Integrated Analysis of Ferroptosis and Immune Infiltration in Ulcerative Colitis Based on Bioinformatics

Introduction

Ulcerative colitis (UC) is a chronic inflammatory bowel disease (IBD) with typical symptoms including diarrhea, mucous bloody stools, and abdominal pain, primarily affecting the colon and rectum.¹ The etiology of UC is unknown, and may be related to genetics, immune dysregulation, and environmental factors. Epidemiologic data show that UC is more common in developed countries, and the global incidence has increased in recent years, with a peak incidence in people in their 20s and 30s.² However, as many as 15% of cases are refractory to drug therapy or secondary to dysplastic chronic colitis requiring surgical treatment.^3,4 Therefore, understanding the exact molecular mechanisms of UC is crucial for developing therapeutic approaches.

As a novel form of programmed cell death, Ferroptosis has attracted much attention in recent years because of its important role in a variety of diseases.⁵ Ferroptosis is triggered by the accumulation of intracellular iron-dependent lipid peroxidation, which involves dysregulation of redox homeostasis and abnormal lipid metabolism.⁶ It has been found that Ferroptosis may play an important role in the pathology of UC.⁷ Intestinal mucosal epithelial cells of UC patients may be more susceptible to Ferroptosis due to oxidative stress, abnormal iron metabolism, and other factors, which may further exacerbate inflammation and tissue damage.⁸ Notable studies have highlighted how ferroptosis contributes to intestinal epithelial cell damage, a hallmark of UC pathogenesis. For instance, research by Xu et al demonstrated that inhibition of ferroptosis could significantly reduce epithelial injury in UC models, suggesting a potential therapeutic pathway.⁹ Despite these advancements, the precise mechanisms by which ferroptosis influences UC’s inflammatory milieu remain incompletely understood. Our study builds on this foundation, aiming to elucidate further the role of specific ferroptosis markers in UC progression and response to therapy, thereby addressing critical gaps in the current understanding and potentially uncovering novel targets for intervention. However, more studies are still needed to elucidate the specific mechanisms regarding the relationship between Ferroptosis and UC. By clarifying the specific mechanisms regulating Ferroptosis in UC, it is expected to provide new therapeutic targets for UC.

In this study, we will unravel the heterogeneity of UC on the basis of Ferroptosis and immunity by identifying homogeneous phenotypes through molecular and immune infiltration data in bioinformatics and validate them with clinical samples, aiming to provide new theoretical and experimental foundations for the treatment of UC.

Materials and Methods

Data Acquisition and Collection

The Gene Expression Omnibus (GEO; www.ncbi.nlm.nih.gov/geo/) database¹⁰ was used to obtain transcriptome data for UC colon samples. The four transcriptional profiles of UC (GSE206171, GSE47908,¹¹ GSE224758 and GSE179285)¹² and their corresponding clinical data were obtained from the GEO database. GSE206171 dataset as the discovery cohort was based on GPL19211 platform, while GSE47908, GSE224758 and GSE179285 datasets as the validation cohort were based on GPL570, GPL16791 and GPL6480 platforms respectively. The GSE206171 dataset includes 66 active UC, 48 inactive UC, and 38 healthy controls. GSE47908 contains 45 active UC colon tissues and 15 healthy controls; GSE224758 consists of 16 active UC colon tissues and 6 healthy controls; a total of 23 active UC patients and 23 healthy controls from the dataset GSE179285 were included in this study (Table 1). Ferroptosis-related genes (FRGs) which drive, suppress, or marker Ferroptosis were retrieved from FerrDb (http://www.zhounan.org/ferrdb),¹³ a widely recognized database of Ferroptosis in academia. The 484 FRGs obtained after removal of duplicate genes were used for subsequent analysis.

Table 1 Details of the UC Datasets

Identification of UC Differentially Expressed Genes (UCDEGs) and UC Differentially Expressed Ferroptosis-Related Genes (UCDE-FRGs)

The gene expression profiles of the GSE206171 dataset were analyzed for UCDEGs between active UC and healthy controls using the R package “limma”. |Log2 fold change (FC)|>1 and False discovery rate (FDR)<0.05 were set as selection criteria for UCDEGs. Subsequently, overlapping genes between UCDEGs and FRGs were identified as UC differentially expressed Ferroptosis-related genes (UCDE-FRGs).

Functional Enrichment of UCDEGs and UCDE-FRGs

The clusterProfiler package in R was used to identify the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG)^14–16 pathways characterizing UCDEGs and UCDE-FRGs, as well as to explore their potential biological processes, cellular components, molecular functions, and important signaling pathways. Metascape (https://metascape.org/gp/index.html#/main/step1) is an analytical website that integrates features such as enriched functional analysis, interactive omics analysis, gene annotation, and member search. It is a comprehensive portal that utilizes more than 40 independent knowledge bases.¹⁷ Metascape was used to determine the characteristics of the hub UCDEGs, and the screening criteria were set to minimum overlap = 3 and minimum enrichment = 1.5. P < 0.05 was considered statistically significant. Gene set enrichment analysis (GSEA) was performed based on the gene list sorted by log2FC obtained from differential expression analysis using gseGO and gseKEGG functions.¹⁸

Construction and Analysis of Correlation Regulation Networks

The STRING database (http://string-db.org/)¹⁹ was used to analyze the interactions between UCDEGs and to construct protein-protein interaction (PPI) network. Visualize the network using Cytoscape software 3.10.1 (http://cytoscape.org/).²⁰ Using the cytoHubba plugin of Cytoscape, the overlap of the top 20 genes based on algorithms such as maximal clique centrality (MCC), maximum neighborhood component (MNC), Stress, Closeness, Degree, and edge percolated component (EPC) algorithms were identified as hub UCDEGs. To investigate the regulatory relationship between hub UCDEGs and miRNAs, we downloaded miRNA-mRNA interactions from the miRTarBase (version 9.0)²¹ database and visualized them by Cytoscape. We analyzed transcription factors (TFs) binding to hub UCDEGs through the TRRUST (Transcriptional Regulatory Relationships Unraveled by Sentence-based Text mining) (https://www.grnpedia.org/trrust/),²² a manually annotated database of transcriptional regulatory networks, and subsequently visualized the hub UCDEGs-TF network using Cytoscape.

Identification of Key UCDE-FRGs by Machine Learning

To further identify the key UCDE-FRGs, three machine learning algorithms, least absolute shrinkage and selection operator (LASSO) logistic regression,²³ support vector machine-recursive feature elimination (SVM-RFE)²⁴ and Random Forest (RF),²⁵ were used in this study. Based on the UCDE-FRGs obtained from the dataset GSE206171 and the Ferroptosis Database, LASSO regressions with the parameter “family = binomial” were run in the “glmnet” package in R, with “nfolds = 10” to prevent combinatorial overfitting. Run the “caret” package in R with the parameter “method = svmLinear”, use SVM for classification, and RFE combined with SVM to screen for key UCDE-FRGs. The “randomForest” package in R was used to run a random forest model with the parameter “ntrees=500” to identify key UCDE-FRGs.

Evaluation and Validation of Key UCDE-FRGs

The GraphPad Prism (version 10.0) software (La Jolla, CA) was used to plot ROC curves and violin plots of gene expression in validation datasets. When the area under the curve (AUC) exceeded 0.8, the gene was considered highly diagnostic for UC. The two-sample t-test was used to compare the gene expression levels between UC and healthy controls.

Single Cell Distribution of Key UCDE-FRGs

The Single Cell Portal (https://singlecell.broadinstitute.org/single_cell) is a single-cell research project database that can be used to query research projects and gene expression in different cell types,²⁶ from which this study obtained and analyzed single-cell sequencing data of key UCDE-FRGs in human UC. In this study, single-cell sequencing data were analyzed for immune cells and epithelial cells in the human UC gut using the SCP259²⁷ dataset, which includes the colonic mucosa of 18 UC patients and 12 healthy individuals.

Assessment of Subtype Distribution for Immune-Infiltrated Cells in UC

The CIBERSORT (https://cibersortx.stanford.edu/) transforms the normalized gene expression matrix into the composition of 22 immune cell types according to a deconvolution algorithm.²⁸ In this study, CIBERSORT was used to calculate the composition of immune cells in UC and healthy samples. The algorithm employed LM22 (22 immune cell types) features and 1000 permutations.

Correlation and Differential Analysis of Immune Infiltration in UC

To assess the correlation between different immune cells, Spearman correlation analysis was performed on the data analyzed by CIBERSORT. The Spearman correlation analysis was performed on the GSE206171 immune-infiltrated cell profiles analyzed by CIBERSORT and the gene expression profiles of this dataset. Correlation between key UCDE-FRGs and immune-infiltrated cells was determined by Spearman correlation coefficient (r) >0.6 and P<0.05.

Immunohistochemistry (IHC) Staining

Paraffin-embedded tissue sections (4–6 µm) were deparaffinized in xylene and rehydrated through graded alcohols to water. Antigen retrieval was performed by heating the sections in citrate buffer (pH 6.0) or EDTA buffer (pH 8.0) at 95°C for 10–15 minutes. After cooling, endogenous peroxidase activity was blocked with 3% hydrogen peroxide for 10 minutes. To minimize non-specific binding, the sections were incubated with 10% goat serum (Cat# C01-03001; Bioss) for 30 minutes. Anti-DUOX2 antibody (Cat# bs-11432R; Bioss) was used at a dilution of 1:200, anti-IDO1 antibody (Cat# b s-15493R; Bioss), and anti-LCN2 antibody (Cat# DF6816; Affinity Biosciences) at a dilution of 1:100, all incubated overnight at 4°C. After washing, the sections were incubated with a biotinylated secondary antibody [Goat Anti-Rabbit IgG (H+L) HRP;1:200; Affinity Biosciences Cat# S0001], followed by HRP-conjugated streptavidin for 30 minutes each. The signal was developed using 3,3ʹ-diaminobenzidine (DAB) (Cat# C-0003; Bioss) as the chromogen, and sections were counterstained with hematoxylin. After dehydration and clearing, the sections were mounted and examined under a light microscope. Expression of DUOX2 was evaluated by assessing staining intensity and the proportion of positive cells. ImageJ (Fiji) software (https://imagej.net/software/fiji) was used to segment the positive areas, and the proportion of positive areas and the mean optical density (OD) of positive areas were calculated. The results were analyzed statistically by GraphPad Prism (version 10.0) software (La Jolla, CA).

Clinical Sample Collection and Ethical Approval

A total of 12 paraffin-embedded specimens were collected from March to June 2024, of which 6 were diagnosed with active UC (Mayo endoscopic score ≥ 2) and 6 with normal tissues. This study was approved by the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (approval number: 2023-474) and performed according to the principles of the Declaration of Helsinki. Written informed consent was signed by each study participant.

Results

Identification of UCDEGs and UCDE-FRGs

The flow chart of the study is shown in Figure S1. The GSE206171 dataset consisted of gene expression profiles of 38 healthy colon tissue samples and 66 active UC colon biopsy tissue samples. After normalization of the data, 341 UCDEGs were identified, including 245 up-regulated genes and 96 down-regulated genes (Figure 1A and B). With the aim of exploring differentially expressed FRGs in the UC, we retrieved 484 FRGs from the FerrDb database. After overlap of UCDEGs and FRGs, totally 18 FRGs were identified as UCDE-FRGs (Figure 1C), and Figure 1D displayed the expression of these 18 UCDE-FRGs in the dataset GSE206171.

Figure 1 Identification of UCDEGs and UCDE-FRGs. (A) Volcano plot showing significantly UCDEGs in GSE206171. Screening conditions were |Log2FC| ≥ l and FDR < 0.05. Blue dots indicate significantly down-regulated genes, pink dots indicate significantly up-regulated genes, and black dots indicate genes with no significant differences. Labels are shown for UCDE-FRGs. (B) Heatmap for the expression of UCDEGs in GSE206171. (C) Venn diagram for UCDE-FRGs. (D) Heatmap for the expression of 18 UCDE-FRGs in the UC dataset GSE206171.

Functional Enrichment of UCDEGs and UCDE-FRGs

Functional enrichment analysis was performed to predict the biological functions of UCDEGs and UCDE-FRGs. GO analysis revealed that UCDEGs were enriched mainly in extracellular region, immune system process, and immune response (Figure 2A); whereas KEGG pathway analysis indicated UCDEGs were significantly enriched in cytokine-cytokine receptor interaction and IL-17 signaling pathway (Figure 2B). GO analysis revealed that UCDE-FRGs were enriched mainly in oxoacid metabolic process and cell death (Figure 2C); whereas KEGG pathway analysis indicated UCDE-FRGs were significantly enriched in biosynthesis of unsaturated fatty acids, IL-17 signaling pathway, and Ferroptosis (Figure 2D). We also compared the respective significantly enriched pathways in the UC and healthy controls. As shown in Figure S2A and B, immune and inflammation-related pathways (such as IL-17 signaling pathway and innate immune response) were enriched in the UC cohort compared to healthy controls.

Figure 2 Functional enrichment of UCDEGs and UCDE-FRGs. (A) Bubble plot of enriched GO terms for UCDEGs. (B) Chord plot of KEGG enrichment for UCDEGs (C) Bubble plot of enriched GO terms for UCDE-FRGs. (D). Chord plot of KEGG enrichment for UCDE-FRGs. Lines are linked to the indicated pathway terms by colored ribbons and are sorted according to the log10 P-value observed, which is shown in decreasing intensity in the blue squares beside the chosen gene.

Construction and Analysis of Correlation Regulation Networks

The PPI network were constructed for UCDEGs based on the STRING database as well as the results were visualized using Cytoscape (Figure 3A). Using the 6 algorithms of Degree, MCC, MNC, EPC, Closeness and Stress in the Cytoscape plug-in cytoHubba (Table 2), the top 20 genes were calculated, and the 11 overlapping genes were identified as hub UCDEGs, including CCL2, ICAM1, TLR2, CXCL9, MMP9, CXCL10, IL1B, CXCL8, PTPRC, FCGR3A, and IL1A (Figure 3B). In further miRNA-mRNA regulatory network analysis, CXCL8, ICAM1, CXCL10 and MMP9 genes were found to play significant regulatory roles (Figure 3C). While for the TF-mRNA regulatory network, MMP9, CXCL8 and ICAM1 played important roles in regulation (Figure 3D).

Table 2 The Top 20 hub UCDEGs by cytoHubba

Figure 3 Construction and analysis of correlation regulation networks. (A) PPI network of UCDEGs (328 nodes and 2266 edges), Circles in Orange are 18 UCDE-FRGs. (B) Hub UCDEGs by cytoHubba. (C) Hub UCDEGs-miRNA regulatory network. (D) Hub UCDEGs-TFs regulatory network.

Functional Enrichment of Hub UCDEGs

GO terms analysis by Metascape revealed that hub UCDEGs were enriched mainly in cellular response to lipopolysaccharide, cell activation, and innate immune response (Figure 4A–C); whereas KEGG pathway analysis by Metascape indicated hub UCDEGs were significantly enriched in cytokine-cytokine receptor interaction and IL-17 signaling pathway (Figure 4D–F).

Figure 4 Functional enrichment of hub UCDEGs. (A) Bar graph of GO analyses of hub UCDEGs. The network of GO enriched terms of hub UCDEGs, colors indicated the same cluster-ID (B) and P-value (C). (D) Bar graph of KEGG analyses of hub UCDEGs. The network of KEGG enriched terms of hub UCDEGs; colors indicated the same cluster-ID (E) and P-value (F).

Identification of Key UCDE-FRGs by Machine Learning

Three machine learning algorithms (LASSO regression, SVM-RFE, and RF) were used in this study to analyze the UCDE-FRGs in order to better optimize the identification of key UCDE-FRGs. The LASSO regression filtered 10 genes in total (Figure 5A and B). The SVM method reaches its highest accuracy when the number of variables is 3 (Figure 5C). The top 10 genes in terms of variable importance were selected using RF (Figure 5D and E). Finally, by overlapping the three algorithms, three genes (LCN2, DUOX2 and IDO1) were identified as key UCDE-FRGs (Figure 5F), and their description and functions see Table 3.

Table 3 The Key UCDE-FRGs and Their Functions

Figure 5 Identification of key UCDE-FRGs by machine learning. (A) The LASSO path diagram shows that the parameter coefficients shrink as the penalty value k increases. (B) Penalty plot for LASSO model, error bars indicate standard errors. (C) Variable screening plot of SVM model, maximum accuracy when variable is 3. (D) Confidence intervals for error rates in random forest model. (E) The importance of genes in random forest model. (F) Venn diagrams for LASSO, SVM and RF algorithms.

Evaluation and Validation of Key UCDE-FRGs

To validate the diagnostic value of these three key UCDE-FRGs, we selected three UC datasets (GSE47908, GSE179285 and GSE224758) and found that all three genes were up-regulated genes in the validation set (Figure 6A, D and G). The diagnostic value of UC is determined by the area under the curve of the ROC analysis based on the key UCDE-FRGs. The results showed that the AUC of all key UCDE-FRGs in GSE47908 were greater than 0.8 (Figure 6B). In the dataset GSE179285, the AUC values of all genes were higher than 0.8 (Figure 6E). In the dataset GSE224758, all key UCDE-FRGs had AUC values higher than 0.8 (Figure 6H). It could be found that in these datasets, the expression of all key UCDE-FRGs was higher in UC and statistically significant, except for IDO1 in dataset GSE224758 (Figure 6C, F and I). Taken together, the key UCDE-FRGs (LCN2, DUOX2 and IDO1) provide better diagnostic results in predicting UC.

Figure 6 Evaluation and validation of key UCDE-FRGs. (A) Volcano plots showing key UCDE-FRGs in datasets GSE47908. (B) The ROC analysis of key UCDE-FRGs in datasets GSE47908. (C) The violin plots show the expression of the key UCDE-FRGs in datasets GSE47908. (D) Volcano plots showing key UCDE-FRGs in datasets GSE179285. (E) The ROC analysis of key UCDE-FRGs in datasets GSE179285. (F) The violin plots show the expression of the key UCDE-FRGs in datasets GSE179285. (G) Volcano plots showing key UCDE-FRGs in datasets GSE224758. (H) The ROC analysis of key UCDE-FRGs in datasets GSE224758. (I) The violin plots show the expression of the key UCDE-FRGs in datasets GSE224758. ns P>0.05; ** P<0.01; ****P<0.0001.

Single-Cell Distribution of Key UCDE-FRGs

To further clarify the cellular distribution of key UCDE-FRGs in UC, we obtained single-cell sequencing data of UC from Single Cell Portal. The data show that in UC, immune cells are distributed in more abundant variety than epithelial cells (Figure 7A and D). In addition, the results showed that LCN2 and DUOX2 were predominantly distributed in epithelial cells in UC, while IDO1 was mainly distributed in immune cells (Figure 7B, C, E and F).

Figure 7 Single-cell distribution of key UCDE-FRGs. (A) Distribution of epithelial cells in SCP259. (B and C) Distribution of key UCDE-FRGs in epithelial cells. (D) Distribution of immune cells in SCP259. (E and F) Distribution of key UCDE-FRGs in immune cells.

Immune Infiltration in UC

Gene expression profiles were analyzed using the CIBERSORT deconvolution method at P<0.05, resulting in 38 healthy controls and 66 active UC on the heat map. Plasma cells, activated memory CD4+ T cells, Monocytes, M0 Macrophages, M1 Macrophages, activated Mast cells and Neutrophils were all expressed in colon tissues of UC patients at higher levels than in healthy controls, while naïve B cells, CD8+ T cells, activated Natural killer (NK) cells, M2 Macrophages, resting Dendritic Cells (DCs), activated DCs and resting Mast cells were the reverse (Figure 8A and D). The distribution of the 22 immune cells in each sample was depicted in detail in Figure 8B. A positive correlation was detected between M2 Macrophages and resting Mast cells (r = 0.63), as well as activated Mast cells and Neutrophils (r = 0.69); Instead, a negative correlation was detected between resting Mast cells and M1 Macrophages (r = −0.60), M2 Macrophages and activated Mast cells (r = −0.72), M2 Macrophages and Neutrophils (r = −0.72), resting Mast cells and activated Mast cells (r = −0.82), as well as resting Mast cells and Neutrophils (r = −0.61) (Figure 8C). The relationship between key UCDE-FRGs and immune-infiltrated cells in UC was evaluated by Spearman correlation. A positive correlation was observed between IDO1 with Neutrophils (r = 0.73), M1 Macrophages (r = 0.77), and activated Mast cells (r = 0.72), respectively. Nevertheless, IDO1 is negatively correlated with resting Mast cells (r = −0.71) and M2 Macrophages (r = −0.74). Resting Mast cells displayed a negative correlation with LCN2 (r = −0.65) and DUOX2 (r = −0.63). LCN2 is positively correlated with resting M1 Macrophages (r = 0.62) (Figure 8E and F).

Figure 8 Immune infiltration in UC. (A) Immune cells contents of healthy and active UC in GSE206171. (B) Relative percentage of 22 subtype of immune cells in GSE206171. (C) Correlation analysis of differences among immune cells in the active UC group; blue indicates a positive correlation and red indicates a negative correlation; the higher is the absolute value, the stronger is the correlation between immune cells. (D) Violin plot of the proportion for each immune cell subtype in healthy and active UC. (E and F) Correlation analysis of the relationship between key UCDE-FRGs and immune-infiltrated cells in UC. - P>0.05; *P<0.05; ** P<0.01; ***P<0.001; ****P<0.0001.

Immunohistochemical (IHC) Analysis of Key UCDE-FRGs in UC

Based on machine learning screening and validation of the relevant UC datasets, 3 key UCDE-FRGs showed good diagnostic efficacy and may be potential UC Ferroptosis-related markers. Thus, we further validated the expression of three key UCDE-FRGs in UC by IHC in the present study. IHC results showed that the expression of DUOX2 in UC was significantly higher than that in the normal controls (NC), which was mainly distributed in intestinal epithelial cells (Figure 9A). IHC staining revealed that IDO1 is mainly distributed in the cytoplasmic membrane of immune cells in UC intestinal tissue, which is consistent with our previous research that IDO1 is mainly distributed in immune cells (Figure 9B). The expression of LCN2 was mainly concentrated in the goblet cells of the UC colon and infiltrating neutrophils, which was higher than that in the normal group (Figure 9C). The distribution of the three key UCDE-FRGs was quantified by the staining intensity (Figure 9D) and proportion of positive area (Figure 9E) for immunohistochemical results, respectively.

Figure 9 Immunohistochemical (IHC) analysis of key UCDE-FRGs expression. IHC staining of DUOX2 (A), IDO1 (B) and LCN2 (C). (D) Statistical analysis of the mean optical density (Mean OD). (E) Statistical analysis of the proportion of positive area. *P<0.05; ** P<0.01; ***P<0.001.

Discussion

Not only does bioinformatics help in the in-depth study of complex polygenic diseases, but it also identifies key genes associated with UC, thus providing new perspectives and understanding to unravel the pathogenesis of UC. In this study, a total of 343 UCDEGs were found to be significantly expressed in UC colon tissues, among which 18 UCDE-FRGs were identified. Functional enrichment analysis of GO terms showed that UCDEGs and UCDE-FRGs were mainly enriched in immune response and cell death, while KEGG pathway analysis showed that UCDEGs and UCDE-FRGs were mainly enriched in the IL-17 signaling pathway and Ferroptosis, all of which suggests that UC is strongly associated with Ferroptosis and immune response.

Through the 6 algorithms of cytoHubba, 11 genes (CCL2, ICAM1, TLR2, CXCL9, MMP9, CXCL10, IL1B, CXCL8, PTPRC, FCGR3A, and IL1A) in the PPI network of UCDEGs obtained high scores and were identified as hub UCDEGs. In the subsequent construction and analysis of the miRNA-mRNA and TF-mRNA networks, we found that MMP9, CXCL8 and ICAM1 perform significant roles in the regulatory correlation network. Functional enrichment of GO terms and KEGG pathway indicated hub UCDEGs were significantly enriched in immune response and IL-17 signaling pathway, which was similar to the enrichment results of UCDE-FRGs, suggesting that these genes might be implicated in UC through ferroptosis or immunity. It has been shown that activated proteases such as plasmin and matrix metalloproteinases (MMPs) are activated in intestinal tissue in patients with active IBD, and that inhibition of plasmin can prevent colitis in mice by inhibiting the release of cytokines from bone marrow cells mediated by MMP9.²⁹ CXCL8 is one of the most important pro-inflammatory factors and plays a vital role in many inflammatory diseases, including UC; numerous studies have shown that the CXCL8-CXCR1/2 axis is involved in the pathogenesis of UC through multiple signaling pathways.³⁰ Intercellular cell adhesion molecule-1 (ICAM1), as an inflammatory factor, is employed as an indicator to assess the response to UC treatment.^31,32

In the present study, three key UCDE-FRGs (LCN2, DUOX2 and IDO1) were screened by machine learning algorithms. The application of three machine learning algorithms (LASSO regression, SVM-RFE, and RF) to give further assurance of the credibility of the identified genes. The subsequent validation sets also demonstrated the good diagnostic efficacy of these three genes. Meanwhile, single-cell sequencing data showed that LCN2 and DUOX2 were mainly distributed in the epithelial cells of UC, while IDO1 was mainly distributed in immune cells. Interestingly, we also obtained similar results in the correlation analysis of immune-infiltrated cells and key UCDE-FRGs, which showed that IDO1 was strongly correlated with immune-infiltrated cells. The balance between M1 and M2 macrophages is critical for maintaining tissue health and normal immune system function, and in a chronic inflammatory disease such as UC, activation of M1 macrophages predominates, whereas in a fibrotic or neoplastic setting, M2 macrophages may be more activated.³³ In the present study, IDO1 was positively correlated with M1 macrophages and negatively correlated with M2 macrophages, which implies that IDO1 may be involved in UC by activation of M1 macrophages. In this study, IDO1 was found to be significantly positively related to neutrophils and activated mast cells. As an immune regulator in autoimmune diseases and chronic inflammation,³⁴ how IDO1 participates in the course of UC through neutrophils and mast cells remains to be further explored. Yang et al found that hypoxia exacerbates colitis by regulating the HIF-1α pathway through LCN2, affecting the polarization of M1 macrophages in glycolysis,³⁵ which is similar to the positive correlation between LCN2 and M1 macrophages observed in our study. The imbalance of immune cell infiltration is considered a hallmark of UC.³⁶ In this study, it was found that the levels of Plasma cells, activated memory CD4+ T cells, Monocytes, M0 Macrophages, M1 Macrophages, activated Mast cells and Neutrophils in the colon tissue of UC patients were higher than those in healthy controls. It has been shown that the characteristic changes of immune cells in UC are an expansion of M0 macrophages and neutrophils.³⁷ Abnormal differentiation of M1/M2 macrophages is an important factor in the development and progression of UC.³⁸ Even so, the complex interactions and regulatory mechanisms between immune cells remain complex, and further research is needed to clarify their specific role in UC.

In the present study, we evaluated the expression levels of three key UCDE-FRGs by IHC, and the results showed that the expression of LCN2, DUOX2 and IDO1 were all higher in UC than in normal controls. LCN2 serves as an inflammatory amplifier in IBD,³⁹ and its importance has been demonstrated in many IBD-related studies. Serum LCN2 shown to be a biomarker of active UC.⁴⁰ Yang et al found that LCN2 adversely affects inflammation in colitis through intestinal epithelial cell pyroptosis mediated by the NF-κB/NLRP3/GSDMD signaling axis.⁴¹ In terms of Ferroptosis, however, Deng et al found that silencing of LCN2 inhibited Ferroptosis events in a model of LPS/IFN-γ-induced inflammation and in a model of dextran sodium sulfate (DSS)-stimulated UC.⁴² Mucosal disorders cause an increase in DUOX2 expression, which may be a sign of mucosal homeostasis disorders in early IBD patients.⁴³ It has been found that seliciclib may improve UC by regulating DUOX2 levels to inhibit ferroptosis in intestinal epithelial cells.⁴⁴ Increased expression of indoleamine-2,3-dioxygenase 1 (IDO1) enhances the production of kynurenine and xanthurenine in intestinal epithelial cells to reduce susceptibility to colitis.⁴⁵ IDO1 enhances hepatocyte ferroptosis in acute immune hepatitis associated with excessive nitrative stress,⁴⁶ and in addition, it has been shown that artesunate induces ferroptosis in melanoma cells by targeting IDO1.⁴⁷ However, the mechanism of ferroptosis by IDO1 in UC or IBD has not been described, and more studies are needed to further clarify it.

There were also some limitations to this study. Firstly, the study is based on a secondary mining and analysis of previously published datasets from the GEO database, which are limited in terms of species representation, sequencing platform, molecular type, sample grouping, and sample quality. Secondly, the CIBERSORT deconvolution algorithm relies on limited genetic data, and due to the varying susceptibility factors and plasticity of disease phenotypes, the accuracy of the results may be affected. In addition, the key UCDE-FRGs obtained in this study only verified their expression levels in UC through simple immunohistochemistry, and did not further explore their detailed mechanisms of exerting influence on UC via ferroptosis by other wet experiments. Nonetheless, our study may still provide strong evidence for further research into the potential of immune-infiltrated cells or immune genes related to ferroptosis in the therapy and diagnostics of UC.

Conclusion

In summary, this study identified three key genes related to UC ferroptosis and immunity, namely DUOX2, IDO1 and LCN2. In addition, we also explored and analyzed their correlation with immune-infiltrated cells, thus providing new insights and evidence for the importance of immunity and ferroptosis in UC.