Integrative analysis and functional validation identified palmitoylated PHGDH as a therapeutic target for colon adenocarcinoma

Abstract

Background

Colon adenocarcinoma (COAD) has high incidence and mortality with poor prognosis, creating an urgent need for novel prognostic models and therapeutic strategies. Protein palmitoylation, an important post-translational modification, exerts pivotal effects on tumorigenesis; however, its prognostic and mechanistic relevance in COAD remains unknown. This study determined palmitoylation-related prognostic biomarkers and probable therapeutic targets to treat COAD.

Methods

We integrated transcriptomic datasets from the TCGA and identified prognostic genes related to palmitoylation by conducting differential expression analysis, weighted gene correlation network analysis, and Cox proportional hazards modeling using the least absolute shrinkage and selection operator (LASSO) method. Based on these genes, we established a robust risk signature and nomogram that consistently stratified COAD patients into distinct risk groups. Multi-omics profiling, including immune infiltration assessment, drug response prediction, and single-cell mapping, revealed distinct molecular and cellular programs within each risk group. In vitro and in vivo studies were conducted using the representative hub gene PHGDH to validate its function in COAD.

Results

A risk score model comprising three palmitoylation-related genes (PHGDH, LAMA2, and PBXIP1) stratified COAD patients into subgroups with distinct survival, immune infiltration, and drug sensitivity profiles. PHGDH was identified as the most dysregulated candidate. Functional studies revealed that depleting PHGDH significantly repressed cell proliferation and induced apoptosis. Critically, pharmacological inhibition of PHGDH palmitoylation markedly suppressed tumor growth in vitro and in xenograft models, indicating its feasibility as a therapeutic target for COAD.

Conclusion

The palmitoylation-related risk model established in this study demonstrates strong prognostic predictive capability for COAD, offering a promising new clinical stratification tool. Furthermore, PHGDH was validated as a key functional target whose activity is regulated by palmitoylation, presenting a novel therapeutic opportunity for COAD treatment.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13402-026-01207-4.

Introduction

Colon adenocarcinoma (COAD) is a major global health concern, characterized by high incidence and mortality rates [1]. Although remarkable progress has been achieved in multimodal treatment approaches encompassing radiotherapy, targeted therapies, and chemotherapy, a substantial proportion of patients experience treatment failure and poor prognosis [2]. This clinical challenge is partly attributable to the profound intertumoral heterogeneity of COAD, which traditional markers (e.g., microsatellite instability [MSI] and tumor mutational burden [TMB]) fail to fully capture, leading to imprecise risk stratification and suboptimal therapeutic decisions [3, 4]. Therefore, developing novel and precise prognostic models is critical to enhance risk assessment and guide personalized treatment strategies in COAD.

In studies on identifying reliable biomarkers and therapeutic targets, epigenetic and protein posttranslational modifications (PTMs)-based signatures have garnered significant attention [5–7]. Among various PTMs, palmitoylation, as a representative form of covalent lipid modification in eukaryotic cells, was first described in the 1970s [8–10]. S-palmitoylation (also termed S-acylation) is the main form of palmitoylation, which refers to the binding of a palmitoyl group to the cysteine residues of proteins [11, 12]. This reversible and dynamic modification profoundly influences the membrane localization of proteins, protein trafficking, and stability, thereby regulating fundamental cellular processes [12–14]. Dysregulated protein palmitoylation promotes carcinogenesis, tumorigenesis, and tumor progression in various cancers by directly modifying key regulatory factors [15, 16]. Notably, Bu et al. showed that ZDHHC17/ZDHHC24-mediated palmitoylation activates AKT and promotes hepatocellular carcinoma development [17]. ZDHHC20-mediated YTHDF3 palmitoylation in pancreatic cancer promotes tumor progression by stabilizing MYC mRNA [18]. Furthermore, in colorectal cancer, ACOX1 dephosphorylation drives tumor progression by enhancing β-catenin palmitoylation [19]. Beyond its intrinsic oncogenic functions, palmitoylation also exerts profound immunomodulatory effects, with PD-L1 palmitoylation inhibition markedly amplifying T cell-mediated antitumor immunity [20], whereas inhibiting TIM-3 palmitoylation enhances CAR-T and natural killer (NK) cell cytotoxicity [21]. However, research on palmitoylation in COAD primarily focuses on individual regulators or pathways; a systematic view of its molecular landscape and clinical significance is therefore required.

Here, we systematically identified three prognostic palmitoylation-related genes (PRGs) and developed a robust risk-scoring model for COAD that enables precise patient stratification while reliably predicting survival outcomes, immunotherapy responses, and drug sensitivity. Among these genes, integrative analyses indicated PHGDH as the most functionally significant. Subsequent in vitro and in vivo studies demonstrated its tumor-promoting role and established that its oncogenic function depends on palmitoylation. Taken together, this study defines a prognostic model involving PRGs and identifies PHGDH as a palmitoylation-regulated oncogenic driver in COAD, providing a mechanistic basis and clinically relevant framework for improved patient stratification and targeted intervention.

Methods

Acquisition and preprocessing of data

This study utilized data from multiple public repositories: (1) TCGA-COAD (https://portal.gdc.cancer.gov/) as the discovery cohort; (2) GEO cohort GSE39582 (https://www.ncbi.nlm.nih.gov/geo/) as a testing set; and (3) a list of 305 PRGs from GeneCards (https://www.genecards.org/) (Table S1), identified using the terms “S-palmitoylation” and “S-acylation.”

Identification and functional analysis of DEGs

The following two steps were used for data cleaning: (1) genes with zero expression in ≥ 50% of the samples were filtered, and (2) duplicate samples from the same patient were filtered. The “DESeq2” R package was employed to screen differentially expressed genes (DEGs) between COAD tumor tissues and adjacent normal tissues under the following thresholds: adjusted p-value < 0.05; |log2-fold change (FC)| > 1. The “clusterProfiler” package was utilized for performing Gene Ontology (GO) enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis; statistical significance was defined at adjusted p-value < 0.05.

Weighted gene co-expression network analysis (WGCNA)

The TCGA-COAD dataset was evaluated with the “WGCNA” package through the following steps. (1) Data preparation: The expression matrix was normalized, and the sample’s top 25% genes exhibiting the highest level of variance were retained. (2) Sample clustering tree construction: Hierarchical clustering dendrograms were pruned at a height cutoff of 130 to exclude expression outliers. (3) Soft threshold selection: Based on β value = 6 as the optimal parameter, a scale-free topology fit index (R²) of > 0.85 was achieved. (4) Gene dendrogram clustering was performed with the following settings: deepSplit = 2, mergeCutHeight = 0.25, and minimum module size = 30. (5) Module-trait correlation heatmaps: The correlation and significance between modules and traits were calculated, and heatmaps were plotted.

Identification of hub genes and risk model development

The intersection of DEGs, PRGs, and specific WGCNA modules was obtained. From this overlap, prognosis-associated candidate genes were screened using univariate Cox regression analysis (p < 0.05). Then, the “glmnet” package was utilized to perform LASSO regression analysis for developing a prognostic risk model.

Nomogram construction and assessment

A prognostic nomogram was developed using the “rms” R package. Subsequently, the nomogram’s performance was assessed using ROC curves, decision curve analysis (DCA), and calibration curves.

TMB and immune cell infiltration analyses

The maftools package in R was utilized to evaluate somatic mutation data from the TCGA-COAD cohort. The mutational landscapes of the high- and low-risk groups were visualized through waterfall plots, followed by a comparison of TMB between the two groups using boxplots. Immune cell variations between the two groups were assessed by CIBERSORT and single-sample gene set enrichment analysis (ssGSEA). The immunotherapy response of samples was predicted based on TIDE, Dysfunction, MSI, and Exclusion scores. Immune, stromal, and tumor purity scores were derived from the ESTIMATE algorithm.

Drug sensitivity prediction

The gene expression (GDSC2 Expr.rds) and drug response data (GDSC2 Res.rds) provided by the “oncoPredict” package were considered the training set for assessing the drug response results of high- and low-risk group samples, with the significance level at p < 0.05.

Single-cell data preparation and assay

Single-cell RNA-seq (scRNA-seq) data were analyzed from 29 CRC samples across the GSE132465 and GSE144735 datasets by using the Seurat package (v5.3.0). Low-quality cells were excluded according to the following criteria: fewer than 200 or more than 6000 genes detected per cell, or mitochondrial gene content > 20%. After normalization, the top 3000 genes with high variance were identified for downstream analysis by using the FindVariableFeatures function. By employing the RunPCA function, we carried out principal component analysis (PCA) for dimensionality reduction; this was followed by cell clustering and t-SNE visualization. Cell types were annotated with reference to the CellMarker 2.0 database (http://www.bio-bigdata.center/index.html). Module scores for PRG-associated genes (PHGDH, LAMA2, and PBXIP1) for each tumor cell were calculated using the AddModuleScore function. According to the median module score, tumor cells were assigned to PRG-high and PRG-low groups.

Cell culture details and treatment procedures

All cell lines for experiments were procured from the American Type Culture Collection. HCT116 and SW620 cells were grown in RPMI 1640 medium (Gibco™), while the remaining cell lines were cultivated in DMEM (Gibco™); 10% fetal bovine serum (ExCell) was added to both culture media. Cell cultivation was achieved at 37℃ in 5% CO₂. Plasmids and small interfering RNAs (siRNAs) were provided by Shanghai GenePharma Co., Ltd.

Quantitative real-time PCR (qRT-PCR)

RNA isolation was performed using RNAex Pro Reagent (Accurate Biotechnology (Hunan) Co., Ltd.). RNA quality and quantity were validated with a NanoDrop 2000 spectrophotometer (Thermo Scientific). The Evo M-MLV kit (Accurate Biotechnology (Hunan) Co., Ltd.) was utilized for reverse transcription; this was followed by qRT-PCR employing SYBR Green Supermix (Bio-Rad, USA). The primers used were as follows: PHGDH (forward) 5′-TGCAAATCTGCGGAAAGTGC-3′; (reverse) 5′-GATGACATCAGCGGTCACCT-3′. β-actin (forward) 5′-CCTGGCACCCAGCACAAT-3′; (reverse) 5′-GGGCCGGACTCGTCATAC-3′.

Western blotting (WB) assay

Cell lysis was performed in RIPA buffer supplemented with PMSF (1 mM) for 20 min; lysed cells were centrifuged at 12,000 g and 4 °C for 30 min. The supernatant protein concentration was assayed by the bicinchoninic acid (BCA) assay (Beyotime), and equal amounts of proteins were separated by SDS-PAGE (10% gel) and electroblotted on PVDF membranes. After blocking with non-fat milk for 1 h, membranes were incubated with primary antibodies overnight at 4 °C. Following washing three times with 1× TBST, the membranes were further incubated with HRP-bound secondary antibodies at room temperature for 1 h. ECL (Biosharp) was utilized for visualizing protein bands. The antibodies used were as follows: PHGDH (Selleck, 1:1000), β-actin (Proteintech, 1:30000), c-Caspase3 (CST, 1:800), c-PARP (Proteintech, 1:5000), goat anti-rabbit IgG (FDbio, 1:5000), and goat anti-mouse IgG (FDbio, 1:5000).

Cell proliferation assay

Transfected cells were seeded onto 96-well plates (1000 cells/well) and cultured for 1–5 days. After adding CCK-8 reagent (Beyotime), the cells were incubated for 2 h at 37 °C; the OD value at 450 nm was estimated. We carried out a colony formation assay by culturing cells (300 cells/well) on 12-well plates for 7-10 days. After fixing the colonies for 20 min with 4% paraformaldehyde, 0.1% crystal violet solution (Solarbio) was used for staining for 10 min.

Apoptosis analysis

After 48 h of siRNA transfection, apoptosis was detected using the Apoptosis Kit (Beyotime). The cells were resuspended in 200 µL of binding buffer and then stained with 5 µL of Annexin V-FITC and 10 µL of propidium iodide. Flow cytometry (Beckman CytoFLEX) was immediately conducted, and the results were quantified by FlowJo software.

Immunoprecipitation-acyl-biotin exchange (IP-ABE)

Cells were co-cultured with ML-348 or 2-bromopalmitoleate (2-BP) (MedChemExpress) for 48 h. Cell lysates were collected, and immunoprecipitation was performed by incubating proteins with magnetic beads and anti-PHGDH antibody overnight. The precipitate was treated with N-acetyl-L-cysteine (20 mM, Sigma) at 4 °C for 6–8 h. The resulting mixture was equally divided into two groups (HAM[-] and HAM[+]) and incubated at room temperature with 0.8 M HAM (Sigma) for 1 h. The precipitates were washed and labeled using 5 µM biotin-HPDP (Macklin) for 1 h. Proteins were determined by WB assay using HRP-streptavidin (1:2000, Beyotime) to detect palmitoylated PHGDH.

Protein half-life and stability assay

After treatment with cycloheximide (CHX, 15 µg/mL; Selleck), the cells were harvested at the indicated time points for WB assay to monitor temporal changes in protein levels. The intracellular protein degradation pathways were clarified by treating cells with proteasome inhibitors (MG132, 5 µM; Selleck) and lysosomal inhibitors (chloroquine [CQ], 10 µM; TargetMol).

Animal experiments

We obtained four-week-old BALB/c nude mice from Guangdong Provincial Medical Laboratory Animal Center. Mice were inoculated with 3 × 10⁶ DLD-1 cells subcutaneously in the right dorsal flank. When the tumor volume reached 40 mm³, 30 mg/kg of 2-BP was intraperitoneally administered every alternate day. The animals were sacrificed after 3 weeks. After being fixed in 4% paraformaldehyde, the harvested tumor tissues were paraffin-embedded and processed for immunohistochemical staining with anti-Ki67 antibodies (1:200 dilution; Abmart).

Statistical analysis

GraphPad Prism software (v10.1.1) and the R program (v4.3.1) were utilized for statistical analysis. Significant differences between two groups and among three or more groups were assessed by an unpaired two-tailed Student’s t-test and ANOVA, respectively. Statistical significance was defined at p < 0.05 (*p < 0.05, **p < 0.01, ***p < 0.001, ns, nonsignificant).

Results

COAD-associated candidate palmitoylation gene screening and identification

From the TCGA-COAD dataset, 9254 DEGs were screened, including 3736 and 5518 downregulated and upregulated genes, respectively (Table S2). WGCNA was then performed to identify COAD-associated modules, with a soft-threshold power of 6 to attain scale-free topology (Fig. 1A). As shown in the cluster dendrogram, the network revealed 22 distinct co-expression modules (Fig. 1B). A module-trait relationship heatmap was subsequently constructed. Among them, MEbrown (r = -0.85, P = 3 × 10^− 126) and MEblue (r = -0.51, P = 1 × 10^− 31) emerged as the top two modules showing the strongest negative correlations with COAD, while MEpink (r = 0.63, P = 5 × 10^− 52) and MEred (r = 0.56, P = 4 × 10^− 39) exhibited the most significant positive associations (Fig. 1C). We further assessed the relationship between intramodular connectivity and phenotypic relevance by correlating module membership with gene significance. Scatterplot analyses demonstrated strong positive correlations within all four key modules (p < 0.001), indicating that highly connected genes were also phenotypically relevant (Fig. 1D, Table S3). Based on these module-level insights, we integrated DEGs, the four WGCNA modules, and predefined PRGs, ultimately identifying 28 overlapping candidate genes (Fig. 1E, Table S4). Among these, three genes (PHGDH, LAMA2, and PBXIP1) were significantly associated with COAD prognosis in a univariate Cox regression analysis (Fig. 1F). Expression validation confirmed remarkable upregulation of PHGDH in tumor tissues and substantial downregulation of LAMA2 and PBXIP1 compared to those in normal controls (Fig. 1G). Expanding our analysis to a pan-cancer context using TCGA and GTEx data demonstrated distinct expression patterns: PHGDH and PBXIP1 displayed cancer type-specific dysregulation, with significant elevation in some malignancies but suppression in other tumors. In contrast, LAMA2 was predominantly downregulated across most cancer types examined (Fig. S1A-C). Consistent with their prognostic value, Kaplan-Meier survival analysis showed that patient survival correlated remarkably with the expression levels of all three genes (Fig. 1H).

Fig. 1.

Identification and validation of prognostic hub genes in COAD. (A) Scale-free topology fitting and mean connectivity trends (optimal soft-threshold: R² = 0.85,  soft power = 6). (B) Gene module identification and clustering dendrogram. (C) Heatmap of module-trait associations. (D) Scatterplots of GS versus MM for key modules. (E) Venn diagram of COAD DEGs, WGCNA modules, and PRGs. (F) Identification of prognostic candidate genes via univariable Cox regression analysis (p < 0.05). (G) Boxplots of LAMA2, PBXIP1, and PHGDH expression levels in COAD tissues: Tumor vs. Normal. (H) Kaplan-Meier survival analysis for patients stratified by the expression of PHGDH, LAMA2, and PBXIP1

GSEA revealed distinct pathway associations for the three hub genes in COAD. Samples with high PHGDH expression showed enrichment in ribosome biogenesis and tRNA metabolism, but negative enrichment in multiple immune response pathways (Fig. S1D). Conversely, high LAMA2 expression was associated with enrichment in extracellular matrix structural constituents, alongside negative enrichment in mitochondrial respiratory chain complex assembly and ATP synthesis (Fig. S1E). Similarly, samples expressing high levels of PBXIP1 were enriched for the collagen-containing extracellular matrix and homophilic cell adhesion but exhibited negative enrichment for oxidative phosphorylation and the structural constituent of the ribosome (Fig. S1F).

Prognostic model construction based on palmitoylation genes

Based on the screened genes, a prognostic model was constructed by LASSO regression. The coefficient trajectories for the three genes are shown in Fig. S2A, and the optimal model identified by cross-validation, along with its corresponding lambda-specific confidence intervals, is displayed in Fig. S2B. The equation for estimating the risk score was as follows: (0.2559 × LAMA2) + (0.3876 × PBXIP1) + (0.2687 × PHGDH). By utilizing the median risk score as the cutoff, TCGA-COAD patients were assigned to high- and low-risk groups; higher risk scores clearly corresponded to increased mortality as reflected by survival outcome and risk score distribution (Fig. 2A-B). Subsequently, Kaplan-Meier survival analysis was carried out to quantify the survival disparity between these subgroups; the results revealed substantially shorter overall survival (OS) for patients in the high-risk group (p < 0.001, Fig. 2C). The model’s predictive accuracy was then determined through time-dependent ROC analysis, yielding area under the curve (AUC) values of 0.653, 0.652, and 0.663 for 1-year, 2-year, and 3-year patient survival, respectively (Fig. 2D). To assess the model’s generalizability and robustness, the prognostic model was validated with the external GSE39582 dataset, the largest publicly available cohort for COAD. Consistent with findings from the TCGA training set, higher risk scores correlated with increased mortality and a significantly poorer prognosis (Fig. 2E-F). The AUC values were 0.785, 0.761, and 0.661 for predicting 1-year, 2-year, and 3-year patient survival in the validation set, respectively (Fig. 2G). Notably, higher expression levels of all 3 genes were noted in high-risk patients (Fig. 2H). Collectively, these observations demonstrate that this three-PRG-based prognostic model could be used accurately for predicting COAD outcomes.

Fig. 2.

Validation of the prognostic risk model in COAD. (A-D) Analyses in the TCGA-COAD cohort: (A) Risk score distribution stratified by survival status. (B) Correlation plot of risk factors with patient risk scores and survival information. (C) Kaplan-Meier survival curves for OS in patients by risk stratification. (D) Time-dependent ROC curves assessing the model’s predictive accuracy at 1, 2, and 3 years. (E-G) External validation in the GSE39582 cohort: risk score distribution plot, Kaplan-Meier survival curve, and ROC curve. (H) Expression of the three-gene signature across risk groups in the TCGA and GSE39582 cohorts

Prognostic nomogram development and validation

By utilizing the key clinical parameters and PRG risk score, a comprehensive nomogram was built to improve prognostic prediction. Univariable and multivariable Cox regression analyses confirmed age, risk score, and clinical stage as independent predictors (p < 0.01; Fig. S2C-D), and these predictors were utilized for constructing the nomogram (Fig. 3A). The model demonstrated accurate calibration (Fig. 3B) and yielded a superior clinical net benefit in DCA (Fig. 3E). Time-dependent ROC analysis demonstrated high predictive accuracy of the nomogram, with AUC values of 0.82, 0.78, and 0.81 for 1-year, 2-year, and 3-year patient survival, respectively (Fig. 3C). According to the nomogram, high-risk group patients exhibited remarkably shorter OS (p < 0.001; Fig. 3D). The model’s robust performance was further confirmed in the external GSE39582 cohort, where the nomogram maintained a 3-year AUC of 0.709, was well-calibrated, and showed significant survival stratification (p < 0.001; Fig. 3F-H).

Fig. 3.

Construction and Validation of the Nomogram in COAD. (A-E) Analyses based on the TCGA-COAD cohort: (A) A nomogram incorporating the prognostic risk score and clinical features. (B) Calibration curves validating the prediction accuracy of the nomogram. (C) Time-dependent ROC curves. (D) Kaplan-Meier survival curves by nomogram-derived risk groups. (E) Decision curve analysis (DCA) of the nomogram. (F-H) External validation of the calibration, ROC curves, and Kaplan-Meier survival curves in the GSE39582 cohort

Immune landscape and drug sensitivity differences between risk groups

The risk model’s association with genomic features in the TCGA-COAD cohort was determined; the results revealed a highly mutated landscape across both risk groups (98.01% vs. 95.31%), with the top 20 mutated genes for each group displayed in Fig. 4A-B. Although the TMB did not significantly differ between the groups (Fig. 4C), an integrated analysis demonstrated that a high-risk score combined with high TMB identified a patient subset with the most unfavorable prognosis, confirming their complementary prognostic value (Fig. 4D). Next, we profiled the association of the immune microenvironment with the risk model. First, the correlation and composition of 22 types of immune cell subsets in the TCGA-COAD cohort were summarized (Fig. S3A-B). We then defined the immune landscape using CIBERSORT and ssGSEA. According to CIBERSORT analysis, the low-risk group demonstrated enrichment in activated CD4⁺ memory and CD8⁺ T cells, whereas the high-risk group showed a higher M0 macrophage abundance (Fig. 4E). Results from ssGSEA further supported an active antitumor immune state in low-risk patients, characterized by elevated levels of activated CD4⁺ and CD8⁺ T cell . However, high-risk patients showed increased proportions of various innate immune and memory cell populations (Fig. 4F). Beyond immune cell infiltration, we assessed the stromal component of the tumor microenvironment (TME), which showed significant enrichment in high-risk patients (Fig. 4G). Subsequently, by applying the TIDE algorithm, the high-risk group was predicted to have a high potential for immune evasion, evidenced by high TIDE, Exclusion, and Dysfunction scores and lower MSI scores (Fig. 4H). Complementing this functional prediction, checkpoint gene expression also differed: CD47 was higher in low-risk patients, while MYD1 and VTCN1 were upregulated in their high-risk counterparts (Fig. S3C).

Fig. 4.

Genomic and immunogenomic characteristics associated with the prognostic risk model. (A-B) Waterfall plots of somatic mutation landscape across risk subgroups. (C) Tumor mutational burden (TMB) in high- and low-risk subgroups. (D) Kaplan-Meier survival analysis stratified by combined TMB and risk status. (E-F) Immune cell infiltration analysis based on CIBERSORT and ssGSEA algorithms. (G) Stromal and immune scores in the tumor microenvironment across risk groups. (H) Prediction of immunotherapy response using the TIDE algorithm

Beyond predicting immunotherapy response, we further explored effective chemotherapeutic and targeted agents for the distinct risk subgroups. The OncoPredict algorithm was applied to GDSC database data for comparing the two risk groups’ sensitivity to 198 drugs, and 86 drugs exhibiting remarkable differences in sensitivity were identified (p < 0.05; Table S5). The high-risk group was significantly more sensitive to conventional chemotherapies (e.g., 5-FU, oxaliplatin, cyclophosphamide, and gemcitabine; Fig. S3D), whereas the low-risk group exhibited an increased responsiveness to targeted inhibitors, including doramapimod, alpelisib, a JAK inhibitor, and JQ1 (Fig. S3E). These findings emphasize the risk model’s potential to guide personalized treatment selection in COAD.

Single-cell analysis based on PRG-associated gene signature

We integrated scRNA-seq data from 29 CRC samples (GSE132465 and GSE144735). After quality control and normalization, t-SNE was used to resolve and visualize seven major cell populations within the TME (Fig. 5A). Cell type annotation was carried out with the established marker genes as follows: epithelial cells (EPCAM, KRT8), myeloid cells (LYZ, CD68), T cells (CD3D, CD3E), plasma cells (JCHAIN, MZB1), stromal cells (DCN, COL1A1), B cells (CD79A, MS4A1), and mast cells (KIT, CPA3) (Fig. 5B). As illustrated in Fig. 5C, PHGDH was predominantly expressed in epithelial and myeloid cells, and PBXIP1 showed elevated expression in both T cells and myeloid cells, whereas LAMA2 exhibited low expression in all cells except myeloid cells. We then employed the AddModuleScore algorithm for determining the risk score for individual cell types by using these three genes; by considering the median score, cells were assigned to two groups: high-risk and low-risk. Determination of risk group distribution across samples showed considerable heterogeneity in cellular risk composition. Notably, mast cells and stromal cells were substantially enriched for high-risk cells; however, low-risk populations showed a predominance of B cells and plasma cells (Fig. 5D). Subsequently, GO and KEGG enrichment analyses for the DEGs between high- and low-risk groups were conducted. GO analysis showed that the DEGs were markedly enriched in biological processes associated with immune response regulation, cell adhesion and migration, and tissue repair (Fig. 5E). KEGG pathway enrichment analysis further showed enrichment in multiple immune-related and infectious disease pathways (Fig. 5F), collectively underscoring a strong immunological basis underlying the differences between the two risk groups.

Fig. 5.

Analysis of hub genes using single-cell data. (A) t-SNE visualization of colorectal cancer samples from the GSE132465 and GSE144735 datasets. (B) Dot plot of key marker gene expression across cell types. (C) Dot plot and t-SNE mapping the expression and distribution of PHGDH, LAMA2, and PBXIP1. (D) Proportions of major cell types stratified by risk group. (E-F) Top 10 significantly enriched terms from (E) GO and (F) KEGG analyses

Silencing PHGDH suppresses cell proliferation and induces apoptosis

Building on the above findings, PHGDH is centrally involved in coordinating core metabolic and immune pathways, and its expression is predominantly enriched within epithelial tumor cells, thus identifying it as a cell-intrinsic key driver of COAD malignancy. However, despite its documented oncogenic role in CRC as the rate-limiting enzyme catalyzing serine biosynthesis [22–24], a critical question remains unanswered: whether and how its function is precisely regulated by palmitoylation. Therefore, we focused on experimentally validating PHGDH. Immunohistochemical data from the Human Protein Atlas database confirmed stronger PHGDH staining in COAD tissues (Fig. S4A). Using 12 paired clinical samples, we further validated PHGDH upregulation at protein and mRNA expression levels by WB assay and qRT-PCR (Fig. S4B-C). Consistent with clinical validation, PHGDH expression was markedly higher in six COAD cell lines than in NCM-460 (normal intestinal epithelial cell line) (Fig. S4D-E). After confirming PHGDH overexpression in COAD, we assessed its functional role by transfecting SW480 and LOVO cells with PHGDH-targeting or control siRNA, and effective knockdown was verified by qRT-PCR and WB assay (Fig. 6A-B). Both the CCK-8 assay (p < 0.001; Fig. 6C) and the colony formation assay (Fig. 6D) demonstrated that cell proliferation was markedly inhibited by PHGDH silencing. Flow cytometry revealed that PHGDH knockdown significantly induced apoptosis (Fig. 6E). Furthermore, Western blot analysis showed that PHGDH knockdown markedly increased the protein levels of cleaved PARP and cleaved Caspase-3 (Fig. S5A), which are established apoptotic markers. These results confirmed that PHGDH silencing triggers apoptosis in COAD cells, suggesting that PHGDH promotes COAD cell survival, at least in part, by suppressing apoptosis.

Fig. 6.

Functional validation of PHGDH in COAD. (A-B) Detection of PHGDH knockdown efficiency in SW480 and LOVO cells. (C) CCK-8 assay to evaluate the effect of PHGDH knockdown on cell proliferation. (D) Colony formation assay of cells after PHGDH knockdown. (E) Flow cytometry analysis of apoptosis induction by PHGDH silencing

Palmitoylation regulates PHGDH stability through proteasomal degradation

To investigate whether PHGDH is regulated by palmitoylation, we exposed cells to the pan-palmitoylation inhibitor 2-BP and the depalmitoylation inhibitor ML348. Treatment with 2-BP caused a concentration-dependent decrease in PHGDH protein levels (Fig. 7A); in contrast, ML348 increased PHGDH protein abundance (Fig. 7B). Notably, these changes in protein expression occurred without any corresponding alterations in PHGDH mRNA levels (Fig. S5B-C), indicating post-transcriptional regulation. Given that protein stability often depends on palmitoylation, we performed cycloheximide (CHX) chase assays. 2-BP substantially shortened the half-life of PHGDH, indicating that palmitoylation inhibition accelerates its degradation (Fig. 7C). We then sought to identify the degradation pathway. In eukaryotic cells, the two primary degradation pathways are autophagy and the ubiquitin-proteasome system [25, 26]. Co-treatment with the proteasome inhibitor MG132, but not the lysosome inhibitor CQ, rescued 2-BP-induced PHGDH loss, demonstrating that the proteasomal pathway mediates this degradation (Fig. 7D). Finally, to directly confirm the modification, we employed an IP-ABE assay, which unequivocally demonstrated that PHGDH is palmitoylated. As expected, the palmitoylation level was reduced by 2-BP and enhanced by ML348 (Fig. 7E).

Fig. 7.

Palmitoylation regulates PHGDH protein stability. (A-B) Assessment of PHGDH protein levels after treatment with 2-BP or ML348. (C) PHGDH protein half-life was detected after treatment with 2-BP. (D) Effects of MG132 or CQ on 2-BP-induced PHGDH degradation. (E) Evaluation of PHGDH palmitoylation levels by the Acyl-Biotin Exchange assay

Inhibiting PHGDH palmitoylation attenuates tumor growth in COAD

Given that palmitoylation regulates PHGDH protein abundance and stability, we next explored whether this regulation affects PHGDH-driven phenotypes. First, we knocked down PHGDH in SW480 and LOVO cells and treated them with ML348. As shown in Fig. S5D-E, PHGDH knockdown significantly suppressed cell proliferation and colony formation, while ML348 treatment partially reversed these inhibitory effects. Consistently, western blot analysis revealed that ML348 partially attenuated apoptosis triggered by PHGDH knockdown (Fig. S5F). Next, we performed a complementary test by overexpressing PHGDH in RKO and DLD1 cells (Fig. 8A-B), which have relatively low endogenous PHGDH expression. Functional assays revealed that PHGDH overexpression significantly enhanced cell proliferation, an effect that was markedly attenuated by 2-BP treatment (Fig. 8C-D). Finally, we established xenograft tumors using DLD1 cells with stable PHGDH overexpression. Tumors derived from PHGDH-overexpressing cells exhibited markedly accelerated growth, as reflected by an increased tumor size, a larger tumor volume on the growth curve, and a higher tumor weight in the endpoint measurement. 2-BP treatment markedly reduced tumor enlargement in the gross images and suppressed both tumor growth kinetics and final tumor mass (Fig. 8E). Immunohistochemistry further confirmed that 2-BP lowered Ki-67 levels induced by PHGDH overexpression (Fig. 8F). Taken together, these findings indicate that PHGDH-driven tumor growth in COAD depends on palmitoylation, both in cultured cells and in vivo.

Fig. 8.

The oncogenic function of PHGDH is impaired by the palmitoylation inhibitor 2-BP. (A-B) Detection of PHGDH overexpression efficiency in RKO and DLD1 cell lines. (C-D) Cell proliferation was assessed using CCK-8 and colony formation assays. (E) Analysis of the DLD-1 subcutaneous tumor model showing the experimental scheme, excised tumors, growth curves, and final tumor weights. (F) Immunohistochemical results of Ki-67 in xenograft tumors

Discussion

Palmitoylation, a major reversible PTM, is a critical regulator in oncogenesis and tumor immunity. Although there are a few reports on palmitoylation modifications in COAD, its precise role in this malignancy remains poorly defined. Here, we introduced a novel risk model utilizing palmitoylation-associated genes (PHGDH, PBXIP1, and LAMA2), which effectively assigns COAD patients to high-risk and low-risk subgroups with unique clinical outcomes. Specifically, high-risk subgroup patients had markedly shorter OS, a conclusion reinforced by consistent results from the external validation cohort GSE39582, confirming the model’s reliability as an independent prognostic factor. We subsequently integrated the independent prognostic factors (age, pathological stage, and the palmitoylation-based risk score) into a clinically practical nomogram. Evaluation of the nomogram revealed not only a superior net benefit over individual factors but also a high accuracy in predicting survival probabilities. This confirms the model’s clinical utility in guiding personalized treatment strategies for COAD patients.

Emerging evidence indicates that palmitoylation critically orchestrates tumor immune microenvironment remodeling and therapeutic responses through the following mechanisms: (i) modulating immunosuppressive molecules, (ii) regulating T-cell functionality, and (iii) facilitating tumor immune evasion [27, 28]. Motivated by these mechanistic insights, we further investigated immune characteristics within the risk-stratified cohorts. The results demonstrated that low-risk group patients showed higher levels of infiltration of activated CD4 memory and activated CD8⁺ T cells as well as CD56dim NK cells—critical mediators of antitumor immune response [29, 30]. Conversely, high-risk patients showed increased infiltration of immunosuppressive cell populations. Application of the TIDE algorithm to these groups demonstrated a significantly enhanced immune escape potential in high-risk patients, as evidenced by elevated TIDE, Exclusion, and Dysfunction scores alongside lower MSI. This profile collectively predicts a diminished likelihood of response to immune checkpoint inhibitor therapy [31]. Given this predicted refractoriness to immunotherapy, we utilized the GDSC database to identify potential chemotherapeutic alternatives. Indeed, high-risk patients showed greater sensitivity to core first-line chemotherapeutic agents (5-FU and oxaliplatin), thereby providing a rational basis for tailoring their treatment regimens.

Among the three candidate genes, phosphoglycerate dehydrogenase (PHGDH), a rate-limiting enzyme catalyzing de novo serine synthesis, has a key function in maintaining cellular metabolic homeostasis and proliferation, which are closely linked with tumor growth and occurrence [32, 33]. According to Rossi et al., PHGDH promotes breast cancer cell metastasis [34]. Shu et al. found that PHGDH promotes mitochondrial translation in hepatocellular carcinoma to drive cancer cell proliferation [35]. Additionally, PHGDH interacts with eIF4A1 and eIF4E (translation initiation factors) to facilitate pancreatic cancer progression [36]. PHGDH’s role in CRC tumorigenesis is being examined, with potential importance for progression and metastasis [37, 38]. In this study, the analysis of clinical samples revealed significantly higher PHGDH expression in COAD than in matched normal tissues. Functional characterization identified a critical role of PHGDH in COAD, showing that its depletion inhibited cellular proliferation and promoted apoptosis.

Previous studies have established that ubiquitination [39, 40], methylation [41, 42], and phosphorylation of PHGDH contribute significantly to its roles in tumorigenesis and cancer progression [43, 44], whereas the impact of palmitoylation on its function in COAD has remained largely unexplored. Palmitoylation represents a biochemically distinct class of modification from those previously described for PHGDH. While phosphorylation mediates acute signaling responses and conformational changes that affect enzymatic activity, and ubiquitination marks proteins for proteasomal degradation, palmitoylation operates through a different mechanism. As a reversible lipid modification, it dynamically controls membrane trafficking, subcellular localization, and protein-protein interactions. Our data identify S-palmitoylation as a pivotal post-translational mechanism that governs the abundance and stability of PHGDH, and more importantly, its oncogenic activity in COAD. This conclusion is strengthened by two pharmacologically distinct inhibitors: inhibition of palmitoylation with 2-BP attenuates the tumor-promoting effects of PHGDH overexpression, but conversely, enhancing palmitoylation by inhibiting depalmitoylation with ML348 could partially rescue the proliferation defect. Together, these findings identify PHGDH palmitoylation as a functionally critical modification and offer a rationale for therapeutic intervention aimed at disrupting this regulatory mechanism in COAD.

This study still has some limitations. First, the reliance on public databases may cause selection bias and warrants further independent clinical validation. Second, our immune-related findings are solely computational. Experimental studies are needed to validate these results and explore the underlying mechanisms. Third, in vivo validation in PHGDH-high COAD models (e.g., SW480-derived xenografts) is still required to further confirm the anti-tumor efficacy of targeting PHGDH palmitoylation. Finally, further identification of the specific ZDHHC palmitoyltransferase(s) responsible for PHGDH palmitoylation would help to refine our understanding of this regulatory mechanism and to facilitate the development of more targeted therapeutic strategies.

Conclusion

Here, we constructed a novel prognostic model for COAD by using three critical PRGs, which provided a valuable biomarker for risk stratification. Beyond this prognostic utility, we mechanistically defined the role of PHGDH (a key gene in model development), demonstrating that its oncogenic activity is critically regulated by palmitoylation. This discovery provides a novel mechanistic basis for PHGDH’s function and suggests new avenues for therapeutic intervention.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

YXS: conceptualization of the study, formal analysis, study methodology, software, and manuscript writing – original draft. LLL and DJL: visualization, software, validation, and data curation. MQZ: study methodology and software. KL and PW: data curation and validation. KQC and AML: supervision, funding acquisition, and writing – review and editing. ZHY: project administration, supervision, and writing – review and editing. All authors have approved the final manuscript version.

Funding

This study was financially aided by the National Natural Science Foundation of China (Grant No. 82373165), Guangzhou Science and Technology Program Jointly Funded by Municipal Schools and Institutes (Grant No. 2023A03J0338), Guangdong Basic and Applied Basic Research Foundation (Grant No. 2024A1515012854), and Plan on Enhancing Scientific Research in GMU (2023, No. 147).

Data availability

The original datasets of the present study are publicly accessible from TCGA and GEO repositories. The corresponding author can provide the analysis code upon reasonable request.

Declarations

Ethical approval

Human subjects: The present study was conducted by following the Declaration of Helsinki guidelines. The Nanfang Hospital ethics committee at Southern Medical University (Guangzhou, China) permitted tissue sample utilization. All participants provided written informed consent. Animals: The First Affiliated Hospital ethics committee at Guangzhou Medical University (Guangzhou, China) approved all animal experiments. Experimental work was carried out according to the institutional guidelines and local legislation.

Competing interests

The authors declare no competing interests.