MCD biomarkers Egfr, Hmox1, Lgmn identified in NAFLD

Abstract

Background

Non-alcoholic fatty liver disease (NAFLD) is associated with metabolic cell death (MCD), and this study aimed to dig deeper into the biomarkers associated with MCD in NAFLD, and to provide new references for the diagnosis and treatment of NAFLD.

Methods

The datasets and MCD-related genes (MCD-RGs) associated with NAFLD were downloaded from the Gene Expression Omnibus (GEO) database and the literature, respectively. Differentially expressed genes (DEGs) between NAFLD and control groups were identified and intersected with MCD-RGs to yield candidate genes. Biomarkers were obtained by screening under four machine learning models, Receiver Operating Characteristic (ROC) curves, and expression validation. Based on the biomarkers, functional enrichment, diagnostic model construction, network modulation, and reverse transcription-quantitative polymerase chain reaction (RT-qPCR) were performed. At the same time, differential infiltration of immune cells in the NAFLD and control groups was analysed.

Results

The 17 candidate genes were mostly involved in processes such as immunity and apoptosis. After screening, Egfr, Hmox1, and Lgmn were identified as biomarkers. Among these, Egfr was down-regulated whereas Hmox1 and Lgmn were up-regulated in NAFLD. Based on these biomarkers, a nomogram diagnostic model was constructed and demonstrated excellent predictive performance (AUC = 0.997). Subsequent enrichment analyses showed enrichment in inflammatory regulation between biomarkers and NAFLD groups. In addition, in the TF-biomarker network, Egfr and Hmox1 co-predicted NF-κB1. SORAFENIB was co-predicted in drug prediction. Meanwhile, five differentially infiltrating immune cells, such as CD8 T cells, were found to be strongly negatively correlated (cor = -0.475) with Egfr in both the NAFLD and control groups.

Conclusion

In this study, Egfr, Hmox1, and Lgmn were used as biomarkers showing transcriptomic correlation with with MCD in NAFLD, and an excellent nomogram diagnostic model was developed accordingly, which is expected to provide a practical tool for diagnosis and treatment of NAFLD.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12902-026-02354-3.

Introduction

Non-alcoholic fatty liver disease (NAFLD) represents a pandemic metabolic disorder, currently affecting 25% of the global population. In China, projected cases will rise from 243.67 million (2016) to 314.58 million (2030), with global prevalence potentially reaching 55.4% by 2040 [1]. NAFLD includes non-alcoholic fatty liver disease (NAFL) and non-alcoholic steatohepatitis (NASH), with NAFL mainly characterized by steatosis with or without mild inflammation, while NASH is characterized by hepatocellular damage and fibrosis [2]. Although considered “benign”, simple steatosis is closely associated with liver fibrosis, which can lead to cirrhosis and hepatocellular carcinoma (HCC), making NAFLD an important factor in mortality from liver-related diseases [3, 4]. This systemic disease arises from dysregulated lipid metabolism, insulin resistance, and gut-liver axis disruption, with persistent inflammation emerging from tripartite interactions among metabolic dysfunction, microbiota imbalance, and immune dysregulation, presenting substantial clinical hurdles. For instance, Pregnane X receptor (PXR) shows a dual role in NAFLD, promoting steatosis while reducing inflammation and fibrosis, highlighting therapeutic complexity [5]. Current international guidelines recommend limited specific medications, mainly vitamin E and pioglitazone, but a combination of lifestyle modifications, increased physical activity, and cessation of smoking and alcohol use is considered beneficial [6, 7]. Therefore, due to the limitations of existing treatment options and the uncertainty of their efficacy, it has become imperative to explore the molecular pathogenesis of NAFLD and identify new therapeutic targets.

Among these mechanisms, metabolic cell death (MCD) has emerged as a relevant yet underexplored contributor to NAFLD pathogenesis. MCD refers to a type of cell death directly induced by metabolic imbalance, encompassing ferroptosis, necroptosis, and autophagy-dependent cell death [8]. It plays a pivotal role in the progression of non-alcoholic fatty liver disease (NAFLD), particularly ferroptosis and necroptosis, which are closely linked to lipid peroxidation, oxidative stress, and inflammatory responses [9, 10].In the progression of NAFLD, ferroptosis serves as an early trigger, inducing liver damage and inflammatory responses. As the disease progresses, this leads to a vicious cycle of iron accumulation and lipid peroxidation: In the NASH stage, enhanced arachidonic acid metabolism triggers a lipid ROS burst, with Fe²⁺synergistically exacerbating lipid peroxidation. Concurrently, oxidative stress damages mitochondria, impairing pancreatic β-cell and hepatic cell function, worsening insulin resistance, and forming an “iron death-oxidative stress-metabolic imbalance” positive feedback loop that accelerates progression to NASH [11, 12]. Beyond ferroptosis, necroptosis, another MCD subtype, also contributes to NAFLD pathogenesis. Necrotic apoptosis regulates hepatic mitochondrial metabolism and adipocyte function through RIPK3 in the development of NAFLD [13] and promotes NAFLD-associated carcinogenesis via a RIP3 kinase-dependent mechanism, exacerbating inflammatory fibrosis and increasing the risk of precancerous nodules. Inhibiting RIP3 signaling improves liver injury and reduces tumorigenesis. However, most studies in this area have focused on animal models, and there is a lack of evidence to validate the mechanisms of MCD at the molecular level and from a clinical patient perspective. How to differentiate iron death from other programmed cell deaths during the disease process for targeted prevention and treatment remains to be explored.

Previous studies have shown that integrated bioinformatics approaches, including network pharmacology and machine learning, can identify gene-pathway interactions and clinically relevant biomarkers in NAFLD [14, 15]. However, most studies have focused on a single form of programmed cell death, such as ferroptosis or apoptosis. In contrast, the present study incorporates multiple metabolic cell death (MCD) processes, including ferroptosis, cuproptosis, disulfidptosis, lysosomal zinc death, and alkaliptosis. Based on this framework, we integrated MCD-related genes with NAFLD transcriptomic data to identify candidate biomarkers and construct diagnostic models. This approach may provide a more comprehensive understanding of MCD-related mechanisms in NAFLD and offer directions for future research.

Materials and methods

Data source

The GSE39549 dataset (platform: GPL6887), sourced from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), served as the training set. It included liver tissue samples from 24 C57BL/6J mice with non-alcoholic fatty liver disease (NAFLD) and another 24 normal C57BL/6J mice (Control). The validation set selected for this study was the GSE52333 dataset (platform: GPL6246), containing liver tissues from 18 NAFLD mice and 18 normal mice, and the strains of mice were consistent with the training set. All datasets were obtained from GEO as preprocessed, normalized matrices (background-corrected, within-array normalized). Analyzed separately (training vs. validation), not merged; no batch correction (ComBat/SVA). Robustness: consistent cross-dataset findings. From the literature [8] for ferroptosis, cuproptosis, disulfidptosis, lysosomal zinc death, and alkaliptosis, five types of related genes, among them, ferroptosis-related genes from the FerrDb website (http://www.zhounan.org/ferrdb) to download, would merge five types of genes to remove duplications, and there were a total of 684 genes after gene conversion to mouse genes, as metabolic death-related genes (MRGs) (Supplementary Table 1).

Screening for differentially expressed genes (DEGs) between NAFLD and controls

To balance sensitivity and specificity, thresholds were set at log₂|fold change (FC)| > 0.5 to capture subtle metabolic changes and an uncorrected p < 0.05 to maximize preliminary gene discovery. Using these criteria, the limma package (v 3.56.2) [16] was used to analyse the DEGs between NAFLD and control groups in the training set, and the ggplot2 package (v 3.4.1) [17] to visualise top 10 up and down DEGs in the log₂|FC| descending order. Meanwhile, the ComplexHeatmap package (v 2.14.0) [18] was used to visualise the expression heatmap of the DEGs.

Exploring the multiple roles of candidate genes

DEGs and MCD-RGs were taken to intersected to get genes associated with both NAFLD and MCD. The intersected genes were noted as candidate genes, and Venn diagram was visualised using the VennDiagram package (v 1.7.1) [19]. Then, clusterProfiler (v 4.2.2) [20] was applied to perform Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses on the candidate genes (with adjusted p-values < 0.05, Benjamini-Hochberg method), aiming to explore the underlying mechanisms of their functions from multiple perspectives. GO enrichment analysis zeroed in on the biological roles of candidate genes, which were categorized into three dimensions: biological process (BP), cellular component (CC), and molecular function (MF). The enrichplot package (version 1.18.3) was utilized to generate plots for the top 5 most prominent features in each category, sorted by ascending adjusted p-values [21]. KEGG enrichment focused on the signaling pathways involved in the candidate genes, and GOplot (v 1.0.2) [22] showed the top 10 most significant pathways (adj p value ascending).

Multi-step biomarker identification: machine learning, ROC analysis, and differential expression validation

To further screen the candidate genes to obtain NAFLD-related characteristic genes, first, four machine learning models were built in the training set using the caret package (v 6.0–93) [23]: Random Forest (RF), Gradient Boosting Machine (GBM), Support Vector Machine (SVM) and generalized linear model (GLM). Subsequently, the DALEX package (v 2.4.3) [24] was used to analyze the residuals of the classifier, and ggplot2 (v 3.4.1) was used to draw a boxplot to show the absolute value of the residual (|residual|) of the model. The smaller the residual, the stronger the prediction ability of the model. Finally, genes were ranked by importance (descending RMSE) from the optimal model, and the top 10 were selected as characteristic biomarkers. This predefined cutoff balanced diagnostic performance and overfitting risk while yielding a clinically manageable and robust subset for nomogram construction.

In the identification of the ability of the characteristic genes to distinguish disease samples, the Receiver Operating Characteristic (ROC) curve of the characteristic genes was drawn by the pROC package (v 1.18.0) [25] in GSE39549 and GSE52333. The area under the curve (AUC) in both datasets was greater than 0.7, indicating that the characteristic gene could distinguish diseases from controls and was recorded as a candidate biomarker. Finally, the Wilcoxon test was used to test differential expression of candidate biomarkers in NAFLD and control samples, and the genes that were differentially expressed and had the same trend in both datasets were used as biomarkers in this study (p < 0.05).

Diagnostic value of biomarkers

Within the training set, the rms package (version 6.5-1) was employed to construct a nomogram diagnostic model grounded in the selected biomarkers [26]. This model leverages the biomarkers to estimate the likelihood of NAFLD in samples; a higher total score corresponds to a greater probability of the disease. Next, the calibration curve for the nomogram was generated, and the Hosmer-Lemeshow test was applied to assess the discrepancy between predicted values and actual outcomes. When the p-value was greater than 0.05 and the Mean absolute error (MAE) was less than 0.1, it implied that the model exhibited robust predictive performance for NAFLD. Regarding the nomogram’s predictive capacity, decision curves and ROC curves were generated using the rmda package (version 1.6) [27]and pROC package (version 1.18.0), respectively, with evaluations focusing on both clinical utility and diagnostic accuracy.

Enrichment analysis of genes and samples

The biomarkers were first analysed by Gene Set Enrichment Analysis (GSEA) by clusterProfiler (v 4.2.2) to explore the signalling pathways in which individual biomarkers are involved. Spearman correlation coefficients were calculated for biomarkers and all genes in the training set of NAFLD samples and arranged in descending order. Next, the “Hallmark gene set” was retrieved from the Molecular Signature Database (MSigDB) (https://www.gsea-msigdb.org/gsea/msigdb) to serve as the reference gene set, with subsequent analysis focusing on the pathways associated with the biomarker. The thresholds were set as false discovery rate (FDR) < 0.05 and p < 0.05. The top 5 signalling pathways were displayed using the enrichplot package (v 1.18.3), sorted in ascending order of p-values.

Similarly, enrichment scores for specific gene sets (Hallmark gene sets) were extracted from the gene expression data of individual samples using the GSVA package (v 1.48.3) [28] and the single sample GSEA (ssGSEA) algorithm, and a p-value less than 0.05 was considered to be a significantly different pathway in the NAFLD and control groups.

The unique immune microenvironment of NAFLD

Using the training set as a basis, the mMCP counter algorithm was applied to quantify the infiltration levels of 12 immune cell types in samples from both the NAFLD and control groups [29]. Differences in immune cell infiltration between the two groups were then compared using the Wilcoxon test from the rstatix package (version 0.7.2; https://CRAN.R-project.org/package=rstatix), with a significance threshold of p < 0.05. Additionally, Spearman’s correlation analysis was used to assess relationships between differentially abundant immune cells and biomarkers, applying criteria of |cor|>0.3 and p < 0.05.

Network regulation and drug prediction for targeted biomarkers

To find genes functionally related to biomarkers, the top 20 most relevant co-expressed genes with biomarkers were searched for in the GeneMANIA database (https://genemania.org/) to map a gene-gene interaction (GGI) network. The upstream and downstream genes of the biomarker and the way of interaction were explored using the DisNor database (https://disnor.uniroma2.it/). Furthermore, the TRRUST database (https://www.grnpedia.org/trrust/) was utilized to predict upstream transcription factors (TFs) targeting the biomarkers, and a TF-biomarker interaction network was visualized with Cytoscape (version 3.8.2) [30].

Reverse transcription quantitative polymerase chain reaction (RT-qPCR)

RNA was extracted from 20 samples using the TRIzol kit, with 11–20 being control samples and 1–10 being NAFLD liver tissue samples, which were collected from the Guizhou University of Traditional Chinese Medicine. The study had been approved by the Ethics Committee of the Guizhou University of Traditional Chinese Medicine. Statistical plots were generated using GraphPad Prism software.

Statistical analysis

All analyses were performed in R software (v 4.1.0). Group differences were evaluated via the Wilcoxon test, while a t-test was used to compare intergroup variations in RT-qPCR data. A p-value < 0.05 was considered statistically significant.

Results

Seventeen candidate genes were identified through screening

Compared with the control group, the NAFLD group showed 258 DEGs, among which 192 were upregulated and 66 were downregulated (Fig. 1A-B). DEGs were intersected with 684 MCD-RGs to obtain 17 candidate genes: Lcn2, Cd74, Cd44, Lgmn, Fads2, Pltp, Il1b, Pcsk9, S100a8, Timp1, Acot3, Hmox1, Lpin1, Gdf15, Fgf21, Chac1, and Egfr (Fig. 1C).

Fig. 1.

Identification of differentially expressed genes and candidate genes in NAFLD (A) Volcano plot of differentially expressed genes between NAFLD and Control. Orange dots indicate genes with increased expression, blue dots indicate genes with decreased expression, and gray dots indicate genes with no significant change in expression. (B) Heat map of differentially expressed genes between NAFLD and Control. The image consists of two parts: the upper part is a heat map of the expression density of sample-specific genes, showing five quantiles and the mean value.The next section shows the heat map of gene expression differences between samples. (C) Identification of candidate genes. (D) Bagua diagram.The inner circle on the left is a bar chart, where the height of the bars represents the significance of the term, with higher bars indicating greater significance. The color of the bars represents the z-score, with darker colors indicating higher z-scores. The z-score is not a standard statistical value, but it can suggest that a biological function is more likely to be upregulation or downregulation; the outer circle shows a scatter plot of the expression levels of each gene in each Term, with red and blue representing upregulation and downregulation, respectively. On the right is the description of the GO enrichment terms. (E) KEGG enrichment results and chord diagram of candidate genes.Note: The colors of the gene bands on the left represent the logFC of the gene, and the different bands on the right represent different pathways in this analysis

Candidate genes involved in 749 biological functions and 10 pathways

GO analysis revealed that the candidate genes were enriched to a total of 749 entries, including 682 BPs, 6 CCs, and 61 MFs. These terms were mainly associated with inflammatory response and wound healing-related pathways. leukocyte aggregation, ERK1/2 activation, neuronal apoptosis, and trauma-induced inflammation. the enrichment results of CCs positioned the candidate genes to exert their functions at late endosome, multivesicular body, membrane raft, membrane microdomain and receptor complex. meanwhile, the results of MFs demonstrated that the candidate genes exerted more detailed regulatory roles such as low-density lipoprotein particle binding, growth factor activity, growth factor receptor binding, lipoprotein particle binding and protein-lipid complex binding (Fig. 1D, Supplementary Table 2). In addition, candidate genes were significantly enriched in the IL-17 and HIF-1 signaling pathways (Fig. 1E, Supplementary Table 3). Enrichment in the IL-17 and HIF-1 signaling pathways suggests potential involvement in inflammation, fibrosis, and hypoxia-related metabolic reprogramming in NAFLD.

Identification of Egfr, Hmox1, and Lgmn as candidate biomarkers

The cumulative residual distributions and residual plots for the four machine learning models are presented in Fig. 2A-B. Figure 2C illustrates the top 10 candidate genes in each model. Combining the above 3 figures, the GBM model possesses the lowest absolute value distribution of residuals and the strongest predictive ability, so the top 10 important candidate genes in the GBM model were taken as characteristic genes (Lgmn, Pltp, Acot3, Chac1, Fgf21, Egfr, Gdf15, Fads2, Lpin1, and Hmox1).

Fig. 2.

Four machine learning classifiers and feature importance analysis charts, (A) Cumulative residual distribution of four machine learning classifiers. (B) Residual plot of four machine learning classifiers (red dots indicate root mean square residuals). The horizontal axis represents the absolute value of the residuals (|residual|), ranging from 0.0 to 0.2. The vertical axis represents the inverse cumulative percentage. (C) Feature importance analysis.Note: The horizontal axis of the left figure represents the loss value after variable replacement, and the vertical axis represents the ten genes with the smallest machine learning loss values for each of the four machine learning methods. (D) ROC analysis of feature genes in the training set GSE39549 (left) and validation set GSE52333 (right). (E) Box plot of candidate gene expression levels in the training set GSE39549 (left) and validation set GSE52333 (right). The horizontal axis represents groups, and the vertical axis represents gene expression levels. Statistical significance is denoted as follows: ns (not significant, p ≥ 0.05), * (p < 0.05), ** (p < 0.01), *** (p < 0.001), and **** (p < 0.0001)

In both datasets, the characteristic genes Lgmn, Acot3, Fgf21, Egfr, Fads2, and Hmox1 had AUC values greater than 0.7 (Fig. 2D), and Egfr, Hmox1, and Lgmn displayed significantly consistent expression patterns between the NAFLD and control groups (p < 0.05) (Fig. 2E). Egfr expression was down-regulated in the NAFLD group, and Hmox1 and Lgmn were up-regulated in the NAFLD group. Therefore, Egfr, Hmox1, and Lgmn were used as biomarkers of NAFLD in this study.

Construction and validation of a nomogram diagnostic model based on biomarkers

The nomogram constructed by biomarkers is shown in Fig. 3A, and its calibration curve had a p-value of 0.674 (p > 0.05) and an MAE value equal to 0.024 (less than 0.1), and the prediction curve is close to overlapping with the ideal curve, indicating that the prediction of the nomogram has high accuracy (Fig. 3B). Compared with individual biomarkers, the gain of the nomogram was far above the biomarkers (Fig. 3C). Notably, the AUC value of the model’s ROC curve was 0.997 (Fig. 3D), which is close to 1, again demonstrating the excellent predictive ability of the nomogram. A tool with excellent potential for the diagnosis of NAFLD in terms of MCD.

Fig. 3.

Nomogram model diagram. (A) Key gene construction of nomogram model Note: The variables in the predictive model are key genes. Each variable is labeled with a scale on the corresponding line segment, representing the range of values that the variable can take, while the length of the line segment reflects the contribution of that factor to the outcome event.The single scores in the figure represent the single scores corresponding to each variable under different values.The total score in the figure represents the sum of the individual scores corresponding to all variable values.The “Risk of NAFLD” in the figure indicates the risk of NAFLD in the sample. (B) Calibration curve of the nomogram model.Note: The horizontal axis represents the nomogram-predicted probability of disease, and the vertical axis represents the actual probability of NAFLD. Perfect prediction corresponds to the black dashed line. The blue solid line indicates the actual prediction results, and the black solid line represents the prediction results after bias correction through Bootstrapping (1000 repetitions). (C) Decision curve.The horizontal axis represents the High Risk Threshold, and the vertical axis represents the Net Benefit. Each curve represents different genes (Egfr, Hmox1, Lgmn) and the nomogram model at different thresholds. (D) ROC curve of the nomogram model

Biomarkers and NAFLD samples enriched to similar signaling pathways

In the context of the Hallmark gene set, Egfr was significantly enriched for 29 pathways. These include genes that define epithelial-mesenchymal transition (linked to wound healing, fibrosis, and metastasis), genes mediating caspase-activated programmed cell death (apoptosis), and genes encoding proteins involved in oxidative phosphorylation; genes regulated by NF-κB in response to TNF and genes up-regulated during transplant rejection(Fig. 4A, Supplementary Table 4); Hmox1 for 34 pathways involving genes up-regulated during transplant rejection, genes regulated by NF-κB in response to TNF; genes mediating programmed cell death (apoptosis) by activation of caspases; genes up-regulated by IL6 via STAT3, e.g., during acute phase response(Fig. 4B, Supplementary Table 5). Lgmn demonstrated enrichment in 30 apoptotic pathways characterized by caspase cascade activation; upregulated in transplant rejection contexts and those under NF-κB transcriptional control upon TNF stimulation (Fig. 4C, Supplementary Table 6). The biomarkers showed co-enrichment in both caspase-dependent apoptotic pathways and TNF-induced NF-κB transcriptional networks. In the same background gene set, there were 28 significantly different pathways between the NAFLD and control groups, including TNF-α signaling via NF-κB, allograft rejection, IL6-JAK-STAT3 signaling, and others(Fig. 4D). Biomarkers are all associated with inflammation and apoptosis, and in addition, both biomarkers and NAFLD are involved in the inflammatory response.

Fig. 4.

Key gene enrichment signaling pathway. (A) EGFR enrichment signaling pathway. The figure can be divided into three parts. First part: The five lines in the upper left are the line graph of the gene Enrichment Score. The vertical axis represents the corresponding Running ES, and there is a peak in the line graph, which is the Enrichment score of this gene set; the genes before the peak are the core genes under this gene set. The horizontal axis represents each gene under this gene set, corresponding to the vertical lines similar to barcodes in the second part. Part 2: The section resembling a barcode is Hits, where each vertical line corresponds to a gene within that gene set.Part Three: The sorting situation of the right-side genes. (B) Hmox1 enrichment signaling pathway. (C) Lgmn enrichment signaling pathway. (D) Results of Hallmark gene set functional enrichment analysis in the NAFLD group and the control group. The left side represents immune cells, and the right side is a box plot

Differential immune cells (DICs) correlate with biomarkers

Comparative immune profiling identified four differentially infiltrated cell populations in NAFLD versus controls: CD8 T lymphocytes, mononuclear phagocytes, NK cells, and pan-T cells (Fig. 5A). All differentially infiltrated immune cells (DICs) excluding NK cells demonstrated significantly elevated levels in NAFLD patients (p < 0.05). Egfr had a strong negative correlation with CD8T cells (cor = -0.475, p < 0.05), and Hmox1 (cor = 0.630) and Lgmn (cor = 0.632) had significant positive correlations with monocytes/macrophages and T cells (p < 0.05) (Fig. 5B) (Supplementary Table 7).

Fig. 5.

Immune cells and key genes correlation diagram. (A) 12 types of immune infiltrating cell abundance based on box plots between high- and low-risk groups. The X-axis represents immune cells, and the Y-axis represents immune scores. (B) Heatmap of Correlation Between Immune Cells and Key Genes.The X-axis represents different cell types, Y-axis represents different genes or features; the color bar on the right indicates the range of correlation values, with the red area representing high positive correlation and the blue area representing high negative correlation

Biomarkers have multiple regulatory relationships

Figure 6A shows 20 functionally related genes associated with biomarkers, such as Btc, Cbl, and Emfi1, that together participate in signalling pathways such as the ERBB signaling pathway (Fig. 6A). Subsequent prediction of the upstream and downstream genes of the biomarkers showed that Egfr was located in the cell membrane and regulates various genes, such as JAK2; Hmox1 and Lgmn were located in the cytoplasm and regulate genes such as BCL2 and TRAF6, respectively, and Hmox1 was also associated with apoptosis and proliferation (Fig. 6B). TFs were also critical for the regulation of biomarkers, and a total of 23 TFs were predicted for Egfr and Hmox1 in the TRRUST database, where the NF-κB1 regulated both Egfr and Hmox1(Fig. 6C). Unfortunately, no corresponding TFs were predicted for Lgmn. Finally, Egfr predicted 189 potential therapeutic agents, 55 of which were already approved for marketing (e.g., GEFITINIB, PEMBROLIZUMAB, CARBOPLATIN, ERLOTINIB, and CETUXIMAB). 6 of the 9 agents predicted by Hmox1 were approved (ASPIRIN, ZINC CHLORIDE, VITAMIN D, SORAFENIB, SELENIUM, and SUNITINIB). Both together predicted SORAFENIB (Fig. 6D, Supplementary Table 8). This work proposes innovative NAFLD treatment paradigms.

Fig. 6.

Key gene network diagram. (A) GeneMANIA database analysis network diagram.The large circle in the middle represents key genes, and the small circles on the outside represent genes related to key genes. (B) Gene intersection network in the DisNor database. Dots: represent proteins or genes; Solid arrows: indicate activation; Dashed arrows: indicate inhibition. (C) Biomarker-TF regulatory network diagram. Red inverted triangles indicate biomarkers; orange circles indicate TF. (D) Key gene-drug relationships. Pink indicates key genes, orange indicates drugs

Differential expression of biomarkers in clinical samples

Quantitative analysis revealed reciprocal expression patterns: Egfr suppression versus Hmox1/Lgmn induction in NAFLD (p < 0.05). (Fig. 7A-C). This was consistent with the expression in the dataset.

Fig. 7.

RT-qPCR validation of Egfr, Hmox1, and Lgmn expression in clinical samples. (A) Egfr expression levels were significantly decreased in NAFLD liver tissue samples compared with control samples. (B) Hmox1 expression levels were significantly elevated in the NAFLD group relative to controls. (C) Lgmn expression levels were significantly elevated in the NAFLD group relative to controls. Data are presented as mean ± standard deviation. Statistical significance was determined by unpaired Student’s t-test. *p < 0.05. NAFLD, non-alcoholic fatty liver disease; RT-qPCR, reverse transcription-quantitative polymerase chain reaction

Discussion

NAFLD has become the most common chronic liver disorder worldwide, yet its underlying mechanisms remain incompletely understood [12]. There remains a scarcity of rigorous studies investigating the disease mechanisms, associated biomarkers, and improved therapeutic interventions. Metabolic cell death represents a unique category of regulated cell death caused by imbalances in cellular metabolism, encompasses various types such as ferroptosis, cuproptosis, disulfide line collapse, osteolytic necrosis, and alkaline collapse [8]. It critically influences the development and progression of NAFLD.

In this study, three candidate biomarkers—Egfr, Hmox1, and Lgmn—were identified through a multi-step bioinformatics screening process based on their transcriptomic co-expression with MCD-related genes in NAFLD datasets. Notably, current evidence is based on expression and enrichment analyses and does not establish causality for these biomarkers in MCD. The column-line diagrams constructed based on these biomarkers demonstrated excellent predictive value. Furthermore, these biomarkers were predominantly associated with inflammatory and apoptotic pathways based on enrichment analysis. These associations collectively predicted the transcription factor NF-κB1 and the drug sorafenib. In addition, significant differences in immunological infiltration were observed between the NAFLD group and the control group. The majority of differentially infiltrated immune cells demonstrated increased abundance in NAFLD patients. It is anticipated that these findings will provide new predictors of NAFLD progression and potentially important targets for prevention and treatment in clinical practice. Specifically, the biological significance and pathological relevance of each identified biomarker in NAFLD are discussed below.

EGFR is a transmembrane glycoprotein receptor that plays an essential role in hepatocyte proliferation and liver regeneration [31]. EGFR serves as the primary receptor for epidermal growth factor ligands. This cell surface protein binds EGF, triggering receptor dimerization and tyrosine autophosphorylation to drive cellular proliferation [32]. EGFR is one of the most highly expressed growth factors in adult liver hepatocytes and plays a crucial role in hepatocyte proliferation and liver regeneration [33].

In the context of NAFLD, downregulation of Egfr expression may represent a key pathological event. Studies have shown that activation of Egfr induces the expression of the hepatic stellate cell (HSC) activation marker α-SMA, promotes the synthesis and deposition of collagen (such as COL-4) and pro-fibrotic factors (such as TGF-β), thereby exacerbating liver fibrosis. In vitro, oxidized low-density lipoprotein (ox-LDL) can induce HSC activation, proliferation, and the expression of fibrosis-related molecules by activating Egfr. Inhibition or silencing of Egfr significantly reduces these fibrosis-related changes [34]. Meanwhile, attenuation of NAFLD pathogenesis involves essential EGFR signaling cascades. In vivo experimental findings demonstrate that stimulating the EGFR-ERK1/2 signaling pathway can enhance LDLR expression. Moreover, in ApoE knockout mice induced by a high-fat diet, this pathway effectively alleviates fatty liver and improves related blood parameters. This protective effect can be achieved through drug interventions such as caffeine [35]. Reduced Egfr levels in NAFLD mouse models relative to controls corroborate bioinformatics-derived evidence. Based on the above evidence, it is suggested that downregulation of Egfr expression in NAFLD may be an important mechanism underlying disease progression, and reactivation of the EGFR signaling pathway through pharmacological interventions may potentially reverse the pathological process.

Hmox1 (Heme Oxygenase 1, HO-1) is an essential enzyme in the catabolism of heme and is protective against programmed cell death and prevents apoptosis. HO-1 is the rate-limiting enzyme for heme degradation and plays an important role in antioxidant defense and cytoprotection in response to cellular stress [36]. Experimental evidence confirms Hmox1-mediated attenuation of liver lipid accumulation and fibrosis by natural dietary supplements in NAFLD models [37, 38]. Enhanced Hmox1 expression with statistical significance was observed in NAFLD groups versus controls via qPCR verification. This Hmox1 upregulation showed a positive correlational pattern consistent with a compensatory response activating hepatic antioxidant defenses against lipid accumulation and oxidative injury, aligning with its documented protective role.

The gene for Lgmn (Legumain, Cysteine Protease 1) encodes a cysteine protease that is strictly specific for hydrolysis of asparagine bonds [39]. Mediates iron death and is aberrantly expressed in obese liver fat [40]. Research is now increasingly focused on its prognostic value in cancer. Although there are currently no studies directly demonstrating the role of Lgmn in NAFLD, we can examine the functions of this gene in related diseases or the liver. For example, adipose tissue macrophages (ATM) critically drive obesity-associated metabolic inflammation, with Legumain predominantly localized to these cells in obese murine models. By depleting Lgmn specifically in macrophages, mice acquire protection against diet-induced obesity while exhibiting attenuated adipose inflammation and improved hepatic metabolic parameters including steatosis and insulin sensitivity [41]. Additionally, The macrophage Lgmn axis critically regulates metabolic inflammation, lipid homeostasis, and energy expenditure dynamics. Lgmn inhibition leads to PKA activation, which subsequently triggers downstream pathway activation. The study identified the important role of Lgmn secreted by macrophages in fat tissue metabolism [29]. In the present research, Lgmn was found to be notably upregulated in the NAFLD group, suggesting that Lgmn may correlate with pro-inflammatory activation of liver macrophages in NAFLAD, exacerbating liver inflammatory responses. However, this hypothesis requires further experimental validation.

For diagnostic models constructed for NAFLD studies, it is commonly used to construct nomograms with copper death [42] and metabolism-related genes [43], However, these do not intuitively reflect the validity of the predictive model of the nomograms. This study found in the decision curves that the combining biomarkers yields the greatest clinical benefit, and the AUC value of the ROC curve was 0.997, which is close to 1, quantifying the diagnostic predictive performance of the model. These findings collectively highlight the clinical value and diagnostic accuracy of our combined biomarker approach for NAFLD detection.

All three biomarkers—Egfr, Hmox1, and Lgmn—demonstrated co-enrichment in caspase-dependent apoptotic pathways, which are well-established drivers of NAFLD/NASH progression. Hepatocyte apoptosis mediated by caspases, Bcl-2 family proteins, and JNK signaling contributes to NAFLD/NASH pathogenesis. Apoptotic hepatocytes stimulate immune cells and hepatic stellate cells through the production of inflammasomes and cytokines, promoting the progression of liver fibrosis [43, 44]. Additionally, Feldstein et al. found that the strong expression of activated caspase 3 and 7 and Fas receptors in NASH samples is closely associated with hepatocyte apoptosis and NASH progression. Activated caspase-3 and consequent hepatocyte apoptosis represent hallmark pathological events across both preclinical NAFLD models and human patients, correlating with progressive disease stages [45]. As previously mentioned, Hmox1 can prevent apoptosis. Studies have shown that Egfr inhibits tumor cell proliferation, apoptosis, and invasion [46], and Lgmn is associated with apoptosis [47, 48]. Collectively, these findings suggest that Egfr, Hmox1, and Lgmn may converge on apoptotic regulation as a shared mechanism underlying their association with NAFLD pathogenesis.

Beyond apoptosis, the NF-κB signaling pathway emerged as another shared regulatory axis among the three biomarkers. NF-κB (Nuclear Factor Kappa B) is a transcription factor primarily involved in immune and inflammatory responses and is associated with the activation of tumor necrosis factor (TNF) [49], which in NAFLD is a major driver in inflammatory response and disease [50]. In addition, the biomarkers Egfr and Hmox1 both showed correlation with the transcription factor NF-κB1, reaffirming the key role of biomarkers in the NF-κB signaling pathway.

Monocyte infiltration marks early NAFLD pathogenesis, with subsequent in situ differentiation into macrophages driving pro-inflammatory cascades [51]. Quantitative analysis demonstrated elevated CD8 + T cell, monocyte/macrophage, and total T cell infiltration in NAFLD specimens relative to controls, indicative of enhanced immune activation. Studies have confirmed the crucial role of macrophages in maintaining heme and iron homeostasis, as evidenced by mouse gene knockout models and analyses of the consequences of human HMOX1 gene mutations. Hmox1-knockout mice develop systemic iron accumulation, persistent inflammatory states, and elevated oxidative stress indicators [52]. The study also showed that argeted suppression of CD8 + tumor-infiltrating lymphocyte apoptosis enhances immunotherapeutic efficacy in EGFR-driven non-small cell lung cancer [53]. Hmox1-knockout mice exhibit systemic iron accumulation, persistent inflammatory responses, and elevated oxidative stress biomarkers [52]. On the other hand, patients with NAFLD who have higher stages and progression of liver fibrosis show an increased frequency of CD25 + CD45+CD4 + T cells, suggesting immune activation [54]. It suggests that modulation of immune cells in NAFLD through biomarkers may also have a positive effect on NAFLD treatment.

The present study proposes a framework linking MCD-related genes with NAFLD transcriptomic data that may inform biomarker discovery. Machine learning with multi-step validation supported robust feature selection, and the nomogram showed high diagnostic performance (AUC = 0.997), suggesting potential clinical utility. The co-prediction of sorafenib by Egfr and Hmox1 may indicate a possible drug repurposing opportunity. Overall, these findings may provide candidate targets for future NAFLD research.

It should be emphasized that the present study establishes correlative rather than causal evidence. Several limitations should be noted: Further detailed clinical features are required to verify the performance of the prediction model, and additional NAFLD samples are necessary to clarify the accuracy of MCD-related markers. Finally, further studies are needed to investigate the mechanism between MCD and immune response.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ATM

Adipose tissue macrophages

AUC

Area under the curve

BP

Biological process

CC

Cellular component

DEGs

Differentially expressed genes

DIC

Differentially infiltrated immune cell

EGFR

Epidermal growth factor receptor

ERK1/2

Extracellular signal-regulated kinase 1/2

FC

Fold change

FDR

False discovery rate

GBM

Gradient boosting machine

GEO

Gene Expression Omnibus

GGI

Gene-gene interaction

GLM

Generalized linear model

GO

Gene Ontology

GSEA

Gene set enrichment analysis

GSVA

Gene set variation analysis

HCC

Hepatocellular carcinoma

HIF-1

Hypoxia-inducible factor 1

HO-1

Heme oxygenase 1

HSC

Hepatic stellate cell

JAK2

Janus kinase 2

JNK

c-Jun N-terminal kinase

KEGG

Kyoto Encyclopedia of Genes and Genomes

LDLR

Low-density lipoprotein receptor

MAE

Mean absolute error

MCD

Metabolic cell death

MCD-RGs

Metabolic cell death-related genes

MF

Molecular function

mMCP

Mouse microenvironment cell population counter

MRGs

Metabolic death-related genes

MSigDB

Molecular Signature Database

NAFLD

Non-alcoholic fatty liver disease

NAFL

Non-alcoholic fatty liver

NASH

Non-alcoholic steatohepatitis

NF-κB

Nuclear factor kappa B

NK

Natural killer (cells)

ox-LDL

Oxidized low-density lipoprotein

PKA

Protein kinase A

PXR

Pregnane X receptor

RF

Random forest

RIPK3

Receptor-interacting protein kinase 3

ROC

Receiver operating characteristic

ROS

Reactive oxygen species

RT-qPCR

Reverse transcription-quantitative polymerase chain reaction

RMSE

Root mean square error

ssGSEA

Single sample gene set enrichment analysis

SVM

Support vector machine

TF

Transcription factor

TGF-β

Transforming growth factor beta

TNF

Tumor necrosis factor

TRAF6

TNF receptor-associated factor 6

Author contributions

Kaiyang Li: Conceptualization, Data curation, Animal experiments, Writing-review & editing. Jie Zhou: Conceptualization, Data curation, Validation, Visualization, Writing-original draft, Writing-review & editing. Mei Yang: Conceptualization, Supervision, Writing-review & editing. Qing Zhang: Validation, Writing-review & editing Yiming Zhao: Animal experiments, Image editing.

Funding

The research reported in this project was generously supported by the National Natural Science Foundation of China under grant agreement number (82160866). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

The transcriptomic, mutation, clinical, and survival data for BLCA were retrieved from Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) include GSE39549 and GSE52333. The transcriptomic, mutation, clinical, and survival data for BLCA were retrieved from FerrDb website (http://www.zhounan.org/ferrdb).

Declarations

Ethics approval and consent to participate

This study was performed in line with the principles of the Declaration of Helsinki and approved by the Ethics Committee of the Ethics Committee of the Guizhou University of Traditional Chinese Medicin. All patients provided written informed consent at the time of clinical sample collection for experiments, ensuring that the research process complied with ethical norms and that patients’ rights and wishes were fully respected.

Consent for publication

Not adapted.

Competing interests

The authors declare no competing interests.