Kidney cancer survival prognosis model based on machine learning combination construction of cell death-immune characteristics and small molecule drug screening method

Abstract

The invention discloses a kidney cancer survival prognosis model with cell death-immune characteristics constructed based on machine learning combination and a small molecule drug screening method, the method adopts 101 machine learning algorithm combinations to analyze gene expression data of KIRC patients, screens out genes closely related to two PCD modes, namely Anikis and LDCD, and finally identifies two key genes, namely HMOX1 and PIK3CG. Finally, it is found that SB505124, Carnosol and PROTAC can be used as an immunotherapeutic agent of KIRC.

Description

Technical Field

[0001] The present invention belongs to the technical field of renal cancer survival prognosis, and specifically is a renal cancer survival prognosis model and a small molecule drug screening method based on a combination of machine learning to construct a cell death-immune feature. Background Art

[0002] Kidney cancer is one of the most common malignancies worldwide, with a steadily increasing incidence and mortality rate. Renal cell carcinoma accounts for up to 90% of all kidney cancer cases, and clear cell renal carcinoma (KIRC) accounts for 70%. The prognosis of KIRC is usually grim, with a recurrence rate of up to 30% for those who undergo surgical treatment. Adjuvant therapy after surgery for localized renal cell carcinoma has always been a challenging research area with limited success.

[0003] Malignancies of the urinary and reproductive systems, including prostate cancer and bladder cancer, have traditionally been treated with radiotherapy and chemotherapy. However, KIRC is unique due to its resistance to conventional chemotherapeutic drugs. KIRC is one of the highly immune-infiltrated cancers and shows low sensitivity to traditional radiotherapy and chemotherapy. Although immunotherapy is effective in some patients, its universality is limited. Studies have revealed that programmed cell death (PCD) can participate in the regulation of anti-tumor immune responses by regulating the activity of immune cells.

[0004] In the TME, immune cells play a key role in tumor growth, invasion, migration, and regulation of anti-cancer immunity. In this case, immunotherapy becomes a crucial component of its treatment strategy. However, the combined use of immunotherapy and immune checkpoint inhibitors only shows positive effects in some patient groups. Existing studies believe that differences in patient response to immunotherapy may be related to the tumor's immune escape mechanism, which results in some patients being unable to benefit from immune checkpoint blockade therapy. Therefore, in-depth research on tumor immune escape mechanisms is crucial to the development of more effective treatment strategies.

[0005] PCD is an orderly process of cellular self-destruction, which is precisely regulated at the genetic level and performs the internal cleaning tasks of cells through a series of specific molecular mechanisms to remove damaged or no longer needed cells to maintain the health and function of tissues. PCD covers a series of different cell death pathways, including apoptosis, pyroptosis, ferroptosis, autophagy, necroptosis, cuproptosis, parthanatos, entotic cell death, netotic cell death, lysosome-dependent cell death (LDCD), alkaliptosis, oxeiptosis, NETosis, immunogenic cell death, Anoikis, paraptosis, methuosis, entosis.

[0006] PCD plays a very critical role in maintaining normal body activities and human health. Notably, PCD also plays an important role in the development of cancer. Studies have shown that PCD can regulate the enrichment of effector immune cells or regulatory immune cells, thereby participating in regulating the anti-tumor immunity of TME. Existing studies have pointed out that there is a complex association between PCD and KIRC immunotherapy response and tolerance. For example, in KIRC, the prognostic risk model associated with copper death revealed that patients in the high-risk group were more sensitive to immunotherapy. In addition, the prognostic risk model constructed based on five genes related to immunogenic cell death can not only map the tumor immune microenvironment of KIRC patients, but also help promote the implementation of personalized treatment and provide potential new targets for immunotherapy. These findings emphasize the importance of intervention strategies targeting PCD in KIRC immunotherapy. Despite this, comprehensive research on the role of PCD in immunotherapy responsiveness in KIRC patients is still insufficient. Summary of the invention

[0007] The purpose of the present invention is to provide a renal cancer survival prognosis model and a small molecule drug screening method based on a combination of machine learning and cell death-immune characteristics. After risk assessment of KIRC patients is performed using the constructed renal cancer survival prognosis model of cell death-immune characteristics, KIRC patients are divided into a low-risk group and a high-risk group, and SB505124, Carnosol and PROTAC are used as immunotherapeutic agents for the treatment of the KIRC high-risk prognosis group, which can provide targeted personalized treatment for KIRC patients.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] A risk assessment model for KIRC patients was developed based on the gene expression data of Anoikis and LDCD patterns. The predictive performance of the model was evaluated using C-index and ROC curve, and risk assessment was performed using a nomogram combined with clinical factors.

[0010] The method for constructing a risk assessment model for KIRC patients comprises the following steps:

[0011] (1) Collect clinical pathological information and gene expression data of KIRC patients: collect 18 programmed cell death modes and their key regulatory genes; download TCGA-KIRC cohort data and RNA sequencing data from the TCGA database; collect clinical pathological information and microarray expression analysis data of the KIRC cohort from the GEO database;

[0012] (2) Identification of differentially expressed genes in KIRC: The “edgeR” package was used to screen differentially expressed genes, with the criteria of adjusted P value < 0.05 and |log2FC| > 1;

[0013] (3) Construction of prognostic models based on comprehensive methods of machine learning:

[0014] Risk score models were constructed in the TCGA-KIRC cohort using a combination of multiple algorithms and leave-one-out cross-validation, and GSE22541 was used for cross-validation. The Harrell consistency index C-index of each model was calculated, and the average C-index or test set C-index>0.7 was selected. We further evaluated the diagnostic efficacy of PCD in the cohort by calculating the area under the ROC curve.

[0015] (4) To verify the value of different PCD patterns as independent prognostic indicators for KIRC patients, univariate and multivariate Cox regression analyses were performed. Based on the TCGA-KIRC cohort and the GSE22541 cohort, the prognostic nomograms were drawn using the R packages "rms" and "replot", and the risk assessment model for KIRC patients was constructed based on the prognostic nomograms.

[0016] The machine learning algorithms include, but are not limited to, random survival forest, elastic net, Lasso, Ridge, stepwise Cox, CoxBoost, partial least squares regression, supervised principal component, generalized boosted regression modeling, and survival support vector machine.

[0017] A method for evaluating the immunotherapy responsiveness of patients with renal clear cell carcinoma, using the risk assessment model for KIRC patients described in claim 1 or the model constructed by the method for constructing the risk assessment model for KIRC patients, specifically comprising the following steps:

[0018] (1) Collect clinicopathological information and gene expression data of KIRC patients;

[0019] (2) inputting the collected clinical pathological information and gene expression data into the risk assessment model for KIRC patients, and scoring using the risk assessment model for KIRC patients;

[0020] (3) Assessing the patient's risk level according to the scoring results of the risk assessment model for KIRC patients.

[0021] Application of HMOX1, PIK3CG, Tfh, and HLA-E as biomarkers for KIRC.

[0022] A biomarker panel for prognosis monitoring of KIRC patients, including HMOX1, PIK3CG, Tfh cell infiltration level, and HLA-E expression level.

[0023] Application of SB505124, Carnosol, and PROTAC as immunotherapeutic agents in the preparation of drugs for the treatment of KIRC.

[0024] A drug combination for treating KIRC, comprising: a TGF-β type I receptor inhibitor, a compound that enhances HMOX1 expression, and a compound that specifically degrades PIK3CG.

[0025] The beneficial effects obtained by the present invention are:

[0026] This invention is the first to conduct an in-depth exploration of 18 PCD types and their related genes in KIRC. An accurate risk assessment model was constructed to identify two PCD patterns, Anoikis and LDCD, which are closely related to the prognosis of KIRC patients, and patients in the high-risk group have a poor prognosis. Immune cell analysis showed that Tfh expression was upregulated in both PCD patterns. Immune checkpoint analysis revealed that HLA-E expression was enhanced in both patterns. TTN and MUC16 gene mutations occurred frequently in the Anoikis pattern, while in the LDCD pattern, although the mutation rate was higher in the high-risk group, there was no significant difference in tumor mutation load. GSEA analysis showed that the primary immunodeficiency pathway was significantly enriched in the Anoikis high-risk group, and the spliceosomal tri-snrnp complex assembly pathway was significantly enriched in the high-risk group in the LDCD pattern. Drug sensitivity analysis showed that SB505124 was sensitive to both PCD patterns. HMOX1 and PIK3CG were identified as genes common to the two key PCD patterns, and molecular docking analysis verified the stable binding between Carnosol and HMOX1, and PROTAC and PIK3CG.

[0027] It was revealed that Anoikis and LDCD can synergistically regulate Tfh cell infiltration and the expression level of immune checkpoint HLA-E by enhancing the expression of HMOX1 and PIK3CG, thereby weakening the immune response effect of KIRC patients. In addition, SB505124, Carnosol and PROTAC showed important potential as potential immunotherapeutic agents, providing new strategies and directions for the personalized treatment of KIRC.

[0028] Through detailed analysis of large-scale data, we have established a strong correlation between PCD and tumor patient prognosis, immunotherapy response, and targeted drug screening. This breakthrough discovery not only provides a new perspective for understanding the complex biological characteristics of KIRC, but also lays a solid foundation for the development of new treatment strategies and biomarkers. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart of the present invention.

[0030] Figure 2A It is an essential regulatory gene for 18 PCD patterns; Figure 2B These are 12,788 genes differentially expressed between normal and tumor tissues in the TCGA-KIRC cohort; Figure 2C It is the intersection gene of 18 PCD patterns and 12788 differentially expressed genes in the TCGA-KIRC cohort.

[0031] Figure 3 shows the results of univariate COX regression analysis of Anoikis, Apoptosis, Autophagy, LDCD, Ferroptosis, Necroptosis, Immunogenic cell death, NETosis, Cuproptosis, Paraptosis, and Pyroptosis.

[0032] Figure 4A , Figure 4B It is the C-index calculation and evaluation results of 11 PCDs in TCGA-KIRC and GSE22541 datasets; Figure 4C , Figure 4D It is the ROC curve calculation and evaluation result of anoikis and LDCD models; Figure 4E , Figure 4F It is the predictive ability and comparison chart of the Risk score model of anoikis and LDCD and traditional clinical features such as clinical stage; Figure 4G , Figure 4H This is a comparison of the prediction capabilities of anoikis and LDCD.

[0033] Figure 5A , Figure 5B is the predictive performance of the nomogram model of anoikis; Figure 5C , Figure 5D is the prediction performance of the nomogram model of LDCD.

[0034] Fig. 6A -L are the survival analysis results of immune cells T cells regulatory (Tregs), T cells follicular helper (Tfh), T cells CD8, T cells CD4 memory resting, T cells CD4memory activated, Plasma cells, Monocytes, Mast cells resting, Mast cells activated, Macrophages M2, Macrophages M0, and Dendritic cells resting in the TCGA-KIRC cohort; Figure 6M It is an inflammatory immune cell subtype of anoikis; Figure 6N It is the inflammatory immune cell subtype of LDCD.

[0035] Fig. 7A This is the result of immune cell infiltration analysis in the high- and low-risk groups of anoikis. Figure 7B is the result of immune cell infiltration analysis in the high- and low-risk groups of LDCD; Fig. 7E , Figure 7F These are the results of immune cell function enrichment analysis in the high- and low-risk groups of anoikis and LDCD, respectively.

[0036] Fig. 8A , Figure 8B They are clinical information under two PCD modes: anoikis and LDCD; Figure 8C , Fig.8D It is the somatic mutation analysis of high- and low-risk groups in anoikis; Fig. 8E , Fig.8F It is an analysis of tumor mutation burden in the high- and low-risk groups of anoikis; Figure 8G , Figure 8H It is the somatic mutation analysis of high- and low-risk groups in LDCD; Figure 8I , Figure 8J is the analysis of tumor mutation burden in the high- and low-risk groups of LDCD; Figure 8I , Figure 8J is the analysis of tumor mutation burden in the high- and low-risk groups of LDCD; Figure 8K , Figure 8L is the expression level of immune checkpoint-related genes in anoikis and LDCD.

[0037] Fig. 9 AI is an analysis of immune checkpoints associated with high- and low-risk groups of anoikis; Fig. 9 JR is the analysis of immune checkpoints related to LDCD high and low risk groups.

[0038] Fig. 10A , Fig. 10B This is the GO analysis graph of Anoikis high and low risk groups; Fig. 10C , Fig. 10D This is the KEGG analysis diagram of Anoikis high and low risk groups; Fig.10E , Fig.10F This is the GO analysis diagram of the high- and low-risk groups of LDCD; Figure 10G This is the KEGG analysis chart of the LDCD low-risk group; Fig. 10H It is the intersection of Anoikis-related genes, LDCD-related genes, and differentially expressed genes (DEGs) in KIRC patients; Fig.10I -J is the ratio of immune stromal components in the tumor microenvironment between Anoikis and LDCD high- and low-risk groups.

[0039] Fig.11A -E is the drug sensitivity of Anoikis high-risk and low-risk groups to dihydrorotenone, OF-1, ibrutinib, SB505124 and P22077; Fig.11F is the drug sensitivity of LDCD high- and low-risk groups in SB505124; Fig.11G -J Schematic diagram of molecular docking (black circles represent binding pockets, and dashed lines of different colors represent potential weak bond interactions between ligands and amino acid sites). DETAILED DESCRIPTION

[0040] The present invention is described below in conjunction with the accompanying drawings. Our research workflow is as follows Figure 1 shown.

[0041] Data collection: First, we collected 1964 essential regulatory genes containing 18 PCD patterns from the literature ( Figure 2A). These genes included 580 genes related to apoptosis, 52 genes related to pyroptosis, 88 genes related to ferroptosis, 367 genes related to autophagy, 101 genes related to necroptosis, 19 genes related to copper death, 9 genes related to partial apoptosis, 15 genes related to enterocyte death, 8 genes related to reticular cell death, 220 genes related to LDCD, 7 genes related to alkali-soluble death, 5 genes related to oxidative stress-induced cell death, 24 genes related to NETosis, 34 genes related to immunogenic cell death, 338 genes related to anchorless death, 66 genes related to paraapoptosis, 8 genes related to methionine-dependent death, and 23 phagocytic cell death genes.

[0042] RNA sequencing data of 614 KIRC samples and their corresponding clinicopathological information were downloaded from the TCGA database (https: / / portal.gdc.cancer.gov / projects / TCGA-KIRC). Differentially expressed genes (DEGs) were screened using the “edgeR” package, with the criteria of adjusted P value < 0.05 and |log2FC| > 1.

[0043] The clinicopathological information and microarray expression analysis data of KIRC cohort GSE22541 were downloaded from the GEO database (GEO Accession viewer (nih.gov)).

[0044] Then, differential analysis was performed on the TCGA-KIRC cohort to identify 12,788 genes that showed differential expression between normal and tumor tissues ( Figure 2B ).

[0045] Then, the intersection of the 18 PCD pattern-related genes and the differentially expressed genes was taken, and it was found that there were 157 intersection genes with Apoptosis, 120 intersection genes with Anoikis, 65 intersection genes with Autophagy, 55 intersection genes with LDCD, 26 intersection genes with Necroptosis, 22 intersection genes with Ferroptosis, 20 intersection genes with Pyroptosis, 16 intersection genes with Paraptosis, 14 intersection genes with Immunogenic cell death, 8 intersection genes with NETosis, 6 intersection genes with Cuproptosis, 4 intersection genes with Entosis, 3 intersection genes with Entoticcelldeath, 3 intersection genes with Methuosis, 2 intersection genes with Netotic cell death, 1 intersection gene with Alkaliptosis, 1 intersection gene with Oxeiptosis, and no intersection gene with Parthanatos ( Figure 2C ).

[0046] Eleven cell death patterns with more than 5 overlapping genes were included. Univariate COX regression analysis revealed that most PCD patterns were risk factors, while most LDCD and cuproptosis-related genes were protective factors ( Figure 3A -K).

[0047] A comprehensive approach based on machine learning to build a prognostic model

[0048] To establish a high-accuracy and stable consensus on PCD-related genes, we integrated 10 machine learning algorithms and 101 algorithm combinations. These algorithms include random survival forest (RSF), elastic network (Enet), Lasso, Ridge, stepwise Cox, CoxBoost, partial least squares regression (plsRcox), supervised principal component (SuperPC), generalized boosted regression modeling (GBM), and survival support vector machine (survival-SVM). The specific algorithm combinations in 101 are shown in Table 1.

[0049] Table 1 Specific combinations of 101 algorithm combinations

[0050]

[0051] The steps of generating feature signature include:

[0052] (a) A prediction model was constructed in the TCGA-KIRC cohort using 101 algorithm combinations and leave-one-out cross-validation.

[0053] (b) All models were further cross-validated using GSE22541;

[0054] (c) The Harrell consistency index (C-index) of each model was calculated, and the models with an average C-index or a test set C-index > 0.7 were considered excellent. The results showed that the models of Anoikis, ferroptosis, LDCD, and autophage showed the highest average C-index or excellent performance with a C-index of more than 0.7 in the test set (see Figure 4A -B).

[0055] We divided KIRC patients into high-risk and low-risk groups using the median scores of the respective cohorts.

[0056] Nomogram construction and evaluation

[0057] To verify the value of different PCD patterns as independent prognostic indicators for KIRC patients, univariate and multivariate Cox regression analyses were performed. Based on the TCGA-KIRC cohort and the GSE22541 cohort, the prognostic nomograms were developed using the R packages “rms” and “replot”, and their performance was evaluated by the receiver operating characteristic (ROC) curve. The results showed that the models of Anoikis and LDCD ranked high (see Figure 4C -F). Through multivariate analysis and C-index calculation of clinical characteristics, we found that Anoikis's Riskscore model surpassed traditional clinical characteristics such as age, gender, tumor grade and clinical stage in predictive ability (see Figure 4E ). Although the C-index of the Risk score model for LDCD is slightly lower than that of the clinical stage, its predictive stability is more prominent.

[0058] In order to further enhance the clinical application value of the model, we combined the Risk score with clinical factors such as gender and age and developed a nomogram as a new prognostic model. The prediction performance of the Anoikis nomogram model reached 0.781, while the prediction performance of the LDCD nomogram model reached 0.770 ( Figure 5A -B). These findings provide new perspectives and tools for the prognostic assessment of KIRC patients.

[0059] Immune infiltration analysis

[0060] The survival analysis of immune cells was performed using the R language "survival" package. The IOBR package combined with the CIBERSORT algorithm was used to conduct a comprehensive immune cell infiltration analysis of the TCGA-KIRC dataset samples. The Pearson correlation analysis method was used to explore the relationship between differentially expressed genes and specific immune cell subsets. The immune gene set immune.gmt was used to perform functional enrichment analysis of immune cells.

[0061] The results showed that when the infiltration levels of T cells CD4 memory resting, Monocytes, Mast cells resting, Macrophages M2, and Dendritic cells resting were high, the overall survival prognosis of patients showed a positive improvement. At the same time, when the infiltration levels of Plasma cells, T cells CD8, T cells CD4 memory activated, Tcells follicular helper, T cells regulatory (Tregs), Macrophages M0, MacrophagesM2, Dendritic cells resting, and Mast cells activated were high, the prognosis of patients was worse. We further analyzed the death patterns of Anoikis and LDCD, and the inflammatory immune cell subtypes in the low-risk group were higher than those in the high-risk group. In the low-risk group, the expression level of inflammatory immune cell subtypes was higher than that in the high-risk group.

[0062] In Anoikis, the high-risk group showed high expression of specific immune cell subsets. Specifically, T cells CD4 memory resting, T cells CD4 memory activated, T cells follicular helper, T cells regulatory (Tregs), Macrophages M0, and Mast cells activated were more common in these patients. Meanwhile, the expression levels of Dendritic cells resting and Mast cells resting were relatively low in the high-risk group.

[0063] In LDCD, we observed that Plasma cells, T cells CD8, T cells follicular helper, and T cells regulatory (Tregs) were expressed at higher levels in the high-risk group. In contrast, T cells CD4 memory resting, Macrophages M2, Mast cells resting, and Mast cells activated were expressed at lower levels in the high-risk group. To further validate these findings, we analyzed the correlation of immune cells in the two PCDs using different software algorithms. Figure 7C -D shows the results of these analyses, which provide us with a quantitative perspective on the expression patterns of immune cell subsets and may reveal the interactions between different immune cell subsets and their impact on the tumor microenvironment. The results of immune cell function analysis showed that the expression of immune checkpoints in the high-risk group was generally higher in both Anoikis and LDCD ( Fig. 7E -F).

[0064] Fig. 8A -B heatmap shows the clinical information of patients in the two PCD patterns. In the TCGA-KIRC cohort, we explored the key biological differences between the different PCD patterns. In the somatic mutation analysis of the Anoikis death pattern, we observed that the high-risk group ( Figure 8C , 53.21%) was significantly higher than that in the low-risk group ( Fig.8D , 41.09%). In particular, the mutation rates of TTN and MUC16 genes were significantly higher in both high-risk and low-risk groups. In addition, the tumor mutation burden in the high-risk group was also higher than that in the low-risk group ( Fig. 8E In the LDCD model, although the high-risk group ( Fig.8F , 82.66%) had a slightly higher overall mutation rate than the low-risk group ( Figure 8G , 81.62%), but there was no significant difference in tumor mutation burden between the two groups ( Figure 8H ).

[0065] Furthermore, we analyzed the relationship between the two PCD patterns and the expression of immune checkpoint-related genes. In the Anoikis pattern, the expression levels of most immune checkpoint-related genes in the high-risk group were significantly increased ( Figure 8I ). Similarly, in the LDCD model, immune checkpoint-related genes were also highly expressed in the high-risk group ( Figure 8J ).

[0066] To gain a deeper understanding of the immune escape mechanism of immunotherapy, we conducted further analysis. The results showed that in Anoikis, microsatellite instability, TIDE score, immune dysfunction, responder status, CD8+ T cells, CD274 (PD-L1) expression, and TAM M2 phenotype were higher in the high-risk group, while immune rejection and CAF (cancer-associated fibroblasts) were not statistically significant in patients with high-risk group ( Fig. 9 In LDCD, a similar pattern was observed, with higher microsatellite instability, TIDE score, immune dysfunction, responder status, CD8+ T cells, CD274 (PD-L1) expression, and TAMM2 phenotype in the high-risk group, while immune rejection, CAF, and CD274 (PD-L1) expression were not statistically significant in the high-risk group ( Fig. 9 JR).

[0067] Tumor Immune Dysfunction and Exclusion (TIDE) is a computational framework http: / / tide.dfci.harvard.edu / that assesses the potential of tumors to escape immune function by evaluating the gene expression profile of cancer samples. The TIDE score consists of two components, immune dysfunction and immune exclusion, and can be used as a surrogate biomarker to predict the response to immune checkpoint blockade in various cancer types. The empirical list of ICGs was obtained based on previous studies.

[0068] We further performed GSEA enrichment analysis on the two key PCD patterns in KIRC. We obtained version 3.0 software from the GSEA official website and divided the samples into high expression (>50%) and low expression (<50%) groups based on the hub gene expression. We downloaded the c2.cp.kegg.v7.4.symbols.gmt gene set for analyzing biological pathways and molecular mechanisms. Through GSEA analysis, the minimum gene set was set to 5 and the maximum was set to 5000, and a thousand simulated samplings were performed. Pathways with P<0.05 and NES> thresholds were screened out, classified, and visualized.

[0069] In the context of Anoikis, GO analysis results highlighted that the high-risk group had significantly enriched genes in the Immunoglobulin complex (see Fig. 10A ). Meanwhile, the gene expression of the low-risk group was significantly enriched in the apical part of cell (see Fig. 10B Further KEGG analysis revealed that the high-risk group was significantly enriched in primary immunodeficiency (see Fig. 10C ), in contrast, the low-risk group was significantly enriched in the ppar signaling pathway (see Fig. 10D ).

[0070] In the LDCD model, GO analysis revealed a significant enrichment of the high-risk group in the spliceosomal tri-snrnpcomplex assembly process (see Fig.10E The low-risk group showed significant enrichment in the inorganic anion transport process (see Fig.10F ). KEGG results showed that the high-risk group had no significant enrichment, while the low-risk group had a continuous enrichment in the metabolism of xenobiotics by cytochrome p450 drug metabolism pathway (see Figure 10G ).

[0071] By using the ESTIMATE algorithm in the estimate package in R language to estimate the ratio of immune stromal components in the tumor microenvironment between high-risk and low-risk groups in the two PCD patterns, we obtained three scores: Immune Score, Stromal Score, and ESTIMATE Score. The high-risk group was positively correlated with immune and stromal components in all these scores (see Fig.10I -J). Given that Anoikis and LDCD showed significant similarities in immune-related analysis, enrichment analysis, and immune matrix components, we conducted a comprehensive analysis of Anoikis-related genes, LDCD-related genes, and differentially expressed genes (DEGs) in KIRC patients and found two key intersection genes, HMOX1 and PIK3CG (see Fig. 10H ), which may play a crucial role in the immune response and therapeutic response of KIRC.

[0072] Since the prognosis of the high-risk group was significantly worse than that of the low-risk group, we performed drug analysis on the low- and high-risk groups of the two PCDs to predict drugs with higher sensitivity in the high-risk group. The results showed that in the high-risk group of Anoikis, dihydrorotenone, OF-1, ibrutinib, SB505124, and P22077 showed higher sensitivity ( Fig.11A -E). In the high-risk group for LDCD, only SB505124 showed higher sensitivity ( Fig.11F Interestingly, we found that SB505124 had high sensitivity in the high-risk groups for both PCDs.

[0073] Since HMOX1 and PIK3CG are two key genes for PCD, we performed molecular docking analysis on HMOX1 and PIK3CG. Through literature search, it was found that Carnosol exerts its antioxidant and protective functions of microvascular endothelial cells by enhancing the expression of HMOX1 and activating the Nrf2 signaling pathway. Therefore, we verified the binding ability of Carnosol to HMOX1 and confirmed that Carnosol can form a stable binding with the active site of HMOX1. At the same time, we also explored the molecular targeting strategy of PIK3CG. According to the literature, a research team has developed a PROTAC heterobifunctional molecule that can specifically degrade PIK3CG, thereby blocking Our molecular docking validation showed that PIK3CG can form a stable binding with the active site of PROTAC molecules ( Fig.11I -J), which provides a new strategy for PIK3CG as a therapeutic target.

[0074] Somatic mutation analysis: Somatic mutation data downloaded from the TCGA database were analyzed using the “maftools” package in R. Tumor mutation burden (TMB) was calculated as the number of nonsynonymous somatic mutations per megabase of cancer samples.

[0075] Tumor microenvironment scoring: The ESTIMATE program was used to analyze the infiltration levels of immune cells and stromal cells in KIRC. The data were derived from the TCGA database, and the stromal cell score, immune cell score and comprehensive score were calculated.

[0076] Drug sensitivity analysis: The anti-tumor drug response of KIRC patients was predicted based on the Genomic Drug Sensitivity of Cancer (GDSC) database, which was reflected by the half-maximal inhibitory concentration (IC50). The differences in IC50 between different groups of KIRC patients were analyzed and displayed using the "ggplot2" R package.

[0077] Docking verification of drug small molecule components and core common target molecules

[0078] Use the PDB database (https: / / www.rcsb.org / ) to download the 3D structure of the core common target (.pdb format file), and then use the PubChem database (https: / / pubchem.ncbi.nlm.nih.gov / ) to collect and analyze the small molecule drug structure (.sdf format file). Use the CB-Dock2 online database (https: / / cadd.labshare.cn / cb-dock2 / php / blinddock.php) to upload the small molecule compound and the target protein to the system for molecular docking. The larger the absolute value of the Vina score, the stronger the binding ability of the ligand molecule to the receptor protein. The specific parameter setting principle is to use the coordinates corresponding to the compound originally bound to the protein target pocket as the center to build the lattice.

[0079] The basic principles of the present invention are as follows:

[0080] We used a combination of 101 machine learning algorithms to analyze the KIRC patients in the TCGA-KIRC cohort and GSE22541, and screened out cell death patterns that were closely related to prognosis from 18 cell death patterns. We combined multi-omics analysis, including immune cell infiltration, typing, functional analysis, immune checkpoint exploration, and gene set enrichment analysis (GSEA) and other bioinformatics analyses to explore the relationship between key cell death patterns and patient immunotherapy response. Finally, we searched for potential drug targets through drug sensitivity screening and molecular docking technology.

[0081] In the field of KIRC research, this is the first time that a combination of 101 machine learning algorithms has been applied to explore the interactions between multiple PCD patterns in KIRC. With this advanced algorithm, we successfully constructed a powerful risk model and identified two PCD patterns that were significantly associated with the prognosis of KIRC patients: Anoikis and LDCD. Through in-depth multi-omics analysis, we revealed the potential mechanisms of these patterns in immunotherapy of KIRC patients, and found that Tfh infiltration and HLA-E expression were closely related to the immunotherapy effect of patients. In addition, we found that the two genes HMOX1 and PIK3CG jointly participated in regulating the key role of immunotherapy response in KIRC patients in Anoikis and LDCD. Drug sensitivity analysis showed that SB505124 showed significant high sensitivity in both PCD patterns, and molecular docking analysis verified the stable binding between Carnosol and HMOX1, and PROTAC and PIK3CG, providing new perspectives and potential therapeutic strategies for immunotherapy of KIRC patients.

[0082] In this study, we first collected 18 essential regulatory genes of PCD patterns and intersected them with differentially expressed genes in the TCGA-KIRC cohort. Then, we included 11 cell death patterns with >5 intersecting genes, laying the foundation for subsequent analysis. Using 10 machine learning algorithms and their 101 combinations, we constructed a comprehensive risk score model, and finally selected the Anoikis and LDCD models that performed well in KIRC diagnosis by calculating the C-index and the area under the ROC curve. In multivariate analysis, the Anoikis model surpassed traditional clinical features in predictive ability, while the LDCD model, although slightly lower in predictive ability than clinical stage, performed well in predicting stability. In order to improve the clinical application value of the model, we combined these Riskscore models with clinical factors and developed a nomogram model, in which the predictive accuracy of the Anoikis model was 0.781 and that of the LDCD model was 0.770. This indicates that our model has significant potential and application value in clinical prognosis assessment.

[0083] KIRC is one of the cancers with high immune infiltration. In the TME, immune cells play a key role in tumor growth, invasion, migration, and regulation of anticancer immunity. Therefore, before studying the effects of the two PCD patterns on immunotherapy in KIRC patients separately, we analyzed the immune status of KIRC patients as a whole. The results showed that in the TCGA-KIRC cohort, high levels of infiltration of specific immune cell subtypes were associated with better overall survival prognosis in patients. In particular, resting T cell CD4 memory type, monocytes, mast cells, and M2 macrophages, these cell subtypes may help inhibit tumor growth and spread by enhancing anti-tumor ability. However, immune cell infiltration also showed a double-edged sword effect. High levels of infiltration of plasma cells, activated T cell CD8, T cell CD4 memory type activation, follicular helper T cells, regulatory T cells (Tregs), M0 macrophages, resting dendritic cells, and activated mast cells were associated with poor prognosis in patients. The study found that high infiltration of CD8+T cells in KIRC was closely associated with poor prognosis. Similarly, there is evidence that infiltration of Tregs and Tfh and M0 macrophages in KIRC predicts a poor prognosis. These findings are consistent with our results and highlight the importance of considering immune cell infiltration in the treatment strategy for KIRC.

[0084] In order to explore the responsiveness of KIRC patients in the high- and low-risk groups in the two PCD patterns to immunotherapy, we performed a series of immune-related analyses. The results of immunophenotyping analysis showed that in both PCD patterns, the inflammatory immune cell subtypes in the high-risk group were significantly lower than those in the low-risk group. This suggests that the high-risk group may be associated with a weaker inflammatory immune response, which may mean that the immune system of these patients has a weaker ability to control tumors. More importantly, this result suggests that the two PCD patterns, Anoikis and LDCD, have similar effects on KIRC patients, and they may jointly shape the responsiveness of KIRC patients to immunotherapy. The results of immune cell infiltration analysis showed that Tfh and Treg showed high levels of expression in the high-risk group of both PCD patterns, which echoed the results of our previous overall analysis of the immune status of KIRC patients. Those with high expression of Tfh and Tregs had a poor prognosis, and it was consistent with the view reported in the literature that high expression of Tfh and Tregs was associated with a poor prognosis. Immune cell function analysis further showed that Tfh immune cells in the high-risk group were highly expressed, whether in Anoikis or LDCD patterns. However, there was no significant difference in Tregs in LDCD. This suggests that Anoikis and LDCD modes may jointly participate in regulating the immune response of KIRC patients through Tfh. In the immune checkpoint analysis, HLA-E showed higher expression levels than other immune checkpoints in both PCD modes, and the expression levels in the high-risk group were slightly lower than those in the low-risk group, which may suggest that HLA-E plays a key role in immune escape in Anoikis and LDCD. As an immune tolerance molecule, HLA-E binds to the CD94 / NKG2A receptor and plays an important role in the immune system to recognize and avoid attacking its own tissues. However, overexpression of HLA-E in renal cell carcinoma is associated with reduced immunogenicity. Studies have found that in KIRC, strong infiltration of NKG2A+CD8+T cells is associated with poor prognosis and response to immunotherapy, and NKG2A blockers combined with PD-1 blockers show better immunotherapy effects. These findings suggest that Anoikis and LDCD patterns can jointly shape the responsiveness of KIRC patients to immunotherapy by affecting Tfh infiltration and HLA-E expression. In particular, the development of targeted immunotherapy strategies against HLA-E and its ligand NKG2A may provide new ideas and directions for future immunotherapy.

[0085] Furthermore, we explored the biological differences between the two PCD patterns in the TCGA-KIRC cohort. In the Anoikis pattern, the overall mutation rate in the high-risk group was significantly higher than that in the low-risk group, indicating genetic instability during tumor development. The mutation rates of TTN and MUC16 genes were significantly higher in both high- and low-risk groups, among which the upregulation of the antisense long noncoding RNA (lncRNA TTN-AS1) of the TTN gene was positively correlated with adverse clinical and pathological characteristics, while the increased level of MUC16 was associated with a significant increase in immunosuppressive characteristics. These findings echoed the results of the previous immune-related analysis and further emphasized the role of the high-risk group in immune escape. In the LDCD pattern, although the overall mutation rate in the high-risk group was slightly higher than that in the low-risk group, there was no significant difference in tumor mutation burden, which means that the increase in mutation rate did not significantly affect the tumor immunogenicity or response to immunotherapy in KIRC patients with the LDCD pattern.

[0086] Next, GSEA analysis revealed that the Anoikis high-risk group was significantly enriched in primary immunodeficiency, and the LDCD high-risk group was significantly enriched in spliceosomal tri-snrnp complex assembly, pointing to abnormalities in RNA splicing and gene expression regulation, which affected the sensitivity and responsiveness of tumor cells to immune signals. This is consistent with the differences in Tfh and HLA-E expression we observed previously, indicating that the high-risk group may have immune response defects. Analysis of the ratio of immune matrix components confirmed that the tumor microenvironment in both PCD patterns was highly activated. These findings further suggest that the two PCD patterns may have complex interactions and influence mechanisms in the tumor microenvironment, which provides important clues and theoretical basis for further exploring tumor immunotherapy strategies.

[0087] In the face of the more severe prognosis of patients in the high-risk group compared with the low-risk group, we conducted an in-depth drug sensitivity analysis of the two PCD models. The purpose of this analysis is to identify drugs that may show better efficacy in the high-risk group. The results showed that SB505124 showed significant sensitivity in the high-risk group of both PCD models. As a selective TGF-β type I receptor inhibitor, SB505124 can block the TGF-β signaling pathway, thereby effectively inhibiting cell proliferation, differentiation and migration. Although no studies have directly explored the specific role of SB505124 in the treatment of KIRC, existing studies have clearly pointed out that TGF-β signaling plays a key role in regulating the invasiveness and metastatic ability of KIRC. This suggests that SB505124 has the potential to slow down the progression and metastasis of KIRC by inhibiting the TGF-β signaling pathway.

[0088] In addition, this study also found that HMOX1 and PIK3CG were expressed in Anoikis, LDCD-related genes, and upregulated differentially expressed genes in KIRC patients, indicating that Anoikis and LDCD may affect the infiltration of Tfh and the expression of HLA-E by enhancing the expression of HMOX1 and PIK3CG, and jointly shape the responsiveness of KIRC patients to immunotherapy. HMOX1 has been shown to play a regulatory role in immunotherapy in patients with advanced renal cell carcinoma, which is consistent with our findings. In addition, through literature research, we found that Carnosol can exert its antioxidant and protective functions of microvascular endothelial cells by enhancing the expression of HMOX1 and activating the Nrf2 signaling pathway. Our experiments further verified the binding ability of Carnosol to the active site of HMOX1 and confirmed its stability. Although there is a lack of literature directly studying the effect of PIK3CG on immunotherapy in KIRC patients, as a key component of the PI3K / Akt / mTOR signaling pathway, its potential role in KIRC treatment cannot be ignored. In particular, mTOR inhibitors have shown significant effects in KIRC treatment. In addition, a research team has developed a PROTAC heterobifunctional molecule that can specifically degrade PIK3CG and block Signaling pathway. Through further molecular docking analysis, we found that PIK3CG can form a stable binding with the active site of the PROTAC molecule. Through further molecular docking analysis, we verified the binding affinity between the active sites of HMOX1 and PIK3CG and potential drug molecules, providing important information for the development of new treatment strategies.

[0089] Overall, although our drug screening strategies focused on different biological targets twice - the first time targeting two PCD patterns and the second time targeting two common key genes - they share the same goal of deepening our understanding of KIRC patients' immunotherapy response and working to improve their treatment outcomes.

[0090] In summary, this study provides a new perspective for a deeper understanding of the immunotherapy response of KIRC. Although previous studies have recognized the independent prognostic value of Anoikis and LDCD in KIRC and help evaluate patients' response to immunotherapy, these studies often regard them as isolated PCD mechanisms. This single perspective may limit our understanding of the complexity of KIRC's immunotherapy response. Our study breaks through this limitation, revealing for the first time the interaction between Anoikis and LDCD and showing how they jointly shape the immunotherapy response of KIRC patients. In particular, we found that Anoikis and LDCD patterns may jointly shape the responsiveness of KIRC patients to immunotherapy by enhancing the expression of HMOX1 and PIK3CG, thereby affecting the infiltration of Tfh and the expression of HLA-E. In addition, we found that SB505124 showed significant sensitivity in the high-risk group of these two PCD patterns. At the same time, we also found that there was a stable binding force between Carnosol and HMOX1, and between PROTAC and PIK3CG. Our study highlights the necessity of considering the interaction of multiple PCD modes in KIRC treatment strategies, which is crucial for discovering key target drugs and developing combination drug regimens, thus bringing more precise and effective treatment options for KIRC patients.

Claims

1. A risk assessment model for KIRC patients, characterized in that: Based on the gene expression data of Anoikis and LDCD patterns, the predictive efficacy of the model was evaluated using C-index and ROC curve, and risk assessment was performed using a nomogram combined with clinical factors.

2. The method for constructing a risk assessment model for KIRC patients according to claim 1, characterized in that: The following steps are involved: (1) Collect clinical pathological information and gene expression data of KIRC patients: collect 18 programmed cell death modes and their key regulatory genes; download TCGA-KIRC cohort data and RNA sequencing data from the TCGA database; collect clinical pathological information and microarray expression analysis data of the KIRC cohort from the GEO database; (2) Identification of differentially expressed genes in KIRC: The “edgeR” package was used to screen differentially expressed genes, with the criteria of adjusted P value < 0.05 and |log2FC| > 1; (3) Construction of prognostic models based on comprehensive methods of machine learning: Risk score models were constructed in the TCGA-KIRC cohort using a combination of multiple algorithms and leave-one-out cross-validation, and GSE22541 was used for cross-validation. The Harrell consistency index C-index of each model was calculated, and the average C-index or test set C-index>0.7 was selected. The diagnostic efficacy of PCD in the cohort was further evaluated by calculating the area under the ROC curve. (4) To verify the value of different PCD patterns as independent prognostic indicators for KIRC patients, univariate and multivariate Cox regression analyses were performed. Based on the TCGA-KIRC cohort and the GSE22541 cohort, the prognostic nomograms were drawn using the R packages "rms" and "replot", and the risk assessment model for KIRC patients was constructed based on the prognostic nomograms.

3. The method for constructing a risk assessment model for KIRC patients according to claim 2, characterized in that: The machine learning algorithms include, but are not limited to, random survival forest, elastic net, Lasso, Ridge, stepwise Cox, CoxBoost, partial least squares regression, supervised principal component, generalized boosted regression modeling, and survival support vector machine.

4. A method for evaluating the responsiveness of patients with renal clear cell carcinoma to immunotherapy, characterized in that: The model constructed by using the risk assessment model for KIRC patients described in claim 1 or the method for constructing the risk assessment model for KIRC patients described in claim 2 specifically comprises the following steps: (1) Collect clinicopathological information and gene expression data of KIRC patients; (2) inputting the collected clinical pathological information and gene expression data into the risk assessment model for KIRC patients, and scoring using the risk assessment model for KIRC patients; (3) Assessing the patient's risk level according to the scoring results of the risk assessment model for KIRC patients.

5. Application of HMOX1, PIK3CG, Tfh, and HLA-E as biomarkers for KIRC.

6. A biomarker combination for monitoring the prognosis of KIRC patients, characterized in that: Including HMOX1, PIK3CG, Tfh cell infiltration levels and HLA-E expression levels.

7. Application of SB505124, Carnosol, and PROTAC as immunotherapeutic agents in the preparation of drugs for the treatment of KIRC.

8. A drug combination for treating KIRC, characterized in that: include: TGF-β type I receptor inhibitors, compounds that enhance HMOX1 expression, and compounds that specifically degrade PIK3CG.

Images