Introduction

Pediatric Crohn’s disease (PCD) is an early-onset form of inflammatory bowel disease (IBD) characterized by chronic transmural inflammation of the gastrointestinal tract. Notably, epidemiological studies have documented a marked rise in the prevalence of PCD in recent decades1. Although the precise etiology of PCD remains elusive, current evidence suggests a complex interplay among environmental triggers2, gut microbiota dysbiosis3,4 and immune dysregulation mediated through genetic susceptibility5,6,7. Clinical manifestations extend beyond gastrointestinal involvement to include systemic complications such as growth retardation, pubertal delay, metabolic bone disease, and psychological comorbidities, all stemming from persistent inflammatory burden and chronic disease progression8. These clinical particularities underscore the imperative need to elucidate the molecular mechanisms underlying PCD pathogenesis, which is critical for establishing robust scientific foundations to develop precision diagnostic criteria and targeted therapeutic strategies.

Ferroptosis, a programmed cell death driven by iron-dependent lipid peroxidation, plays a pivotal role in gastrointestinal pathophysiology9. The pathological hallmarks of this process include intracellular iron overload, systemic glutathione depletion, functional inactivation of glutathione peroxidase 4 (GPX4), and accumulation of membrane-damaging lipid peroxides10,11. Of particular relevance to IBD pathogenesis, intestinal epithelial cells (IECs) serve as the frontline defense by maintaining both physical barrier integrity and immunoregulatory functions, thus preserving intestinal homeostasis and preventing bacterial translocation12. Emerging experimental evidence has implicates that excessive iron deposition within the intestinal mucosa may trigger lipid peroxidation-mediated ferroptosis in IECs, disrupting the epithelial barrier integrity and potentiating IBD development13,14. Notably, dysregulated ferroptosis signatures have been documented in adult-onset CD patients15, preclinical studies have shown that pharmacological inhibition of ferroptosis with Ferrostatin-1 ameliorates intestinal inflammation in TNBS-induced murine models of CD-like colitis16. Increasing evidence suggests that ferroptosis is a key pathogenic mechanism in the progression of IBD17,18. However, most current studies on FRGs are predominantly based on adult CD, with very limited research conducted in pediatric populations.

Given the marked differences in clinical manifestations and disease progression between adult CD and PCD, the underlying molecular mechanisms likely differ as well. To address this research gap in the pediatric population, we employed bioinformatics methods to systematically identify and validate key FRGs implicated in the pathogenesis of PCD. This study aims to provide new mechanistic insights into the ferroptotic processes underlying PCD progression, while concurrently exploring FRGs as potential diagnostic biomarkers and therapeutic targets for pediatric patients.

Materials and methods

Data sources

The gene expression dataset GSE117993 was retrieved from the GEO database (https://www.ncbi.nlm.nih.gov/geo/), which comprises intestinal tissue samples from 92 PCD patients and 55 healthy controls. Because the data came from public sources, there was no requirement for approval from the local ethics committee.

Data processing

The “Limma” package in R (v4.2.2) was employed to normalize the data and identify significantly DEGs between PCD and control samples in the GSE117993 dataset. Expression values for duplicate genes were averaged, and non-informative genes were subsequently excluded from the analysis. Stringent filtering criteria were applied, requiring an absolute log2 fold change (|log2FC|) > 1 and adjusted p-value < 0.01. Results were visualized using ggplot2 for volcano plots and pheatmap for hierarchical clustering heatmaps. To investigate ferroptosis-related pathways, we retrieved FRGs from the FerrDb database (http://www.zhounan.org/ferrdb). After removing duplicate entries, this yielded 264 ferroptosis drivers, 238 suppressors, and 9 markers. The intersection between DEGs and FRGs was subsequently determined using Venn analysis to identify FRGs-DEGs.

Functional enrichment analysis

The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses19,20 were performed using the DAVID bioinformatics platform (https://david.ncifcrf.gov/) with the following parameters: Homo sapiens as the species, official gene symbols as identifiers, and a significance threshold of P < 0.05. The top five significantly enriched terms in each category—Biological Process (BP), Molecular Function (MF), and Cellular Component (CC) — were selected based on ascending P-values.

Construction of PPI network

The FRGs-DEGs were uploaded to the STRING12.0 online website to build interaction connections between the proteins encoded by the genes, which may play a significant role in PCD pathogenesis. Interaction values greater than 0.4 were deemed significant, and disconnected nodes in the network were concealed. The resulting PPI network was visualized using Cytoscape 3.9.1. The hub genes were identified using the CytoHubba plugin in Cytoscape. To ensure a robust analysis, we employed the following five standard algorithms: MCC, Maximum Neighborhood Component (MNC), Degree, Edge Percolated Component (EPC) and Closeness.

Immune infiltration analysis

The ssGSEA algorithm was used to analyze the infiltration levels of 28 immune cell subtypes in the PCD and normal group samples from the GSE117993 dataset, with subsequent visualization through a hierarchically clustered heatmap. Box plots were drawn to demonstrate the differential expression levels of the 28 immune infiltrating cells. Spearman’s rank correlation coefficients (ρ) were calculated to assess associations between hub gene expression and the 28 immune infiltrating cells. The “ggplot2” package was used to display the results. The P-value < 0.05 was confirmed to be statistically significant.

Validation of the hub genes

Two independent validation cohorts (GSE126124 and GSE62207) were retrieved from GEO to verify hub gene expression differences. The database GSE126124 contains intestinal tissue samples from 37 PCD patients and 19 healthy children, whereas GSE62207 has samples from 259 PCD patients and 51 healthy children. The ROC curves were constructed using the pROC package, with area under the curve (AUC) values calculated to evaluate diagnostic performance. Higher AUC indicates better diagnostic performance.

Establishment and validation of the nomogram for PCD

Based on these hub genes, a clinical prediction model for PCD was constructed employing the rms package in R. The predictive accuracy of the nomogram was evaluated using calibration curves, and its discriminative ability was further assessed via ROC analysis. External validation was performed using the GSE126124 and GSE62207 datasets to verify the accuracy of the model.

Statistical analysis

All analyses were conducted in R4.2.2 and SPSS 26.0. T-tests were used to assess the differences in hub gene expression between PCD and healthy samples, and Spearman analysis was used to calculate correlation. All statistical approaches were considered significant at a P-value < 0.05.

Results

Identification of DEGs

A total of 1074 DEGs were screened from the GSE117993 database (|log2FC| > 1 and P < 0.01), including 788 upregulated and 286 downregulated genes. Figure 1A shows the volcano plots of the DEGs. In Fig. 1B, the heatmap displayed the top 50 DEGs that were up-and down-regulated between PCD group and the normal group.

Fig. 1
figure 1

The analysis of DEGs. (A) The volcano plot of DEGs. Red dots represented upregulated DEGs, and blue dots represented downregulated DEGs. (B) The heatmap of the top 50 DEGs that were up-and down-regulated.

Full size image

Identification of FRGs-DEGs

The FerrDb contained a total of 511 FRGs. Through comparative analysis of the 511 FRGs and the 1074 DEGs using a Venn diagram approach, we identified 21 overlapping FRG-DEGs (Fig. 2). Notably, 19 genes were significantly upregulated while only 2 genes showed downregulation.

Fig. 2
figure 2

Venn diagram of FRGs and DEGs.

Full size image

Functional enrichment analysis

To elucidate the biological significance of the 21 FRGs-DEGs, we performed comprehensive functional enrichment analyses. The GO analysis revealed that these genes were significantly enriched in three BP: linoleic acid metabolism (P = 1.17E-06), response to lipopolysaccharide (P = 3.57E-04) and regulation of insulin secretion(P = 8.59E-04). The CC analysis revealed predominant associations with apical part of cell (P = 0.003), endoplasmic reticulum membrane (P = 0.004) and basolateral plasma membrane (P = 0.027). The MF analysis identified key roles in heme binding (P = 5.24E-04), linoleoyl-CoA desaturase activity (P = 0.002) and arachidonate 15-lipoxygenase activity (P = 0.003). The top five enriched terms for each GO category are visually summarized in Fig. 3A-C. The KEGG pathway analysis further indicated significant enrichment in three major pathways: metabolic pathways (P = 5.66E-04), and IL-17 signaling pathway (P = 6.08E-04). Figure 3D illustrates the five most statistically significant enriched pathways based on p-values.

Fig. 3
figure 3

Functional enrichment analysis. (A) BP. (B) CC. (C) MF. (D) KEGG pathway enrichment analysis19,20.

Full size image

PPI network construction and analysis

A total of 21 FRG-DEGs were subjected to PPI network analysis using the STRING database. The analysis was performed with default parameters (interaction score threshold > 0.4, corresponding to medium confidence), yielding a network comprising 14 interconnected nodes and 62 edges after excluding unconnected protein nodes. In the STRING-generated network, nodes represent genes, while edges indicate functional interactions between them. The PPI network was subsequently visualized and analyzed using Cytoscape software (Fig. 4A). The top 10 hub genes were identified using the cytoHubba plugin in Cytoscape, employing five algorithms: MCC, MNC, Degree, EPC, and Closeness (Table 1). Integrating these results identified five hub genes: PTGS2, IFNG, IL1B, IDO1, and NOS2. (Fig. 4B).

Fig. 4
figure 4

PPI networks. (A) PPI network of FRGs-DEGs (14 nodes and 62 edges). (B) The hub genes screened by the Cytoscape plugin Cytohubba.

Full size image
Table 1 The top 10 hub genes rank in five algorithms.
Full size table

Immune infiltration analysis

To evaluate differences in immune cell infiltration between PCD patients and healthy controls, we utilized the ssGSEA algorithm. A heatmap was generated to visualize the relative enrichment levels of 28 immune cell subtypes across PCD and control samples (Fig. 5A). The box plot comparison further suggested that the enrichment scores of MDSCs and neutrophils were significantly higher in the PCD group compared to the control group (Fig. 5B). To explore potential functional relationships, Spearman’s rank correlation analysis was performed to assess associations between the identified hub genes (IL1B, IDO1, PTGS2, NOS2, and IFNG) and immune cell subtypes. Notably, IL1B, IDO1, and PTGS2 exhibited positive correlations with MDSCs, neutrophils, activated CD4 + T cells, and activated dendritic cells. However, NOS2 and IFNG only showed significant positive correlations exclusively with MDSCs and neutrophils (Fig. 5C).

Fig. 5
figure 5

Immune infiltration analysis. (A) Hierarchical clustering of the distribution of the 28 immune cells in the GSE117993 samples. (B) Box plots of the proportions of different immune cells in the PCD group and normal group, respectively in the GSE117993 samples. (C) Heatmap of correlation analysis between immune cell infiltration and 5 hub genes in the GSE117993 samples.

Full size image

Validation of hub gene expression

To validate the diagnostic potential of hub genes in predicting disease-related outcomes, we selected two additional PCD datasets from the GEO database (GSE126124 and GSE62207) for independent verification. Differential expression analysis demonstrated significantly elevated transcript levels of PTGS2, IFNG, IL1B, IDO1, and NOS2 in intestinal tissues of PCD patients compared to normal controls (p < 0.01) across both validation cohorts (Fig. 6A-E and G-K). The ROC curve analysis was subsequently performed to assess the diagnostic sensitivity and specificity of these hub genes. All five hub genes exhibited robust discriminative capacity, with the AUC values exceeding 0.76 in both validation sets (Fig. 6F and L). Notably, IL1B exhibits the highest specificity and sensitivity (AUC > 0.86 in both validation cohorts), establishing it as the most promising diagnostic biomarker among the five hub genes. These findings collectively indicate that the identified hub genes display marked differential expression between PCD patients and healthy controls, thus supporting their potential utility as diagnostic biomarkers for PCD.

Fig. 6
figure 6

Validation of hub genes. (A–E) Differential expression of hub genes between PCD patients and healthy controls in the validation dataset GSE126124. (F) The ROC curves of hub genes in the validation dataset GSE126124. (G–K) Differential expression of hub genes between PCD patients and healthy controls in the validation dataset GSE62207. (L) The ROC curves of hub genes in the validation dataset GSE62207.

Full size image

Establishment and validation of the nomogram for PCD

Building upon prior findings, we constructed a risk prediction model for PCD incorporating the genes PTGS2, IFNG, IL1B, IDO1, and NOS2 (Fig. 7A). The model demonstrated robust diagnostic efficacy for PCD. Calibration curves indicated that the nomogram closely aligned with the ideal reference line, supporting the reliability of its predictions (Fig. 7B). The model demonstrated excellent performance, with an AUC of 0.895. (Fig. 7C). Furthermore, to validate the robustness of the model, external datasets GSE126124 and GSE62207 were employed. The AUC values of the risk score in these cohorts reached 0.945 and 0.932, respectively, reaffirming the model’s outstanding discriminative ability (Fig. 7D-E).

Fig. 7
figure 7

Construction of the prediction model. (A) Nomogram of PCD patients based on 5 hub genes. (B) Calibration curve of nomogram prediction. (C) ROC curve of prediction model. (D) ROC curve of validation set in GSE126124. (E) ROC curve of validation set in GSE62207.

Full size image

Discussion

PCD, as a chronic and refractory IBD, exerts profound and multifaceted impacts on children’s health. PCD is often more severe, and has greater morbidity compared with adult Crohn’s disease21. These challenges underscore the urgent need for developing more accurate diagnostic modalities with enhanced specificity, patient convenience, and non-invasive characteristics for pediatric populations.

In this study, we analyzed the GSE117993 dataset from the GEO database, comprising transcriptomic profiles of PCD patients. Differential expression analysis identified 1,074 significantly dysregulated genes, including 21 FRGs. The GO analysis revealed that the BP of these FRG-DEGs were predominantly enriched in the linoleic acid metabolic process. Linoleic acid is a polyunsaturated fatty acid (PUFA) that is oxidized by endogenous enzymes and reactive oxygen species in the circulation. Recent research has shown that an increase in the synthesis of PUFA enhances lipid peroxidation and creates ferroptosis signals22,23. Li et al. demonstrated that 6-gingerol can inhibit ferroptosis by modulating the metabolism of linoleic acid24. These findings suggest a close association between the linoleic acid metabolic pathway and ferroptosis. The CC analysis revealed that the FRGs in PCD were primarily localized to the apical region of cells. Notably, iron absorption occurs at the apical membranes of intestinal epithelial cells25. Previous studies have demonstrated that cytochrome b reductase, located on the apical membrane of small intestinal enterocytes, facilitates the reduction of Fe³⁺ to Fe²⁺25. The resultant Fe²⁺ can subsequently react with PUFAs in cellular membranes, leading to excessive lipid reactive oxygen species (ROS) accumulation and ultimately triggering ferroptosis26. The MF analysis highlighted heme binding as a key enrichment term. The heme-binding protein hemopexin (HPX), a 60-kDa acute-phase reactant with ultra-high heme affinity, exerts cytoprotective effects by upregulating heme oxygenase-1 (HO-1) transcription27. HO-1, the rate-limiting enzyme in heme degradation, catalyzes heme breakdown into biliverdin, carbon monoxide, and free Fe²⁺. Substantial evidence underscores HO-1’s pleiotropic protective roles, including antioxidant, anti-inflammatory, and anti-apoptotic activities28. Preclinical studies demonstrate that HO-1 induction ameliorates gastrointestinal inflammation in IBD models via activation of anti-inflammatory cytokine networks29. Crucially, HO-1 overexpression suppresses ferroptosis and attenuates intestinal inflammation in IBD, suggesting a mechanistic link between heme metabolism and ferroptosis regulation30. In addition, KEGG enrichment analysis of this study demonstrated that the FRGs-DEGs were predominantly associated with metabolic pathways and IL-17 signaling pathway. Metabolomics analysis showed that 36 metabolites were significantly different between IBD patients and healthy controls31. Furthermore, metabolic pathways contribute to the regulation of ferroptosis sensitivity32. IL-17, a key pro-inflammatory cytokine that is increased in IBD33. Moreover, inhibition of the IL-17 signaling pathway can effectively suppress ferroptosis34. These findings are all consistent with our results. Besides the two pathways mentioned above, we unexpectedly observed significant enrichment of the Leishmania infection pathway in the KEGG analysis. Although this enrichment is statistically significant (P = 3.28E-04), it appears biologically implausible within the context of the current study. It may represent an analytical artifact potentially arising from shared gene sets. These results for both GO terms and KEGG pathways indicated that the FRG-DEGs identified in this study may participate in the progression of PCD through the aforementioned pathways. These findings lay the foundation for subsequent investigations into molecular mechanisms and may inform targeted interventions for PCD.

We also used Cytoscape’s Cytohubba plugin to analyze the FRGs-DEGs. Finally, the top five hub genes (PTGS2, IFNG, IL1B, IDO1, and NOS2) were chosen based on the scores, and their diagnostic potential was confirmed using ROC curves in the validation groups. PTGS2, also known as cyclooxygenase 2 (COX-2), is a specific biomarker of ferroptosis that worsens intestinal inflammation in IBD patients by promoting the release of prostaglandins and leukotriene E435. COX-2 expression is up-regulated in inflammatory circumstances because of the stimulation by several inflammatory factors. Histopathological analyses reveal 6–8 fold elevation of COX-2 expression in inflamed colonic tissues from ulcerative colitis and CD patients compared to healthy controls36. Notably, the COX-2 inhibitor NS-398 can ameliorate colonic dysmotility in rodent models of Crohn’s-like colitis37,38. Interferon gamma (IFN-γ, namely IFNG) is the sole member of the type II interferon family and plays a pivotal role in amplifying pro-inflammatory signaling by priming macrophages for enhanced antimicrobial responses39. Emerging evidence highlights the involvement of IFN-γ in the regulation of ferroptosis: T cell-derived IFN-γ synergizes with arachidonic acid to induce ferroptosis in immunogenic tumor cells40, while JAK1-2/STAT1/SLC7A11 axis mediates IFN-γ-driven ferroptosis in retinal epithelium41. Given its established role in CD pathogenesis42,43, we hypothesize that IFN-γ may also contribute to the progression of PCD through ferroptosis-related mechanisms. IL-1B is a critical pro-inflammatory cytokine that plays a role in a variety of autoimmune inflammatory responses as well as cellular functions such as cell proliferation, differentiation, and apoptosis44. Clinical studies have demonstrated significant upregulation of IL-1B in patients with CD compared to healthy controls45, aligning with our findings of its etiological significance in PCD. Targeted IL-1B neutralization thus represents a promising therapeutic strategy warranting further investigation. Indoleamine 2,3-dioxygenase 1 (IDO1) is a heme enzyme involved in the oxidation of L-tryptophan. IDO1, a key enzyme in tryptophan metabolism, was significantly up-regulated in both human IBD and animal models of colitis46. Preclinical studies demonstrate colitis attenuation through IDO1 inhibition47, suggesting its dual role as both disease biomarker and therapeutic target in PCD management. Nitric oxide synthase (NOS) comprises three distinct isoenzymes: neuronal NOS (nNOS/NOS1), inducible NOS (iNOS/NOS2), and endothelial NOS (eNOS/NOS3). Among these, NOS2 has been identified as a critical mediator of inflammatory responses48. NOS2 expression is significantly elevated in active CD49. Our immune infiltration analysis suggested a positive correlation between NOS2 and neutrophils (r = 0.69), which is consistent with its role in neutrophil recruitment and inflammatory cytokine production50. These findings suggest that NOS2 may influence PCD by regulating neutrophils in the intestine. Although these genes have been clearly demonstrated to play defined roles in adult CD, we independently screened and validated through bioinformatics approaches that these well-established genes indeed represent the most core drivers of ferroptosis in PCD. This finding corroborates the presence of a tightly coordinated regulatory relationship between classical inflammatory pathways and ferroptosis in PCD, thereby providing new perspectives for its diagnosis and treatment.

Given the immunological nature of PCD, we conducted focused investigations into immune cell participation in disease progression. Our analyses revealed significant enrichment of MDSCs and neutrophils in PCD patients, suggesting their potential role in maintaining intestinal immune homeostasis through regulatory mechanisms. MDSCs, also known as immature myeloid cells, including neutrophils and monocytes, MDSCs regulate immune responses in various pathological conditions, one of which is IBD51. Haile LA et al. found that MDSCs had a direct immune regulatory effect on IBD, and the frequency of MDSCs was increased in the peripheral blood of IBD patients52. Neutrophils, as first responders to microbial invasion, exhibit important roles in intestinal immunity. While essential for pathogen clearance, excessive neutrophil infiltration may induce mucosal damage and disrupt epithelial barrier integrity - pathological features strongly associated with CD progression. Notably, Gavriilidis et al. demonstrated a direct correlation between neutrophil aggregation density and CD lesion severity53, aligning with our current findings in PCD pathogenesis. Sha X et al. demonstrated that ferroptosis can promote the accumulation of MDSCs54, while Bao C et al. proposed that inhibiting ferroptosis can suppress the activation of neutrophils55. These findings indicate a close connection between ferroptosis and the regulation of immune cells, potentially providing a theoretical foundation for PCD-based therapeutic strategies. Nevertheless, the precise molecular mechanisms and pathways by which ferroptosis modulates immune cell function demand further elucidation.

In this investigation, we identified PTGS2, IFNG, IL1B, IDO1, and NOS2 as core ferroptosis regulators in PCD pathogenesis. These genes may drive disease progression through dual mechanisms: inducing ferroptosis in intestinal epithelium and orchestrating pro-inflammatory immune microenvironment via MDSC and neutrophil infiltration. Among these five hub genes, IL1B and IFNG demonstrate the most significant biological and clinical relevance in PCD. As a core pro-inflammatory mediator, IL1B directly drives the extensive acute inflammatory responses and tissue damage commonly observed in children with the condition. Meanwhile, IFNG, primarily produced by activated Th1 lymphocytes and NK cells, serves as a key cytokine promoting chronic intestinal inflammation and tissue injury. These genes not only function as robust biomarkers for disease activity but also represent highly promising therapeutic targets.

The clinical detection of these biomarkers requires careful consideration of accuracy, feasibility, and patient compliance in pediatric settings. Currently, two main approaches are employed: invasive and non-invasive techniques. Endoscopic biopsy remains an indispensable method. By obtaining affected intestinal mucosal tissue, proteins such as PTGS2, IL1B, and IFNG can be localized and semi-quantitatively analyzed using immunohistochemistry (IHC), or the mRNA expression levels of these genes can be detected via quantitative PCR (qPCR). In addition to endoscopic biopsy, non-invasive detection methods have also become a research focus. Serum levels of cytokines including IFNG and IL1B can be quantified using enzyme-linked immunosorbent assay (ELISA) or electrochemiluminescence immunoassays to assess systemic inflammatory status. IDO1 activity can be estimated indirectly in blood by measuring the kynurenine-to-tryptophan ratio. Intestinal inflammation may also be monitored through stool-based assays, which detect expression products (e.g., proteins or metabolites) of genes such as PTGS2, IL1B, and NOS2, or via RNA extraction from fecal cells followed by qPCR analysis. PCD is characterized by high heterogeneity and significant influence from developmental stage. Therefore, clinical implementation requires the establishment of age-stratified reference intervals and the development of minimally invasive, child-friendly testing technologies—such as ultra-low-volume blood collection and at-home stool sampling kits—to enable earlier and more precise disease management.

Our findings may provide a new theoretical framework for advancing precision diagnosis and developing targeted therapeutic strategies for this challenging pediatric condition. While our results reveal significant associations, several limitations must be acknowledged. First, the relatively small clinical cohort size may limit the statistical power of subgroup analyses. Second, it is important to note that the ssGSEA algorithm provides enrichment scores that are relative measures. Therefore, its results should be considered an indication of correlation, not a conclusive outcome. Lastly, the computational nature of this study requires subsequent experimental validation through in vitro and in vivo models. Future research should prioritize experimental validation of candidate genes and conduct longitudinal evaluations of ferroptosis biomarkers in larger patient cohorts.

Conclusion

In summary, this study establishes ferroptosis as a critical factor in PCD pathogenesis. We have identified five hub genes (PTGS2, IFNG, IL1B, IDO1, and NOS2) as potential diagnostic biomarkers and therapeutic targets, while also elucidating their mechanistic links to immune cell infiltration patterns. Among these, IL1B represents the most promising diagnostic biomarker for PCD due to its demonstrated superior specificity and sensitivity. These findings not only deepen our understanding of the molecular underpinnings of PCD pathogenesis, but also propose innovative strategies for precision diagnostics and targeted therapy.