Integrating machine learning and multi-omics analysis to unveil key programmed cell death patterns and immunotherapy targets in kidney renal clear cell carcinoma

Abstract

Kidney renal clear cell carcinoma (KIRC), a cancer characterized by substantial immune infiltration, exhibits limited sensitivity to conventional radiochemotherapy. Although immunotherapy has shown efficacy in some patients, its applicability is not universally effective. Studies have indicated that programmed cell death (PCD) can modulate the activity of immune cells and participate in the regulation of antitumor immune responses. However, systematic research on how various PCD patterns in KIRC affect the responsiveness to immunotherapy is lacking and requires in-depth investigation. We utilized a combination of 101 machine learning algorithms to analyze the TCGA-KIRC cohort and the GSE22541 KIRC patients, screening for cell death patterns closely associated with prognosis from 18 potential modes. Integrating multi-omics analysis, including immune cell infiltration, phenotyping, functional analysis, immune checkpoint exploration, and gene set enrichment analysis (GSEA), we explored the relationship between key cell death patterns and patients' responses to immunotherapy. Finally, potential drug targets were identified through drug sensitivity screening and molecular docking techniques. Our sophisticated risk assessment model successfully identified two PCD patterns, Anoikis and lysosome-dependent cell death (LDCD), closely associated with the prognosis of KIRC patients, with the high-risk group exhibiting poor outcomes. Immune cell analysis revealed upregulated expression of T follicular helper (Tfh) cells in both PCD patterns. Analysis of immune checkpoints disclosed enhanced expression of human leukocyte antigen E (HLA-E) across both patterns. Frequent mutations in the TTN and MUC16 genes were observed in the Anoikis pattern, whereas in the LDCD pattern, although the high-risk group had a higher mutation rate, there was no significant difference in tumor mutational burden. GSEA analysis indicated significant enrichment of the primary immunodeficiency pathway in the Anoikis high-risk group and significant enrichment of the spliceosomal tri-snrnp complex assembly pathway in the LDCD high-risk group. Drug sensitivity analysis showed notable sensitivity to SB505124 in both PCD patterns. HMOX1 and PIK3CG were identified as common genes in the two key PCD patterns, and molecular docking analysis confirmed stable binding affinity between Carnosol and HMOX1, and between PROTAC and PIK3CG. Our study identifies Anoikis and LDCD as prognostic PCD patterns in KIRC, with key immune cells, genetic mutations, and drug sensitivity profiles. HMOX1 and PIK3CG are common genes with stable binding to Carnosol and PROTAC, respectively, while SB505124 shows significant sensitivity to both PCD modes, suggesting potential therapeutic targets.

Keywords: Immune checkpoints; KIRC; Machine learning algorithms; Programmed cell death.

Fig. 1

A graphic abstract of this study.

Fig. 2

Preliminary screening of PCD of KIRC patient. (A) Essential regulatory genes for 18 modes of PCD. (B) 12,788 DEGs between normal and tumor tissues in the TCGA-KIRC cohort. (C) Intersections of 18 PCD modes with differentially expressed genes in the TCGA-KIRC cohort.

Fig. 3

Univariate COX regression analysis results for 11 PCD modes. (A) Anoikis, (B) Apoptosis, (C) Autophagy, (D) LDCD, (E) Ferroptosis, (F) Necroptosis, (G) Immunogenic cell death, (H) NETosis, (I) Cuproptosis, (J) Paraptosis, (K) Pyroptosis.

Fig. 4

101 Machine learning combinations unveil key patterns of PCD in KIRC. (A-B) Model construction of TCGA-KIRC and GSE2254. (C-D) ROC curves and survival analysis for anoikis. (E–F) ROC curves and survival analysis for LDCD. (G-H) Predictive ability and comparative diagram of anoikis and LDCD.

Fig. 5

Nomogram models. (A, B) Anoikis. (C, D) LDCD.

Fig. 6

Immune cell survival analysis and phenotyping. Immune cell survival analysis is presented for the following panels: (A) Tregs, (B) Tfh, (C) CD8 T cells, (D) CD4 memory resting T cells, (E) CD4 memory activated T cells, (F) Plasma cells, (G) Monocytes, (H) Resting mast cells, (I) Activated mast cells, (J) Macrophages M2, (K) Macrophages M0, and (L) Resting dendritic cells. Immune cell phenotyping for (M) Anoikis and (N) LDCD.

Fig. 7

Immune cell subtypes and immune checkpoint expression. (A, B) Analysis of immune cell infiltration for Anoikis and LDCD. (C, D) Analysis of immune cell infiltration using diverse software algorithms. (E, F) Functional enrichment analysis of immune cells.

Fig. 8

Molecular characteristics and immune escape mechanisms. (A, B) Clinical information of patients under Anoikis and LDCD. (C, D) Somatic mutation analysis for Anoikis. (E, F) TMB analysis for Anoikis. (G, H) Somatic mutation analysis for LDCD. (I, J) TMB analysis for LDCD. (K, L) Expression levels of immune checkpoint-related genes in Anoikis and LDCD.

Fig. 9

Analysis of immune evasion correlation. (A–I) Analysis of immune checkpoint associations in high and low-risk Anoikis groups: (A) MSI, (B) TIDE, (C) Dysfunction, (D) Exclusion, (E) CAF, (F) Responder, (G) CD8, (H) CD274, (I) TAM M2. (J–R) Analysis of immune checkpoint associations in high and low-risk LDCD groups: (J) MSI, (K) TIDE, (L) Dysfunction, (M) Exclusion, (N) CAF, (O) Responder, (P) CD8, (Q) CD274, (R) TAM M2.

Fig. 10

GSEA and assessment of the tumor microenvironment. (A) GO analysis of high-risk groups for Anoikis. (B, C) Ratios of immune stroma components in the tumor microenvironment among high and low-risk groups for Anoikis and LDCD. (D) Intersection of Anoikis, LDCD, and DEGs in KIRC patients.

Fig. 11

Molecular docking verification of HMOX1 and PIK3CG with related drug small molecule components. (A–E) Anoikis exhibits sensitivity to the drugs dihydrorotenone, OF-1, ibrutinib, SB505124, and P22077. (F) LDCD shows sensitivity to SB505124. (G–J) Schematic diagrams of molecular docking (black circles represent binding pockets, and dashed lines of different colors represent potential weak interactions between ligands and amino acid sites).