Diagnostic potential of the B9D2 gene in colorectal cancer based on whole blood gene expression data and machine learning

Abstract

Background

Colorectal cancer (CRC) is a malignant tumor marked by high prevalence and a challenging early detection landscape. While tools such as colonoscopy and serum biomarkers enhance screening efficacy, their invasive nature and inadequate sensitivity and specificity hamper their broad adoption. There is a pressing need for non-invasive, precise biomarkers for early diagnosis. The B9D2 gene, which is essential for ciliary function, has been rarely explored in CRC. This study is the first to investigate the diagnostic potential of B9D2 in CRC, using bioinformatics and machine learning to uncover its novel role in early detection, with implications for clinical translation.

Methods

Gene expression data from whole blood samples obtained from the GEO database were analyzed to identify differentially expressed genes (DEGs) associated with CRC, using a adjusted p-value threshold of < 0.05 and an absolute logFC > 0.5. The biological functions of these genes were investigated through Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), the Human Protein Atlas (HPA), and Gene Set Enrichment Analysis (GSEA). Additionally, three machine learning methods—Random Forest (RF), LASSO regression, and Support Vector Machine Recursive Feature Elimination (SVM-RFE)—were employed for feature selection and to evaluate the robustness and predictive power of the selected features, with diagnostic efficacy evaluated through Receiver Operating Characteristic (ROC) curves.

Results

Through this analysis, we identified five key genes—B9D2, CR2, DNMT3B, FOS, and PTGS2—from the GSE203024 dataset. Four of these genes have been previously linked to CRC, typically in tissue samples. Our study strengthens their significance as biomarkers by showing their expression in peripheral blood, a non-invasive source, and using multiple analytical methods. Notably, no previous studies have connected B9D2 to CRC, making this discovery a novel contribution. B9D2 expression was significantly upregulated in CRC patients, with an AUC of 0.797 in ROC analysis. This finding was further validated in the GSE47756 dataset, with an AUC of 0.756, confirming its potential as a reliable diagnostic biomarker for CRC. Further IHC staining showed significant different expression of B9D2 between normal and CRC tissue.

Conclusion

This study highlights the diagnostic potential of the B9D2 gene in CRC, marking the first time it has been proposed as a biomarker for early detection in CRC. This provides a foundation for its potential application in non-invasive diagnostic methods, such as liquid biopsy. Further experimental and clinical studies are needed to validate B9D2 as a reliable biomarker for early CRC detection and screening.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-03415-0.

Introduction

CRC arises from the epithelial lining of the colon and ranks as the third most common cancer globally. The insidious nature of early-stage CRC often leads to delayed diagnosis, as nearly 60% of patients remain asymptomatic until advanced stages [1, 2]. This diagnostic delay results in poor prognosis, accounting for approximately 900,000 deaths each year globally [3]. Studies demonstrate a striking survival disparity: while patients diagnosed at early stage exhibit a 95% five-year survival rate, this plummets to below 8% for late-stage disease [4]. Given this 12-fold survival difference between early and late diagnoses, improving detection timelines becomes paramount [5].

Current diagnostic paradigms face dual challenges: colonoscopy, while accurate, suffers from procedural invasiveness and poor population compliance. Conversely, non-invasive alternatives like FOBT demonstrate suboptimal sensitivity and specificity, creating a critical diagnostic gap that demands innovative solutions [6]. There is, therefore, an urgent need to develop a highly sensitive, specific, non-invasive, and convenient new screening approach for early CRC diagnosis to reduce mortality rates.

In recent years, transcriptomic profiling of peripheral blood has emerged as a promising non-invasive strategy for cancer detection, providing insights into systemic changes associated with tumor development [7]. Whole blood contains immune cell populations that respond to tumor-related signals, enabling detection of disease-associated gene expression signatures. Several studies have successfully employed blood-based transcriptomic data to identify potential diagnostic and prognostic biomarkers for CRC [8–10]. Nonetheless, most of these studies have focused on genes already known to be associated with CRC, and the discovery of novel, robust biomarkers—especially from blood samples—remains limited.

Traditional differential expression analysis can yield a large number of candidate genes, many of which may not be biologically or clinically relevant [11]. To address this, machine learning (ML) methods have become increasingly valuable in bioinformatics research, enabling the identification of key biomarkers with high predictive potential from high-dimensional datasets. Algorithms such as LASSO (Least Absolute Shrinkage and Selection Operator), Random Forest (RF), and Support Vector Machine Recursive Feature Elimination (SVM-RFE) are widely used for feature selection, as they help prioritize genes most strongly associated with disease status while minimizing noise and redundancy [12, 13]. These techniques enhance the reliability and clinical utility of DEGs by improving robustness, reducing overfitting, and capturing nonlinear patterns often missed by traditional statistical methods.

In this study, we applied bioinformatics and machine learning to whole blood gene expression data from CRC patients to identify novel diagnostic biomarkers, aiming to improve non-invasive early detection of colorectal cancer.

Methods

Data collection

Gene expression data were sourced from the GEO database (http://www.ncbi.nlm.nih.gov/geo), specifically from datasets GSE203024 (platform GPL570) and GSE47756 (platform GPL10558). Dataset GSE203024 contains whole blood gene expression data from various cancer types, including CRC. For our study, we used all CRC samples from this dataset, which includes 205 CRC samples and 296 healthy controls. Dataset GSE47756 focuses exclusively on CRC and includes 93 individuals: 38 healthy volunteers and 55 CRC patients, which are further classified into 27 non-metastatic CRC patients and 28 patients with metastatic CRC. All samples were used for validation in this study. Detailed sample information can be found in Supplementary Tables S1 and S2.

Identification of DEGs

The R package limma was employed to perform DEG analysis between the two groups of whole blood samples. Genes with |logFC| >1 and adjusted P value < 0.05 were defined as DEGs. The results were visualized with ggplot2 and pheatmap packages to generate volcano plots and heatmaps.

GO and KEGG pathway analysis

GO term identification and analysis were conducted in R using the clusterProfiler package, with gene identifiers set to “official_gene_symbol” and the species set to Homo sapiens. The org.Hs.eg.db package was used for human gene annotation, and the enrichplot and ggplot2 packages were employed for visualizing the enriched GO terms. The significance threshold for enrichment was set at p < 0.05. For KEGG pathway analysis, the clusterProfiler package was used to identify relevant pathways, and pathway diagrams were visualized using the pathview package [14, 15].

PPI network analysis

This study employed STRING (version 12.0) for PPI network analysis of DEGs. STRING includes 24,584,628 proteins from 5090 organisms, capturing both known and predicted interactions among over 932 million proteins, including humans [16]. The “multiple proteins” function was selected, specifying the species as “human.” Interactions were considered significant with a p-value < 0.05 and a confidence score > 0.7. The PPI network was visualized using Cytoscape software (version 3.9.0), and the CytoHubba plugin identified the top 50 hub genes as candidate DEGs for further analysis [17].

Selection of hub genes using machine learning

Feature selection and validation of the top 50 hub genes were conducted using RF, LASSO, and SVM-RFE algorithms in R. RF is effective for ranking feature importance, LASSO helps select the most predictive genes while preventing overfitting, and SVM-RFE optimizes feature selection by eliminating the least important ones. The combination of these methods ensures a robust and reliable identification of key genes [12, 13].

In the RF model, the number of trees was set to 500, with the maximum number of features per tree set to sqrt(p) and a minimum node sample size of 1. LASSO utilized α = 1, with λ determined through 10-fold cross-validation. SVM-RFE employed a radial basis function (RBF) kernel, with the penalty coefficient C determined through 10-fold cross-validation [18, 19]. The selection yielded 10, 30, and 15 candidate genes, respectively; a Venn diagram was generated using the “VennDiagram” package to identify the final five candidate genes. Boxplots and ROC curves were produced using the “ggpubr” and “pROC” packages, with a 95% confidence interval (CI) set.

Gene enrichment analysis

GSEA (www.gsea-msigdb.org/gsea/index.jsp) was performed on the up-regulated and down-regulated genes in the GSE203024 dataset and on the B9D2 gene. A p-value < 0.05 was considered statistically significant [20].

Immunohistochemical (IHC) staining

IHC images were obtained from the HPA database, which includes representative protein expression data from approximately 20 common cancers [21]. These images were utilized to assess the differential expression of the B9D2 gene in normal versus CRC tissue samples.

For IHC staining of clinical samples, formalin-fixed, paraffin-embedded (FFPE) tissue sections of CRC and matched adjacent normal tissues were collected from patients at our hospital. Section (4 μm thick) were deparaffinized in xylene and rehydrated through a graded ethanol series. Antigen retrieval was performed by heating the sections in citrate buffer (pH 6.0) at 95 °C for 20 min. Endogenous peroxidase activity was quenched using 3% hydrogen peroxide for 10 min at room temperature.

After blocking with 5% bovine serum albumin (BSA) for 30 min, sections were incubated overnight at 4 °C with a primary antibody against B9D2 (Polyclonal antibody, Proteintech, Cat# 22058-1-AP, 1:100 dilution). The next day, sections were incubated with a biotinylated secondary antibody and then with a streptavidin–HRP complex, followed by color development with diaminobenzidine. Finally, sections were counterstained with hematoxylin, dehydrated, and mounted.

Staining intensity and distribution were evaluated independently by two pathologists who were blinded to the clinical data. Expression levels were compared between tumor and adjacent normal tissues.

Statistical analysis

All data analyses were conducted using R (version 4.4.1) and its associated packages. ROC curve analysis was performed to evaluate diagnostic performance, with a 95% CI. Independent two-sample t-tests were employed to compute differences between groups, with significance set at p < 0.05 unless otherwise noted. For DEGs, p-values were adjusted using the Benjamini-Hochberg method to control the false discovery rate.

Results

Differential expression gene analysis

This study analyzed 296 healthy and 205 CRC whole blood samples from the GSE203024 dataset. Detailed clinical information is available in Supplementary Table S1. A total of 21,642 genes were analyzed before filtering, and 569 DEGs were identified relative to the control group, with 365 genes upregulated and 204 genes downregulated (Fig. 1A). The volcano plot displays these DEGs, with the top 10 up-regulated and down-regulated genes prominently labeled (Fig. 1B).

Fig. 1.

Differential Gene Expression Analysis. A. Heatmap displaying 569 DEGs in the GSE203024 dataset. B Volcano plot showing distribution of DEGs in the GSE203024 dataset, highlighting the top 10 upregulated and downregulated genes

Functional enrichment analysis of GSE203024 dataset

GSEA was conducted to assess the gene sets within the GSE203024 dataset, pinpointing key signaling pathways that are significantly upregulated or downregulated in CRC. The analysis revealed several enriched pathways, offering insights into the molecular mechanisms driving the development and progression of CRC.

A mountain plot was utilized to provide a comprehensive view of the enrichment of multiple pathways, emphasizing critical pathways such as the VEGF signaling pathway, TNF signaling pathway, chemokine signaling pathway, and ribosome biogenesis (Fig. 2A). VEGF signaling plays a critical role in angiogenesis and vascular permeability, facilitating tumor progression and metastasis [22]. The TNF signaling pathway promotes CRC progression by activating NF-κB and MAPK/ERK, enhancing cell survival, proliferation, and resistance to apoptosis, while also supporting metastasis through epithelial-mesenchymal transition (EMT) and immune cell infiltration in CRC [23]. The MAPK signaling pathway, driven by mutations in KRAS and BRAF, plays a crucial role in tumorigenesis and metastasis by promoting cell proliferation and invasion [24]. Overall, these pathways are crucial drivers of CRC development and progression.

Fig. 2.

Functional Enrichment Analysis of the GSE203024 Dataset using GSEA. A Ridge plot providing a comprehensive overview of the enriched biological pathways with GSEA analysis. B Diagram highlighting the top five upregulated pathways in CRC based on GSEA analysis. C Diagram highlighting the top five downregulated pathways in CRC based on GSEA analysis

Further focused on the top five upregulated and downregulated pathways implicated in CRC. Five pathways—methotrexate resistance, autophagy, bladder cancer, and the VEGF signaling pathway—were found to be significantly upregulated and are considered critical in promoting tumor growth and dissemination through mechanisms such as increased angiogenesis, resistance to autophagy, and drug tolerance. Conversely, the five down-regulated pathways—herpes simplex virus 1 infection, olfactory transduction, ribosome biogenesis, and the spliceosome—are linked to reduced cellular transcription and protein synthesis, suggesting these pathways may inhibit the progression of CRC (Fig. 2B, C).

These enriched pathways provide crucial insights into the molecular mechanisms of CRC, such as angiogenesis, tumor growth, and resistance to therapy, reinforcing their importance as potential therapeutic targets for the treatment of CRC.

Functional enrichment analysis based on DEGs

Based on the DEG from the blood samples of CRC patients and healthy individuals, we conducted a comprehensive functional enrichment analysis using GO and KEGG. This analysis explored changes related to biological processes (BP), cellular components (CC), molecular functions (MF), and signaling pathways.

In the BP enrichment analysis, processes significantly enriched included muscle system processes, vascular processes, and secretion regulation. These findings suggest that CRC may be closely associated with immune (e.g., humoral immune responses and acute inflammatory responses) and hormonal responses (e.g., responses to glucocorticoids and mineralocorticoids). Additionally, neuroregulatory processes (e.g., calcium ion responses and positive regulation of synaptic plasticity) were significantly enriched, indicating a potential role of neural signaling in cancer initiation and progression (Fig. 3A, B).

Fig. 3.

GO and KEGG Analysis Based on DEGs. A, B GO-BP enrichment analysis showing the top enriched biological processes associated with the DEGs identified in CRC. C, D GO-CC enrichment analysis highlighting the cellular components most affected by the DEGs. E, F GO-MF enrichment analysis revealing the key molecular functions of the DEGs. G, H KEGG pathway analysis showing the top enriched pathways and genes involved in CRC. For dot plots (A, C, E, G), the color bar represents the p-value, and the dot size reflects the number of enriched genes. For circos plots (B, D, F, H), the size of the dots indicates the number of enriched genes, and the color bar represents logFC value for DEGs. The color of the lines represents different pathways involved in CRC

In the CC enrichment analysis, components such as specific granules and their lumen, cytoplasmic vesicle lumen, and adhesion junctions were significantly enriched. These results underscore the critical role of immune cells and their secretory granules in the tumor microenvironment, highlighting complex interactions between cancer cells and the immune system (Fig. 3C, D).

The MF enrichment analysis showed significant enrichment of DEGs in functions like ligand binding, protein kinase activity, hormone binding, and chemoreception. These functional alterations suggest that genes in CRC may promote cancer cell survival and dissemination by activating signaling pathways and regulating metabolic pathways (Fig. 3E, F).

KEGG analysis identified pathways closely related to CRC, with significant enrichment observed in the PI3K-Akt signaling pathway, cytoskeletal regulation, and neuroactive ligand-receptor interactions. The enrichment of pathways related to substance addiction (e.g., cocaine and amphetamine addiction) indicates metabolic and neural signaling alterations associated with cancer (Fig. 3G, H, Supplementary Figure S1).

PPI network and identification of hub genes

To further explore the relationships between DEGs in blood samples from CRC patients and healthy individuals, this study utilized the STRING database and Cytoscape software (version 3.9.0) to construct a PPI network. The top 50 central node genes were selected as candidate DEGs for subsequent analysis (Fig. 4A, B).

Fig. 4.

The PPI network analysis and the identified hub genes. A PPI network based on DEGs. B Top 50 hub genes identified

Validation was conducted using three machine learning methods: RF, LASSO, and SVM-RFE

Based on the top 50 candidate genes from the PPI network analysis, we employed RF, LASSO, and SVM-RFE algorithms for feature selection and validation to identify marker genes highly correlated with CRC.

In the RF model, as the number of trees increased, both training and testing errors stabilized (Fig. 5A). The RF model identified key marker genes with high importance scores, which were visualized for the top 10 genes (Fig. 5B). These genes are likely critical in the onset and progression of CRC. The LASSO model indicated 30 genes, including B9D2, BRCA1, CBX8, CCL24, and CCR10, as closely associated with CRC (Fig. 5C, D). Furthermore, we applied the SVM-RFE method for recursive feature elimination to optimize the predictive model’s performance. The SVM-RFE model achieved the highest cross-validation accuracy (0.871) and the lowest error (0.129) utilizing 15 features (Fig. 5E, F).

Fig. 5.

Validation Using RF, LASSO, and SVM-RFE Machine Learning Models. A, B RF analysis for DEGs. Diagram showing the relationship between the total number of trees in the random forest and the error rates (A). Top 10 genes were visualized in the order of significance (B). C, D LASSO analysis for DEGs. Path diagram showing the coefficients of various genes (C). The lines represent the paths of different genes, indicating how their coefficients change with varying levels of regularization. Binomial Deviance plot for the LASSO model (D). The dashed vertical lines indicate the range of λ values within one standard error of the minimum deviance. E, F SVM-RFE model for recursive feature elimination. Plot showing the 5-fold cross-validation (5 x CV) accuracy as a function of the number of features (E). Plot showing the 5-fold cross-validation (5 x CV) error as a function of the number of features (F)

Validation of key genes and expression analysis

Based on the feature selection from three machine learning models—RF, LASSO, and SVM-RFE—five key genes were identified: B9D2, CR2, DNMT3B, FOS, and PTGS2 (Fig. 6A). Further validation assessed their diagnostic potential in CRC. The B9D2 gene was highlighted as a primary candidate for detailed analysis, while results for the other four genes are available in Supplementary Figure S2.

Fig. 6.

Expression and Validation of Candidate Genes. A Venn diagram illustrating overlaps among the three machine learning methods. B Box plot of B9D2 gene expression. C ROC curve for evaluating the diagnostic performance of B9D2 in the GSE203024 dataset

In this study, B9D2 showed significantly higher expression levels in CRC patients’ whole blood compared to healthy controls (Fig. 6B), demonstrating statistical significance (p < 0.05). Additionally, ROC curve analysis showed that the AUC for B9D2 was 0.797 (95% CI: 0.758–0.835), indicating strong diagnostic performance (Fig. 6C).

Among the other four genes, CR2 and DNMT3B were downregulated in CRC patients, with AUC values of 0.729 and 0.686, respectively. Conversely, FOS and PTGS2 were significantly upregulated, with ROC analyses yielding AUC values of 0.730 and 0.789, respectively (Supplementary Figure S2). Notably, aside from B9D2, the other four genes have been previously reported in CRC studies, further validating their potential as biomarkers or therapeutic targets.

Functional enrichment analysis of B9D2

Among these, CR2, DNMT3B, FOS, and PTGS2 have been well-documented in CRC literature. However, research on the B9D2 gene’s role in CRC is relatively scarce, making it a novel target for investigation. Furthermore, unlike the other four genes, which are generally involved in broader molecular signaling pathways, B9D2 is directly linked to ciliary structure and function. This unique feature provides a new angle in studying CRC, as ciliary dysfunction has been implicated in tumorigenesis through mechanisms such as immune evasion and tumor cell proliferation [25].

For the target gene B9D2, GSEA was employed to explore potential biological functions, and enrichment analysis of its up-regulated and down-regulated pathways was conducted using the KEGG database. B9D2 was significantly enriched in up-regulated pathways related to chemokine signaling, glutathione metabolism, glycan biosynthesis, starch and sucrose metabolism, and type 2 diabetes, all linked to metabolic and immune processes. These pathways suggest B9D2’s involvement in CRC progression through metabolic regulation and signal transduction (Fig. 7A).

Fig. 7.

Functional Enrichment Results for the B9D2 Gene. A Visualization of the top five upregulated enriched pathways. B Visualization of the top five downregulated enriched pathways

Conversely, B9D2 was enriched in down-regulated pathways related to antigen processing and presentation, graft-versus-host disease, autoimmune thyroid disease, and ribosome pathways, which are associated with immune responses and protein synthesis. This suggests B9D2 may contribute to the immune evasion mechanisms of CRC (Fig. 7B).

Validation of B9D2 in GSE47756 dataset

B9D2 was validated in the GSE47756 dataset, assessing its expression differences between healthy individuals and CRC patients, as well as its diagnostic value. The dataset was selected for its consistent data type with our analysis, and its well-structured sample groups, including both non-metastatic and metastatic CRC patients. Detailed clinical information for the dataset can be found in Supplementary Table S2. It was found that B9D2 expression was significantly higher in both the metastatic CRC group and the non-metastatic CRC group compared to the healthy group. However, no statistically significant difference was observed between the metastatic and non-metastatic CRC groups (Fig. 8A, B).

Fig. 8.

Expression and Validation of the B9D2 Gene. A, B Box plots showing the expression of B9D2 across the indicated groups. B ROC curve for assessing the diagnostic performance of B9D2 in the GSE47756 dataset

To further evaluate B9D2’s diagnostic value, a ROC curve was constructed. The results revealed an AUC of 0.756 (95% CI 0.655–0.848), indicating good predictive ability for distinguishing CRC patients from healthy individuals (Fig. 8C). These findings support B9D2’s considerable diagnostic potential within the GSE47756 dataset, endorsing its use as a biomarker.

B9D2 gene IHC staining results

In this study, IHC staining for the B9D2 gene utilized antibody HPA042618 from the HPA to investigate its expression in CRC and healthy colon tissues. The IHC staining of healthy colon tissue showed that B9D2 expression was relatively high, predominantly localized in glandular epithelial cells. Conversely, IHC staining of CRC tissue indicated that B9D2 expression was weak in cancerous tissues, primarily localized in specific areas, without significant up-regulation (Fig. 9A). A comparison of the images reveals that B9D2 expression is higher in normal tissue, particularly in glandular epithelial cells.

Fig. 9.

IHC Staining of the B9D2 Gene. A IHC staining showing the expression of B9D2 in normal tissue and CRC tissue from the HPA dataset. B IHC staining and quantification bar chart showing the expression of B9D2 in paracancerous and CRC tissue of the clinical samples from patients at our hospital

To further validate these findings, we incorporated clinical samples from patients at our hospital and performed additional IHC staining for B9D2. Consistently, B9D2 expression was significantly higher in paracancerous tissues compared to tumor tissues (Fig. 9B), supporting the initial observations from the HPA data.

Interestingly, this contrasts with the observed up-regulation of B9D2 in whole blood, suggesting that B9D2 expression patterns may vary between systemic circulation and tumor tissue. This disparity could be attributed to different regulatory mechanisms affecting B9D2 in systemic immune responses versus the localized tumor microenvironment, or it may indicate the distinct functions of B9D2 in various tissues.

Discussion

CRC is a highly prevalent and lethal malignancy worldwide, with early diagnosis challenges significantly contributing to its poor prognosis. Due to the subtlety of early symptoms, most patients are diagnosed at advanced stages, limiting treatment options and leading to high mortality rates [26]. Although colonoscopy is the gold standard for CRC diagnosis, its invasiveness, high costs, and low patient compliance limit its utility as a routine screening tool [27, 28]. Traditional methods like fecal occult blood tests (FOBT) and serum marker detection (e.g., CEA) lack the sensitivity and specificity required for early diagnosis, rendering them insufficient for mass screening [29]. Thus, there is an urgent need for an efficient, non-invasive, and convenient screening method to improve early detection rates and reduce mortality.

This study’s functional enrichment analysis of the GSE203024 dataset identified key molecular mechanisms in CRC. GSEA results highlighted significant up-regulation of the VEGF and MAPK signaling pathways in CRC. VEGF pathway activation in CRC is strongly associated with tumor progression and metastasis, as elevated VEGF levels promote angiogenesis, enabling tumor growth and invasion [23]. Similarly, MAPK pathway activation has been linked to increased cell proliferation and migration in CRC [30]. These insights not only highlight the importance of VEGF and MAPK pathways in CRC biology but also suggest that targeting these pathways could be a promising therapeutic strategy. The clinical significance of these pathways is further supported by the established use of VEGF inhibitors such as Bevacizumab and ongoing research into MAPK inhibitors for CRC treatment [31]. DEG analysis also revealed significant enrichment of immune-related BP, such as humoral immune responses and acute inflammatory responses, illustrating the immune system’s dual role in CRC progression. This dual role may support anti-tumor immune responses while also promoting cancer progression through chronic inflammation [32]. These findings align with recent advances in CRC immunotherapy (e.g., PD-1/PD-L1 inhibitors), suggesting that immune regulatory pathways could serve as effective therapeutic targets [33]. Notably, the enrichment of neuroregulatory pathways implies that the influence of neural signaling on CRC may be underestimated. Studies suggest that interactions between neurons and the tumor microenvironment might affect tumor progression through neurotransmitter modulation, an area deserving further investigation [34]. KEGG analysis also supported the mechanisms by which cancer cells enhance survival and dissemination through metabolic pathways and cellular structural changes, especially emphasizing the enrichment of the PI3K-Akt signaling pathway and cytoskeletal regulatory pathways [35].

Among the five genes ultimately selected in this study—B9D2, CR2, DNMT3B, FOS, and PTGS2—all have been previously reported to be closely associated with CRC. CR2, primarily known for immune regulation, may also influence tumor microenvironment modulation and immune evasion [36, 37]. DNMT3B regulates de novo DNA methylation and is linked to CRC initiation and the CpG island methylator phenotype [38, 39]. PTGS2, encoding COX-2, promotes inflammation and angiogenesis through NF-κB and MAPK signaling [40] and its up-regulation is linked to chronic inflammation in cancer patients and can serve as a prognostic marker [40–42]. FOS, involved in cell proliferation and apoptosis, interacts with Wnt signaling in CRC stem cells [43] and FOS is correlated with low expression and progression in CRC [44]. B9D2, a component of ciliary membranes, is implicated in the Hedgehog signaling pathway [45]. The correlation of these genes further validates the reliability and biological significance of the DEGs identified in this study.

For the B9D2 gene, significant up-regulation was observed in CRC patients, with consistent performance across the GSE203024 and GSE47756 datasets, achieving AUC values of 0.797 and 0.756, respectively, suggesting good diagnostic capabilities for distinguishing CRC patients from healthy individuals. However, research on the B9D2 gene is relatively limited, and its functions are not fully elucidated. This study suggests that B9D2 may be involved in ciliary function, crucial in various cellular signaling pathways, including those related to cancer signaling and immune microenvironment regulation [46]. Ciliary dysfunction may accelerate tumorigenesis by promoting immune evasion and tumor cell proliferation [47–50]. GSEA analysis of B9D2 showed significant involvement in metabolic and immune-related pathways, up-regulated pathways included chemokine signaling and glucose metabolism, suggesting that B9D2 may promote tumor cell survival and dissemination through metabolic regulation. Conversely, down-regulated pathways included antigen processing and presentation and immune evasion, indicating B9D2’s role in regulating tumor immune evasion.

An interesting observation in this study was the significant up-regulation of B9D2 in whole blood samples, whereas IHC analysis from the HPA database showed lower expression in CRC tissues compared to higher expression in healthy colon tissues.

This variation in B9D2 expression may be linked to the distinct regulatory mechanisms that govern gene activity in different environments. In the bloodstream, immune cells such as macrophages or dendritic cells could contribute to the observed upregulation of B9D2, whereas in the tumor microenvironment, factors such as hypoxia, inflammation, or immune evasion mechanisms may suppress its expression [51]. Additionally, the tumor microenvironment often harbors immune suppressive factors, which could inhibit the expression of B9D2 in cancer cells while still allowing their upregulation in peripheral blood [52].

Another key factor in this discrepancy could be differences in RNA stability. The stability of mRNA can vary between tissues, and this may affect the levels of B9D2 detected in CRC versus normal tissues. In the blood, B9D2 mRNA might be more stable or actively protected by immune cells, leading to higher observed expression [53]. In contrast, in tumor tissues, RNA degradation or post-transcriptional modifications could reduce gene expression levels. Additionally, the tumor microenvironment, with its distinct features such as hypoxia, inflammation, and immune suppression, could further influence the stability and expression of B9D2 mRNA, resulting in lower detectable levels in CRC tissue [54].

These findings also suggest that B9D2 may play distinct roles in different tissues, possibly linked to immune modulation or tumor suppression [55, 56]. Understanding the regulatory mechanisms governing B9D2 expression in both systemic circulation and tumor tissues could provide valuable insights into its potential as a biomarker or therapeutic target in CRC. Further research is needed to better elucidate the factors influencing B9D2 expression in the context of cancer and its broader implications in immune responses and tumor progression [57].

Limitation

Although this study highlighted the potential of B9D2 and other genes in CRC diagnosis, it has some limitations. First, while we performed immunohistochemical validation of B9D2 expression using clinical CRC and adjacent normal tissues, we did not further investigate the underlying molecular mechanisms or functional role of B9D2 in colorectal cancer progression. Future studies are warranted to explore the biological significance of B9D2 through in vitro and in vivo experiments, as well as to validate its clinical utility in larger, independent cohorts. Second, the relatively small sample size may affect the reliability of statistical results. Additionally, although the original dataset is from an American population and the validation dataset is from a European population, demographic differences such as age, gender distribution, and ethnicity could introduce variability in gene expression. Third, environmental and lifestyle factors, such as diet and exposure, may also affect the observed results. Furthermore, clinical pathological confounders, such as tumor staging, treatment interference, comorbidity screening, and other factors, may also contribute to variability in the data and should be considered in future studies. Finally, the differential expression of B9D2 in whole blood and tumor tissues highlights complex regulatory mechanisms, necessitating more in-depth molecular experiments to explore its specific roles in different tissues. Future research should aim to elucidate the molecular mechanisms of B9D2 and its role in the tumor microenvironment. Developing non-invasive diagnostic methods based on B9D2 may provide new tools for early CRC screening. Additionally, the potential application of B9D2 as a therapeutic target, particularly in tumor immunotherapy, warrants further investigation. Overall, the discovery of the B9D2 gene not only presents new avenues for early diagnosis of CRC but also establishes the groundwork for personalized and targeted treatment approaches.

Conclusion

This study identified five key genes with diagnostic potential in CRC through comprehensive bioinformatics analysis and machine learning methods: B9D2, CR2, DNMT3B, FOS, and PTGS2. The significant up-regulation of the B9D2 gene is notable, demonstrating its potential as a biomarker in CRC diagnosis. ROC curve analysis revealed that B9D2 exhibited high diagnostic performance across different datasets, further confirming its effectiveness in distinguishing CRC patients from healthy individuals. Importantly, B9D2 was identified through analysis of peripheral blood from both CRC patients and healthy individuals, highlighting its feasibility as a non-invasive biomarker for early CRC detection. This makes B9D2 a promising candidate for early diagnosis and monitoring of CRC progression through blood-based tests.

Moreover, the findings regarding CR2, DNMT3B, FOS, and PTGS2 align with existing literature, indicating that these genes are involved in various BP such as immune regulation, epigenetic control, cancer progression, and inflammation. Gene enrichment analysis underscored the key roles of B9D2 in metabolic and immune pathways, suggesting that it may influence CRC progression through the regulation of immune evasion and metabolic pathways. Despite the identification of several key genes with diagnostic and therapeutic potential, this study has limitations, including the lack of in vivo and in vitro experimental validation and a relatively small sample size. Therefore, future research should incorporate more experimental approaches to validate the specific functions of the B9D2 gene and its clinical application potential, especially in liquid biopsy and targeted therapies for CRC.

In conclusion, the results of this study provide new potential targets for early diagnosis and personalized treatment of CRC, particularly highlighting the translational potential of B9D2 as a liquid biopsy biomarker. This offers significant promise for clinical applications in CRC detection and management.

Author contributions

All authors contributed to the study conception and design. NL provided ideas and carried out research and design. Material preparation, data collection and analysis were performed by ZW, and YF. The first draft of the manuscript was completed by ZW, and YF., modified by HZ, and NL, and determined by NL. All authors read and approved the final manuscript.

Funding

2024 Zhengzhou Medical and Health Technology Innovation Guidance Program Project: Research on Exploring Novel Auxiliary Diagnostic Biomarkers for Colorectal Cancer Based on Fecal Microbiome Sequencing and Metabolomics Analysis(2024YLZDJH292). National Natural Science Foundation of China (82303343). Henan Province Medical Science and Technology Research Plan (LHGJ20230723).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.