Regulation mechanisms of disulfidptosis-related genes in ankylosing spondylitis and inflammatory bowel disease

Abstract

Introduction: Disulfidptosis is a recently identified form of cell death that contributes to maintaining the internal environment balance of an organism. However, the molecular basis of disulfidptosis in ulcerative colitis (UC), ankylosing spondylitis (AS), and Crohn's disease (CD) has not been thoroughly explored.

Methods: Firstly, the differentially expressed genes (DEGs) and disulfidptosis-associated genes (DAGs) were obtained through differential analysis between diseases (AS, CD, and UC) and control groups. After the disulfidptosis score was acquired using the single-sample gene set enrichment analysis (ssGSEA) algorithm, the DE-DAGs were screened by overlapping DAGs and DEGs of the three diseases. Next, the feature genes were selected through a combination of machine learning algorithms, receiver operating characteristic (ROC) curves, and expression analysis. Based on these feature genes, nomograms were created for AS, CD and UC. The co-feature genes were then identified by taking the intersections of the genes featured in all three diseases. Meanwhile, single-gene set enrichment analysis (GSEA) and the TF-mRNA-miRNA network were utilized to investigate the molecular mechanisms of the co-feature genes. To validate the expression differences of the co-feature genes between healthy controls and patients (AS and IBD), RT-PCR was performed. Lastly, mendelian randomization (MR) analysis was utilized to explore the causality between genetic variants of S100A12 with AS, UC and CD.

Results: In this study, 11 DE-DAGs were obtained. Functional enrichment analysis revealed their involvement in cytokine production and fatty acid biosynthesis. Latterly, AS/CD/UC -feature genes were derived, and they all had decent diagnostic performance. Through evaluation, the performance of the nomogram was decent for three diseases. Then, 2 co-feature genes (S100A12 and LILRA5) were obtained. The GSEA enrichment results indicated that the co-feature genes were mainly enriched in the cytokine-cytokine receptor interaction and drug metabolism cytochrome P450. As shown by functional experiments, there was a correlation between the mRNA expression of S100A12 with AS, UC and CD. Additionally, a causal connection between S100A12 and IBD was detected through MR analysis.

Discussion: In this study, 2 co-feature genes (S100A12 and LILRA5) were screened, and their functions were investigated in AS, CD and UC, providing a basis for further research into diagnosis and treatment.

Keywords: Crohn’s disease; ankylosing spondylitis; bioinformatics; disulfidptosis; mendelian randomization; ulcerative colitis.

Figure 1

Identification of DEGs associated with AS, CD and UC. (A, B) The volcano plot and heatmap plot of DEGs identified in GSE25101 (AS, n=32, p < 0.05); (C, D) The volcano plot and heatmap plot of DEGs identified in GSE75214 (CD, n=62, p < 0.05); (E, F) The volcano plot and heatmap plot of DEGs identified in GSE75214 (UC, n=85, p < 0.05).

Figure 2

Identification of DEGs associated with disulfidptosis involved in AS, CD and UC. (A, B) The volcano plot and heatmap plot of DAGs involved in AS (p < 0.05); (C, D) The volcano plot and heatmap plot of DAGs involved in CD (p < 0.05); (E, F) The volcano plot and heatmap plot of DAGs involved in UC (p < 0.05).

Figure 3

Identification and functional analysis of co-DE-DAGs associated with AS, CD and UC. (A) Bubble plot of GO functional enrichment; (B) Network Diagram of GO functional enrichment; (C) Bubble plot of KEGG functional enrichment; (D) Network Diagram of KEGG functional enrichment.

Figure 4

Machine learning screening of candidate feature genes involved in AS, CD and UC. (A–C) 9 XGBoost-feature genes1 (AS), 8 XGBoost-feature genes2 (CD) and 7 XGBoost-feature genes3 (UC) on the basis of DE-DAGs; (D–I) 5 RF-feature genes 1 (AS), 9 RF-feature genes 2 (CD) and 7 RF-feature genes 3 (UC) on the basis of DE-DAGs; (J–L) 9 SVM-RFE-feature genes1 (AS), 6 SVM-RFE-feature genes 2 (CD) and 6 SVM-RFE-feature genes 3 (UC) on the basis of DE-DAGs.

Figure 5

(A) 5 AS-candidate-feature genes, (B) 5 CD-candidate-feature genes and (C) 5 UC-candidate-feature genes were filtered via overlapping XGBoost-feature genes, RF-feature genes and SVM-RFE-feature genes.

Figure 6

Verification of feature genes between control and disease (AS, CD and UC) groups. (A, B) 5 AS-candidate-feature genes expression levels between AS and control groups (training & validation sets); (C, D) Plotted ROC curves for 5 AS-candidate-feature genes in AS group (training & validation sets); (E, F) 5 CD-candidate-feature genes expression level differences between CD and controls (training & validation sets); (G, H) 5 UC-candidate feature genes expression level differences between UC and controls (training & validation sets). ns, no significance; *, **, ***, and **** indicate the significance of gene expression differences, the more asterisks there are, the greater the significance of the difference; *: p < 0.05; **: p < 0.01; ***: p < 0.001; ****: p < 0.0001.

Figure 7

The AUC values of candidate feature genes involved in AS, CD and UC. (A, B) Plotted ROC curves for 5 CD-candidate-feature genes in CD groups (training & validation sets); (C, D) Plotted ROC curves for 6 UC-candidate-feature genes in UC groups (training & validation sets).

Figure 8

Nomogram model construction and evaluation of candidate feature genes involved in AS, CD and UC. (A–C) the nomogram model construction on the basis of feature genes to predict the likelihood of disease in patients with AS, CD and UC (p>0.05; MAE<0.05); (D–F) Calibration curve showed the accuracy of the nomogram was relatively high and validated the model performance of AS, CD and UC; (G–I) DCA curve suggested net profit of the constructed model is better than the default method.

Figure 9

GSEA analysis of each co-feature genes. (A, B) Single-gene GSEA of S100A12 and LILRA5 on pathways in AS; (C, D) Single-gene GSEA of S100A12 and LILRA5 on pathways in CD; (E, F) Single-gene GSEA of S100A12 and LILRA5 on pathways in UC.

Figure 10

Screening for co-feature genes and immune-infiltration analysis. (A–C) The proportion of the 22 immune cells in each sample in AS, CD and UC; (D–F) The difference in the proportion of immune cells infiltrating between the three diseases and control group (Wilcoxon method); ns, no significance; *, **, ***, and **** indicate the significance of gene expression differences, the more asterisks there are, the greater the significance of the difference; *: p < 0.05; **: p < 0.01; ***: p < 0.001; ****: p < 0.0001. (G–I) The correlation analysis between the proportion of immune cells infiltrating and co-feature genes in AS, CD and UC.

Figure 11

TF-mRNA-miRNA network of S100A12 and LILRA5.

Figure 12

The mRNA expression differences of S100A12 (A) and LILRA5 (B) between healthy controls and diseases. S100A12 mRNA expression was significantly up regulated in the IBD and AS patients compared with healthy individuals. ns, no significance.

Figure 13

The significant outcomes of MR effect regarding S100A12 on UC. (A) Scatterplot: The x-axis represents the effect of SNPs on exposure, and the y-axis represents the effect of SNPs on the outcome. A slope greater than 0 indicates that the exposure factor is an adverse factor for the outcome. (B) Forest plot: A value greater than 0 implies a positive association between the SNP position and the outcome, while a value less than 0 suggests a negative association. (C) Funnel plot. (D) Leave-one-out: Leave-one-out analysis did not result in the exclusion of any instrumental variable, and the model’s effects remained statistically significant without significant deviations.

Figure 14

The significant outcomes of MR effect regarding S100A12 on CD. (A) Scatterplot: The x-axis represents the effect of SNPs on exposure, and the y-axis represents the effect of SNPs on the outcome. A slope greater than 0 indicates that the exposure factor is an adverse factor for the outcome. (B) Forest plot: A value greater than 0 implies a positive association between the SNP position and the outcome, while a value less than 0 suggests a negative association. (C) Funnel plot. (D) Leave-one-out: Leave-one-out analysis did not result in the exclusion of any instrumental variable, and the model’s effects remained statistically significant without significant deviations.