Environmental toxicant ochratoxin A induces psoriasis based on network toxicology machine learning and molecular docking analyses

Abstract

Psoriasis is a chronic skin disease influenced by genetic susceptibility and environmental factors. Ochratoxin A (OTA) is a ubiquitous foodborne mycotoxin known for its immunotoxicity, however, its specific role in psoriasis pathogenesis remains underexplored. This study implemented an integrated systems biology framework that combined network toxicology, machine learning, and molecular dynamics to elucidate the mechanisms of OTA-induced psoriasis. By intersecting OTA-associated targets with psoriasis-related genes, we identified 242 potential targets that were significantly enriched in the IL-17 and TNF signaling pathways. We utilized a weighted gene co-expression network analysis combined with nine machine learning algorithms to identify five hub genes based on their high feature importance and diagnostic robustness: PNP, LCN2, HSPE1, TYMP, and CXCR2, with an area under the curve of 0.988 in the training set and 1.00 in the external validation. These hub genes positively correlated with activated dendritic cells and eosinophils, suggesting that they mediate a pro-inflammatory microenvironment. Molecular docking and dynamics simulations demonstrated stable binding affinities (up to − 8.8 kcal/mol) between OTA and the corresponding proteins. Our findings establish a mechanistic link whereby OTA directly interacts with key regulatory proteins to drive immune dysregulation, providing novel biomarkers and a theoretical basis for managing OTA-induced psoriasis.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37417-x.

Introduction

Psoriasis is a chronic, immune-mediated inflammatory skin disease characterized by erythematous, disfiguring skin plaques, extensive inflammatory cell infiltration, and frequent systemic comorbidities such as cardiovascular disease, all of which contribute to a markedly reduced quality of life^1,2. Although current therapeutic strategies, including biologics and systemic immunosuppressants, have significantly improved patient outcomes, they are often associated with high costs, long-term side effects, and high relapse rates². This underscores the urgent need to identify modifiable environmental triggers to improve prevention and personalized management.

Growing evidence indicates that environmental toxicants play a major role in the initiation and progression of chronic inflammatory diseases, including psoriasis^3–7. Yu et al. reported that polybrominated diphenyl ethers have a propensity to preferentially accumulate in the skin, thereby contributing to psoriasis⁵. Bisphenol is involved in the pathogenesis of psoriasis via immune regulatory pathway modulation. Another prevalent environmental toxicant frequently detected in contaminated food products and animal feed, ochratoxin A (OTA) is a mycotoxin produced by Aspergillus and Penicillium species^8,9. It exhibits a prolonged biological half-life and is commonly detected in human plasma at the nanomolar level, raising concerns regarding its potential chronic toxicity due to long-term exposure¹⁰. Multiple studies have suggested that persistent exposure to OTA can lead to multiorgan toxicity, including hepatotoxicity, nephrotoxicity, and reproductive dysfunction^11–13. In renal tissue, OTA preferentially accumulates in the proximal tubular epithelial cells, where it initiates cellular injury via oxidative stress, DNA damage, apoptosis, and inflammatory responses¹³. Furthermore, Boonen et al. demonstrated that, among various mycotoxins, OTA possesses the highest transdermal absorption efficiency in vitro, indicating its potential to exert direct toxic effects on the skin¹⁴.

The disruption of immune homeostasis is widely recognized as a major risk factor for psoriasis^1,2. OTA induces oxidative stress and immunotoxicity, both of which are key drivers of immune imbalance¹³. However, the specific molecular mechanism linking OTA exposure to the dysregulation of psoriatic skin remain unclear. This study aimed to fill this knowledge gap by employing an integrated systems biology approach to identify OTA-sensitive biomarkers and delineate the toxin-immune-skin axis.

In this study, we employed a comprehensive and integrative strategy combining network toxicology, machine learning, weighted gene co-expression network analysis (WGCNA), molecular docking, and immune infiltration analysis to elucidate the potential toxicity and molecular mechanisms of OTA in psoriasis. First, we constructed an integrated network toxicology framework to identify potential OTA-induced psoriasis targets and to delineate relevant signaling pathways involved in psoriatic pathogenesis. Subsequently, five robust key targets associated with OTA exposure and psoriasis-related clinical characteristics were identified using the combined application of WGCNA and machine learning. Lastly, molecular docking analysis provided evidence of potential direct interactions between OTA and the five identified hub proteins. Collectively, this study offers novel mechanistic insights into the role of OTA in psoriasis and provides a theoretical basis for early risk assessment and development of targeted preventive or therapeutic strategies for environmentally induced psoriatic disease.

Materials and methods

Screening of OTA-associated targets

The chemical structure and SMILES (Simplified Molecular Input Line Entry System) notation of OTA were obtained from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). To comprehensively predict potential molecular targets of OTA, we queried three complementary databases: ChEMBL (https://www.ebi.ac.uk/chembl/, SwissTargetPrediction (http://swisstargetprediction.ch/) and PharmMapper (https://lilab-ecust.cn/pharmmapper/). After integrating the results and removing duplicates (Supplementary data 1), a non-redundant list of OTA-associated targets was generated, serving as a foundational dataset for subsequent network and pathway enrichment analyses. The workflow is illustrated in (Fig. 1).

Fig. 1.

The workflow of this research.

Identification of potential psoriasis-related targets

Psoriasis-related targets were retrieved from the Gene Cards databases (https://www.genecards.org/) using the keyword “psoriasis”. A total of 4969 candidate genes (Supplementary data2) were identified and subsequently used for downstream analyses.

Functional enrichment analysis of OTA-associated psoriasis targets

Using the “clusterProfiler” R package, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed to investigate the pathways and molecular mechanisms associated with OTA-associated psoriasis targets¹⁵. The GO analysis was conducted across three functional categories: biological process, cellular component, and molecular function. To provide comprehensive overview of the functional roles of the identified genes, the top 10 significantly enriched GO terms and the top 15 KEGG pathways were selected and visualized.

Integration of multiple datasets and differential expression analysis of core genes

Transcriptomic data were obtained from three publicly available GEO datasets (https://www.ncbi.nlm.nih.gov/): GSE13355, GSE14905, and GSE30999, all of which include gene expression profiles from psoriatic and non-lesional skin samples. To minimize inter-dataset variability, the three datasets were integrated, and batch effects were corrected using the “ComBat” R package¹⁶. The merged dataset was then normalized using the “Normalization” function to ensure comparability across samples. Differentially expressed genes (DEGs) associated with OTA-associated psoriasis targets were identified using the “limma” R package¹⁷, with statistical significance defined as an adjusted p-value < 0.05 and an absolute log2 foldchange (|log₂FC|) > 1.

Weighted gene co-expression network analysis

We conducted WGCNA using the “WGCNA” R package to identify gene modules associated with psoriasis and explore underlying molecular network¹⁸. After filtering out low-expression genes and outlier samples, a sample clustering dendrogram was constructed to detect and exclude outliers and ensue network robustness. An appropriate soft-thresholding power (β) was selected based on the scale-free topology criterion to construct a weighted adjacency matrix. This matrix was then transformed into a topological overlap matrix, which quantified the network interconnectedness among the genes. Gene modules were identified using hierarchical clustering and dynamic tree cutting. Module eigengenes (MEs), which represented the first principal component of each module, were used to summarize the overall expression profile of genes within a given module. The association between gene modules and psoriasis was evaluated by calculating module significance (MS). Gene significance (GS), defined as the correlation between individual gene expression and clinical traits, was calculated to identify candidate hub genes with potentially biological relevance.

Construction of diagnostic models using multiple machine learning methods

To identify hub genes with robust diagnostic value, predictive models were developed using the following nine machine learning algorithms: SVM, KNN, RF, XGB, GLM, GBM, LASSO, NNET, and DT. Key OTA-associated psoriasis genes were used as input features, and the output label was the disease status. A total of 347 samples were randomly divided into training (70%, n = 244) and validation sets (30%, n = 103). The model training and hyperparameter tuning were performed using the caret R package¹⁹. All models were executed using default hyperparameters and evaluated via five-fold cross-validation to ensure reliability and generalizability. The diagnostic performance of each model was assessed using the area under the receiver operating characteristic curve (AUC), with visualization conducted using the pROC R package²⁰. The optimal model was selected based on the highest AUC values, and the top five most important features identified by the model were considered core diagnostic biomarkers associated with OTA-associated psoriasis pathogenesis.

Construction and validation of a nomogram

To facilitate individualized risk prediction, the core diagnostic genes were then utilized to construct a nomogram using the “rms” R package. In this model, each gene was assigned a weighted score according to its relative contribution, and the cumulative score was used to estimate the individual risk of psoriasis. The predictive performance and clinical applicability of the nomogram were systematically evaluated using calibration curves to assess the agreement between the predicted and observed outcomes. Decision curve analysis (DCA) was conducted to quantify the net clinical benefit of the nomogram across a range of threshold probabilities. Lastly, ROC curve analysis was conducted on the GSE54456 dataset to validate the diagnostic value of the model.

Evaluation of immune cell infiltration

To estimate the relative abundance of immune cell types within psoriatic tissue, the “CIBERSORT” R package was used in conjunction with the LM22 signature matrix²¹. This deconvolution approach enabled the quantification of 22 distinct immune cell subsets based on normalized gene expression profiles. Only samples with a CIBERSORT output p-value < 0.05 were deemed reliable for immune infiltration analysis. The inferred immune cell fractions were normalized to a total of 1, ensuring comparability across samples.

Molecular Docking

To explore the potential binding interactions between OTA and key target proteins, molecular docking simulations were performed to predict binding conformations and affinities.The two-dimensional chemical structure of OTA was obtained from PubChem, and the three-dimensional structures of human target proteins were sourced from the UniProt (https://www.uniprot.org/) and Protein Data Bank databases (https://www.rcsb.org/). Subsequently, molecular preprocessing, including the removal of water molecules, addition of hydrogen atoms, and charge calculations, was carried out using PyMOL (version 2.6). And, the receptor and ligand preparation, as well as definition of the docking grid box, were performed using AutoDock Tools (version 1.5.7). Lastly, docking simulations were executed with AutoDock, Vina, and the resulting complexes were visualized and analyzed using PyMOL software to assess the binding conformations and affinities between OTA and its core targets.

Molecular dynamics simulations

Molecular dynamics (MD) simulations were performed using GROMACS 2022. The proteins and ligands were described using the AMBER14SB and GAFF2 force fields, respectively. Ligand parameters were generated using Antechamber, and partial charges were calculated using the RESP method. The system was solvated with TIP3P water (≥ 1 nm from protein surface) and neutralized with 0.15 M NaCl. Long-range electrostatic interactions were treated with PME (cutoff = 1 nm), and bond constraints were enforced using LINCSx²². Energy minimization was performed in two stages: (1) constrained minimization of solvent and ions with protein/ligand fixed (3,000 steepest descent + 2,000 conjugate gradient steps); and (2) unconstrained minimization of the entire system (3,000 steepest descent + 7,000 conjugate gradient steps, with a convergence criterion of F_max < 1,000 kJ/(mol·nm)). The equilibration consisted of 100 ps NVT at 310 K (Nose-Hoover thermostat) followed by 100 ps NPT at 310 K and 1 bar (Parrinello-Rahman barostat, coupling constant = 0.5 ps). Production MD simulations were performed in an NPT ensemble for 50 ns at 310 K and 1 bar (time step = 2 fs; output frequency = 10 ps). The RMSD, RMSF, hydrogen bonds, R_g, and SASA were calculated using GROMACS tools to evaluate system stability and protein–ligand interactions.

Results

Identification of potential OTA targets and their overlap with psoriasis-related genes

To investigate the potential molecular mechanisms by which OTA contributes to psoriasis, we obtained the chemical structure of OTA from the PubChem database (Fig. 2A). To comprehensively predict the potential biological targets of OTA, three complementary databases were used: ChEMBL, SwissTargetPrediction, and PharmMapper (Fig. 2B). Following data integration and the elimination of duplicate entries, 556 unique potential OTA-associated targets were identified (Fig. 2C). Additionally, 4,969 unique psoriasis-related genes were obtained from the GeneCards database using “psoriasis” as the search term. Intersection analysis between OTA targets and psoriasis-related genes identified 242 overlapping genes (Supplementary data 3), accounting for 4.6% of the total psoriasis genes. These overlapping targets likely reflect the core molecular pathways through which OTA exerts pathogenic effects in psoriasis (Fig. 2C).

Fig. 2.

Identification of potential target proteins of OTA. (A) Chemical structure of OTA. (B) Predicted OTA targets were obtained from ChEMBL, PharmMapper, and SwissTargetPrediction databases. (C) Venn diagram illustrating the overlap between OTA-associated targets and psoriasis-associated genes.

Functional enrichment analysis of OTA-associated psoriasis targets

GO and KEGG pathway enrichment analyses were performed to systematically elucidate the biological functions and signaling pathways associated with OTA-induced psoriasis. GO analysis revealed that the overlapping targets were significantly enriched in processes related to immune response, oxidative stress, and epidermal regulation. Key biological processes included cellular responses to chemical and oxidative stress and cytokine-mediated signaling pathways (Fig. 3A). In the cellular component category, enrichment was observed in collagen-containing extracellular matrix, secretory granule lumen, and vesicle lumen (Fig. 3B). In the molecular function caegory, the targets were enriched in protease binding, serine-type endopeptidase activity, and ubiquitin-like protein ligase binding, indicating a critical role in protein modification and immune-related enzyme activity (Fig. 3C). KEGG pathway analysis revealed that OTA-associated psoriasis genes were primarily enriched in inflammatory and immune regulation pathways, including the IL-17 signaling pathway, C-type lectin receptor signaling, MAPK signaling pathway, and FoxO signaling pathway. Additionally, enrichment was observed in pathways related to apoptosis, oxidative stress (e.g., fluid shear stress and atherosclerosis), and viral infection (Fig. 3D). Several pathways were linked to lipid metabolism and endocrine resistance, suggesting that OTA has a broader involvement in systemic inflammatory and proliferative disorders. Lastly, the protein–protein interaction network based on these overlapping targets was constructed and visualized using Cytoscape software (Fig. 3E).

Fig. 3.

Gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) pathway enrichment analyses of OTA-induced psoriasis targets. (A) biological process (BP); (B) cellular component (CC); (C) molecular function (MF); (D) KEGG pathway enrichment analysis highlighting the top significantly enriched pathways related to the identified targets. (E) Protein–protein interaction network visualizes interactions among overlapping genes. Nodes are sized and colored based on their degree centrality to highlight network hubs.

Collectively, the OTA targets and psoriasis-associated genes networks converge on critical molecular processes involving inflammation, oxidative damage, and cell death, which may represent the critical mechanistic links underlying OTA-induced psoriasis pathogenesis.

DEGs of OTA-associated psoriasis targets

To explore the transcriptional landscape associated with OTA-induced psoriasis genes, we conducted a differential expression analysis between psoriatic and non-lesional skin samples across three datasets (Fig. 4A). Principal component analysis (PCA) revealed a clear separation between psoriatic and non-lesional samples along PC1 (12.9%), indicating substantial intergroup transcriptomic variation (Fig. 4B). A set of significantly dysregulated genes was observed in psoriatic samples, with upregulation of LCN2, CXCL8, HPSE, and CXCR2, and downregulation of RORA, AGTR1, and FABP4 (Fig. 4C). In total (Supplementary data 4), 21 genes were upregulated, and 3 were downregulated. A heatmap of these DEGs (Fig. 4D) showed distinct gene expression signatures, with many immune-related genes markedly upregulated in psoriatic lesions, including TNF, MMP9, GZMA, and LCK.

Fig. 4.

Differential expression analysis of OTA-associated psoriasis genes between psoriasis and non-lesional samples. (A) Boxplot showing the normalized expression values of genes in psoriasis (green) and non-lesional (red) groups; (B) PCA demonstrating clear separation between psoriasis and non-lesional samples; (C) Volcano plot of DEGs. Red represents significantly upregulated genes, and blue indicate downregulated genes, and gray dots represent non-significant changes (adjusted p < 0.05, |log2FC| > 1); (D) Heatmap of the DEGs. Red and blue colors indicate higher and lower expression levels, respectively.

Key gene modules and OTA-associated psoriasis targets

To elucidate the potential gene modules associated with OTA-induced psoriasis, we performed WGCNA to identify clusters of highly correlated genes and explore their relationships with disease traits. A soft-thresholding power of 9 was selected to ensure a scale-free network topology that met the criteria of scale independence (R² > 0.9) while maintaining sufficient mean connectivity (Fig. 5A). Using average linkage hierarchical clustering and dynamic tree cutting, eight distinct gene co-expression modules were identified and labeled with different colors (Fig. 5C).

Fig. 5.

WGCNA reveals key co-expression modules associated with OTA-associated psoriasis signatures. (A) Analysis of scale-free topology index and mean connectivity for various soft-thresholding powers; power = 9 was selected. (B) Dendrogram of module eigengenes showing clustering among modules. (C) Gene clustering dendrogram with assigned module colors using dynamic tree cutting. (D) Heatmap showing correlations between module eigengenes and sample traits (control vs. psoriasis). (E) Scatterplot of module membership vs. gene significance in the turquoise module (cor = 0.95, p < 1e–200). (F) Venn diagram displaying the overlap between DEGs and turquoise module genes.

Next, we examined the correlation between each module’s eigengenes and clinical traits. Among them, the turquoise module showed the strongest positive correlation with psoriasis and the strongest negative correlation with control samples, suggesting that it plays a crucial role in the disease process (Fig. 5D). Hierarchical clustering of module eigengenes confirmed the distinctiveness of the turquoise module (Fig. 5B). Furthermore, the gene significance for psoriasis exhibited a strong positive correlation (cor = 0.95, p < 1e–200) with module membership in the turquoise module, indicating that genes highly correlated with the trait were also highly representative of the module’s expression pattern (Fig. 5E). This highlights the biological importance of the turquoise module as a hub for psoriasis-associated transcriptional regulation.

Lastly, we intersected the genes in the turquoise module with the previously identified DEGs of OTA-associated psoriasis targets. Nine overlapping genes, representing 2.8% of the module genes, were identified via Venn diagram (Fig. 4F). These common genes are likely critical OTA-related targets in the pathogenesis of psoriasis.

Machine learning models and hub gene selection

To comprehensively evaluate the predictive value of the nine key genes associated with OTA-induced psoriasis, we used the following nine machine learning algorithms including RF, SVM, XGB, KNN, GBM, GLM, LASSO, NNET, and DT. As shown in Fig. 6A, the reverse cumulative distribution of absolute residuals revealed that LASSO and XGB models exhibited a greater proportion of low residuals, suggesting superior overall predictive accuracy and tighter error control. Correspondingly, the SVM, RF, and KNN model achieved the lowest root mean square error (RMSE), with RF model as the most stable performance across repeated sampling (Fig. 6B). To ensure the stability of the identified biomarkers, we implemented a consensus-based selection strategy. Genes were ranked based on their feature importance scores (RMSE loss after permutation) for each of the nine algorithms. The top five genes (PNP, LCN2, HSPE, TYMP, and CXCR2) were selected based on their consistently high ranking across most models (Fig. 6C).

Fig. 6.

Construction and performance evaluation of nine machine learning models, including RF, SVM, XGB, KNN, GBM, GLM, LASSO, NNET, and DT. (A) Cumulative residual distribution of each machine learning model. (B) Boxplots of residuals across models; red dots indicate the RMSE. (C) Top-ranked predictive features identified by each model. (D) ROC analysis of nine machine learning models in the testing cohort used to assess their diagnostic performance.

Lastly, ROC curve analysis was conducted to assess the diagnostic performance of each model (Fig. 6D). The LASSO model achieved the highest AUC value (AUC = 0.988), followed by the GLM (AUC = 0.981) and RF (AUC = 0.976) models, indicating its superior discriminative power in distinguishing psoriatic samples. In contrast, the DT model had the lowest AUC (AUC = 0.913). Despite its relatively high RMSE, the LASSO model was selected as the optimal diagnostic model because of its robust classification accuracy and effective feature selection capability.

Validation of the LASSO model based on five hub genes

To evaluate the clinical applicability of the five genes (PNP, LCN2, HSPE, TYMP, and CXCR2) derived from the LASSO model, we constructed a nomogram to predict the risk of psoriasis in individual patients (Fig. 7A). Each gene was assigned a corresponding score, and the total score enabled quantitative risk estimation for clinical decision-making. Calibration analysis demonstrated excellent agreement between the predicted and observed probabilities, indicating the high calibration accuracy of the nomogram (Fig. 7B). Lastly, decision curve analysis (DCA) confirmed that the five-gene nomogram provided a superior net clinical benefit across a wide range of threshold probabilities, supporting its potential utility in guiding personalized risk assessment (Fig. 7C).

Fig. 7.

Validation of the five hub genes-based LASSO model. (A) Nomogram constructed based on the five-gene signature selected by LASSO regression for predicting psoriasis risk. (B,C) Construction of calibration curve (B) and DCA (C) used to evaluate the predictive efficiency of the nomogram model. (D) ROC curve analysis for external validation of the five-gene model using the GSE54456 dataset.

This nomogram was applied to an independent external dataset (GSE54456) to validate the robustness of the model. The ROC curve revealed a strong discriminatory capacity, with an AUC of 1, demonstrating the stability and generalizability of the model across different cohorts (Fig. 7D).

Correlation between hub genes and immune cell infiltration in psoriatic lesions

To explore the immune microenvironment associated with OTA-induced psoriasis, we examined the relative distributions of 22 immune cell types in psoriatic lesions and normal skin tissues using the CIBERSORT algorithm. Significant differences were observed in the proportions of multiple immune subsets, including memory B cells, plasma cells, CD4⁺ memory T cells (both resting and activated), follicular helper T cells, natural killer (NK) cells, M1 macrophages, dendritic cells, and resting mast cells (Fig. 8A, B).

Fig. 8.

Landscape of immune cell infiltration in psoriatic lesions versus healthy skin. (A,B) Comparison of the relative abundance of 22 distinct immune cell types between psoriatic lesions and normal skin tissues, as estimated by the CIBERSORT algorithm. (C) Spearman correlation analysis between the five core hub genes (PNP, LCN2, HSPE1, TYMP, and CXCR2) and immune cell infiltration in psoriatic lesions.

Subsequently, we analyzed the relationships between the five hub targets and immune cell infiltration patterns (Fig. 8C). Correlation analysis revealed that the five hub genes were positively associated with the infiltration of activated dendritic cells and eosinophils. Moreover, PNP, LCN2, HSPE1, and CXCR2 positively correlated with activated CD4⁺ memory T cells. In contrast, these core genes negatively correlated with resting mast cells and naive CD4⁺ T cells, suggesting a potential role in shaping the pro-inflammatory immune landscape of psoriasis.

Molecular docking of OTA with the identified psoriasis targets

We performed molecular docking analysis to elucidate the potential interactions between OTA and the hub proteins. OTA exhibited strong binding affinities with all targets, with binding energies ranging from − 8.0 to − 8.8 kcal/mol (Fig. 9A–E). Specifically, OTA showed the highest binding affinity with HPSE (–8.8 kcal/mol), forming hydrogen bonds with key residues such as ARG-272, GLU-225, THR-97, GLN-383, and ASP-62 (Fig. 9A). In the case of CXCR2 (–8.7 kcal/mol), OTA interacted with the following residues, suggesting a stable ligand–receptor interaction: Gly-100, Ser-36, Ser-99, and Trp-104 (Fig. 9B). Lastly, docking with TYMP revealed a binding energy of − 8.0 kcal/mol, with OTA forming interactions with ASP-301, LYS-279, PHE-473, and PRO-470 (Fig. 9E). Collectively, these results indicated that OTA has the ability to form stable interactions with all five identified targets, supporting their potential roles as direct molecular mediators in the pathogenesis of OTA-induced psoriasis.

Fig. 9.

Molecular docking analysis of OTA with five hub target proteins. (A) OTA and HPSE. (B) OTA and CXCR2. (C) OTA and LCN2. (D) OTA and PNP. (E) OTA and TYMP.

Molecular dynamics of OTA and the identified psoriasis targets

The root mean square deviation (RMSD) is a standard metric for assessing protein-ligand conformational stability. As shown in Fig. 10A, the HPSE-OTA complex remained stable for 0–30 ns, fluctuating below 3.5 Å. The LCN2-OTA complex reached equilibrium after 10 ns with an RMSD value approximately 2.2 Å. The PNP-OTA complex equilibrated after 30 ns, stabilizing approximately 2 Å. The TYMP-OTA complex equilibrated was stable 40 ns, with an RMSD value approximately 3 Å. In contrast, the CXCR2-OTA complex exhibited continuous fluctuations, indicating conformational changes. The radius of gyration (Rg) characterizes the overall protein compactness, all five complexes exhibited minor Rg fluctuations during simulation (Fig. 10B), suggesting modest conformational changes upon ligand binding.

Fig. 10.

Molecular dynamics simulations of the protein-ligand complex. (A) RMSD as a function of simulation time. (B) Rg versus time. (C) SASA during the MD simulation. (D) Hydrogen bond formation between protein and ligand over time. (E) RMSF of the protein-ligand complex atoms.

The solvent-accessible surface area (SASA) reflects changes in the protein surface environment upon ligand binding. Figure 10C shows minor SASA fluctuations for HPSE-OTA, LCN2-OTA, TYMP-OTA, and CXCR2-OTA, indicating that small-molecule binding alters the local binding microenvironment. Hydrogen bonds play a crucial role in ligand-protein binding. Figure 10D shows hydrogen bond counts during the MD simulation: HPSE–OTA (mean, 2; range, 0–5), LCN2–OTA (mean, 2; range, 0–5), PNP–OTA (mean, 2; range, 0–4), TYMP–OTA (mean, 2; range, 0–6), and CXCR2–OTA (mean, 1; range, 0–3). These results demonstrate favorable hydrogen bonding interactions between OTA and the target proteins.

The root mean square fluctuation (RMSF) reflects amino acid residue flexibility. As shown in Fig. 10E, all complexes exhibited relatively low RMSF values (mostly < 3 Å), indicating high structural stability and low flexibility. Specifically, HPSE-OTA exhibited the lowest RMSF (mostly < 2.3 Å). In summary, OTA displayed stable binding and favorable hydrogen bonding interactions with HPSE, LCN2, PNP, and TYMP, indicating effective target engagement.

Discussion

OTA is a widespread environmental mycotoxin found in contaminated food products and animal feed. Studies have highlighted its immunotoxicity and organ-toxicity^8,9,13 however, its role in the pathogenesis of inflammatory skin diseases, particularly psoriasis, remains largely unknown. In this study, we investigated the contribution of chronic OTA exposure to the pathogenesis of psoriasis from the perspective of network toxicology. By integrating WGCNA with multiple machine learning models, we identified five robust OTA-associated hub genes (PNP, LCN2, HSPE1, TYMP, and CXCR2) that were closely linked to immune dysregulation and demonstrated strong discriminatory power in psoriasis risk. These findings underscore the potential link between OTA exposure and psoriasis, providing a basis for the development of novel diagnostic and therapeutic strategies.

Environmental toxicants have been increasingly recognized for their involvement in the pathogenesis of various diseases. An epidemiological study revealed that long-term exposure to NO₂ and NOₓ is significantly associated with an elevated risk of developing psoriasis⁶. Using network toxicology, Feng et al. reported that bisphenol exposure increases the risk of psoriasis⁷. Additionally, particulate matter 2.5 exposure exacerbates psoriatic severity by upregulating the expression of keratin 17, S100a8, and S100a7a, which is associated with activation of the AKT/mTOR/HIF-1α signaling pathway²³. Other studies have also demonstrated that particulate matter can disrupt keratinocyte differentiation and upregulate genes associated with psoriatic skin disease²⁴. With regard to OTA, exposure has been shown to impair both innate and adaptive immunity by affecting immune organs²⁵. Notably, OTA influences the differentiation of T cells into Th1 and Th17 subsets, particularly during the regulation of humoral immune responses²⁶. Kumar et al.. reported that non-cytotoxic concentrations of OTA promote DNA synthesis in primary murine keratinocytes by activating EGFR signaling and its downstream MAPK signaling pathways²⁷. Collectively, these findings suggest an association between OTA exposure and the development of psoriasis.

In this study, we identified 242 potential toxicological targets associated with OTA-induced psoriasis, underscoring the significant convergence between mycotoxin-associated pathways and gene signature characteristics of psoriatic skin lesions. GO and KEGG enrichment analyses revealed that these targets were significantly enriched in immune-related pathways, including oxidative stress responses, cytokine signaling (e.g., IL‑17 signaling pathway, TNF signaling pathway), and the MAPK/FOXO signaling pathways. These pathways are well-established drivers of keratinocyte hyperproliferation and immune cell activation in the pathogenesis of psoriasis^2,28. Moreover, the high transdermal permeability of OTA suggests that cutaneous exposure may directly disrupt these molecular pathways, thereby establishing a mechanistic link between environmental toxin exposure and skin inflammation¹⁴.

To further refine the key molecular signatures of OTA-induced psoriasis, we used WGCNA in combination with nine machine learning algorithms to identify five hub genes: PNP, LCN2, HSPE1, TYMP, and CXCR2. These genes are involved in critical pathogenic processes underlying psoriasis. To facilitate clinical translation, we constructed a nomogram based on this five-gene signature. External validation using the independent GSE54456 cohort achieved an AUC of 1.00, highlighting its exceptional predictive performance. The high predictive accuracy observed in the external datasets substantiates the robustness of our model, likely reflecting the synergistic value of integrating toxicogenomic and transcriptomic approaches.

Among these targets, each gene serves as a potential molecular bridge linking OTA-induced oxidative stress and immunotoxicity to key pathogenic processes in psoriasis. LCN2 plays a well-established role in the pathogenesis of psoriasis. Shao et al. reported that reported that LCN2 is significantly upregulated in psoriatic skin lesions and promotes neutrophil chemotaxis, keratinocyte activation, and pro-inflammatory cytokine release via ERK1/2 and p38-MAPK pathway activation²⁹. Furthermore, serum LCN2 levels are significantly associated with the intensity of itching in patients with psoriasis³⁰. CXCR2 is a receptor for chemokines such as IL-8. In imiquimod-induced mouse models of psoriasis, CXCR2 and BLT1 facilitate early phase neutrophil infiltration into the skin, a key event for initiating keratinocyte activation and inflammation³¹. Cataisson et al. reported similar results, wherein CXCR2 deficiency significantly impaired neutrophil infiltration into mouse skin³². Therefore, given its critical role in mediating neutrophil-driven inflammation, CXCR2 is a promising therapeutic target for the treatment of psoriasis. PNP is a purine-metabolizing enzyme, and PNP deficiency triggers apoptotic cell death in T cells. Several studies have suggested targeting PNP as a potential therapeutic approach for psoriasis³³. Bantia et al. reported that the PNP inhibitor RO5092888 induces apoptosis in both T and B lymphocytes. Indeed, this inhibitor is currently in early clinical trials for the treatment of psoriasis³⁴. TYMP markedly enhances endothelial cell proliferation and neovascularization^35,36. Marked microvascular dilation and tortuous and elongated capillaries are hallmark features of psoriatic lesions². The upregulation of TYMP in these lesions likely facilitates localized vascular remodeling processes, contributing to the persistence of a chronic inflammatory microenvironment.

The immune cell infiltration analysis revealed abnormal immune cell infiltration in psoriatic lesions, including increased numbers of memory B cells, CD4⁺ T cell subsets, M1 macrophages, and dendritic cells, alongside a reduction in resting mast cells. These findings imply that OTA-induced immunotoxicity establishes a feed-forward loop in which oxidative stress and direct protein interactions disrupt keratinocyte homeostasis, promoting dendritic cell maturation and a Th17/Th1-mediated immune response that subsequently exacerbates epidermal hyperproliferation and inflammation^2,37. Additionally, we simulated molecular docking of OTA with the three-dimensional structures of the five hub proteins to explore direct biochemical mechanisms. The simulations demonstrated stable binding between OTA and HSPE, a critical regulator of cellular redox homeostasis^38,39. Such interactions may impair the mitochondrial chaperone function of HSPE and exacerbat oxidative damage. The binding of OTA to CXCR2 and TYMP may interfere with chemokine-mediated neutrophil recruitment and nucleotide metabolic pathways, respectively, both of which are critical in psoriatic pathophysiology^32,35. Despite the limitations of computational docking and MD simulations in reflecting physiological conditions, the observed binding interactions offer testable hypotheses for experimental validation and potentially therapeutically actionable targets.

Inevitably, this study has several limitations that should be considered. First, the large-scale human data used in this study was sourced from publicly available datasets. To minimize potential bias, we integrated multiple datasets to explore the association between OTA exposure and psoriasis. However, in vitro and in vivo experiments are necessary to confirm the direct effects of OTA on skin and immune cell populations in psoriasis. Second, the immune cell composition was inferred using the CIBERSORT algorithm, which may be biased by gene expression variability. Lastly, the constructed nomogram demonstrated excellent predictive performance in the GSE54456 cohort; however, its clinical applicability requires further validation in larger multicenter cohorts. Future work should focus on the following aspects: (1) epidemiological studies correlating dietary OTA exposure with the incidence of psoriasis to establish population-level causality; (2) exposing primary keratinocytes or organotypic skin cultures to OTA and assessing oxidative stress markers, cytokine secretion, and gene expression of the five hub genes; and (3) constructing specific knockouts of LCN2 or CXCR2 to clarify their roles in OTA-induced inflammation.

Conclusion

In conclusion, our integrated network toxicology and multi-omics analyses suggest that OTA may serve as an under-recognized environmental trigger for psoriasis. WCGNA and machine learning algorithms identified five target genes that may be central mediators of OTA-induced psoriatic effects. Molecular docking and MD simulations confirmed the stable interactions between OTA and these core targets. Collectively, these findings provide a basis for an in-depth study of OTA-induced psoriasis and offer novel insights and potentially actionable therapeutic targets.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Jian Hu: Methodology, Network toxicology analysis, Writing—Reviewing and Editing. Ming Tang: Supervision, Funding acquisition, Writing- Reviewing. Quan-you Zheng: Data curation, Molecular docking, Software. Shen-ju Liang: Visualization, Writing—Original draft preparation. Gui-lian Xu: Writing—Original draft preparation, Validation. Ke-qin Zhang: Supervision, Funding acquisition, Formal Analysis.

Funding

This work was funded by the National Natural Science Foundation of China (Youth Program, Grant No. 82300855), China Postdoctoral Science Foundation (General Program, Grant No. 2021M700634) and the Natural Science Foundation of Chongqing (Grant No. CSTB2022NSCQ-MSX0099).

Data availability

The datasets GSE13355, GSE14905, GSE30999, and GSE54456 were sourced from the GEO databases (https://www.ncbi.nlm.nih.gov/geo/).

Declarations

Competing interests

The authors declare no competing interests.