Integrative systems biology identifies PRELP and TAGLN as prognostic biomarkers and immune modulators in colorectal cancer Authors: Amir Ali Judaki, Mehran Aghemiri, Zahra Yousefli, Sajjad Tavakkoli, Amir Taherkhani, Elham Nazari, Afsaneh Arefi-Oskouie Published: 2025 ABSTRACT AbstractAim:To identify prognostic biomarkers in colorectal cancer (CRC) using an integrative systems biology approach combining transcriptomics, network topology, and immune profiling.Background:Advanced CRC prognosis remains poor due to a lack of biomarkers capturing tumor-immune interactions. Current single-gene markers often overlook the network-based nature of tumorigenesis.Methods:We analyzed TCGA-COAD data (n = 463 tumor, n = 85 normal) using GEPIA3 with thresholds of |log_2FC| > 2 and adjusted p. value < 0.05. Protein-protein and gene-gene interaction networks were constructed via STRING (confidence > 0.4) and GeneMANIA. Hub genes were prioritized by intersecting degree and betweenness centrality using Cytoscape. Prognosis was assessed using cSurvival (Kaplan-Meier) and immune correlations via TISDB. Protein expression was verified in the Human Protein Atlas and external validation was performed to justify accuracy of candidate biomarkers.Results:Analysis revealed 992 DEGs (498 up-regulated and 494 down-regulated) enriched in extracellular matrix organization and innate immune pathways. Eight dual hub genes were identified; PRELP and TAGLN significantly predicted overall survival. High PRELP (p= 0.014) and TAGLN (p = 0.039) expression predicted poor prognosis. Both positively correlated with innate immune cells (NKT/NK cells) but negatively with activated CD4+ T cells, suggesting immune exclusion.Conclusion: PRELP and TAGLN are novel dual prognostic biomarkers linking stromal remodeling to immunosuppression in CRC. High tissue expression serves as a marker of desmoplastic stroma and poor survival, emphasizing context-dependent biomarker evaluation.Key Words: Colorectal cancer, Systems biology, Tumor microenvironment, PRELP, TAGLN, Prognosis Introduction Colorectal cancer is the third most commonly diagnosed malignancy and the second leading cause of cancer-related death worldwide (1). Global incidence is rising, particularly in developing countries adopting Westernized lifestyles. While early detection and curative surgery improve outcomes, many patients present with advanced disease and experience relapse despite adjuvant chemotherapy (2). Accurate prognostic biomarkers are urgently needed to stratify patients according to risk, guide treatment decisions, identify patients suitable for immunotherapy, and discover novel therapeutic targets (3). Traditional biomarker discovery has largely focused on single-gene associations. However, cancer is driven by complex intracellular and intercellular networks rather than isolated genetic events. Systems biology approaches, which integrate transcriptomics with protein–protein interaction (PPI) networks, offer a powerful framework to deconvolute this complexity (4). By analyzing network topology, researchers can identify "hub genes", central nodes that regulate essential biological processes, which often have higher prognostic penetrance than peripheral genes (5). Furthermore, the tumor microenvironment (TME), particularly the infiltration of immune cells, is a decisive factor in patient survival and immunotherapy response. Therefore, there is a specific research gap in identifying biomarkers that are not only topologically central but also predictive of the immune landscape (6). In this study, we addressed this gap by applying a comprehensive systems-biology workflow to TCGA data. Unlike previous studies that often overlook the intersection between network topology and immune infiltration, we specifically sought hub genes that satisfy three criteria: differential expression, network centrality, and dual prognostic/immunomodulatory capacity. We used GEPIA3 for differential gene expression analysis, DAVID for gene ontology enrichment, STRING and GeneMANIA for constructing PPI and GGI networks, Cytoscape for centrality analysis and hub gene identification, cSurvival for survival analysis, TISDB for immune correlation analysis, and the Human Protein Atlas for protein-level validation. Our goal was to systematically prioritize genes with strong network centrality, significant associations with overall survival, and meaningful correlations with immune infiltration. Through this workflow, we identified two key candidate biomarkers and discuss their biological functions, prognostic significance, and implications for CRC immunobiology and therapeutic targeting. Methods Study design and reporting guidelines This computational study followed the STREGA (STrengthening the REporting of Genetic Association studies) and MINSEQE (Minimum Information about a high-throughput SEQuencing Experiment) guidelines. These standards recommend clearly describing sample sizes, data sources, preprocessing steps, statistical thresholds, and analytical methods to enhance transparency and reproducibility. All public databases were accessed between May and September 2025. Data acquisition and preprocessing Gene expression profiles for colon adenocarcinoma (TCGA-COAD) and normal colon tissues from GTEx were accessed through GEPIA3, which integrates TCGA and GTEx datasets. A total of 463 tumor samples and 85 normal colon samples from GTEx were analyzed (7). GEPIA3 uses TCGA and GTEx expression data that have been uniformly reprocessed by the UCSC Xena project to reduce differences arising from distinct processing pipelines and enable tumor vs. normal comparisons across sources. Expression levels are provided in TPM and are analyzed after log₂ transformation as log₂ (TPM + 1) (8). Metadata were retrieved from cBioPortal and are shown in Table S1. Differential Expression Analysis Differential expression between tumor and normal samples was assessed using the limma method as implemented in GEPIA3. P-values were adjusted for multiple testing using the Benjamini-Hochberg method to control the false discovery rate (FDR). Genes with |log₂ FC| ≥ 2 and FDR < 0.05 were considered differentially expressed. These thresholds are standard in transcriptomic biomarker studies and prioritize genes with both substantial effect sizes and statistical significance. Functional Enrichment Analysis Gene ontology (GO) enrichment analysis was performed using DAVID (database for annotation, visualization, and integrated discovery). Up- and down-regulated genes were analyzed. Enrichment was considered significant at FDR-adjusted p < 0.05. Biological Process (BP), Molecular Function (MF), and Cellular Component (CC) categories were visualized using bar charts. Terms related to extracellular matrix organization, cell adhesion, immune response, and inflammatory signaling were particularly enriched. Network construction and hub gene identification To prioritize genes with high regulatory potential, we constructed two interaction networks. A Protein-Protein Interaction (PPI) network was constructed using STRING online database with a minimum confidence score of 0.4 (medium confidence), incorporating both physical and functional associations. A Gene-Gene Interaction (GGI) network was constructed using GeneMANIA, incorporating co-expression, co-localization, and genetic interactions (9). Both networks were visualized in Cytoscape (v3.10.3). The cytoHubba plugin was used to calculate centrality scores. We utilized both Degree and Betweenness centrality metrics to rank nodes. The top 30 genes from both the PPI and GGI networks were selected based on the principle of balancing specificity and sensitivity in hub gene identification; selecting the top 30 genes from each network provides sufficient representation of highly central nodes while maintaining a manageable set for intersection analysis and downstream functional validation, a threshold commonly employed in network biology studies to identify genes with robust topological importance across multiple centrality metrics. These top 30 genes from both networks were intersected via a Venn diagram to identify common "hub" genes that occupy central positions across multiple interaction modalities, thereby highlighting genes with consolidated functional importance in the biological system under investigation. Survival analysis The prognostic value of the identified hub genes was evaluated using cSurvival based on TCGA-COAD clinical data (n=463 with available survival data) (10). Patients were stratified into "High" and "Low" expression groups based on the median gene expression value. Kaplan–Meier survival curves were generated, and statistical significance was assessed using the log-rank test (p < 0.05). Hazard Ratios (HR) with 95% Confidence Intervals (CI) were calculated to quantify the risk magnitude. Immune correlation analysis The Tumor Immune System Database (TISDB) provides pre-computed correlations between gene expression and immune-cell abundances estimated from TCGA cohorts (11). We conducted a comprehensive analysis examining correlations between candidate gene expression and multiple immune cell types. The immune cells showing significant correlation with PRELP and TAGLN expression were identified and reported, including natural-killer-T (NKT) cells, natural-killer (NK) cells, activated B cells, and activated CD4⁺ T cells. Spearman correlation coefficients (ρ), 95% confidence intervals (CIs), and p-values were retrieved for these significant associations. Spearman correlation assumes a monotonic relationship; scatter plots with regression lines were generated. To improve interpretability, 95% CIs and sample size (n = 463) were reported alongside each correlation coefficient. Protein-level validation Immunohistochemical (IHC) validation was performed using images from the Human Protein Atlas (HPA). CRC tissue sections stained for the candidate biomarker TAGLN (using antibody CAB001447) were downloaded, as suitable images for PRELP were not available for illustration. Two independent observers assessed staining intensity, quantity of positive tumor cells, and subcellular location based on HPA annotations. Examples of medium staining (25-75% cells positive) and absent expression (0% cells positive) were selected to illustrate inter-patient heterogeneity in TAGLN expression. Antibody clone information, validation scores, and staining protocols are detailed in the HPA database. External validation in independent GEO datasets To assess the generalizability of findings and evaluate the robustness of candidate biomarkers, we conducted external validation across four independent gene expression datasets from the Gene Expression Omnibus (GEO) database. Four independent GEO datasets (GSE113513, GSE156355, GSE44076, and GSE200427) comprising 290 samples in total were analyzed for external validation of candidate biomarkers. Differential expression analysis was performed independently for each dataset using the GEO2R online analytical tool. To control for multiple testing, p-values were adjusted using the Benjamini–Hochberg False Discovery Rate (FDR) method. Genes were considered significantly differentially expressed when meeting both criteria: adjusted p-value < 0.05 and |log₂FC| ≥ 1, consistent with the statistical thresholds applied in the primary TCGA-COAD analysis. Expression levels of candidate genes were extracted separately for each dataset to enable direct comparison with discovery cohort findings. Data availability All datasets and analysis codes used in this study are publicly available. The TCGA-COAD’s RNA-seq data are available from the GDC Data Portal. GEPIA3, STRING, GeneMANIA, cSurvival, and TISDB are open-access web resources. Detailed parameters are listed above to facilitate reproduction. Results Identification of DEGs and functional enrichment Using GEPIA3 with the limma method and FDR correction, we identified 992 DEGs between 463 colon cancer and 85 normal samples. Of these, 498 genes were up-regulated and 494 genes were down-regulated (|log₂FC| ≥ 2, FDR < 0.05). GO enrichment analysis showed that down-regulated genes were predominantly associated with extracellular-matrix organization, collagen-containing extracellular matrix, and extracellular exosomes, whereas up-regulated genes were enriched in innate immune response, inflammatory response, cell adhesion, apoptotic processes, and positive regulation of cell proliferation. These terms underscore the dual importance of matrix remodeling and immune activation in CRC progression. The top GO categories are displayed in Figure 1. Gene ontology enrichment. Bar plots showing significant GO terms for DEGs across Biological Process (BP), Cellular Component (CC), and Molecular Function (MF) categories. Down-regulated genes (red bars) are predominantly enriched in extracellular matrix organization, collagen-containing extracellular matrix, and extracellular exosomes. Up-regulated genes (blue bars) are enriched in innate immune response, inflammatory response, cell adhesion, apoptotic process, and positive regulation of cell proliferation. Network analysis and hub-gene identification We constructed PPI and GGI networks for the DEGs. Calculating degree and betweenness centrality metrics identified the top 30 most central genes in each network, a threshold selected to maintain comprehensive representation of network topology while ensuring robust statistical power for intersection analysis. Intersection of these lists produced eight hub genes (LUM, FBLN1, FMOD, DPT, MFAP4, PRELP, MMP2, and TAGLN), indicating that these genes occupy central positions at both protein and gene-interaction levels. The identification of genes present in both networks highlights their consolidated importance across diverse interaction modalities, suggesting broader functional relevance in the biological system. These eight hubs include PRELP, TAGLN, and six other genes. The network workflow and Venn diagram are shown in Figure 2. Hub gene identification through network analysis. (A) Protein–Protein Interaction (PPI) network constructed using STRING showing interconnected nodes representing genes. (B) Gene–Gene Interaction (GGI) network constructed using GeneMANIA. (C) Venn diagram illustrating the intersection of the top 30 genes from the PPI network and the top 30 genes from the GGI network, identifying 8 common hub genes with consolidated topological importance across multiple interaction modalities. Prognostic value of PRELP and TAGLN Kaplan–Meier curves generated by cSurvival revealed that PRELP and TAGLN among the eight hub genes were significantly associated with overall survival (OS) in COAD. As shown in Figure 3, patients with low PRELP expression had significantly better survival than those with high expression (log-rank p = 0.014). For PRELP, patients with low expression (median split) exhibited better overall survival compared to those with high expression. For TAGLN, patients with high expression showed reduced overall survival compared to those with low expression (log-rank p = 0.039). Likewise, high TAGLN expression predicted poorer survival. These associations remained significant after FDR adjustment (PRELP FDR value = 0.034, TAGLN FDR value = 0.048). The Kaplan–Meier curves clearly demonstrate the prognostic stratification capacity of both genes, with 95% confidence intervals around survival estimates provided to establish the precision of these survival associations. Kaplan–Meier survival analysis. (A) Overall survival curve for PRELP comparing low expression (blue) versus high expression (yellow) patients (Log-rank p < 0.05). Patients with low PRELP expression demonstrate significantly improved overall survival. (B) Overall survival curve for TAGLN comparing low expression (blue) versus high expression (yellow) patients (Log-rank p < 0.05). High TAGLN expression correlates with reduced survival. Shaded areas represent 95% Confidence Intervals around the survival curves. Correlation of candidate genes with tumor-infiltrating immune cells Spearman correlation analysis was conducted to examine associations between PRELP/TAGLN expression and specific tumor-infiltrating immune cell types in COAD samples (Figure 4). A comprehensive screen of immune cell correlations revealed that statistically significant associations were concentrated in four immune cell types. Immune correlation analysis. Scatter plots showing Spearman correlation between gene expression and immune cell abundance in COAD. (A) PRELP expression versus NKT cell abundance (ρ = 0.492, p < 2.2 × 10⁻¹⁶). (B) PRELP expression versus activated CD4⁺ T cell abundance (ρ = -0.196, p = 2.39 × 10⁻⁵). (C) TAGLN expression versus NK cell abundance (ρ = 0.620, p < 2.2 × 10⁻¹⁶). (D) TAGLN expression versus activated B cell abundance (ρ = 0.306, p = 2.48 × 10⁻¹¹). (E) TAGLN expression versus activated CD4⁺ T cell abundance (ρ = -0.123, p = 0.00862). Each point represents a patient sample; solid lines indicate linear regression fits. 95% Confidence Intervals are displayed. PRELP expression demonstrated a strong positive association with NKT cells (ρ = 0.492, p < 2.2 × 10⁻¹⁶, 95% CI: 0.430–0.549). In contrast, PRELP expression showed a negative association with activated CD4⁺ T cells (ρ = -0.196, p = 2.39 × 10⁻⁵, 95% CI: -0.274 to -0.116). These correlations suggest that high PRELP expression is associated with a microenvironment enriched in innate-like lymphocytes (NKT cells) but depleted of activated adaptive immune cells (CD4⁺ T cells). Although these correlations are statistically significant, they represent moderate effect sizes, indicating subtle but consistent patterns across the patient cohort. TAGLN expression demonstrated a strong positive association with NK cells (ρ = 0.620, p < 2.2 × 10⁻¹⁶, 95% CI: 0.569–0.667). Additionally, TAGLN expression showed a moderate positive association with activated B cells (ρ = 0.306, p = 2.48 × 10⁻¹¹, 95% CI: 0.240–0.370). TAGLN expression was negatively associated with activated CD4⁺ T cells (ρ = -0.123, p = 0.00862, 95% CI: -0.214 to -0.030). The stronger correlations observed for TAGLN (particularly with NK cells, ρ = 0.620) suggest more prominent immunological associations. Together, these correlations imply that high TAGLN expression is associated with increased infiltration of innate and B-cell compartments but reduced infiltration of activated helper T cells. Protein-level validation in the human protein atlas Immunohistochemistry images from the Human Protein Atlas were examined to evaluate TAGLN protein expression in colon cancer tissues (Figure 5). A sample from a 46-year-old female patient (patient ID 1454) with colon adenocarcinoma displayed medium cytoplasmic/membranous staining, with 25-75% of tumor cells showing positive staining (intensity: moderate). In contrast, a sample from a 72-year-old male patient (patient ID 2096) with colon adenocarcinoma exhibited no detectable staining (intensity: negative; quantity: none). These images illustrate the inter-patient heterogeneity of TAGLN protein expression and corroborate the transcriptomic findings for this biomarker, though protein-level validation was limited to the available samples in the HPA database and should be expanded in future studies to include PRELP and additional cohorts. Protein-level validation in the human protein atlas. Immunohistochemistry images from colorectal cancer tissues showing differential protein expression of TAGLN (antibody CAB001447). (A) Sample from a 46-year-old female patient (patient ID 1454) with colon adenocarcinoma showing medium cytoplasmic/membranous staining (intensity: moderate) with 25-75% of tumor cells positive. (B) Sample from a 72-year-old male patient (patient ID 2096) with colon adenocarcinoma showing no detectable staining (intensity: negative; quantity: none). These images demonstrate inter-patient heterogeneity in TAGLN protein expression patterns and corroborate the transcriptomic findings at the protein level. External validation of PRELP and TAGLN expression across independent datasets To establish the robustness and generalizability of PRELP and TAGLN as colorectal cancer biomarkers, we conducted independent validation using four GEO datasets comprising a total of 290 samples. Differential expression analysis using GEO2R was performed on each dataset with consistent analytical parameters (p-value adjustment: Benjamini-Hochberg FDR, threshold: adjusted p < 0.05, |logFC| ≥ 1). External validation analysis demonstrated strong alignment between PRELP and TAGLN expression patterns in the independent GEO datasets and the primary TCGA-COAD cohort discovery findings. Results for each dataset are detailed in Table.1. Table 1 PRELP and TAGLN expression patterns in the independent GEO datasetsDataset TAGLN log 2 FC PRELP log 2 FC Sample DetailDataset TypeTCGA-COAD-3.54-3.92463 tumor sample/ 85 normal sampleRNA-seqGSE113513-1.35-1.7714 tumor samples/ 14 normal samplesMicroarrayGSE156355-1.00-2.096 tumor samples/ 6 normal samplesMicroarrayGSE44076-1.61-1.6798 tumor samples/ 98 normal adjacent tissueMicroarrayGSE200427-2.93-1.802 tumor samples/ 2 normal samplesRNA-seq PRELP and TAGLN expression patterns in the independent GEO datasets In the TCGA-COAD discovery cohort, PRELP exhibited a log₂FC = -3.92 (down-regulated in tumors) and TAGLN demonstrated a log₂FC = -3.54 (down-regulated in tumors). Across all four independent GEO datasets, both genes demonstrated consistent down-regulation in tumor samples compared to normal tissue, establishing directional consistency across distinct cohorts and expression profiling platforms. The consistency of PRELP and TAGLN down-regulation across all four independent validation datasets, spanning different expression profiling platforms and distinct patient populations, strongly supports the robustness of these genes as candidate biomarkers for colorectal cancer. Discussion By integrating transcriptomic, network, and clinical data, this study prioritized PRELP and TAGLN as dual prognostic and immunomodulatory biomarkers in colon cancer. Both genes emerged as hubs in PPI and GGI networks, were significantly associated with overall survival, and correlated with specific immune-cell populations. To our knowledge, this is the first study to combine these criteria and to report their associations with immune infiltration in a single analysis. PRELP represents a particularly intriguing case of context-dependent gene expression with seemingly contradictory roles in cancer biology. In our analysis, low PRELP expression was associated with better overall survival in COAD patients, which appears to contradict the established literature characterizing PRELP as a tumor suppressor when expressed in epithelial cancer cells. This apparent paradox can be resolved by considering the cellular source of PRELP expression (12, 13). Single-cell transcriptomic studies have revealed that PRELP expression in colon cancer is not uniformly associated with tumor suppression. Instead, PRELP+ cancer-associated fibroblasts (CAFs) represent a distinct stromal population predominantly enriched in metastatic liver lesions. These PRELP+ CAFs exhibit a transcriptomic signature characterized by activation of pro-tumor pathways including TGF-β and Wnt signaling, and their presence correlates with poor prognosis and an immunosuppressive tumor microenvironment (14). While PRELP loss in epithelial cancer cells promotes EMT and activates oncogenic pathways (supporting its tumor-suppressive role), bulk gene expression measures capture PRELP transcripts from both epithelial and stromal sources. In metastatic/advanced colon cancer, the pro-tumoral PRELP+ CAF population may dominate, conferring poor prognosis at the tissue level. This context-dependency highlights the limitation of bulk transcriptomic analyses and the necessity of single-cell approaches to distinguish cellular sources of gene expression (15). The positive correlation between PRELP expression and NKT-cell abundance, combined with negative correlation to activated CD4⁺ T cells, suggests that high PRELP expression may be associated with a microenvironment exhibiting reduced adaptive immunity. PRELP+ CAFs may influence immune composition through secretion of immunomodulatory factors, ECM remodeling that alters immune-cell trafficking, or production of metabolites that suppress T-cell activation. Further experimental validation using in vitro co-culture systems and conditional knockout mouse models is required to establish causality (14). In contrast to PRELP, TAGLN's role as a poor prognostic factor in colon cancer is well-established and unambiguous (16), (17). Our finding that high TAGLN expression predicts worse overall survival is consistent with prior studies demonstrating that transgelin promotes proliferation, invasion, and metastatic potential. Mechanistically, transgelin acts as an actin-binding protein that remodels the cytoskeleton to enhance cell motility and invasion, and it interacts with PARP1 to modulate Rho signaling, a pathway critical for metastasis (18). The strong positive correlation between TAGLN and NK-cell abundance (ρ = 0.620) is noteworthy. Although high NK-cell infiltration is generally associated with improved survival in solid tumors, the association between TAGLN and NK infiltration may reflect a compensatory immune response to high TAGLN expression and tumor aggressiveness, rather than a protective effect. In other words, tumors with high TAGLN expression may provoke enhanced NK-cell recruitment in an attempt to control aggressive disease, but this innate immune response proves insufficient to overcome TAGLN-driven progression. This interpretation is supported by the concurrent negative correlation between TAGLN and activated CD4⁺ T cells, suggesting that high TAGLN expression is associated with a microenvironment deficient in adaptive immunity. Both PRELP and TAGLN negatively correlate with activated CD4⁺ T-cell abundance. CD4⁺ T cells are central to antitumor immunity, providing help to cytotoxic T lymphocytes and B cells and orchestrating coordinated immune responses (19), (2). The reduced infiltration of activated CD4⁺ T cells in tumors with high PRELP or TAGLN expression may contribute to poor patient outcomes. However, this association could reflect tumor-cell composition, leukocyte infiltration patterns driven by other factors, or sampling bias rather than direct functional immune modulation. Mechanistically, PRELP and TAGLN might influence immune-cell trafficking through several speculative mechanisms: ECM proteoglycans like PRELP interact with heparan sulfate and bind chemokines, serving as co-receptors that present chemokines to immune cells and establish chemotactic gradients. Altered PRELP levels or modifications could disrupt chemokine-mediated recruitment of T cells, selectively affecting the trafficking of different T-cell subsets. High PRELP expression by CAFs might sequester chemokines locally, preventing effective T-cell infiltration into tumor core regions while allowing innate lymphocyte infiltration through alternative pathways (20). PRELP functions as a basement membrane anchor through interaction with collagens and perlecan. Remodeling of the ECM by PRELP-expressing CAFs could alter expression or availability of adhesion molecules (e.g., ICAM-1, VCAM-1) required for T-cell extravasation. TAGLN, as an actin-binding protein and key regulator of cellular contractility, could influence the expression or surface presentation of integrins and selectins on stromal and immune cells, affecting their interaction and trafficking capacity (12), (21). High PRELP expression and TAGLN-mediated cytoskeletal remodeling may establish a physical microarchitecture of tumors, creating dense collagen-rich barriers that impede T-cell infiltration into core tumor regions. A stiffened, PRELP-rich extracellular matrix is known to sequester immune cells at the tumor-stromal interface, creating an immune-excluded phenotype (22). CAF-derived PRELP and cytoskeletal remodeling may facilitate metabolite sequestration (e.g., glucose, glutamine), creating a nutrient-depleted microenvironment unfavorable for activated T-cell expansion. These mechanisms are speculative and require experimental validation through in vitro co-culture systems, in vivo functional studies, and mechanistic dissection of chemokine gradients, adhesion molecule expression, ECM composition, and metabolic states in PRELP-/TAGLN- high versus PRELP-/TAGLN- low tumors (23). Recent bioinformatics analyses of colon cancer have identified multiple candidate biomarkers through transcriptomic, proteomic, and genomic approaches. Comprehensive bioinformatics analyses have characterized molecular subtypes based on gene-expression profiles and identified distinct immune-infiltration patterns. However, few studies have explicitly integrated hub-gene identification from network analysis with survival prediction and immune-cell correlations in a unified framework. Our systems-biology approach provides several methodological advantages: It reduces dimensionality through network analysis to prioritize genes with broad functional importance rather than isolated associations; it combines multiple independent databases (STRING, GeneMANIA, cSurvival, TISDB) to triangulate evidence across diverse data sources; it explicitly examines immune correlations to identify biomarkers with mechanistic relevance to tumor-immune interactions; and it addresses the PRELP paradox by acknowledging context-dependent expression roles and the limitations of bulk transcriptomic analyses. Conclusion We identified PRELP and TAGLN as dual prognostic and immune-associated biomarkers in colon cancer. Their expression patterns stratify patients according to survival risk and correlate with specific immune-cell populations, likely reflecting a dense, immunosuppressive stromal microenvironment. These genes represent promising targets for stromal-depletion therapies or strategies to enhance immunotherapy efficacy in colorectal tumors. Limitations This study has limitations inherent to in silico analyses. First, the results are based on retrospective public data; prospective validation in an independent clinical cohort is required. Second, determining survival groups via median split may obscure non-linear relationships. Third, bulk RNA-seq cannot definitively distinguish between tumor and stromal expression without single-cell resolution. Finally, correlation does not imply causation; the observed immune associations may reflect tumor purity or sampling bias rather than direct functional modulation. Conflict of interests There is no conflict of interest for authors of this article. References Associated Data Data Availability Statement All datasets and analysis codes used in this study are publicly available. The TCGA-COAD’s RNA-seq data are available from the GDC Data Portal. GEPIA3, STRING, GeneMANIA, cSurvival, and TISDB are open-access web resources. Detailed parameters are listed above to facilitate reproduction.