Introduction
Renal cell carcinoma, as one of the most common tumors of the urinary system, accounts for 3% of all cancers, with the highest incidence in Western countries. In general, the incidence has increased by about 2% per year globally and in Europe. Clear cell renal cell carcinoma is the most common solid lesion in the kidney, accounting for about 90% of all renal malignancies. It includes different RCC subtypes with specific histopathological and genetic characteristics. The male to female ratio is 1.5: 1. The average age of patients with the disease tends to be younger (Sung et al.
2021). Several proven risk factors have been identified, including smoking, obesity and high blood pressure. These are considered clear risk factors for RCC (Cairns
2010). Clear cell renal cell carcinoma is the largest pathological subtype of renal carcinoma, accounting for more than 75%. It is usually found during surgery that the tumor incision surface is golden yellow, often accompanied by bleeding and necrosis (Cohen and McGovern
2005). Loss of the 3p chromosome and mutation of the von Hippel-Lindau (VHL) gene on the 3p25 chromosome are often found. The loss of VHL protein function contributes to the occurrence, progression and metastasis of tumors. The 3p locus contains at least four additional ccRCC tumor suppressor genes (UTX, JARID1C, SETD2, PBRM1) (Gossage et al.
2015; Thompson et al.
2018). ccRCC generally has a poorer prognosis than other classifications, but this difference disappears after adjustment for stage and grade. Therefore, ccRCC itself is heterogeneous and has a different prognosis (Jonasch et al.
2021). Based on this idea, clinicians continue to identify potential subtypes of renal cancer and develop its potential ability to guide prognosis and clinical therapy.
In addition to being the key organelles of energy generation in the cell, mitochondria also participate in the metabolic processes such as apoptosis, free radical production and lipid metabolism. Several studies have reported that abnormal mitochondrial function contributes to the pathology of many common diseases, including neurodegeneration, metabolic diseases, heart failure, ischemia–reperfusion injury, infections in protozoa and cancer (Annesley and Fisher
2019). Mitochondria are, therefore, an important drug target for these highly prevalent diseases. Several strategies aimed at therapeutically restoring mitochondrial function are emerging, and a handful of drugs have entered clinical trials. Mitochondria are maternally inherited and originated as organelles of symbiotic bacteria. They co-evolved with the host, so most mitochondrial proteins are nuclear encoded. However, mitochondria retain a small 16 kb DNA genome that encodes tRNAs, rRNA, and proteins essential for respiration. Cells have hundreds of mitochondria and can be wild-type or a mixture of wild-type and mutant types, a state known as heterogeneity. Mitochondria are important bioenergy and biosynthesis factories that are essential for normal cell function and human health (Nunnari and Suomalainen
2012). Otto Warburg proposed that mitochondrial respiratory defects were a potential basis for aerobic glycolysis and cancer, known as the Warburg effect (Vaupel et al.
2019). However, in fact, the Warburg effect can only be used as the basis for FDG-PET tumor imaging, and not all tumors have this aerobic glycolytic property (Czernin et al.
2013). Mitochondrial respiratory defects are not usually the cause of aerobic glycolysis, nor are they usually selected for during tumor evolution. In most cancers, it is carcinogenic driver mutations such as activation of K-ras, c-Myc, and phosphatidylinositol-3 (PI3) kinases or loss of phosphatase and tensin homologues and p53 that promote glycolysis, rather than mutations in the inactivated mitochondrial respiratory complex. Most cancers always preserve mitochondrial function, including respiration. Some tumors have high levels of oxidative phosphorylation, while others still retain mitochondrial respiration and other functions. Quantified by flux analysis in cultured cells, it was found that AKT conversion did not significantly affect respiration, while Ras conversion reduced respiration, but most ATP was still produced by oxidative phosphorylation. Functional tests of mitochondrial activity requirements in cancer have revealed their importance. The inactivation of the mitochondrial transcription factor Tfam depletes the mitochondria in tumor cells, thus impairing the growth of K-ras lung tumors. Depleting the mtDNA of tumor cells by poisoning mtDNA replication to produce r0 cells can significantly disrupt tumor development. In addition, selection for recovery of MTDNA-depleted r0 tumor growth was associated with horizontal transfer of the mitochondrial genome in host tissue and respiratory recovery (Kroemer and Pouyssegur
2008; Wallace
2012; Klein et al.
2020; Missiroli et al.
2020). These and other findings suggest that the role of mitochondria in cancer is not as simple as Warburg thought. Instead, they point to the importance of mitochondrial function for tumor growth. Therefore, we decided to identify potential subtypes of ccRCC and construct prognostic models based on the expression levels of regulatory genes of mitochondrial composition and function, and reveal the guiding significance of mitochondrial regulatory mechanisms for the clinical therapy of ccRCC.
Discussion
Renal cell carcinoma (RCC) is a kind of malignant tumor that is not sensitive to radiotherapy or chemotherapy. Currently, effective tumor therapy mainly relies on a variety of molecular targeted therapeutic drugs targeting vascular endothelial cell growth factor (VEGF), platelet-derived growth factor (PDGF) and mammalian target protein of rapamycin (mTOR) and immunotherapy targeting immune checkpoints such as PD-1 and PD-L1 (Yoon
2017; Chen et al.
2019; Braun et al.
2020; Lai et al.
2021; Qi et al.
2022). Renal cell carcinoma (RCC) is also considered to be a metabolic disease in many studies, mainly due to the presence of large amounts of carbohydrate, cholesterol, and fat metabolic reprogramming in renal cell (Wettersten et al.
2017). In normal cells, a large portion of glucose is metabolized to pyruvate through the TCA (Krebs) cycle in the mitochondria and oxidative phosphorylation, which is almost completely oxidized to CO2, resulting in a large amount of ATP (Tsvetkov et al.
2022). Pyruvate can be metabolized into lactic acid only when oxygen is restricted. Instead, most cancer cells convert most glucose into lactic acid, regardless of oxygen availability (the Warburg effect). In addition, tumor cells increase ROS production, thereby enhancing their antioxidant defenses to avoid oxidative damage and maintain ROS homeostasis. Because of this, key enzyme proteins and intermediates in the TCA cycle and oxidative phosphorylation have become potential targets for many cancer targeting drugs. Clear cell renal cell carcinoma is the most common pathological type of renal cell carcinoma. In 70–90% of patients with clear cell renal cell carcinoma, the VHL gene is inactivated, resulting in significantly increased hypoxia-inducing factor (HIF) levels in the cancer cells in the normoxic state (Zhang and Zhang
2018; Thompson et al.
2018). HIF can inhibit mitochondrial glucose oxidation by up-regulating the expression of pyruvate dehydrogenase kinase (PDK), a key protein kinase that regulates mitochondrial glucose oxidation metabolism, and then up-regulating the expression level of intracellular glycolytic enzyme. The inhibition of mitochondrial function in cancer cells in this anaerobic state can inhibit the apoptosis process of the mitochondrial pathway, reduce the levels of alpha-ketoglutarate, a circulating metabolite of tricarboxylate, and mitochondria-related ROS, and thus inhibit the function of P53. P53 has been proved to have tumor suppressor function, which can inhibit the expression of pyruvate dehydrogenase kinase 2 (PDK2), thus activating mitochondrial oxidative metabolism and promoting TCA cycle (Zhang et al.
2013; Harlander et al.
2017). In addition, p53 can induce mitochondrial GLS2 expression to enhance GSH synthesis and alpha- ketoglutarate, thereby promoting TCA cycling. P53 function is often impaired in tumors. Idasanutlin (RG7388), a small molecule that blocks the negative regulation of P53 in rat double microgene 2 (Mdm2), is currently in Phase III trials (Konopleva et al.
2020). It has been shown that RG7388 effectively reduces cell proliferation and induces p53-dependent pathways, cell cycle arrest and apoptosis, thereby inhibiting tumor growth. Meanwhile, ALRN-6924, a dual-targeted inhibitor of Mdm2/MdmX, has been tested in Phase I clinical trials (Saleh et al.
2021). The present results suggest that it stably activates p53-dependent transcription at the single-cell and single-molecule levels, and has good tolerance and antitumor activity in patients with solid tumors or lymphomas carrying wild-type TP53.
Studies have shown that OXPHOS can provide ATP for tumor proliferation. The electron transport chain (ETC) is an important component of OXPHOS, which consists of the complex I-IV, CoQ, and Cyt c and is required for tumor growth. As a major producer of proton gradients in ETC, complex I is a suitable target for the development of OXPHOS inhibitors. Early metformin and BAY87-2243 received much attention for their ability to inhibit complex I, but their low potency and severe side effects prevented their further development (Foretz et al.
2014; Mallik and Chowdhury
2018; Du et al.
2022). Petasin (PT) is a complex I inhibitor that mainly inhibits tumor growth in animal models with high efficiency and low toxicity (Heishima et al.
2021). In addition, the human epidermal growth factor receptor 2 (ERBB2) inhibitor mubritinib has anticancer effects by inhibiting complex I (Baccelli et al.
2019).
Multidrug resistance (MDR) in tumor cells is also related to mitochondria. MDR is one of the main causes of chemotherapy failure. The occurrence of MDR is associated with a variety of proteins on the cell membrane, such as the energy-dependent P-gp protein, which can expel chemotherapy drugs from the cell with the help of ATP. The ATP needed for P-gp to function comes mainly from the mitochondria. With high energy demand, mitochondria produce more ATP through glycolysis (Kopecka et al.
2020). Considering the multitude of drugs available for the clinical treatment of ccRCC, we decided to analyze the sensitivity of ccRCC samples with differential cuprotosis expression to these commonly used drugs, starting with common chemotherapy drugs and targeted therapies for kidney cancer. Therefore, all 12 drugs selected from the GDSC database are first and second-line treatments for kidney cancer. The IC50 prediction results for all drugs show significant differences between the high and low cuprotosis expression groups, with most drugs in the three groups showing a trend of increased or decreased IC50 values. It is gratifying to note that the first-line drugs in this subgroup: Sorafenib, Sunitinib, and pazopanib all show good resistance differences. This indicates that the classic ccRCC drugs used in clinical practice are related to cuprotosis. This preliminary result confirms our hypothesis from a clinical treatment perspective and suggests that the differential expression of cuprotosis still has guiding significance in the selection of currently used therapeutic drugs.
Based on these current research hotspots, we believe that mitochondria play an important role in the alteration of glucose and lipid metabolism in cancer. The key regulatory genes of mitochondria must play a key role in ccRCC, a type of cancer with obvious metabolic variation. MitoCarta3.0 which was published in 2021 is an emerging mitochondrial gene database. At present, the database contains a total of 1136 human mitochondrial pathway genes and 1140 mouse mitochondrial pathway genes, which is the most comprehensive gene bank for explaining mitochondrial function, structure and metabolism in the public database. To screen key genes broadly, we decided to include all human mitochondrial pathway genes in the initial study, rather than focusing on certain key pathways, to ensure maximum refinement of the final predictive model's ability to interpret mitochondria. The 7 mitochondrial pathway genes with large regulatory range ABCB6, ACSL1, ALDH4A1, ATP5MF, BIK, CPT1C and GCSH obtained by the final iterative LASSO screening also suggests that the final prediction model has a broad explanatory ability for mitochondrial function, rather than being limited to certain mitochondrial functional pathways. After comparison with Mitocarta database and literature review, we found that ABCB6, as a regulatory factor of ATP binding box, is mainly responsible for the transport of metal ions, cofactors and small molecules. ACSL1 is mainly responsible for regulating lipid metabolism and fatty acid oxidation balance (Quan et al.
2021), while ALDH4A1 and GCSH play important roles in amino acid metabolism, and are responsible for regulating the production and transport of proline and glycine, respectively (Lorenzo et al.
2021). ATP5MF is an important component in the regulation of oxidative phosphorylation: Complex 5 (Zhang et al.
2022). BIK is directly related to mitochondrial apoptosis (Chinnadurai et al.
2008). CPT1C is mainly responsible for carnitine transport and lipid metabolism (Fadó et al.
2023). It can be seen that almost most of the genes are related to the changes of mitochondria in cancer cells, and it can be considered that these genes have a potential regulatory relationship with the proliferation and metastasis of cancer cells. A more interesting phenomenon is that most of the genes that were screened were concentrated in cellular metabolic functions. For this phenomenon, we believe that there are several possibilities: 1.Gene regulation in cell metabolism: The mitochondrial pathway is closely related to cell metabolism. To maintain normal biological activities, cells must carry out various metabolic processes, such as energy production, organic synthesis, decomposition, and so on. Therefore, the enrichment of metabolism-related genes in the mitochondrial pathway is expected. 2. The main function of the mitochondria: The mitochondria is an organelle within the cell whose main function is to produce the energy required by the cell. Mitochondria provide energy through ATP produced during cellular respiration. ATP is involved in many metabolic pathways in the cell, including fatty acid metabolism, glycolysis, ketone body synthesis, etc. Therefore, gene functions associated with these metabolic processes may be highly enriched in the mitochondrial pathway. 3. The relationship between mitochondrial pathway and metabolic diseases: The mitochondrial pathway is closely related to the development and progression of many metabolic diseases. In these diseases, gene mutations or functional abnormalities related to energy metabolism and cellular respiratory function may lead to the occurrence of the disease. Therefore, genes that play an important role in metabolic function may receive more attention in the functional enrichment of mitochondrial pathways.
At present, the construction of prediction models based on gene screening is based on model variables such as forest map, nomogram and so on. Here, we hope to find different model display methods to interpret the model we constructed from a new perspective. At present, the large-scale development of machine learning has provided great help for the construction of clinical prediction models. In this paper, we use cox regression and random survival forest, a well-known variant of random forest in machine learning algorithms, to compare two different model construction methods, and use the survex R package to quantify the model (Taylor
2011). The results show that the machine learning random forest algorithm is ahead of the traditional cox regression algorithm in many aspects, and thanks to the help of survex, the former 's complex internal algorithm can be explained globally and locally through their separate variables, which is convenient for more clinicians to understand the significance of the prediction model. Survival analysis models typically output functions (survival or risk functions) rather than point predictions like regression and classification models. This makes interpreting these models a challenging task, especially with Shapley values. To do this, we apply SurvSHAP, a new model agnostic algorithm, to interpret survival models that predict survival curves. The algorithm is based on finding patterns in the predictive survival curve that will identify significantly different survival behaviors, and utilizing proxy models and SHAP methods to explain these different survival behaviors. Experiments on both synthetic and real datasets show that SurvSHAP is able to capture the underlying factors of survival patterns. In addition, the SurvSHAP results of the Cox proportional risk model are compared with the weights of the model to show that we provide a more realistic overall explanation and a more refined explanation of subpopulations. Non-linear machine learning survival models using SurvSHAP can better model the data and provide better interpretations compared to linear models.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.