Skip to main content
NCBI home page
World Journal of Surgical Oncology logoLink to World Journal of Surgical Oncology
. 2026 Jan 6;24:67. doi: 10.1186/s12957-025-04169-3

Development of a prognostic model for thyroid cancer based on mitochondrial metabolism-related genes and immune profiling

Guorui Wang 1, Qinchao Feng 1, Guohua Zhu 1,✉
PMCID: PMC12871013  PMID: 41495747

Background

The preliminary study found that mitochondrial metabolism and structure are abnormal in thyroid cancer (THCA) patients. Therefore, this study systematically investigates the relationship between mitochondrial metabolism-related genes (MMRGs) and the prognosis of THCA patients, while establishing a prognostic model for THCA.

Methods

This study utilized THCA transcriptome data from the UCSC Xena database, performed differential expression analysis using the “limma” package, and intersected differentially expressed genes (DEGs) with MMRG to identify differentially expressed MMRGs (DEMMRGs). THCA prognostic genes were identified using the least absolute shrinkage and selection operator (LASSO) regression and multivariate Cox regression analysis, and a prognostic model was constructed. Using ssGSEA, CIBERSORT and Immune Phenotype Score methods, we compared differences in immune cell infiltration levels and anti-tumor immune response capacity between distinct risk groups. Furthermore, molecular subtypes of THCA were identified through consensus clustering analysis.

Results

This study systematically identified nine MMRGs to construct a robust prognostic prediction model for THCA. Enrichment analysis revealed that patients in the low-risk group exhibited significant enrichment in multiple immune-related pathways, such as T cell-mediated immune responses to tumor cells, and demonstrated stronger responsiveness to anti-CTLA-4 and anti-PD-1 immunotherapies compared to the high-risk group. Further analysis identified two distinct molecular subtypes of THCA: Group 2 exhibited upregulation of immune checkpoint molecules, elevated ESTIMATEScore and StromalScore, and lower TumorPurity.

Conclusion

This study adopts the unique perspective of MMRGs to elucidate their pivotal role and molecular basis within the THCA tumor microenvironment, offering novel insights for deepening our understanding of the disease’s pathogenesis and developing innovative therapeutic strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12957-025-04169-3.

Keywords: Thyroid cancer, Mitochondrial metabolism-related genes, Molecular subtypes, Prognostic model

Highlights

Based on prognostic genes, we successfully classified thyroid cancer into two molecular subtypes with distinct immune heterogeneity.

In this study, nine mitochondrial metabolism-related genes associated with thyroid cancer were systematically identified and utilized to construct prognostic models.

Patients with low-risk thyroid cancer exhibit higher survival rates, and their tumor microenvironments also demonstrate a more active state of immune cell infiltration.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12957-025-04169-3.

Introduction

Thyroid cancer (THCA) ranks among the most common malignancies globally, and its mortality rate has been steadily increasing in recent years [1]. The majority of THCA cases arise from epithelial cells and are categorized into several subtypes, including papillary thyroid carcinoma (PTC), follicular thyroid carcinoma (FTC), anaplastic thyroid carcinoma (ATC), and a smaller proportion of medullary thyroid carcinomas (MTC) [2, 3]. According to the International Agency for Research on Cancer (IARC), PTC is the most common form of THCA, with a significant gender discrepancy, as it is more frequently diagnosed in males than females [4]. Despite advances in medical interventions such as surgical resection, chemotherapy, and targeted therapies, issues such as limited clinical efficacy, tumor latency, and the tumor microenvironment continue to challenge treatment outcomes [5–7]. Furthermore, excessive treatment with THCA may harm patients, including psychological stress from cancer diagnoses, unnecessary surgeries that may trigger complications, and a decline in quality of life following surgery. Therefore, a deeper understanding of THCA’s molecular mechanisms is needed, as well as the identification of relevant molecular subtypes and biomarkers, is essential for advancing precision medicine in the treatment of this cancer.

As the central hub of cellular energy metabolism, mitochondrial dysfunction is closely linked to tumor initiation and progression [8, 9]. Cancer cell proliferation and resistance are influenced by mitochondrial properties, with mitochondrial energy metabolism closely associated with the reprogrammed energy production pathways characteristic of tumor cells [10]. Throughout tumor development, mitochondria release energy through their own metabolic processes and glucose degradation, providing sustained power for cytoskeletal formation, organelle construction, and membrane maturation in cancer cells [11, 12]. Research indicates that mitochondrial DNA mutations and metabolic pathway alterations may influence tumor cell proliferation and apoptosis, thereby affecting treatment efficacy [13, 14]. Abnormal alterations in mitochondrial metabolic pathways not only lead to energy metabolism remodeling in tumor cells, enhancing their proliferation and invasive capabilities, but also profoundly influence the tumor immune microenvironment [15]. For example, mitochondrial metabolic disorders can affect CD8 T cell migration and thus indirectly regulate the proliferation of tumor cells [16]. Knockdown of the mitochondrial-associated pathway gene NDUFB10 not only directly inhibits tumor growth but also significantly enhances the efficacy of anti-PD-1 therapy. These findings suggest that targeting mitochondrial metabolism-related genes (MMRGs) and metabolic pathways may offer novel therapeutic strategies for treating oncological diseases [17].

Growing evidence indicates that mitochondrial dysfunction is a key driver of tumor progression and metastasis, indicating that targeting mitochondrial-related genes and metabolic pathways could offer novel therapeutic strategies for neoplastic disease treatment. In recent years, an increasing body of research has demonstrated a close association between the initiation and progression of THCA and mitochondrial metabolic abnormalities. Within the hypoxic THCA tumor microenvironment, mitochondria generate elevated levels of reactive oxygen species (ROS), which in turn cause oxidative damage to biomolecules, inducing genomic instability and metabolic reprogramming [1]. In terms of treatment, targeting mitochondrial metabolism has demonstrated promising potential. For instance, the antimalarial drug artemether effectively inhibits the proliferation of chemotherapy-resistant ATC cells by targeting mitochondrial metabolic processes [18]. Tigecycline exhibits high selectivity towards THCA cells by inhibiting the mitochondrial respiratory chain, suppressing mitochondrial respiration and ATP synthesis more significantly than in normal thyroid cells [19]. In summary, advancing research into THCA and mitochondrial metabolism holds significant scientific and clinical value.

Bioinformatics, as a key discipline integrating multi-omics data with computational analysis, plays an increasingly vital role in deciphering the molecular mechanisms of cancer metabolic reprogramming [20]. With the advancement of this technology, efficient processing of large-scale genomic data has become feasible. Against this backdrop, this study conducted systematic bioinformatics analysis based on TCGA data, identifying a set of differentially expressed genes (DEGs) closely associated with mitochondrial function. Leveraging these findings, we developed and validated a prognostic model for THCA, offering novel insights for the development of more targeted and effective treatment strategies in the future.

Materials and methods

Data sources and preprocessing

This study obtained transcriptome data for THCA patients (TCGA-THCA) from the UCSC Xena database (https://xena.ucsc.edu/), which included 59 normal samples and 513 tumor samples. Corresponding patient clinical information and copy number variation data were acquired from the TCGA database (https://portal.gdc.cancer.gov/). Patient samples were randomly divided into training and validation sets at a 7:3 ratio. Additionally, 1,234 MMRGs were collected from the MSigDB database (https://www.gsea-msigdb.org/gsea/msigdb).

Preliminary screening of THCA prognostic genes

To screen for prognostic genes in THCA, this study employed the “limma” package (version 3.62.2) to perform differential analysis between the normal and tumor groups. Differentially expressed genes (DEGs) were then intersected with MMRGs to identify differentially expressed MMRGs (DEMMRGs). Additionally, candidate prognostic genes were obtained through univariate Cox regression analysis (survival package, version 3.5-8).

Prognostic feature selection and model construction

To prevent overfitting, least absolute shrinkage and selection operator (LASSO) regression analysis was performed on the candidate prognostic genes using the “glmnet” package (version 4.1-9). The optimal penalty parameter, lambda, was selected through cross-validation to eliminate highly correlated differential genes and reduce model complexity. Subsequently, multivariate regression analysis was performed on the LASSO-selected genes using the “survival” package (version 3.5-8).

Patients were stratified into high- and low-risk groups based on the median risk score derived from the expression levels of prognostic genes and their corresponding regression coefficients. Kaplan-Meier (K-M) survival analysis was conducted to evaluate prognostic differences between the groups. Time-dependent ROC curves generated using the “timeROC” package (version 0.4) were used to assess the model’s predictive accuracy at 1-, 3-, and 5-year intervals by calculating AUC values. Additionally, risk score distribution and survival status plots were created to visually illustrate differences between the two risk groups.

Gene enrichment analysis

Gene Set Enrichment Analysis (GSEA, version 4.3.2) was performed to assess pathway enrichment differences between the two groups. DEGs were subjected to Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses using the “clusterProfiler” package (version 4.14.6).

Construction and evaluation of nomogram

Univariate and multivariate Cox regression analyses were conducted to evaluate the clinical relevance of the model, and results were visualized with forest plots. A nomogram predicting the risk of developing THCA was constructed based on clinical characteristics and RiskScore. The model’s clinical utility for predicting 1-year, 3-year, and 5-year outcomes was evaluated using calibration curves and Decision Curve Analysis (DCA).

Immune infiltration analysis and prediction of immune therapy response

To characterize the tumor immune microenvironment, we employed ssGSEA (GSVA package, version 2.0.7), ESTIMATE (estimate, version 1.0.13), and CIBERSORT algorithms to assess immune infiltration and scores between risk groups. Further analysis examined the expression levels of immune checkpoint molecules and the Pearson correlation coefficients between immune cell infiltration levels and the expression of prognostic genes. Furthermore, the model was validated using the IMvigor210 cohort to evaluate its predictive value in anti-PD-L1 immunotherapy outcomes.

Tumor mutational burden (TMB) profiling and drug sensitivity prediction

Waterfall plots generated with the “maftools” package (version 2.22.0) were used to visualize the top 20 mutated genes in both the high- and low-risk groups. To explore potential therapeutic targets, drug sensitivity analyses were conducted. The CellMiner database (https://discover.nci.nih.gov/cellminer/home.do) was used to identify correlations between drug response and prognostic genes, while the “pRRophetic” package (version 0.5) predicted the half maximal inhibitory concentration (IC50) values for various chemotherapeutic agents, with lower IC50 values indicating higher sensitivity and therapeutic potential.

Construction of competitive endogenous RNA (ceRNA) regulatory networks

Upstream transcription factors (TFs) regulating the feature genes were predicted using NetworkAnalyst (https://www.networkanalyst.ca), and miRNA–lncRNA interactions were retrieved from MiRTarBase http://mirtarbase.mbc.nctu.edu.tw/php/index.php)(https://mirtarbase.cuhk.edu.cn/). Based on these datasets, a comprehensive ceRNA regulatory network—encompassing lncRNA–miRNA–mRNA interactions—was constructed and visualized using Cytoscape software (version 3.9.1).

Identification of THCA molecular subtypes based on MMRGs

This study employed consensus clustering analysis (ConsensusClusterPlus package, version 1.70.0) on tumor samples from the training set based on prognostic genes to delineate distinct molecular subtypes of THCA. Furthermore, the reliability and validity of the clustering results were assessed using cumulative distribution function (CDF) plots, delta curves, and principal component analysis (PCA).

Statistical analysis

Data analysis and visualization were performed using R (version 4.4.0). Statistical differences were evaluated with the Wilcoxon rank-sum test, and visualizations were primarily created using the “ggplot2” package (version 3.5.2). Correlations between variables were assessed using Pearson methods. Significance was defined at P < 0.05, with the following notation: **** for P < 0.0001, *** for 0.0001 < P < 0.001, ** for 0.001 < P < 0.01, * for 0.01 < P < 0.05, and ns for P > 0.05.

Results

Preliminary screening of THCA prognostic genes

This study began with a differential expression analysis both tumor and normal groups, identifying 2,013 DEGs in total, including 1,070 upregulated and 943 downregulated genes (Fig. 1A). Following this, an intersection analysis was conducted both the DEGs and MMRGs, screen out the identification of 116 DEMMRGs (Fig. 1B). GO enrichment analysis indicated that the DEMMRGs were significantly associated with processes comprise fatty acid metabolism, organic acid biosynthesis, and carboxylic acid biosynthesis (Fig. 1C). KEGG enrichment analysis indicated that DEMMRGs were prominently enriched in the PPAR signaling pathway, glycerophospholipid metabolism, and biosynthesis of cofactors (Fig. 1D).

Fig. 1.

Fig. 1

Open in a new tab

Preliminary Screening of THCA Prognostic Genes. A Volcano plot of DEGs. B Intersection analysis of both DEGs and MMRGs. C GO and (D) KEGG enrichment. E Univariate forest plot of DEMMRGs. F Expression boxplot of candidate prognostic genes. G Correlation heatmap of candidate prognostic genes. H Chromosomal location circos plot of genes

Subsequently, 19 candidate prognostic genes for THCA were identified through univariate regression analysis (Fig. 1E, Supplementary Table S1). Box plots were generated to illustrate the markedly expressed levels of the 19 genes in two groups including tumor and normal control groups (Fig. 1F). Correlation analysis illustrated that strong correlations among the 19 genes (Fig. 1G). Lastly, the study explored the chromosomal distribution of these genes and found that the distribution was not uniform (Fig. 1H).

Construction and validation of a THCA prognosis model

To further screen for prognostic-related genes in THCA, this study conducted the following analyses based on 19 candidate prognostic genes. First, LASSO regression analysis preliminarily screened 12 genes (Fig. 2A-B), followed by multivariate Cox regression analysis to ultimately identify nine THCA prognostic genes (Fig. 2C). Based on this gene set, we constructed a prognostic risk assessment model. ROC curve analysis demonstrated that this model exhibited good predictive performance in both the validation set (Fig. 2D) and the training set (Fig. 2E-F). K-M survival analysis results further indicated that patients in the high-risk group had significantly lower overall survival rates in the training set (Fig. 2G), and this finding was also observed in the validation set (Fig. 2H-I). To visually illustrate differences between risk groups, we plotted risk score versus survival status distributions (Fig. 2J-L). Additionally, these nine prognostic genes exhibited significant differences between the tumor group and the normal group (Fig. 2M). Gene expression analysis revealed (Supplementary Figure S1A) that compared to the low-risk group, the high-risk group exhibited significantly upregulated expression levels of AOX1, AKR1B15, PCK1, PPM1L, PTGIS, and TNFAIP8L3, while CDC45, PLAAT5, and THRSP showed a significant downregulation trend. Further K-M survival analysis (Supplementary Figure S1B) demonstrated that all nine prognostic genes independently predicted survival outcomes.

Fig. 2.

Fig. 2

Open in a new tab

Construction and Validation of a THCA Prognosis Model. A-B LASSO regression coefficient distribution and spectrum. C Multivariate regression forest plot. D-F ROC curves for training, validation, and overall cohorts. G-I K-M curves for training, validation, and overall cohorts. J-L Survival status and risk score distributions. M Boxplot of feature genes expression in tumor and control groups

Gene enrichment analysis

This study employed gene enrichment analysis to systematically elucidate the differential molecular characteristics between high- and low-risk groups. GSEA enrichment analysis indicated that the high-risk group exhibited distinctly significant enrichment, including in the cellular response to glucocorticoid stimulus, insulin-like growth factor receptor signaling, regulation of cell migration involved in sprouting angiogenesis, and response to epidermal growth factor (Fig. 3A). In contrast, the low-risk group showed significant enrichment in pathways related to dendritic cell differentiation, interferon-mediated signaling, positive regulation of T cell apoptosis, and T cell-mediated immune responses to tumor cells. This finding indicated the potential importance of these genes in cell differentiation and T cell-related immune functions (Fig. 3B). Next, gene enrichment analysis was performed on the DEGs across different risk groups. GO enrichment results indicated that these genes were enriched in processes such as epidermis development, hormone metabolic processes, inorganic anion transport, and monoatomic anion transport (Fig. 3C). KEGG enrichment analysis showed that these genes were predominantly involved in like neuroactive ligand-receptor interactions, cytokine-cytokine receptor interactions, neuroactive ligand signaling, and cornified envelope formation (Fig. 3D).

Fig. 3.

Fig. 3

Open in a new tab

Gene Enrichment Analysis. A-B GSEA results of two groups. C GO and (D) KEGG enrichment

Construction of nomogram

To evaluate the prognostic significance of the model, univariate and multivariate Cox regression analyses were performed by integrating risk scores with clinical parameters. These results showed that the risk score served as an independent and vital predictor of patient survival outcomes (Fig. 4A-B). Then we structured a nomogram incorporating important factors such as risk score, age, disease stage, and TNM classification, to forecast survival probabilities in 1, 3, and 5 years (Fig. 4C). The DCA for these time points demonstrated the nomogram’s strong predictive power (Fig. 4D). The calibration curves at each time point demonstrated a strong similarity both the predicted and actual survival outcomes (Fig. 4E).

Fig. 4.

Fig. 4

Open in a new tab

Construction of Nomogram. A univariate and (B) multivariate regression analysis. C Line graph of prognostic model. D DCA. E Risk prediction calibration curve

Immune infiltration analysis and prediction of immune therapy response

To investigate the immune characteristics of the high-risk and low-risk groups, the following analyses were conducted. Immune infiltration analysis based on the ssGSEA algorithm revealed significant differences in neutrophil and Treg expression between two groups (Fig. 5A). With the exception of CCR and T helper cells, all other immune-related genes were differentially expressed (Fig. 5B). Immune infiltration analysis based on the CIBERSORT algorithm revealed that the infiltration level of B cells naive was significantly higher in the low-risk group than in the high-risk group (Fig. 5C). The immune checkpoint analysis indicated the expression levels of most immune checkpoints were significantly higher in the low-risk group than in the high-risk group (Fig. 5D). Finally, correlation analysis indicated the expression level of PPM1L is positively correlated with the degree of monocyte infiltration (Fig. 5E).

Fig. 5.

Fig. 5

Open in a new tab

Immune Characteristics Analysis. A Boxplot of ssGSEA-derived immune cell scores. B Boxplot illustrating immune function activity. C CIBERSORT-based immune cell proportion boxplot. D Boxplot of immune checkpoint gene expression across groups. E Heatmap depicting correlations both model genes and immune cell infiltration levels

These results revealed that the high-risk group had higher Stromal Scores, while the low-risk group showed elevated Immune Scores (Fig. 6A). Immune Phenotype Score (IPS) analysis revealed that patients in the low-risk group demonstrated significantly higher treatment response rates to anti-CTLA-4 and anti-PD-1 therapies compared to those in the high-risk group (Fig. 6B). Patients in the low-risk group demonstrated significantly improved survival compared to those in the high-risk group (Fig. 6C). Then, we classified complete response (CR) and partial response (PR) as responders (R), while stable disease (SD) and progressive disease (PD) were grouped as non-responders (NR). Notably, the response rate was substantially higher in the low-risk group than in the high-risk group (Fig. 6D). Further statistical analysis of risk scores between NR and R groups showed that responders had markedly lower risk scores than non-responders (Fig. 6E).

Fig. 6.

Fig. 6

Open in a new tab

Immunotherapy Response in Two Groups. Comparison of ESTIMATE (A)and IPS scores (B) revealed significant differences between two groups. C Survival analysis of the IMvigor210. D Bar chart illustrating immune response rates. E Cloud-rain plot of risk scores in the NR and R groups

TMB and drug sensitivity analysis

The top 20 most frequently mutated genes were all identified and compared between the different groups. A waterfall plot showed that significantly different in mutation rates of TMB genes both the two groups (Fig. 7A-B). The IC50 results for different drugs revealed higher IC50 values for Dabrafenib, Dasatinib, and Trametinib in the high-risk group, while Sorafenib was more higher in the low-risk group. Drug sensitivity correlations with prognostic genes were further explored using CellMiner. The PTGIS gene was positively correlated with Lapatinib (Cor = 0.378) and negatively correlated with Crizotinib (Cor = -0.269). The TNFAIP8L3 gene revealed a positive correlation with SGX-523 (Cor = 0.417), and AOX1 was positively correlated with Dasatinib (Cor = 0.420) (Fig. 7D). All data are available in Supplementary Table S4.

Fig. 7.

Fig. 7

Open in a new tab

TMB and Drug Sensitivity Analysis. Waterfall plots of the top 20 mutated genes in both (A) high- and (B) low-risk group. C Violin plot depicting sensitivity to four different drugs. D Correlation plot illustrating the relationship between prognostic genes and drug sensitivity predictions from the CellMiner database

Construction of miRNA-mRNA/TF network

We first identified 131 miRNAs that target nine key prognostic genes and subsequently found 99 lncRNAs that regulate these miRNAs. Using Cytoscape software, we constructed a ceRNA regulatory network involving mRNA-miRNA-lncRNA interactions, and visualised the core network with an interaction score of 3.8 (Fig. 8A). Notably, the miRNA hsa-miR-5011-5p, which targets PCK1, was regulated by multiple lncRNAs. Additionally, using NetworkAnalyst, we identified 47 upstream TFs. This analysis suggested that PCK1, PTGIS, and PPM1L may regulate multiple genes simultaneously (Fig. 8B).

Fig. 8.

Fig. 8

Open in a new tab

Construction of Regulatory Networks Based on prognostic genes. A ceRNA network of prognostic genes, red circles represent mRNAs, blue triangles represent miRNAs, and green diamonds represent lncRNAs. B TF-mRNA interaction network, red circles represent mRNA, yellow circles represent TFs

Identification of THCA molecular subtypes

Based on the prognostic genes set selected through model screening, this study performed cluster analysis on the THCA training samples, successfully dividing them into two molecular subtypes with significant differences (Fig. 9A). The robustness of this classification was further validated through CDF and delta area curve analysis (Fig. 9B-C). PCA results also supported the classification of these molecular subtypes (Fig. 9D). K-M survival analysis indicates poorer survival rates in the group2 (Fig. 9E). Differential expression analysis between subtypes identified DEGs, which were further explored via GO and KEGG enrichment analysis. GO enrichment analysis indicated that these DEGs were involved in processes such as taxis, chemotaxis, external encapsulating structure organization, and extracellular matrix organization (Fig. 9F). KEGG enrichment analysis suggested these genes were primarily enriched in specific pathways close to the cytoskeleton in muscle cells and hematopoietic cell lineage (Fig. 9G). Differential expression analysis across subtypes revealed distinct expression patterns of candidate target genes (Fig. 9H). Group2 exhibited significantly higher ESTIMATE and Stromal Scores, while group1 had notably higher Tumor Purity (Fig. 9I). Immune cell infiltration analysis based on the ssGSEA algorithm revealed significantly higher infiltration levels of B cells, macrophages, and Th1 cells in group2 (Fig. 9J). The MCP-counter analysis revealed that group1 had higher levels and greater diversity of immune cell infiltration (Fig. 9K).

Fig. 9.

Fig. 9

Open in a new tab

Identification of THCA Molecular Subtypes. A-C Consensus Clustering Analysis using ConsensusClusterPlus. D PCA plot of tumor samples. E Survival curves comparing subtypes. F GO and (G) KEGG enrichment analysis. H Boxplot showing prognostic genes expression differences between subtypes. I Violin plots comparing ESTIMATE Score, ImmuneScore, StromalScore, and TumorPurity across subtypes. J Differential immune infiltration between subtypes shown in boxplots. K Proportional distribution of immune cell types across subtypes

Discussion

THCA is among the top five most common malignancies in American women, with its global incidence steadily increasing [21]. Altered mitochondrial metabolism has been demonstrated to significantly influence the tumor microenvironment, thereby promoting tumor progression and metastasis [22]. Therefore, targeting mitochondrial metabolism has become a research hotspot in current cancer treatment strategies [23]. Against this backdrop, this study integrates multiple bioinformatics approaches to investigate the potential role of MMRGs in THCA.

This study identified nine MMRGs (AOX1, TNFAIP8L3, CDC45, PPM1L, THRSP, PTGIS, PCK1, PLAAT5, and AKR1B15) closely associated with THCA prognosis and constructed a robust prognostic risk model. Among these, thyroid hormone-responsive protein (THRSP) exhibited significantly higher mRNA expression in THCA tissue compared to normal tissue [24], with its expression positively correlated with insulin sensitivity, glucose tolerance, and lipid synthesis [25]. Combined with our findings linking the high-risk group to insulin-like growth factor receptor signaling, and literature reporting that overexpression of insulin-like growth factor receptor 1 (IGF-1R) and insulin receptor (IR) in THCA cells promotes cellular transformation, proliferation, and inhibits apoptosis [26], it is suggested that THRSP may participate in THCA progression by regulating the insulin-like growth factor receptor signaling pathway. The specific mechanism of action requires further elucidation. Aldehyde oxidase 1 (AOX1) participates in the metabolism of nitrogen-containing heterocyclic compounds and lipids, and its expression is downregulated in PTC tissues [27, 28], potentially affecting metabolic homeostasis within tumors. Aldo-keto reductase family 1 member B15 (AKR1B15) is an aldehyde-ketone reductase that exhibiting potent reductase activity toward retinal and 17-ketosteroids, suggesting its potential involvement in the metabolic processes of steroids and retinal within the body [29, 30]. Phosphoenolpyruvate carboxykinase 1 (PCK1), as a key enzyme in gluconeogenesis, can prevent glucose from entering the glycolytic pathway and suppress the Warburg effect in hepatocellular carcinoma (i.e., the phenomenon where tumor cells preferentially metabolize glucose via glycolysis under aerobic conditions rather than generating energy through oxidative phosphorylation), thereby inhibiting tumor cell proliferation and promoting tumor cell death [31, 32]. TNF alpha-induced protein 8-like 3 (TNFAIP8L3) is a member of the tumor necrosis factor family, playing a crucial role in tumor immunoregulation. It participates in regulating inflammatory responses, maintaining immune homeostasis, and influencing cancer progression [33, 34]. TNFAIP8L3 has been found to inhibit tumor cell occurrence and progression by disrupting PGAM5-mediated mitochondrial dysfunction [35]. Protein phosphatase, Mg²⁺/Mn²⁺ dependent 1 L (PPM1L) exerts tumor-suppressive effects in colorectal tumor progression by negatively regulating TGF-β and BMP signaling pathways [36, 37]. Prostaglandin I2 synthase (PTGIS) exhibits reduced expression levels in colorectal cancer cells. Its overexpression inhibits apoptosis whilst promoting cancer cell proliferation, invasion, and migration capabilities [38]. Cell division cycle 45 (CDC45) is essential for the initiation of chromosomal DNA replication, and studies have indicated that metformin treatment downregulates the helicase activity of the CDC45-MCM2-7-GINS complex and DNA replication interfix genes in THCA cells, suggesting a potential role of CDC45 in tumor cell growth and development [39–41]. In summary, these key MMRGs may play a role in the occurrence and development of THCA by regulating biological processes such as metabolic reprogramming, cell migration, and proliferation. Subsequent studies should further elucidate the specific mechanisms of action of each gene in THCA, providing new clues for clinical prognosis assessment and targeted therapy.

The immune tumor microenvironment creates a favorable environment for tumor development and growth, acting as a critical site for cancer cell proliferation [42]. Immunotherapy, which leverages the human immune system to combat cancer, has emerged as a promising therapeutic strategy. Current cancer treatments primarily target this immune microenvironment to indirectly regulate tumor cell proliferation and progression [43–45]. Our research shows that the two subtypes exhibit markedly different immune responses. The high-risk subgroup demonstrates strong immune activation, while the other group shows pronounced immunosuppressive characteristics. Moreover, immune checkpoint therapy represents a transformative approach in cancer treatment, enhancing the immune system’s capacity to suppress or eliminate tumor cells by modulating immune pathways [34]. Our study observed significantly elevated expression of NRP1 in the high-risk patient cohort, whilst ITGAL, LAG3, and TIGIT exhibited upregulation in the low-risk group. This differential expression pattern suggests heterogeneous distribution characteristics of immune checkpoint molecules within the tumor microenvironment across distinct risk stratifications. Based on these findings, we propose adopting personalized immunotherapy strategies according to risk subgroups: for high-risk patients, inhibitors targeting the NRP1 signaling pathway should be priority consideration; whereas for low-risk patients, immune checkpoint blockade therapies targeting ITGAL, LAG3, and TIGIT may be more suitable. This approach enables more precise immune intervention and enhances treatment response rates.

To evaluate the model’s ability to predict drug sensitivity, we conducted sensitivity analyses for both high-risk and low-risk groups. The results revealed that high-risk patients exhibited lower IC50 values for sorafenib, confirming its significant therapeutic efficacy in treating tumors. Conversely, low-risk patients showed weaker responses to these drugs, underscoring sorafenib’s superior efficacy in high-risk patients. Sorafenib, an oral enzyme inhibitor widely used in liver cancer treatment, inhibits tumor cell proliferation and improves survival in patients with advanced liver cancer [46]. Studies have also shown that combining camrelizumab and rivoceranib results in statistically significant and clinically meaningful improvements over sorafenib monotherapy [47]. Moreover, sorafenib has been found to regulate tumor growth by inhibiting ferroptosis [48]. Additionally, dabrafenib and trametinib have proven effective in treating gliomas [49–51], while dasatinib, a potent tyrosine kinase inhibitor, is usually used in the support of chronic myeloid leukemia [52, 53]. In our study, people classified as low-risk patients exhibited lower IC50 values for sorafenib, dabrafenib, trametinib, and dasatinib, suggesting these drugs may offer therapeutic potential for this group as well.

In conclusion, this study successfully identified nine MMRGs and developed a predictive scoring model for assessing patient survival risk, providing potential new therapeutic targets and strategies for personalized treatment, several limitations remain. Firstly, although THCA subtypes defined by molecular characteristics exhibit significant differences in immune infiltration patterns, their potential as biomarkers or clinical tools for predicting treatment response requires further validation through clinical studies. Secondly, the key molecular mechanisms driving malignant progression in THCA remain incompletely elucidated. Consequently, future research should integrate in vitro cellular and molecular experiments with in vivo animal models to systematically reveal the multi-faceted clinical translational value of prognostic genes in THCA.

Conclusion

This study employed multi-omics analysis to identify MMRGs in THCA, highlighting the pivotal role of these genes in THCA treatment. A prognostic risk model was developed, integrating survival prediction and immunotherapy response assessment, while also identifying potential therapeutic targets. The results above indicate that they provide novel insights into the role of mitochondrial metabolism in THCA and establish a foundation for enhanced patient stratification and the development of personalized therapeutic strategies.

Supplementary Information

12957_2025_4169_MOESM1_ESM.docx (263.7KB, docx)

Supplementary Material 1: Supplementary Figure S1: Analysis of Prognostic Gene Expression in THCA and Patient Survival Prognosis.

12957_2025_4169_MOESM2_ESM.xls (68KB, xls)

Supplementary Material 2: Supplementary Table S1: Results of Univariate Cox Regression Analysis. Supplementary Table S2: LASSO Regression Analysis Results. Supplementary Table S3: Multivariate Cox Regression Analysis Results. Supplementary Table S4: Drug Sensitivity Prediction for Prognostic Genes from the Cellminer Database.

Acknowledgements

Not applicable.

Authors’ contributions

WGR, FQC and ZGH contributed to the study design. WGR conducted the literature search. FQC acquired the data. WGR wrote the article. ZGH performed data analysis. ZGH revised the article and gave the final approval of the version to be submitted. All authors read and approved the final manuscript.

Funding

The Study was supported by Jiangsu Provincial Medical Key Discipline (Laboratory), Project number: ZDXYS202211, project implementation period: 2022–2025.

Data availability

The data and materials in the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  • 1.Dabravolski SA, Nikiforov NG, Zhuravlev AD, Orekhov NA, Mikhaleva LM. A.N. Orekhov the role of altered mitochondrial metabolism in thyroid cancer development and Mitochondria-Targeted thyroid cancer treatment. Int J Mol Sci. 2021;23(1):460. [DOI] [PMC free article] [PubMed]
  • 2.Martínez-Hernández R, Sánchez N, de la Blanca P, Sacristán-Gómez A, Serrano-Somavilla JL, De Muñoz F, Sánchez, Cabo, et al. Unraveling the molecular architecture of autoimmune thyroid diseases at Spatial resolution. Nat Commun. 2024;15(1):5895. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Laha D, Nilubol N. Boufraqech new therapies for advanced thyroid cancer. Front Endocrinol. 2020;11:82. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Miranda-Filho A, Lortet-Tieulent J, Bray F, Cao B, Franceschi S, Vaccarella S, et al. Thyroid cancer incidence trends by histology in 25 countries: a population-based study, the lancet. Diabetes Endocrinol. 2021;9(4):225–34. [DOI] [PubMed] [Google Scholar]
  • 5.Pace-Asciak P, Russell JO. Tufano the treatment of thyroid cancer with radiofrequency ablation. Tech Vasc Interv Radiol. 2022;25(2):100825. [DOI] [PubMed] [Google Scholar]
  • 6.RP Tufano, P Pace-Asciak, JO Russell, C Suárez, GW Randolph, F López, et al. Update of radiofrequency ablation for treating benign and malignant thyroid Nodules. The future is now. Front Endocrinol. 2021;12:698689. [DOI] [PMC free article] [PubMed]
  • 7.Ancker OV, Krüger M, Wehland M, Infanger M. D. Grimm multikinase inhibitor treatment in thyroid cancer. Int J Mol Sci. 2019;21(1):10. [DOI] [PMC free article] [PubMed]
  • 8.Sugarman AJ, Huynh LD, Di Shabro A. Cristofano anaplastic thyroid cancer cells upregulate mitochondrial one-carbon metabolism to Meet purine demand, eliciting a critical targetable vulnerability. Cancer Lett. 2023;568:216304. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Jiang W, Zhang M, Gao C, Yan C, Gao R, He Z, et al. A mitochondrial EglN1-AMPKα axis drives breast cancer progression by enhancing metabolic adaptation to hypoxic stress. EMBO J. 2023;42(20):e113743. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Li Y, Sun Q, Chen S, Yu X, Jing H. Monensin inhibits anaplastic thyroid cancer via disrupting mitochondrial respiration and AMPK/mTOR signaling. Anti-cancer Agents Med Chem. 2022;22(14):2539–47. [DOI] [PubMed] [Google Scholar]
  • 11.Doolittle WKL, Park S, Lee SG, Jeong S, Lee G, Ryu D, et al. Non-genomic activation of the AKT-mTOR pathway by the mitochondrial stress response in thyroid cancer. Oncogene. 2022;41(44):4893–904. [DOI] [PubMed] [Google Scholar]
  • 12.Xing J, Qi L, Liu X, Shi G, Sun X. Yang roles of mitochondrial fusion and fission in breast cancer progression: a systematic review. World J Surg Oncol. 2022;20(1):331. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Vasan K, Werner M. Chandel mitochondrial metabolism as a target for cancer therapy. Cell Metabol. 2020;32(3):341–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Frattaruolo L, Brindisi M, Curcio R, Marra F, Dolce V. A.R. Cappello targeting the mitochondrial metabolic network: A promising strategy in cancer treatment. Int J Mol Sci. 2020;21(17):6014. [DOI] [PMC free article] [PubMed]
  • 15.Cabané P, Correa C, Bode I, Aguilar R. A.A. Elorza biomarkers in thyroid cancer: emerging opportunities from Non-Coding RNAs and mitochondrial space. Int J Mol Sci. 25(12):6719. [DOI] [PMC free article] [PubMed]
  • 16.Simula L, Fumagalli M, Vimeux L, Rajnpreht I, Icard P, Birsen G, et al. Mitochondrial metabolism sustains CD8(+) T cell migration for an efficient infiltration into solid tumors. Nat Commun. 2024;15(1):2203. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Zhang P, Zhang M, Liu J, Zhou Z, Zhang L, Luo P et al. Mitochondrial pathway signature (MitoPS) predicts immunotherapy response and reveals NDUFB10 as a key immune regulator in lung adenocarcinoma. J Immunother Cancer. 2025;13(7):e012069. [DOI] [PMC free article] [PubMed]
  • 18.Ma L. Fei antimalarial drug Artesunate is effective against chemoresistant anaplastic thyroid carcinoma via targeting mitochondrial metabolism. J Bioenerg Biomembr. 2020;52(2):123–30. [DOI] [PubMed] [Google Scholar]
  • 19.Xiong Y, Liu W, Huang Q, Wang J, Wang Y, Li H, et al. Tigecycline as a dual inhibitor of retinoblastoma and angiogenesis via inducing mitochondrial dysfunctions and oxidative damage. Sci Rep. 2018;8(1):11747. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Mohanty V. Inferring cancer metabolism from gene-expression data. Nat Rev Cancer. 2024;24(4):230. [DOI] [PubMed] [Google Scholar]
  • 21.Boucai L, Zafereo M. M E Cabanillas Thyroid Cancer: Rev Jama. 2024;331(5):425–35. [DOI] [PubMed] [Google Scholar]
  • 22.Díaz-Gago S, Vicente-Gutiérrez J, Ruiz-Rodríguez JM, Calafell J, Álvarez-Álvarez A, Lasa M, et al. Autophagy sustains mitochondrial respiration and determines resistance to BRAF(V600E) Inhibition in thyroid carcinoma cells. Autophagy. 2024;20(6):1383–97. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Sainero-Alcolado L, Liaño-Pons J, Ruiz-Pérez MV. Arsenian-Henriksson targeting mitochondrial metabolism for precision medicine in cancer. Cell Death Differ. 2022;29(7):1304–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Yu ZX, Xiang C, Xu SG. Zhang the clinical significance of thyroid hormone-responsive in thyroid carcinoma and its potential regulatory pathway. Medicine. 2022;101(31):e29972. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Anderson GW, Zhu Q, Metkowski J, Stack MJ, Gopinath S. Mariash the Thrsp null mouse (Thrsp(tm1cnm)) and diet-induced obesity. Mol Cell Endocrinol. 2009;302(1):99–107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Vella V. R. Malaguarnera the emerging role of insulin receptor isoforms in thyroid cancer: clinical implications and new perspectives. Int J Mol Sci. 2018;19(12):3814. [DOI] [PMC free article] [PubMed]
  • 27.Yu Y, Wang S, Zhang X, Xu S, Li Y, Liu Q, et al. Clinical implications of TPO and AOX1 in pediatric papillary thyroid carcinoma. Translational Pediatr. 2021;10(4):723–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Sigruener A, Buechler C, Orsó E, Hartmann A, Wild P.J., Terracciano L, et al. Human aldehyde oxidase 1 interacts with ATP-binding cassette transporter-1 and modulates its activity in hepatocytes. Hormone Metabolic Res = Hormon- Und Stoffwechselforschung = Horm Et Metab. 2007;39(11):781–9. [DOI] [PubMed] [Google Scholar]
  • 29.Giménez-Dejoz J, Weber S, Fernández-Pardo Á, Möller G, Adamski J, Porté S, et al. Engineering aldo-keto reductase 1B10 to mimic the distinct 1B15 topology and specificity towards inhibitors and substrates, including retinoids and steroids. Chemico-Biol Interact. 2019;307:186–94. [DOI] [PubMed] [Google Scholar]
  • 30.Weber S, Salabei JK, Möller G, Kremmer E, Bhatnagar A, Adamski J, et al. Aldo-keto reductase 1B15 (AKR1B15): a mitochondrial human aldo-keto reductase with activity toward steroids and 3-keto-acyl-CoA conjugates. J Biol Chem. 2015;290(10):6531–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Liu N, Zhu XR, Wu CY, Liu YY, Chen MB. Gu PCK1 as a target for cancer therapy: from metabolic reprogramming to immune microenvironment remodeling. Cell Death Discovery. 2024;10(1):478. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Liu R, Gou D, Xiang J, Pan X, Gao Q, Zhou P, et al. O-GlcNAc modified-TIP60/KAT5 is required for PCK1 deficiency-induced HCC metastasis. Oncogene. 2021;40(50):6707–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Day TF, Mewani RR, Starr J, Li X, Chakravarty D, Ressom H, et al. Transcriptome and proteome analyses of TNFAIP8 knockdown cancer cells reveal new insights into molecular determinants of cell survival and tumor progression. (Clifton N J). 2017;1513:83–100. Methods in molecular biology. [DOI] [PubMed] [Google Scholar]
  • 34.Yue C, Zheng X, Ye Y, Li D, Zhou Y. TNFAIP8L3 regulation of the TGF-β signaling pathway affects the proportion of macrophages during tumor antigen presentation and affects the prognosis of ovarian cancer. Front Mol Biosci. 2025;12:1566363. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Chen W, Chen X, Wang L, Yang R, Zhang W, Zhang S, et al. TIPE3 represses head and neck squamous cell carcinoma progression via triggering PGAM5 mediated mitochondria dysfunction. Cell Death Dis. 2023;14(4):251. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Song H, Tian D, Sun J, Mao X, Kong W, Xu D, et al. circFAM120B functions as a tumor suppressor in esophageal squamous cell carcinoma via the miR-661/PPM1L axis and the PKR/p38 MAPK/EMT pathway. Cell Death Dis. 2022;13(4):361. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Thean LF, Loi C, Ho KS, Koh PK, Eu KW. Cheah Genome-wide scan identifies a copy number variable region at 3q26 that regulates PPM1L in APC mutation-negative Familial colorectal cancer patients. Genes Chromosomes Cancer. 2010;49(2):99–106. [DOI] [PubMed] [Google Scholar]
  • 38.Ding H, Wang KY, Chen SY, Guo KW. Qiu validating the role of PTGIS gene in colorectal cancer by bioinformatics analysis and in vitro experiments. Sci Rep. 2023;13(1):16496. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Ouyang J, Feng Y, Zhang Y, Liu Y, Li S, Wang J, et al. Integration of metabolomics and transcriptomics reveals Metformin suppresses thyroid cancer progression via inhibiting Glycolysis and restraining DNA replication. Biomed pharmacotherapy = Biomedecine Pharmacotherapie. 2023;168:115659. [DOI] [PubMed] [Google Scholar]
  • 40.Cheng L, Tan Z, Huang Z, Pan Y, Zhang W. Wang expression profile and prognostic values of Mini-Chromosome maintenance families (MCMs) in breast cancer. Med Sci Monitor: Int Med J Experimental Clin Res. 2020;26:e923673. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Cui X, Liu R, Duan L, Zhang Q, Cao D. Zhang exogenous hydrogen sulfide (H(2)S) exerts therapeutic potential in triple-negative breast cancer by affecting cell cycle and DNA replication pathway. Biomed pharmacotherapy = Biomedecine Pharmacotherapie. 2023;161:114488. [DOI] [PubMed] [Google Scholar]
  • 42.Han PZ, Ye WD, Yu PC, Tan LC, Shi X, Chen XF et al. A distinct tumor microenvironment makes anaplastic thyroid cancer more lethal but immunotherapy sensitive than papillary thyroid cancer. JCI Insight. 2024;9(8):e173712. [DOI] [PMC free article] [PubMed]
  • 43.Ramirez-Valdez RA, Baharom F, Khalilnezhad A, Fussell SC, Hermans DJ, Schrager AM, et al. Intravenous heterologous prime-boost vaccination activates innate and adaptive immunity to promote tumor regression. Cell Rep. 2023;42(6):112599. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Pinheiro Neto A, Lucchesi HL, Valsecchi V, Ward LS. L.L. Cunha immunotherapy for patients with thyroid cancer: a comprehensive appraisal. Chin Clin Oncol. 2024;13(3):36. [DOI] [PubMed] [Google Scholar]
  • 45.Wang Q, Pan G, Zhang Y, Ni Y, Mu Y. Luo emerging insights into thyroid cancer from immunotherapy perspective: A bibliometric analysis. Hum Vaccines Immunotherapeutics. 2024;20(1):2403170. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Kong FH, Ye QF, Miao XY, Liu X, Huang SQ, Xiong L, et al. Current status of Sorafenib nanoparticle delivery systems in the treatment of hepatocellular carcinoma. Theranostics. 2021;11(11):5464–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Qin S, Chan SL, Gu S, Bai Y, Ren Z, Lin X, et al. Camrelizumab plus Rivoceranib versus Sorafenib as first-line therapy for unresectable hepatocellular carcinoma (CARES-310): a randomised, open-label, international phase 3 study. Lancet (London England). 2023;402(10408):1133–46. [DOI] [PubMed] [Google Scholar]
  • 48.Li Q, Chen K, Zhang T, Jiang D, Chen L, Jiang J, et al. Understanding sorafenib-induced ferroptosis and resistance mechanisms: implications for cancer therapy. Eur J Pharmacol. 2023;955:175913. [DOI] [PubMed] [Google Scholar]
  • 49.Bouffet E, Hansford JR, Garrè ML, Hara J, Plant-Fox A, Aerts I, et al. Dabrafenib plus Trametinib in pediatric glioma with BRAF V600 mutations. N Engl J Med. 2023;389(12):1108–20. [DOI] [PubMed] [Google Scholar]
  • 50.Bharadwaj M, Nagda D. Goyal mudras in medicine: A role for dance in appreciating nonverbal communication in the clinical encounter. J Clin Oncology: Official J Am Soc Clin Oncol. 2023;41(1):128–31. [DOI] [PubMed] [Google Scholar]
  • 51.Bouffet E, Geoerger B, Moertel C, Whitlock JA, Aerts I, Hargrave D, et al. Efficacy and safety of Trametinib monotherapy or in combination with Dabrafenib in pediatric BRAF V600-Mutant Low-Grade glioma. J Clin Oncology: Official J Am Soc Clin Oncol. 2023;41(3):664–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Alqattan Y, Ali S, Almohammad R, Kayali N. Alhuraiji Dasatinib-induced chylothorax in chronic myeloid leukemia. Gulf J Oncol. 2022;1(40):74–7. [PubMed] [Google Scholar]
  • 53.Crounse RG. The diagnostic value of microscopic examination of human hair. Arch Pathol Lab Med. 1987;111(8):700–2. [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

12957_2025_4169_MOESM1_ESM.docx (263.7KB, docx)

Supplementary Material 1: Supplementary Figure S1: Analysis of Prognostic Gene Expression in THCA and Patient Survival Prognosis.

12957_2025_4169_MOESM2_ESM.xls (68KB, xls)

Supplementary Material 2: Supplementary Table S1: Results of Univariate Cox Regression Analysis. Supplementary Table S2: LASSO Regression Analysis Results. Supplementary Table S3: Multivariate Cox Regression Analysis Results. Supplementary Table S4: Drug Sensitivity Prediction for Prognostic Genes from the Cellminer Database.

Data Availability Statement

The data and materials in the current study are available from the corresponding author on reasonable request.

Articles from World Journal of Surgical Oncology are provided here courtesy of BMC

RESOURCES