Machine learning-derived mast cell-associated angiogenesis features can serve as prognostic targets for clear cell renal cell carcinoma

Introduction

Renal cell carcinoma (RCC) is the sixth most common malignant tumor in males and the ninth most frequently diagnosed in females, accounting for approximately 5% and 3% of all newly diagnosed cancer cases in men and women, respectively (1). Clear cell renal cell carcinoma (ccRCC) represents the most prevalent and malignant subtype of renal cancer, with a 5-year overall survival rate of 50–70% (2). The disease has a poor prognosis, as around 45% of ccRCC patients present with or develop metastatic lesions at diagnosis (3), and about 30% of those undergoing surgery eventually experience relapse (4-6). Despite advancements in surgical resection, immunotherapy, and novel targeted agents that have improved progression-free and overall survival, tumor heterogeneity and drug resistance continue to limit overall efficacy. The majority of cases still develop long-term recurrence (7-9). Therefore, the development of novel and reliable molecular markers is of great importance for optimizing the prognostic evaluation system of ccRCC patients, accurately predicting overall survival and guiding personalized treatment.

Mast cells are tissue-resident immune cells prevalent in barrier tissues like the lungs, urinary tract, intestines, and skin (10,11). Mast cells play a crucial role in tumorigenesis and progression by activating multiple signaling pathways, which convert local microenvironmental stimuli into controlled release of active mediators (12). Mast cells release cytokines, chemokines, and bioactive amines, modulating the immune response within the tumor microenvironment and affecting tumor growth, invasion, metastasis, and treatment response (13,14). Notably, mast cells also serve a key role in the process of tumor angiogenesis (15). Mast cells can synthesize and release a variety of molecules that regulate angiogenesis, possessing both pro-angiogenic and anti-angiogenic activities, but pro-angiogenic factors are predominant, so mast cells are key cells that amplify the angiogenic response in tumors (16). They can secrete various pro-angiogenic factors such as FGF-2, IL-8, TNF-α, TGF-β, VEGF, and NGF (17-20). Research using animal models has shown that mouse mast cells and their granules can enhance angiogenesis in the chick embryo chorioallantoic membrane assay, and this pro-angiogenic effect can be partially blocked by antibodies against FGF-2 and VEGF (21). Serine proteases, including tryptase and chymase, stored in mast cell granules, facilitate endothelial cell proliferation and vascular lumen formation. They also enhance angiogenesis by degrading the extracellular matrix (ECM) to release VEGF and FGF-2 (22,23). Mast cells possess TIMPs that control MMP-driven ECM degradation and the activation of angiogenic factors (24,25). In lung cancer, mast cell density exhibits a positive correlation with microvessel density in tumor regions, further supporting their promoting role in angiogenesis (26).

Angiogenesis constitutes one of the critical biological processes underpinning cancer cell proliferation, invasion, and metastasis. Inhibiting aberrant angiogenesis has emerged as a pivotal strategy in contemporary oncology therapeutics, wherein monoclonal antibodies targeting VEGF signaling pathways and tyrosine kinase inhibitors (TKIs) have been most widely employed (27,28). The clinical application of these anti-angiogenic drugs has not only improved treatment outcomes for patients but also promoted the establishment of a novel anti-tumor therapeutic model centered on vascular normalization.

The advancement of multi-omics technologies, such as proteomics, genomics, and metabolomics, alongside the promising role of biomarkers in clinical diagnosis and precision therapy, has significantly advanced research on molecular markers for RCC (29,30). Multiple candidate biomarkers have been proposed to reveal tumor heterogeneity, predict therapeutic response, and evaluate drug toxicity, offering new directions for personalized clinical management (31). The biological roles and prognostic significance of mast cell-mediated angiogenesis in ccRCC are not well understood. This study seeks to address the gap by offering new insights into the association between the tumor microenvironment and disease progression. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0248/rc).

Methods

Data acquisition

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Transcriptomic profiles of ccRCC derived from the TCGA-KIRC cohort were retrieved through the UCSC Xena platform (https://xena.ucsc.edu). Following filtration of samples with integrated survival data, a total of 525 patients were finally enrolled as the study cohort for subsequent analyses. Additionally, two independent ccRCC datasets were included for validation: the GSE167573 dataset (n=56) acquired from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) and the E-MTAB-1980 dataset (n=101) obtained from ArrayExpress (https://www.ebi.ac.uk/biostudies/arrayexpress). Meanwhile, a panel of angiogenesis-associated genes was extracted from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database.

Single-cell analysis

We employed the TISCH portal (http://tisch.compbio.cn/home/) for single-cell analysis of the GSE171306 dataset. This portal, as a mature single-cell data visualization tool, is widely embraced by the scientific community (32). For detailed usage instructions, please refer to the developers’ original article (33) due to space limitations.

Construction of the mast cell-associated angiogenesis score (MCAS) prognostic model

A comprehensive literature review guided the incorporation of ten machine learning algorithms for the construction of a reliable prognostic signature, termed the MCAS (34-37). Specifically, stepwise Cox modeling was performed with the survival package using AIC-based variable selection; RSF was implemented via randomForestSRC with grid search and 10-fold cross-validation optimizing ntree and mtry; Enet, Lasso, and Ridge were executed using glmnet; SuperPC employed superpc for survival-oriented dimension reduction; plsRcox utilized plsRcox for partial least squares feature extraction; CoxBoost was fitted with CoxBoost; survival-SVM was run via survivalsvm; and GBM was trained with gbm iteratively combining weak learners. The TCGA-KIRC cohort was utilized for model development, with the GSE167573 and E-MTAB-1980 cohorts employed for external validation. Kaplan-Meier (KM) and time-dependent receiver operating characteristic (ROC) curves were utilized to assess prognostic stratification performance. Subsequently, SurvSHAP(t) explanation was applied using the survex package to interpret model predictions at the individual level.

Functional pathway enrichment and immune cell infiltration analysis

Differential expression analysis utilized the limma package, applying criteria of |log₂FC| >1 and an adjusted P value <0.05 to identify differentially expressed genes (DEGs). Enrichment analyses for Gene Ontology (GO) terms and KEGG pathways were performed using the DEGs. Gene set enrichment analysis (GSEA) was utilized to compare Hallmark gene sets between risk groups, highlighting functional differences. The ESTIMATE R package was utilized to compute stromal, immune, and combined scores, characterizing the tumor microenvironment’s cellular composition. To quantify immune cell infiltration in ccRCC patients, the CIBERSORT algorithm was employed. A Spearman correlation analysis was conducted to evaluate the association between model genes and immune cell infiltration levels. Furthermore, the study analyzed variations in HLA family member expression and single-sample gene set enrichment analysis (ssGSEA)-based immune function scores between risk subgroups.

Quantitative real-time polymerase chain reaction (qRT-PCR) and immunohistochemistry

Our prior study has detailed the specific protocols for total RNA extraction and qRT-PCR (37). The primer sequences utilized in the present study are listed in Table 1.

Paired tumor and adjacent non-tumor tissue specimens were collected from patients with pathologically confirmed ccRCC who underwent surgical resection at the Department of Urology, The First Affiliated Hospital of Nanjing Medical University. Adjacent non-tumor tissues were obtained from macroscopically non-cancerous renal tissue located away from the tumor margin. Patients who had received preoperative anti-tumor therapy, including targeted therapy, immunotherapy, radiotherapy, or chemotherapy, were excluded. Cases with insufficient tissue quality, incomplete clinicopathological information, or non-ccRCC histology were also excluded. Total RNA was extracted from paired tissue samples, and qRT-PCR was performed to validate the expression of selected genes. The study was approved by the Ethics Committee of The First Affiliated Hospital of Nanjing Medical University (No. 2021-SR-430), and written informed consent was obtained from each patient prior to tissue collection.

Statistical analysis

Statistical analyses and graphical representations were performed using R software (version 4.3.3). Statistical significance was set at P<0.05, with the Benjamini-Hochberg procedure applied for multiple testing corrections to control the false discovery rate (FDR) as needed. Between-group comparisons were performed using the Wilcoxon rank-sum test, and correlations were evaluated via Spearman’s rank correlation coefficient.

Results

Analysis of the single-cell microenvironment of mast cells in ccRCC

Single-cell RNA sequencing analysis of ccRCC samples yielded 25 distinct cell clusters through clustering (Figure 1A), which were subsequently annotated into 10 major cell types (Figure 1B). Functional analysis of mast cell subpopulations revealed significant upregulation of TNF-α signaling activity via the NF-κB pathway in this cell type (Figure 1C), suggesting potential importance within the tumor microenvironment. When examining cell population-specific transcription factors, we identified STAT1 and GATA1 as most critical in mast cells (Figure 1D). Concurrently, we observed that mast cells demonstrate extensive cellular interactions with all other cell populations (Figure 1E). Specifically, the SPP1-CD44 ligand-receptor interaction exhibits high intensity between malignant cells (Malignant_C6, Malignant_C9) and mast cells, while the FN1-CD44 interaction demonstrates elevated intensity between endothelial cells (Endothelial_C19) and mast cells (Figure 1F). Additionally, high interaction intensity is observed for CLEC2C-KLRB1 between mast cells and natural killer (NK) cells (NK_C0) (Figure 1G).

Figure 1 The single-cell landscape of mast cells in clear cell renal cell carcinoma. (A) Cell clustering in GSE171306. (B) Cell type annotations in GSE171306. (C) TNF-α via NF-κB pathway activity in GSE171306. (D) Ranking of the importance of TFs in mast cell subpopulations. (E) Cell communication between mast cells and other cell populations. (F) Ligand-receptor interaction strength from other cell types to mast cells. (G) Ligand-receptor interaction strength from mast cells to other cell types. NF-κB, nuclear factor κB; NK, natural killer; TF, transcription factor; TNF-α, tumor necrosis factor-α.

Construction of prognostic model

A joint prognostic model was developed by integrating ten machine learning algorithms using intersecting genes. Given that the Lasso + CoxBoost model demonstrated the highest mean concordance index (C-index), this model was identified as the optimal prognostic model (Figure 2A). Consequently, we derived a risk score designated as MCAS. Scatter plots demonstrated a correlation between higher MCAS and decreased patient survival, with more deaths observed in the high-score group compared to the low-score group (Figure 2B-2D). Furthermore, KM curves confirmed significant prognostic stratification between high- and low-MCAS groups (Figure 2E-2G). The predictive capacity of MCAS for patient survival was evaluated through ROC curves, showing area under the curve (AUC) values above 0.8 in the GSE167573 cohort and typically over 0.7 in the E-MTAB-1980 and TCGA-KIRC cohorts, occasionally exceeding 0.8 (Figure 2H-2J).

Figure 2 Construction of the MCAS prognostic model. (A) C-index heatmap for combinations of machine learning algorithms. (B-D) Scatter plots of survival time versus survival status in the entire, training, and validation cohorts. (E-G) KM curves for overall survival in TCGA-KIRC (E), E-MTAB-1980 (F), and GSE167573 (G). (H-J) Time-dependent ROC curves in the TCGA-KIRC (H), E-MTAB-1980 (I), and GSE167573 (J). AUC, area under the curve; C-index, concordance index; KM, Kaplan-Meier; MCAS, mast cell-associated angiogenesis score; ROC, receiver operating characteristic; TCGA-KIRC, The Cancer Genome Atlas Kidney Renal Clear Cell Carcinoma.

Independent validation of prognostic value

We employed pie charts to organize and display the distribution of clinical analyses in the TCGA cohort. As expected, the high-risk subgroup exhibited a higher proportion of progressive clinical features (Figure 3A). Similarly, box plot results demonstrated that the risk score increased with advancing clinical grade (Figure 3B). Subsequent KM curves validated the model’s remarkable capacity for risk stratification among subgroups with different clinical features (Figure 3C). Cox regression analysis further demonstrated that risk stratification represented an independent prognostic indicator in each cohort (Figure 3D,3E). The calibration curves showed that MCAS was highly consistent with the ideal curve, highlighting the excellent performance of MCAS in predicting survival status. In addition, the 1-, 3-, and 5-year decision curve analysis (DCA) curves indicated that MCAS provided certain clinical benefits in clinical decision-making (Figure S1).

Figure 3 Clinical characteristic analysis of MCAS. (A) Pie charts of clinical characteristics. (B) Comparison of risk score among levels of clinical characteristics. (C) KM curves for risk stratification in subgroups with different clinical characteristics. (D) Univariate Cox regression of risk score and clinical characteristics. (E) Multivariate Cox regression of risk score and clinical characteristics. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001; ns, not significant. KM, Kaplan-Meier; MCAS, mast cell-associated angiogenesis score.

Analysis of functional differences between groups

GO-based enrichment analysis identified significant enrichment in transmembrane substance transport and cell membrane functions, particularly in organic anion transport, the apical cell region, apical plasma membrane, secondary active transmembrane transporter activity, and organic anion transmembrane transporter activity (Figure 4A,4B). KEGG enrichment analysis revealed notable dysregulation in the degradation of valine, leucine, and isoleucine, as well as in the PPAR signaling pathway and complement and coagulation cascades (Figure 4C). Further in‑depth GSEA verified that the IL-6/JAK/STAT3 signaling pathway was significantly enriched in the high‑score group. In contrast, fatty acid metabolism, bile acid metabolism and oxidative phosphorylation were the dominant pathways in the low-score group (Figure 4D,4E).

Figure 4 Functional differences between high- and low-risk groups. (A,B) The GO enrichment analysis of DEGs. (C) The KEGG pathway analysis. (D,E) GSEA between the two groups. BP, biological process; CC, cellular component; DEGs, differentially expressed genes; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; MCAS, mast cell-associated angiogenesis score; MF, molecular function; PPAR, peroxisome proliferator-activated receptor.

Immune infiltration analysis

The high-risk subgroup exhibited markedly elevated stromal score, immune score, and ESTIMATE score (Figure 5A). The CIBERSORT algorithm was applied to assess the infiltration levels of various immune cell subsets in the TCGA training cohort, providing insights into the tumor immune microenvironment. Differential infiltration of these immune cells was illustrated using a box plot (Figure 5B). Macrophages M1 were substantially more concentrated in the low-risk group, while the high-risk group showed increased levels of regulatory T cells (Tregs). We utilized the ssGSEA method to quantitatively assess the functional activity of various immune cells, visualized through box plots (Figure 5C). The study found that in the high-risk subgroup, enrichment scores for immune function modules associated with checkpoint, HLA, and Treg were significantly increased, while the score for the mast cells subset was notably decreased. We observed a marked decrease in the expression levels of key HLA molecules, such as HLA-A, HLA-B, and HLA-C, within the high-risk subgroup (Figure 5D). Correlation analysis indicated a significant association between model gene expression levels and the infiltration levels of various immune cell subsets in the tumor microenvironment (Figure 5E). EMCN was inversely associated with Tregs infiltration, whereas TIMP1 exhibited a positive association with Tregs infiltration. NRP1 demonstrated a negative correlation with NK cells activated infiltration, whereas EMCN was positively correlated with monocyte infiltration.

Figure 5 Differences in immune infiltration between high- and low-risk groups. (A) The TME scores contrasting low- and high-risk patients. (B) Comparison of immune cell abundance between high- and low-risk groups. (C) Comparison of immune functions between high- and low-risk groups. (D) Comparison of HLA expression between high- and low-risk groups. (E) Correlation of model genes with immune cell abundance. *, P<0.05; **, P<0.01; ***, P<0.001. HLA, human leukocyte antigen; MCAS, mast cell-associated angiogenesis score; TME, tumor microenvironment; Treg, regulatory T cell.

Interpretation of the mutational landscape

Our study found that the high-risk subgroup had an increased tumor mutational burden (TMB) (Figure 6A), and TMB levels showed a significantly positive correlation with risk score (Figure 6B). KM survival analysis indicated that patients with low TMB had a more favorable prognosis compared to those with high TMB (Figure 6C). Further integrated prognostic stratification analysis demonstrated that patients with both low risk and low TMB exhibited the most favorable clinical outcomes. Stratification analysis at approximately 2,160 days indicated that patients with high risk and elevated TMB had the worst prognosis at this time point. Beyond this time point, patients with high risk and low TMB displayed the poorest prognosis (Figure 6D). Comparison of mutational landscapes across different risk subgroups (Figure 6E,6F) demonstrated that the top four mutated genes were identical in both groups, with SETD2 mutations notably ranking higher in the high-risk group. BAP1 mutations occurred more frequently in the high-risk group, whereas PBRM1 mutations were more prevalent in the low-risk subgroup. A comparative assessment of tumor signaling pathway activation among different risk subgroups showed no statistically significant differences in mutation frequency across groups (Figure 6G,6H).

Figure 6 Differences in somatic mutation landscapes between high- and low-risk groups. (A) Comparison of TMB between high- and low-risk groups. (B) Correlation of risk score with TMB. (C) KM curves for TMB stratification. (D) KM curves for TMB-risk stratification. (E) Ranking of mutated genes in the low-risk group. (F) Ranking of mutated genes in the high-risk group. (G) Ranking of affected pathways in the low-risk group. (H) Ranking of affected pathways in the high-risk group. **, P<0.01. KM, Kaplan-Meier; MCAS, mast cell-associated angiogenesis score; TMB, tumor mutational burden.

Analysis of gene expression patterns in models

We visualized the expression patterns of model genes from a single-cell perspective using expression distribution maps and violin plots (Figure 7A,7B). For example, NOTCH4 and EMCN showed high expression levels in endothelial cells, COL4A2 was expressed in both endothelial cells and fibroblasts, whereas VCAN had increased expression in neutrophils and cancerous cells. Expression analysis revealed decreased EMCN levels in tumor samples with elevated expression of other genes (Figure 8A). The qRT-PCR expression levels of model genes in ccRCC samples and normal renal tissue samples were fully consistent with our bioinformatics analysis results (Figure 8B). Immunohistochemical profiles derived from the HPA database (https://www.proteinatlas.org/) also validate this at the protein level (Figure 8C). Detailed information of all HPA-derived images is listed in Table S1.

Figure 7 Single-cell expression landscape of model genes. (A) Model gene expression landscape in GSE171306. (B) Comparison of model gene expression among different cell types in GSE171306. The color bar represents the normalized gene expression level. NK, natural killer.

Figure 8 Validation of differential expression of prognostic model genes. (A) Comparison of model gene expression levels between normal and tumor samples. (B) qRT-PCR experimental validation. (C) Immunohistochemical staining of model genes in normal and tumor tissues. Antibody IDs, sample information, and source URLs for all images are provided in Table S1. Image credit: Human Protein Atlas. *, P<0.05; **, P<0.01; ***, P<0.001. qRT-PCR, quantitative real-time polymerase chain reaction; TPM, transcripts per million.

Characterization of the core model gene

Diagnostic ROC analysis indicated that the model genes generally exhibited AUC values exceeding 0.8 (Figure 9A), indicating their robust diagnostic efficacy for ccRCC and potential clinical utility. The SurvSHAP(t) method was employed to assess the influence of various model components on temporal survival predictions. The findings identified EMCN as the primary influencing factor, with TIMP1 and NRP1 following in significance (Figure 9B,9C). Survival analysis revealed that patients with low TIMP1 expression had improved survival, whereas those with high expression of NRP1, EMCN, COL4A2, NOTCH4, and ANGPTL4 consistently demonstrated better survival outcomes. The expression levels of VCAN, FSTL1, and CCND2 exhibited no significant association with patient overall survival (Figure 9D).

Figure 9 Clinical characteristics of model genes. (A) Diagnostic ROC curves. (B) SurvSHAP(t) model feature importance ranking. (C) Beeswarm plot showing the ranking of aggregated SurvSHAP(t) values. (D) KM curves for survival in high- and low-expression groups of model genes. AUC, area under the curve; FPR, false positive rate; KM, Kaplan-Meier; ROC, receiver operating characteristic; TPR, true positive rate.

Expression and functional enrichment analysis of core model genes EMCN and TIMP1

We utilized box plots to illustrate the expression of EMCN across various clinicopathological characteristics in ccRCC patients (Figure 10A). The results demonstrated a marked negative correlation between EMCN expression and tumor stage, histological grade, and TNM stage. Further univariate and multivariate Cox regression analyses showed that EMCN is an independent protective factor for patients with ccRCC, suggesting that EMCN has potential prognostic value (Table 2). Patients were stratified into high- and low-expression subgroups based on the median EMCN expression value. DEGs between these groups were identified, followed by GO and KEGG enrichment analyses (Figure 10B, Table 3). The findings revealed a substantially enriched expression of EMCN in neuroactive ligand-receptor interactions, gated channel activity, and passive transmembrane transporter activity. GSEA revealed notable upregulation of TGF-β and Wnt/β-catenin signaling pathways in the high EMCN expression group, while the low EMCN expression group exhibited significant upregulation of immune-related pathways, including KRAS signaling, coagulation, and complement cascade (Figure 10C). These results suggest that EMCN may exert a protective role in ccRCC by inhibiting the activation of pro-tumor signaling pathways and regulating immune responses within the tumor microenvironment.

Figure 10 Functional analysis of core genes in the prognosis model. (A) Box plots showing EMCN expression across different clinicopathological features. (B) Volcano plot of GO/KEGG enrichment analysis for genes co-expressed with EMCN. (C) GSEA plot showing the enrichment of key signaling pathways in the EMCN high-expression group. (D) Box plots showing TIMP1 expression across different clinicopathological features. (E) Volcano plot of GO/KEGG enrichment analysis for genes co-expressed with TIMP1. (F) GSEA plot showing the enrichment of key signaling pathways in the TIMP1 high-expression group. *, P<0.05; **, P<0.01; ***, P<0.001. BP, biological process; CC, cellular component; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function; TPM, transcripts per million.

Similarly, we presented TIMP1 expression across different clinicopathological characteristics of ccRCC patients using a box plot (Figure 10D). The results demonstrated that TIMP1 expression levels showed significant positive correlations with tumor stage, histological grading, and TNM staging. Further univariate and multivariate Cox regression analyses showed that TIMP1 was an independent risk factor for patients with ccRCC, suggesting its potential prognostic value (Table 4). Patients were stratified into high- and low-expression subgroups based on the median value of TIMP1 expression. Differential genes between the groups underwent GO and KEGG enrichment analysis (Figure 10E, Table 5). Results indicated a significant enrichment of TIMP1 in the collagen-containing ECM, structural constituents of the ECM, and the organization of external encapsulating structures. GSEA revealed notable downregulation of oxidative phosphorylation pathway in the TIMP1 high-expression group, whereas epithelial-mesenchymal transition (EMT), inflammatory response, apical junction, and hypoxia pathways were significantly upregulated (Figure 10F). The results indicate that TIMP1 potentially promotes tumorigenesis in ccRCC by activating signaling pathways that facilitate stromal remodeling and inflammatory responses in the tumor microenvironment.

Discussion

Previous studies have highlighted the important roles of pathway-based molecular subtyping, such as mTOR-related signaling, in angiogenesis, metabolic regulation, and therapeutic stratification in RCC (38). These findings provide valuable references for exploring treatment potential based on tumor biological characteristics. In line with this concept, our study integrates single-cell and bulk transcriptomic analyses to characterize mast cell-associated angiogenesis in ccRCC and identifies TIMP1 and EMCN as key prognostic molecules with potential therapeutic implications. Our findings further support the feasibility of biologically driven risk stratification for personalized treatment in ccRCC.

In recent years, several prognostic models for ccRCC have been established from different perspectives, including deep learning-based models using histopathological images, multi-omics-based signatures of methylated DEGs, and machine learning models based on intra-tumor heterogeneity (39-41). Compared with the above models, our proposed MCAS specifically focuses on the crosstalk between mast cells and angiogenesis—a critical but underexplored aspect of ccRCC progression. By integrating single-cell transcriptomics and ten machine learning algorithms, MCAS has demonstrated good generalizability across multiple independent cohorts, complementing existing prognostic signatures and identifying potential therapeutic targets such as TIMP1 and EMCN.

The study demonstrated that TNF-α expression is increased in mast cells through the NF-κB signaling pathway. TNF-α primarily activates the NF-κB signaling pathway by binding to TNF-R1, thereby regulating inflammatory response, cell survival, and proliferation; its persistent activation promotes tumorigenesis and development, constituting the core mechanism through which TNF-α exerts its pro-tumor effects. In nearly all cell types, NF-κB is activated upon exposure to TNF-α, resulting in the upregulation of a large number of inflammation-associated genes. NF-κB is transiently activated by cytokine stimulation, triggering an inflammatory response. However, its persistent activation is linked to tumorigenesis, encompassing cellular proliferation of cancer cells, inhibition of apoptosis under drug-resistant conditions, and elevated tumor angiogenesis and metastatic potential. The content underscores the crucial role of the TNF-α via NF-κB signaling pathway in the progression of malignant tumors (42). Mast cells, important inflammatory immune cells in the tumor microenvironment, contribute to local inflammation and tumor growth by releasing pro-inflammatory substances like TNF-α. Targeting TNF-α via NF-κB signaling in mast cells may inhibit their pro-inflammatory and pro-tumorigenic effects, offering a novel strategy to reduce inflammatory activation in the tumor microenvironment and limit tumor progression.

STAT1 and GATA1 are two key transcription factors in mast cells that collectively regulate their biological functions. Abnormal expression and dysfunction of these factors are closely associated with tumorigenesis and development. Among these, the activation of STAT1 is crucial in cellular proliferation, apoptosis, autophagy, and immune regulation. Its overactivation or deficiency can trigger pathological alterations in the organism and is closely associated with tumorigenesis and progression (43); GATA1 mRNA expression was significantly downregulated in ccRCC tissues. The downregulation of this gene was significantly associated with poor clinicopathological characteristics such as tumor advancement and metastatic spread, markedly increased patients’ risk of disease recurrence, and served as an independent prognostic marker for ccRCC recurrence (44).

The study identified significant ligand-receptor interactions involving mast cells and adjacent cell populations, notably SPP1-CD44 interactions with malignant cells and CLEC2C-KLRB1 interactions with NK cells. SPP1, or OPN, is a phosphorylated glycoprotein that engages with integrins and CD44 receptors in the tumor-promoting microenvironment. This interaction triggers downstream signaling pathways, promoting cancer cell migration, invasion, and metastasis (45). Prior studies demonstrate the significant role of the CLEC2C-KLRB1 axis in forming an immunosuppressive tumor microenvironment. GNB1 contributes to an immunosuppressive microenvironment by enhancing CLEC2C-KLRB1 interactions between memory B cells and CD8⁺ T cells, reducing the effectiveness of PD-1 inhibitors (46). Additionally, decreased immunomodulatory interactions in tumor tissue, linked to molecules like CLEC2C and KLRB1, weaken immune activation and obstruct peripheral immune cell infiltration (47). Furthermore, NMPR treatment induces the expansion of hyperactivated regulatory T cells (hyper-Treg) and directly suppresses KLRB1-positive exhausted precursor CD8⁺ T cells via the CLEC2C-KLRB1 axis, thereby blocking antitumor immune responses (48). These interactions collectively constitute a pro-tumor cellular communication network that regulates multiple mechanisms of tumor progression. This study suggests that the SPP1-CD44 and CLEC2C-KLRB1 signaling pathways, associated with mast cells, are promising targets for precision immuno-oncology therapies.

This study established a robust renal cancer prognostic stratification model using Lasso combined with CoxBoost machine learning algorithms, based on intersection genes between mast cell marker genes and angiogenesis genes. The nine genes included are ANGPTL4, EMCN, NRP1, COL4A2, NOTCH4, TIMP1, CCND2, FSTL1, and VCAN. Increased circulating ANGPTL4 levels are linked to a higher risk of various cancers, indicating its potential as a pan-cancer biomarker for risk stratification and early prevention (49). Specifically, in ccRCC, ANGPTL4 can regulate the proliferation of ccRCC cells through the ERK/P38 pathway (50). This study found its elevated expression in tumor samples, further suggesting its potential oncogenic role in ccRCC. EMCN is a glycoprotein expressed in endothelial cells (51). The prognostic effect of EMCN in ccRCC remains unclear. In this study, this gene was downregulated in tumor samples, serving as a protective factor for ccRCC. EMCN was inversely correlated with Treg infiltration and positively correlated with monocyte infiltration. This suggests that EMCN may reduce immune escape by suppressing Treg infiltration while promoting monocyte-mediated anti-tumor immune responses, thereby inhibiting ccRCC progression. In this study, NRP1 was highly expressed and functioned in a cancer-promoting role, consistent with previous experimental findings. Aberrant GFAP expression, limited vasculature-astrocyte interaction, and endothelial NRP1 positivity may contribute to dysregulated angiogenesis and tumor progression (52). Its elevated expression is negatively correlated with activated NK cell infiltration, suggesting that NRP1 may also facilitate malignant progression through modulation of the tumor immune microenvironment. COL4A2, a subunit of type IV collagen, shows marked clinical correlations in various cancers (53-55). While our study validated COL4A2 as an unfavorable prognostic biomarker for ccRCC, its precise regulatory signaling pathways in this disease warrant further investigation. NOTCH4, a crucial NOTCH family member, uniquely regulates tumor cell and vasculature behaviors. Previous research has shown that pathway activation by NOTCH4 promotes cancer, aligning with the high NOTCH4 expression found in tumor samples in this study. In hepatocellular carcinoma, Notch4 enhances vasculogenic mimicry by increasing MT1-MMP, MMP-2, and MMP-9 expression and activity, which boosts HCC’s invasive and metastatic potential (56). Notch4 activation in gastric cancer cells stimulates Wnt1, β-catenin, and downstream targets c-Myc and cyclin D1, enhancing cell growth in vitro and in vivo (57). TIMP1 enhances ccRCC malignancy by directly mediating EMT and indirectly promoting tumor progression through a positive feedback loop involving inflammatory response regulation and IL-6/JAK/STAT3 pathway activation (58). TIMP1 was identified as a risk factor in this study, consistent with its role as a risk factor in other cancers reported in previous research. The pathway activated in the TIMP1 high-expression group showed strong correlation with cancer progression. Furthermore, TIMP1 exhibited positive correlation with Tregs, potentially facilitating immune escape through promoting Treg infiltration, thereby contributing to ccRCC progression. CCND2 is commonly regarded as an oncogene in multiple tumor types. CCND2 overexpression has been documented in testicular germ cell tumor cell lines (59), colorectal cancer (60), and gastric cancer, correlating with tumor progression and poor prognosis (61,62). This aligns with the elevated CCND2 expression detected in tumor samples in this study. FSTL1 potentially acts as a tumor suppressor by inhibiting NF-κB and HIF-2α signaling pathways (63). The study revealed no notable variation in the gene expression levels between tumor and normal samples. In this research, VCAN showed high expression in tumor samples and functions as a pro-cancer factor. VCAN is an ECM proteoglycan macromolecule that promotes cancer progression in ccRCC through the ‘ADAMTS1-VCAN-EGFR signaling axis’. This pathway disrupts cellular dependence on death signals and enhances the invasiveness of cancer cells (64).

The mTOR pathway has received increasing attention in renal cancer because of its important roles in tumor metabolism, proliferation, immune regulation, and therapeutic response. Several genes in our prognostic model may be connected to this pathway from different biological perspectives. Previous studies have shown that FSTL1 can bind TLR4 and activate mTOR signaling to promote malignant progression and metastasis in hepatocellular carcinoma (65), while CCND2 overexpression activates mTOR signaling and cooperates with AML1-ETO to drive leukemogenesis in a pre-leukemic mouse model (66). TIMP1 may also serve as a molecular link among ECM remodeling, mitochondrial metabolism, mTOR signaling, and immune modulation. In non-small cell lung cancer, TIMP1 regulates mitochondrial STAT3 acetylation through the CD44-STAT3 axis, thereby altering mitochondrial metabolism and contributing to chemoresistance (67). In addition, tumor-derived TIMP1 has been reported to activate AKT/mTOR signaling in macrophages via CD63/β1-integrin, promoting M2-like macrophage polarization and the formation of an immunosuppressive pre-metastatic niche (68). Although these findings are not specific to renal cancer, they provide mechanistic clues suggesting that our prognostic signature may partly reflect mTOR-related metabolic reprogramming, mitochondrial dysfunction, ECM remodeling, and immune microenvironment regulation, which may collectively contribute to renal cancer progression and unfavorable outcomes.

Functional pathway enrichment revealed mechanisms underlying prognostic differences between risk groups, notably upregulating pathways such as organic anion transmembrane transporter activity, complement and coagulation cascades, and IL-6/JAK/STAT3 signaling. Organic anion transporters (OATs) are mainly part of the SLC superfamily, including the SLC22 family. These proteins recognize and transport organic anions through specific domains and mechanisms. Research shows that key SLC22 genes, particularly those in the OAT and OAT-related groups, demonstrate reduced expression as renal cancer progresses from healthy kidney function to advanced metastatic stages (69). The complement cascade and hypercoagulability create a self-perpetuating cycle through the formation of NETs. Within this feedback loop, they can induce tumor-promoting phenotypes in immune cells and shield tumor cells from immune attack, thereby facilitating tumor development, progression, and metastasis (70). The IL-6/JAK/STAT3 pathway influences tumor progression by promoting migration, invasion, and angiogenesis (71). In ccRCC, the IL-6/JAK/STAT3 signaling pathway is notably activated in the high-risk group. It interacts with pathways like epithelial-mesenchymal transition and PI3K-Akt signaling, forming a distinct gene expression profile linked to tumor aggressiveness (72). The high-risk subgroup exhibits hyperactivation of the IL-6/JAK/STAT3 pathway, disrupted transmembrane transport functions, and abnormal activation of the complement-coagulation system. These pathways collectively promote tumor proliferation, invasion, metastasis, and immune escape, ultimately leading to poorer prognosis in ccRCC patients.

In contrast, in the low-risk group, fundamental metabolic pathways such as fatty acid metabolism, oxidative phosphorylation and bile acid metabolism were significantly enriched. This suggests that the tumor microenvironment in low-risk ccRCC patients retains a relatively normal metabolic profile and has not undergone complete ‘metabolic reprogramming’ towards a malignant phenotype. Metabolic reprogramming is a key pathological feature of ccRCC and is closely associated with tumor progression. Previous transcriptomic, proteomic and metabolomic studies have confirmed that, during the progression of ccRCC, there is a widespread upregulation of oncogenic metabolic pathways such as aerobic glycolysis, the pentose phosphate pathway, fatty acid synthesis, glutamine and glutathione, accompanied by a downregulation of metabolic pathways such as the tricarboxylic acid cycle, fatty acid β-oxidation and oxidative phosphorylation, as well as their products (73-75). In this study, the sustained activation of multiple normal metabolic pathways in the low-risk group indicates that this subtype of ccRCC, which is less malignant, has not yet fully established tumor-specific abnormal metabolic programmes. Its metabolic homeostasis remains largely intact, and its potential for invasiveness and immunosuppression is relatively weak; this also provides a molecular explanation for the better prognosis observed in patients in the low-risk group.

We performed an in-depth analysis of the immune microenvironment to uncover new mechanisms driving ccRCC progression. The high-risk subgroup showed a significant enrichment of Tregs, whereas the low-risk subgroup was characterized by predominant infiltration of M1 macrophages. Previous research indicates that ccRCC is marked by both heightened immune infiltration and an immune exhaustion state. This phenomenon is mechanistically primarily associated with regulatory T cell-mediated immunosuppression (76). Tregs are a key type of immunosuppressive cells (77), whose substantial infiltration in tumor tissue has been confirmed to be significantly associated with tumor progression (78). M1 macrophages are key anti-tumor effector cells within the tumor immune microenvironment, inhibiting tumor growth through synergy with T cells and interferon signaling pathways (79,80).

The high-risk subgroup exhibited decreased expression of several HLA molecules, especially HLA-A class molecules. Reduced expression of HLA-A class molecules is associated with poor prognosis and disease progression in ccRCC (81), while increased HLA-A expression is linked to longer recurrence-free survival (82). In summary, these findings indicate that HLA-A constitutes a clinically significant biomarker for ccRCC. These discoveries provide a theoretical basis for the marked prognostic disparities observed between distinct risk groups.

In terms of clinical interpretation, the AUC values of our model were all above 0.70, indicating acceptable discrimination, which is consistent with the commonly used framework for transcriptomic machine learning models (83). This suggests that the model is not only statistically capable of distinguishing patients with different prognostic risks, but also has potential clinical utility. Specifically, MCAS may assist in patient risk stratification by identifying individuals at higher risk of adverse outcomes, thereby informing follow-up intensity, imaging surveillance, and subsequent treatment planning. From a clinical perspective, such a model may complement conventional pathological staging and clinicopathological features by helping clinicians recognize patients who require closer management at an earlier stage. Nevertheless, our findings are based on retrospective data, and the clinical value of the model still needs to be further validated in prospective, multicenter cohorts.

While the findings are promising, it is important to recognize the study’s limitations. Patient data mainly originated from public datasets, potentially leading to population selection bias due to limited racial diversity representation. To address this limitation, future studies will incorporate larger multi-ethnic cohorts to enhance the model’s generalizability and validity. Furthermore, our single-cell analyses were primarily derived from a single dataset (GSE171306). Consequently, the robustness and generalizability of the identified mast cell clusters across different datasets and patient populations remain unclear, as the results may be dependent on specific clustering parameters or batch effects. We also plan to integrate multiple independent single-cell datasets using advanced batch effect correction algorithms (e.g., Harmony, scVI) and validate the mast cell signature using cross-dataset mapping and automatic annotation tools (e.g., SingleR, CellTypist) to systematically assess the stability of these clusters. Moreover, this research currently lacks prospective clinical data support. Future investigations will prioritize clinical cohort studies centered on therapeutic response to evaluate differential treatment benefits across distinct risk subgroups. Concurrently, as this study primarily relies on bioinformatics analysis, it fails to elucidate the precise molecular mechanisms through which mast cell-associated angiogenesis genes regulate ccRCC progression. Future research will focus on collaborating with specialized laboratories for detailed molecular and cellular experiments. This will validate the biological functions of mast cell-related angiogenesis genes in ccRCC, further elucidate their molecular basis in mediating tumor progression, and provide experimental evidence for discovering novel therapeutic targets.

Conclusions

Our study is the first to characterize mast cell-associated angiogenesis features in ccRCC using an integrated single-cell and bulk transcriptomic approach, and to construct a robust prognostic model (MCAS) based on ten machine learning algorithms, with its risk-stratification ability validated across multiple independent cohorts. By integrating multidimensional analyses of functional pathway enrichment, immune infiltration, and somatic mutation landscapes, we provided an in-depth discussion of the heterogeneity affecting prognosis, revealing that activation of the TNF-α/NF-κB signaling pathway in mast cells and interactions such as SPP1-CD44 collectively contribute to the formation of an immunosuppressive microenvironment. Subsequently, the core genes TIMP1 and EMCN, identified by SurvSHAP(t) analysis, were recognized by us as promising prognostic markers and potential therapeutic targets for ccRCC. In conclusion, our study provides valuable insights by pointing the way for subsequent experimental research to deeply analyze the malignant progression of ccRCC and by offering new avenues for personalized treatment.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0248/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0248/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0248/prf

Funding: This study was supported by the National Natural Science Foundation of China (No. 82573240).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0248/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The First Affiliated Hospital of Nanjing Medical University (No. 2021-SR-430), and written informed consent was obtained from each patient prior to tissue collection.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin 2024;74:12-49. [Crossref] [PubMed]
2. Nezami BG, MacLennan GT. Clear Cell Renal Cell Carcinoma: A Comprehensive Review of its Histopathology, Genetics, and Differential Diagnosis. Int J Surg Pathol 2025;33:265-80. [Crossref] [PubMed]
3. Hsieh JJ, Purdue MP, Signoretti S, et al. Renal cell carcinoma. Nat Rev Dis Primers 2017;3:17009. [Crossref] [PubMed]
4. Nerich V, Hugues M, Paillard MJ, et al. Clinical impact of targeted therapies in patients with metastatic clear-cell renal cell carcinoma. Onco Targets Ther 2014;7:365-74. [Crossref] [PubMed]
5. Rini BI. Metastatic renal cell carcinoma: many treatment options, one patient. J Clin Oncol 2009;27:3225-34. [Crossref] [PubMed]
6. Makhov P, Joshi S, Ghatalia P, et al. Resistance to Systemic Therapies in Clear Cell Renal Cell Carcinoma: Mechanisms and Management Strategies. Mol Cancer Ther 2018;17:1355-64. [Crossref] [PubMed]
7. Young M, Jackson-Spence F, Beltran L, et al. Renal cell carcinoma. Lancet 2024;404:476-91. [Crossref] [PubMed]
8. Molina AM, Lin X, Korytowsky B, et al. Sunitinib objective response in metastatic renal cell carcinoma: analysis of 1059 patients treated on clinical trials. Eur J Cancer 2014;50:351-8. [Crossref] [PubMed]
9. Crusz SM, Tang YZ, Sarker SJ, et al. Heterogeneous response and progression patterns reveal phenotypic heterogeneity of tyrosine kinase inhibitor response in metastatic renal cell carcinoma. BMC Med 2016;14:185. [Crossref] [PubMed]
10. Paivandy A, Pejler G. Novel Strategies to Target Mast Cells in Disease. J Innate Immun 2021;13:131-47. [Crossref] [PubMed]
11. Kalesnikoff J, Galli SJ. New developments in mast cell biology. Nat Immunol 2008;9:1215-23. [Crossref] [PubMed]
12. Espinosa-Riquer ZP, Segura-Villalobos D, Ramírez-Moreno IG, et al. Signal Transduction Pathways Activated by Innate Immunity in Mast Cells: Translating Sensing of Changes into Specific Responses. Cells 2020;9:2411. [Crossref] [PubMed]
13. Englund A, Molin D, Enblad G, et al. The role of tumour-infiltrating eosinophils, mast cells and macrophages in Classical and Nodular Lymphocyte Predominant Hodgkin Lymphoma in children. Eur J Haematol 2016;97:430-8. [Crossref] [PubMed]
14. Antsiferova M, Martin C, Huber M, et al. Mast cells are dispensable for normal and activin-promoted wound healing and skin carcinogenesis. J Immunol 2013;191:6147-55. [Crossref] [PubMed]
15. Ribatti D, Tamma R, Vacca A. Mast Cells and Angiogenesis in Human Plasma Cell Malignancies. Int J Mol Sci 2019;20:481. [Crossref] [PubMed]
16. Solimando AG, Desantis V, Ribatti D. Mast Cells and Interleukins. Int J Mol Sci 2022;23:14004. [Crossref] [PubMed]
17. Abdel-Majid RM, Marshall JS. Prostaglandin E2 induces degranulation-independent production of vascular endothelial growth factor by human mast cells. J Immunol 2004;172:1227-36. [Crossref] [PubMed]
18. Boesiger J, Tsai M, Maurer M, et al. Mast cells can secrete vascular permeability factor/ vascular endothelial cell growth factor and exhibit enhanced release after immunoglobulin E-dependent upregulation of fc epsilon receptor I expression. J Exp Med 1998;188:1135-45. [Crossref] [PubMed]
19. Grützkau A, Krüger-Krasagakes S, Baumeister H, et al. Synthesis, storage, and release of vascular endothelial growth factor/vascular permeability factor (VEGF/VPF) by human mast cells: implications for the biological significance of VEGF206. Mol Biol Cell 1998;9:875-84. [Crossref] [PubMed]
20. Qu Z, Huang X, Ahmadi P, et al. Synthesis of basic fibroblast growth factor by murine mast cells. Regulation by transforming growth factor beta, tumor necrosis factor alpha, and stem cell factor. Int Arch Allergy Immunol 1998;115:47-54.
21. Ribatti D, Crivellato E, Candussio L, et al. Mast cells and their secretory granules are angiogenic in the chick embryo chorioallantoic membrane. Clin Exp Allergy 2001;31:602-8. [Crossref] [PubMed]
22. Metcalfe DD, Baram D, Mekori YA. Mast cells. Physiol Rev 1997;77:1033-79. [Crossref] [PubMed]
23. Blair RJ, Meng H, Marchese MJ, et al. Human mast cells stimulate vascular tube formation. Tryptase is a novel, potent angiogenic factor. J Clin Invest 1997;99:2691-700.
24. Koskivirta I, Rahkonen O, Mäyränpää M, et al. Tissue inhibitor of metalloproteinases 4 (TIMP4) is involved in inflammatory processes of human cardiovascular pathology. Histochem Cell Biol 2006;126:335-42. [Crossref] [PubMed]
25. Tanaka A, Yamane Y, Matsuda H. Mast cell MMP-9 production enhanced by bacterial lipopolysaccharide. J Vet Med Sci 2001;63:811-3. [Crossref] [PubMed]
26. Nakae S, Suto H, Kakurai M, et al. Mast cells enhance T cell activation: Importance of mast cell-derived TNF. Proc Natl Acad Sci U S A 2005;102:6467-72. [Crossref] [PubMed]
27. Guo Z, Jing X, Sun X, et al. Tumor angiogenesis and anti-angiogenic therapy. Chin Med J (Engl) 2024;137:2043-51. [Crossref] [PubMed]
28. Liu ZL, Chen HH, Zheng LL, et al. Angiogenic signaling pathways and anti-angiogenic therapy for cancer. Signal Transduct Target Ther 2023;8:198. [Crossref] [PubMed]
29. Lucarelli G, Loizzo D, Franzin R, et al. Metabolomic insights into pathophysiological mechanisms and biomarker discovery in clear cell renal cell carcinoma. Expert Rev Mol Diagn 2019;19:397-407. [Crossref] [PubMed]
30. Chen YY, Hu HH, Wang YN, et al. Metabolomics in renal cell carcinoma: From biomarker identification to pathomechanism insights. Arch Biochem Biophys 2020;695:108623. [Crossref] [PubMed]
31. Fotia G, Stellato M, Guadalupi V, et al. Current Status of Predictive Biomarker Development in Metastatic Renal Cell Carcinoma. Curr Oncol Rep 2023;25:671-7. [Crossref] [PubMed]
32. Dai H. Friend or foe? Multi-omics analysis unravels the complex identity of key pyroptosis factor CASP9 in clear cell renal cell carcinoma. Clin Cancer Bull 2025;4:18.
33. Han Y, Wang Y, Dong X, et al. TISCH2: expanded datasets and new tools for single-cell transcriptome analyses of the tumor microenvironment. Nucleic Acids Res 2023;51:D1425-31. [Crossref] [PubMed]
34. Liu Z, Liu L, Weng S, et al. Machine learning-based integration develops an immune-derived lncRNA signature for improving outcomes in colorectal cancer. Nat Commun 2022;13:816. [Crossref] [PubMed]
35. Dai H, Zhang X, Yin L, et al. Integrating machine learning and multi-omics analysis to explore Treg-associated programmed cell death features in clear cell renal cell carcinoma. Cancer Cell Int 2026;26:15. [Crossref] [PubMed]
36. Dai H, Wang X, Wu Q, et al. Integrating genetic crosstalk between atherosclerosis and lung adenocarcinoma to advance precision diagnosis and treatment. Cell Div 2025;20:27. [Crossref] [PubMed]
37. Dai H, Zhao K, Zhao Y, et al. Machine learning model in multi-omics perspective demystifies the prognostic significance of crotonylation heterogeneity in clear cell renal cell carcinoma. BMC Urol 2025;25:229. [Crossref] [PubMed]
38. Marquardt A, Solimando AG, Kerscher A, et al. Subgroup-Independent Mapping of Renal Cell Carcinoma-Machine Learning Reveals Prognostic Mitochondrial Gene Signature Beyond Histopathologic Boundaries. Front Oncol 2021;11:621278. [Crossref] [PubMed]
39. Hu F, Zeng W, Liu X. A Gene Signature of Survival Prediction for Kidney Renal Cell Carcinoma by Multi-Omic Data Analysis. Int J Mol Sci 2019;20:5720. [Crossref] [PubMed]
40. Filho VOC, Passos PRC, Noronha MM, et al. Development and Validation of an Intratumor Heterogeneity-Based Prognostic Model for Clear Cell Renal Cell Carcinoma. JCO Precis Oncol 2025;9:e2500709. [Crossref] [PubMed]
41. Tabibu S, Vinod PK, Jawahar CV. Pan-Renal Cell Carcinoma classification and survival prediction from histopathology images using deep learning. Sci Rep 2019;9:10509. [Crossref] [PubMed]
42. Wu Y, Zhou BP. TNF-alpha/NF-kappaB/Snail pathway in cancer cell migration and invasion. Br J Cancer 2010;102:639-44. [Crossref] [PubMed]
43. Bruyère C, Madonna S, Van Goietsenoven G, et al. JLK1486, a Bis 8-Hydroxyquinoline-Substituted Benzylamine, Displays Cytostatic Effects in Experimental Gliomas through MyT1 and STAT1 Activation and, to a Lesser Extent, PPARγ Activation. Transl Oncol 2011;4:126-37. [Crossref] [PubMed]
44. Peters I, Dubrowinskaja N, Tezval H, et al. Decreased mRNA expression of GATA1 and GATA2 is associated with tumor aggressiveness and poor outcome in clear cell renal cell carcinoma. Target Oncol 2015;10:267-75. [Crossref] [PubMed]
45. Feng Y, Shao X, Xu ZF, et al. SPP1 regulates tumor progression through modulation of signaling pathways and the tumor microenvironment. Discov Oncol 2025;17:115. [Crossref] [PubMed]
46. Qi F, Gao N, Li J, et al. A multidimensional recommendation framework for identifying biological targets to aid the diagnosis and treatment of liver metastasis in patients with colorectal cancer. Mol Cancer 2024;23:239. [Crossref] [PubMed]
47. Tan L, Huang Z, Chen Y, et al. Integration of Single-Cell Analysis and Bulk RNA Sequencing Data Using Multi-Level Attention Graph Neural Network for Precise Prognostic Stratification in Thyroid Cancer. Cancers (Basel) 2025;17:2411. [Crossref] [PubMed]
48. Huang S, Yan J, Liu W, et al. Progenitor CD8(+) T cells and hyper-Treg crosstalk: a driver of immune checkpoint inhibitor resistance in esophageal squamous cell carcinoma. Apoptosis 2026;31:44. [Crossref] [PubMed]
49. Cheng X, Zhao S, Du M, et al. Associations between plasma ANGPTL4 levels and the risk of 24 cancers: a prospective cohort study. Cytokine 2026;198:157089. [Crossref] [PubMed]
50. Ma B, Qin L, Sun Z, et al. The single-cell evolution trajectory presented different hypoxia heterogeneity to reveal the carcinogenesis of genes in clear cell renal cell carcinoma: Based on multiple omics and real experimental verification. Environ Toxicol 2024;39:869-81. [Crossref] [PubMed]
51. Zahr A, Alcaide P, Yang J, et al. Endomucin prevents leukocyte-endothelial cell adhesion and has a critical role under resting and inflammatory conditions. Nat Commun 2016;7:10363. [Crossref] [PubMed]
52. Santana-Bejarano MB, Reyes-Mata MP, Guerrero-García JJ, et al. NRP1 and GFAP Expression in the Medulloblastoma Microenvironment: Implications for Angiogenesis and Tumor Progression. Cancers (Basel) 2025;17:2417. [Crossref] [PubMed]
53. Feng Z, Qiao R, Ren Z, et al. Could CTSK and COL4A2 be specific biomarkers of poor prognosis for patients with gastric cancer in Asia?-a microarray analysis based on regional population. J Gastrointest Oncol 2020;11:386-401. [Crossref] [PubMed]
54. He L, Han W, Yue K, et al. COL4A2 enhances thyroid cancer cell proliferation through the AKT pathway. Oncol Res 2024;32:1467-78. [Crossref] [PubMed]
55. Liu Y, Zhang J, Chen Y, et al. The correlation and role analysis of COL4A1 and COL4A2 in hepatocarcinogenesis. Aging (Albany NY) 2020;12:204-23. [Crossref] [PubMed]
56. Cheng R, Cai XR, Ke K, et al. Notch4 inhibition suppresses invasion and vasculogenic mimicry formation of hepatocellular carcinoma cells. J Huazhong Univ Sci Technolog Med Sci 2017;37:719-25. [Crossref] [PubMed]
57. Qian C, Liu F, Ye B, et al. Notch4 promotes gastric cancer growth through activation of Wnt1/β-catenin signaling. Mol Cell Biochem 2015;401:165-74. [Crossref] [PubMed]
58. Shi SN, Qin X, Wang S, et al. Identification of potential novel differentially-expressed genes and their role in invasion and migration in renal cell carcinoma. Aging (Albany NY) 2020;12:9205-23. [Crossref] [PubMed]
59. Sicinski P, Donaher JL, Geng Y, et al. Cyclin D2 is an FSH-responsive gene involved in gonadal cell proliferation and oncogenesis. Nature 1996;384:470-4. [Crossref] [PubMed]
60. Mermelshtein A, Gerson A, Walfisch S, et al. Expression of D-type cyclins in colon cancer and in cell lines from colon carcinomas. Br J Cancer 2005;93:338-45. [Crossref] [PubMed]
61. Takano Y, Kato Y, Masuda M, et al. Cyclin D2, but not cyclin D1, overexpression closely correlates with gastric cancer progression and prognosis. J Pathol 1999;189:194-200. [Crossref] [PubMed]
62. Takano Y, Kato Y, van Diest PJ, et al. Cyclin D2 overexpression and lack of p27 correlate positively and cyclin E inversely with a poor prognosis in gastric cancer cases. Am J Pathol 2000;156:585-94. [Crossref] [PubMed]
63. Liu Y, Tan X, Liu W, et al. Follistatin-like protein 1 plays a tumor suppressor role in clear-cell renal cell carcinoma. Chin J Cancer 2018;37:2. [Crossref] [PubMed]
64. Wen YC, Lin YW, Ho KH, et al. The oncogenic ADAMTS1-VCAN-EGFR cyclic axis drives anoikis resistance and invasion in renal cell carcinoma. Cell Mol Biol Lett 2024;29:126. [Crossref] [PubMed]
65. Loh JJ, Li TW, Zhou L, et al. FSTL1 Secreted by Activated Fibroblasts Promotes Hepatocellular Carcinoma Metastasis and Stemness. Cancer Res 2021;81:5692-705. [Crossref] [PubMed]
66. Mou J, Huang Q, Liu X, et al. AML1-ETO and CCND2 overexpression cooperate to drive acute myeloid leukemia initiation and progression. J Leukoc Biol 2025;117:qiaf072. [Crossref] [PubMed]
67. Xiao W, Ahluwalia P, Wang L, et al. TIMP-1 Dependent Modulation of Metabolic Profiles Impacts Chemoresistance in NSCLC. Cells 2022;11:3036. [Crossref] [PubMed]
68. Liu J, Zhao L, Wang L, et al. Integrin-Mediated TIMP1 Signaling Reprograms Liver Macrophages and Accelerates Colorectal Cancer Metastasis. Cells 2025;15:29. [Crossref] [PubMed]
69. Whisenant TC, Nigam SK. Organic Anion Transporters (OAT) and Other SLC22 Transporters in Progression of Renal Cell Carcinoma. Cancers (Basel) 2022;14:4772. [Crossref] [PubMed]
70. Guglietta S, Rescigno M. Hypercoagulation and complement: Connected players in tumor development and metastases. Semin Immunol 2016;28:578-86. [Crossref] [PubMed]
71. Ni JS, Zheng H, Ou YL, et al. Surg Oncol 2020;34:113-20. [Crossref] [PubMed]
72. Li S, Wang C, Li F, et al. Noninvasive prediction of occult pT3a upstaging in localized ccRCC with radiogenomic insights and prognostic relevance. NPJ Precis Oncol 2026;10:104. [Crossref] [PubMed]
73. Comprehensive molecular characterization of clear cell renal cell carcinoma. Nature 2013;499:43-9.
74. Hakimi AA, Reznik E, Lee CH, et al. An Integrated Metabolic Atlas of Clear Cell Renal Cell Carcinoma. Cancer Cell 2016;29:104-16. [Crossref] [PubMed]
75. Wettersten HI, Hakimi AA, Morin D, et al. Grade-Dependent Metabolic Reprogramming in Kidney Cancer Revealed by Combined Proteomics and Metabolomics Analysis. Cancer Res 2015;75:2541-52. [Crossref] [PubMed]
76. Yao Y, Liu Y, Lu B, et al. Construction and validation of a regulatory T cells-based classification of renal cell carcinoma: an integrated bioinformatic analysis and clinical cohort study. Cell Oncol (Dordr) 2025;48:591-615. [Crossref] [PubMed]
77. Tay C, Tanaka A, Sakaguchi S. Tumor-infiltrating regulatory T cells as targets of cancer immunotherapy. Cancer Cell 2023;41:450-65. [Crossref] [PubMed]
78. Ohue Y, Nishikawa H, Regulatory T. Treg) cells in cancer: Can Treg cells be a new therapeutic target? Cancer Sci 2019;110:2080-9. [Crossref] [PubMed]
79. Cao J, Liu C. Mechanistic studies of tumor-associated macrophage immunotherapy. Front Immunol 2024;15:1476565. [Crossref] [PubMed]
80. Gunassekaran GR, Poongkavithai Vadevoo SM, Baek MC, et al. M1 macrophage exosomes engineered to foster M1 polarization and target the IL-4 receptor inhibit tumor growth by reprogramming tumor-associated macrophages into M1-like macrophages. Biomaterials 2021;278:121137. [Crossref] [PubMed]
81. Kitamura H, Honma I, Torigoe T, et al. Down-regulation of HLA class I antigen is an independent prognostic factor for clear cell renal cell carcinoma. J Urol 2007;177:1269-72.
82. Yuan J, Liu S, Yu Q, et al. Down-regulation of human leukocyte antigen class I (HLA-I) is associated with poor prognosis in patients with clear cell renal cell carcinoma. Acta Histochem 2013;115:470-4. [Crossref] [PubMed]
83. Costa Filho VO, Passos PRC. Precision over quantity: identifying trustworthy genomic risk stratification models. Einstein (Sao Paulo) 2025;23:eCE1726. [Crossref] [PubMed]