Deciphering oxidative stress-related heterogeneity and developing a prognostic signature for colorectal cancer

Abstract

While recent studies have highlighted oxidative stress (OS) as a pivotal factor influencing tumor dynamics, its specific interactions within the tumor microenvironment (TME) of colorectal cancer (CRC) remain elusive. This study seeks to unveil the impact of OS on the CRC TME and to develop an advanced OS-related risk signature (OSRRS) model. We analyzed OS-related pathway activities using both single-cell and bulk RNA-seq data. An unsupervised clustering algorithm was utilized to identify OS-related subtypes. Based on genes associated with the OS pathways, we constructed an OSRRS model employing the LASSO Cox analysis. For validation, we employed quantitative real-time polymerase chain reaction (qRT-PCR) coupled with immunohistochemical (IHC) analyses on tissue microarrays (TMA) to confirm the expression of the identified gene. Examination of the single-cell RNA-seq GSE132465 dataset revealed a universal elevation in OS-associated pathway activities within malignant cells. By integrating this with the bulk RNA-seq TCGA-CRC dataset, we identified two unique OS-specific clusters. This led to the establishment of a 12-gene OSRRS using the LASSO Cox method. The robustness of our model was further verified using the GSE39582 and GSE17538 cohorts. Notably, increased expression of the UCN gene was observed in CRC specimens, as confirmed by qRT-PCR and IHC assays on TMA. In this research, we delineated two distinct subtypes of CRC associated with OS. The developed OSRRS holds promise as a candidate prognostic and stratification tool for CRC management. Collectively, these results shed light on the intricate role of OS in CRC pathology.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-46560-4.

Introduction

Colorectal cancer (CRC) is one of the leading causes of cancer-related mortalities, accounting for over 900,000 deaths in 2020¹. It is estimated that by 2040, there will be approximately 3.2 million new cases and 1.6 million deaths, with over 80% of these cases predicted to occur in countries with higher human development index². While the mechanisms driving CRC development remain incompletely understood, several risk factors have been identified, which include colon polyps, age, alcohol intake, red meat consumption, smoking, obesity, and presence of family history³. Moreover, carcinogenesis of CRC is a multifaceted and spontaneous pathogenic process, with oxidative stress (OS) playing a significant role^4–6.

OS, characterized by a redox imbalance that favors oxidative burden, is a critical biological process in tumor progression and is increasingly recognized as a hallmark of various cancers^7,8. The onset of numerous diseases involves a myriad of cellular and molecular changes triggered by a spectrum of external and internal stimuli, among which reactive oxygen species (ROS), being a source of oxidative stress, have a crucial role⁹. Furthermore, the effects of ROS are multifaceted, functioning as a double-edged sword in carcinogenesis^10–12. Elevated ROS levels may induce oxidative impairment to lipids, proteins and DNA, thereby becoming detrimental to cells. However, at lower levels, ROS may serve as cellular signaling messengers, regulating a range of cellular functions including stress response, proliferation, differentiation, and gene expression¹³. Essentially, the balance of ROS levels dictates the intensity of OS, impacting either the cellular compromise or survival-associated functions.

Genetic mutations and DNA damage induced by ROS substantially influence the development of various cancers, including CRC¹⁴. Moreover, OS triggers a cascade of reactions including inflammation of the intestinal mucosa, genetic susceptibility, and modifications in the intestinal immune response—factors that are fundamentally associated with the development of CRC^15,16. OS is also linked to another form of mutation observed in CRC, namely microsatellite instability (MSI), which leads to defective DNA repair mechanisms during replication. Studies suggest that OS impairs DNA repair systems, resulting in MSI-Low and, consequently, the development of CRC in patients with ulcerative colitis^17,18. Thus, detecting novel biomarkers remains to be a crucial yet unaddressed necessity for the effective treatment of CRC.

Currently, the field of cancer biology has evolved from a perspective focused on cancer cells alone to a more holistic approach that views cancer cells as constituents of tumor microenvironment (TME)¹⁹. This transition is characterized by cancer cells exhibiting elevated OS levels due to oncogenic transformations, which include genetic and metabolic changes as well as modifications in the TME²⁰. Notably, the dynamic equilibrium of OS not only regulates intricate cell signaling processes in malignant cells but also impacts other TME components¹³. In the context of cancer therapeutics, the integration of ROS-modulating agents and immunotherapy is gaining prominence as a potent approach. However, no previous research has investigated the role of OS-related pathways in the TME, a crucial area to elucidate for developing effective CRC treatment strategies.

Here, we evaluated the activity of OS-related pathways at both cellular and bulk levels. Furthermore, we employed the genes related to OS pathways for classification and developed an OS-related risk signature (OSRRS).

Materials and methods

Data acquisition

We accessed the (Gene Expression Omnibus) GEO database (https://www.ncbi.nlm.nih.gov/geo/) to download the single-cell RNA sequencing (scRNA-seq) dataset (GSE132465; n = 23) and bulk RNA sequencing datasets (GSE39582, n = 585; GSE17538, n = 244). Moreover, we obtained the TCGA-CRC dataset from The Cancer Genome Atlas (TCGA, n = 612) repository (https://portal.gdc.cancer.gov/). The somatic mutation information was also downloaded from TCGA repository. Furthermore, we acquired 10 OS-related pathways and the pertinent gene sets from MSigDB database (https://www.gsea-msigdb.org/gsea/msigdb).

ScRNA-seq data processing

We utilized the Seurat package (v4.2.1) to convert scRNA-seq data into the Seurat object and filtered out low-quality cells with ≥ 5% mitochondrial genes or ≤ 200 detected genes²¹. Then, the NormalizeData function was invoked to standardize the data, while the FindVariableFeatures function was designated to identify 2,000 highly variable genes. Following data scaling with the ScaleData function, we employed RunPCA function to reduce the dimensionality and RunUMAP function for enhanced visual representation. In subsequent steps, we employed the FindClusters function (with a resolution parameter set to 0.5) to determine the clusters. To identify marker genes specific to each cluster, we utilized the FindAllMarkers function. Cell type annotations for these clusters were then derived by referencing the CellMarker 2.0 database (http://bio-bigdata.hrbmu.edu.cn/CellMarker/), which enabled us to assign biological identities based on established marker gene signatures. To validate the annotation accuracy and functional prominence of each cell type, we applied the RunDEtest function to identify marker genes and used the FeatureHeatmap function for heatmap elucidation. To evaluate the activity of OS-related pathways within the TME, the escape package was employed to perform single-sample gene set enrichment analysis (ssGSEA), illustrating in a heatmap that showcased the pathway activities across cell types.

Pathway analysis in bulk RNA-seq

To compare the activities of OS-related pathways between tumor and normal tissues, we utilized the ssGSEA algorithm to calculate the activity of 10 OS-related pathways in TCGA-CRC samples and depicted the differences in boxplots²².

Consensus clustering

To refine the molecular stratification of CRC with heightened precision, we used the ConsensusClusterPlus package to divide TCGA-CRC samples into two clusters based on OS pathway-related genes²³. We compared the survival differences between these clusters using Kaplan–Meier (K–M) analysis. Moreover, we visualized the distribution of prognostic OS pathway-related genes and clinical information between the two clusters in a heatmap using the pheatmap package.

Functional enrichment analysis

We conducted gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) analyses on the differentially expressed genes (DEGs) between the two distinguished clusters to examine the functional enrichment status^24–27. Furthermore, we employed gene set enrichment analysis (GSEA) algorithm on the C2 cluster to gain further insight into its critical functions. All the aforementioned functional enrichment analyses were executed using the clusterProfiler package²⁸. We also employed the Gene Set Variation Analysis (GSVA) algorithm, using the GSVA package, to delineate the contrasting functional dynamics between the two clusters²².

Tumor mutation burden analysis

To elucidate the mutational difference between the two clusters, we utilized the maftools package to compute the tumor mutation burden (TMB) for each TCGA-CRC sample and portrayed the TMB landscape with a waterfall plot²⁹.

Immune landscape analysis

To further investigate the immune landscape, we deployed the ssGSEA algorithm to compare the immune infiltration ratios of 23 distinct immune cells across the two clusters. Then, we calculated the Estimation of STromal and Immune cells in MAlignant Tumour tissues using Expression data (ESTIMATE) score of the two clusters using ESTIMATE package. Furthermore, to estimate the potential responsiveness to immunotherapy, we compared expressions of eight major immune checkpoint inhibitors (ICIs)-related genes between the two clusters.

Construction of OSRRS

Utilizing the TCGA-CRC dataset (training cohort), we incorporated OS pathway-related genes into univariate Cox and LASSO Cox analyses to screen for genes that could construct the OSRRS³⁰. The risk score of OSRRS was calculated:

To authenticate the potency and reliability of our OSRRS, we included the GSE39582 and GSE17538 datasets, designating them as validation cohorts.

Nomogram

TCGA-CRC samples were stratified into a high-risk group (HRG) and a low-risk group (LRG) based on the median risk value. We compared the risk score across different sub-classifications of clinical features. Subsequently, we developed a nomogram combining both the risk score and clinical features identified by Cox analysis. Moreover, we utilized calibration and receiver operating characteristic (ROC) curves to assess the efficacy and precision of our nomogram.

Functional, mutational and immune analyses between HRG and LRG

To explore the functional disparities between HRG and LRG, we conducted GSVA analysis and highlighted the most significant pathways in a heatmap. Then, we evaluated the gene mutation frequency in both HRG and LRG and integrated the risk score with TMB to estimate survival probability in CRC patients. In a further exploration, we executed (tumor immune dysfunction and exclusion) TIDE analysis to estimate the predisposition towards ICIs therapeutic responsiveness^31,32. We also compared the expression of ICIs-related genes across HRG and LRG. Lastly, we deployed several algorithms (TIMER, EPIC, MCPCounter, quanTIseq) to investigate the correlation between risk score and immune cell infiltration.

Quantitative real-time PCR

From the Sixth Affiliated Hospital of Sun Yat-Sen University, we sourced 15 paired samples from CRC tumors and their adjacent normal tissues to probe the expression profiles of UCN via quantitative real-time polymerase chain reaction (qRT-PCR). The acquisition of patient samples was sanctioned by the Ethics Committee of the Sixth Affiliated Hospital of Sun Yat-Sen University (no. 2022ZSLYEC-143). This study complies with the guidelines set forth in Declaration of Helsinki. Written informed consent was obtained from all participants. The qRT-PCR analysis employed the EZBioscience® PCR system (EZBioscience, Roseville, USA), strictly following the provided protocol. Utilizing the 2^(−∆∆Ct) approach and considering GAPDH as the reference gene, we quantified gene expression level. The primers are listed in Supplementary Table 1.

Immunohistochemistry validation based on tissue microarray

A tissue microarray (TMA) encompassing 75 matched CRC and adjacent normal tissues (HColA150CS02) was procured from Shanghai Outdo Biotech Co., Ltd. for immunohistochemistry (IHC) analysis. Following deparaffinization and rehydration, the TMA slides underwent a 30-min incubation in bovine serum albumin (BSA) and subsequently with rabbit UCN antibody (21951-1-AP; Proteintech) overnight at 4 °C. The slides were then exposed to a secondary antibody at ambient conditions for 50 min, developed using 3, 3′-diaminobenzidine (DAB), and counterstained with hematoxylin.

Utilizing the Aipathwell software (Wuhan Servicebio Technology Co.), we scanned and analyzed the IHC staining on the slide. Both staining intensity and the percentage of positively stained area informed our interpretation. Staining intensity was scored as follows: 0 for no staining, 1 for weak, 2 for moderate, and 3 for strong. The proportion of positively stained cells was graded on a scale: 0 for 0%, 1 for 1–25%, 2 for 25–50%, 3 for 50–75%, and 4 for 75–100%. A composite score, ranging from 0 to 12, was derived by multiplying the two aforementioned scores.

Anti-cancer drugs prediction

We employed the oncoPredict package to estimate IC50 values for both HRG and LRG, thereafter pinpointing potential therapeutic agents optimal for CRC patient interventions³³.

Statistical analysis

We conducted our statistical evaluations using R software (version 4.2.2) and GraphPad Prism (version 9.0). The Wilcoxon Rank-Sum Test served for bi-group comparisons, while the Kruskal–Wallis Test was utilized for multi-group distinctions. Kaplan–Meier log-rank methods determined survival outcomes. Spearman’s rank correlation was used to evaluate the interrelation of coefficients. Statistical significance was defined as an FDR-adjusted P value of less than 0.05.

Results

Single-cell transcriptome analysis unveils the TME

From the GSE132465 dataset, we curated a total of 63,252 cells from 23 primary CRC samples and their 10 matched normal mucosal counterparts, prepared for downstream dissection. After pre-processing, all cells were stratified into a spectrum of 50 clusters and visualized using the UMAP algorithm (Fig. 1A). Concurrently, we identified the marker genes specific to each cluster. By cross-referencing the CellMarker 2.0 database, these clusters were further annotated into eight major cell types: T cell, epithelial cell, B cell, macrophage, stromal cell, plasma cell, endothelial cell and mast cell (Fig. 1B). Moreover, UMAP visualization was employed to depict the sample classes of tumor cells and normal cells (Fig. 1C). The heatmap illustrated a well-annotated cell type classification and showcased the biological process of GO terms of each cell type (Fig. 1D).

Fig. 1.

Analysis of scRNA-seq data. (A) 50 clusters were identified. (B) Annotation of individual cell types. (C) Distribution of CRC sample cells normal mucosa cells. (D) Heatmap representation highlighting marker genes specific to each identified cell type. CRC, colorectal cancer.

Pathway activity analysis

We obtained 10 OS-related pathways of GO biological process from MSigDB database. Contextualized within the sample class, we utilized single cell analysis to delineate the differences in the activity of OS-related pathways across various cell types in normal and tumor samples at the single-cell level. Our findings revealed that the malignant cells (epithelial cells in tumor sample) manifested a markedly enhanced activity than that of normal epithelial cells (Fig. 2A). To substantiate the differential pathway activity within the TME, we assessed the ssGSEA scores across cell types in both normal and tumor classes. Notably, malignant cells exhibited a significantly higher ssGSEA score relative to their normal epithelial counterparts (Supplementary Fig. 1).

Fig. 2.

Evaluation of OS-related pathways. (A) Heatmap demonstrated activity of OS-related pathways were up-regulated in malignant epithelial cells. (B) 9 out of 10 OS-related pathways were activated in tumor. OS, oxidative stress.

Moreover, the endothelial cells, macrophages and stromal cells evinced a modest up-regulation in OS-related pathways, whereas T cells and B cells exhibited conspicuous down-regulation. Using bulk RNA transcriptome analysis, we compared the activity of 10 OS-related pathways between tumor and normal TCGA-CRC samples at tissue level. Intriguingly, a preponderance of the OS-related pathways (9 out of 10) demonstrated significantly elevated activity in tumor tissue compared to normal tissue. However, the negative regulation of response to oxidative stress pathway did not exhibit a statistically significant divergence (Fig. 2B).

Consensus clustering

Based on the consensus cumulative distribution function (CDF) in conjunction with delta area assessment, we identified the optimal k-value of 2, thereby partitioning the TCGA-CRC samples into two distinct clusters: C1 (n = 347) and C2 (n = 147), as illustrated in Fig. 3A-C. Furthermore, the K-M analysis illustrated a pronouncedly adverse prognosis for C2 compared to C1 (Fig. 3D). Moreover, employing Cox regression analysis, we screened OS pathway-related genes with prognostic value and depicted their expression patterns alongside the distribution of clinical features in a heatmap (Fig. 3E).

Fig. 3.

Identification of two OS-related clusters in CRC. (A) CDF from consensus clustering. (B) Delta area plot. (C) Consensus matrix representation with k = 2. (D) The K–M curve illustrating survival difference between C1 and C2. (E) Distribution of clinicopathological characteristics and differential expression of OS-prognostic genes between C1 and C2. OS, oxidative stress; CRC, colorectal cancer; CDF, cumulative distribution function.

Functional enrichment analysis

To gain insight into the pivotal roles that DEGs played, we performed GO and KEGG analysis. The GO analysis revealed that the DEGs are predominantly enriched in biological processes of extracellular structure organization, extracellular matrix organization, regulation of membrane potential, axonogenesis, regulation of cell morphogenesis involved in differentiation, and modulation of chemical synaptic transmission (Fig. 4A). While the KEGG analysis highlighted the involvement of DEGs in PI3K-Akt signaling pathway, neuroactive ligand-receptor interaction signaling pathway, calcium signaling pathway, cAMP signaling pathway, and ECM-receptor interaction (Fig. 4B).

Fig. 4.

Functional and mutational analysis between C1 and C2. (A) GO analysis. (B) KEGG analysis²⁵. (C) GSEA in C2. (D) Heatmap showed the difference of GSVA between C1 and C2. GO, Gene Ontology; KEGG, kyoto encyclopedia of genes and genomes; BP, biological process; CC, cellular component; MF, molecular function; GSEA, gene set enrichment analysis; GSVA, gene set variation analysis.

Furthermore, the GSEA analysis indicated that axon guidance, calcium signaling pathway, ECM-receptor interaction, cytokine-cytokine receptor interaction, and neuroactive ligand-receptor interaction signaling pathways are enriched in C2 cluster (Fig. 4C).

We utilized the GSVA algorithm to compare the differences in pathway enrichment between C1 and C2. The results indicate that oxidative phosphorylation is predominantly enriched in C1, while C2 exhibits higher enrichment of the TGF-beta signaling pathway, neuroactive ligand-receptor interaction, ECM-receptor interaction, MAPK signaling pathway, and the Hedgehog signaling pathway (Fig. 4D).

TMB and immune landscape

Through TMB analysis, our findings illuminate that C1 is characterized by a pronounced mutational frequency and an elevated TMB compared to C2 (Fig. 5A and B). The result of ESTIMATE analysis revealed that C1 possesses a notably higher tumor purity than that of C2, whereas C2 distinctly exhibits a higher stromal score, immune score and ESTIMATE score (Fig. 5C). The ssGSEA analysis underscores a predominant immune cell infiltration in C2, in stark contrast to C1 (Fig. 5D). The expression levels of all eight ICIs-related genes (HAVCR2, SIGLEC15, CTLA4, PDCD1, PDCD1LG2, TIGIT, CD274, LAG3) were remarkably higher in C2 (Fig. 5E).

Fig. 5.

Mutation landscape and immune microenvironment differences between C1 and C2. (A–B) Waterfall plots depict the mutation profiles for C1 (mutations identified in 300 of 315 samples, 95.24%) and C2 (mutations in 137 of 140 samples, 97.86%). (C) ESTIMATE analysis. (D) Differences in the infiltration levels of 23 immune cell types. (E) Differential expression of ICIs-related genes between C1 and C2. TMB, tumor mutation burden. ICIs, immune checkpoint inhibitors; HLA, human leukocyte antigen; ESTIMATE, estimation of STromal and immune cells in MAlignant tumor tissues using expression data. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Construction of OSRRS

A total of 424 genes were obtained as OS pathway-related genes (Supplementary Table 2). Posteriorly, 42 genes emerged as sentinel prognostic markers via Cox analysis (Fig. 6A). Finally, PPARGC1A, NCOA7, DAPK1, ERCC3, FKBP1B, SLC7A11, GSTP1, KRT1, SMPD3, UCN, WRN, and ADIPOQ were selected for the construction of OSRRS via LASSO Cox analysis (Fig. 6B-C). The coefficients for each gene were illustrated in Supplementary Table 3. We further assessed the correlation between the risk scores and ssGSEA scores for these pathways, which demonstrated a significant association (Supplementary Fig. 2). Utilizing the median OSRRS score as a fulcrum, we bifurcated the TCGA-CRC samples into HRG and LRG strata. Survival analyses of both training cohort (TCGA-CRC) and validation cohort (GSE39582, GSE17538) illuminated a somber outlook for the HRG patients compared to LRG patients (Fig. 6D–F). Following this, the distribution of risk score, survival status and gene expression patterns for training and validation cohorts were illustrated in Fig. 6G–I. Notably, DAPK1, ERCC3, SMPD3, UCN and ADIPOQ were highly expressed in HRG.

Fig. 6.

Constructing and validating the OSRRS. (A) Genes identified with prognostic relevance through Cox analysis. (B–C) Prognostic genes derived from LASSO Cox regression. Survival differences depicted in the TCGA (D), GSE39582 (E), and GSE17538 (F) cohorts. Distribution of risk scores, survival statuses, and gene expression patterns in TCGA (G), GSE39582 (H), and GSE17538 (I) cohorts. OSRRS, Oxidative stress-related risk signature; TCGA, The cancer genome atlas.

Nomogram

Initially, we conducted a comparative analysis of the clinical characteristics. Remarkably, while classifications based on age and gender did not exhibit significant differences, pronounced differences emerged within the parameters of stage, T stage, and N stage (Fig. 7A–E). Subsequently, incorporating these clinical features into Cox analyses, we pinpointed three independent prognostic variables: age, stage, and risk score (Fig. 7F, G). Consequently, a nomogram was constructed to predict the survival probability based on the risk stratification, age and stage (Fig. 7H). The calibration curve underscored the impeccable efficacy of our nomogram in predicting the prognosis (Fig. 7I). The area under the curve (AUC) in predicting 1-, 3-, and 5-year survival stood at 0.825, 0.815 and 0.835 respectively, furnishing compelling evidence of our nomogram’s precision in charting CRC survival outcomes (Fig. 7J). Moreover, the AUC for risk, age and stage are 0.825, 0.648 and 0.729, corroborating their pivotal role as risk determinants in our nomogram (Fig. 7K). Furthermore, we compared the risk score between C1 and C2, the box plot illustrated that the C2 exhibited a significantly higher risk score than that of C1 (Fig. 7L). To investigate the discriminatory potency of our OSRRS concerning survival probability, we performed the stratified survival analysis. The K-M survival plots underscored the capability of our OSRRS to function as a robust discriminator, distinguishing survival differences across gender (male versus female), age groups (≤ 65 years versus > 65 years), and disease stages (stage I-II versus stage III-IV) as illustrated in Fig. 7M–R.

Fig. 7.

Nomogram. (A–E) Differences in clinical features. (F–G) Clinical features with prognostic significance identified via Cox analyses. (H) Construction of the nomogram. (I) Calibration curve demonstrating the model’s robust performance. (J–K) ROC curves predicting overall survival. (L) C2 displays an elevated risk score compared to C1. (M–R) Sub-stratified survival analyses based on clinical features. ROC, receiver-operating characteristic.

GSVA and immune landscape between risk groups

Through GSVA analysis, HRG manifested a marked enrichment of the NOTCH signaling pathway, whereas LRG demonstrated pronounced enrichment in oxidative phosphorylation and metabolisms of fatty acids, butanoate, and propanoate (Fig. 8A). Turning to the TIDE analysis, a higher TIDE score was observed in HRG (Fig. 8B). The analysis of TMB indicated that TCGA-CRC patients exhibiting higher TMB could potentially experience a more unfavorable prognosis than those with lower TMB (Fig. 8C). Notably, the synergistic evaluation of both TMB and risk stratification provides a powerful tool in discerning survival outcomes for TCGA-CRC patients (Fig. 8D). Moreover, an inverse relationship crystallized between TMB and risk score (Fig. 8E). Subsequently, a comparative landscape of eight ICIs-related gene expressions across HRG and LRG unveiled SIGLEC15 as being dominantly expressed in HRG, while CD274 showed higher expression level in LRG (Fig. 8F). The waterfall plot revealed that LRG had a universally lower mutation frequency compared to HRG (Fig. 8G–H). Finally, when investigating the correlation between risk score and specific immune cells, the results showed that CD8 + T cells negatively correlated with the risk score, whereas cancer-associated fibroblasts and M2 macrophages displayed a positive correlation with the risk score (Fig. 8I).

Fig. 8.

Functional, mutational and immune analyses based on the OSRRS. (A) GSVA analysis. (B) TIDE analysis. (C) Survival difference between high- and low -TMB samples. (D) Survival stratified by the interplay of TMB and risk score. (E) Correlation between TMB and risk score. (F) ICIs-related genes expression. (G–H) Waterfall plots represent mutation landscapes of HRG (mutations noted in 220 out of 232 samples, 94.83%) and LRG (mutations in 217 of 223 samples, 97.31%). (I) Correlation between risk score and major immune cell population. OSRRS, Oxidative stress-related risk signature; GSVA, gene set variation analysis; TIDE, Tumor Immune Dysfunction and Exclusion; HRG, High-risk group; LRG, Low-risk group; ICIs, Immune Checkpoint Inhibitors; TMB, tumor mutation burden. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Experimental validation

Based on our analyses, UCN demonstrated a higher hazard ratio (HR = 1.92, 95% CI 1.302–2.832) and significant prognostic value (P < 0.001) among the candidate genes. Taken together with its significantly higher expression in tumor versus normal samples, we selected UCN for further experimental validation via IHC staining on a tissue microarray containing 75 paired CRC and adjacent normal specimens. (Supplementary Fig. 3, 4; Fig. 9A). Notably, the UCN H-score was significantly elevated in CRC tissues (Fig. 9B). To further validate our findings, we employed qRT-PCR to assess the mRNA expression of UCN in 15 matched CRC and adjacent normal samples. Our findings revealed a significant overexpression of UCN in CRC tissues (Fig. 9C).

Fig. 9.

Experimental validation of UCN and drugs screening. (A) IHC staining of UCN. (B) Elevated H-scores of UCN observed in CRC specimens. (C) qRT-PCR analysis. (D) 8 promising drugs were screened. IHC, immunohistochemistry; CRC, colorectal cancer; qRT-PCR, quantitative reverse transcription PCR. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Anti-cancer drugs prediction

Utilizing the oncoPredict algorithm, we identified eight anti-cancer drugs that demonstrated reduced IC50 values in the LRG, suggesting a potential therapeutic advantage for patients within this group (Fig. 9D).

Discussion

CRC is a highly heterogeneous disease in clinical and molecular features, causing great variation in tumor progression and treatment response³⁴. Stratification strategies were conducted to screen promising biomarkers and to apply precise treatment to CRC patients as well as predict the prognosis^35–39. However, the underlying role that OS played in TME of CRC patients remains understudied.

Previous studies have attempted to search OS-related biomarkers and develop innovative models for CRC, yet they neglected the important role of TME in CRC and failed to identify the molecular subtypes^40,41. To our knowledge, no research has investigated the activity of OS-related pathways in TME. In the era of single-cell sequencing, exploring pathway activities at both cellular and tissue levels represents a novel and robust approach for model construction and biomarker discovery. In the current study, we aim to elucidate the potential role that OS played in TME by evaluating 10 OS-related pathway activities. All 10 OS-related pathways exhibited an up-regulated activity in malignant epithelial cells compared to normal mucosa cells, indicating that OS could be a key player in tumorigenesis and progression. This finding was further validated at tissue level using bulk RNA-seq data. Thus, targeting OS could be a promising strategy in CRC classification and screening of novel biomarkers. Moreover, we observed that there exhibit a prominent up-regulation of OS-related pathways in stromal cells and a slight up-regulation in macrophages, which are two important immune cells. We therefore speculate that the OS might be involved in tumorigenesis and progression of CRC by mediating immune cells in TME.

Unsupervised consensus clustering is an effective method to determine subtypes of CRC^42,43. We identified two distinct subtypes via the consensus clustering algorithm, namely C1 and C2, which showed a distinct functional enrichment. Moreover, C2 is characterized as a worse prognosis subtype with higher mutation frequency and abundant immune cell infiltration, indicating that C2 patients could possibly be more responsive to immunotherapy. Our clustering showed a solid classification of CRC patients and pave the way for future exploration of precision medicine and prognosis prediction.

Based on the OS pathway-genes, we screened 12 prognostic significant genes, namely, PPARGC1A, NCOA7, DAPK1, ERCC3, FKBP1B, SLC7A11, GSTP1, KRT1, SMPD3, UCN, WRN, and ADIPOQ and constructed an OSRRS to make risk stratification for CRC patients. Due to the multidomain structure, DAPK1 was deemed a highly pleiotropic molecule^44–46. Previous studies reveled that DAPK1 could be a potential regulator for the metastasis of CRC by increasing the sensitivity of cancer cells to apoptotic signals and modulating the TME^47,48. Moreover, DAPK1 has been shown to play a role in the interaction between cancer cells and macrophages and loss of DAPK1 could potentially drive tumor invasion in CRC^49,50. SMPD3 was reported to inhibit CRC liver metastasis by producing exosomal ceramide⁵¹. Emerging studies have integrated ADIPOQ into risk models, the association of ADIPOQ with CRC risk remains a point of ongoing scientific discourse^52–56.

Our study has unveiled that this risk scoring system is highly effective in stratifying patients based on various clinical features. Furthermore, the sub-stratification analysis illuminates the efficacy of this scoring system in adeptly distinguishing patient prognoses. Integrating age, stage, and risk score enables an accurate prediction of survival probabilities for CRC patients. In addition, our investigation also identified eight promising anti-cancer drugs, namely Savolitinib, Obatoclax Mesylate, Linsitinib, Fludarabine, Erlotinib, Entinostat, Cyclophosphamide, and Carmustine. Notably, Linsitinib, Fludarabine, Erlotinib, and Cyclophosphamide have been featured in high-quality clinical trials and randomized controlled trials^57–60. Moreover, the current research has revealed that ICIs-related genes exhibit high expression levels in the C2 group, suggesting that patients within this group may have better reactivity to immunotherapy. However, the hypothesis that the OSRRS can predict benefits from ICI therapy in real-world CRC cohorts remains further validation. ICIs have transformed the clinical management of various malignancies, and targeting both oncogene and non-oncogene dependencies could significantly invigorate the TME, thereby potentially optimizing the efficacy to ICIs^61,62. Besides, exosomes, disulfidptosis, and gut microbiota may also influence the effectiveness of cancer immunotherapies, particularly ICIs^63–65.

Ma et al.⁶⁶ reported that UCN has emerged as a notable biomarker in renal cancer, demonstrating its potential in promoting renal tumor cell progression and its association with OS. In CRC, while several investigations have identified UCN as an innovative prognostic gene with corresponding models developed, there remains a scarcity of comprehensive studies in CRC^52,67,68. Utilizing qRT-PCR and IHC analyses on TMA, our research demonstrates an up-regulation of UCN in CRC cells compared to their adjacent normal mucosa, underscoring UCN’s potential as a therapeutic candidate for CRC intervention.

Generally, abundant immune infiltration is considered a favorable biomarker for predicting the efficacy of immunotherapy in CRC, whereas stromal infiltration is often associated with a pejorative prognostic outcome^69–72. In our current study, the C2 group, which exhibited a worse prognosis, demonstrated both high immune cell and stromal infiltration, underscoring the complexity of the TME. Recent research supports the notion that a prominent stromal component could neutralize the beneficial effects of immune responses^73–75. Additionally, while T cell infiltration is generally linked to better clinical outcomes, some phase III clinical trials have reported contrary results^76–78. TMB has emerged as a novel biomarker for predicting the prognosis of various cancers^79,80. Previous studies have indicated that tumors with high mutation levels possess higher levels of tumor neoantigens and exhibit increased immunogenicity, leading to a more favorable response to immunotherapy^80–82. Nonetheless, there are also findings suggesting that a high mutation burden does not consistently predict a positive immunotherapy response^83,84. Contradictorily, our study found that C2 patients with high TMB experienced worse prognoses. These observations suggest that while high TMB often correlates with an increased neoantigen load and potentially higher immunogenicity, which are advantageous under immunotherapy, its efficacy as a universal prognostic marker remains under debate.

Admittedly, some limitations merit consideration in this study. Primarily, the data analyzed for our study was retrieved from publicly accessible repositories, potentially introducing a selection bias that may not comprehensively mirror the broader patient cohort. Additionally, while our insights predominantly stem from bioinformatics analyses, a deeper exploration into the underlying molecular mechanisms remains imperative.

Conclusion

Taken together, our preliminary analysis shed light on the activities of 10 OS pathways within the TME of CRC at a single-cell resolution. This exploration culminated in the delineation of two distinct OS-associated subtypes. Subsequently, we constructed an OSRRS model using 12 genes derived from OS pathway-related genes, which offers prognostic insights for CRC patients. While our study highlights the intricate dynamics of the TME and identifies promising therapeutic avenues, it is limited by the reliance on publicly available cohort data. Future research should focus on validating these findings in larger, diverse cohorts and exploring the mechanistic roles of identified biomarkers in CRC progression and treatment response.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

L.M., W.L., and H.G. conceptualized and designed the study. E.L., S.Z., L.Z. and Z.L. interpreted the data. S.Z. and Q.G. conducted the experiment. L.M. and W.L. draft the initial manuscript. W.L., Z.L. and H.G. revised the manuscript.

Funding

This work was supported by The Youth Project of Shandong Provincial Natural Science Foundation (No. ZR2024QH655 and No. ZR2023QH414).

Data availability

This study utilized publicly available data for analysis. This study utilized four colorectal cancer datasets: the TCGA-CRC dataset, GSE132465 dataset, GSE39582 dataset, and GSE17538 dataset. All datasets are freely accessible and available for download. For any additional data requests or inquiries related to this study, please contact the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.