Figure 1 exhibitied the workflow of our study. Microarray datasets of UC came from the GEO database (http://www.ncbi.nlm.nih.gov/geo), and GEO include numerous high-throughput datasets and expression microarray datasets. We searched related gene expression profiles and UC patients information with the key words “colon mucosal” and “ulcerative colitis” in the GEO database. We got the transcriptome data, patients information and healthy control data from the TCGA database. First, gene expression profiling should include case and control groups. Second, to maintain the accuracy of WGCNA, more than 10 samples were included in each dataset. Third, all the samples were colon tissues. Fourth, these gene matrices must provide raw data for further analysis. Finally, TCGA-COAD in the TCGA dataset and GSE87473, GSE87466, GSE39582, GSE92415 and GSE20916 in the GEO dataset were chosen for subsequent research. The probe expression matrix was converted into a gene expression matrix based on platform annotation files.

WGCNA explored gene module associations with disease traits by analyzing coexpressed gene modules with high biological significance (20). Hence, modules associated with UC and CRC were detected in the GSE87473 and TCGA-COAD datasets by using “WGCNA package”. In our study, we selected the first 10,000 genes with large variation based on their variance for further analysis. To omit unqualified genes and samples, Hclust (hierarchical clustering) analysis was performed in R prior to analysis. First, based on the criterion of R² > 0.85, a candidate soft-thresholding power β (ranging from 1-20) was computed for the scale-free topology using the pickSoftThreshold function. The subsequent step involved the construction of an adjacency matrix utilizing the soft power value and gene correlation matrix derived from Pearson analysis. This adjacency matrix was then transformed into a topological overlap and corresponding dissimilarity matrix. Subsequently, coexpressed gene modules were determined via hierarchical clustering, and subsequently, we acquired a dynamic tree. The minimum number of modules was 30. Finally, to determine each module’s expression profile, we calculated the module eigengene and its association with clinical features. Other parameters were as follows: mergeCutHeight = 0.2, deepSplit = 2 and networkType = “unsigned”.

For acquiring the potential inner mechanisms and functions of coexpressed genes, Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were performed with the “ClusterProfiler package”. GO included biological process (BP), cellular component (CC) and molecular function (MF) (21). KEGG is a comprehensive database containing valuable information on gene pathways (22). Furthermore, barplots and circos plots were employed to display the outcome of enrichment analysis.

The online search tool STRING (http://string-db.org) (version 12.0) was applied to search for interrelationship among protein-coding genes of interest (minimum required interaction score: 0.15). PPI network could be visualized via the utilization of Cytoscape (version 3.9.1) (23). As a plugin, MCODE was applied to filter critical functional interaction genes, with K-core=2, degree cut-off=2, max depth=100 and node score cut-off=0.2 (24). CytoHubba was utilized to figure out the hub genes. For the purpose of evaluating and selecting hub genes, five common algorithms (MNC, DMNC, degree, MCC and EPC) were applied. The subsequent step involved the utilization of GeneMANIA (http://www.genemania.org/), which is used for internal association detection within gene sets and for establishing coexpression networks for central genes (25).

Expression validation of these genes in GSE87466 and GSE39582 datasets, respectively. The GSE87466 dataset consisted of 21 normal people and 87 ulcerative colitis patients. GSE39582 is composed of 19 normal people and 566 CRC patients. The GSE87466 and GSE39582 datasets were obtained from human intestinal tract tissue. We used the t test to compare the two datasets.

With the “pROC” package, Receiver Operating Characteristic (ROC) curves were established in order to estimate the diagnosis capacity of key genes (26). The AUC values and corresponding 95% CIs were calculated, with an AUC > 0.7 indicating high diagnostic value.

To evaluate disease immune microenvironment and immune cell infiltration, the “cibersort” package was utilized (27). The distribution of immune cells across different samples was visually represented using a bar plot. Additionally, a correlation heatmap depicting the interrelationships among 22 immune cell types was generated utilizing the “corrplot” package. Vioplotting was conducted to compare the proportions of diverse immune cells in the UC and healthy cohorts. Finally, Spearman’s correlation was performed to reveal the relationships between hub biomarker expression and infiltrating immune cells.

The fresh colon tissues were fixed in 4% paraformaldehyde solution overnight, dehydrated and embedded in paraffin. Subsequently, paraffin section were cut into 5-μm sections and immersed in a hematoxylin-eosin blend for uniform staining. Using an microscope (Olympus, Japan), we observed and photographed colon tissue slices.

After fixation in 4% paraformaldehyde for 12 hours, Paraffin sections of the colon tissues were prepared. A thermostat set at 80°C was applied to heat the sections overnight. Afterwards, the sections were dewaxed in xylene and dehydrated in graded ethanol, followed by subsequent inactivation of endogenous peroxidase activity using a 3% H2O2 solution and then antigen retrieval with ethylenediaminetetraacetic acid. Subsequently, they were blocked with 5% goat serum (ABS933, Absin, Shanghai, China) for 30 minutes and incubated overnight at 4°C with Primary antibodies (anti-CXCL1[1:100, 12335-1-AP, Proteintech] and anti-BCL6 [1:1000, 66340-1-Ig, Proteintech]). The following day, the sections were incubated at 37°C without light for 1 hour with secondary antibodies (goat anti-rabbit [RGAR011, Proteintech] and anti-mouse [RGAM011, Proteintech]). After washing again with PBS, the chromogenic substrate diaminobenzidine (ZLI-9017, ZSGB-BIO, China) was applied to the sections. Staining was finished when positive cells exhibited deep staining while maintaining an unstained or lightly stained background under a microscope (Olympus, Japan). The paraffin sections were subsequently counterstained with hematoxylin followed by thorough rinsing in water. Dehydration was achieved using gradient alcohol and xylene before the samples were sealed appropriately in resin. Finally, the dried sections were photographed for analysis by ImageJ software and GraphPad Prism 8 software.

In liquid nitrogen, samples of colon tissues were thoroughly ground and in lysis buffer, contain protease inhibitors, samples of colon tissues were homogenized. The tissues were placed in a centrifuge tube and centrifuged at 12,000 rpm for 20 min at 4°C. The BCA approach was applied to measure the total protein concentration in the supernatant. We heated and denatured the protein solution at 100°C for 5 minutes. After removing solution, placed them on ice and stored at a temperature of 20°C. We further transferred the protein samples to a polyvinylidene fluoride (PVDF) membranes. The PVDF membrane was blocked with 5% skim milk. The membrane was washed twice with TBST for 5 minutes each and subsequently incubated overnight at 4°C with primary antibodies against CXCL1 (1:500, 12335-1-AP, Proteintech), BCL6 (1:1000, 66340-1-Ig, Proteintech), and β-actin (1:1000, 7A-09, ab8226, ZSGB-BIO). On the following day, the PVDF membrane was exposed to the secondary antibodies (goat anti-rabbit [RGAR001, Proteintech] and anti-mouse [RGAM001, Proteintech]) for 1 hour at 37°C. Subsequently, the protein bands were immersed in enhanced chemiluminescence solution, and images were captured using a gel imaging system.

After the colon tissues were rinsed with saline solution, they were gently dried using filter paper to eliminate any residual moisture. The tissue blocks were then carefully sectioned into 1 cm3 pieces and embedded in optimal cutting temperature compound (OCT). Frozen colon tissues were sliced into 5-μm sections, followed by treatment with 0.1% Triton X-100 to enhance cell permeability. After blocking with 5% goat serum (ABS933, Absin, China) for 30 minutes at a temperature of 37°C, primary antibodies, including anti-CXCL1 (1:100, 12335-1-AP, Proteintech) and anti-BCL6 (1:1000, 66340-1-Ig, Proteintech), were added and incubated overnight at 4°C. On the following day, goat anti-rabbit secondary antibody (RGAR002; Proteintech, China) and anti-mouse secondary antibody (RGAM002; Proteintech, China) were applied to the colon sections, which were then incubated for one hour at a temperature of 37°C in the dark. Nuclei staining was performed using DAPI (Beyotime, China) for 5 min before images were captured under a laser confocal microscope (Zeiss, Germany). The results were semiquantitatively analyzed with GraphPad Prism 8.2.1 and Image J software.

The statistical analysis and image preparation were completed via R studio (version 4.1.3) and GraphPad Prism 8.2.1, respectively. Continuous variables with a normal distribution are described as the mean ± standard deviation (SD). One-way analysis of variance was employed for data analysis involving more than two groups. Figures were edited using Adobe Illustrator (AI), PhotoShop (PS), and Image J. Statistical significance was defined when P < 0.05.

Five GEO datasets and one TCGA dataset, namely, GSE87473, TCGA-COAD, GSE87466, GSE92415, GSE39582 and GSE20916, were selected. Information on each dataset is shown in Table 1 . We further coupled the GSE87473 and TCGA-COAD cohorts as training sets for WGCNA and the other cohorts as verification sets to validate the expression of hub genes.

We identified 11 functional modules in GSE87473 by WGCNA between the UC and control groups. We chose an optimal β value = 20 (scale-free R² = 0.85) as the soft threshold according to the scale-free topology criterion and average connectivity ( Figure 2A ). The Spearman correlation coefficient was used to create a heatmap that illustrates the relationships between modules and traits. This allowed for an evaluation of the associations between the modules and the disease ( Figures 2C, E ). Three modules, namely, ‘skyblue’, ‘plum1’, and ‘gray60’, showed significantly positive correlations with UC and were considered UC-associated modules. (sky blue module: r= 0.52, p = 3e−10, genes= 116; plum1: r=0.52, p= 5e−10, genes= 90; grey60: r= 0.59, p = 5e−13, genes=169). By setting the soft threshold to 7 ( Figure 2B ), 32 modules were identified in TCGA-COAD, with the modules “blue” (r = 0.81, p = 2e−119, genes = 480), “pink” (r = 0.45, p = 1e−26, genes = 834) and “darkturquoise” (r=0.45, p = 2e−26, genes = 206) being significantly correlated with CRC ( Figures 2D, F ). A Venn diagram was constructed, and 33 intersecting genes were obtained ( Figure 3A ).

GO and KEGG analysis were applied to explore the underlying information and pathways associated with coexpressed genes. The genes were predominantly enriched in humoral immune response, molecular mediators of immune response production and biological processes related to immunoglobulin production. In CC, the genes were predominantly enriched in immunoglobulin complexes, immunoglobulin complexes and blood microparticles. Molecular function analysis demonstrated significant enrichment in antigen binding and immunoglobulin receptor binding ( Figures 3B, C ). Furthermore, the coexpressed genes were mainly involved in the NF−kappa B pathway, chemokine signaling pathway, TNF signaling pathway, and cytokine–cytokine receptor interaction ( Figure 3D ). Supplementary Tables S1 , S2 show the GO and KEGG analyses. These findings strongly suggest the relationship of proinflammatory factors and disease development.

We constructed a network of interactions among 33 overlapping gene proteins, with combined scores exceeding 0.15 ( Supplementary Figure S1 ). Eighteen nodes and twenty-six edges in the interaction network were visualized with Cytoscape ( Figure 3E ). The MCODE was applied to find subnetworks and gene cluster modules (scores: 4.0, 5 nodes and 10 edges) ( Figure 3F ).

We identified the top five crucial genes (CXCL1, BCL6, CXCL2, LCN2, and IRF4) from the PPI network ( Figure 3G ). Topological analysis algorithms (MCC, DMNC, EPC, MNC and Degree) were applied to comprehensively estimate and identify crucial genes in PPI networks. Detailed information on these genes is provided in Supplementary Table S3 ( Supplementary Table S3 ). Using the GeneMANIA database, we analyzed the coexpression networks and functions of these genes. Our findings revealed a complex protein-protein interaction (PPI) network comprising physical interactions (77.64%), coexpression patterns (8.01%), predicted associations (5.37%), colocalization relationships (3.63%), genetic interactions (2.87%), and pathway connections (1.88%), as depicted in Supplementary Figure S2 . The predicted genes could be found in the outer ring, whereas core genes were located in the inner ring. In Figure 3H , these genes were enriched in neutrophil migration, humoral immune response, cellular response to chemokines, regulation of T-cell differentiation, response to chemokines and cytokine activity.

Differential expression verification was performed in the validation cohort GSE87466 for UC and GSE39582 for CRC. We found that all hub genes (CXCL1, BCL6, CXCL2, and LCN2) showed predominant up-regulation in the UC and CAC patient cohorts in contrast with the controls ( Figures 4A–H ).

ROC curve analysis of the GSE92415 dataset revealed BCL6 (AUC: 0.883, CI: 0.817−0.939), CXCL1 (AUC: 0.955, CI: 0.918−0.984), CXCL2 (AUC: 0.935, CI: 0.878−0.977) and LCN2 (AUC: 0.965, CI: 0.925−0.995) ( Figures 4I–L ). Furthermore, GSE20916 was used to confirm the diagnostic ability of the hub genes BCL6 (AUC: 0.772, CI: 0.661−0.872), CXCL1 (AUC: 0.948, CI: 0.905−0.981), CXCL2 (AUC: 0.947, CI: 0.902−0.983) and LCN2 (AUC: 0.884, CI: 0.809−0.947) ( Figures 4M–P ). The AUC values of four of the five hub genes exceeded 0.85, which indicates that these genes have high diagnostic value.

In UC samples of GSE87473, macrophage infiltration was strongly correlated with CXCL1, BCL6 and CXCL2. In addition, CXCL1, BCL6 and CXCL2 were associated with the expression of activated dendritic cells and T regulatory cells ( Figure 5A ). Specifically, these genes were strongly related to M0 macrophages (BCL6, R=0.598450467, p=1.36E-13; CXCL1, R=0.487848406, p=6.87E-09; and CXCL2, R=0.402050537, p=3.06E-06) ( Figure 5B ). Moreover, all genes were positively linked to activated dendritic cells (BCL6, R=0.365708474, p=2.54E-05; CXCL1, R=0.426961846, p=6.17E-07; and CXCL2, R=0.349592704, p=6.01E-05) ( Figure 5B ). As shown in Figure 5D , the bar plot shows 22 different immune cell type proportions of each sample from the UC and control groups ( Figure 5D ). Violin plots revealed that the UC group had greater numbers of neutrophils, M0 macrophages, M1 macrophages, activated memory CD4⁺ T cells and activated mast cells and lower numbers of M2 macrophages, resting mast cells, CD8⁺ T cells and resting dendritic cells ( Figure 5C ). In addition, resting mast cells were positively related to M2 macrophages (r=0.54), and Monocytes were positively related to resting dendritic cells (r=0.53) and Regulatory T Cells (r=0.62) ( Figure 5E ). Moreover, M0 macrophages were negatively correlated with M2 macrophages (=-0.58), and activated mast cells were negatively linked to resting mast cells (r=-0.55) ( Figure 5E ).

Compared to the controls, the UC group expressed more CXCL1 (P < 0.001), while the CRC group expressed more CXCL1 (P<0.0001), and the CRC group expressed more CXCL1 than UC cohort did (P<0.01) ( Figure 6Q ). Moreover, the level of BCL6 in colon tissues demonstrated that the BCL6 in UC patients was markedly greater than CRC patients (P <0.001), CRC patients was greater than the control group (P <0.001) and in UC patients was greater than the controls (P<0.0001) ( Figure 6R ).

We applied immunohistochemical staining to determine the mRNA expressions of CXCL1 and BCL6 in nontumor adjacent, UC and cancer tissues ( Figures 6K–P ). CXCL1 in CRC group expressed greater than the UC group (P<0.05), CRC group was markedly greater than controls (P<0.001), and in UC was greater than the controls (P<0.05) ( Figures 6K–M, S ). The levels of BCL6 in UC patients were markedly greater than the CRC patients (P < 0.05) and the controls (P < 0.001). Moreover, the level of BCL6 in CRC patients was higher than controls (P < 0.01) ( Figures 6N–P, T ).

Immunofluorescence assays demonstrated that, compared with that in the UC cohort (P<0.001) and control group (P<0.0001), CXCL1 expression in cancer patients was greater and markedly upregulated. The level of CXCL1 in the UC patients was markedly greater than the controls (P < 0.0001) ( Figures 7A–J ). The level of BCL6 in the UC patients was greater than the cancer patients (P < 0.0001). In contrast with the controls, the level of BCL6 in the UC and cancer patients dramatically increased (P < 0.0001) ( Figures 7K–T ).