Prediction of Distant Metastases in Patients with Kidney Cancer Based on Gene Expression and Methylation Analysis

Abstract

Clear cell renal cell carcinoma (ccRCC) is the most common and aggressive histological type of cancer in this location. Distant metastases are present in approximately 30% of patients at the time of first examination. Therefore, the ability to predict the occurrence of metastases in patients at early stages of the disease is an urgent task aimed at personalized treatment. Samples of tumor and paired histologically normal kidney tissue from patients with metastatic and non-metastatic ccRCC were studied. Gene expression was analyzed using real-time PCR. The level of gene methylation was evaluated using bisulfite conversion followed by quantitative methylation-specific PCR. Two groups of genes were analyzed in this study. The first group includes genes whose expression is significantly reduced during metastasis: CA9, NDUFA4L2, EGLN3, and BHLHE41 (p < 0.001, ROC analysis). The second group includes microRNA genes: MIR125B-1, MIR137, MIR375, MIR193A, and MIR34B/C, whose increased methylation levels are associated with the development of distant metastases (p = 0.002 to <0.001, ROC analysis). Based on the data obtained, a combined panel of genes was formed to identify patients whose tumors have a high metastatic potential. The panel can estimate the probability of metastasis with an accuracy of up to 92%.

Correctly identifying the cells of origin is essential for tailoring treatment and understanding the emergence of cancer stem cells and late metastases

“…Thus, not only neuroendocrine carcinomas but also the more benign variant NETs often show metastases at an asymptomatic stage ( 75 , 76 ). This pattern is akin to CCRCCs ( 77 ), which most likely also are of neuroendocrine origin ( 21 , 22 , 33 ). Melanoma, originating from melanocytes derived from neural crest cells, also demonstrates early metastasis as well as cellular plasticity ( 78 ).…”

Interpretable AI models for predicting distant metastasis development based on genetic data: Kidney cancer example

“…Following the established practical application guidelines [10], the detection of three or more concurrent markers from the protein-coding gene panel was indicative of increased metastatic potential. This was particularly the case when gene expression levels dropped below ROC-analysis-defined cutoffs.…”

“…However, AI models based on previously described algorithms are non-interpretable and therefore cannot be directly used in clinical practice for making medical decisions. To enhance the interpretability of developed ML models, we applied a set of comparative rules as outlined in [10]. According to the guidelines, simultaneous occurrence of at least three markers from the protein-coding gene panel signifies a heightened risk of metastasis, especially when their expression levels fall below the threshold identified through ROCanalysis for each respective gene.…”

“…These determinations rely on specific ROC-analysis-derived threshold values for each gene. Furthermore, in instances where a combined gene panel is used, the identification of six or more abnormalitieswhether as decreased expression levels in protein-coding genes or increased methylation levels in miRNA genesserves as an indicator of significant metastatic potential in the tumor [10]. The predictive validity of these markers for metastasis was ascertained through evaluation of sensitivity and specificity metrics.…”

The Metastasis Pattern of Renal Cell Carcinoma Is Influenced by Histologic Subtype, Grade, and Sarcomatoid Differentiation

“…In addition to histologic subtypes, numerous studies have been conducted to predict the prognosis of RCC but the most commonly used are Fuhrman nuclear grades, World Health Organization/International Society of Urological Pathology (WHO/ISUP) nuclear grades, and the presence or absence of sarcomatous differentiation [13][14][15][16][17][18]. Much research has been conducted to predict the prognosis of renal cancer, including attempts to use genetic analysis and artificial intelligence to predict the prognosis [5,19,20].…”