Identification of prognostic signatures in pheochromocytomas and paragangliomas based on mitochondrial autophagy and ferroptosis in TCGA and GEO datasets

Abstract

Pheochromocytomas (PCCs) and paragangliomas (PGLs), collectively PPGLs, are rare tumors with significant molecular heterogeneity, which complicates prognosis and treatment. This study analyzed publicly available datasets (TCGA-PPGLs, GSE19422, GSE60459) to identify differentially expressed genes (DEGs) related to mitochondrial autophagy and ferroptosis in PPGLs. We identified 6,286 DEGs, including 31 mitochondrial ferroptosis-related DEGs (MFRDEGs). A prognostic model based on four genes (AMBRA1, EIF2S1, SRC, PHGDH) demonstrated high predictive accuracy (AUC > 0.9). Functional enrichment analysis highlighted key pathways, including mitophagy and Fc epsilon receptor I (FcεRI) signaling. Protein-protein interaction (PPI) and ceRNA network analyses revealed potential regulatory mechanisms. Calibration and decision curve analyses confirmed the model’s clinical utility. These findings offer insights into PPGL molecular mechanisms, suggest prognostic biomarkers, and propose candidate therapeutic targets to improve risk stratification and personalized treatment. However, experimental validation is required to confirm their biological relevance before clinical application.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-026-04393-7.

Introduction

Pheochromocytomas (PCCs) and paragangliomas (PGLs), collectively referred to as PPGLs, are uncommon neuroendocrine neoplasms. They originate from chromaffin cells within the adrenal medulla or from extra-adrenal sympathetic and parasympathetic ganglia. PPGLs are characterized by the secretion of catecholamines, leading to symptoms such as hypertension, headache, and diaphoresis [1]. The prevalence of PPGLs is estimated at 2–8 cases per 100,000 individuals [2], but they are often underdiagnosed due to the nonspecific nature of their symptoms [3]. Approximately 25% of PPGLs are malignant, with metastatic cases showing a 5 year survival rate of approximately 50% [4, 5]. Clinically, PPGLs are associated with hypertension, cardiovascular complications, and potential metastasis, and are frequently linked to genetic mutations in key genes such as SDHB, RET, and VHL [6]. Despite advancements in diagnostic techniques, challenges remain in accurately predicting malignancy preoperatively, as there is no histological gold standard [7]. Therefore, the identification of reliable predictive markers for PPGLs, coupled with a comprehensive approach that combines genetic and histological risk stratification, is essential for differentiating high-risk patients who may require aggressive, personalized treatment strategies.

To address this critical gap, it is essential to explore the underlying molecular mechanisms that drive PPGL aggressiveness. A growing body of evidence indicates that metabolic reprogramming is a hallmark of cancer, and PPGLs, particularly those with mutations in mitochondrial-related genes like SDHx, exhibit profound metabolic shifts [8, 9]. This mitochondrial dysfunction points toward two interconnected cellular processes as potential key players in tumorigenesis: mitophagy and ferroptosis. Mitophagy is the selective degradation of mitochondria by autophagy, a critical process for maintaining mitochondrial quality control and cellular homeostasis [10]. Ferroptosis is a distinct form of regulated, iron-dependent cell death characterized by the accumulation of lipid peroxides [11]. Investigating how these fundamental cellular pathways are dysregulated in PPGLs offers a promising strategy for identifying biologically relevant molecular markers that could enhance our ability to predict tumor behavior and clinical outcomes.

Mitophagy and ferroptosis are not isolated events but are deeply intertwined. This interplay is central to determining cell fate. For instance, mitophagy can modulate ferroptosis sensitivity in several ways. On one hand, by clearing damaged mitochondria, it can limit the production of reactive oxygen species (ROS), thereby protecting cells from lipid peroxidation and ferroptosis. On the other hand, mitophagy can also promote ferroptosis by degrading ferritin (a process known as ferritinophagy), which releases free iron into the cytoplasm, a key catalyst for the Fenton reaction that drives lipid peroxidation [12]. Therefore, a disruption in the delicate balance between these two pathways could create a cellular environment that favors tumor survival and progression, making their combined investigation particularly relevant to PPGL pathogenesis.

While the roles of specific genes in PPGLs have been investigated, such as the tumor suppressor KIF1Bβ in regulating apoptosis [13] or the functions of ferroptosis-related genes like SLC7A11 and ACSL4 in other cancers [14], research in PPGLs has been limited [15]. Most studies have focused on isolated gene mutations or have been conducted on a small scale, without systematically exploring the complex interplay between entire molecular pathways. Specifically, a comprehensive analysis of how the coordinated dysregulation of mitophagy and ferroptosis contributes to PPGL progression and clinical outcomes is currently lacking. This knowledge gap prevents the translation of molecular insights into robust clinical tools for risk assessment and treatment selection.

Given the foundational roles of mitochondrial quality control and iron-dependent cell death in cancer biology, we hypothesized that the expression patterns of genes governing mitophagy and ferroptosis could serve as powerful prognostic indicators in PPGLs. A systematic investigation of these interconnected pathways is necessary to improve upon current risk stratification models. Therefore, the primary objective of this study was to identify and validate a novel molecular signature based on mitophagy- and ferroptosis-related genes. We aimed to determine if this signature could effectively predict patient prognosis and provide a more accurate, biologically informed basis for the clinical management of PPGLs.

Methods

Data source

The TCGA-PPGLs dataset was acquired from The Cancer Genome Atlas (TCGA) using the R package TCGAbiolinks (Version 2.30.0) [16]. Following the exclusion of samples lacking clinical information, a total of 183 PPGL samples and 3 control samples were included for analysis. The sequencing data were normalized to FPKM format, while the corresponding clinical data were obtained from the UCSC Xena database (https://xena.ucsc.edu/) [17], as summarized in Table S1. Additionally, we downloaded the GSE19422 and GSE60459 datasets from the GEO database using the R package GEOquery [18] (Version 2.70.0). These datasets, derived from Homo sapiens adrenal tissue, included 84 PPGL samples and 6 control samples (GSE19422), as well as 24 PPGL samples and 3 control samples (GSE60459), with detailed descriptions provided in Table 1.

Table 1.

Details of GEO microarray chip

Item	GSE19422	GSE60459
Platform	GPL6480	GPL1015 and GPL13607
Species	Homo sapiens	Homo sapiens
Tissue	adrenal tissue	adrenal tissue
Samples in PPGL group	84	24
Samples in control group	6	3

To identify mitochondrial autophagy and ferroptosis-related genes (MFRGs), we searched the GeneCards database [19] (https://www.genecards.org/) for “Mitophagy” and “Ferroptosis” genes, retaining protein-coding genes with a relevance score > 1. The intersection of 1641 mitophagy-related genes (MRGs) and 805 ferroptosis-related genes (FRGs) resulted in 192 MFRGs (Table S2). Batch effects in the GEO datasets were eliminated using the sva R package (Version 3.50.0), resulting in a merged dataset consisting of 108 PPGLs and 9 control samples. The combined data were normalized using the limma package (Version 3.58.1), and Principal Component Analysis (PCA) was used to evaluate batch effect correction.

Screening of differentially expressed genes (DEGs) related to mitophagy and ferroptosis in PPGLs

Using the sample grouping defined in the TCGA-PPGLs dataset, the samples were divided into PPGLs and control groups. Differential expression analysis was performed using the R package DESeq2 (v1.42.0) [20]. Genes with |logFC| > 1 and adj.p < 0.05 were identified as DEGs. These DEGs were subsequently intersected with the previously identified mitochondrial autophagy and ferroptosis-related genes (MFRGs). A Venn diagram was constructed to visualize the overlap, enabling the identification of mitophagy- and ferroptosis-related differentially expressed genes (MFRDEGs).

Construction of the prognostic risk model

Univariate Cox regression analysis was conducted utilizing the R package survival (Version 3.5–7.5) to identify variables significantly associated with the outcome (p-value < 0.05). These significant variables were then incorporated into a subsequent multivariate Cox regression analysis for further evaluation. The R package glmnet [21] (v4.1–8) was used to perform LASSO regression on the MFRDEGs identified in the univariate analysis, yielding model genes for constructing the prognostic risk model. The risk score was calculated using the risk coefficients from the LASSO regression. The formula for the risk score calculation is as follows:

Evaluation and validation of the prognostic risk model

Time-dependent ROC curves were generated using the R package survivalROC (v1.0.3.1) to assess the predictive performance of the risk score in relation to overall survival (OS) in the TCGA-PPGLs dataset. The area under the curve (AUC) was calculated to evaluate the accuracy of 1-year, 2-year, and 3-year survival predictions for PPGLs samples. Additionally, a nomogram was constructed using the R package rms (v6.7-1) to visualize the relationships between the risk score and clinical variables derived from the multivariate Cox regression analysis. The predictive accuracy of the model was further validated using a calibration curve, which compared the actual and predicted probabilities. To assess the clinical utility of the risk score, decision curve analysis (DCA) was performed with the R package ggDCA (v1.1). The DCA plot quantifies the net benefit of the model at various threshold probabilities, demonstrating its potential in clinical decision-making.

Enrichment analysis

Gene Ontology (GO) analysis [22] is a widely used method for large-scale functional enrichment studies, classifying genes into Biological Process (BP), Cellular Component (CC), and Molecular Function (MF) categories. Kyoto Encyclopedia of Genes and Genomes (KEGG) [23] is another well-established database that provides information on genomes, biological pathways, diseases, and drugs. In this study, GO and KEGG enrichment analyses were conducted on model genes using the R package clusterProfiler [24] (Version 4.10.0). The screening criteria for inclusion were set at adj.p < 0.05 with Benjamini-Hochberg (BH) correction applied to adjust p values, and p < 0.05.

GSEA and GSVA analyses

PPGLs samples from the TCGA- PPGLs dataset were divided into high and low-risk groups according to the median risk score. Differential expression level was analyzed using DESeq2 (Version 1.42.0), and gene expression patterns were visualized using ggplot2 (Version 3.4.4) for volcano plots and pheatmap (Version 1.0.12) for heatmaps. Gene set enrichment analysis (GSEA) was performed using the clusterProfiler package, with the MSigDB c2 gene sets as the reference. The criteria applied were adjusted p-value (adj.p) < 0.05 and false discovery rate (FDR) < 0.25.

Gene Set Variation Analysis (GSVA) [25] was employed to evaluate differences in pathway enrichment between the HighRisk and LowRisk groups. GSVA was performed using the GSVA package (Version 1.50.0) and MSigDB c2 gene sets (GMT file: Cp. V2023.2.Hs.Symbols), with adj.p < 0.05 and BH correction.

PPI and CeRNA network

The Protein-Protein Interaction (PPI) network, composed of interacting proteins, plays a key role in biological processes, including signaling, gene expression, metabolism, and cell cycle regulation. By analyzing these interactions, we gain insights into protein functions and their involvement in diseases. In this study, we utilized the STRING database [26] (https://cn.string-db.org/) to build a PPI network based on model genes, setting a minimum interaction score of 0.150. Additionally, the GeneMANIA database [27] (https://genemania.org/) was used to predict gene functions and identify genes with similar functions based on their interactions.

For the ceRNA (competing endogenous RNAs) network, we utilized the ENCORI [28] database (https://starbase.sysu.edu.cn/) to explore miRNA, ncRNA, mRNA, and RNA-RNA interactions. The miRDB database [29] was used to predict miRNA target genes. miRNAs with a Target Score > 85 were intersected with miRNA-mRNA interactions in ENCORI to form the mRNA-miRNA network. We also predicted lncRNAs interacting with miRNAs using ENCORI, applying clipExpNum ≥ 30 as the screening criterion to generate the miRNA-lncRNA interaction network. These networks were then integrated to form a comprehensive mRNA-miRNA-lncRNA interaction network.

Statistical analysis

All statistical analyses were conducted in R (version 4.3.0). Continuous variable comparisons followed distribution-dependent approaches: two-sample Student’s t-tests assessed normally distributed data between groups, while non-parametric alternatives were employed for non-normal distributions-Mann-Whitney U tests for pairwise comparisons and Kruskal-Wallis tests for multi-group analyses. Intermolecular associations were quantified through Spearman’s rank-order correlation. A two-tailed α-level of 0.05 defined statistical significance throughout the study.

Results

Workflow chart

The workflow chart (Fig. 1) outlines the key steps of our study, starting with the identification of MFRDEGs via differential expression analysis. A prognostic model was developed to classify TCGA-PPGL patients into high- and low-risk groups. Functional enrichment and gene interaction network analyses explored the roles of key genes in biological processes, signaling pathways, and their potential as biomarkers and therapeutic targets for PPGLs.

Fig. 1.

Flowchart for the comprehensive analysis of MFRDEGs in PPGLs

Differential expressed genes analysis

The R package sva was employed to remove batch effects from the PPGL datasets GSE19422 and GSE60459, resulting in a combined GEO dataset. To assess the efficacy of batch effect correction, we generated distribution boxplots (Figure S1A, S1B) and principal component analysis (PCA) projections (Figure S1C, S1D), contrasting expression profiles and low-dimensional feature representations pre- and post-correction. These complementary visualization strategies collectively validated the complete elimination of technical batch variability across the PPGL cohort. In TCGA-PPGLs dataset, 6,286 differentially expressed genes (DEGs) were identified, including 4,197 upregulated and 2,089 downregulated genes (Fig. 2A). After intersecting the DEGs with known MFRGs, 31 MFRDEGs were obtained (Fig. 2B, Table S2). The top 20 MFRDEGs, ranked by logFC absolute value, were further analyzed, and a heatmap was generated to visualize their expression patterns (Fig. 2C).

Fig. 2.

Differential gene expression analysis. A Volcano plot of DEGs between the PPGLs and control groups in the TCGA- PPGLs dataset. B Venn diagram showing the intersection of DEGs and MFRGs in the TCGA–PPGLs dataset. C Heatmap of the top 20 MFRDEGs based on logFC absolute value in the TCGA- PPGLs dataset. In the heat map, pink represents the control group, yellow represents the PPGLs group, red indicates high expression, and blue indicates low expression

Construction of prognostic risk models for PPGLs

To develop a prognostic risk model for PPGLs, we initially conducted a univariate Cox regression analysis on 31 MFRDEGs. Four genes-AMBRA1, EIF2S1, SRC, and PHGDH were associated with prognosis (p < 0.05, Fig. 3A). To refine the model, LASSO regression analysis was was performed, which identified the same four genes as prognostic markers (Fig. 3B and C). The risk score formula, derived from the LASSO regression analysis, was calculated as follows:

Fig. 3.

Cox regression analysis. A Forest plot of univariate Cox regression analysis showing the association of four differentially expressed genes (DEGs) related to MFRDEGs with prognosis. B-C. LASSO regression model for constructing the prognostic risk model: (B) LASSO model diagram and (C) variable trajectory plot

risk score = AMBRA1 * 0.78 + EIF2S1 * 1.31 + SRC * (−1.41) + PHGDH * 0.595.

Evaluation and validation of the prognostic risk model for PPGLs

We further developed a prognostic risk model for PPGLs using TCGA–PPGLs dataset. The time-dependent ROC curves (Fig. 4A) showed strong predictive accuracy, with AUC values exceeding 0.9 at 1, 2, and 3 years, respectively. It should be noted that the stepped appearance of the ROC curves is primarily a reflection of the low number of events (deaths) in the TCGA-PPGL cohort, a common characteristic in studies of rare tumors. Univariate and multivariate Cox regression analyses (Fig. 4B and C; Table 2) identified the risk score as a significant prognostic factor (p < 0.05). A nomogram (Fig. 4D) based on these analyses highlighted risk score as the most influential variable, outperforming sex in predicting prognosis. Calibration curves (Figure S2A-S2C) showed high concordance between predicted and actual survival probabilities, particularly at 2 years, indicating optimal model performance at this time point. Finally, decision curve analysis (DCA) (Figure S2D- S2F) confirmed the model’s clinical utility, with the LASSO regression model providing the greatest net benefit at 3 years, followed by 2 and 1 years.

Fig. 4.

Prognostic analysis of PPGLs. A Time-dependent ROC curves for PPGLs samples from TCGA- PPGLs dataset, showing model accuracy at 1, 2, and 3 years. B, C Forest plots of univariate (B) and multivariate (C) Cox regression analyses for risk score and clinical variables. D Nomogram based on risk score and clinical factors from Cox regression models

Table 2.

Univariate and multivariate Cox analyses

Characteristics	Total (N)	Univariate analysis		Multivariate analysis
		HR (95% CI)	P value	HR (95% CI)	P value
Age	183	1.025 (0.981–1.071)	0.273
Gender	183
Female	102	Reference		Reference
Male	81	4.493 (0.879–22.957)	0.071	2.795 (0.419–18.650)	0.289
Risk. score	183	3.770 (2.212–6.427)	< 0.001	3.323 (1.943–5.684)	< 0.001

General HR > 1 indicates that the variable is a risk factor, and HR < 1 is a protective factor. Univariate p values < 0.1 were included in the analysis

HR: Hazard ratio

Differential expression and ROC curve analysis

The TCGA-PPGLs dataset was divided into high- and low-risk groups based on the median risk score derived from the prognostic model. Figure 7A illustrates the differential expression of the four model genes between these groups (p < 0.05; Fig. 5A). ROC curves were constructed using the R package pROC (Fig. 5B and E). The ROC analysis indicated moderate classification accuracy for EIF2S1, SRC, and PHGDH (0.7 < AUC < 0.9), whereas AMBRA1 demonstrated lower accuracy (0.5 < AUC < 0.7) in distinguishing between the groups. These results were consistent in the combined GEO dataset, where PHGDH, EIF2S1, and SRC showed significant expression differences between the high- and low-risk groups (p < 0.05; Figure S3A). ROC analysis of the GEO dataset further confirmed moderate classification accuracy for EIF2S1 and SRC (0.7 ≤ AUC < 0.9), while PHGDH exhibited lower accuracy (0.5 < AUC < 0.7) in differentiating risk groups (Figure S3B-S3D). Collectively, these findings suggest that the model genes can effectively differentiate between high- and low-risk groups across both datasets, although the classification accuracy varied.

Fig. 7.

GSEA for risk group. A Volcano plot of differentially expressed genes between the high- and low-risk groups in PPGLs samples from the TCGA-PPGL dataset. B Heatmap of the Top 10 upregulated and Top 10 downregulated differentially expressed genes, ranked by logFC absolute value, between the High Risk and Low Risk groups. C Biological function mountain map of Gene Set Enrichment Analysis (GSEA) in the TCGA- PPGLs dataset. D-G GSEA results showing significant enrichment of genes in the following pathways: Fc Epsilon Receptor (FCER1) signaling (D), FcεRI-mediated Ca²⁺ mobilization (E), Role of LAT2 in calcium mobilization (F), and FCGR3A-mediated IL-10 synthesis (G). In the heat map, purple represents the high-risk group, blue represents the low-risk group, red indicates high expression, and blue indicates low expression

Fig. 5.

Differential expression and ROC curve analysis. A Group comparison of model genes between high- and low-risk groups in the TCGA- PPGLs dataset . B-E ROC curves. ROC curves for AMBRA1 (B), EIF2S1 (C), PHGDH (D), and SRC (E) in the PPGLs samples. ** stands for p value < 0.01; *** represents p value < 0.001

GO and KEGG analyses

GO and KEGG enrichment analyses were performed to systematically explore the biological processes, cellular components, molecular functions, and signaling pathways linked to the four model genes within PPGLs. The results in Table S3 show that the four model genes were primarily enriched in processes such as the positive regulation of phosphatidylinositol 3-kinase activity, lipid kinase activity, and dephosphorylation. They were also enriched in cellular components including dendrite terminus, focal adhesion, and postsynaptic specialization; and molecular functions such as BMP receptor binding and phospholipase activator activity. Additionally, the genes were significantly enriched in the mitophagy-animal pathway. These findings were visualized in bubble plots (Fig. 6A) and network diagrams (Fig. 6B and E), with larger nodes representing greater gene counts in specific categories, highlighting the significant enrichment in the mitophagy-animal pathway.

Fig. 6.

GO and KEGG enrichment analysis for model genes. A Bubble plot of GO and KEGG enrichment analysis showing biological process (BP), cellular component (CC), molecular function (MF), and KEGG pathways, with bubble size reflecting gene number and color indicating p-value (red: smaller p-value, blue: larger p-value). B-E Network diagrams for GO (BP, CC, MF) and KEGG analysis of model genes, with orange nodes representing terms, green nodes representing molecules, and lines indicating relationships

GSEA and GSVA analyses

The TCGA-PPGLs cohort was stratified into high-risk and low-risk subgroups using the median risk score derived from the prognostic model. Differential expression analysis, performed with the DESeq2 R package, identified 131 differentially expressed genes (DEGs) meeting the thresholds of |logFC| > 2 and adj.p < 0.05. Among these, 109 genes were classified as upregulated, while 22 were downregulated. The results were visualized through a volcano plot (Fig. 7A), and the top 10 DEGs ranked by logFC absolute value were highlighted in a heatmap (Fig. 7B). Gene set enrichment analysis (GSEA) of genome-wide expression profiles revealed significant enrichment of biological processes, including FcεRI signaling” and “FcεRI-mediated Ca²⁺mobilization (Fig. 9C and G; detailed in Table S4).

Fig. 9.

PPI and ceRNA network analysis. A Protein-protein interaction (PPI) network of Model Genes was calculated using the STRING database. B Interaction network of functionally similar genes predicted by GeneMANIA for the four model genes, with circles representing the model genes and their functionally similar genes. The colors of the lines correspond to the types of interrelated functions. C The orange oval is mRNA, the green circular block is miRNA, and the purple oval is lncRNA

Gene Set Variation Analysis (GSVA) employing the c2.cp.v2023.2.hs.symbols.gmt gene set identified 20 significantly enriched pathways (adj.p < 0.05) between risk subgroups. A heatmap of pathways with the highest logFC absolute value is shown in Fig. 8A, while intergroup differences were validated using the Mann-Whitney U test (Fig. 8B, Table S5). Notably, pathways associated with RAS/ERK signaling, particularly those involving mutations and activation of key genes such as FGFR, EGFR, and KIT, demonstrated statistical significance (p < 0.05). These results underscore distinct molecular pathway dysregulations between high- and low-risk PPGLs subgroups.

Fig. 8.

GSVA analysis. A Heatmap of Gene Set Variation Analysis (GSVA) results between the High Risk and Low Risk groups in PPGL samples from the TCGA-PPGLs dataset. B Group comparison map of GSVA results between the high- and low-risk groups. *** indicates p < 0.001, highly statistically significant

Protein-protein interaction (PPI) and CeRNA network analysis

A protein-protein interaction (PPI) network for the four model genes (AMBRA1, EIF2S1, SRC, and PHGDH) was generated using the STRING database (Fig. 9A), suggesting their central roles within the network. Further functional interaction analysis via GeneMANIA predicted and assembled a network comprising these genes and their functionally associated partners (Fig. 9B). Co-expression relationships and shared protein domains, represented by color-coded edges, revealed interactions among the four model genes and 20 functionally analogous proteins.

To investigate regulatory mechanisms, mRNA-miRNA interaction data were extracted from the ENCORI and miRDB databases, enabling the prediction of miRNAs targeting the four model genes. The ENCORI database was further employed to identify lncRNAs associated with these miRNAs. The resulting mRNA-miRNA-lncRNA regulatory network, visualized using Cytoscape (Fig. 9C), included four core mRNAs (AMBRA1, EIF2S1, SRC, PHGDH), 35 miRNAs (e.g., hsa-miR-24-3p, hsa-miR-200b-3p, hsa-miR-9-5p, hsa-miR-200c-3p), and 11 lncRNAs. Specific interaction pairs are comprehensively listed in Table S6.

Discussion

Pheochromocytomas and paragangliomas (PPGLs) are rare neuroendocrine tumors originating from chromaffin cells, whose notable clinical and immunological heterogeneity are central to prognosis and therapy. Clinically, their excessive catecholamine secretion causes hypertension, and approximately 80% of cases are driven by germline mutations, classifying them into distinct molecular clusters [30]. The pseudohypoxia cluster (Cluster 1), associated with mutations in genes like SDHB and VHL, carries the highest metastatic risk and is often marked by adverse pathological features [31]. This clinical aggressiveness is closely linked to a unique tumor microenvironment (TME) that is typically immunologically “cold,” characterized by a low tumor mutational burden and poor T-cell infiltration [32]. Specifically, the high-risk pseudohypoxia cluster often exhibits a more immunosuppressive TME, with a predominance of M2 macrophages and sparse PD-L1 expression, which may limit immunotherapy efficacy. Conversely, other subtypes may show stronger immune cell presence [33]. Thus, the interplay between specific genetic drivers, clinical presentation, and the resulting immune landscape defines the biological behavior of PPGLs and underscores the need for stratified management strategies.

This study systematically explored the molecular characteristics of PPGLs using publicly available datasets (TCGA and GEO) to identify differentially expressed genes (DEGs) associated with mitochondrial autophagy and ferroptosis. Furthermore, we screened four candidate genes (AMBRA1, EIF2S1, SRC, and PHGDH) and developed a prognostic risk model that demonstrated high accuracy in stratifying patients into high-risk and low-risk groups in both the TCGA and GEO datasets. These findings suggest novel biomarkers for prognosis and highlight promising therapeutic avenues for personalized treatment strategies in patients with PPGLs.

Several biomarkers in our gene signature have been implicated in various cancers, but their specific roles in mitochondrial autophagy and ferroptosis in PPGLs have yet to be fully explored. For example, AMBRA1 has been identified as having comprehensive oncogenic functions that promote tumorigenesis. Its involvement in the regulation of c-MYC, a well-known oncogene, is particularly noteworthy. AMBRA1 facilitates the dephosphorylation and degradation of c-MYC, thereby influencing cell proliferation and tumor growth [34]. Studies have shown that AMBRA1 deficiency correlates with increased tumorigenesis, particularly in models of melanoma where its absence leads to accelerated tumor growth and enhanced invasion capabilities [35, 36]. EIF2S1 is often overexpressed in cancer tissues compared to matched non-cancerous tissues, suggesting its potential as a biomarker for cancer prognosis [37]. In hepatocellular carcinoma (HCC), elevated levels of EIF2S1 are associated with poor patient outcomes [38]. SRC, a cytoplasmic tyrosine kinase and prominent member of the Src family kinases (SFKs), plays a crucial role in cellular processes like proliferation, survival, and metabolism, with its overexpression and hyperactivation common in various cancers [39]. Increased Src activity is commonly observed across multiple breast cancer subtypes, particularly in triple-negative breast cancer (TNBC), where it correlates with enhanced invasive capabilities and lower cellular viability due to mitochondrial dysfunction [40]. PHGDH, an enzyme involved in serine biosynthesis, has recently been implicated in mitochondrial metabolism and ferroptosis. Increased PHGDH expression in various cancers has been associated with altered metabolic pathways that support cell proliferation and survival, particularly under oxidative stress [41]. Ferroptosis is a form of regulated cell death characterized by iron-dependent lipid peroxidation. Studies indicate that PHGDH can inhibit ferroptosis by stabilizing mRNA levels of SLC7A11, a gene that plays a role in cystine uptake and glutathione synthesis. This inhibition allows cancer cells to survive under conditions that would typically induce ferroptosis [42]. In summary, the biological functions associated with mitochondrial autophagy and ferroptosis of the novel gene signature will need to be investigated.

This study aimed to refine risk stratification for PPGLs patients by developing a risk score prognostic model that incorporates clinical and pathological characteristics, with the risk score emerging as the primary determinant. The multivariate Cox regression analysis further demonstrated that the risk score exhibited a significantly stronger prognostic value for overall survival compared to traditional factors such as gender and other clinicopathological features. These findings underscore that the four-gene (AMBRA1, EIF2S1, SRC, and PHGDH) signature serves as a powerful risk factor for predicting overall survival in PPGLs patients. To quantify individual patient risk, we developed a nomogram incorporating the risk score alongside other clinicopathological features. Calibration analysis revealed that the nomogram provided highly accurate survival predictions, closely aligning with actual survival outcomes. Moreover, time-dependent ROC analysis confirmed that the nomogram exhibited the most stable and robust survival prediction capabilities, with the nomogram exhibiting robust survival prediction capabilities across various follow-up time points, demonstrated by AUC values exceeding 0.9 at 1, 2, and 3 years.

The biological processes and pathways associated with these genes, such as phosphatidylinositol 3-kinase regulation, BMP receptor binding, and the mitophagy-animal pathway, offer insights into the underlying molecular mechanisms driving PPGLs. These pathways are critical in regulating cellular processes like apoptosis, immune response, and metabolism, all of which may contribute to tumor progression and response to therapy. The enrichment of the Mitophagy-animal pathway, in particular, points to the potential therapeutic implications of targeting mitophagy in PPGLs treatment strategies. This pathway has been studied in the context of diverse malignancies, including lung cancer and breast cancer, where mitophagy is involved in regulating tumor progression and treatment resistance [43, 44].

Moreover, the identification of differentially expressed genes (DEGs) and enriched pathways between the high-risk and low-risk groups, including those involved in immune responses and cell signaling (e.g., Fc Epsilon Receptor signaling, RAS/ERK signaling, and WNT5Adependent signaling), underscores the importance of immune modulation and cellular communication in tumor progression. These pathways are similarly implicated in immune evasion mechanisms in other tumor types, such as in neuroendocrine tumors [45], but our study extends these findings to PPGLs, providing new insights into the role of immune modulation in this specific malignancy.

We acknowledge that this study is based solely on computational analysis and that experimental validation is essential to confirm the biological relevance of our findings. Future work will focus on validating the expression of key hub genes using qRT-PCR and Western blot analyses on patient-derived samples, as well as conducting in vitro functional assays (e.g., knockdown/overexpression experiments) to explore their mechanistic roles in mitochondrial autophagy and ferroptosis. Additionally, pathway-specific assays, including ROS and iron quantification, will be employed to confirm the predicted signaling pathways. These efforts will provide a more comprehensive understanding of the molecular mechanisms and strengthen the translational significance of our findings. However, several limitations inherent to this study must be acknowledged. First, as a retrospective analysis of public datasets, our findings may be influenced by sample selection bias and limited clinical data, such as the lack of detailed tumor stage information in the TCGA-PPGL cohort. Second, and most importantly, our conclusions are currently derived solely from bioinformatics analyses and thus lack experimental confirmation. Finally, the precise biological mechanisms linking these genes to the interplay of mitophagy and ferroptosis in PPGLs remain to be fully elucidated. Therefore, future work must include the proposed experimental validation and functional studies, ultimately requiring confirmation through prospective clinical trials before this gene signature can be considered for clinical application.

Conclusions

In this study, we developed a gene signature based on mitophagy- and ferroptosis-related genes to classify patients with PPGLs into high- and low-risk groups. By integrating the risk score with clinicopathological variables, we constructed a nomogram to quantitatively estimate individual patient risk. These findings provide a promising framework for identifying high-risk PPGL patients and guiding further research into targeted adjuvant therapies. However, experimental validation and prospective studies will be essential before clinical implementation can be considered.

Author contributions

Qingke Chen : Conceptualization, Methodology, Software, Writing-original draft. Qian Zou : Conceptualization, Formal analysis, Methodology, Writing-original draft. Zhiyong Xian : Validation, Writing-review & editing, Supervision. Junming Bi : Data curation, Formal analysis, Funding acquisition, Validation.

Funding

This research was funded by the Guangdong Basic and Applied Basic Research Foundation (Grant No. 2021A1515111219) and the Science and Technology Projects in Guangzhou (Grant No. 202201010897).

Data availability

GSE19422 dataset: Available from the GEO database (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE19422). GSE60459 dataset: Available from the GEO database (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi).

Declarations

Ethics approval and consent to participate

No applicable. This study was based on publicly available datasets (TCGA and GEO), in which all patient data had been anonymized in advance; therefore, approval from an institutional review board was not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.