Integrative analysis of angiogenesis-related genes highlights prognostic indicators in colorectal cancer through single-cell sequencing and immune infiltration analysis

Abstract

Angiogenesis plays a pivotal role in colorectal cancer (CRC) progression and is closely intertwined with the tumor microenvironment (TME) and immune infiltration. This study aimed to identify key angiogenesis-related genes (ARGs) with prognostic significance in colorectal cancer using integrative bioinformatics and single-cell transcriptomic analysis. 526 common differentially expressed genes (DEGs) were extracted from three Gene Expression Omnibus (GEO) datasets and intersected with ARGs from MSigDB, resulting in identification of 18 candidate genes. A protein–protein interaction (PPI) network was constructed using the STRING database, followed by the extraction of 10 hub genes using the Cytoscape software. Five hub genes (MMP14, CXCL12, SPP1, TIMP1, and VCAN) showed significant association with poor overall survival. Expression analysis using UALCAN revealed significant upregulation of these genes, and their correlation with tumor-stage-specific expression. Utilizing the Tumor Immune Estimation Resource (TIMER) database immune infiltration analysis was carried out to explore the immune landscape. Tumor Immune Single-cell Hub2 (TISCH2), Tumor Immunotherapy Gene Expression Resource (TIGER), IMMUcan scDB and single-cell TIME (scTIME) databases revealed the expression of these hub genes in key TME components including fibroblasts, macrophages, and endothelial cells, and their link to immunosuppressive landscapes. Additionally, we discovered a substantial positive correlation between the expression of these hub genes and immune infiltration cells, such as macrophages, myofibroblasts and regulatory T cells (Treg). Notably, CXCL12 and SPP1 were implicated in immune cell recruitment, while TIMP1 and MMP14 were associated with ECM remodeling and myeloid cell differentiation. This study highlights the relevance of ARGs in tumor progression, prognosis and immune infiltration in CRC, offering potential targets for novel therapeutic interventions.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-03614-9.

Introduction

Colon cancer has the second-highest global rate of mortality and is the third most prevalent cancer worldwide, mostly due to the fact that it spreads extensively and is inherently heterogeneous [1]. The global burden of CRC is projected to rise over 2.2 million new cases and over 1.1 million deaths by 2030, presenting an alarming risk to public health [2]. Establishing an optimal approach at the individual level is extremely challenging because of CRC's heterogeneity. Robust studies must therefore be carried out promptly in order to identify high-risk CRC patients and to uncover new molecular targets.

Angiogenesis plays a role in the development and progression of multiple cancers, including colorectal cancer [3]. A multitude of genes and signaling pathways have been recognized as essential regulators of angiogenesis, with many exhibiting frequent overexpression in tumors, thereby intensifying angiogenic signaling [4]. Early diagnosis and accurate patient therapy will benefit from a thorough understanding of the occurrence and progression of angiogenesis in CRC. CRC research is now being conducted to investigate combinations of immune checkpoint inhibitors and antiangiogenic treatments [5]. In particular, RAB17 was identified to be a crucial angiogenesis regulator in human dermal microvascular endothelial cells using a combined single-cell RNA-seq and bulk-seq analysis [6]. Numerous research based on bioinformatics have revealed that angiogenesis-related genes (ARGs) could be potential prognostic indicators for a number of malignancies [7–9]. Nevertheless, the potential use of ARGs as prognostic indicators for colorectal cancer remains uncertain.

Tumour micro environment (TME) is an intricate network of interactions that includes blood arteries, fibroblasts, endothelial cells, stromal cells, immune cells, secretory factors, and extracellular matrix (ECM) [10]. Recent research has also highlighted the clinical significance of the TME, particularly the degree and phenotype of tumor-infiltrating immune cells, in predicting patient outcomes and treatment responses [11]. Notably, Pyroptosis-related gene signatures in glioblastoma enabled gene signature-based cancer research by revealing immunological microenvironmental characteristics and potential treatment targets [12]. Their critical function in the tumor microenvironment has been highlighted by targeted imaging of tumor-associated macrophages (TAMs) in breast cancer, underscoring the significance of immune component analysis in cancer research [13]. TME-associated genes in ovarian cancer show prognostic value, highlighting the importance of microenvironment-focused analysis [14].Understanding the complex interactions within the TME is essential for developing effective cancer therapies and improving patient prognosis [15, 16]. While previous studies have indicated that ARGs contribute to immune cell infiltration in CRC, the specific mechanisms underlying their involvement in shaping the TME and their prognostic significance have not been systematically elucidated, particularly at the single-cell resolution.

In recent years, bioinformatics analysis of high-throughput data, such as gene expression patterns, has become essential to understanding the pathophysiology of human Cancer. Previous studies on integrative multi‑omics and machine‑learning approaches have shown strong prognostic and predictive power across cancers; In lung adenocarcinoma, Zhang et al. created an AI-network signature based on 26 algorithms from tumor-infiltrating immune cell genes that reliably predicts survival and treatment response [17]. Single-cell transcriptomics has uncovered shared immunosuppressive features like dysfunctional T cells and CAF-S1 fibroblasts in mouse and human neuroblastoma, underscoring its value in tumor microenvironment analysis [18]. In ICI-related pneumonitis, single-cell transcriptomics has revealed pro-inflammatory monocytes and pathogenic Th17.1 cells, suggesting its ability to disrupt down immune-related pathogenesis [19].

While previous studies have utilized microarray data to identify key hub genes associated with angiogenesis in CRC, their potential roles in modulating immune cell infiltration within the tumor microenvironment remain insufficiently explored [20–22]. In our study through comprehensive and systematic bioinformatic investigation, we identified overlapping molecular gene signatures linked to angiogenesis and immune infiltration in CRC patient samples. Identifying key regulatory ARGs or ‘hub genes’ that orchestrate angiogenic signaling and immune interactions could not only improve prognostic assessments but also uncover potential therapeutic targets for combinatorial strategies involving anti-angiogenic and immunotherapy agents. Since angiogenesis is a major modifier of the TME regulating immune suppression, nutrient supply, and tumor cell migration, decoding the expression and functional context of ARGs within the TME is critical to understand CRC progression and therapeutic resistance. Leveraging multi-dataset integration, protein–protein interaction networks, and single-cell transcriptomic platforms, we aimed to pinpoint robust ARGs with clinical and immunological significance in CRC. DEGs were found by comparing normal colorectal tissues with CRC tissues obtained from three independent Gene Expression Omnibus (GEO) databases. Gene sets associated with angiogenesis were obtained from MSigDB. We used the STRING database to construct a protein–protein interaction (PPI) network in order to find important hub genes. The expression patterns and prognostic significance of these hub genes in CRC were validated using the Gene Expression platforms like University of Alabama at Birmingham Cancer Data Analysis Portal (UALCAN) and GEPIA. Additionally, we provided evidence for the correlation between the expression patterns of specific hub genes and the immunosuppressive microenvironment in CRC. Single-cell RNA-sequencing data revealed that the hub genes were differently expressed across TME components, providing insight into their functional role.

Thus, our study integrates bulk transcriptomic datasets with curated angiogenesis signatures to identify differentially expressed ARGs in CRC. By constructing co-expression networks and conducting immune infiltration and single-cell analyses, we have uncovered key ARGs that serve as potential biomarkers and modulators of immune suppression within the TME. This study lays the groundwork for developing targeted therapeutic interventions aimed at reshaping the TME and improving clinical outcomes in colorectal cancer.

Methods

Ethics approval

The current research did not include any studies with human or animal participants performed by any of authors. The bioinformatic analysis did not require approval from any ethics committee.

Microarray dataset identification and screening

The Gene Expression Omnibus (GEO) database allowed us to access the transcriptomics data of colorectal cancer patients (https://www.ncbi.nlm.nih.gov/GEO/). To find relevant datasets that align with the objectives of our study, we employed a stringent screening strategy by using the keywords "colorectal cancer," "homo sapiens," and "expression profiling by array". We selected datasets that comprised both tumor and normal tissues. Establishing statistical robustness and minimizing potential biases associated with limited sample sizes required a minimum of 20 samples.

The datasets not coinciding with such specifications were excluded. Three datasets GSE44076, GSE41258, and GSE81558 were selected that matched our set of criteria and included a diverse range of normal and primary tumor tissue types. GSE44076 comprised gene expression profiles of 98 tumor samples from patients and 98 normal colon samples. The GSE41258 dataset contained 186 primary tumors and 54 samples were of adjacent normal colon, the GSE81558 dataset comprised of 23 tumor samples and 9 normal colon samples. The detailed information the data sets used in this study and the annotation platform is listed in Supplementary Table S1.

GEO2R to determine differentially expressed genes (DEGs)

The datasets were screened for differentially expressed genes (DEGs) in primary tumor v/s normal patient tissues in CRC using the GEO2R online program (https://www.ncbi.nlm.nih.gov/geo/geo2r/), with threshold values of log fold change (FC) ≥ 1 and adjusted p-value ≤ 0.05. The data was subjected to Benjamini–Hochberg correction GEOquery (version 2.66.0) and limma (version 3.54.0). R packages were applied to get p-value ≤ 0.05 cut-off and to reduce the false positive rates. Only the genes with p-values lower than 0.05 for the aforementioned parameter, which is the cutoff for significant differential expression, were selected for further analysis. The detailed DEG list of all GEO data sets between all the cancers is provided as supplementary Excel files (Namely; 1_CRC_GSE44076, 2_CRC_GSE41258 and 3_CRC_GSE81558). In addition, a volcano plot was made in order to assess the DEGs. Genes that had an adjusted P value of less than or equal to 0.05 and a log2 fold change (FC) ≥ 1 were deemed to be upregulated and log₂ FC ≤ -1 were considered downregulated genes. The detail of common upregulated and downregulated genes is provided in the Supplementary excel file (Namely; Common upregulated and downregulated genes).

MSigDB

Next, ARGs were intersected with the significant DEGs of the three datasets, thereby obtaining angiogenesis-related DEG. 140 ARGs have been identified using a PubMed search for studies concerning angiogenesis and from the Molecular Signatures Database (MSigDB) [23] https://www.gsea-msigdb.org/gsea/msigdb/. After removing any duplicates and merging, 18 ARGs in total were found overlapping DEGs and ARGs using the Venny 2.1 web tool (https://bioinfogp.cnb.csic.es/tools/venny/).

Protein–protein interaction network construction

STRING database [24] (https://cn.string-db.org/), was used to create protein–protein interaction (PPI) networks. This network was further analyzed to investigate the molecular mechanism involved in cellular processing. Cytoscape software, version 3.9.1 (https://www.cytoscape.org) [25], was used to analyze the network for analysis and visualization. The top 10 genes were identified using the maximum clique centrality (MCC) parameter by the CytoHubba plug-in within the Cytoscape software.

External verification of the hub genes

Hub gene expression was verified by using two independent datasets GSE31279 and GSE50760 in the GEO database (https://www.ncbi.nlm.nih.gov/geo/) to confirm the specific expression of the obtained crucial hub genes. GSE31279 contains 52 primary tumor and 54 normal colon samples while GSE50760 had 18 tumor and 18 normal samples.

Cancer hallmark analysis

By merging 6763 genes from multiple mapping databases, Cancer Hallmarks created a list of genes (https://cancerhallmarks.com/) [26]. This hallmark-based approach enhances the utility of the concept as an organizational tool. Only significantly enriched hallmarks (adjusted P < 0.05) are highlighted, with each of the ten cancer hallmarks represented by a uniquely colored slice. The degree of enrichment in relation to the specified reference cancer signature gene set can be determined by the size of the slices. A graph with a minimal p-value of less than 0.0001 shows three distinct layers, while if it is less than 0.001, it shows two. For hub genes, hallmark enrichment plots were generated.

Survival analysis of hub genes

Using KM Plotter [27] (http://kmplot.com/), we verified the predictive significance of the 10 hub genes in CRC patients. We created KM survival maps, verified the overall survival linked to each of the 10 hub genes, and categorized the CRC patients into high- and low-expression groups based on the auto cutoff expression level of each hub gene. The best cut-off provided in the KM plotter is the best performing cut-off with the most statistically significant p-value (Cox regression analysis) from all the possible cut-offs computed automatically by the database between the lower and upper quartiles. P values were produced using the log-rank tests and the hazard ratios and related 95% CIs. A result of P < 0.05 on the log-rank test was deemed statistically significant. A user-friendly database and web tool focused on cancer prognostic biomarkers Dosurvive (http://dosurvive.lab.nycu.edu.tw/) [28] was used to perform the multivariate survival. Cox proportional hazards model and Accelerated failure-time (AFT) model was used to check the prognostic values of the ARGs.

Transcription level expression and protein levels of genes associated with prognosis

The web tool UALCAN (http://ualcan.path.uab.edu/analysis.html) [29] was used to examine the mRNA expression of the genes linked to angiogenesis in colon adenocarcinoma (COAD) tissues as well as normal tissues. UALCAN was also utilized to perform expression analysis of the extracted hub genes at different stages of the tumor for COAD patients. GEPIA2 was used for further validation of ARGs expression levels at mRNA level. The online resource GEPIA2 (http://gepia2.cancer-pku.cn/) [30] contains RNA sequencing (RNA-Seq) data from the Genotype-Tissue Expression (GTEx) project and The Cancer Genome Atlas (TCGA) of 9,736 tumors and 8,587 normal samples. All of the data were analyzed using conventional processing methods. The expression of each ARGs was then compared between CRC and normal colon tissues.

Hub genes expression levels in immune cells at single cell level in colorectal cancer

The IMMUcan scDB (https://immucanscdb.vital-it.ch/) [31] features 144 datasets that cover 56 various cancer types, each meticulously annotated with clinical, technological, and biological information across 50 fields. Additionally, the TIGER database (http://tiger.canceromics.org/#/) [32] provides bulk transcriptome gene expression data for 1,508 tumor samples across 8 cancer types, along with clinical immunotherapy data from 20 published studies, and includes 11,057 tumor/normal samples from 33 cancer types. In our study, we systematically utilized the IMMUcan database (which allows searches by gene and major annotation). Expression of hub genes in TME based on IMMUcan database utilized CRC_IMM_SS2_GSE146771 dataset. TIGER database was used to examine the heterogeneity within the TME. For single-cell transcriptomes of tumor immunological microenvironment, the Single-cell TIME (scTIME) [33] portal (http://scTIME.sklehabc.com), an online database and tool was used.GSE108989 (colon rectal) datasets was chosen to study the immune association of hub genes and to explore and analyze the correlation between hub and TME at the single-cell level. The expression of hub genes in several individual single CRC cells from different datasets was determined from the TISCH2 database (http://tisch.comp-genomics.org/) [34]. TISCH2 offers comprehensive, single-cell level annotations, facilitating the in-depth exploration of the TME across a spectrum of cancer types. It encompasses an extensive dataset of 6,297,320 cells derived from 190 distinct studies.

Immune infiltration analysis

To comprehensively examine the immune infiltration in COAD patients, we used the Tumor Immune Estimation Resource (TIMER2.0) [35], an interactive web-based platform. The TIMER2.0 database provides a thorough evaluation of tumor-infiltrating lymphocyte (TIL) levels using six sophisticated algorithms. The study employed immunedeconv, an R package that incorporates six cutting-edge algorithms, including TIMER, xCell, MCP-counter, CIBERSORT, EPIC, and quanTIseq, to accurately evaluate immune cell fractions from bulk RNA sequencing data. This platform is particularly valuable for analyzing TCGA or tumor-related data and provides accurate estimates of tumor purity. The levels of immune cell infiltration in each cancer sample were evaluated using TIMER, which were correlated with expression of hub genes.

Statistical analysis

The statistical analysis was carried out utilizing a range of tools and software. ANOVA or the Student's t-test were used for comparisons. All data used in this study were sourced from the Bioinformatics Online Database. GEO2R statistics obtained by the GEOquery and limma R tools were used to identify DEGs from the GEO database in order to conduct differential expression analysis. The differential expression of all the genes in CRC tissues was evaluated using Student’s t-test. The multiple testing correction Benjamini-Hochberg (FDR) approach was used in GEO2R as it offers a reasonable trade-off between limiting false positives and finding statistically significant genes when comparing two sample groups. The MCC method is used for identifying hub genes in PPI networks. Box plots were created using transcripts per million (TPM) with a p-value cut-off of 0.05 and log2FC threshold of 1.0. Univariate and multivariate analyses were performed using Cox regression.

The significance of prognostic differences among different groups was assessed using the log-rank test. Statistical analyses were conducted using R software (version 4.2.0). The Kruskal–Wallis test was utilized to examine the relationship between hub gene’s expression and various cell types. Pearson’s correlation coefficient was calculated to evaluate the strength of gene–gene associations. A p < 0.05 indicated statistically significant for all analyses (*p < 0.05; ** p < 0.01; ***p < 0.001).

Results

Identification of DEGs

The GEO2R tool was used in the current study to investigate microarray datasets of CRC patient samples and normal colon. Three GEO datasets were employed namely GSE44076, GSE41258, and GSE89287 and their microarray data was screened as a preprocessing step. Genes with log2 FC > 1 were identified to be up-regulated among significant genes, whereas those with log2 FC ≤ -1 were determined as significantly down-regulated. A total of 4446, 1048 and 2984 significant DEGs were obtained from GSE44076, GSE41258 and GSE81558 respectively based on strict criteria of adjusted p-value (≤ 0.05). The results of the DEGs in each dataset have been represented by volcano plots in Supplementary Figure S1.A-C. DEGs organized according to both biological and statistically significant dimensions.

The volcano plot showed the distribution of gene expression between primary tumor and normal colon. Genes with an adjusted p-value ≤ 0.05 and log2FC ≥ 1 were considered as differentially expressed genes. Each dot represented one gene wherein red dots indicated significantly upregulated genes and blue dots indicated significantly downregulated genes. Common upregulated and downregulated genes in the three GEO sets identified after intersection of three DEGs obtained from GEO2R. (Supplementary Figure S1. D-E). A total 526 genes were obtained after the intersection out of which 193 were upregulated and 333 were downregulated.

To identify overlapping ARGs from all microarray datasets we intersected ARGs from MSigDB and previously identified DEGs using Venny 2.1 tool. Total of 18 overlapping ARGs, comprising 14 up-regulated and 4 down-regulated genes were obtained (Supplementary Figure S1. F-G). Supplementary table S2 sheds light on the functions of all the 18 ARG genes in angiogenesis.

Protein–protein interaction network analysis and identification of central hub genes

Given the well-established role of angiogenesis in cancer, we hypothesized that these 18 ARGs could play a crucial role in the complex networks of CRC involving migration and angiogenesis. Therefore, they were selected for further investigation. Using STRING software, a PPI network diagram was created for 18 ARGs, with a minimum correlation coefficient > 0.400 as the criteria (Supplementary Figure S2. A). STRING database identified 18 nodes and 64 edges with PPI enrichment P value < 1.0E−16, average clustering coefficient of 0.674 and average node degree of 7.11. The hub genes, which are selected based on their high scores in the MCC algorithm ranking in CytoHubba. The network shows that the genes interact to varying degrees. TIMP1, MMP9, MMP3, MMP1, SPP1, CXCL12, MMP7, MMP14, MMP12 and VCAN were the key hub genes (Supplementary Figure S2.B) that were identified from the protein–protein interaction network analysis of the 18 genes using the CytoHubba plug-in MCC. Genes are color coded with dark red, red, orange, and yellow shades, with a deeper red shade indicating higher MCC scores and stronger connectivity (Supplementary Figure S2.B).

A hallmark enrichment plot for 10 interacting ARGs was obtained. It highlighted tissue invasion and metastasis, tumor promoting inflammation, resisting cell death, sustained angiogenesis, sustaining proliferation signalling as the common cancer hall mark. Other hallmarks like replicative immortality and resisting cell death were also visible (Supplementary Figure S2.C).

Verification of the hub genes

To verify the validity of the ARGs obtained after Cytoscape, we searched another GEO database and used the external dataset GSE31279 and GSE50760 to further verify the differential expression between the two groups. The result revealed the 3 ARGs (SPP1, CXCL12 and TIMP1) as hub genes in this validation dataset as previously identified by us (Supplementary Figure S2). This highlights the importance of these genes in angiogenesis in CRC. The detailed analysis is provided in the supplementary file namely _External verification data.

Overall survival analysis of hub genes

To decipher the prognostic relevance of the identified hub genes, the KM plotter tool was utilized for survival analysis. The mRNA expression profile of all the ten hub genes was analysed in relation to overall survival. By the construction of Kaplan–Meier plot, utilizing the best cutoff auto selected by the KM plotter tool for allotting the patients into low and high groups. Kaplan–Meier curves showed that higher expression of MMP14, CXCL12, SPP1, TIMP1 and VCAN genes was significantly related to worse survival, implying their potential as prognostic markers in (Fig. 1A–J). The top five ARGs with significant log p- rank values and effect on the overall survival were selected for further analysis i.e.; MMP14 (HR = 1.67, p = 5.1e−07), CXCL12 (HR = 1.41, p = 0.00099), SPP1 (HR = 1.83, p = 2.6e−08), TIMP1 (HR = 1.34, p = 0.0044) and VCAN (HR = 1.65, p = 1.5e−06). The Cox regression results for all investigated genes in all available colon cancer specimens with OS data are provided in Supplementary Table S3. As all five genes are known to be involved in angiogenesis, these findings suggest a strong link between high expression of ARGs and adverse outcomes in COAD patients. Supplementary table S4. gives the details of Intergene survival difference of 10 ARGs.

Fig. 1.

Association between candidate gene expression levels and overall survival (OS) in colorectal cancer. The OS of CRC participants with high (red lines) and low (black lines) expression of the five potential genes is shown in Kaplan–Meier survival plots. A–E The OS of MMP14, CXCL12, SPP1, TIMP1 and VCAN varied significantly. Time in months is shown on the x-axis, while survival probability is shown on the y-axis. Each analysis is given a hazard ratio (HR) and a statistical significance value (P < 0.05), with the log-rank test used to obtain P-values

Similarly, the results of multivariate analysis showed that ARGs were consistently determined to be significant at stage by both the Cox regression and AFT model, MMP14 (HR = 3.92, CI [2.48, 6.17], p-val = 4.13E−09), CXCL12 (HR = 3.91, CI [2.50, 6.13], p-val = 2.41E−09), SPP1 (HR = 3.79, CI [2.42, 5.93], p-val = 6.24E−09), TIMP1 (HR = 3.92, CI [2.48, 6.17], p-val = 4.13E−09) and VCAN (HR = 3.83, CI [2.45, 5.99], p-val = 3.93E−09). Advanced stage leads to significantly shorter time to survival with all the ARGs, as shown in Fig. 2A and B. ARGs were consistently identified in the AFT model as a substantial risk factor. Overall, the MMP14, CXCL12, SPP1, TIMP1 and VCAN were adverse prognostic factors and independent prognostic markers.

Fig. 2.

Multivariate Cox analysis of the relationship between MMP14, CXCL12, SPP1, TIMP1 and VCAN and overall survival in CRC patients A Cox regression analysis was used to compute the hazard ratios for each of the factors under test results were represented using forest plots B forest plots and a table of the time ratios and confidence intervals calculated from AFT model

Clinical relevance of expression of hub genes

Using UALCAN, the relationship between the clinical features of patients with colon adenocarcinoma and the mRNA expression of the above five hub genes was examined, including the cancer stages of the patients. (Supplementary Figure S3). The expression of significant hub genes in TCGA was compared between colon adenocarcinoma and normal colon samples for their clinical relevance. Interestingly, it was observed that tumor samples had higher expression levels of four of these genes (MMP1, SPP1, TIMP1, VCAN) than normal ones while CXCL12 which was significantly downregulated in the tumor sample (Supplementary Figure S3A). We further wanted to check if the expression of these five hub genes alters during different stages of tumour progression. We therefore, analysed the expression of all key hub genes in COAD’s pathological staging (Supplementary Figure S3B). We discovered that the mRNA levels of MMP14, SPP1, TIMP1 and VCAN were elevated in advanced stages of CRC while the expression of CXCL12 was significantly downregulated with the CRC stages. Among the 5 selected hub genes, the expression of MMP14, SPP1 and VCAN showed significant variation between different tumor stages when compared with the adjacent normal tissue. On the other hand, expression levels of MMP14 showed significant variation in stages 2, and 3 compared to that in adjacent normal tissue. Expression of MMP14, SPP1, TIMP1 and VCAN were observed to exhibit significant variation between stage 1 and stage 3 of the tumor whereas the expression of CXCL12 was significantly downregulated with the progression of cancer as seen in the stagewise comparative analysis. Supplementary Table S5. provides the significant variation of the hub genes across different stages of COAD patient sample data available on UALCAN database.

To verify these findings, further analysis at the transcript levels of the 5 ARGs was performed using the GEPIA2 database, the results are shown in Supplementary Figure S4. The mRNA expression levels of all four ARGs i.e., MMP14, SPP1, TIMP1 and VCAN were demonstrated to be upregulated in CRC tissues compared with normal samples, whereas CXCL12 expression was downregulated in tumor tissue consistent with the results obtained from UALCAN.

Taken together, the overall survival data and stage specific expression analysis strongly suggest that during cancer progression, the expression of four of these five hub genes increases and is associated with poor prognosis. In contrast, the expression of CXCL2 is progressively downregulated.

These expression patterns highlight the potential of these genes as prognostic biomarkers, capable of distinguishing low-survival-probability patients from those with better outcomes. Moreover, elevated expression levels of these genes are significantly correlated with advanced stages of colorectal cancer and reduced overall survival (See Fig. 1 and Supplementary Figure S3).

Hub genes and immune infiltration analysis using TIMER

Recent studies have emphasized the complex interplay between angiogenesis and immune cell infiltration in colorectal cancer [10–12]. We therefore, checked if there was any correlation between the mRNA expression of MMP14, CXCL12, SPP1, TIMP1 and VCAN with infiltrating immune cells in COAD patients using the TIMER database. Interestingly, expression of all the hub genes showed a significant positive correlation with macrophages while a weak or insignificant correlation was observed with B cells. MMP14 also showed significant positive correlation with CD4 + T cell (R = 0.301, P = 3.539e−07), myeloid dendritic cell (R = 0.672, P = 1.68e−37), and neutrophil (R = 0.583, P = 1.88e−26), (Fig. 3A) suggesting its involvement in inflammation or tumour invasion. Similarly, expression of CXCL12, a chemokine important for immune cell recruitment, showed strong positive correlation with CD4 + T cell (R = 0.401, P = 4.92e−12), CD8 + T cell (R = 0.23, P = 3.77e−04), and myeloid dendritic cell (R = 0.591, P = 2.79e−27). (Fig. 3B). SPP1 (osteopontin) gene showed significant positive correlation only with macrophage (R = 0.534, P = 1.16e−21) and neutrophil (R = 0.528, P = 3.50e−21) while an insignificant or negative correlation, observed with CD8 + and CD4 + T cells, and B cells suggesting its pro-tumour immune microenvironment role favouring immune evasion (Fig. 3C). Recent findings suggest that macrophages and neutrophils contribute to immunosuppressive inflammation, thereby influencing anti-tumor immune responses [36]. Both TIMP1 and VCAN showed strong positive correlation with macrophage, neutrophils and dendritic cells and both genes exhibited negative correlation with tumour purity suggesting their expression in stromal and immune cells (Fig. 3D, E). Both VCAN and TIMP1 may be associated with poor prognosis that is also supported by Kaplan–Meier survival plots (see Fig. 1).

Fig. 3.

Correlation analysis between hub genes and immune cell infiltration in COAD samples from TCGA cohort. A–E Correlations between the 5 hub genes and the abundance of the immune cells, deconvoluted with TIMER. Correlation analysis of A MMP14, B CXCL12, C SPP1, D TIMP1 and E VCAN and B cells, CD4 + T cells, CD8 + T cells, myeloid DCs, macrophages and neutrophil. Pearson correlation was applied to analyze the correlation between the mRNA level of hub genes and tumor purity and immune infiltration level

Analysis of correlation between hub gene expression and tumor immune microenvironment at single-cell level

IMMUcan SingleCell RNAseq Database (IMMUcan scDB) was utilized to explore the genes expression patterns of the hub genes in the TME. The IMMUcan scDB displays a heatmap of the average gene expression in each cell type within multiple datasets. We searched for all the five hub genes (MMP14, CXCL12. SPP1, TIMP1 and VCAN) using “annotation major” as cell type resolution. The expression of hub genes in various cells of TME in four CRC datasets i.e.; CRC_IMM_10X_GSE146771, CRC_IMM_SS2_GSE146771, CRC_UNB_10X_GSE132465 and CRC_UNB_10X_GSE144735 is shown in Fig. 4. MMP14 exhibited the highest expression in fibroblasts, mast cells, and myofibroblasts. CXCL12 was predominantly expressed in fibroblasts and endothelial cells, while SPP1 showed elevated levels in macrophages and endothelial cells. TIMP1 was highly expressed across multiple immune-related cell types, including myofibroblasts, fibroblasts, and macrophages. VCAN demonstrated strong expression in both macrophages and fibroblasts (Fig. 4A–E). The distinct expression of hub genes across TME cell types highlights their potential involvement in immune modulation and colorectal cancer progression. The relationship between hub gene expression and the diverse single cell landscapes was also examined using the TIGER database (CRC/Qian J et al. cell Res 2020, jun20) [37]. The cell populations were separated into nine primary cell types: B cells, endothelial cells, enteric glial cells, fibroblasts, malignant cells, mast cells, myeloid cells, and plasma cells and T cells (Fig. 5A–E). TISCH2 database and Single Cell RNAseq database was employed to further analyze the distribution of each gene in TME across the four different datasets i.e.; CRC_IMM_10X_GSE146771, CRC_IMM_SS2_GSE146771, CRC_GSE166555 and CRC_GSE179784. Furthermore, the IMMUcan SingleCell RNAseq Database was utilized to investigate the distribution of hub genes expression in the TME using UMAP (Fig. 5 left panel) and violin plots (Fig. 5 right panel), as depicted in Fig. 5. MMP14 was predominantly expressed in myofibroblasts, with detectable levels in endothelial and fibroblast cells. CXCL12 was mainly confined to endothelial cells, while SPP1 showed enrichment in dendritic cells, endothelial cells, mast cells, and epithelial cells. VCAN was prominently expressed in fibroblasts, and TIMP1 was highly expressed in both myofibroblasts and endothelial cells, though present at varying levels across most cell types. These expression patterns were consistent across all datasets (Fig. 5).

Fig. 4.

The expression of five hub genes in TME based on IMMUcan Single Cell RNAseq database. A–E The distribution of MMP14, CXCL12, SPP1, TIMP1 and VCAN expression in various datasets downloaded from IMMUcan Single Cell RNAseq database

Fig. 5.

Expression patterns of hub genes in the tumor microenvironment as revealed by single-cell RNA sequencing. A–E Visualization of MMP14, CXCL12, SPP1, TIMP1, and VCAN expression across tumor microenvironment (TME) cell populations using datasets retrieved from the TISCH2 database. Data from the CRC_IMM_SS2_GSE146771 single-cell RNA-seq dataset are presented, highlighting gene expression across annotated TME cell types. Uniform Manifold Approximation and Projection (UMAP) plots and violin plots depict gene expression intensity and distribution. Cell populations are color-coded based on their primary classification to illustrate cell-type-specific expression profiles

Single-cell characterization of hub gene expression and cell–cell interactions in CRC

To better characterize the various cell types that express ARGs, we then examined scRNA-seq data from CRC tumors. To investigate the cellular context and functional implications of hub gene expression in CRC, we analyzed the GSE146771 dataset from the Tumor Immune Single-cell Hub 2 (TISCH2) database using single-cell RNA sequencing data (Figs. 6, 7). The bar chart shows the proportion of different cells in each patient (Fig. 6A). The pie chart shows the overall composition of different cell types (Fig. 6B). Through cell–cell interaction (CCI) analysis, the intercellular communication network can be systematically decoded, which helps to reveal the regulatory mechanisms of the communicating cells and ultimately explains the function of tissues in homeostasis and their changes in tumors. The interactions between various cell types were anticipated using TISCH2's cell chat algorithm. The study revealed robust intercellular signalling particularly among fibroblast, endothelial cells, and malignant cells (Fig. 6C) suggesting these populations may coordinate key processes in the CRC microenvironment.

Fig. 6.

Single-cell transcriptomic analysis of hub genes and intercellular signalling in colorectal cancer. A, B Bar plot showing the distribution of major cell types across individual patients and pie chart summarizing the overall cell-type proportions in the CRC cohort. C Heatmap of predicted ligand–receptor interactions between annotated cell clusters, revealing potential intercellular communication networks. D Transcription factors regulated by hub genes in the GSE146771 dataset, with color intensity reflecting predicted regulatory strength. E Cell–cell interaction network visualizing the number and strength of predicted communications among different cell populations. F Dot plot illustrating the expression levels of representative marker genes across identified cell types. G UMAP embedding of single-cell transcriptomes, coloured by cluster identity and annotated cell type. H–J Gene set enrichment analysis (GSEA) showing pathway-level enrichment of hallmark signatures related to inflammatory response, angiogenesis and epithelial–mesenchymal transition (EMT) across cell subsets

Fig. 7.

Single-cell expression and pathway enrichment analysis of hub genes in colorectal cancer. A Violin plots showing the expression distribution of CXCL12, MMP14, SPP1, TIMP1, and VCAN across individual cells in the GSE146771 dataset, derived from the TISCH2 single-cell RNA sequencing database. B UMAP visualization displaying the spatial distribution and expression intensity of the five hub genes across annotated cell populations at single-cell resolution. C, D Heatmaps illustrating significantly enriched upregulated (C) and downregulated (D) hallmark pathways identified from differentially expressed genes in each cell type. E, F Heatmaps showing enriched upregulated (E) and downregulated (F) KEGG pathways across cell populations, revealing functional heterogeneity within the tumor microenvironment based on gene expression dynamics

The onset and spread of CRC are significantly influenced by a number of transcription factors. We deduced the transcription factors controlling gene expression in each cell cluster using the local markers of spatial relationship approach in TISCH2. We used heatmaps to show the degree to which transcription factors were expressed in each of the dataset's cell clusters. Figure 6D shows the expression patterns of several key transcription factor in macrophages, malignant cells and endothelial cells. Cell communication networks predict the number and intensity of interactions between different cell subsets. Complementary interaction network data further highlighted strong bidirectional signaling between malignant cells, fibroblasts, and endothelial subsets (Fig. 6E). These findings indicate that malignant cells played a crucial role in CRC immunity-related pathways and were worthy of further investigation. The cells were initially clustered according to the expression profile, that led to identification of 34 transcriptionally distinct subpopulations (Fig. 6F). To better understand the biological significance of these cell populations, these were further consolidated into 10 broad immune cell types based on canonical markers (Fig. 6G). GSEA analysis demonstrated the enrichment of single-cell datasets in the inflammatory response, angiogenesis and EMT gene sets. Single-cell sequencing research revealed that fibroblast cells were characterized by the presence of unfavourable major Hallmarks, including HALLMARK_ EPITHELIAL_MESENCHYMAL_TRANSITION and HALLMARK _ANGIOGENESIS. In contrast, monocytes and macrophages were shown to have a prevalence of immune-related gene sets, such as HALLMARK_INFLAMMATORY_RESPONSE (Fig. 6h–J).

Figure 7 shows the distribution and expression of hub genes within the TME at single-cell resolution. The results revealed that MMP14 and CXCL12 were expressed majorly in epithelial and fibroblast cells. SPP1 was upregulated in several immune cell types including monocytes, macrophages, malignant cells, endothelial and fibroblast cells. TIMP1 was expressed majorly in all the cell types like macrophages, monocytes and plasma cells, whereas the expression levels of VCAN was predominantly expressed in the monocytes, macrophages and fibroblast cell cluster (Fig. 7A, B).

The distinctive roles of hub genes were further characterized using TISCH2 database in order to clarify the precise role that these genes may play in CRC and their synergistic connection with other cells. The hub genes were shown to be strongly correlated with several hallmark genes sets. The hub genes were predominantly expressed in fibroblast and macrophages in controlling the angiogenesis and were downregulated in B cells in case of hallmark gene sets (Fig. 7C, D) consistent with the TIMER results (see Fig. 3). According to Fig. 7E, F, the heatmaps display increased KEGG hallmark gene sets that are mostly prevalent in mononuclear/macrophage and fibroblast populations while downregulated KEGG pathways were more diffuse across cell types (Fig. 7F).

Single-cell transcriptomic analysis reveals hub gene expression dynamics in the tumor Immune microenvironment of colorectal cancer

To investigate the relationships between cellular specificity and the tumor microenvironment of key hub genes in colorectal cancer, we utilized the GSE108989 dataset for single- cell transcriptome visualization and the online scTIME tool for genetic analysis of tumor immune microenvironment specificity. This enabled the detailed mapping of gene expression patterns and the reconstruction of pseudo-temporal cell differentiation trajectories among diverse tumor-infiltrating cell populations (Figs. 8, 9, 10, 11, 12). Unsupervised clustering identified twenty main cell types within the TME (Figs. 8A, 9, 10, 11, 12A). The findings showed that MMP14 was mostly expressed in fibroblasts, myeloid cells, and plasma cells, indicating a role in stromal and innate immune compartments. (Fig. 8B, C). Single-cell landscape and expression characteristics of the CXCL12 gene are shown in Fig. 8B, C. CXCL12 expression was localized to fibroblasts, malignant epithelial cells, and endothelial cells (Fig. 9B, C), suggesting a multifaceted role in both tumor and vascular niches. The expression of SPP1 was elevated in SPP1 in fibroblasts, enteric glial cells, endothelial cells, and malignant cells (Fig. 10B, C), supporting its involvement in both structural and neoplastic components of the TME. TIMP1 demonstrated robust expression in fibroblasts and myeloid cells (Fig. 11B, C), implicating it in extracellular matrix remodeling and immune modulation in colorectal cancer. VCAN was prominently enriched in fibroblasts, endothelial cells, and enteric glial cells (Fig. 12B, C), reinforcing its stromal-centric role.

Fig. 8.

Single-cell level characterization of MMP14 and its association with the tumor immune microenvironment in colorectal cancer. A Distinct cell populations identified in colorectal cancer (CRC) using single-cell transcriptomic data from the scTIME database. B Cell clusters exhibiting predominant expression of MMP14, highlighting its distribution within the tumor landscape C Expression levels of MMP14 across immune cell subsets, as profiled in scTIME, illustrating its immune context within the CRC microenvironment. D Pseudotime trajectory analysis mapping the dynamic progression of endothelial cells, fibroblasts, malignant cells, myeloid cells, plasma cells and T cells. Each dot represents a single cell, colored according to its annotated type. E Distribution of MMP14 expression along pseudotime trajectories for the six cell types, suggesting a potential role in cellular differentiation or state transitions. F, G Association between MMP14 expression and immune infiltration levels, with specific enrichment in CD8-ZNF683 and NK-FCGR3A cell subsets, indicating a possible link to cytotoxic immune regulation in CRC

Fig. 9.

Single-cell transcriptomic insights into CXCL12 expression and its association with the tumor immune microenvironment in colorectal cancer. A Identification of distinct cellular clusters in colorectal cancer (CRC) based on scTIME single-cell RNA sequencing data. B Cell populations demonstrating notable CXCL12 expression, highlighting its preferential distribution across specific compartments of the tumor microenvironment. C Expression profile of CXCL12 across various immune cell subsets as derived from the scTIME database, emphasizing its immune-context specificity. D Pseudotime trajectory analysis depicting the inferred differentiation paths of endothelial cells, fibroblasts, malignant cells, myeloid cells, plasma cells and T cells. Each point represents a single cell, color-coded by cell type. E Distribution of Cxcl12 expression along pseudotime trajectories across the six major cell types, suggesting its potential involvement in dynamic cellular transitions within the tumor milieu

Fig. 10.

Single-cell resolution analysis of SPP1 expression and its association with the tumor immune microenvironment in colorectal cancer. A Identification of distinct cellular clusters in colorectal cancer (CRC) using scTIME-based single-cell transcriptomic profiling. B Visualization of cell populations with prominent SPP1 expression, indicating its preferential localization within specific components of the tumor microenvironment. C Expression landscape of SPP1 across immune cell subsets, as defined by the scTIME database, highlighting its variable role in immune modulation. D Pseudotime trajectory reconstruction illustrating the differentiation dynamics of endothelial cells, fibroblasts, malignant cells, myeloid cells, plasma cells and T cells. Each point represents a single cell, color-coded according to its annotated type. E Association between SPP1 expression and pseudotime progression across the six major cell types, suggesting context-specific functional roles during cellular transitions. F Correlation between SPP1 expression and immune cell infiltration, with increased expression observed in CD4⁺ CTLA4⁺ regulatory T cells (Tregs), implying a potential immunosuppressive function in the CRC microenvironment

Fig. 11.

Single-cell characterization of TIMP1 expression and its association with the tumor immune microenvironment in colorectal cancer. A Identification of distinct cellular populations in colorectal cancer (CRC) using scTIME-derived single-cell RNA sequencing data. B Visualization of cell clusters with elevated TIMP1 expression, indicating its preferential enrichment in specific tumor-associated cell types. C Distribution of TIMP1 expression across various immune cell subsets based on scTIME annotations, highlighting its immunological relevance within the tumor context. D Pseudotime trajectory analysis depicting the differentiation dynamics of endothelial cells, fibroblasts, malignant cells, myeloid cells, plasma cells, and T cells. Each point represents an individual cell, color-coded by its classified cell type. E Expression trend of TIMP1 along pseudotime trajectories across the major cell types, suggesting potential roles in cellular progression or state transitions within the tumor microenvironment. F Association between TIMP1 expression and immune cell infiltration, with increased expression observed in CD8⁺ PDCD1⁺ T cells, implicating a possible link to T cell exhaustion or immune suppression in CRC

Fig. 12.

Single-cell analysis of VCAN expression and its association with the tumor immune microenvironment in colorectal cancer. A Cell clusters identified in colorectal cancer (CRC) using scTIME single-cell RNA-seq data. B Predominant VCAN expression observed in specific CRC-associated cell clusters. C Expression levels of VCAN across various immune cell subsets, based on scTIME database annotations. D Pseudotime trajectory illustrating the differentiation pathways of endothelial cells, fibroblasts, malignant cells, myeloid cells, plasma cells, and T cells. Each point represents a single cell, color-coded by cell type. E Association between VCAN expression and pseudotime progression in each cell type, indicating its involvement in cellular state transitions. F–H Correlation between VCAN expression and immune infiltration, with elevated levels detected in Cycling T cells, CD8⁺ SLC4A10⁺ MAIT cells, CD8⁺ Transitional cells, and CD8⁺ ZNF683⁺ cells, suggesting immunomodulatory functions in the CRC microenvironment

Furthermore, pseudotime analysis was utilized to evaluate CRC cell development. There are two development pathways in CRC (P1: fibroblast cells to B cells; P2: fibroblast cells to malignant cells), as shown by the two-dimensional tree structure that started with fibroblast cells and finished with B cells or malignant cells (Figs. 8, 9, 10, 11, 12D, E). Expression of MMP14 along these trajectories further supports its role in fibroblast differentiation and transition to immune and malignant phenotypes. CXCL12 expression is predominantly localized within fibroblasts, myeloid, and endothelial cell lineages, but is largely absent in malignant cells. High expression in fibroblasts suggests the acquisition of a cancer-associated fibroblast phenotype. The upregulation of TIMP1 in differentiated myofibroblasts, fibroblasts, and macrophages along pseudotime trajectories suggests its role in orchestrating ECM remodeling and stromal activation during CRC progression, linking terminal cell states to tumor-promoting processes. SPP1 and VCAN are terminally expressed in endothelial and macrophages. Overall, trajectory-based analysis showed that during stromal and immune cell development, MMP14, CXCL12 SPP1, TIMP1 and VCAN are dynamically regulated. Their expression patterns are consistent with functional shifts toward active, tumor-promoting characteristics. This highlights their significance in influencing the tumor microenvironment of colorectal cancer via angiogenesis, immunological regulation, and extracellular matrix remodeling.

Our data further confirmed the critical role of hub genes in the differentiation of fibroblast cells. Correlative analyses further clarified hub gene roles within the TME. MMP14 showed significant positive correlation with CD8_ZNF683(p value = 0.02306) and NK_FCGR3A (p value = 0.00856) Fig. 8F, G. The expression of SPP1 was positively corelated with CD4-CTLA4-Treg cells (p value = 0.00869) in colorectal cancer (Fig. 10F). A positive correlation was also observed between TIMP1 and CD8-PDCD1(p value = 0.02235) (Fig. 11F). A positive correlation was observed between VCAN and Cycling-T(p = 0.03764), CD8-SLC4A10-MAIT(p = 0.00525), CD8-Transition (p value = 0.00971) and CD8_ZNF683(p value = 0.0483) (Fig. 12F). We observed that CD8_ PDCD1 clusters had relatively high weights in the network in almost all the cases, which indicated that CD8_PDCD1 played an important role in interacting with TIME. Gene expression quantification and cell type annotation may be impacted by technological constraints that are intrinsic to single-cell RNA sequencing, such as dropout events and batch effects. Technical restrictions might be the cause of the inconsistent results described above.

Overall, these results show that these hub genes are expressed not only in fibroblasts and endothelial cells but also in some immune cells and tumor cells. The high expression of TIMP1 in nearly all immune cell types proposed their crucial role in regulating the immune suppression process within the TME. This may include the inhibition of immune cell activation, promoting the proliferation and function of immune inhibitory cells, and modulation of signaling pathways associated with immune evasion. The analysis of single-cell data, in conclusion, reveals notable variations in hub gene expression levels among distinct cell subpopulations. The aforementioned results accentuate the diversity that exists in cell populations while providing functional specificity to gene expression within the colorectal cancer microenvironment.

Discussion

Angiogenesis plays a crucial role in colorectal cancer (CRC) development and progression [38]. Previous research was primarily focused on identifying CRC biomarkers through comparisons between the expression of genes in tumor and normal tissues [39, 40]^. In this study we have identified and characterised ARGs using a comprehensive bioinformatics analysis of transcriptomic data. We observed that 18 ARGs were consistently dysregulated in CRC samples, with 14 upregulated and 4 downregulated genes. Ten important ARGs implicated in CRC pathogenesis. TIMP1, MMP9, MMP3, MMP1, SPP1, CXCL12, MMP7, MMP14, MMP12 and VCAN 10 ARGs were identified in this study, leveraging a comprehensive and robust bioinformatics pipeline.

Protein–protein interaction network analysis showed five key hub genes, MMP14, CXCL12, SPP1, TIMP1, and VCAN that displayed strong connectivity and were implicated in multiple hallmark cancer pathways. These genes were particularly enriched in hallmarks such as angiogenesis, invasion, and immune modulation, emphasizing their relevance to colon cancer pathogenesis [41]. Based on our results, we delineated five important genes (MMP14, SPP1, TIMP1, and VCAN, CXCL2) that contributed to progression and prognosis of colorectal cancer. These results were further validated using GSE31279 and GSE50760 external GEO datasets of colorectal cancer. Expression profiling using TCGA dataset revealed that most of these hub genes were significantly overexpressed in tumor tissues compared to normal counterparts, except for CXCL12, which showed downregulation. Importantly, MMP14, SPP1, TIMP1, and VCAN demonstrated progressive upregulation with advancing tumor stages, suggesting their association with tumor aggressiveness [42, 43]. In addition to offering important insights into their possible involvement in the advancement of colorectal cancer, this systematic approach strengthens confidence in TME associated target identification. Survival analysis further confirmed that the high expression of top five hub genes; MMP14, CXCL12, SPP1, TIMP1 and VCAN was linked to a poor patient prognosis in COAD [44]. Although CXCL12 was downregulated in tumours, it was associated with poor prognosis. This paradoxical role of CXCL12 has also been observed earlier where it displays both anti and pro-tumorigenic effects in a context dependent manner [45, 46]. We validated the role of these five hub genes in CRC across multiple platforms including TCGA, TIMER, UALCAN, and single-cell RNA sequencing datasets.

Several hub genes have been reported to be linked to immune infiltration in tumors and influences the sensitivity of the cancers to anticancer medications [47]. Furthermore, the examination of immune cell infiltration by TIMER, demonstrated significant correlation between hub gene expression and immune cell populations, especially macrophages, CD4 + T cells, and dendritic cells. Unique immune cell abundance patterns and their association with patient outcome, highlights the complex interplay between the tumor microenvironment and the advancement of colorectal cancer. Our observations support the hypothesis that these genes are integral to the tumor immune microenvironment. For instance, SPP1 and TIMP1 showed strong expression in macrophages and were negatively correlated with tumor purity, suggesting expression by stromal or immune cells rather than tumor epithelial cells alone.

The crucial involvement of these hub genes in the pathophysiology of colorectal cancer were further validated by their inclusion in single-cell RNA sequencing data sets, offering a strong basis for further study and possible therapeutic uses. Fibroblasts, endothelial cells, and various immune cells such as macrophages and dendritic cells were found to express these hub genes. The pseudotime trajectory analyses also revealed dynamic gene expression patterns during cell differentiation, particularly from fibroblast to malignant or immune phenotypes, reinforcing their roles in tumor progression and immune evasion [48].

Matrix metalloproteinases (MMPs) play a crucial role in colorectal cancer (CRC) progression and angiogenesis. MMP2 is overexpressed in CRC and is associated with poor outcomes [49]. This MMP, along with others like MMP7, MMP10, MMP11, MMP12, and MMP14, are consistently upregulated in CRC tissues [50]. Their involvement in angiogenesis and early carcinogenesis has been established through preclinical and clinical trials, showing a direct correlation between elevated MMP expression and poor survival in endometrial cancer patients [51, 52]. Our investigation revealed higher expression levels of MMP14 in CRC tumor tissue than the adjacent normal colon tissue, we also showed a strong correlation between high levels of MMP14 and lower survival rates in CRC patients (P < 0.05). This highlights the important role of MMP14 in the CRC. Although MMPs are well established to play an important role in tumour progression, However, developing effective and safe MMP inhibitors remains challenging due to their complex regulation and high homology within the MMP family. Future research should focus on development of selective MMP inhibitors and antibodies to improve cancer treatment strategies.

TIMP1 plays a significant role in colorectal cancer (CRC) progression and angiogenesis. As a valuable diagnostic and prognostic biomarker, TIMP1 is associated with cancer signaling pathways, immune cell infiltration, and immune checkpoint gene expression [53, 54]. A meta-analysis by Lee et al. (2011) confirmed that high plasma or serum TIMP1 levels predict poor survival outcomes in CRC [55]. TIMP1EV is enriched in CRC-derived extracellular vesicles and can be detected in patient serum, serving as a potential non-invasive biomarker for risk stratification in colorectal liver metastases [56]. In glioblastoma, TIMP1 is upregulated by Sp1 and correlates with immune cell infiltration and cancer progression [57]. These findings highlight TIMP1's multifaceted role in CRC angiogenesis and its potential as a therapeutic target.

CXCL12, a chemokine constitutively expressed by stromal cells, plays a crucial role in angiogenesis and immune cell infiltration in various pathological conditions, including rheumatoid arthritis and cancer [58]. CXCL12 enhances angiogenesis by recruiting endothelial progenitor cells and stimulating endothelial cell migration and growth [59]. In the current study CXCL12 expression was elevated in endothelial cells and was also involved in its differentiation. SPP1 overexpression has been linked to a poor outcome in tissues from ovarian and hepatocellular carcinomas [60]. In both cancer types, SPP1 expression positively correlates with infiltrating levels of various immune cells, including CD4 + T cells, CD8 + T cells, macrophages, neutrophils, and dendritic cells [60, 61]. VCAN proteolysis strongly correlates with CD8 + T cell infiltration in colorectal cancer, while its accumulation is associated with T cell exclusion [62]. Although most cancers seem to exhibit minimal lymphocyte infiltration, macrophages are often among the first that invade colorectal tumors, followed by T and B cells. Those CRC tumors that have high densities of endothelial cells and fibroblasts tend to show worse prognosis [63].

Recent research has shown that T cell exhaustion is considered a mechanism of resistance for cellular immunotherapies [64–66]. All the five ARGs were correlated with T cell infiltration, highlighting their impact on T cell exhaustion. Our results with cell–cell communication analysis highlighted a dense network of interactions among fibroblasts, endothelial cells, and malignant cells, suggesting a coordinated regulation of angiogenesis and stromal remodeling. M2 macrophages are generally associated with angiogenesis and tissue repair. Studies on the roles of macrophages in the tumor microenvironment remains highly discordant; nonetheless, a number of investigations suggest that improved survival in CRC is linked to macrophage infiltration [67]. Almost all the ARGs were correlated with the macrophages at varied degrees suggesting the interplay between macrophages and ARGs to regulate the progression of CRC.

The findings may be made more robust and reliable by combining data from multiple sources, which can reveal cellular heterogeneity and phenotypes. Our extensive integration and characterization of cell subpopulations in the TME has produced a vast and comprehensive data that includes a variety of tumors and different technologies. This variety creates batch effects that are unique to each dataset also We did not experimentally verify the expression of ARGs in immune cells in the current study. Thus, future experiments can be designed to determine whether the immune cells express these ARGs. Future studies incorporating larger, more diverse CRC cohorts as available in public resources (e.g., TCGA) or prospective patient recruitment would enhance robustness, enable validation of existing findings, and reveal novel candidates. Overexpression or siRNA knockdown of top hub genes in CRC cell lines to assess effects on proliferation, apoptosis, migration, and immune pathway markers via qPCR, flow cytometry and immunohistochemistry. By acknowledging current limitations and outlining a systematic plan for cohort expansion and mechanistic validation, this study lays a clearer foundation for future translational and therapeutic exploration.

Collectively, our findings emphasize the multifaceted roles of MMP14, CXCL12, SPP1, TIMP1, and VCAN in CRC. Their expression not only reflects tumor progression and prognosis but also indicates involvement in immune modulation and angiogenesis. These genes may serve as potential biomarkers for CRC diagnosis and staging, as well as therapeutic targets for anti-angiogenic and immune-modulatory strategies^68,69.

Limitations

We acknowledge a number of limitations in our work, despite the comprehensiveness of our integrative single-cell analysis and our identification of significant cellular subpopulations. This study has several limitations. First, it primarily relies on public transcriptome datasets with relatively small sample sizes, which may affect the reliability of the model and overlook potential biomarkers. Larger clinical cohorts are needed to validate the diagnostic performance and clinical utility of the identified hub genes. Second, the lack of in vivo and in vitro experiments limits further investigation into the mechanisms of these genes. While our findings provide a theoretical foundation and new perspectives for treating angiogenesis in CRC, rigorous experimental validation is essential for the development of new drugs and therapies. This study limited itself to ARGs as potential genes; nevertheless, there are several interactions between other molecule types. Additionally, to support our results, expression of the five angiogenesis-related genes should be confirmed using colorectal cancer tissue samples and analyzed in relation to clinicopathological features and prognosis. The mechanisms underlying angiogenesis and immune infiltration in CRC remain unclear and require further exploration through experimental approaches. Future studies should involve overexpression or silencing of these genes in cellular and animal models to assess their roles in tumor progression, angiogenesis, and the tumor microenvironment. Therefore, large-scale, multicenter studies are necessary to validate and expand upon our findings. Despite its limitations, this study provides a solid scientific basis for future research and contributes to ongoing efforts in developing more effective treatments for invasive and metastatic cancers, particularly CRC. There are currently insufficiently annotated samples from patients undergoing immunotherapy in the publicly accessible databases of CRC gene expression data, such as TCGA and GEO. This has limited our ability to verify the model's prediction performance directly. Nevertheless, we indirectly evaluated the model's possible relevance for predicting immunotherapy responses by multidimensional validation using external datasets. In order to further improve the model's translational applicability in clinical settings, future research will be focused on resolving these limitations.

Conclusion

Overall, the study pinpointed certain molecular targets that influences TME, highlighting the potential impact of angiogenesis and immune infiltration in CRC progression. Our investigation, which was based on an array of bioinformatics algorithms, we identified hub genes and their association with immune infiltration cells in CRC. These hub genes exhibited significant upregulation in CRC tissues, particularly in advanced tumor stages, and their elevated expression levels were associated with poorer overall survival in COAD patients. Their expression also showed strong correlations with immune cell infiltration, particularly macrophages, dendritic cells, and fibroblasts, highlighting a potential role in modulating the TME. Single-cell RNA sequencing analyses confirmed their distinct localization to stromal and immune compartments, revealing diverse expression patterns across fibroblasts, endothelial cells, macrophages, and malignant epithelial cells. Importantly, pseudotime trajectory analyses demonstrated the involvement of these hub genes in fibroblast lineage differentiation toward malignant and immune cell phenotypes, suggesting their roles in driving cellular plasticity and immune suppression within the TME. The interplay between angiogenesis, immune evasion, and extracellular matrix remodeling underscores the multifaceted function of these genes in CRC progression.

In summary, MMP14, CXCL12, SPP1, TIMP1, and VCAN represent promising biomarkers for prognosis and potential therapeutic targets in CRC. Their expression profiles and functional roles in the TME warrant further validation in experimental models and clinical cohorts, which could pave the way for novel anti-angiogenic and immunomodulatory interventions in colorectal cancer management.

Abbreviations

CRC

Colorectal cancer

DEG

Differentially expressed gene

GEO

Gene expression omnibus

HR

Hazard ratio

logFC

Logarithmic fold change

FDR

False discovery rate

MSigDB

Molecular signature database

STRING

Search tool for the retrieval of interacting genes

PPI

Protein–protein interaction

MCC

Maximum clique centrality

TCGA

The cancer genome atlas

OS

Overall survival

ECM

Extracellular matrix

Author contributions

NB: Conceived, designed, and performed the Data analysis and interpretation, writing manuscript and reviewing; SB and CB: performed data analysis and manuscript writing; DS: Conception and designing of the study, data analysis & interpretation, overall supervision, manuscript writing, reviewing and editing. All authors approved the final version of the manuscript.

Funding

The authors received no funds from any grant to carry out this work.

Data availability

The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author. The public datasets GSE44076, GSE41258, GSE81558, GSE31279 and GSE50760 used in this paper are available on the NCBI website (https://www.ncbi.nlm.nih.gov/geo/).

Declarations

Ethics approval and consent to participate

In this study, we utilized publicly available databases and datasets. So ethical approval is not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.