Transcriptomic exploration of key genes related to mitochondria and ferroptosis in inflammatory bowel disease and experimental validation.

Firstly, a sum of 465 DEGs were obtained between IBD and control groups, of which were 280 upregulated and 185 downregulated genes in IBD group ( p  < 0.05) (Fig.  1 a) (Supplementary Table 4). Additionally, the circular heatmap constructed with DEGs the top 20 genes demonstrated that these genes could effectively distinguish between the IBD and control groups (Fig.  1 b). Fig. 1 Identification of 8 candidate genes and function enrichment analysis. ( a ) 465 DEGs were obtained between IBD and control groups,280 upregulated and 185 downregulated genes; ( b ) The circular heatmap with DEGs the top 20 genes; ( c ) The candidate gene Venn diagram represented the intersection of differentially expressed genes, mitochondrial-related genes, and ferroptosis-related genes. A total of 8 candidate genes were identified; ( d ) A total of 799 GO entries were enriched based on thesegenes, which included 753 in BPs, 19 in CCs as well as 27 in MFs; e There were found 37 related pathways in KEGG enrichment analysis. Identification of 8 candidate genes and function enrichment analysis. ( a ) 465 DEGs were obtained between IBD and control groups,280 upregulated and 185 downregulated genes; ( b ) The circular heatmap with DEGs the top 20 genes; ( c ) The candidate gene Venn diagram represented the intersection of differentially expressed genes, mitochondrial-related genes, and ferroptosis-related genes. A total of 8 candidate genes were identified; ( d ) A total of 799 GO entries were enriched based on thesegenes, which included 753 in BPs, 19 in CCs as well as 27 in MFs; e There were found 37 related pathways in KEGG enrichment analysis. The 465 DEGs, 2030 MRGs, and 431 FRGs were intersected, resulting in the identification of 8 candidate genes (AQP8, ACSF2, FADS1, HIF1A, GJA1, IL6, ACSL4, and IL1B) (Fig.  1 c). A total of 799 GO entries were enriched based on these genes, which included 753 in BPs (such as amide transport), 19 in CCs (such as apical plasma membrane) as well as 27 in MFs (such as CoA − ligase activity) (Fig.  1 d) (Supplementary Table 5). In addition, there were found 37 related pathways in KEGG enrichment analysis, which were mainly related to antifolate resistance and Th17 cell differentiation(Fig.  1 e) (Supplementary Table 6). To further identify candidate keygenes, machine learning models were constructed. Through the LASSO algorithm,6 feature genes (AQP8, ACSF2, HIF1A, GJA1, ACSL4, and IL1B) were selected at theoptimal lambda value of 0.0074 (lambda.min)(Fig.  2 a, b), while 6 feature genes (IL1B, IL6, ACSF2, AQP8, ACSL4, and FADS1) were selected at the highest accuracy value in 10-fold cross-validation through the SVM-RFE algorithm (Fig.  2 c). The intersection of the feature genes obtained from the 2 algorithms was taken, resulting in 4 candidate key genes (AQP8, ACSF2, ACSL4, and IL1B) (Fig.  2 d). Following this, it was found that in both datasets, the 4 genes had AUC values greater than 0.7, indicating good diagnostic value for IBD (Fig.  2 e, f). In addition, ACSL4 and IL1B were found to be upregulated in IBD group, while ACSF2 and AQP8 were found to be upregulated in control group, with consistent expression trends in both datasets ( p  < 0.001)(Fig.  2 g, h). Therefore, the 4 genes (AQP8, ACSF2, ACSL4, and IL1B) were identified as key genes. Fig. 2 Identification of 4key genes. a , b ,Through the LASSO algorithm,6 feature genes (AQP8, ACSF2, HIF1A, GJA1, ACSL4, and IL1B) were selected; c ,6 feature genes (IL1B, IL6, ACSF2, AQP8, ACSL4, and FADS1) were selected through the SVM-RFE algorithm; d ,The intersection of the feature genes obtained from the 2 algorithms was taken, resulting in 4 candidate key genes (AQP8, ACSF2, ACSL4, and IL1B) ; e - f ,In both datasets, the 4 genes had AUC values greater than 0.7, indicating good diagnostic value for IBD, e, GSE75214 , f, GSE59071 ; g - h ,Expression verification of key genes in the dataset. ACSL4 and IL1B were upregulated in IBD group, while ACSF2 and AQP8 were upregulated in control group.g, GSE75214 , h, GSE59071 . Identification of 4key genes. a , b ,Through the LASSO algorithm,6 feature genes (AQP8, ACSF2, HIF1A, GJA1, ACSL4, and IL1B) were selected; c ,6 feature genes (IL1B, IL6, ACSF2, AQP8, ACSL4, and FADS1) were selected through the SVM-RFE algorithm; d ,The intersection of the feature genes obtained from the 2 algorithms was taken, resulting in 4 candidate key genes (AQP8, ACSF2, ACSL4, and IL1B) ; e - f ,In both datasets, the 4 genes had AUC values greater than 0.7, indicating good diagnostic value for IBD, e, GSE75214 , f, GSE59071 ; g - h ,Expression verification of key genes in the dataset. ACSL4 and IL1B were upregulated in IBD group, while ACSF2 and AQP8 were upregulated in control group.g, GSE75214 , h, GSE59071 . An ANN model was then constructed using the 4key genes in GSE75214 to evaluate the potential diagnostic value. These genes were connected to IBD through 3 unspecified biomarkers (H1, H2, H3) by red lines, indicating a strong association between them (Fig.  3 a). Additionally, the AUCs in GSE75214 and GSE59071 were 0.855 (0.759–0.951) and 0.859 (0.721–0.997), respectively, demonstrated that the ANN model which was constructed using the 4key genes had excellent diagnostic performance (Fig.  3 b, c). Fig. 3 Evaluation of disease prediction accuracy using ANN model constructed with key genes. ( a ) Neural network topology diagram, with lines represented the weights between each layer and each connection connections, with red lines represented positive connection weights and gray lines represented negative connection weights; ( b )–( c ) the AUCs inGSE75214 and GSE59071 were 0.855 (0.759–0.951) and 0.859 (0.721–0.997), respectively. Evaluation of disease prediction accuracy using ANN model constructed with key genes. ( a ) Neural network topology diagram, with lines represented the weights between each layer and each connection connections, with red lines represented positive connection weights and gray lines represented negative connection weights; ( b )–( c ) the AUCs inGSE75214 and GSE59071 were 0.855 (0.759–0.951) and 0.859 (0.721–0.997), respectively. GSEA in GO revealed that pathways associated with the expression of ACSF2, ACSL4, AQP8, and IL1B were 2,299, 2,761, 1,582, and 2,556, respectively. In KEGG analysis, the pathways linked to the expression of these genes were 179, 190, 136, and 194, respectively. In the top 5 GO terms, ACSF2 was mainly associated with fatty acid catabolic processes, extracellular matrix structural components and other pathways (Fig.  4 a, Supplementary Table 7); ACSL4 was mainly enriched in adaptive immune response, myeloid leukocyte activation(Fig.  4 b, Supplementary Table 8); AQP8 was involved in adaptive immune response and lymphocyte-mediated immunity(Fig.  4 c, Supplementary Table 9); and IL1B was enriched in adaptive immune response, positive regulation of cytokine production (Fig.  4 d, Supplementary Table 10).In top 5 KEGG pathways, four genes have been found to be associated with rheumatoid arthritis, malaria, hematopoietic cell lineage, inflammatory bowel disease, and osteoclast differentiation. In addition, ACSL4 was associated with viral protein-cytokine and cytokine receptor interaction pathways, AQP8 was involved in the IgA pathway of the intestinal immune network, and IL1B was also associated with leishmaniasis (Fig.  4 e-h) (Supplementary Tables 11–14). Fig. 4 The GSEA analysis results of key genes. The first part showed that the horizontal axis represents genes and the vertical axis represents the corresponding Running ES. The second part was the hit diagram, where each vertical line represents the gene in the pathway, sorted from left to right by expression level. The third part was the distribution map of gene rank values. The gene enrichment score on the left side was positive, while that on the right side was negative. ( a–d ) GO entry enrichment analysis; ( e–h ) KEGG entry enrichment analysis. The GSEA analysis results of key genes. The first part showed that the horizontal axis represents genes and the vertical axis represents the corresponding Running ES. The second part was the hit diagram, where each vertical line represents the gene in the pathway, sorted from left to right by expression level. The third part was the distribution map of gene rank values. The gene enrichment score on the left side was positive, while that on the right side was negative. ( a–d ) GO entry enrichment analysis; ( e–h ) KEGG entry enrichment analysis. Moreover, 166, 163, 126, and 152 differential pathways were linked to the expression of ACSF2, ACSL4, AQP8, and IL1B, respectively. In addition, it was found that low expression of ACSF2 and AQP8 was associated with activation of the citric acid cycle, TCA cycle, and inositol phosphate metabolism, whereas high expression of ACSL4 and IL1B was associated with activation of the citric acid cycle, TCA cycle, and inositol phosphate metabolism (Fig.  5 a-d) (Supplementary Tables 15–18). Fig. 5 GSVA enrichment analysis of key genes. a ACSF2, b ACSL4, c AQP8, d IL1B. GSVA enrichment analysis of key genes. a ACSF2, b ACSL4, c AQP8, d IL1B. In immune infiltration analysis, stacked and violin plots indicated significant differences in the levels of 24 immune cell types, including activated CD4 T cells and macrophages, between the IBD and control groups ( p  < 0.05) (Fig.  6 a-b). A correlation analysis revealed that ACSL4 and IL1B were positively correlated with the most differential immune cells, whereas ACSF2 and AQP8 were negatively correlated with 24 and 23 immune cell types, respectively (|cor| > 0.3, p  < 0.05). The strongest positive correlation was observed between ACSL4 and immature dendritic cells (cor = 0.93, p  < 0.01), while ACSF2 showed the strongest negative correlation with effector memory CD8 T cells (cor = -0.81, p  < 0.01) (Fig.  6 c). Fig. 6 Analysis of immune cell infiltration in IBD. ( a–b ) The stacked plot and violin plot results showed that infiltration levels of 24 types of immune cells were significantly different between IBD and control groups.ns indicates p  > 0.05, * indicates p  < 0.05, ** indicates p  < 0.01, ***indicates p  < 0.001, **** indicates p  < 0.0001. ( c ) Correlation analysis of key genes and immune cells. The strongest positive correlation between ACSL4 and immature dendritic cells (cor = 0.93, p  < 0.01), the strongest negative correlation between ACSF2 and effector memory CD8 T cells (cor =-0.81, p  < 0.01). Analysis of immune cell infiltration in IBD. ( a–b ) The stacked plot and violin plot results showed that infiltration levels of 24 types of immune cells were significantly different between IBD and control groups.ns indicates p  > 0.05, * indicates p  < 0.05, ** indicates p  < 0.01, ***indicates p  < 0.001, **** indicates p  < 0.0001. ( c ) Correlation analysis of key genes and immune cells. The strongest positive correlation between ACSL4 and immature dendritic cells (cor = 0.93, p  < 0.01), the strongest negative correlation between ACSF2 and effector memory CD8 T cells (cor =-0.81, p  < 0.01). Chromosomal localization showed that ACSF2, ACSL4, AQP8, and IL1B were located on chromosomes 17, X, 16, and 2, respectively (Fig.  7 a). Through the CTD database, potential drugs were predicted for each of these genes: 54 for ACSF2, 91 for ACSL4, 15 for AQP8, and 1,078 for IL1B (Supplementary Table 19). Notably, based on target prediction, MIBX may be involved in IBD regulation by acting on ACSF2 and IL1B, but its therapeutic effect is speculative and requires experimental verification(Fig.  7 b). Fig. 7 Chromosomal localization and drug prediction of key genes. ( a ) Key genes ACSF2, ACSL4, AQP8, and IL1B were located on chromosome 17, chromosome X, chromosome 16, and chromosome 2, respectively. ( b ) Potential drugs were separately predicted for ACSF2, ACSL4, AQP8, and IL1B;1-Methyl-3-isobutylxanthinesimultaneously targeted ACSF2 and IL1B. Chromosomal localization and drug prediction of key genes. ( a ) Key genes ACSF2, ACSL4, AQP8, and IL1B were located on chromosome 17, chromosome X, chromosome 16, and chromosome 2, respectively. ( b ) Potential drugs were separately predicted for ACSF2, ACSL4, AQP8, and IL1B;1-Methyl-3-isobutylxanthinesimultaneously targeted ACSF2 and IL1B. RT-qPCR experiments showed that the expression of AQP8 and ACSF2 was significantly decreased in IBD compared to the control group ( p  < 0.01) (Fig.  8 a-b), while the expression of ACSL4 and IL1B was significantly increased ( p  < 0.05) (Fig.  8 c-d). These findings were consistent with the results observed in both the training and validation sets. Fig. 8 Differential expression of key genes in clinical samples. ( a–b ) The expression levels of AQP8 and ACSF2 in IBD were significantly decreased. ( c–d ) the expression levels of ACSL4 and IL1B in IBD were significantly increased. * indicates p  < 0.05, ** indicates p  < 0.01, **** indicates p  < 0.0001. Differential expression of key genes in clinical samples. ( a–b ) The expression levels of AQP8 and ACSF2 in IBD were significantly decreased. ( c–d ) the expression levels of ACSL4 and IL1B in IBD were significantly increased. * indicates p  < 0.05, ** indicates p  < 0.01, **** indicates p  < 0.0001.

IBD related transcriptome dataset ( GSE75214 and GSE59071 ) was obtained from GEO database. GSE75214 (platform: GPL6244 ) consisted of colon and terminal ileum mucosal samples from 172IBD patients (colon: 105, terminal ileum: 67) and 22 healthy controls (colon: 11, terminal ileum: 11). On the other hand, GSE59071 (platform: GPL6244 ), as the validation set, consisted of colon mucosal samples from 105IBD patients and 11 healthy controls. Mitochondria-related genes (MRGs) and ferroptosis-related genes(FRGs) were separately obtained from relevant literature 7 , 8 , including 2,030 and 431 genes, respectively (Supplementary Tables 1, 2). Firstly, by using limma package (v 3.58.1) 9 , differential expression analysis was performed on all samples from GSE75214 between the IBD and control groups (|log 2 fold change (FC)| >1, adj. p < 0.05) so as to yield differentially expressed genes (DEGs). Subsequently, ggplot2 (v 3.4.4) 10 , and circlize (v 0.4.16) 11 packages were separately applied to visualize the results by plotting a volcano plot, with the top 10 genes highlighted based on |log 2 FC| from largest to smallest, and a circular heatmap for the top 20 genes (upregulated: downregulated = 10: 10)based on |log 2 FC| from from largest to smallest. The intersection of DEGs, MRGs, and FRGs was determined using the ggvenn package (v 0.1.10) ( https://rdocumentation.org/packages/ggvenn/ ), yielding candidate genes. Gene Ontology (GO) analysis ( p < 0.05) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis (pvalueCutoff = 0.05, qvalueCutoff = 0.2) were conducted to explore the potential biological functions and pathways of these genes 12 – 14 , using the cluster Profiler package (v 4.10.0) 15 and org.Hs.eg.db package (v 3.16.0) 16 . The top five most significant biological functions in each category (biological process [BP], cellular component [CC], molecular function [MF]) in GO and the 30 most significant KEGG pathways were plotted based on p-values and adj.p-values. To further refine the candidate genes, machine learning models were applied using the GSE75214 dataset, employing the LASSO and SVM-RFE algorithms to identify feature genes. LASSO logistic regression analysis was performed using the glmnet package (v 4.1-8)( 10-fold cross-validation nfold = 10, nlambda = 200, binary logistic regression family = “binomial”) 17 , and SVM-RFE analysis was conducted using the e1071 package (v 1.7–14)(10-fold cross-validation k = 10, and the halving strategy is adopted when the number of features exceeds 200 halve.above = 200) 18 . The intersection of the feature genes identified by both algorithms was used to pinpoint key candidate genes. Subsequently, receiver operating characteristic (ROC) analysis was conducted using the pROC package (v 1.18.4)( smooth = FALSE, ci = FALSE, and legacy.axes = TRUE) 19 to calculate the area under the curve (AUC), evaluating the ability of the candidate key genes to differentiate between IBD and control group samples from GSE75214 and GSE59071 (AUC > 0.7). The expression levels of the candidate key genes were further analyzed using the Wilcoxon rank-sum test ( p 0.7, significant differences, and consistent expression trends across the two datasets were identified as key genes ( p < 0.05). To assess the overall disease prediction performance of the key genes, an ANN model was developed using all samples from GSE75214 via the neuralnet package (v 1.44.2)(The number of hidden layer nodes is hidden = 3, the activation function is act.fct = “logistic”, and the output layer is set linear.output = FALSE) 20 . The model’s predictive capability was then evaluated using ROC analysis with the pROC package (v 1.18.5), and the AUC was calculated for both GSE75214 and GSE59071 (AUC > 0.7). To further investigate the biological pathways associated with the key genes, Spearman correlation coefficients between each key gene and all other genes in the GSE75214 samples were calculated using the corrplot package (v 0.92) 16 . This resulted in a gene ranking list for each key gene, ordered by descending correlation coefficients. GSEA was then conducted to separately evaluate the enrichment of key genes in GO and KEGG(|NES| > 1, adj. p < 0.05) based on the ranking results and by using cluster Profiler package (v 4.10.0).GO: use all GO categories ont = “ALL”, use human annotation database OrgDb = org.Hs.eg.db, pvalueCutoff = 0.05, eps = 1e-10. KEGG: use human KEGG database organization = “hsa”, keyType = “kegg”. pvalueCutoff = 0.05, eps = 1e-10. The top five enriched terms and pathways were visualized using the GseaVis package (v 0.0.5) 21 , sorted by |NES| from highest to lowest. Based on the median differential expression in IBD and control samples from GSE75214 , each key gene was categorized into high and low expression groups to assess the differential activation of pathways. GSVA scores were calculated for each IBD sample using the GSVA package (v 1.50.0) 22 and the reference gene set “c2.cp.kegg.v7.2.symbols.gmt”. Using the single-sample ssGSEA method method= ‘ssgsea’, assuming that the input data follows a Gaussian distribution, kcdf= ‘Gaussian’, and normalizing the ssGSEA score ssgsea.norm = T. Differences in GSVA scores between the high and low expression groups were computed using the limma package (v 3.58.1) (|t| > 2, p < 0.05). The top 50 differential pathways (high: low = 25:25) were visualized using the ggplot2 package (v 3.4.4), sorted by |t|-values from largest to smallest. In addition, immune cell infiltration was assessed across all GSE75214 samples using the GSVA package (v 1.50.0) and the ssGSEA algorithm, with infiltration scores for 28 immune cell types calculated ( p < 0.05) 23 . Results were visualized using the pheatmap package (v 1.0.12) 24 . Differences in immune cell infiltration between groups were compared using the Wilcoxon rank-sum test ( p 0.3, p < 0.05), and the results were visualized using the ggplot2 package (v 3.4.4). Chromosomal locations of key genes were explored using the RCircos package (v 1.2.2) 25 , based on gene annotation information. Key gene positions were extracted and visualized through a circos plot to illustrate their locations on the chromosomes. In addition, the CTD database was utilized in the study to screen potential drugs targeting key genes, and only human-related data were retained (i.e., screened entries with Organism listed as “Homo sapiens” in the ctd data box); the data were then grouped by gene symbols, and the top 30 drugs were selected for each group. The final results of the screening and sorting were visualized using Cytoscape software (v 3.9.1) 26 . Tissue samples from five patients with IBD and five healthy controls were collected for analysis. The IBD samples were obtained from colonoscopy biopsy samples of 4 cases of ulcerative colitis and 1 case of Crohn’s disease. The source of the normal control sample was the adjacent non-inflammatory tissue of colon cancer patients. The sample of this study followed the Helsinki Declaration and was approved by the Ethics Committee of the Second Affiliated Hospital of Zhengzhou University (approval number: 2022221). All participants signed an informed consent form. The relative expression levels of key genes were assessed through RT-qPCR. The primer sequence list is shown in the Supplementary Table 3. Total RNA was extracted from the tissue samples using TRIzol reagent, followed by reverse transcription with the Hifair ® Ⅲ 1st Strand cDNA Synthesis SuperMix for qPCR kit (Yeasen, China). The reverse transcription was conducted on a S1000™ Thermal Cycler PCR machine (BIO-RAD, USA) using the following program: 25 °C for 5 min, 55 °C for 15 min, 85 °C for 5 min, and then held at 4 °C. Three microliters of cDNA were used to prepare the PCR reaction system, according to the Servicebio ® 2×Universal Blue SYBR Green qPCR Master Mix protocol (Servicebio, China). The PCR reaction was performed for 40 cycles using a CFX Connect real-time PCR instrument. GAPDH served as the internal reference, and the relative expression levels of the key genes were calculated using the 2–△△CT method (Table  1 ). Table 1 Primer sequences Primer Sequence AQP8 F GGTACGAACGGTTTGTGCAG AQP8 R TACGGGAGGAGCATCACCA ACSF2 F CGCTTCCTCAGGTGTCTGTG ACSF2 R CCACCGAGATGACTGTGGTC ACSL4 F TTGGCTACTTGCCTTTGGCT ACSL4 R TGCTGGACTGGTCAGAGAGT IL1B F AACCTCTTCGAGGCACAAGG IL1B R AGATTCGTAGCTGGATGCCG H-GAPDH F ATGGGCAGCCGTTAGGAAAG H-GAPDH R AGGAAAAGCATCACCCGGAG Primer sequences A t-test was conducted in GraphPad Prism 10 to compare the expression differences of key genes between the groups, with graphical representation of the results. Bioinformatics analyses were conducted using R (v 4.2.2), with differences between two groups assessed using the Wilcoxon rank-sum test ( p  < 0.05). The differences in RT-qPCR experimental results between groups were also evaluated using a t-test ( p  < 0.05).

In this study, four key genes (AQP8, ACSF2, ACSL4, and IL1B) were identified as being associated with ferroptosis and mitochondrial dysfunction in patients with IBD. Enrichment analysis revealed that these key genes are linked to the adaptive immune response, the TCA cycle, and inositol phosphate metabolism in IBD. Additionally, 1-methyl-3-isobutylxanthine was identified as a potential therapeutic agent for IBD. The mechanisms involving mitochondria and ferroptosis associated with these four genes and the drug suggest that they may help alleviate ferroptosis in IBD, offering a promising avenue for treatment.

IBD is a multifaceted systemic disorder 1 . Both mitochondrial dysfunction and ferroptosis have been closely linked to the disease 6 , 27 , but their potential collaborative role in IBD remains unclear. This study utilized publicly available transcriptome data from IBD and control samples, while 2,030 MRGs and 431 FRGs were compiled from relevant literature. Using machine learning and two distinct algorithms, four key genes—AQP8, ACSF2, ACSL4, and IL1B—were identified and analyzed. An ANN model was developed based on these key genes, demonstrating strong diagnostic performance for IBD. Enrichment analysis revealed that the TCA cycle and inositol phosphate metabolism pathways were activated in the low-expression groups of ACSF2 and AQP8, and in the high-expression groups of ACSL4 and IL1B. Notably, ACSL4 showed the strongest positive correlation with immature dendritic cells, while ACSF2 exhibited the strongest negative correlation with effector memory CD8 T cells. These key genes were validated through RT-qPCR experiments. A cross-analysis of DEGs, MRGs, and FRGs identified eight candidate genes (AQP8, ACSF2, FADS1, HIF1A, GJA1, IL6, ACSL4, and IL1B). KEGG enrichment analysis identified 37 relevant pathways, with a particular focus on antifolate resistance and Th17 cell differentiation. Recent studies have highlighted the involvement of short-chain fatty acids, Th17 cell metabolites, and microbial activity in the pathogenesis of IBD 28 , 29 . In related studies, Haolong Zhang et al. performed differential gene analysis on endometriosis (EM) and IBD transcriptome datasets using the limma software package within the R environment. They constructed protein-protein interaction (PPI) networks, applying machine learning techniques and topological analysis algorithms (MCC and Degree) through CytoHubba plug-ins to identify key genes involved in the co-pathogenic mechanisms of EM and IBD 30 . Similarly, Zhaohui Zhang et al. employed bioinformatics and machine learning strategies to identify potential characteristic genes in nonalcoholic fatty liver disease 31 . Lianxiang Luo et al. developed a PPI network and two machine learning algorithms to identify central genes in hypoxic-ischemic brain injury 32 . These studies highlight the reliability of machine learning-based approaches for key gene identification. In this study, AQP8, ACSF2, ACSL4, and IL1B were successfully identified as key genes through machine learning, ROC analysis, and dataset expression validation. ACSF2 (acyl-CoA synthetase family member 2) activates medium-chain fatty acid-CoA ligase activity and plays a role in fatty acid metabolism 33 . ACSF2 is also involved in mitochondrial dysfunction and may contribute to the progression of diabetic nephropathy 34 . Given that mitochondria serve as the primary site for fatty acid β-oxidation 35 , it was hypothesized that aberrant expression of ACSF2 may impair mitochondrial utilization of fatty acids, thereby contributing to the pathogenesis of IBD. Furthermore, Ferroptosis is significantly implicated in the progression of UC, and ACSF2, a FRG, holds potential as both a diagnostic and therapeutic marker for UC 36 . Given that UC was a major subtype of IBD 37 , this finding also implied a close interconnection between ACSF2 and ferroptosis in IBD. Through screening analysis, this study identified ACSF2 as a key gene associated with both ferroptosis and mitochondrial dysfunction in IBD patients. Furthermore, we substantiated its involvement in mediating mitochondrial dysfunction and the ferroptosis metabolic pathway in these patients. ACSL4 (acyl-CoA synthetase long chain family member 4) encodes a protein that functions as an isoenzyme of the long-chain fatty acid CoA ligase family. This enzyme is responsible for converting free long-chain fatty acids into fatty acyl CoA esters, playing a critical role in lipid biosynthesis and fatty acid degradation. In the context of UC, ACSL4, a gene associated with ferroptosis, is highly expressed 38 . ACSL4-mediated activation of the arachidonic acid (AA) metabolic pathway contributes to ferroptosis susceptibility, impeding mucosal healing in UC 39 . Shuo Gao et al. suggested that vitamin D could mitigate UC by negatively regulating ACSL4, thereby inhibiting ferroptosis in both mice and cell models 40 . It is hypothesized that ACSL4 may exacerbate intestinal mucosal damage in IBD, particularly UC, by promoting ferroptotic lipid peroxidation. In this study, high expression of ACSL4 was found to activate the TCA cycle and inositol phosphate metabolism. Studies indicated that aberrant activation of the tricarboxylic acid (TCA) cycle may impair the activity of cellular energy sensors (e.g., AMPK), thereby disrupting key mitochondrial quality control processes such as mitophagy 41 . In the context of IBD, these alterations likely contribute to mitochondrial dysfunction, compromising the ability of cells to effectively respond to the intestinal inflammatory microenvironment and ultimately accelerating disease progression. AQP8 (aquaporin 8) is an aquaporin that facilitates the passage of water and uncharged molecules, such as hydrogen peroxide and neutral ammonia (NH3), across cell membranes. It is present in tissues such as the plasma membrane, inner mitochondrial membrane, and endoplasmic membrane. In the small intestine and colon, AQP8 mediates water transport through the mitochondria and the apical membrane of epithelial cells 42 . AQP8 gene expression is known to increase tight junction proteins (TJs), reduce epithelial cell apoptosis, and enhance the intestinal mucosal barrier 43 . In this study, enrichment analysis of key genes revealed that Low expression of AQP8 was associated with the activation of the TCA cycle and inositol phosphate metabolism. Furthermore, studies indicated that mitochondria, as central hubs for cellular energy metabolism, played an indispensable role in the TCA cycle 44 . Concurrently, metabolites of myo-inositol phosphate (IP) metabolism mediated the synthesis and distribution of mitochondrial membrane phospholipids, supplied substrates for lipid peroxidation, and thereby compromised mitochondrial membrane stability 45 . Additionally, during ferroptosis, activation of the TCA cycle promoted ferroptosis induced by cystine deprivation 46 , and IP metabolism-related products were also implicated in regulating lipid peroxidation in ferroptosis 47 . Collectively, these findings demonstrated a close interconnection between mitochondria, ferroptosis, the TCA cycle, and IP metabolism. This further suggested that AQP8 may participate in the pathogenesis of IBD by modulating the TCA cycle and IP metabolism, thereby interfering with mitochondrial function and ferroptosis processes. IL1B (interleukin 1 beta) encodes a protein that belongs to the interleukin-1 cytokine family. Initially synthesized as an inactive precursor by activated macrophages, it is cleaved by the enzyme caspase 1 (CASP1/ICE) into its active form. As a central mediator of inflammation, IL1B regulates cellular processes such as proliferation, differentiation, and apoptosis. In some patients with UC, IL1B+/LYZ + bone marrow cells may contribute to resistance to anti-integrin biotherapy 48 . Gasdermin D-mediated release of IL-1B-containing extracellular vesicles has been implicated in the pathogenesis of intestinal inflammation in IBD 49 . Studies have demonstrated that IL-1β can impair intestinal epithelial barrier integrity by activating the NF-κB and MLCK pathways 50 , and can also induce apoptosis via the mitochondrial pathway 51 . It is hypothesized that within the IBD intestinal microenvironment, the overproduction of IL-1β may similarly impair mitochondrial function in intestinal epithelial cells and immune cells, disrupt energy homeostasis, and thereby exacerbate inflammation and tissue damage. Furthermore, research indicates that inflammatory stimuli can alter the expression of ferroptosis regulators such as GPX4. As a key enzyme inhibiting ferroptosis, reduced expression or activity of GPX4 enhances cellular susceptibility to ferroptosis 52 . This suggests that in IBD, the IL-1β-mediated inflammatory microenvironment may promote ferroptosis in intestinal cells through this mechanism, accelerating mucosal damage and disease progression. Gene enrichment analysis in this study suggested that IL1B was highly expressed during the pathogenesis of IBD and correlates with the upregulation of the TCA cycle and inositol phosphate metabolism. Pengbei Fan 53 successfully assessed the overall predictive performance of key genes (SOCS1 and HSPB1) for COPD using an ANN. Ruan Y 54 demonstrated that the generalized linear model (GLM) was the most accurate method for predicting Alzheimer’s disease (AD) and its subtypes through ANN analysis, providing a reliable approach for evaluating the risk of AD. Yang Y 55 employed an ANN model and ROC curve to assess the diagnostic efficacy of feature genes for UC. These studies validate the reliability of using ANN for predicting key genes and diseases. In this study, an ANN model was constructed using four key genes from GSE75214 , yielding AUC values of 0.855 (0.759–0.951) for GSE75214 and 0.859 (0.721–0.997) for GSE59071 , confirming the strong diagnostic performance of the ANN model based on these four key genes. IBD is a chronic condition characterized by persistent inflammation of the intestinal lining, driven by both innate and adaptive immune mechanisms. GSEA and GSVA enrichment analyses in this study revealed that four key genes—ACSF2, ACSL4, AQP8, and IL1B—are associated with the adaptive immune response and rheumatoid arthritis. The cholinergic output of tuft cells is crucial for the restoration of damaged intestinal structures. Park SE 56 discovered that mice with intestinal epithelial cell-specific inositol polyphosphokinase (IPMK) knockout were more susceptible to colitis due to reduced cholinergic output from tuft cells, caused by impaired inositol phosphate metabolism. In patients with acute severe UC (ASUC), cyclosporine enhanced neutrophil glycolysis and TCA cycling, while significantly reducing apoptosis and migration 57 . These findings suggest that inositol phosphate metabolism and the TCA cycle play significant roles in the pathogenesis of IBD. This study also observed that low expressions of ACSF2 and AQP8, along with high expressions of ACSL4 and IL1B, activate the TCA cycle and inositol phosphate metabolism. RT-qPCR experiments confirmed that the expression of AQP8 and ACSF2 was significantly reduced in IBD, while ACSL4 and IL1B were significantly upregulated. IBD is a complex genetic disorder influenced by various genetic and environmental factors that disrupt the immune-microbiome axis. Among these, cytokines derived from immune cells play a pivotal role in the pathogenesis of IBD by sustaining chronic inflammation through the multifaceted regulation of immune responses. Cytokines such as tumor necrosis factor (TNF), IL-12p40, and IL-23 are critical in IBD pathogenesis, with corresponding antibodies showing promising results in clinical treatment 58 , 59 . Gadjalova I et al. 60 identified the CD86 expression cluster on activated B cells as a key factor in amplifying pro-inflammatory cytokine production in intestinal CD4 T cells. In this study, IBD was found to be associated with the infiltration of 24 types of immune cells, including activated CD4 T cells and macrophages. Notably, for the first time, this study reports that ACSL4 shows the strongest positive correlation with immature dendritic cells, while ACSF2 exhibits the strongest negative correlation with effector memory CD8 T cells. This study also mapped the key genes—ACSF2, ACSL4, AQP8, and IL1B—to chromosomes 17, X, 16, and 2, respectively. Using the CTD database, potential drugs targeting these genes were predicted. Among these, MIBX may modulate IBD by targeting ACSF2 and IL1B; however, its therapeutic effects remain speculative. MIBX, an alkaloid commonly found in tea and coffee, is known to stimulate the central nervous system. However, its use in treating IBD has not been reported. MIBX can delay apoptosis in vascular endothelial cells induced by acid fibroblast growth factor (aFGF) and serum depletion 61 . Recent literature has proposed MIBX as a potential therapeutic for triple-negative breast cancer (TNBC) 62 and recurrent pelvic organ prolapse (POP) 63 . Grader-Beck et al. 64 demonstrated that MIBX elevates intracellular cAMP levels, thereby suppressing CD3- and CD28-mediated T cell activation and cytokine production. Given that iron death is linked to IBD, it is hypothesized that MIBX may alter the tumor microenvironment by inhibiting ferroptosis, potentially offering therapeutic benefits for IBD. However, further research is required to confirm the therapeutic effects of this drug in IBD treatment. This study utilized transcriptome data from patients with IBD available in public databases to identify key genes related to mitochondrial dysfunction and ferroptosis, further exploring their potential biological functions. The findings provide new insights that could enhance the management and diagnosis of IBD. Furthermore, during the clinical validation phase of this study, adjacent non-inflammatory tissue from colorectal cancer patients was used as controls. Although these tissues were pathologically confirmed to be non-inflammatory, they may still exhibit differences in gene expression profiles and immune composition compared to intestinal tissue from healthy individuals. This discrepancy could potentially introduce a slight bias in the interpretation of differential key gene expression.When inferring immune cell infiltration levels and analyzing their correlation with key genes, the algorithms employed rely on pre-defined signature gene sets. Consequently, the estimated immune cell scores are subject to inherent mathematical limitations and error. Moreover, bulk sequencing techniques cannot resolve the cellular origin of gene expression. Therefore, the correlations observed represent potential associations rather than definitive cell-specific interactions. The causal relationships between candidate genes and immune cells require further validation through CRISPR-based knockout and co-culture experiments. The drug predictions (e.g., the proposed mechanism of MIBX in regulating ACSF2 and IL1B) are solely based on bioinformatic analyses and lack in vitro or in vivo experimental support. Their potential associations with ferroptosis and IBD must be confirmed using cellular assays and animal models. Additionally, the sample size of the RT-qPCR validation cohort was relatively small, which may limit the statistical power and generalizability of the results. Future studies should expand the cohort size to enhance the reliability of the conclusions. Meanwhile, normal intestinal mucosa samples from healthy individuals should be included as controls, and functional validation studies must be conducted. This includes performing knockout or overexpression of key genes (such as AQP8, ACSF2, ACSL4, and IL1B) in cellular models to evaluate their effects on IBD-related phenotypes, as well as employing Western blotting to confirm the consistency between protein expression and transcriptional levels.

Inflammatory bowel disease (IBD) is a chronic, immune-mediated, systemic disorder that can affect multiple organ systems 1 , 2 . Current IBD treatments predominantly focus on suppressing inflammation and targeting specific inflammatory mediators. However, prolonged use of immunosuppressive therapies may elevate the risk of infections and malignancies 3 . Mitochondria play a pivotal role in the metabolic reprogramming of immune responses and inflammation. The mitochondrial function in the gastrointestinal epithelium is essential for maintaining intestinal health. Recent studies have highlighted that mitochondrial dysfunction is implicated in the pathogenesis of IBD 4 . In colitis, T cells exhibit impaired mitochondrial respiration, compromised mitochondrial biogenesis, and abnormal mitochondria-associated membrane formation 5 . Ferroptosis, an iron-dependent form of cell death driven by lipid peroxidation, has also been linked to the pathogenesis of IBD 6 . In addition, studies have found that mitochondrial functional status regulates the process of ferroptosis through membrane potential and metabolic pathways. Moreover, the mitochondrial-targeted antioxidant SkQ1 can significantly inhibit ferroptosis by suppressing the generation of ROS in complex I, reducing mitochondrial lipid peroxidation levels by 55%. Ferroptosis induction provides an additional substrate for the production of mitochondrial reactive oxygen species, leading to an increase in the generation of reactive oxygen species and subsequently promoting the accumulation of lipid peroxidation products. These studies suggest that mitochondrial dysfunction and ferroptosis may affect the occurrence and development of IBD through a synergistic effect. However, the interaction between mitochondrial dysfunction and ferroptosis in IBD, as well as in other IBD-like conditions such as irritable bowel syndrome and immune-mediated inflammatory diseases, remains underexplored. Given the significant role of these factors, their combined influence in IBD provides a compelling area for investigation. In this study, transcriptome data from patients with IBD were analyzed to identify candidate genes associated with mitochondria and ferroptosis. Key genes involved in mitochondrial function and ferroptosis were selected through machine learning algorithms and dataset-based expression validation. The potential biological roles of these key genes in IBD were further explored using artificial neural networks (ANNs), enrichment analysis, immune infiltration analysis, and chromosomal localization, thus providing a novel theoretical foundation for the diagnosis and treatment of IBD.

Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6 Supplementary Material 7 Supplementary Material 7 Supplementary Material 8 Supplementary Material 8 Supplementary Material 9 Supplementary Material 9 Supplementary Material 10 Supplementary Material 10 Supplementary Material 11 Supplementary Material 11 Supplementary Material 12 Supplementary Material 12 Supplementary Material 13 Supplementary Material 13 Supplementary Material 14 Supplementary Material 14 Supplementary Material 15 Supplementary Material 15 Supplementary Material 16 Supplementary Material 16 Supplementary Material 17 Supplementary Material 17 Supplementary Material 18 Supplementary Material 18 Supplementary Material 19 Supplementary Material 19