The flowchart of the study is shown in Figure 1 . Mendelian randomization (MR) analysis was performed using the MRbase online platform (https://app.mrbase.org) (21) to explore the causal relationship between UC and PD. UC data were sourced from several datasets available through the European Bioinformatics Institute (EBI), including ebi-a-GCST000964, ebi-a-GCST003045, ebi-a-GCST90018933, ebi-a-GCST90020072, as well as from the IEU OpenGWAS project, including ieu-a-32, ieu-a-968, ieu-a-970, and ieu-a-973. PD data were sourced from the finn-b-K11_PERIODON_CHRON dataset. All datasets are derived from European populations. None of the single nucleotide polymorphisms (SNPs) associated with PD met the threshold of P < 5 × 10⁻⁸; therefore, SNPs with P < 5 × 10⁻⁶ were selected as instrumental variables (IVs), following the criteria of previous studies (13, 22). For UC, SNPs with P < 5 × 10⁻⁸ were employed as IVs. Subsequently, MR analysis was performed on the website.

The combined odds ratios (OR) and 95% confidence intervals (CI) were calculated using R (version 3.6.0) with the Meta package (version 4.13-0). A fixed-effect model was applied to generate the forest plots. Data heterogeneity was assessed using the I² statistic, with I² ≥ 50% indicating high heterogeneity. Bias was evaluated using the Begg and Egger tests. A fixed-effect model was chosen based on low heterogeneity (I² < 50%) among studies.

Microarray data for UC and PD were obtained from eight GEO databases, as detailed in Supplementary Table S1 . All datasets underwent standardized preprocessing including background correction, log2 transformation, and quantile normalization using the “limma” package in R. Datasets selected for inclusion in this study were required to meet the following strict criteria: Adequate sample size: datasets with ≥10 samples per group (patients and controls). Clearly annotated clinical status: confirmed diagnoses of UC or PD and matched healthy controls. Transcriptomic profiles derived specifically from inflamed mucosal tissues (colonic tissues for UC and gingival tissues for PD). Availability of raw or properly normalized expression data suitable for integration and further analysis. Previous utilization in peer-reviewed immunological research studies to ensure biological relevance. Datasets were excluded if they had incomplete clinical annotation, insufficient sample size (<10 per group), data derived from non-mucosal tissues, or incompatible normalization status. For the analysis, data from GSE107499 and GSE16134 were preprocessed and normalized. Differentially expressed genes (DEGs) between disease and control groups were identified using the “limma” package in R, a standard tool for analyzing microarray data.

To visualize the DEGs, we used the “ggplot2” package to generate volcano plots and heatmaps. DEGs were considered statistically significant if they met the following criteria: adjusted p-value (adj. P) < 0.05 and absolute log2 fold change (|log2FC|) > 1. To identify common DEGs across the datasets, we intersected the DEGs from the two databases using a Venn diagram tool.

To further investigate the biological significance of the shared DEGs identified from the datasets, functional enrichment analyses were performed. The analyses included Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis, and Gene Set Enrichment Analysis (GSEA). All analyses were conducted using the “clusterProfiler” R package (23), and the “Hallmarks” pathway was selected for GSEA. To control for multiple hypothesis testing, false discovery rate (FDR) correction was applied using the Benjamini-Hochberg method in all enrichment analyses, including GO, KEGG, and GSEA. Enriched terms or pathways with an adjusted p-value (FDR) < 0.05 were considered statistically significant.

To identify immune-related differentially expressed genes (IRDEGs), we obtained a comprehensive list of immune-related genes from the ImmPort database (https://www.immport.org/home), which provided a total of 2483 immune-related genes (IRGs). Then, we intersected the list of IRGs with the up-DEGs from UC and PD through Venn diagram tool.

We employed three machine learning methods: Random Forest (RF), Least Absolute Shrinkage and Selection Operator (LASSO), and Support Vector Machine Recursive Feature Elimination (SVM-RFE). For RF algorithm (24), we used the randomForest package in R with default parameters and set the number of trees to 500. The importance score of each gene was evaluated based on the mean decrease in accuracy and Gini index. For LASSO analysis, the glmnet package in R was utilized with cross-validation to determine the optimal regularization parameter (λ). Genes with non-zero coefficients were identified as key predictors (25). The e1071 package in R was used to perform SVM-RFE with a radial basis function (RBF) kernel. The number of features was sequentially reduced until the optimal number of features was determined (26). Hub genes identified by RF, LASSO, and SVM-RFE in both UC and PD datasets were intersected to find common genes across methods and conditions. Then, CXCL6 was highlighted as a critical gene due to its consistent identification across multiple machine learning approaches, suggesting its potential as a significant biomarker or therapeutic target in UC and PD.

CXCL6 expression levels were validated in PD using GSE10334 and in UC using GSE47908. The sensitivity and specificity of CXCL6 as a diagnostic marker were evaluated using operating characteristic curves (ROC) and the area under the curve (AUC) by “pROC” package.

Based on the median expression value of CXCL6, UC and PD patients were categorized into different subgroups. DEGs between these subgroups were analyzed using the “limma” package, followed by Gene Set Enrichment Analysis (GSEA) (27) with Hallmark pathways. Additionally, immune infiltration and immune function across subgroups was assessed using single-sample GSEA (ssGSEA) using “GSVA” package, as previously mentioned (2).

Vedolizumab (VDZ) and Infliximab (IFX) are the most commonly used biologics for UC patients. We compared CXCL6 expression between responders and non-responders at baseline prior to biological therapy. The value of CXCL6 in predicting the efficacy of biological therapy was assessed using ROC curves and AUC. Additionally, we analyzed the changes in CXCL6 expression before and after treatment in responders.

Male Sprague-Dawley rats (6–8 weeks old, weighing 200–250 g) were obtained from Beijing Huafukang Biotechnology Co., Ltd. (Beijing, China). All animals were housed in specific pathogen-free (SPF) conditions with a 12 h light/dark cycle and had ad libitum access to standard chow and water. The study protocol was approved by the Animal Ethics Committee of The Fifth Hospital of Shanxi Medical University (approval number: No. 361 of 2025).

To investigate the impact of periodontitis on the development of colitis, rats were randomly assigned to two experimental groups (n = 6 per group). In the DSS group, ulcerative colitis was induced by administering 3% dextran sulfate sodium (DSS)(MP Biomedicals, USA) in the drinking water for 7 consecutive days. In the UC + periodontitis group, rats underwent experimental periodontitis induction through repeated tooth abrasion. Specifically, the occlusal surfaces of the maxillary left molars were gently ground using a low-speed dental drill every 2–3 days for a total of 23 days. During the final 7 days of this period, the rats also received 3% DSS in their drinking water to induce colitis, concurrently modeling both periodontitis and UC.

Colon tissues were fixed in 4% paraformaldehyde, embedded in paraffin, sectioned at 5μm, and stained with hematoxylin and eosin (H&E). The scoring criteria included inflammatory cell infiltration, mucosal damage, glandular architecture distortion, and ulceration extent, each graded on a 0–4 scale, with a total maximum score of 16. Higher scores indicated more severe histopathological damage.

Paraffin-embedded colon tissues were sectioned at 5μm thickness for immunohistochemical analysis. After standard deparaffinization, rehydration, antigen retrieval, and blocking, sections were incubated overnight at 4°C with primary antibodies targeting Occludin, ZO-1, Claudin-1, and CXCL6 (all from Affinity Biosciences, China). The next day, slides were incubated with HRP-conjugated secondary antibody, followed by DAB development and hematoxylin counterstaining.

Stained sections were examined under a light microscope. Immunoreactivity was evaluated using the Hiscore semi-quantitative grading system, which assesses both the staining intensity and the proportion of positively stained cells across representative high-power fields.

All statistical analyses were conducted using R software (version 3.6.0). Two-sided tests were applied throughout the study. Wilcoxon rank-sum tests were employed for non-normally distributed data, while Student’s t-tests were used for normally distributed data. A p-value of less than 0.05 was deemed statistically significant.

Among the eight UC databases used as exposures, MR analysis results were available for seven, with three databases (ebi-a-GCST003045, ieu-a-968, ieu-a-970) indicating that UC is a risk factor for PD, as assessed using the Inverse Variance Weighted (IVW) method. The meta-analysis demonstrated that UC is a risk factor for PD (OR, 1.04; 95% CI, 1.02–1.05; P = 0.001), with an I² of 4.58%. Both Egger’s test (P = 0.802) and Begg’s test (P = 0.881) revealed no evidence of bias ( Figure 2A ).

For UC as the outcome, MR analysis results were available from five databases, all of which showed that periodontitis is not a risk factor for UC. The meta-analysis yielded an OR of 1.02 (95% CI, 0.97–1.08; P = 0.47), with an I² of 0%. Neither Egger’s test (P = 0.724) nor Begg’s test (P = 0.624) indicated bias ( Figure 2B ).

The differential gene expression analysis of the GSE107499 dataset identified 865 upregulated and 387 downregulated genes in UC ( Figure 3A ). A heatmap ( Figure 3B ) visualizes these expression profiles across UC and HC samples. For GSE16134 (PD data), the analysis revealed distinct upregulated and downregulated gene sets ( Figures 3C, D ).

A Venn diagram ( Figure 3E ) shows 84 upregulated and 6 downregulated genes common to both conditions, totaling 90 shared genes. These were subjected to functional enrichment analysis, revealing significant involvement in immune-related processes such as B cell activation and activation of immune response ( Figure 3F ). KEGG pathway analysis ( Figure 3G ) highlights the involvement of these genes in key immune pathways, including TNF signaling pathway and IL-17 signaling pathway, both crucial in UC and PD pathogenesis.

The Venn diagram illustrates the intersection between upregulated DEGs and immune-related genes (IRGs), identifying 25 candidate genes ( Figure 4A ). LASSO analysis of the GSE107499 dataset (UC data) identified 11 genes ( Figure 4B ). SVM-RFE analysis selected 14 genes with notable accuracy (0.953) and a minimal error rate (0.0468) ( Figure 4C ). For RF analysis, genes with an importance score greater than 1 are considered significant ( Figure 4D ). For the GSE16134 dataset (PD data), LASSO analysis identified 10 genes ( Figure 4E ), while SVM-RFE analysis selected 11 genes with an accuracy of 0.933 and a minimal error rate of 0.0668 ( Figure 4F ). Random Forest results for PD are shown in ( Figure 4G ). The intersection of the results from these machine learning methods across both datasets led to the identification of CXCL6 as the hub gene ( Figure 4H ).

The reverse cumulative distribution plot and boxplots illustrate the residual expression patterns of machine learning models in UC, highlighting model variations ( Figure 5A ). Diagnostic performance metrics show high accuracy for UC, with AUC values of 0.979 for RF, 0.958 for SVM, and 0.983 for LASSO ( Figure 5B ). Similarly, residual expression patterns for machine learning models in PD are shown ( Figure 5C ), with the models demonstrating significant diagnostic value for PD ( Figure 5D ).

We further validated CXCL6 expression in disease and control groups using the GSE47908 dataset (UC data) and the GSE10334 dataset (PD data). ROC curve analysis demonstrated the diagnostic potential of CXCL6. The results indicate that CXCL6 is significantly upregulated in both UC ( Figure 6A ) and PD patients ( Figure 6B ) and exhibits strong diagnostic value.

We divided UC patients from the GSE107499 dataset into CXCL6 high-expression and low-expression groups based on the median expression of CXCL6. Differential gene analysis between these groups is shown ( Figure 7A ). GSEA using Hallmark pathways revealed that the CXCL6 high-expression group had higher activation levels in immune pathways, such as TNF signaling via NF-kB and inflammatory response, compared to the low-expression group ( Figure 7B ). Immune infiltration analysis indicated a significant increase in immune cells involved in UC development, such as macrophages and neutrophils, in the CXCL6 high-expression group ( Figure 7C ). Similarly, various immune-related pathways, including APC stimulation and IFN response, were enriched in the CXCL6 high-expression group ( Figure 7D ).

In a comparable manner, the PD cohort from the GSE16134 dataset was analyzed for CXCL6 high-expression and low-expression groups, revealing 1397 upregulated and 2111 downregulated genes ( Figure 8A ). GSEA showed that the CXCL6 high-expression group exhibited significant enrichment in immune pathways ( Figure 8B ). Immune cell infiltration analysis also indicated a notable increase in immune cell levels in the CXCL6 high-expression group ( Figure 8C ). Additionally, immune function scoring demonstrated elevated immune activity in the CXCL6 high-expression group ( Figure 8D ).

In the analysis of GSE16879 ( Figure 9A ), GSE23597 ( Figure 9B ), GSE73661 ( Figure 9C ), and GSE73661 ( Figure 9D ), CXCL6 expression was assessed to evaluate its potential as a predictive biomarker for biologic treatment efficacy, particularly for VDZ (AUC > 0.9). The results revealed that patients with elevated CXCL6 levels, indicating a higher inflammatory burden, were more likely to be non-responders to biologics. Furthermore, in responders, both IFX and VDZ treatments significantly reduced CXCL6 expression. These findings highlight the potential benefit of biologics for UC patients with periodontitis, as CXCL6 serves as a common hub gene linking UC and PD. It is essential to recognize that the oral-gut axis suggests that managing intestinal inflammation may also play a role in mitigating oral diseases. Consequently, therapies such as VDZ, which primarily focus on the gastrointestinal tract, may inadvertently benefit periodontitis related to ulcerative colitis.

To validate our bioinformatic prediction of CXCL6 upregulation under compound disease conditions, we established a rat model of colitis induced by DSS in combination with PD treatment. Histological examination revealed markedly heightened immune cell infiltration and epithelial disruption in the DSS+PD group relative to DSS alone ( Figure 10A ). CXCL6 expression, assessed by immunohistochemistry, was significantly elevated in colonic tissues from the DSS+PD group ( Figure 10B ). Given the enhanced inflammatory response, we next evaluated intestinal barrier integrity by staining for the tight junction proteins Claudin-1, ZO-1, and Occludin. All three markers were reduced in the DSS+PD group, indicating compromised epithelial barrier function ( Figure 10C ). These findings confirm that CXCL6 is upregulated under dual inflammatory challenge and coincides with more severe epithelial damage. While the mechanistic link remains to be established, the parallel increase in CXCL6 expression and loss of barrier proteins raises the possibility that CXCL6 may contribute to intestinal barrier dysfunction in this context.