The datasets analyzed in this study were obtained from the GEO database, which included T2DM datasets GSE20966 (11), GSE25724 (12), and one AP dataset, GSE194331 (13). The T2DM datasets GSE20966 and GSE25724 were merged for analysis by the “sva” package (14). The |log2 Fold change (FC)| > 0.585 and adjust p <0.05 were set as the criteria for identifying differentially expressed genes (DEGs) of AP and T2DM using the “limma” package (15). Finally, the Venn map was applied to select their common genes.

The STRING database (https://cn.string-db.org/) was used to investigate protein interactions, with the validity of such interactions being determined by a composite score greater than 0.15 (16). Meanwhile, GeneMANIA database (https://genemania.org/) prioritized genes for functional tests (17). Functional enrichment studies were performed by Metascape database (https://metascape.org/), which was designed to provide an extensive resource for annotating and analyzing gene lists to investigate the biological roles and routes implicated in certain genes (18).

Mendelian randomization analysis was conducted using the R package “TwoSampleMR (v.0.5.6)” and strictly adhered to the three fundamental assumptions of Mendelian randomization (MR): (I) There exists a strong association between the instrumental variables (IVs) and the exposure; (II) The IVs are not associated with potential confounding factors; (III) The IVs influence the outcome solely through the exposure. The inverse-variance weighted (IVW) method was employed as the primary analytical approach (19), complemented by four additional methods. The datasets utilized in this study were sourced from the IEU OPEN GWAS (https://gwas.mrcieu.ac.uk/). Specifically, the GWAS for HDL-C included 94,595 participants and a total of 2,418,527 single nucleotide polymorphisms (SNPs); the GWAS for LDL-C comprised 173,082 participants and 2,437,752 SNPs; the GWAS for triglycerides (TG) involved 177,861 participants and 2,439,433 SNPs; and the GWAS for apolipoprotein A-I (APOA-I) encompassed 393,193 participants and 12,321,875 SNPs. The threshold for instrumental variable selection was set at p<5E-08, r²<0.001, with a clumping distance of 10,000 kb. Heterogeneity testing was performed using the “mr_heterogeneity” function, and horizontal pleiotropy testing was conducted via the “mr_pleiotropy_test” function and “MR-PRESSO” R package (19), and sensitivity analysis was executed using the “leave-one-out” method.

LASSO, RF, and SVM-RFE were performed to filter genes in both AP and T2DM, respectively. To mitigate overfitting among genes, LASSO regression analysis was applied, followed by cross-validation by the package “glmnet” (20, 21). The “Random Forest” R software was used to conduct RF (22). Genes with importance > 2 in AP samples and importance > 1 in T2DM samples were selected as feature genes. In addition, SVM-RFE was performed using the R package “e1071”, aiming to optimize the learning performance by minimizing the empirical error (23). The hub genes for the following studies were then selected from the intersection of the three subsets. The intersection of the core genes of AP and T2DM was used as the biomarkers for the two diseases. The expression levels of the core genes in the disease and control groups are shown in boxplots (wilcox test).

The ROC curves were computed using the “pROC” program in order to assess the predictability of the model (24). Using the “rms” package, a nomogram incorporating model genes was created (25). In addition, the model’s predicted accuracy was assessed using the DCA and calibration curves (26).

We downloaded the human T2DM dataset GSE95849 (27) to further validate the expression levels of core genes in T2DM. The wilcox test was used to compare the difference in expression of core genes between disease and control groups, with p < 0.05 considered statistically significant. However, due to the lack of another human AP dataset in the public database, we chose mouse dataset GSE77983 (28) for validation. We used the GEO2R (based on the R package “limma”) online tool of the GEO database to analyze GSE77983 to verify the differential expression of core genes between the AP and control groups.

The mouse pancreatic acinar cell line 266-6 (ATCC; VA, USA), were cultured in DMEM with 10% fetal bovine serum, 100 IU/ml penicillin and 100 μg/ml streptomycin. To induce pancreatitis in vitro, the 266-6 cells were stimulated with 250μM Sodium taurocholate (STC) (29). Twenty-four hours later, cells were collected for the following qPCR analysis.

Total RNA from cells was extracted using Trizol reagent (Accurate Biology, Hunan), and the circRNAs were reversely transcribed using the Evo M-MLV for qPCR (Accurate Biology, Hunan). Then, cDNAs were synthesized and quantified via SYBR Green Pro Taq HS (Accurate Biology, Hunan) under the following cycle scheme: 95°C for 30s, then 95°C for 5 s and 60°C for 30 s for 40 cycles. The RNA expression levels were analyzed and quantified using the ΔΔCt method, and the expression levels of the target genes were compared between the two groups using the t-test. The following primers were employed: IL-6 primer (forward: 5’-GAGAGGAGACTTCACAGAGGATACC-3’; reverse: 5’-TCATTTCCACGATTTCCCAGAGAAC-3’), IL-1β primer (forward: 5’-AGGTCGGTGTGAACGGATTTG-3’; reverse: 5’-TGAGAAGAGGCTGAGACATAGGC-3’), SLPI primer (forward: 5’-GAAGCCACAATGCCGTACTGAC-3’; reverse: 5’-GGAACAGGATTCACGCACTTGG-3’), ARHGEF9 primer (forward: 5’-GAAGCAGTGCCGAAAGAGAAGG-3’; reverse: 5’-ACGAAGCCCATCTGAAATCTGTATATG-3’), and Actin primer (forward: 5’-ACTGCCGCATCCTCTTCCTC-3’; reverse: 5’-AACCGCTCGTTGCCAATAGTG-3’).

Gene set enrichment analysis (GSEA) is utilized to elucidate the molecular mechanisms between high- and low-core gene expression samples, and results with a p value < 0.05 were considered significant (30). Meanwhile, the xCell method was also used to assess the correlation between significantly different enriched immune cell types and characteristic genes, which was considered to be correlated at p < 0.05.

The TargetScan, miRDB, and miRanda databases were used to anticipate miRNA-mRNA pairs in order to identify the ceRNA network that might be influenced by model genes. Genes that were simultaneously listed in three databases were the only ones that were thought to be possible mRNA targets for further research. To predict miRNA-lncRNA pairs, the spongeScan database was used. At last, the ceRNA network could be seen using Cytoscape (31). Meanwhile, the Human Protein Atlas (HPA: https://www.proteinatlas.org/) was utilized to examine the model genes’ immunofluorescence and single-cell type atlases.

The CMap database (https://clue.io/) can link diseases, genes and drugs based on similar or opposite gene expression profiles (32). Commonly upregulated DEGs in AP and T2DM were entered into the CMap database to identify potential small molecule compounds for the treatment of AP and T2DM. Then, the protein structures of the feature genes were obtained from the PDB database, and the AutoDock tool was applied to calculate the protein hydrogenation and charge. PubChemdatabase to download the chemical structure of the drug’s active ingredient. The AutoDock tool is used to check the charge balance and rotatable bonds of tiny molecules. To generate docking energy, AutoDock Vina runs docking simulations. Finally, PyMol software was used to check the docking complex.

A flowchart demonstrating the study protocol is presented in Figure 1 . Initially, we merged two T2DM datasets and corrected batch effects using the “sva” software package. As shown in Figures 2A, B , the differences between batches were effectively eliminated after data normalization, indicating that the two datasets could be merged. The volcano map presented in Figures 2C, D shows DEGs in T2DM and AP (|log2 FC| > 0.585 and adjusted p < 0.05). Venn diagram, the up-regulated and down-regulated genes of the two datasets were crossed, respectively, and 14 up-regulated DEGs and 12 down-regulated DEGs were obtained ( Figures 2E, F ). Figure 2G shows the locations of these common DEGs on chromosomes.

The 26 common DEGs were imported into the STRING database to construct a PPI network ( Figure 3A ). Subsequently, GeneMANIA was then used to further analyze DEGs for co-localization, co-expression, and shared protein domains ( Figure 3B ). The genes in the PPI network were mainly enriched in glycosyl compound metabolism, insulin-like growth factor binding, and complement activation. Furthermore, we used Metascape to determine biological processes and pathways related to the DEGs. According to the results, the DEGs were involved in immune effector processes and blood vessel development ( Figure 3C ) and were closely related to conditions such as hypertriglyceridemia and dyslipidemia ( Figure 3D ).

MR analysis was performed to assess the potential relationship between dyslipidemia and T2DM or AP ( Supplementary Figures S1A, B ). IVW analysis showed that HDL-C and Apoa-I exhibited significant protective effects against both T2DM (P = 1.57E-09 and 0.0007, respectively) and AP (P = 0.0269 and 0.0232, respectively). Notably, TG (P = 0.0001) were identified as a significant risk factor for T2DM. These results remained consistent in most models, demonstrating their robustness. However, some degree of heterogeneity was observed ( Supplementary Table S1 ). To address this issue, we used a random-effect model to minimize potential biases and errors. In addition, we used the MR-PRESSO method to identify and eliminate potential outliers, ensuring non-pleiotropy and the accuracy of the results. Finally, we validated the sensitivity of the results using the leave-one-out test ( Supplementary Figures S1C–G ). Altogether, the results suggested that dyslipidemia may be a common underlying mechanism of T2DM and AP.

To identify key genes associated with the development of both AP and T2DM, we constructed three machine-learning models based on the 26 DEGs. In AP samples, ten key genes were identified using the LASSO regression ( Figures 4A, B ). Seventeen genes extracted from these genes by the SVM-RFE were identified as the best genes for AP patients ( Figures 4C, D ). According to the RF, six genes with importance greater than 2 were included in the subsequent analysis ( Figures 4E, F ). Then, five AP characteristic genes were screened out by the Venn diagram ( Figure 4G ).

Similarly, the LASSO regression has shown eleven genes as potential indicators for diagnosis ( Figures 5A, B ). Using the SVM-RFE, seven genes were identified from these genes as potential biomarkers ( Figures 5C, D ). Eight genes with importance higher than 1 were included in the subsequent analysis based on the RF ( Figures 5E, F ). A Venn diagram was constructed to intersect these three gene sets, resulting in the identification of 4 key genes associated with T2DM ( Figure 5G ). Among the key genes identified in AP and T2DM, we found two common genes, SLPI and ARHGEF9, at the intersection of the Venn diagrams ( Supplementary Figure S2 ). These two genes may serve as a link between AP and T2DM, playing a key role in the development of both conditions.

A box plot was constructed to demonstrate the expression levels of the two key genes in the disease and control groups ( Figures 6A, B ). Specifically, the expression of ARHGEF9 was significantly lower in the AP group than in the control group, whereas the expression of SLPI showed the opposite trend (p < 0.001). As shown in the ROC curve in Figure 6C , both SLPI and ARHGEF9 had high diagnostic value in AP. The AUC value of the 2-gene prediction model was 0.928, which demonstrated the high diagnostic value of both genes ( Figure 6D ). In addition, the DCA curve of the two genes showed a better overall clinical benefit than if none or all of the tests were used for diagnosing AP ( Figure 6E ). Furthermore, we developed a nomogram to assess the possible risks associated with AP ( Figure 6F ). Figure 6G shows that there was a minimal difference between the actual and anticipated risk for AP as indicated by the calibration curve.

The expression levels of ARHGEF9 and SLPI in the T2DM group were consistent with those in the AP group ( Figures 7A, B ). The ROC curve showed that both SLPI and ARHGEF9 had high diagnostic value in T2DM ( Figure 7C ). The AUC value of the 2-gene model was 0.985, which emphasized the diagnostic value of the two genes ( Figure 7D ). Furthermore, the DCA curve of the two genes ( Figure 7E ) showed a better overall clinical benefit than if all or none of the tests were used for diagnosing T2DM. Subsequently, a nomogram to gauge the possible harm that T2DM individuals could cause ( Figure 7F ). The calibration curve indicates that our nomogram also has a good predictive value for T2DM ( Figure 7G ).

The GSE95849 dataset was used to validate the expression levels of SLPI and ARHGEF9 in T2DM. The expression of SLPI was higher in the T2DM group than in the control group, whereas that of ARHGEF9 was significantly lower in the T2DM group than in the control group ( Supplementary Figures S3A, B ). These results were consistent with those observed in the training set. Owing to the lack of another suitable human AP dataset, we analyzed the mouse AP dataset GSE77983 using the GEO2R tool to validate the expression levels of the two key genes in AP. As shown in Supplementary Figures S3C, D , the expression of ARHGEF9 was lower in the AP group than in the control group, whereas that of SLPI showed the opposite trend (p < 0.05). These results were consistent with those observed in the training set.

To additionally verify the expression levels of the two genes in AP, we developed a cell model of pancreatitis by stimulating the mouse pancreatic acinar cell line 266-6 with STC. As shown in Supplementary Figure S3E , the expression levels of IL-6 and IL-1β were significantly higher in the model group than in the control group, indicating that the AP model was successfully established. Subsequently, we evaluated the expression levels of SLPI and ARHGEF9 in the cells. The results showed that SLPI was upregulated in the AP group (p < 0.05), which is consistent with the results observed in the training set. However, no significant difference in ARHGEF9 expression was observed between the two groups ( Supplementary Figure S3F ).

GSEA was used to determine the biological functions of the two key genes in AP and T2DM. According to the results of GSEA in the AP group, ARHGEF9 was significantly downregulated in pathways related to DNA regulation and metabolism ( Figures 8A, B ), whereas SLPI was significantly upregulated in pathways related to the regulation of protein response and localization ( Figures 8C, D ). Subsequently, we investigated the relationship between the two key genes and the immune environment of AP. The results showed that ARHGEF9 expression was significantly negatively correlated with the proportions of resting NK T cells, endothelial cells, and aDCs, whereas SLPI expression was significantly positively correlated with the proportions of epithelial cells, M1 and M2 macrophages, endothelial cells, and basophils ( Figure 8E ).

According to the results of GSEA in the T2DM group, ARHGEF9 was significantly downregulated in pathways related to amino acid metabolism and regulation ( Figures 9A, B ), whereas SLPI was significantly upregulated in pathways related to hormone regulation and cell interactions ( Figures 9C, D ). With regard to the relationship between the two genes and the immune environment of T2DM, ARHGEF9 expression was negatively correlated with the proportions of resting NK T cells, endothelial cells, aDCs, and mesangial cells, whereas SLPI expression was positively correlated with the proportions of epithelial cells, M1 macrophages, mast cells, and basophils ( Figure 9E ).

SLPI- and ARHGEF9-based ceRNA networks were separately constructed using various public databases. Eventually, 14 objective miRNAs and 47 objective lncRNAs interacting with ARHGEF9 ( Figure 10A ) and 1 objective miRNA and 5 objective lncRNAs interacting with SLPI ( Figure 10B ) were identified. The ceRNA networks constructed based on these miRNAs and lncRNAs revealed transcriptional regulatory mechanisms for the two genes. Furthermore, SLPI was found to be distributed primarily in exocrine gland cells and endothelial cells, whereas ARHGEF9 was found to be distributed primarily in duct cells ( Figures 10C, D ). With regard to their locations in cells, SLPI was detected in mitochondria, whereas ARHGEF9 was detected in mitochondria and endoplasmic reticulum ( Figures 10E, F ).

To identify potential drugs for the treatment of AP and T2DM, we imported the 14 upregulated DEGs common between AP and T2DM into the CMAP database for analysis. The top 10 small-molecule compounds are shown in Table 1 . Notably, the tyrosine kinase inhibitor imatinib had the highest negative connectivity score. Subsequently, we performed molecular docking of these 10 small-molecule compounds with SLPI and ARHGEF9 ( Supplementary Figures S4 , S5 ). The minimum binding energies of all docked complexes are shown in Table 1 . The minimum binding energies of complexes were typically less than -6.0 kcal/mol, which indicated that both proteins had a good binding affinity for the 10 small-molecule compounds. In particular, the SLPI–imatinib (-9.1 kcal/mol) and ARHGEF9–imatinib (-10.4 kcal/mol) complexes had the lowest free binding energies. These results suggest that imatinib is a promising therapeutic agent for both AP and T2DM.