig. 3F).

Fig. 3.

Prognostic stratification of CRC patients using the RSF model. A Distribution of RSF risk scores, survival status, and gene expression in GSE39582. B Distribution of RSF risk scores, survival status, and gene expression in TCGA. C Kaplan-Meier survival curves illustrating overall survival differences between high-risk and low-risk groups stratified by RSF model scores in GSE39582. D Time dependent ROC curve of RSF risk scores in GSE39582. E Kaplan-Meier survival curves illustrating overall survival differences between high-risk and low-risk groups stratified by RSF model scores in TCGA. F Time dependent ROC curve of RSF risk scores in TCGA

The Cox regression forest plot indicated that the high expression of PCOLCE2 and COMP was significantly associated with poor prognosis (Fig. 4A). Notably, COMP was significantly involved in the ECM-receptor interaction, Focal adhesion, and PI3K-Akt signaling pathways, highlighting its critical role in tumor progression. Kaplan-Meier survival curve analysis further confirmed that patients with high COMP expression had worse prognoses (Fig. 4B).

Fig. 4.

Cox regression forest plot and Kaplan-Meier survival analysis. Cox regression forest plot showing the hazard ratios (HR) and 95% confidence intervals (CI) for EMT-related genes in RSF model of GSE39582 (A) and TCGA (B). Kaplan-Meier survival curves for patients stratified by COMP expression levels in GSE39582 (C) and TCGA (D)

Single-cell transcriptomic analysis reveals tumor-specific expression of COMP and EMT-DEGs

To further dissect the cell-type-specific expression of COMP and EMT-related genes in CRC, we performed scRNA-seq analysis and identified 23 transcriptionally distinct clusters (Figure S2A). A total of 10 major cell types were identified following dimensionality reduction and unsupervised clustering, including epithelial cells, fibroblasts, macrophages, CD4 + and CD8 + T cells, B cells, Tregs, plasma cells, endothelial cells, mast cells, and natural killer T (NKT) cells (Figure S2B, S2C). Feature plots (Figure S2D) showed that COMP expression was specifically localized to fibroblasts and epithelial with clusters enriched for EMT-related genes, including COL1A1, MMP2, TIMP1, and EPCAM.

Detection in primary and metastatic CRC patients

H&E staining showed that the stromal regions in metastatic CRC tissues were looser, with significantly higher inflammatory cell infiltration compared to primary CRC tissues (Fig. 5A). RT-qPCR results demonstrated that in metastatic CRC tissues, the mRNA levels of COMP, Collagen 1, Laminin, Hyaluronan, EpCAM, and N-cadherin were significantly higher, while E-cadherin expression was significantly lower compared to primary CRC tissues (Fig. 5B). Western blot results showed that the protein levels of COMP, Collagen 1, Laminin, Hyaluronan, EpCAM, N-cadherin, MMP2, MMP9, and TIMP1 were significantly upregulated, while E-cadherin expression was significantly downregulated in metastatic CRC tissues compared to primary CRC tissues (Fig. 5C, Figure S3).

Fig. 5.

Analysis of pathology and molecular in primary and metastatic CRC. A Representative hematoxylin and eosin (HE) staining of primary and metastatic CRC. Bar = 100 μm. B RT-qPCR analysis of key EMT-related genes in primary and metastatic CRC tissues. n = 20. Data are shown as mean ± SD. ***P < 0.001. C Western blot analysis of key EMT-related genes in primary and metastatic CRC tissues. Original blots are presented in Figure S3. n = 20. Data are shown as mean ± SD. ***P < 0.001

Effect of COMP on the function of HT-29 cells

RT-qPCR and Western blotting demonstrated that HT-29 cells showed the highest COMP mRNA expression, followed by SW480 and HCT116 (Figure S4).

After transfecting HT-29 cells with si-COMP, the knockdown efficiency was confirmed using RT-qPCR and Western blot. Transfection with siRNA 1574 significantly reduced COMP mRNA and protein expression levels (Figure S5A). RT-qPCR results indicated that after COMP knockdown, the mRNA levels of Collagen 1, Laminin, Hyaluronan, EpCAM, and N-cadherin were significantly decreased, while E-cadherin levels were significantly increased in HT-29 cells (Fig. 6A). Western blot results showed that in COMP-knockdown HT-29 cells, the protein levels of Collagen 1, Laminin, Hyaluronan, EpCAM, N-cadherin, MMP2, MMP9, and TIMP1 were significantly reduced, whereas E-cadherin expression was significantly elevated (Fig. 6B, Figure S5B).

Fig. 6.

Effect of gene expression through regulating COMP in HT-29 cells. A RT-qPCR analysis of EMT-related genes after COMP knockdown. n = 9. Data are shown as mean ± SD. **P < 0.01, ***P < 0.001. B Western blot analysis of EMT-related genes after COMP knockdown. Original blots are presented in Figure S5B. n = 9. Data are shown as mean ± SD. ***P < 0.001. C RT-qPCR analysis of EMT-related genes after COMP overexpression. n = 9. Data are shown as mean ± SD. **P < 0.01, ***P < 0.001. D Western blot analysis of EMT-related genes after COMP overexpression. Original blots are presented in Figure S6. n = 9. Data are shown as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001

To further validate the role of COMP in promoting EMT, we performed COMP overexpression experiments in HT-29 cells. RT-qPCR showed a significant increase in COMP, Collagen 1, Laminin, Hyaluronan, EpCAM, and N-cadherin and a significant reduced in E-cadherin mRNA expression following transfection with pcDNA3.1-COMP (Fig. 6C). Western blot also confirmed a significant increase in COMP, Collagen 1, Laminin, Hyaluronan, EpCAM, N-cadherin, MMP2, MMP9, and TIMP1 protein expression, and a significant reduced in E-cadherin following transfection with pcDNA3.1-COMP (Fig. 6D, Figure S6).

Additionally, we observed that E-cadherin expression was markedly downregulated, whereas both N-cadherin and EpCAM levels were significantly upregulated following TGF-β1 stimulation (Figure S7). Interestingly, COMP knockdown further increased E-cadherin expression, while reduced N-cadherin and EpCAM levels, compared to TGF-β1-treated controls. To investigate whether COMP affects EMT-related gene expression at the transcriptional level, we conducted dual-luciferase reporter assays targeting the promoter regions of EpCAM. The results showed that COMP overexpression significantly enhanced luciferase activity (Figure S8), suggesting transcriptional regulation.

These findings were further supported by migration and invasion assays. Transwell experiments demonstrated that COMP knockdown significantly reduced both migration and invasion capabilities, while COMP overexpressing significantly increased both migration and invasion capabilities of HT-29 cells (Fig. 7A and B), suggesting that suppression of COMP impairs the EMT-mediated motility phenotype. Additionally, CCK-8 assay results showed that COMP knockdown significantly inhibited the proliferation of HT-29 cells, with the proliferation rates at 24 and 48 h post-si-COMP transfection, and COMP overexpressing HT-29 cells exhibited increased proliferation (Fig. 7C). TUNEL assay demonstrated that COMP knockdown significantly increased the apoptosis rate of HT-29 cells, and COMP overexpressing reduced apoptosis rates (Fig. 7D).

Fig. 7.

Effects of COMP on HT-29 cell invasion, migration, proliferation and apoptosis. A Representative images and quantification of invaded cells from the Transwell invasion assay. Bar = 100 μm. B Representative images and quantification of migrated cells from the Transwell migration assay. Bar = 100 μm. C Cell proliferation detected by CCK-8. D Cell apoptosis detected by Tunel. Bar = 100 μm. n = 9. Data are shown as mean ± SD. **P < 0.01, ***P < 0.001

COMP promotes CRC liver metastasis in vivo

To confirm the in vitro findings, we established an in vivo mouse model of CRC liver metastasis. Bioluminescence imaging revealed that mice in the COMP-OE group developed significantly higher hepatic tumor burden compared to controls, while the sh-COMP group exhibited markedly reduced liver metastatic signals (Fig. 8A). Molecular analysis of metastatic liver lesions by RT-qPCR demonstrated that the COMP-OE group showed significantly increased mRNA levels of Collagen I, Laminin, Hyaluronan, EpCAM, and N-cadherin, with a concomitant reduction in E-cadherin expression, indicating activation of the EMT program. In contrast, the sh-COMP group showed the opposite expression trend, consistent with EMT suppression (Fig. 8B).

Fig. 8.

In vivo validation of COMP-mediated promotion of CRC liver metastasis and EMT activation. A Representative bioluminescence imaging of Liver metastases in 8 weeks after intrasplenic injection of HT-29 cells stably overexpressing COMP (COMP-OE), COMP knockdown (sh-COMP), or control vector. B RT-qPCR analysis of ECM and EMT-related genes in liver metastatic lesions. C Western blot analysis of key ECM/EMT markers in liver metastases. Original blots are presented in Figure S9. Data are presented as mean ± SD, n = 6 mice per group. **P < 0.01, ***P < 0.001

Western blot analysis further confirmed these findings at the protein level, the COMP-OE group exhibited elevated expression of Collagen I, Laminin, Hyaluronan, EpCAM, N-cadherin, MMP2, MMP9, and TIMP1, while E-cadherin was significantly downregulated. Conversely, the sh-COMP group showed a notable reduction in these ECM/EMT markers and restoration of E-cadherin expression (Fig. 8C, Figure S9).

COMP modulates PI3K-AKT signaling pathway

Furthermore, we examined PI3K-AKT signaling activity. Western blotting revealed that COMP knockdown significantly reduced the levels of phosphorylated PI3K and AKT, while total PI3K and AKT remained unchanged. Conversely, COMP overexpression enhanced the phosphorylation of both PI3K and AKT, indicating activation of the pathway (Fig. 9A, Figure S10A).

Fig. 9.

Detection of PI3K-AKT signaling pathway. A Western blot analysis of PI3K-AKT in HT-29 cells. Original blots are presented in Figure S10A. n = 9. Data are shown as mean ± SD. *** P<0.001. B Western blot analysis of PI3K-AKT in liver metastases. Original blots are presented in Figure S10B. Data are presented as mean ± SD, n = 6 mice per group. ***P < 0.001

Importantly, we also observed that phosphorylation of PI3K and AKT was significantly enhanced in the COMP-OE group in CRC mice, whereas total PI3K and AKT levels remained unchanged. In the sh-COMP group, phosphorylation of both proteins was markedly suppressed (Fig. 9B, Figure S10B).

Discussion

This study systematically investigated the role of EMT in the progression and metastasis of CRC. Through integrated bioinformatics analysis, machine learning model construction, and experimental validation, we identified EMT-related genes, particularly COMP, as key factors associated with CRC progression, metastasis, and poor prognosis.

Firstly, enrichment analysis revealed significant enrichment of EMT in CRC, suggesting that EMT might be a critical driver of CRC development [20]. Analysis of primary and metastatic CRC tissues further demonstrated altered expression of COMP and other EMT-related markers in metastatic CRC compared to primary CRC. These findings support the pivotal role of EMT in CRC metastasis. The decrease in E-cadherin and increase in N-cadherin expression drive EMT, promoting CRC invasion and migration [21].

Further differential expression analysis identified 36 EMT-DEGs potentially involved in CRC metastasis. These genes were significantly enriched in the ECM-receptor interaction, Focal adhesion, and PI3K-Akt signaling pathways, highlighting their central role in CRC progression. The ECM-receptor interaction and Focal adhesion pathways regulate tumor cell adhesion, motility, and survival, influencing distant metastasis of CRC [22]. Interactions between the ECM and cell surface receptors modulate cell behavior and play vital roles in intercellular communication, proliferation, adhesion, and migration, making these pathways crucial in CRC progression and metastasis [23]. Focal adhesion kinase promotes tumor progression primarily by altering tumor cell invasiveness, EMT, and stemness [24]. The PI3K-AKT pathway is essential for CRC development, playing a critical role in proliferation, metastasis, survival, and angiogenesis [25].

In the machine learning analysis, the RSF model demonstrated the highest predictive accuracy. Moreover, its robust ability to stratify CRC patients into high-risk and low-risk groups highlights its potential application in personalized medicine. RSF is considered a suitable analytical model for survival data and is frequently utilized in modeling for CRC analysis [26].

Using the RSF model, 22 EMT-related genes were identified as significantly associated with patient prognosis, with COMP emerging as a key factor. Cox regression analysis revealed that patients with high COMP expression had a significantly increased risk of poor survival. Further Kaplan-Meier survival curve analysis confirmed that high COMP expression was strongly associated with unfavorable outcomes.

COMP is highly expressed in several solid tumor types, including colon cancer, colorectal cancer, prostate cancer, liver cancer, and breast cancer, where its overexpression correlates with poorer prognosis [27]. COMP has been associated with clinical and pathological features of patients and is linked to shorter overall survival (OS) [28]. Additionally, COMP is a characteristic marker of colorectal cancer liver metastasis [29]. COMP plays a pivotal role in the ECM-receptor interaction and PI3K-Akt signaling pathways. It enhances the ability of tumor cells to degrade surrounding ECM by upregulating MMP9 expression, thereby promoting EMT and tumor metastasis [30]. Furthermore, COMP activates the Akt pathway to stimulate cell proliferation, impacting the prognosis of CRC patients [31]. These findings suggest that COMP is not only a central molecule in EMT regulation but also a crucial driver of CRC progression and metastasis. This underscores its potential as a therapeutic target in CRC.

We demonstrated that COMP is intricately involved in modulating CRC-specific ECM components (such as Collagen I, Laminin, and Hyaluronan), EMT markers, and related metastatic behaviors. Moreover, the expression of COMP to specific cells underscores its functional and spatial specificity in CRC. Fibroblasts and epithelial not only exhibit high COMP expression but also co-activate canonical EMT drivers, suggesting that COMP is a hallmark of EMT-prone subpopulations. The co-localization of COMP with EMT markers at the single-cell level validates its mechanistic involvement in tumor progression. Recent evidence suggests that tumor-associated fibroblasts can significantly alter epithelial tumor cell behavior by secreting ECM components and soluble factors [32]. Consequently, COMP expressed by fibroblasts might interact functionally with epithelial tumor cells, thereby promoting EMT-driven invasiveness and metastatic colonization specifically in CRC. Similarly, epithelial-derived COMP may modulate the stromal compartment, reinforcing ECM stiffness and enhancing tumor progression [33, 34]. Although the overall expression of COMP appeared low across total epithelial and fibroblast populations, subcluster analysis revealed enrichment in specific EMT-prone subpopulations. These subclusters also exhibited elevated expression of mesenchymal markers such as COL1A1 and MMP2, suggesting that COMP may be selectively expressed in pro-invasive cell states.

Functionally, COMP silencing in HT-29 cells inhibited EMT marker expression, reduced cell migration, invasion, and proliferation, and increased apoptosis. In contrast, COMP overexpression enhanced these malignant phenotypes. EMT is characterized by the transition from E-cadherin to N-cadherin, which enhances colorectal cancer metastasis [35]. COMP knockdown effectively inhibited EMT activation, demonstrating COMP’s regulatory role in the EMT process. Furthermore, COMP knockdown significantly reduced HT-29 cell proliferation, migration, and invasion, while markedly promoting apoptosis. Our study suggests that COMP contributes to CRC metastasis by modulating key components of the extracellular matrix and engaging downstream effectors involved in EMT, such as MMP2, MMP9, and N-cadherin. Mechanistically, COMP appears to enhance ECM stiffness and remodeling, which in turn may activate integrin-mediated focal adhesion and PI3K-Akt signaling pathways, fostering an EMT-permissive environment. COMP has been shown to induce EMT and cancer stem cell-like properties, facilitating tumor invasion and metastasis [36]. In colorectal cancer, COMP is co-expressed with EMT transcription factors and is associated with poor patient survival rates [37]. In terms of therapeutic relevance, silencing COMP not only reversed EMT markers but also sensitized CRC cells to apoptosis. These dual effects underscore the potential of COMP as a druggable target. Our results confirm that COMP not only serves as an EMT-associated prognostic biomarker but also functions as an active driver of CRC cell migration and invasion through modulation of EMT transcriptional programs and ECM components. These findings support the therapeutic potential of targeting COMP to suppress CRC metastasis.

To validate these observations in vivo, we established a CRC liver metastasis mouse model using HT-29 cells stably overexpressing or silencing COMP. Mice injected with COMP-overexpressing cells exhibited significantly enhanced hepatic colonization, elevated ECM and EMT marker expression, and activation of the PI3K-AKT pathway. In contrast, COMP knockdown suppressed these effects. These in vivo results not only confirm our in vitro findings but also reinforce the oncogenic function of COMP in driving metastatic progression.

Notably, although COMP has been reported to promote tumor progression and EMT in other malignancies, we demonstrated that COMP not only correlates with poor prognosis and EMT marker expression in CRC, but also actively regulates ECM remodeling and downstream signaling pathways associated with metastatic potential. These findings highlight the unique oncogenic function of COMP in the CRC context and support its potential as a CRC-specific prognostic biomarker and therapeutic target.

This study has some limitations. Although resource constraints currently limit expansion to small sample size, we are actively working on a multicenter validation cohort and plan to incorporate these additional data in a follow-up study. Second, the absence of in vivo validation limits the translational scope of our findings. Although the in vitro data suggest a pro-metastatic function of COMP, we did not assess its therapeutic relevance in animal models. Future studies employing orthotopic or liver metastasis mouse models will be essential to determine whether COMP silencing impairs tumor growth and metastatic colonization in vivo. Then, our current functional experiments primarily relied on siRNA-mediated knockdown of COMP in HT-29 cells, without complementary rescue or COMP overexpression assays. Such additional experiments would further strengthen our mechanistic claims. Moreover, the predictive model based on the RSF algorithm showed excellent performance in the training dataset (GSE39582) but exhibited moderate performance in the validation dataset (TCGA), suggesting potential overfitting. Future studies should incorporate regularization and feature reduction strategies to enhance model robustness and generalizability.

Conclusion

COMP has emerged from our analyses as a potentially valuable prognostic marker and candidate therapeutic target for CRC. Although our data strongly support COMP’s role in CRC progression and EMT regulation in vitro, further validation through animal studies and larger clinical cohorts is required. The current findings offer preliminary insights into CRC biology and suggest COMP targeting as a potential strategy worth exploring to mitigate CRC metastasis.

Authors’ contributions

He Huang contributed to design the study and write the manuscript. Lei Wang collected the samples and generated the data. Song Gao performed experiment and analyzed data. Haijiang Wang drafted and revised the manuscript. All authors read and approved the final manuscript.

Funding

Not applicable.

Data availability

The datasets generated and/or analyzed during the current study are available from corresponding author.

Declarations

Ethics approval and consent to participate

The study protocol was reviewed and approved by the Ethics Review Committee of the Fifth Affiliated Hospital of Xinjiang Medical University (Approval No. XYDWFYLSk-2024-149). All patients provided written informed consent, and the sample collection process strictly adhered to the principles of the Declaration of Helsinki. All animal procedures were approved by the Institutional Animal Care and Use Committee of Xinjiang Medical University (No. IACUC20240615-1).

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.