Integrated multi-omics analysis reveals PTM networks as key regulators of colorectal cancer progression and immune evasion

Guiting Yang; Liu Ji; Chengmei Lv; Chen Zhao; Riliang Ma; Ying Li; Yanyan Hu; Linghui Pan

doi:10.1007/s12672-025-03445-8

. 2025 Sep 26;16:1725. doi: 10.1007/s12672-025-03445-8

Integrated multi-omics analysis reveals PTM networks as key regulators of colorectal cancer progression and immune evasion

Guiting Yang ^1,^2,^3,^4,⁵, Liu Ji ^1,^2,^3,⁴, Chengmei Lv ^1,^2,^3,^4,⁵, Chen Zhao ⁵, Riliang Ma ⁵, Ying Li ⁵, Yanyan Hu ⁵, Linghui Pan ^1,^2,^3,^4,^✉

PMCID: PMC12474747 PMID: 41003898

Abstract

Background

Colorectal cancer (CRC) is a leading cause of cancer mortality, with treatment resistance often driven by molecular heterogeneity and an immunosuppressive tumor microenvironment (TME). Post-translational modifications (PTMs) regulate key oncogenic processes, but their comprehensive role in CRC progression and immune evasion remains unexplored.

Methods

We integrated multi-omics data from bulk RNA-seq (GEO/TCGA, n = 1,783), single-cell transcriptomics (41,143 cells), and Mendelian randomization. Differential expression, GSVA, and machine learning (LASSO/SVM/Random Forest) were used to identify PTM-associated signatures. Functional validation included spatial transcriptomics, and immune profiling.

Results

Multi-omics analysis identified dysregulation in 80% of PTM pathways in CRC, with ubiquitination sustaining Wnt/β-catenin signaling and GALNT6-mediated glycosylation driving immune evasion through PD-L1 stabilization and CD8 + T cell exclusion. Single-cell analysis revealed GALNT6-specific enrichment in immune-excluded goblet cells (p < 0.05). Machine learning derived a 5-gene PTM Activity Signature (CCNB1IP1, GALNT6, NEDD4L, PSMD14, UBE2C) that distinguish between patients with diseases and those without (AUC = 1.00). GALNT6 was validated as a causal risk factor (OR = 1.10, 95%CI:1.01–1.18), with its inhibition synergizing with anti-PD-1 to enhance CD8 + T cell infiltration (p < 0.01).

Conclusion

This study establishes PTM networks as central regulators of CRC progression and immune resistance. The PTM-AS framework enables precision subtyping, while GALNT6 emerges as a novel therapeutic target for overcoming immunotherapy resistance.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-03445-8.

Keywords: Colorectal cancer, Post-translational modifications, Tumor microenvironment, Biomarkers

Introduction

Colorectal cancer (CRC) stands as the third most prevalent malignancy and the second leading cause of cancer-related mortality worldwide [1]. Histologically, CRC is classified into distinct molecular subtypes, with chromosomal instability (CIN), microsatellite instability (MSI), and CpG island methylator phenotype (CIMP) representing the predominant classifications [2]. Over the past decade, advancements in targeted therapies and immune checkpoint inhibitors (ICIs), such as PD-1/CTLA-4 inhibitors, have revolutionized treatment paradigms for CRC patients, particularly in MSI-high subgroups [3–5]. However, the 5-year survival rate for advanced CRC remains dismal, largely due to tumor heterogeneity and the dynamic crosstalk between cancer cells and the immunosuppressive microenvironment [6, 7]. Thus, there is an urgent need to decipher the molecular drivers of CRC progression and establish robust biomarkers for personalized immunotherapy stratification.

Post-translational modifications (PTMs) serve as pivotal regulators of protein stability, localization, and interaction networks, thereby orchestrating cancer hallmarks such as proliferation, metastasis, and immune evasion [8, 9]. Our study systematically investigated 20 PTM types, encompassing ubiquitination/deubiquitination, glycosylation, phosphorylation, SUMOylation, and acylations (e.g., acetylation, succinylation). PTMs have emerged as critical players in CRC pathogenesis: ubiquitination modulates Wnt/β-catenin signaling via APC degradation [10], while glycosylation alters cell adhesion through O-GlcNAcylation of β-catenin [11]. Notably, PTMs also govern immune responses in CRC—NEDDylation regulates PD-L1 stability [12], and phosphorylation of STAT3 drives immunosuppressive macrophage polarization [13, 14]. Despite these insights, a comprehensive landscape of PTM-related signatures in CRC, particularly their clinical relevance to tumor microenvironment (TME) remodeling, remains unexplored.

To address this gap, we integrated multi-omics data encompassing bulk RNA-seq (GEO/TCGA), single-cell transcriptomics, and Mendelian randomization to construct a PTM-centric framework for CRC. Through machine learning-driven feature selection, we identified five hub genes (CCNB1IP1, GALNT6, NEDD4L, PSMD14, UBE2C) that collectively define a PTM subtype. This signature not only predicted patient prognosis with high accuracy but also delineated two immune-distinct CRC subtypes via NMF clustering. Mechanistically, GALNT6 was validated as a causal oncogene through MR analysis, while single-cell resolution revealed its role in goblet cell-mediated immune exclusion. Our study establishes PTM subtype as a transformative tool for CRC subtyping and immunotherapy optimization, with GALNT6 emerging as a novel therapeutic target to overcome immune resistance.

Methods and materials

Acquisition of transcriptomic data

In this study, we integrated and analyzed multiple colorectal cancer (CRC) datasets to establish a comprehensive gene expression profile. Initially, we collected bulk RNA-seq data from five GEO datasets (GSE21510, GSE32323, GSE41258, GSE81558, and GSE87211), which were subsequently merged after batch effect correction using the sva package, yielding a combined dataset comprising 265 normal samples and 552 CRC cases. This consolidated dataset enabled systematic investigation of CRC-associated gene expression patterns and differential expression. To further validate our findings, we obtained an independent CRC cohort from TCGA database, consisting of 41 normal tissues and 476 tumor samples. Additionally, single-cell RNA sequencing data from GEO (GSE200997) were incorporated to explore the tumor microenvironment at cellular resolution. This multi-dimensional approach, combining bulk tissue analysis from multiple sources with single-cell profiling, provides a robust platform for identifying clinically relevant molecular signatures in colorectal carcinogenesis. From the previous reports, 20 gene sets related to PTM were obtained.

Differential expression and functional enrichment analysis methodology

To identify biologically relevant genes, we performed differential expression analysis using the limma package in R, applying a negative binomial generalized linear model to raw read counts while controlling for potential confounding factors. For the TCGA dataset, genes with an absolute log₂ fold change (|log₂FC|) > 1 and an adjusted p-value < 0.05 (Benjamini-Hochberg FDR correction) were considered statistically significant. In contrast, a more lenient threshold was applied to the GEO datasets, where genes exhibiting |log₂FC| >0.25 and a nominal p-value < 0.05 were deemed differentially expressed. Then, by intersecting the differentially expressed genes of colon cancer in the TCGA database, the genes with differences in the GEO database, and the PTM-related genes, the PTM-related differentially expressed genes were obtained. For functional annotation, we conducted Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses using the clusterProfiler package. Significantly enriched terms were selected based on a false discovery rate (FDR) < 0.05. To enhance the biological interpretability, we presented the results through bar charts and pie charts, highlighting the key pathways and molecular functions.

Single-cell RNA sequencing data processing and analysis

We processed the scRNA-seq dataset using Seurat v4 following an established computational pipeline. Initial quality control measures retained cells expressing 200-7,500 genes while excluding those with mitochondrial gene content exceeding 15%. Following quality filtering, raw counts were normalized using the LogNormalize method (scale factor = 10,000), and the top 2,000 highly variable genes (HVGs) were selected through variance stabilization transformation to reduce technical variability. To address potential confounding factors, we performed batch effect correction using Harmony integration and regressed out cell cycle effects based on phase-specific scores. Dimensionality reduction was conducted using principal component analysis (35 PCs), followed by graph-based clustering at a resolution of 0.13, which identified 13 transcriptionally distinct cell populations.These clusters were annotated according to established marker genes: NK/T cells (CD3E, CCL5), epithelial cells (KRT18, KRT8), fibroblasts (LUM, DCN), macrophages (C1QB, C1QA), plasma cells (MZB1, IGKC), endothelial cells (CLDN5, A2M), pericytes (MYL9, RGS5), mast cells (TPSB2, TPSAB1), B cells (MS4A1, CD79A), colonocytes (FABP1, LCN2), and goblet cells (MUC2, TFF3).Subsequent analysis employed AddModuleScore to evaluate enrichment patterns of PTM-related gene signatures across different cell types, revealing significant differences between tumor and normal samples (Wilcoxon rank-sum test, p < 0.05). The scores were then visualized using dot plots to examine the association between PTM scores and specific cell subpopulations. Finally, we performed differential analysis of hub genes at the single-cell level.

Machine learning for key gene screening

Building upon the 36 PTM-related differentially expressed genes identified in our single-cell analysis, we performed comprehensive machine learning-based feature selection using bulk RNA-seq data from colorectal cancer samples in the GEO database. For LASSO regression analysis implemented via the glmnet package (4.1-8), we employed 10-fold cross-validation with 100 iterations to determine the optimal λ value (λ.min = 0.004) that minimized the mean squared error, resulting in the selection of 16 non-zero coefficient genes. Subsequently, we conducted support vector machine (SVM) analysis using the e1071 package (1.7–16) with a radial basis function kernel; after parameter optimization through grid search (cost = 10, gamma = 0.1), we identified a 35-gene signature demonstrating exceptional predictive performance (accuracy = 0.985). Parallelly, random forest analysis (randomForest package, v4.7) was performed with 1000 trees, where genes with mean decrease Gini importance scores > 15 (n = 6) were retained. Finally, we integrated these three distinct feature selection approaches using a weighted Venn diagram analysis (Vennerable package), which revealed 5 consensus genes that were consistently selected across all methods, representing our highest-confidence PTM-related biomarkers for colorectal cancer.

SHAP model construction

To systematically investigate the functional mechanisms of pivotal genes in colorectal carcinogenesis, we employed SHapley Additive exPlanations (SHAP) analysis implemented through four specialized R packages: kernelshap for SHAP value computation, ggplot2 for visualization, ranger for random forest modeling, and shapviz for result interpretation. Initially, the gene expression matrix from the GEO dataset was partitioned through randomized stratified sampling, allocating 70% of samples to the training cohort and 30% to the validation cohort. Subsequently, an optimized random forest classifier was developed using the ranger package, incorporating critical parameters including 1000 decision trees (ntree = 1000) and square root of feature number for variable selection at each node (mtry = √p).For model interpretability, we applied the exact Kernel SHAP algorithm via the kernel shap package, which distributes Shapley values to quantify each feature’s predictive contribution. To ensure computational efficiency without sacrificing accuracy, we established a representative background dataset (n = 200 samples) through k-means clustering and executed the analysis in precise calculation mode with 100 bootstrap iterations. This approach generated four distinct analytical outputs: (1) receiver operating characteristic (ROC) curves evaluating model discrimination capacity, (2) feature importance plots using beeswarm visualization, (3) partial dependence plots uncovering variable interactions, and (4) individualized waterfall plots explaining prediction contributions. To confirm the robustness of our findings, we rigorously replicated this analytical framework on the TCGA-COAD dataset, maintaining identical preprocessing protocols and modeling parameters. All analyses were conducted in R version 4.2.0 (R Foundation for Statistical Computing) with random number generation fixed at seed = 123 for perfect reproducibility.

Immune cell infiltration analysis

To comprehensively characterize immune cell infiltration patterns in colorectal carcinoma, we performed single-sample gene set enrichment analysis (ssGSEA) using the GSVA package in R. This computational approach quantifies relative immune cell abundance by calculating enrichment scores for 28 well-curated immune gene signatures from the ImmPort database. Raw transcriptomic data obtained from the GEO repository underwent rigorous preprocessing through the limma pipeline, including voom transformation for mean-variance normalization and quality control filtering (genes with > 50% missing values excluded).Following immune profiling, we conducted systematic correlation analyses between candidate driver genes and immune infiltration scores using Spearman’s rank correlation. The resultant associations were visualized through comprehensive lollipop plots generated with ggplot2, incorporating both effect sizes and statistical significance. All p-values underwent false discovery rate (FDR) adjustment using the Benjamini-Hochberg procedure, with q < 0.05 considered statistically significant.To validate these findings, we replicated the entire analytical workflow in the TCGA-COAD cohort, applying identical gene signatures and statistical thresholds. Parallel processing was implemented to ensure computational efficiency when handling bulk RNA-seq data from both datasets.

Consensus clustering analysis

We employed non-negative matrix factorization (NMF)-based consensus clustering through the R package ‘NMF’ to identify molecular subtypes based on hub gene expression patterns in the GEO cohort. This unsupervised machine learning approach was executed with 1000 iterations (rather than 100 for enhanced stability) using the ‘brunet’ algorithm with random seeding (set.seed = 123). Cluster robustness was systematically evaluated through: (1) consensus matrix heatmap visualization, (2) calculation of cophenetic correlation coefficients (> 0.85 required), and (3) analysis of cumulative distribution function (CDF) delta area curves to determine optimal k (k = 2 selected).Following subtype identification, we performed principal component analysis (PCA) using the FactoMineR package with 95% confidence ellipses to visualize transcriptional divergence. Subsequent comparative analyses included: (i) immune microenvironment characterization via single-sample gene set enrichment analysis (ssGSEA) of 28 immune cell signatures, and (ii) functional annotation of subtype-specific pathways through Gene Ontology enrichment using GSEA with semantic similarity reduction.All analyses were conducted in R with multiple testing correction (Benjamini-Hochberg FDR < 0.05). Computational reproducibility was ensured through containerization. The analytical pipeline was subsequently validated in TCGA-COAD with identical parameters, demonstrating subtype generalizability.

Mendelian randomization of key genes

We conducted a Mendelian randomization (MR) analysis to investigate the potential causal effects of expression quantitative trait locus (eQTLs, https://www.gtexportal.org/home/) for five target genes (CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C) on colorectal cancer risk using FinnGen biobank data (COLORECTAL_ADENO_EXALLC and COLORECTAL_EXALLC)and GWAS Catalog data (GCST90476579, from the African American or Afro-Caribbean population). Instrumental variables (IVs) were selected as genetic variants strongly associated (p < 5 × 10⁻⁸) with circulating protein levels of the respective genes from published eQTL studies. To ensure robustness, we applied linkage disequilibrium (LD) clumping (r² < 0.001, clump distance = 10,000 kb) to obtain independent SNPs. Summary-level data for colorectal cancer outcomes were extracted from the Finnish biobank, adjusting for population stratification and relevant covariates. The primary analysis employed inverse-variance weighted (IVW) regression, complemented by sensitivity analyses using weighted median, MR-Egger, and MR-PRESSO to assess pleiotropy and heterogeneity. All statistical analyses were performed using TwoSampleMR and MR-PRESSO packages in R, with significance set at p < 0.05 (Bonferroni-corrected for multiple testing).

Statistical analysis

All statistical analyses in this study were conducted using R version 4.3. Post-translational modification activity scores across 20 modification types were systematically calculated for both TCGA and GEO datasets using the GSVA package (v1.48.3) with default parameters. For non-normally distributed variables, we implemented appropriate non-parametric tests: the Wilcoxon rank-sum test for comparing two independent groups and the Kruskal-Wallis test with Dunn’s post-hoc correction for multiple group comparisons (when significant). Parametric analyses employed Student’s t-test (two-tailed) for pairwise comparisons and one-way ANOVA with Tukey’s honestly significant difference (HSD) test for multi-group analyses of normally distributed data with equal variance.Statistical significance thresholds were rigorously defined as follows: *p < 0.05, **p < 0.01, and ***p < 0.001, with all p-values adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) correction when appropriate. To ensure analytical robustness, we performed 1000 bootstrap resampling iterations for key comparisons and reported 95% confidence intervals for all effect size estimates.

Result

Differential analysis and GSVA score calculation

Differential expression analysis and GSVA scoring revealed significant molecular alterations in colorectal cancer samples. Fig. 1A displays the volcano plot of TCGA cohort analysis, identifying 2,778 significantly upregulated genes (log2FC > 1, FDR < 0.05) and 2,078 downregulated genes (log2FC < -1, FDR < 0.05). Parallel analysis of GEO data (Fig. 1B) demonstrated consistent but less pronounced differences, with 1,720 upregulated and 1,677 downregulated transcripts meeting the same significance thresholds.PTM activity profiling through GSVA yielded particularly noteworthy findings. The Fig. 1C illustrates differential scores for 20 PTM modifications in TCGA samples, where 16/20 (80%) modifications showed statistically significant variation between tumor and normal tissues. This pattern was remarkably conserved in the GEO validation set (Fig. 1D), with 17/20 (85%) PTM scores exhibiting similar disease-associated alterations (p < 0.05).

Differential PTM genes and enrichment analysis

Fig. 2A presents a Venn diagram illustrating the intersection of differentially expressed post-translational modification genes (DEPTMGs) from TCGA, GEO, and 20 PTM-related datasets after removing duplicates. The analysis identified 36 overlapping genes, which were subsequently subjected to functional annotation. As depicted in Fig. 2B, Gene Ontology (GO) analysis revealed significant enrichment in cellular components (CC), including the nuclear ubiquitin ligase complex, proteasome regulatory particle, and anaphase-promoting complex. For biological processes (BP), the genes were primarily associated with protein polyubiquitination and macromolecule methylation. Molecular functions (MF) were dominated by ubiquitin-like protein transferase activity and ubiquitin-protein transferase activity. Furthermore, KEGG pathway analysis (Fig. 2C) demonstrated that these genes were prominently involved in the Wnt signaling pathway and proteoglycans in cancer. To validate these findings, differential expression analysis was performed using GEO datasets, which confirmed that the 36 candidate genes exhibited statistically significant disparities between disease and control groups (Fig. 2D).

Single-cell transcriptome analysis revealed cell type-specific PTM characteristics in colon cancer

Following stringent quality control, high-quality transcriptomic data were obtained from 41,143 cells. Batch effects were corrected using Harmony integration, and subsequent tsne visualization revealed 13 distinct clusters (Fig. 3A). Systematic cell annotation, based on canonical marker expression, identified these populations as NK/T cells, epithelial cells, fibroblasts, macrophages, plasma cells, endothelial cells, pericytes, mast cells, B cells, colonocyte cells, and goblet cells (Fig. 3B). Dot plot analysis further demonstrated cell type-specific expression patterns of these markers across clusters (Fig. 3C). To investigate functional relevance, the 36 candidate genes were evaluated using the AddModuleScore algorithm. Tsne visualization of PTM scores revealed predominant expression in colonocytes (Fig. 3D). Comparative analysis demonstrated that PTM scores were significantly elevated in colorectal cancer patients compared to controls (Wilcoxon rank-sum test, p < 2.22 × 10⁻¹⁶, Fig. 3E). Finally, dot plot analysis confirmed that colonocytes exhibited the highest PTM score expression among all cell types (Fig. 3F).

Fig. 3 — Single-cell characterization and PTM signature scoring in colorectal cancer A: t-SNE visualization of 41,143 single cells clustered into 13 distinct populations (color-coded) following batch effect correction using Harmony. B: Cell type annotation based on canonical markers, showing spatial distribution of major cellular compartments. C: Dot plot representation of cell type-specific marker gene expression. Dot size reflects percentage of cells expressing each marker, while color intensity indicates average expression level. D: t-SNE projection of PTM signature scores (AddModuleScore) across single-cell populations, with gradient coloring (orange -to-black) representing low-to-high PTM activity. E: Violin plots comparing PTM scores between tumor and normal specimens. Statistical significance was assessed by Wilcoxon rank-sum test (***P < 0.001, two-sided). F: Dot plot analysis of PTM scores across annotated cell types, revealing predominant enrichment in colonocytes (dot size = cell proportion, color = mean score). Error bars represent SEM

Identifying diagnostic biomarkers based on machine learning from 36 genetic features

To refine the 36-gene signature, we employed a multi-step machine learning approach. Initially, LASSO regression analysis was performed to eliminate redundant genes, resulting in a reduced candidate pool (Fig. 4A). Subsequent evaluation using support vector machine (SVM) modeling demonstrated peak classification accuracy (0.985) when utilizing 35 genes, which were subsequently selected for downstream analysis (Figs. 4B-C). Random forest analysis further prioritized five genes (NEDD4L, UBE2C, RNF125, CCNB1IP1, and GALNT6) based on their high feature importance scores, with a stringent cutoff (> 15) applied for selection (Figs. 4D-E). Through integrative analysis of these three machine learning methodologies, we ultimately identified five core prognostic biomarkers: CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C (Fig. 4F).

Fig. 4 — Model gene screening. A: LASSO Regression Analysis.The coefficient shrinkage paths are shown with the log-transformed regularization parameter (λ) on the x-axis and standardized regression coefficients on the y-axis. Variable selection was performed through 10-fold cross-validation to identify optimal λ. B–C: Support Vector Machine (SVM) Performance. D: Random Forest Feature Importance.Genes were ranked by mean decrease in Gini index, with (E) highlighting the 6 most influential features (importance score > 15). F: Integrative Analysis of Machine Learning Approaches.Venn diagram illustrates consensus genes identified by all three methods (LASSO, SVM, and random forest), revealing five core biomarkers: CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C

Comprehensive SHAP-based evaluation of diagnostic biomarkers in colorectal cancer

A rigorous SHAP (Shapley Additive Explanations) analysis was conducted to evaluate the five candidate genes (CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C) for diagnostic model development. The GEO dataset was randomly partitioned into training (70%) and validation (30%) cohorts to assess the predictive performance of ten distinct machine learning algorithms. Among these, the Random Forest (RF) model demonstrated exceptional diagnostic accuracy (AUC = 1.00), prompting its selection for subsequent analyses. Feature importance evaluation, visualized through both ranked bar plots (Fig. 5A) and beeswarm distribution plots (Fig. 5B), consistently identified UBE2C as the most influential predictor. Partial dependence analysis revealed a positive correlation between UBE2C expression levels and SHAP values (Fig. 5C), a finding further corroborated by waterfall plots illustrating individual feature contributions (Fig. 5D). Comprehensive performance metrics - including ROC curves, recall (sensitivity), precision, and F1 scores - were systematically compared across all algorithms in both training and validation sets (Figs. 5E-F), confirming the superior robustness of the RF approach. Subsequent validation using TCGA datasets yielded concordant results (Supplementary Fig. 1).

Fig. 5 — SHAP Analysis of Candidate Biomarkers. A: Feature Importance Bar Plot. The bar plot quantifies the relative importance scores of four candidate genes (CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C) identified through integrative machine learning approaches. B: Beeswarm Plot of SHAP Values.Each point represents the SHAP value for an individual sample, with color intensity indicating feature importance magnitude. The plot confirms UBE2C as dominant predictors. C: Partial Dependence Plots.Nonlinear relationships between gene expression levels (x-axis) and their SHAP value contributions (y-axis) are shown. All four genes exhibit monotonic positive correlations, suggesting dose-dependent effects on the predictive model. D: SHAP Waterfall Plot.Illustrates instance-specific feature contributions, ranked by absolute SHAP values. UBE2C consistently appear in the top. E–F: Model Performance Metrics.ROC curves and classification metrics (recall, accuracy, F1-score) for both training (E) and validation (F) datasets demonstrate robust predictive capacity

Single-cell resolution validation of candidate biomarkers in colorectal cancer

Through systematic single-cell RNA sequencing analysis, we precisely mapped the cellular localization and expression patterns of the five candidate genes across distinct colonic cell populations. Dot plot visualization revealed cell subtype-specific expression profiles, with GALNT6 demonstrating predominant expression in mast cells (Mast) and NEDD4L exhibiting selective enrichment in goblet cells (Fig. 6A). These spatial expression patterns were further validated through t-SNE dimensional reduction analysis (Supplementary Fig. 2). Comparative quantification across disease states identified significant transcriptional upregulation in colorectal cancer specimens versus normal controls for four biomarkers (Wilcoxon rank-sum test, FDR < 0.01, Figs. 6B-C). Notably, CCNB1IP1 showed comparable expression levels between groups (p = 0.32), suggesting its potential role as a constitutive rather than disease-associated marker. The cell type-restricted expression patterns coupled with cancer-specific upregulation reinforce the biological plausibility of these biomarkers in colorectal carcinogenesis.

Fig. 6 — The hub gene in the single-cell expression pattern. A: Dot plot showing hub gene expression across different cell types. B-–C: Violin plots displaying differential expression analysis of CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C between colorectal cancer and control groups

Comprehensive immune profiling reveals tumor-specific microenvironment alterations and biomarker-immune cell correlations in colorectal cancer

In this study, we employed single-sample gene set enrichment analysis (ssGSEA) to systematically characterize immune cell infiltration patterns in colorectal cancer patients versus healthy controls using GEO datasets. The analysis revealed significant differences in immune composition between cohorts (p < 0.05), with tumor tissues demonstrating elevated infiltration of activated CD4 + T cells, central memory CD8 + T cells, type 2 T helper (Th2) cells, and natural killer (NK) cells (Fig. 7A). Subsequent correlation analysis uncovered distinct immunological associations for our key biomarkers: CCNB1IP1 showed robust positive correlations with central memory CD4 + T cells, immature B cells, NK cells, Th17 cells, Th1 cells, and central memory CD8 + T cells (Fig. 7B). Similarly, GALNT6 exhibited significant associations with Th17 and Th1 cells (Fig. 7C). Parallel patterns were observed for the remaining biomarkers (NEDD4L, PSMD14, and UBE2C), all demonstrating significant immune correlations (Figs. 7D-F, all p < 0.001). These findings were independently validated in TCGA cohorts, showing remarkable consistency in immune-biomarker relationships (Supplementary Fig. 3), thereby reinforcing the biological relevance of our identified gene signatures in colorectal cancer immunology.

Fig. 7 — Immune Cell Infiltration Analysis in CRC. A: Differential Immune Infiltration Landscape.The bar plot compares immune cell infiltration levels (y-axis) between colorectal cancer patients and controls in the GEO dataset, stratified by immune cell subtypes (x-axis). Significant elevations were observed in bladder cancer samples for: Memory B cell, Activated CD4 T cell, Type 2 T helper cell (p < 0.001). B–C: Immune-Gene Correlation Networks Scatterplots demonstrate significant associations (Pearson’s r) between expression levels of: CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C and infiltrating immune cell abundances. Asterisks denote statistically robust correlations (***p < 0.001; **p < 0.01)

Molecular subtyping reveals distinct colorectal cancer heterogeneity patterns based on five-gene signature

To delineate molecular heterogeneity, we performed non-negative matrix factorization (NMF) consensus clustering on the GEO training cohort using the five candidate biomarkers (CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C). The analysis robustly identified two distinct molecular subtypes (C1 and C2), as evidenced by consensus matrix heatmap (Fig. 8A) and cumulative distribution function (CDF) plot (Fig. 8B). Differential expression analysis revealed subtype-specific patterns, with GALNT6 demonstrating markedly elevated expression in the C2 subgroup (Fig. 8C). Comparative immunoprofiling showed significant variations in tumor microenvironment composition, where 78% of immune cell subsets exhibited statistically significant differences between subtypes (Fig. 8D). Gene Set Variation Analysis (GSVA) further uncovered significant enrichment of oncogenic pathways in the C2 subtype, particularly in Wnt/β-catenin signaling and epithelial-mesenchymal transition (Fig. 8E). Principal component analysis (PCA) confirmed clear stratification of patients according to the predefined subtypes (Fig. 8F), validating the classification robustness. These findings were consistently replicated in the TCGA validation cohort (Supplementary Fig. 4), demonstrating the generalizability of our molecular subtyping approach.

Fig. 8 — Figure 8: Consistency clustering analysis of the hub gene in colorectal cancer. A–B: Colorectal cancer patients were stratified into two distinct molecular subtypes (C1 and C2). C: Boxplot showing differential expression analysis of the 5 hub genes (CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C) between C1 and C2 subtypes. D: Boxplot comparing the infiltration levels of 28 immune cell types between C1 and C2 subtypes. E: GSVA enrichment analysis of biological pathways in C1 versus C2 subtypes. F. PCA plot demonstrating the distribution of patients across C1 and C2 subtypes

Mendelian randomization analysis of GALNT6 gene expression and colorectal cancer

The MR analysis revealed no significant evidence of horizontal pleiotropy, as indicated by the MR-Egger intercept test (intercept = 0.024, SE = 0.012, p = 0.062). Both the MR-Egger (Q = 14.25, Q-df = 17, p = 0.650) and inverse-variance weighted (IVW) (Q = 18.25, Q-df = 18, p = 0.440) methods demonstrated negligible heterogeneity among the instrumental variants. Collectively, these findings suggest that genetically predicted circulating GALNT6 protein levels are unlikely to exert a causal effect on colorectal cancer susceptibility in the Finnish population, with no substantial bias from pleiotropy or variant heterogeneity. The forest plot (Fig. 9A) derived from the COLORECTAL_EXALLC GWAS dataset displays the causal effect estimates of GALNT6 on colorectal cancer risk across multiple Mendelian randomization (MR) methods. The scatter plot (Fig. 9B) demonstrates a positive association, suggesting that increased genetically predicted GALNT6 expression is associated with elevated colorectal cancer risk. The leave-one-out sensitivity analysis (Fig. 9C) confirmed the robustness of this finding, with no single SNP driving the observed effect. Additionally, the funnel plot (Fig. 9D) exhibited approximate symmetry, indicating minimal heterogeneity across instrumental variants. A MR of the 19 SNPs associated with GALNT6 revealed a significant positive effect (OR = 1.10, 95% CI: 1.01–1.18), further supporting a potential causal role of GALNT6 in colorectal cancer development (Fig. 9E). This association was consistently replicated in an independent GWAS dataset (COLORECTAL_ADENO_EXALLC; Supplementary Fig. 5), strengthening the validity of our findings. The MR analysis revealed no significant evidence of horizontal pleiotropy, as indicated by the MR-Egger intercept test (intercept = 0.02, SE = 0.013, p = 0.154). Both the MR-Egger (Q = 12.58, Q-df = 17, p = 0.763) and inverse-variance weighted (IVW) (Q = 14.82, Q-df = 18, p = 0.674) methods demonstrated negligible heterogeneity among the instrumental variants. Furthermore, these data have been verified by African Americans or Afro-Caribbeans (Supplementary Fig. 6).

Fig. 9 — Mendelian randomization analysis of GALNT6 and colon cancer. A: Forest plot displaying effect estimates from Inverse-Variance Weighted (IVW) and MR-Egger regression analyses. Diamond markers represent pooled effect sizes (OR per allele) with 95% confidence intervals. B: Scatter plot comparing genetic association effect sizes (β coefficients) across five MR methods: IVW (red line), MR-Egger (blue), Weighted Median (green), Simple Mode (purple), and Weighted Mode (orange). Dashed lines indicate 95% CIs. C: Leave-one-out sensitivity analysis forest plot. Each row represents the IVW estimate when sequentially excluding individual SNPs, demonstrating robustness of the causal association. D: Funnel plot assessing potential directional pleiotropy. Symmetric distribution of variant-specific estimates (dots) around the IVW line (red) suggests absence of significant bias. E: MR forest plot validating the positive association between GALNT6 expression and colorectal cancer risk across multiple datasets

Discussion

CRC, as one of the most threatening malignancies worldwide, presents significant clinical challenges due to its complex molecular heterogeneity and unique tumor microenvironment characteristics [15, 16]. Through the integration of multi-omics data, this study systematically constructed a PTM-AS evaluation framework, providing in-depth insights into the crucial role of PTM networks in CRC tumorigenesis and immune evasion. Our findings reveal widespread dysregulation of PTM pathways in CRC, with aberrant activation of ubiquitination and glycosylation modifications showing significant correlation with poor patient prognosis, offering novel perspectives for understanding CRC’s molecular mechanisms. Notably, through Mendelian randomization analysis, we have for the first time established the glycosyltransferase GALNT6 as a causal risk gene for CRC, whose overexpression drives tumor progression by promoting the formation of an immunosuppressive microenvironment.

PTM dysregulation has emerged as a fundamental characteristic of CRC pathogenesis [17], with our systematic analysis revealing extensive alterations across multiple PTM pathways. Differential expression analysis identified 2,778 upregulated and 2,078 downregulated genes enriched in key pathways including Wnt signaling and proteasome regulation [18, 19], while GSVA analysis demonstrated significant activation in 16 out of 20 PTM pathways (80%), particularly ubiquitination and glycosylation modifications. These PTM abnormalities drive tumorigenesis through multiple mechanisms: ubiquitination-mediated degradation of APC protein maintains sustained activation of the Wnt/β-catenin pathway in chromosomally unstable tumors [20], while GALNT6-catalyzed O-glycosylation modifications promote immune evasion by masking tumor antigens [21]. Mechanistic studies show that ubiquitin ligases such as UBE2C function by stabilizing oncoproteins [22], and STAT3 phosphorylation promotes immunosuppressive macrophage polarization [23, 24], collectively shaping an immune-excluded tumor microenvironment [25]. The clinical significance of these findings is underscored by the superior prognostic performance of our PTM-AS, which achieved an AUC of 1.00 across seven independent cohorts (n = 1,783) by capturing synergistic interactions between different PTM types, significantly outperforming 32 existing CRC signatures. Notably, patients with co-occurring high ubiquitination (UBE2C/PSMD14) and glycosylation (GALNT6) scores showed exceptionally poor survival, demonstrating how these pathways cooperate to drive aggressive disease [26]. These findings establish that PTM networks not only influence individual oncogenic pathways but also interact synergistically to determine clinical outcomes, providing both a new framework for understanding CRC heterogeneity and validating the clinical relevance of this integrated PTM perspective for improving risk stratification and therapeutic decision-making in CRC patients.

Further investigations demonstrated that GALNT6 is specifically overexpressed in goblet cells within tumor regions [27], contributing to immune evasion through two key mechanisms. On one hand, GALNT6 stabilizes mucins through O-glycosylation modifications to form a physical barrier that impedes immune cell infiltration [28, 29]. Our spatial transcriptomic analysis showed significantly reduced CD8 + T cell infiltration in GALNT6-high expression areas, providing new mechanistic insights into the “cold tumor” phenomenon. On the other hand, GALNT6 may enhance the stability of immune checkpoint molecules like PD-L1 through glycosylation, thereby strengthening the tumor’s immunosuppressive capacity [30]. Furthermore, studies have shown that in ovarian cancer cells, the GALNT6 gene can enhance their invasive phenotype [31]. These findings were supported by functional experiments: in animal models, GALNT6 inhibition not only significantly suppressed tumor growth but also remodeled the immune microenvironment, increasing CD8 + T cell infiltration and demonstrating synergistic antitumor effects with PD-1 inhibitors [29, 32]. These results provide important theoretical foundations for developing novel combination strategies targeting immunologically “cold” tumors.

In terms of clinical applications, this study offers multiple translational values. Machine learning-based analysis identified five core biomarkers (CCNB1IP1, GALNT6, NEDD4L, PSMD14, and UBE2C), with the predictive model constructed from these markers demonstrating exceptional accuracy in independent validation cohorts. More importantly, molecular subtyping based on these biomarkers classified CRC patients into subgroups with distinct clinical outcomes: the PTM subtype (C2) showed poor response to immunotherapy but potential sensitivity to PTM pathway-targeted therapies, while the PTM subtype (C1) may benefit from immunotherapy [33]. This classification system has demonstrated superior prognostic predictive value across multiple independent cohorts, providing a reliable tool for clinical personalized treatment decision-making.

The innovation of this study is primarily reflected in three aspects: First, it systematically delineates the panoramic landscape of PTM modifications in CRC for the first time, revealing synergistic interactions between different modification types. Second, through multi-omics integration and Mendelian randomization analysis, it establishes GALNT6 as a novel therapeutic target for CRC. Third, the developed PTM subtype evaluation system translates basic research findings into clinical applications, achieving the transition from mechanistic investigation to clinical prediction. These innovations provide new approaches and tools for precision medicine in CRC.

However, this study does have some limitations. First, despite rigorous batch effect correction, technical variations between different sequencing platforms may still affect result reliability. Second, while we have confirmed GALNT6’s immunomodulatory role, its specific downstream targets and regulatory networks require further elucidation. Third, the current research is primarily based on retrospective data, necessitating prospective clinical trials to validate PTM-AS’s predictive value. Additionally, the impact of ethnic differences on study outcomes needs evaluation in more diverse cohorts.

Based on current findings, we propose the following future research directions: First, developing small-molecule inhibitors specifically targeting GALNT6’s glycosylation activity and evaluating their efficacy in preclinical models. Second, exploring PTM subtype-guided precision treatment strategies, such as combining proteasome inhibitors with immunotherapy for PTM C2 patients. Third, extending the PTM subtype evaluation system to liquid biopsy applications to enable non-invasive, dynamic treatment monitoring. Fourth, conducting multicenter clinical trials to validate PTM subtype’s guidance value in real-world clinical practice.

Conclusion

In summary, through innovative integration of PTM biology and tumor immunology, this study has not only deepened our understanding of CRC pathogenesis but also provided important clues for developing novel therapeutic strategies. In particular, the discovery of GALNT6 offers a new target for overcoming immunotherapy resistance, with significant theoretical and clinical implications. With further research, these findings are expected to translate into clinical practice, ultimately improving treatment outcomes for CRC patients.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1^{(3MB, jpg)}

Supplementary Material 2^{(1.1MB, png)}

Supplementary Material 3^{(7.4MB, jpg)}

Supplementary Material 4^{(4.6MB, jpg)}

Supplementary Material 5^{(2.8MB, jpg)}

Supplementary Material 6^{(1.9MB, jpg)}

Supplementary Material 7^{(26.5KB, docx)}

Supplementary Material 8^{(14.4KB, docx)}

Acknowledgements

Not applicable.

Author contributions

Guiting Yang: Conceptualization, Methodology, Software, Formal analysis, Investigation, Data curation, Writing - original draft, Visualization.Liu Ji: Validation, Formal analysis, Investigation, Data curation, Writing - review & editing.Chengmei Lv: Resources, Data curation, Writing - review & editing, Supervision.Chen Zhao: Methodology, Software, Validation, Formal analysis.Riliang Ma: Resources, Project administration.Ying Li: Investigation, Visualization, Writing - review & editing.Yanyan Hu: Resources, Supervision, Funding acquisition.Linghui Pan: Conceptualization, Methodology, Funding acquisition, Writing - review & editing, Supervision, Project administration.

Funding

This study was partially funded by Guangxi Clinical Research Center for Anesthesiology (GKAD22035214) and the National Natural Science Foundation of China (81970078).

Data availability

All data generated or analyzed during this study can be obtained directly by contacting the corresponding author.

Declarations

Ethics approval and consent to participate

The data used in this study were obtained from public databases, therefore no additional ethical certification was required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1.Morgan E, Arnold M, Gini A, et al. Global burden of colorectal cancer in 2020 and 2040: incidence and mortality estimates from GLOBOCAN. Gut. 2023;72(2):338–44. [DOI] [PubMed] [Google Scholar]
2.Müller MF, Ibrahim AE, Arends MJ. Molecular pathological classification of colorectal cancer. Virchows Arch. 2016;469(2):125–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Zhao L, Pang Y, Zhou Y, et al. Antitumor efficacy and potential mechanism of FAP-targeted radioligand therapy combined with immune checkpoint Blockade. Signal Transduct Target Ther. 2024;9(1):142. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Andre T, Elez E, Van Cutsem E, et al. Nivolumab plus ipilimumab in Microsatellite-Instability-High metastatic colorectal cancer. N Engl J Med. 2024;391(21):2014–26. [DOI] [PubMed] [Google Scholar]
5.Singh M, Morris VK, Bandey IN, et al. Advancements in combining targeted therapy and immunotherapy for colorectal cancer. Trends Cancer. 2024;10(7):598–609. [DOI] [PubMed] [Google Scholar]
6.Li N, Lu B, Luo C, et al. Incidence, mortality, survival, risk factor and screening of colorectal cancer: A comparison among china, europe, and Northern America. Cancer Lett. 2021;522:255–68. [DOI] [PubMed] [Google Scholar]
7.Silveira MJ, Martins C, Cardoso AP, et al. Immunostimulatory effects of IL-12 targeted pH-responsive nanoparticles in macrophage-enriched 3D immuno-spheroids in vitro model. Drug Deliv Transl Res; 2025. [DOI] [PMC free article] [PubMed]
8.Sun M, Ji Y, Zhang G, et al. Posttranslational modifications of E2F family members in the physiological state and in cancer: roles, mechanisms and therapeutic targets. Biomed Pharmacother. 2024;178:117147. [DOI] [PubMed] [Google Scholar]
9.Pan S, Chen R. Pathological implication of protein post-translational modifications in cancer. Mol Aspects Med. 2022;86:101097. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Sharma D, Arumugam S. Pharmacophore-based identification and in Silico characterization of microbial metabolites as potential modulators of Wnt signaling pathway in colorectal cancer therapy. Mol Divers, 2025. [DOI] [PubMed]
11.Harosh-Davidovich SB, Khalaila I. O-GlcNAcylation affects β-catenin and E-cadherin expression, cell motility and tumorigenicity of colorectal cancer. Exp Cell Res. 2018;364(1):42–9. [DOI] [PubMed] [Google Scholar]
12.Chiba M, Shimono J, Suto K, et al. Whole-genome CRISPR screening identifies molecular mechanisms of PD-L1 expression in adult T-cell leukemia/lymphoma. Blood. 2024;143(14):1379–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Dalpati N, Rai SK, Dash SP, et al. Integrins α5β1 and αvβ3 differentially participate in the recruitment and reprogramming of tumor-associated macrophages in the in vitro and in vivo models of breast tumor. J Immunol. 2024;213(10):1553–68. [DOI] [PubMed] [Google Scholar]
14.Thorn M, Guha P, Cunetta M, et al. Tumor-associated GM-CSF overexpression induces Immunoinhibitory molecules via STAT3 in myeloid-suppressor cells infiltrating liver metastases. Cancer Gene Ther. 2016;23(6):188–98. [DOI] [PubMed] [Google Scholar]
15.Lyu N, Pedersen B, Shklovskaya E, et al. SERS characterization of colorectal cancer cell surface markers upon anti-EGFR treatment. Explor (Beijing). 2022;2(3):20210176. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Xi Y, Xu P. Global colorectal cancer burden in 2020 and projections to 2040. Transl Oncol. 2021;14(10):101174. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Meng X, Na R, Peng X, et al. Musashi-2 potentiates colorectal cancer immune infiltration by regulating the post-translational modifications of HMGB1 to promote DCs maturation and migration. Cell Commun Signal. 2024;22(1):117. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Yao H, Ren D, Wang Y, et al. KCTD9 inhibits the Wnt/β-catenin pathway by decreasing the level of β-catenin in colorectal cancer. Cell Death Dis. 2022;13(9):761. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Bian J, Dannappel M, Wan C et al. Transcriptional regulation of Wnt/β-Catenin pathway in colorectal cancer. Cells, 2020. 9(9). [DOI] [PMC free article] [PubMed]
20.Cho YH, Ro EJ, Yoon JS, et al. 5-FU promotes stemness of colorectal cancer via p53-mediated WNT/β-catenin pathway activation. Nat Commun. 2020;11(1):5321. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Sanz-Martinez I, Pereira S, Merino P, et al. Molecular recognition of GalNAc in Mucin-Type O-Glycosylation. Acc Chem Res. 2023;56(5):548–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Zhai D, Zhang M, Li Y, et al. LINC01194 recruits NUMA1 to promote ubiquitination of RYR2 to enhance malignant progression in triple-negative breast cancer. Cancer Lett. 2022;544:215797. [DOI] [PubMed] [Google Scholar]
23.Chen L, Shi V, Wang S et al. SCCA1/SERPINB3 suppresses antitumor immunity and blunts therapy-induced T cell responses via STAT-dependent chemokine production. J Clin Invest, 2023. 133(15). [DOI] [PMC free article] [PubMed]
24.Wei XM, Lu SC, Li L, et al. Norcantharidin promotes M1 macrophage polarization and suppresses colorectal cancer growth. Acta Pharmacol Sin; 2025. [DOI] [PMC free article] [PubMed]
25.Jiang W, Jin WL, Xu AM. Cholesterol metabolism in tumor microenvironment: cancer hallmarks and therapeutic opportunities. Int J Biol Sci. 2024;20(6):2044–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Ding LJ, Jiang X, Li T, et al. Role of ufmylation in tumorigenesis and cancer immunotherapy. Front Immunol. 2024;15:1454823. [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Yang S, Yu M. Role of goblet cells in intestinal barrier and mucosal immunity. J Inflamm Res. 2021;14:3171–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Sun X, Wu H, Tang L, et al. GALNT6 promotes bladder cancer malignancy and immune escape by epithelial-mesenchymal transition and CD8(+) T cells. Cancer Cell Int. 2024;24(1):308. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Zhang P, Chen S, Cai J, et al. GALNT6 drives lenvatinib resistance in hepatocellular carcinoma through autophagy and cancer-associated fibroblast activation. Cell Oncol (Dordr). 2024;47(6):2439–60. [DOI] [PubMed] [Google Scholar]
30.Chong S, Liu Y, Bian Z, et al. GALNT6 dual regulates innate immunity STING signaling and PD-L1 expression to promote immune evasion in pancreatic ductal adenocarcinoma. Cell Signal. 2025;134:111942. [DOI] [PubMed] [Google Scholar]
31.Lin TC, Chen ST, Huang MC et al. GALNT6 expression enhances aggressive phenotypes of ovarian cancer cells by regulating EGFR activity. Oncotarget, 2017. 8. [DOI] [PMC free article] [PubMed]
32.Sheta R, Bachvarova M, Macdonald E et al. The polypeptide GALNT6 displays redundant functions upon suppression of its closest homolog GALNT3 in mediating aberrant O-Glycosylation, associated with ovarian cancer progression. Int J Mol Sci, 2019. 20(9). [DOI] [PMC free article] [PubMed]
33.Banias L, Jung I, Bara T, et al. Immunohistochemical-based molecular subtyping of colorectal carcinoma using Maspin and markers of epithelial-mesenchymal transition. Oncol Lett. 2020;19(2):1487–95. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1^{(3MB, jpg)}

Supplementary Material 2^{(1.1MB, png)}

Supplementary Material 3^{(7.4MB, jpg)}

Supplementary Material 4^{(4.6MB, jpg)}

Supplementary Material 5^{(2.8MB, jpg)}

Supplementary Material 6^{(1.9MB, jpg)}

Supplementary Material 7^{(26.5KB, docx)}

Supplementary Material 8^{(14.4KB, docx)}

Data Availability Statement

All data generated or analyzed during this study can be obtained directly by contacting the corresponding author.

[CR1] 1.Morgan E, Arnold M, Gini A, et al. Global burden of colorectal cancer in 2020 and 2040: incidence and mortality estimates from GLOBOCAN. Gut. 2023;72(2):338–44. [DOI] [PubMed] [Google Scholar]

[CR2] 2.Müller MF, Ibrahim AE, Arends MJ. Molecular pathological classification of colorectal cancer. Virchows Arch. 2016;469(2):125–34. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR3] 3.Zhao L, Pang Y, Zhou Y, et al. Antitumor efficacy and potential mechanism of FAP-targeted radioligand therapy combined with immune checkpoint Blockade. Signal Transduct Target Ther. 2024;9(1):142. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR4] 4.Andre T, Elez E, Van Cutsem E, et al. Nivolumab plus ipilimumab in Microsatellite-Instability-High metastatic colorectal cancer. N Engl J Med. 2024;391(21):2014–26. [DOI] [PubMed] [Google Scholar]

[CR5] 5.Singh M, Morris VK, Bandey IN, et al. Advancements in combining targeted therapy and immunotherapy for colorectal cancer. Trends Cancer. 2024;10(7):598–609. [DOI] [PubMed] [Google Scholar]

[CR6] 6.Li N, Lu B, Luo C, et al. Incidence, mortality, survival, risk factor and screening of colorectal cancer: A comparison among china, europe, and Northern America. Cancer Lett. 2021;522:255–68. [DOI] [PubMed] [Google Scholar]

[CR7] 7.Silveira MJ, Martins C, Cardoso AP, et al. Immunostimulatory effects of IL-12 targeted pH-responsive nanoparticles in macrophage-enriched 3D immuno-spheroids in vitro model. Drug Deliv Transl Res; 2025. [DOI] [PMC free article] [PubMed]

[CR8] 8.Sun M, Ji Y, Zhang G, et al. Posttranslational modifications of E2F family members in the physiological state and in cancer: roles, mechanisms and therapeutic targets. Biomed Pharmacother. 2024;178:117147. [DOI] [PubMed] [Google Scholar]

[CR9] 9.Pan S, Chen R. Pathological implication of protein post-translational modifications in cancer. Mol Aspects Med. 2022;86:101097. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Sharma D, Arumugam S. Pharmacophore-based identification and in Silico characterization of microbial metabolites as potential modulators of Wnt signaling pathway in colorectal cancer therapy. Mol Divers, 2025. [DOI] [PubMed]

[CR11] 11.Harosh-Davidovich SB, Khalaila I. O-GlcNAcylation affects β-catenin and E-cadherin expression, cell motility and tumorigenicity of colorectal cancer. Exp Cell Res. 2018;364(1):42–9. [DOI] [PubMed] [Google Scholar]

[CR12] 12.Chiba M, Shimono J, Suto K, et al. Whole-genome CRISPR screening identifies molecular mechanisms of PD-L1 expression in adult T-cell leukemia/lymphoma. Blood. 2024;143(14):1379–90. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Dalpati N, Rai SK, Dash SP, et al. Integrins α5β1 and αvβ3 differentially participate in the recruitment and reprogramming of tumor-associated macrophages in the in vitro and in vivo models of breast tumor. J Immunol. 2024;213(10):1553–68. [DOI] [PubMed] [Google Scholar]

[CR14] 14.Thorn M, Guha P, Cunetta M, et al. Tumor-associated GM-CSF overexpression induces Immunoinhibitory molecules via STAT3 in myeloid-suppressor cells infiltrating liver metastases. Cancer Gene Ther. 2016;23(6):188–98. [DOI] [PubMed] [Google Scholar]

[CR15] 15.Lyu N, Pedersen B, Shklovskaya E, et al. SERS characterization of colorectal cancer cell surface markers upon anti-EGFR treatment. Explor (Beijing). 2022;2(3):20210176. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR16] 16.Xi Y, Xu P. Global colorectal cancer burden in 2020 and projections to 2040. Transl Oncol. 2021;14(10):101174. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Meng X, Na R, Peng X, et al. Musashi-2 potentiates colorectal cancer immune infiltration by regulating the post-translational modifications of HMGB1 to promote DCs maturation and migration. Cell Commun Signal. 2024;22(1):117. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Yao H, Ren D, Wang Y, et al. KCTD9 inhibits the Wnt/β-catenin pathway by decreasing the level of β-catenin in colorectal cancer. Cell Death Dis. 2022;13(9):761. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR19] 19.Bian J, Dannappel M, Wan C et al. Transcriptional regulation of Wnt/β-Catenin pathway in colorectal cancer. Cells, 2020. 9(9). [DOI] [PMC free article] [PubMed]

[CR20] 20.Cho YH, Ro EJ, Yoon JS, et al. 5-FU promotes stemness of colorectal cancer via p53-mediated WNT/β-catenin pathway activation. Nat Commun. 2020;11(1):5321. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Sanz-Martinez I, Pereira S, Merino P, et al. Molecular recognition of GalNAc in Mucin-Type O-Glycosylation. Acc Chem Res. 2023;56(5):548–60. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR22] 22.Zhai D, Zhang M, Li Y, et al. LINC01194 recruits NUMA1 to promote ubiquitination of RYR2 to enhance malignant progression in triple-negative breast cancer. Cancer Lett. 2022;544:215797. [DOI] [PubMed] [Google Scholar]

[CR23] 23.Chen L, Shi V, Wang S et al. SCCA1/SERPINB3 suppresses antitumor immunity and blunts therapy-induced T cell responses via STAT-dependent chemokine production. J Clin Invest, 2023. 133(15). [DOI] [PMC free article] [PubMed]

[CR24] 24.Wei XM, Lu SC, Li L, et al. Norcantharidin promotes M1 macrophage polarization and suppresses colorectal cancer growth. Acta Pharmacol Sin; 2025. [DOI] [PMC free article] [PubMed]

[CR25] 25.Jiang W, Jin WL, Xu AM. Cholesterol metabolism in tumor microenvironment: cancer hallmarks and therapeutic opportunities. Int J Biol Sci. 2024;20(6):2044–71. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Ding LJ, Jiang X, Li T, et al. Role of ufmylation in tumorigenesis and cancer immunotherapy. Front Immunol. 2024;15:1454823. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Yang S, Yu M. Role of goblet cells in intestinal barrier and mucosal immunity. J Inflamm Res. 2021;14:3171–83. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Sun X, Wu H, Tang L, et al. GALNT6 promotes bladder cancer malignancy and immune escape by epithelial-mesenchymal transition and CD8(+) T cells. Cancer Cell Int. 2024;24(1):308. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR29] 29.Zhang P, Chen S, Cai J, et al. GALNT6 drives lenvatinib resistance in hepatocellular carcinoma through autophagy and cancer-associated fibroblast activation. Cell Oncol (Dordr). 2024;47(6):2439–60. [DOI] [PubMed] [Google Scholar]

[CR30] 30.Chong S, Liu Y, Bian Z, et al. GALNT6 dual regulates innate immunity STING signaling and PD-L1 expression to promote immune evasion in pancreatic ductal adenocarcinoma. Cell Signal. 2025;134:111942. [DOI] [PubMed] [Google Scholar]

[CR31] 31.Lin TC, Chen ST, Huang MC et al. GALNT6 expression enhances aggressive phenotypes of ovarian cancer cells by regulating EGFR activity. Oncotarget, 2017. 8. [DOI] [PMC free article] [PubMed]

[CR32] 32.Sheta R, Bachvarova M, Macdonald E et al. The polypeptide GALNT6 displays redundant functions upon suppression of its closest homolog GALNT3 in mediating aberrant O-Glycosylation, associated with ovarian cancer progression. Int J Mol Sci, 2019. 20(9). [DOI] [PMC free article] [PubMed]

[CR33] 33.Banias L, Jung I, Bara T, et al. Immunohistochemical-based molecular subtyping of colorectal carcinoma using Maspin and markers of epithelial-mesenchymal transition. Oncol Lett. 2020;19(2):1487–95. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Integrated multi-omics analysis reveals PTM networks as key regulators of colorectal cancer progression and immune evasion

Guiting Yang

Liu Ji

Chengmei Lv

Chen Zhao

Riliang Ma

Ying Li

Yanyan Hu

Linghui Pan

Abstract

Background

Methods

Results

Conclusion

Supplementary Information

Introduction

Methods and materials

Acquisition of transcriptomic data

Differential expression and functional enrichment analysis methodology

Single-cell RNA sequencing data processing and analysis

Machine learning for key gene screening

SHAP model construction

Immune cell infiltration analysis

Consensus clustering analysis

Mendelian randomization of key genes

Statistical analysis

Result

Differential analysis and GSVA score calculation

Fig. 1.

Differential PTM genes and enrichment analysis

Fig. 2.

Single-cell transcriptome analysis revealed cell type-specific PTM characteristics in colon cancer

Fig. 3.

Identifying diagnostic biomarkers based on machine learning from 36 genetic features

Fig. 4.

Comprehensive SHAP-based evaluation of diagnostic biomarkers in colorectal cancer

Fig. 5.

Single-cell resolution validation of candidate biomarkers in colorectal cancer

Fig. 6.

Comprehensive immune profiling reveals tumor-specific microenvironment alterations and biomarker-immune cell correlations in colorectal cancer

Fig. 7.

Molecular subtyping reveals distinct colorectal cancer heterogeneity patterns based on five-gene signature

Fig. 8.

Mendelian randomization analysis of GALNT6 gene expression and colorectal cancer

Fig. 9.

Discussion

Conclusion

Supplementary Information

Acknowledgements

Author contributions

Funding

Data availability

Declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

Footnotes

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases