Identification and validation of an immune-related programmed cell death signature for predicting prognosis and immunotherapy in large-scale multicenter cohorts for lung adenocarcinoma

Introduction

Lung cancer, with its high global mortality rate, exhibits a relatively low five-year survival rate and generally poor prognosis for patients (1,2). Lung adenocarcinoma (LUAD) is a major subtype of lung cancer (3). Despite significant advancements in the diagnosis and treatment of LUAD in recent years, the prognosis for patients with LUAD remains unsatisfactory (4,5). Currently, clinical prediction of LUAD prognosis primarily relies on surgeons’ empirical judgment based on traditional pathological and clinical variables, which exhibit notable limitations in precision and personalization (6,7). Therefore, there is an urgent need to develop a more accurate method for predicting the prognosis of patients with LUAD and stratifying them into different risk groups to optimize treatment strategies and improve patient survival rates.

Immunotherapy has revolutionized cancer treatment. However, less than 10–20% of cancer cases exhibit durable responses to immune checkpoint blockade (ICB) (8,9). Commonly used biomarkers for predicting the efficacy of immunotherapy in clinical practice include programmed death ligand 1 (PD-L1) protein expression (10), tumor mutation burden (TMB) (11,12) and microsatellite instability (MSI) (13). Recent studies indicate that CD8⁺ T cells not only play a critical role in anti-tumor responses during immunotherapy but also that their dynamic changes and subpopulation characteristics offer valuable insights for assessing treatment efficacy and predicting patient prognosis (14,15). But these markers have certain limitations in predicting the effectiveness of immunotherapy, such as multiple technical issues in PD-L1, high variability and significant costs in TMB and low frequency of MSI in LUAD (10,16,17). These limitations highlight the critical need for identifying predictive biomarkers for immunotherapy to enable more precise clinical intervention and personalized treatment.

Programmed cell death (PCD) is a regulated process whereby cells undergo death through pre-programmed molecular mechanisms governed by intracellular pathways (18). This process includes various forms such as necroptosis, pyroptosis, ferroptosis, entotic cell death, netotic cell death, parthanatos, lysosome-dependent cell death, autophagy-dependent cell death, alkaliptosis, and oxeiptosis (18). Accumulating evidence underscores the pivotal role of PCD in the onset and progression of diseases, particularly cancer (19,20). Pyroptosis, a form of PCD distinct from apoptosis, has recently been shown to enhance anti-tumor immune responses by modulating inflammation and the immune microenvironment. This characteristic positions pyroptosis as a significant target for immunotherapy (21-23). Although single-cell death pattern models have already been developed, many of these studies have encountered limitations, including simplistic models, small cohort sizes, and the absence of independent validation cohorts (24-26).

In this study, we integrated 18 PCD signatures and identified immune-related PCD genes. We then utilized 101 combinations of 10 machine learning algorithms to develop the immune-related programmed cell death signature (IRPCDS). Our results demonstrated that IRPCDS exhibited stable and robust performance across multiple independent validation cohorts, accurately predicting patient outcomes. Additionally, IRPCDS outperformed traditional clinicopathological factors and previously published signatures in predicting values. Our findings further suggest that low-risk patients may benefit from immunotherapy. Overall, this study underscores the significant role of PCD in the development, prognosis, and therapeutic response of LUAD. We have developed a precise method for predicting the prognosis of LUAD and stratifying patients, thereby improving clinical outcomes for individual LUAD. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1015/rc).

Methods

Patients and tissue samples

In this study, we collected pathologically confirmed paired LUAD and adjacent normal lung tissue samples from Qingdao Traditional Chinese Medicine Hospital between June and December 2024. A total of six patients were enrolled in this study, including four males and two females, with ages ranging from 49 to 72 years. The samples included six primary tumor tissues and six paired adjacent normal tissues. Tumor tissues were collected from the lesion core, while adjacent normal tissues were obtained from pulmonary regions at least 5 cm beyond the tumor margin. Following tissue collection, total protein was extracted using a commercial total protein extraction kit. According to the staging criteria jointly established by the American Joint Committee on Cancer (AJCC) and the Union for International Cancer Control (UICC), the cohort included three cases of stage I, one case of stage II, and two cases of stage III disease. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Qingdao Traditional Chinese Medicine Hospital (No. 2022HC05LS001) and informed consent was obtained from all individual participants.

Publicly available data collection and processing

The RNA-seq data and corresponding clinical pathological information for The Cancer Genome Atlas (TCGA)-LUAD were obtained from TCGA (accessible at https://portal.gdc.cancer.gov/). The RNA-seq data, protein abundance levels, and corresponding clinical pathological information for Clinical Proteomic Tumor Analysis Consortium (CPTAC)-LUAD were sourced from the study conducted by Petralia et al. (27). To address the substantial missing data in protein abundance profiles, we implemented a two-step preprocessing pipeline: first, proteins with non-detectable expression in ≥50% of samples were excluded, and subsequently, missing values in the retained dataset were imputed using the impute package in R statistical software. Additionally, clinical pathological information and gene expression data for three other LUAD cohorts (GSE31210, GSE50081, and GSE37745) and the immunotherapy cohorts (GSE126044 and GSE91061) were retrieved from the Gene Expression Omnibus (GEO, accessible at https://www.ncbi.nlm.nih.gov/gds/) database. Details of baseline information for all cohorts are provided in Table S1. For the expression matrices in raw read count format, we converted the data to transcripts per kilobase million (TPM) and subsequently applied a log₂ transformation to all expression matrices. In addition, we also downloaded the immunohistochemical information of LUAD and adjacent tissues from the Human Protein Atlas (HPA), with the website: https://www.proteinatlas.org/.

Single-sample gene set enrichment analysis (ssGSEA)

The ssGSEA algorithm, implemented in the R package GSVA, was used to evaluate the scores of 18 PCD signatures, the activity of 10 classical cancer-related signaling pathways, and the relative infiltration levels of 28 immune cells (28).

Consensus clustering

Consensus clustering was performed using the ConsensusClusterPlus R package, employing the infiltration patterns of 28 immune cells to stratify LUAD patients into clusters ranging from 2 to 6 (29). Specifically, 80% of the tumor samples were randomly subsampled without replacement and partitioned into three major clusters using the K-means algorithm with Pearson’s correlation as metric.

Weighted gene co-expression network analysis (WGCNA)

WGCNA package was utilized to construct co-expression gene networks associated with PCD in the TCGA-LUAD dataset (30). Initially, we calculated an appropriate soft threshold β to ensure the network met the criteria for a scale-free topology. Next, we transformed the weighted adjacency matrix into a topological overlap matrix (TOM). Gene modules were identified using the dynamic tree cutting algorithm. Finally, we screened for PCD-associated gene modules that exhibited the highest correlation with immune clusters for in-depth analysis.

Construction and validation of IRPCDS

To develop a highly accurate and stable prognostic model, we employed a machine learning framework that integrated 10 machine learning algorithms [including Random Survival Forest (RSF), Elastic Net (Enet), the least absolute shrinkage and selection operator (LASSO), Ridge, Stepwise Cox, CoxBoost, partial least squares regression for Cox proportional hazards (plsRcox), Supervised Principal Components (SuperPC), Generalized Boosted Regression Modeling (GBM), and Survival Support Vector Machine (survival-SVM)] and 101 algorithm combinations. The process comprised the following steps:

• The identified immune-related PCD genes were subjected to univariate Cox regression analysis to identify those with prognostic value. These prognostic immune-related PCD genes were then used as initial features and input into 101 algorithm combinations to construct predictive models using leave-one-out cross-validation (LOOCV) in the TCGA-LUAD cohort.
• All models were further validated using four independent validation datasets (CPTAC-LUAD, GSE31210, GSE50081 and GSE37745).
• For each model, the Harrell’s concordance index (C-index) was calculated in TCGA-LUAD training dataset and four validation datasets, and the model with the highest average C-index was deemed optimal.

The optimal predictive model integrated LASSO and plsRcox algorithms, incorporating 17 immune-related PCD genes, namely TG1IP, GNAI3, KRT8, CDCP1, TRIM6, G0S2, CIDEB, PEBP1, RNF5, SIRT2, NCOA4, ATP6V1B2, FUCA1, BTK, ADRB2, GPR15, and CX3CR1. We conducted a comprehensive evaluation of the predictive efficacy of the model across multiple cohorts using time-dependent receiver operating characteristic curve (tROC) analysis, which evaluated its discriminative accuracy for survival outcomes at specified time intervals via the timeROC tool. The optimal threshold for IRPCDS was determined using the surv_cutpoint function from the survminer package in TCGA-LUAD training dataset. Based on the threshold, patients in all validation cohorts were categorized into high-risk and low-risk groups. Survival differences between the two risk groups were analyzed and visualized using the survminer package.

Collection and comparison of previously published LUAD prognostic signatures

We performed a literature search on PubMed for articles published up to August 1, 2024, focusing on prognostic signatures for LUAD (available online: https://cdn.amegroups.cn/static/public/tcr-2025-1015-1.xls). In all cohorts, we calculated risk scores based on the gene expression values and the corresponding coefficients provided in the literature. The predictive performance of these models was assessed using univariate Cox regression analysis.

Tumor microenvironment (TME) and immune infiltration analysis

The Estimate package was employed to calculate the stromal score, immune score, and tumor purity score, with the gene expression matrix serving as input data. We assessed the immune infiltration of 22 cell types using the CIBERSORT (31) algorithm and evaluated the infiltration of 6 immune cells using the TIMER (32) algorithm.

Single-cell RNA-sequencing (scRNA-seq) data collection and processing

The LUAD scRNA-seq dataset (accession number GSE137804) was obtained from the GEO database. For downstream analyses, the dataset was processed using the Seurat 4.0 R package. Cells were selected based on specific criteria: gene counts between 250 and 2,500, total transcript counts exceeding 500, and a mitochondrial gene proportion below 15%. We then identified the top 2,000 highly variable genes for scaling. Harmony batch effect correction was subsequently applied, followed by uniform manifold appproximation and projection (UMAP) dimensionality reduction and clustering. Cell types were annotated based on typical markers referenced in the original literature. Additionally, we performed cell interaction analysis using the R package CellChat (33).

Western blot

Proteins were extracted using the Total Protein Extraction Maxi Kit (Solarbio, Beijing, China; catalog number: BC3710). Following centrifugation at 12,000 ×g for 15 minutes at 4 ℃, the supernatant was collected as the total protein sample. Protein concentrations were quantified using the BCA Kit (Beyotime, Shanghai, China; catalog number: P0010). A 40 μg aliquot of the protein sample was subjected to electrophoresis on a 12.5% SDS-PAGE gel. The membrane was blocked with TBST containing 5% non-fat milk for 1 hour at room temperature. Subsequently, it was incubated overnight at 4 ℃ with primary antibodies, including anti-CDCP1 (1:1,000 dilution, catalog number: 12754-1-AP, Proteintech, Wuhan, China), anti-PEBP1 (1:1,000 dilution, catalog number: ab76582, Abcam, Shanghai, China), and anti-GAPDH (1:50,000 dilution, catalog number: 60004-1-Ig, Proteintech). After washing with TBST, horseradish peroxidase (HRP)-conjugated goat anti-mouse/rabbit IgG secondary antibodies were applied, and the membrane was incubated for 1 hour at room temperature. Protein bands were visualized using enhanced chemiluminescence (ECL) reagent and detected with a chemiluminescence imaging system. Density analysis was performed, and results were expressed as quantitative data from three independent experiments.

Statistical analysis

Statistical analysis was conducted using R software (version 4.2.6). Pearson’s correlation coefficients were employed to evaluate the associations between two continuous variables. Categorical data were analyzed using Chi-squared tests, while continuous variables were compared using the Wilcoxon rank-sum test or the t-test. Statistical significance was established at a P value of <0.05.

Results

PCD signatures were closely associated with the prognosis and immune infiltration in LUAD

We systematically collected 18 PCD signatures from published literature and identified 1,964 PCD-associated genes (34,35). The types of PCD and the proportion of genes contained in each type are depicted in Figure 1A. To elucidate the biological roles of these 18 PCD signatures in LUAD, we compared the ssGSEA scores of all 18 PCD signatures between tumor and adjacent non-tumor tissues. Our analysis revealed that immunogenic cell death, entotic cell death, and NETosis were the most enriched cell death signatures in tumor tissue, whereas entosis, autophagy, and cuproptosis were the most enriched in normal tissue in TCGA-LUAD, CPTAC-LUAD and GSE31210 cohorts (Figure 1B). Additionally, we examined the prognostic significance of all 18 PCD modes in all LUAD cohorts. Univariate Cox regression analyses across LUAD cohorts demonstrated significant correlations between overall survival (OS) and ssGSEA scores of 18 PCD signatures (Figure 1C). Pearson correlation analysis indicated that the majority of PCD modes were associated with tumor-related pathways (Figure 1D). For example, lysosome-dependent cell death was positively correlated with the MAPK signaling pathways, PI3K-Akt signaling pathway, and Ras signaling pathway, while negatively correlated with the cell cycle signaling pathway. Ferroptosis exhibited positive correlations with the cell cycle signaling pathway and NRF2 pathway but negative correlations with hippo signaling pathway, PI3K-Akt signaling pathway, and Ras signaling pathway.

Figure 1 The relationship between 18 PCD signatures and the occurrence, progression, and prognosis of LUAD. (A) 18 PCD modes and the gene proportion for each PCD type. (B) Heatmap showing the differences in ssGSEA scores of 18 PCD signatures between tumor and adjacent non-tumor tissues in the TCGA-LUAD, CPTAC-LUAD, and GSE31210 cohorts. Differential analysis was performed using the limma package. Log₂FC >0, P<0.05 indicates enrichment in tumor tissues, while log₂FC <0, P<0.05 indicates enrichment in normal tissues. (C) Univariate Cox regression analysis of ssGSEA scores of 18 PCD signatures in the TCGA-LUAD, CPTAC-LUAD, GSE31210, GSE50081, and GSE37745 cohorts. (D) Pearson correlation heatmap depicting the relationships between ssGSEA scores of 10 tumor-related signaling pathways and 18 PCD signatures. CPTAC, clinical proteomic tumor analysis consortium; FC, fold change; LUAD, lung adenocarcinoma; PCD, programmed cell death; ssGSEA, single-sample gene set enrichment analysis; TCGA, The Cancer Genome Atlas.

Accumulating evidence has revealed a relationship between PCD and immune checkpoint inhibitors (ICIs). Consequently, we explored the relationship between PCD and immune cell infiltration. Our results showed that the majority of PCD modes were significantly associated with immune infiltration. For instance, immunogenic cell death was positively correlated with myeloid-derived suppressor cells (MDSC), regulatory T cells (Tregs), and activated B cells, but negatively correlated with CD56^bright natural killer (NK) cells, immature dendritic cells (DCs), and plasmacytoid DCs. Conversely, Autophagy was negatively correlated with MDSC and Tregs, but positively correlated with plasmacytoid DCs, activated CD4(+) T cells, and activated CD8(+) T cells (Figure S1).

Identification of immune-related PCD genes

Next, we employed a consensus clustering approach, utilizing the infiltration patterns of 28 immune cells in TCGA-LUAD cohort, to stratify LUAD patients into clusters ranging from 2 to 6. By assessing the cumulative distribution function (CDF) curve and the proportion of ambiguous clustering (PAC) statistic, we determined the optimal number of clusters to be 3. The clustering heatmap revealed distinct variations in immune infiltration across these clusters (Figure 2A), and principal component analysis (PCA) effectively differentiated the immune clusters (Figure 2B), with their immune scores ranked in descending order: C3 > C2 > C1 (Figure 2C). Subsequent analyses demonstrated significant differences in immune cell infiltration among the clusters (Figure 2D). Specifically, cluster C1 was enriched with Th2, Th17, CD56^bright NK cells, and CD56^dim NK cells. Cluster C2 exhibited enrichment primarily in memory CD8(+) T cells and NK cells. Cluster C3 was characterized by significant enrichment of activated B cells, activated CD8(+) T cells, and immature B cells.

Figure 2 Identification of immune-related PCD genes in the TCGA-LUAD cohort. (A) The infiltration abundance of 28 immune cells evaluated by ssGSEA for three clusters. (B) PCA of 28 immune cells infiltration among the three clusters. (C) Violin plots showing the differences in immune scores among the three clusters. (D) The distribution of 28 immune cells infiltration among the three clusters. (E) Heatmap of the correlations between MEs and clinical traits. Correlation coefficients and corresponding P values are shown in the rectangles and the brackets, respectively. NS, P>0.05; **, P<0.01; ***, P<0.001. LUAD, lung adenocarcinoma; ME, module eigengene; PCA, principal component analysis; PCD, programmed cell death; ssGSEA, single-sample gene set enrichment analysis; TCGA, The Cancer Genome Atlas.

To identify immune-related PCD genes, we conducted a WGCNA procedure using the expression profiles of PCD genes in TCGA-LUAD cohort. We balanced the two key criteria: the scale-free fit index and the mean connectivity, ultimately selecting 9 as the soft threshold β. This choice ensures sufficient power for constructing the co-expression network (Figure S2). Ultimately, we obtained seven modules, each characterized by a distinct color (Figure 2E). We then calculated the correlations of these modules with various clinical traits, including immune clusters, age, gender, T stage, node (N) stage, metastasis (M) stage, stage, homologous recombination deficiency (HRD), TMB. The results indicated that the red and blue modules showed the strongest correlations with the immune clusters. Ultimately, we identified 315 immune-related PCD genes, comprising 48 genes from the red module and 267 genes from the blue module.

Construction and validation of IRPCDS

Next, we utilized the expression profiles of 315 immune-related PCD genes and further identified 45 prognostic genes through univariate Cox regression analysis. Subsequently, we employed 101 combinations of 10 machine-learning algorithms to develop the IRPCDS using these 45 genes as initial features. In the TCGA-LUAD dataset, we fitted 101 prediction models via the LOOCV framework and calculated the C-index for each model across all validation datasets. The optimal model was a combination of LASSO and plsRcox algorithms, comprising 17 immune-related PCD genes, with the highest average C-index (0.685) (Figure 3A). For each patient, a risk score based on the model was obtained. All patients were stratified into high- and low-risk groups according to the optimal cut-off value determined by the survminer package in TCGA-LUAD cohort. Compared to patients in the low-risk group, those in the high-risk group exhibited significantly poorer OS in both the TCGA-LUAD training set (Figure 3B) and four validation sets: CPTAC-LUAD (Figure 3C), GSE31210 (Figure 3D), GSE50081 (Figure 3E), and GSE37745 (Figure 3F). The ROC analyses revealed areas under the ROC curves (AUCs) of 0.764, 0.777, and 0.822 for 1-, 3-, and 5-year OS in the TCGA-LUAD training set (Figure 3G); 0.742, 0.638, and 0.712 for 1-, 3-, and 5-year OS in the CPTAC-LUAD (Figure 3H); 0.644, 0.644, and 0.699 for 1-, 3-, and 5-year OS in the GSE31210 (Figure 3I); 0.737, 0.725, and 0.689 for 1-, 3-, and 5-year OS in the GSE50081 (Figure 3J); and 0.730, 0.633, and 0.687 for 1-, 3-, and 5-year OS in the GSE37745 (Figure 3K). These results indicated that the IRPCDS exhibited stable and robust performance across various independent cohorts, effectively predicting patient prognosis.

Figure 3 Construction and validation of IRPCDS via machine learning algorithm. (A) C-indices of multiple models derived from various machine-learning algorithm combinations in TCGA-LUAD, CPTAC-LUAD, GSE31210, GSE50081, GSE37745 cohorts. Time-dependent ROC curve analysis of the IRPCDS in TCGA-LUAD (B), CPTAC-LUAD (C), GSE31210 (D), GSE50081 (E), GSE37745 (F) cohorts. Kaplan-Meier curves of OS between high- and low-risk groups defined by IRPCDS in TCGA-LUAD (G), CPTAC-LUAD (H), GSE31210 (I), GSE50081 (J), GSE37745 (K) cohorts. AUC, area under the curve; CPTAC, clinical proteomic tumor analysis consortium; IRPCDS, immune-related programmed cell death signature; LUAD, lung adenocarcinoma; TCGA, The Cancer Genome Atlas.

The predictive value of IRPCDS for OS outperformed clinicopathological factors and previously published signatures

In conventional clinical practice, patient prognosis was typically evaluated using various clinical indicators such as tumor location and TNM staging. However, this approach often relied on the subjective judgment of surgeons, potentially limiting accuracy. We conducted a comparative analysis between the IRPCDS and the clinical factors commonly used by clinicians for empirical prognostic assessment, as well as molecular variables, in predicting patient prognosis. The results revealed that the IRPCDS achieved a significantly higher C-index than the traditional method leveraging clinical indications (Figure 4A), indicating its superior performance in predicting clinical outcomes.

Figure 4 Evaluation of the prognostic performance of the IRPCDS model. (A) The predictive value for OS when comparing IRPCDS with clinicopathological factors. Statistical tests: two-sided z-score test. *, P<0.05; **, P<0.01; ***, P<0.001. (B) The predictive value for OS when comparing IRPCDS with previously published LUAD prognostic signatures. CPTAC, clinical proteomic tumor analysis consortium; HRD, homologous recombination deficiency; IRPCDS, immune-related programmed cell death signature; LUAD, lung adenocarcinoma; OS, overall survival; pM, pathological metastasis stage; pN, pathological node stage; pT, pathological tumor stage; TCGA, The Cancer Genome Atlas.

We also manually curated 52 published gene panel-based prognostic signatures by systematically reviewing the literature. The comparative analysis consistently showed that the IRPCDS had higher accuracy in predicting prognosis in the TCGA-LUAD training dataset and all validation datasets when compared with these established signatures (Figure 4B). These findings underscore the potential of the IRPCDS to enhance the precision and reliability of prognostic assessments in LUAD.

Patients in the low-risk group exhibited high immune infiltration and potential benefit from ICB therapy

The heterogeneity of the tumor immune microenvironment (TIME) influences tumor progression, prognosis, and response to immunotherapy (36). Therefore, we assessed the differences in immune infiltration between the high- and low-risk groups. Initially, Pearson correlation analysis indicated that the IRPCDS was associated with the expression of a broad spectrum of immune checkpoint-related genes (Figure 5A). Stromal scores and immune scores were significantly higher in the low-risk group compared with the high-risk group, which were significantly negatively correlated with the risk score (Figure 5B,5C). Conversely, tumor purity was significantly higher in the high-risk group and positively correlated with the risk score (Figure 5D). Compared to patients in the high-risk group, those in the low-risk group exhibited significantly higher overall immune infiltration abundance (Figure 5E,5F). The significant differences in immune infiltration between the two risk groups suggested that there may be differences in their sensitivity to immunotherapy. Furthermore, we utilized the tumor immune dysfunction and exclusion (TIDE) algorithm to assess the response of LUAD to immunotherapy. The results indicated that the TIDE score was lower in the low-risk group (Figure 5G). Consistently, the cytotoxic activity (CYT), an immunotherapy biomarker reflecting the anti-tumor immune activity of CD8⁺ cytotoxic T cells and macrophages, was significantly higher in the low-risk group compared with the high-risk group (Figure 5H), these results suggested that patients in low-risk group might benefit from ICIs therapy. We further confirmed the IRPCDS’s power to predict the response to ICIs using the GSE126044 and GSE91061 cohorts, which were treated with anti-PD-1 therapy. The results consistently showed that the IRPCDS scores were significantly lower in responders compared with non-responders (Figure 5I). Additionally, we calculated AUC for the IRPCDS in predicting treatment response outcomes in the GSE126044 and GSE91061 cohorts. The results demonstrated that the IRPCDS exhibited strong predictive performance in both cohorts (Figure 5J,5K). These findings underscore the potential utility of IRPCDS as a predictive biomarker for immunotherapy in LUAD.

Figure 5 Comparison of the TIME and immune therapy efficacy between high- and low-risk groups. (A) Heat map of the correlation between the IRPCDS and immune checkpoint-related genes in the TCGA-LUAD cohort. (B-D) Violin plots showing the differences in stromal score, immune score, and tumor purity between the high- and low-risk groups (left), and the correlation between stromal score, immune score, tumor purity, and risk score (right) in the TCGA-LUAD cohort. (E,F) Box plots showing the differences in the infiltration abundance of immune cells evaluated by CIBERSORT and TIMER algorithms between the high- and low-risk groups in the TCGA-LUAD cohort, respectively. (G,H) Comparison of TIDE score and CYT activity between the high- and low-risk groups (left), and the correlation between TIDE score and CYT activity and risk score (right) in the TCGA-LUAD cohort. (I) Comparison of risk scores between responders and PD or SD patients in the GSE126044 (non-small cell lung cancer) cohort (left), and comparison of risk scores between CR or PR and PD or SD patients in the GSE91061 (melanoma) cohort (right). (J-K) ROC plots illustrate the performance of the IRPCDS in predicting responses to anti-PD-1 therapy for the GSE91061 (J) and GSE126044 (K) cohorts, respectively. NS, P>0.05; *, P<0.05; **, P<0.01; ***, P<0.001. CR, complete response; CYT, cytotoxic activity; IRPCDS, immune-related programmed cell death signature; LUAD, lung adenocarcinoma; PD, progressive disease; PR, partial response; ROC, receiver operating characteristic; SD, stable disease; TCGA, The Cancer Genome Atlas; TIME, tumor immune microenvironment; TIDE, tumor immune dysfunction and exclusion.

Genome alteration landscape of IRPCDS

To explore the genomic heterogeneity between the high- and low-risk groups, we performed a detailed comparative analysis of mutation alterations in the TCGA-LUAD dataset. In total, 163 genes were statistically different in mutation frequencies between the high- and low-risk groups (available online: https://cdn.amegroups.cn/static/public/tcr-2025-1015-2.xls). The high-risk group exhibited higher frequencies of lung cancer driver gene mutations such as TP53, SMARCA4, and MMP2 compared with the low-risk group. Additionally, genes involved in key pathways, including NCOR2 in the Notch signaling pathway and COL11A1 in the EPITHELIAL_MESENCHYMAL_TRANSITION pathway, also had high mutation frequencies in the high-risk group (Figure 6A). Survival analysis further indicated that mutations in FMN2, CTTNBP2, SMARCA4, and NCOR2 are associated with a poor prognosis in LUAD (Figure 6B). In the TCGA-LUAD and CPTAC-LUAD cohorts, compared to patients in the low-risk group, those in the high-risk group had a higher total number of non-synonymous mutations (Figure 6C). The HRD score was also higher in the high-risk group within the TCGA-LUAD dataset (Figure 6D). We further analyzed the differences in mutation signatures and observed that DNA mismatch repair deficiency-related (dMMR) mutations (signature 20) and APOBEC somatic mutations (signature 13) were enriched in the high-risk group. Smoking-related mutations (signature 4) did not differ between the two risk groups (Figure 6E). These significant mutational differences were likely contributing factors to the poor prognosis observed in the high-risk group.

Figure 6 Genome alteration landscape of IRPCDS. (A) Waterfall plot depicting the common mutations exhibiting differential frequencies between the high- and low-risk groups. (B) From left to right, Kaplan-Meier survival curves analysis for FMN2, CTTNBP2, SMARCA4, and NCOR2 gene mutations and wild-type groups. (C) Comparison of the total number of non-synonymous mutations between the high- and low-risk groups in TCGA-LUAD (left) and CPTAC-LUAD (right). (D) Comparison of HRD scores between the high- and low-risk groups in TCGA-LUAD. (E) Comparison of Signature.20, Signature.13, and Signature.4 mutation signatures between the high- and low-risk groups in the TCGA-LUAD cohort. *, P<0.05; **, P<0.01; ***, P<0.001. CPTAC, clinical proteomic tumor analysis consortium; HRD, homologous recombination deficiency; IRPCDS, immune-related programmed cell death signature; LUAD, lung adenocarcinoma; OS, overall survival; TCGA, The Cancer Genome Atlas.

scRNA-seq analysis reveals high IRPCDS scores in endothelial cells and T/NK cells within the TME

To investigate the distribution of IRPCDS among cell types in the TME of LUAD, we analyzed scRNA-seq data from 9 LUAD samples. Following quality control and batch effect removal, we identified 115,245 cells and 23 distinct cell clusters. These identified clusters were labeled into different cell types according to the canonical marker genes obtained from the original text (Figure 7A,7B). The IRPCDS for each cluster was calculated based on the coefficients of the genes in the model and the corresponding gene expression data. The top two cell types with the highest IRPCDS scores were Endothelial cells and T/NK cells, while mast cells exhibited the lowest IRPCDS (Figure 7C,7D). To further explore the interactions among cell types within the TME in LUAD, we performed a cell-cell interaction analysis by assessing the probabilities of cell communication. The results revealed significant and robust interactions between T/NK cells and myeloid cells, myeloid and epithelial cells, as well as between tumor-associated fibroblasts and T/NK cells (Figure 7E). Additionally, we investigated the interactions among cell types within the COLLAGEN (37,38), EGF (39), and TNF (40,41) signaling pathways, which were implicated in the onset, progression, and poor prognosis of LUAD. In the COLLAGEN signaling pathway, fibroblasts interacted with multiple cell types, primarily acting as senders and influencers. Epithelial cells predominantly engaged with Mast cells, primarily fulfilling receiving and influencing roles. In the TNF signaling pathway, Myeloid cells mainly interact with T/NK cells, serving as senders, mediators, and influencers (Figure 7F,7G).

Figure 7 scRNA-seq analysis revealed the roles of different cell types in pathway networks. (A) UMAP analysis plot illustrating the identified cell types. (B) The bubble plot illustrates the expression levels of marker genes across various cell types. (C,D) Distribution of IRPCDS across various cell types. (E) The number and strength of interaction in cell-cell communication network. (F) Circos plots showing cell-cell communication interaction in COLLAGEN, EGF, TNF signaling pathway. (G) Heatmap showing the roles of different cell types playing in the pathway network. IRPCDS, immune-related programmed cell death signature; scRNA-seq, single‑cell RNA‑sequencing; UMAP, uniform manifold approximation and projection.

Expression of IRPCDS genes was significantly different between cancer and adjacent non-cancerous samples

These genes in IRPCDS primarily originate from apoptosis, autophagy, lysosome-dependent cell death, ferroptosis, paraptosis, and anoikis (Figure 8A). Among these, the expression levels of genes such as PTTG1IP, GNAI3, KRT8, CDCP1, TRIM6, and G0S2 were positively correlated with better prognosis. Conversely, the expression levels of genes including CIDEB, PEBP1, RNF5, SIRT2, NCOA4, ATP6V1B2, FUCA1, BTK, ADRB2, GPR15, and CX3CR1 were positively correlated with poor prognosis (Figure 8B). In the TCGA-LUAD, CPTAC-LUAD, and GSE31210 cohorts, the mRNA expression levels of PTTG1IP, KRT8, CDCP1, and TRIM6 were elevated in tumor tissues compared with normal tissues, whereas the mRNA expression levels of PEBP1, BTK, and ADRB2, CX3CR1 were decreased in tumor tissues (Figure 8C-8E). Consistently, in the CPTAC-LUAD cohort, we observed elevated protein abundance of PTTG1IP, KRT8, and CDCP1 in tumor tissues, while the protein abundance of BTK and PEBP1 was decreased (Figure 8F). Survival analysis indicated that patients with higher abundance of CDCP1 and GNAI3 proteins exhibited poorer prognosis. Conversely, patients with higher abundance of BTK and SIRT2 proteins demonstrated better prognosis. Although there was a trend towards poorer prognosis in patients with high expression of KRT8 protein, this was not statistically significant (Figure 8G). Furthermore, we demonstrated through the HPA database that the protein expression of PTTG1IP, KRT8, and CDCP1 was also elevated in tumor tissues compared with normal lung tissues (Figure 8H-8J). Furthermore, we verified the expression of the CDCP1 and PEBP1 proteins in 6 LUAD specimens and 6 adjacent non-tumorous tissue samples. Western blot analyses showed that CDCP1 was upregulated and PEBP1 was downregulated in tumor tissues in comparison with adjacent non-tumorous tissues (Figure 8K).

Figure 8 Comparison of gene expression differences between tumor and adjacent non-tumor tissues within the IRPCDS model. (A) Chord diagram showing the PCD mode origins of the 17 genes in IRPCDS. (B) Bar plot showing the coefficients of the 17 genes finally obtained in the combination of LASSO and plsRcox algorithms. (C-E) Box plots showing the mRNA expression differences of the 17 genes in tumor and adjacent non-tumor tissues in the TCGA-LUAD, CPTAC-LUAD, and GSE31210 cohorts, respectively. (F) Box plots showing the protein abundance differences of the genes in IRPCDS in tumor and adjacent non-tumor tissues in the CPTAC-LUAD cohort. (G) From left to right, Kaplan-Meier survival curves analysis for high and low proteins expression groups of CDCP1, GNAI3, KRT8, BTK, and SIRT2, with the optimal cut-off value determined by the survminer package. (H-J) Immunohistochemical staining of PTTG1IP (H), CDCP1(I), and KRT8 (J) in LUAD patients based on HPA, PTTG1IP tumor sample (image available from https://www.proteinatlas.org/ENSG00000183255-PTTG1IP/cancer/lung+cancer#img); PTTG1IP normal sample (image available from https://www.proteinatlas.org/ENSG00000183255-PTTG1IP/tissue/lung#img). CDCP1 tumor sample (image available from https://www.proteinatlas.org/ENSG00000163814-CDCP1/cancer/lung+cancer#img); CDCP1 normal sample (image available from https://www.proteinatlas.org/ENSG00000163814-CDCP1/tissue/lung#img). KRT8 tumor sample (image available from https://www.proteinatlas.org/ENSG00000170421-KRT8/cancer/lung+cancer#img); KRT8 normal sample (image available from https://www.proteinatlas.org/ENSG00000170421-KRT8/tissue/lung#img). Scale bars: 200 μm. (K) Western blot analysis of CDCP1 and PEBP1. NS, P>0.05; *, P<0.05; **, P<0.01; ***, P<0.001. CPTAC, clinical proteomic tumor analysis consortium; HPA, Human Protein Atlas; IRPCDS, immune-related programmed cell death signature; LASSO, least absolute shrinkage and selection operator; LUAD, lung adenocarcinoma; PCD, programmed cell death; plsRcox, partial least squares regression for Cox proportional hazards; TCGA, The Cancer Genome Atlas.

Discussion

PCD plays a pivotal role in the occurrence, progression, prognosis, and response to immunotherapy in tumors (19,20,42,43). The relationship between PCD as a form of cell death in LUAD is poorly understood. In this study, we systematically compiled 18 PCD signatures and elucidated their roles in LUAD. Our findings revealed that PCD signatures serve as significant risk factors for OS in LUAD, strongly correlating with the activation of cancer-related pathways and immune cell infiltration.

To enhance precise treatment decisions and prognosis prediction in LUAD, we curated immune-related PCD genes and employed 10 machine learning algorithms in 101 combinations to develop the IRPCDS. We validated its prognostic performance across multiple validation cohorts, effectively stratifying LUAD patients into distinct risk groups to inform clinical interventions. Furthermore, we benchmarked the model against clinical features and previously published signatures, revealing that IRPCDS outperformed clinicopathological factors and previously published signatures. These results suggest that the IRPCDS had higher predictive accuracy and robust stability for prognosis.

Over the past decade, immunotherapy has transformed cancer treatment, achieving significant success in oncology (8,44). However, only a small fraction (less than 10–20%) of patients exhibit sustained responses to ICIs. Current predictive markers for immunotherapy, such as TMB, require whole exome or whole genome sequencing, making them costly. Additionally, MSI has limited prevalence in LUAD, restricting its clinical utility. To address these challenges, we employed WGCNA combined with consensus clustering algorithms to identify immune-related PCD genes, thereby developing the IRPCDS. IRPCDS has demonstrated a significant correlation with the expression of a wide array of immune checkpoint-related genes. Notably, the TIME exhibits marked differences between high- and low-risk groups stratified by IRPCDS. Patients in the low-risk group exhibit higher total stromal scores, immune scores, and immune infiltration, but lower TIDE score and higher CYT activity. These results suggested that low-risk patients may benefit from ICB therapy. We further validated these findings in two additional immunotherapy cohorts, confirming that risk scores were lower in the immunotherapy responders compared with non-responders. Consequently, IRPCDS held potential as a practical tool for predicting immunotherapy responses. The differential expression of genes between cancer and adjacent tissues was evaluated at both the transcriptomic and proteomic levels. Our findings consistently showed mRNA and protein expression of PTTG1IP, KRT8, and CDCP1 were elevated in tumor tissues, while BTK and PEBP1 exhibited decreased expression. Survival analysis indicated that patients with higher CDCP1 protein levels had a poorer prognosis, whereas those with lower BTK and PEBP1 protein levels had a better prognosis. CUB domain-containing protein 1 (CDCP1), encoding a transmembrane protein involved in tyrosine phosphorylation-dependent cellular events, is relevant to tumor invasion and metastasis. Consistent with our findings, Mamat et al. reported that elevated CDCP1 expression correlates with poor prognosis in LUAD (45). Dagnino et al. found that circulating CDCP1 protein levels were associated with an increased risk of lung cancer (46). Lin et al. demonstrated that CDCP1 was linked to lung cancer metastasis (47). Phosphatidylethanolamine binding protein 1 (PEBP1), also known as Raf kinase inhibitory protein (RKIP), a member of the phosphatidylethanolamine-binding protein family, modulates critical signaling pathways, including MAPK (48) and NF-κB (48,49). Studies have shown that RKIP as a tumor suppressor, and its low expression levels are associated with tumor initiation, invasion, and metastasis, as well as with the inhibition of lung squamous cell carcinoma differentiation (50). In addition to its role in lung cancer, PEBP1 also plays a regulatory role in various other cancers, including hepatocellular carcinoma (51), breast cancer (52,53) and prostate cancer (54). Bruton tyrosine kinase (BTK) plays a crucial role in B-cell development and influences the microenvironment of solid tumors, such as squamous cell carcinoma and pancreatic cancer (55). The introduction of BTK inhibitors into clinical practice is rapidly transforming the treatment of B-cell malignancies in chronic lymphocytic leukemia (56).

There are several limitations in this study. The data primarily rely on retrospective analyses from public databases, lacking validation from prospective multicenter cohorts. The predictive value of IRPCDS in immunotherapy has only been validated in non-small cell lung cancer and melanoma cohorts. Further validation using more LUAD immunotherapy datasets is necessary. While we validated the protein expression of CDCP1 and PEBP1 using clinical specimens, further in vivo and in vitro experiments are essential to elucidate the roles of the 17 IRPCDS genes in the prognosis of LUAD and to understand their complex mechanisms in modulating immunotherapy outcomes. Therefore, more multicenter randomized controlled trials with high quality, large sample size, and adequate follow-up are required for additional validation.

Conclusions

We systematically explored the role of PCD signatures in LUAD. Leveraging a comprehensive suite of bioinformatics and machine learning algorithms, we developed a robust and powerful signature for predicting prognosis and response to ICIs by integrating 18 PCD signatures. The IRPCDS model represents a promising tool to enhance decision-making and surveillance protocols for individual LUAD patients.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1015/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1015/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1015/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1015/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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Qingdao Traditional Chinese Medicine Hospital (No. 2022HC05LS001) and informed consent was obtained from all individual participants.

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. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
2. Brody H. Lung cancer. Nature 2020;587:S7. [Crossref] [PubMed]
3. Travis WD, Brambilla E, Noguchi M, et al. International association for the study of lung cancer/american thoracic society/european respiratory society international multidisciplinary classification of lung adenocarcinoma. J Thorac Oncol 2011;6:244-85. [Crossref] [PubMed]
4. Little AG, Gay EG, Gaspar LE, et al. National survey of non-small cell lung cancer in the United States: epidemiology, pathology and patterns of care. Lung Cancer 2007;57:253-60. [Crossref] [PubMed]
5. Xu S, Wang Y, Ren F, et al. Impact of genetic alterations on outcomes of patients with stage I nonsmall cell lung cancer: An analysis of the cancer genome atlas data. Cancer Med 2020;9:7686-94. [Crossref] [PubMed]
6. Alvarez-Jimenez C, Sandino AA, Prasanna P, et al. Identifying Cross-Scale Associations between Radiomic and Pathomic Signatures of Non-Small Cell Lung Cancer Subtypes: Preliminary Results. Cancers (Basel) 2020;12:3663. [Crossref] [PubMed]
7. Ludwig JA, Weinstein JN. Biomarkers in cancer staging, prognosis and treatment selection. Nat Rev Cancer 2005;5:845-56. [Crossref] [PubMed]
8. Zhang Y, Zhang Z. The history and advances in cancer immunotherapy: understanding the characteristics of tumor-infiltrating immune cells and their therapeutic implications. Cell Mol Immunol 2020;17:807-21. [Crossref] [PubMed]
9. Reckamp KL, Redman MW, Dragnev KH, et al. Phase II Randomized Study of Ramucirumab and Pembrolizumab Versus Standard of Care in Advanced Non-Small-Cell Lung Cancer Previously Treated With Immunotherapy-Lung-MAP S1800A. J Clin Oncol 2022;40:2295-306. [Crossref] [PubMed]
10. Gibney GT, Weiner LM, Atkins MB. Predictive biomarkers for checkpoint inhibitor-based immunotherapy. Lancet Oncol 2016;17:e542-51. [Crossref] [PubMed]
11. Chan TA, Yarchoan M, Jaffee E, et al. Development of tumor mutation burden as an immunotherapy biomarker: utility for the oncology clinic. Ann Oncol 2019;30:44-56. [Crossref] [PubMed]
12. Samstein RM, Lee CH, Shoushtari AN, et al. Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nat Genet 2019;51:202-6. [Crossref] [PubMed]
13. Ganesh K. Optimizing immunotherapy for colorectal cancer. Nat Rev Gastroenterol Hepatol 2022;19:93-4. [Crossref] [PubMed]
14. Casalegno Garduño R, Spitschak A, Pannek T, et al. CD8+ T Cell Subsets as Biomarkers for Predicting Checkpoint Therapy Outcomes in Cancer Immunotherapy. Biomedicines 2025;13:930. [Crossref] [PubMed]
15. Li H, Zandberg DP, Kulkarni A, et al. Distinct CD8(+) T cell dynamics associate with response to neoadjuvant cancer immunotherapies. Cancer Cell 2025;43:757-775.e8. [Crossref] [PubMed]
16. Mino-Kenudson M, Schalper K, Cooper W, et al. Predictive Biomarkers for Immunotherapy in Lung Cancer: Perspective From the International Association for the Study of Lung Cancer Pathology Committee. J Thorac Oncol 2022;17:1335-54. [Crossref] [PubMed]
17. Cancer immunotherapy: the quest for better biomarkers. Nat Med 2022;28:2437. [Crossref] [PubMed]
18. Tang D, Kang R, Berghe TV, et al. The molecular machinery of regulated cell death. Cell Res 2019;29:347-64. [Crossref] [PubMed]
19. Bedoui S, Herold MJ, Strasser A. Emerging connectivity of programmed cell death pathways and its physiological implications. Nat Rev Mol Cell Biol 2020;21:678-95. [Crossref] [PubMed]
20. Fuchs Y, Steller H. Programmed cell death in animal development and disease. Cell 2011;147:742-58. [Crossref] [PubMed]
21. Wang J, Su H, Wang M, et al. Pyroptosis and the fight against lung cancer. Med Res Rev 2025;45:5-28. [Crossref] [PubMed]
22. Huang C, Li J, Wu R, et al. Targeting pyroptosis for cancer immunotherapy: mechanistic insights and clinical perspectives. Mol Cancer 2025;24:131. [Crossref] [PubMed]
23. Xiao Z, Xie J, Zhao X, et al. Role of Pyroptosis in inflammatory bowel disease. Int Immunopharmacol 2025;155:114619. [Crossref] [PubMed]
24. Ma X, Deng Z, Li Z, et al. Leveraging a disulfidptosis/ferroptosis-based signature to predict the prognosis of lung adenocarcinoma. Cancer Cell Int 2023;23:267. [Crossref] [PubMed]
25. Huang F, Zheng Y, Li X, et al. Ferroptosis-related gene AKR1C1 predicts the prognosis of non-small cell lung cancer. Cancer Cell Int 2021;21:567. [Crossref] [PubMed]
26. Ma C, Gu Z, Ding W, et al. Crosstalk between copper homeostasis and cuproptosis reveals a lncRNA signature to prognosis prediction, immunotherapy personalization, and agent selection for patients with lung adenocarcinoma. Aging (Albany NY) 2023;15:13504-41. [Crossref] [PubMed]
27. Petralia F, Ma W, Yaron TM, et al. Pan-cancer proteogenomics characterization of tumor immunity. Cell 2024;187:1255-1277.e27. [Crossref] [PubMed]
28. Charoentong P, Finotello F, Angelova M, et al. Pan-cancer Immunogenomic Analyses Reveal Genotype-Immunophenotype Relationships and Predictors of Response to Checkpoint Blockade. Cell Rep 2017;18:248-62. [Crossref] [PubMed]
29. Wilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics 2010;26:1572-3. [Crossref] [PubMed]
30. Langfelder P, Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 2008;9:559. [Crossref] [PubMed]
31. Newman AM, Liu CL, Green MR, et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods 2015;12:453-7. [Crossref] [PubMed]
32. Li T, Fan J, Wang B, et al. TIMER: A Web Server for Comprehensive Analysis of Tumor-Infiltrating Immune Cells. Cancer Res 2017;77:e108-10. [Crossref] [PubMed]
33. Jin S, Guerrero-Juarez CF, Zhang L, et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun 2021;12:1088. [Crossref] [PubMed]
34. Cao K, Zhu J, Lu M, et al. Analysis of multiple programmed cell death-related prognostic genes and functional validations of necroptosis-associated genes in oesophageal squamous cell carcinoma. EBioMedicine 2024;99:104920. [Crossref] [PubMed]
35. Qin H, Abulaiti A, Maimaiti A, et al. Integrated machine learning survival framework develops a prognostic model based on inter-crosstalk definition of mitochondrial function and cell death patterns in a large multicenter cohort for lower-grade glioma. J Transl Med 2023;21:588. [Crossref] [PubMed]
36. Bagaev A, Kotlov N, Nomie K, et al. Conserved pan-cancer microenvironment subtypes predict response to immunotherapy. Cancer Cell 2021;39:845-865.e7. [Crossref] [PubMed]
37. Hanley CJ, Waise S, Ellis MJ, et al. Single-cell analysis reveals prognostic fibroblast subpopulations linked to molecular and immunological subtypes of lung cancer. Nat Commun 2023;14:387. [Crossref] [PubMed]
38. Zhu G, Wang Y, Wang Y, et al. Myofibroblasts derived type V collagen promoting tissue mechanical stress and facilitating metastasis and therapy resistance of lung adenocarcinoma cells. Cell Death Dis 2024;15:493. [Crossref] [PubMed]
39. Zhang F, Cui X, Yang K, et al. Activin A inhibits the migration of human lung adenocarcinoma A549 cells induced by EGF. Int Immunopharmacol 2024;142:113170. [Crossref] [PubMed]
40. Zhang C, Zhang G, Sun N, et al. Comprehensive molecular analyses of a TNF family-based signature with regard to prognosis, immune features, and biomarkers for immunotherapy in lung adenocarcinoma. EBioMedicine 2020;59:102959. [Crossref] [PubMed]
41. Fujiwara Y, Takahashi RU, Saito M, et al. Oncofetal IGF2BP3-mediated control of microRNA structural diversity in the malignancy of early-stage lung adenocarcinoma. Proc Natl Acad Sci U S A 2024;121:e2407016121. [Crossref] [PubMed]
42. Lu J, Tai Z, Wu J, et al. Nanomedicine-induced programmed cell death enhances tumor immunotherapy. J Adv Res 2024;62:199-217. [Crossref] [PubMed]
43. Hsu SK, Li CY, Lin IL, et al. Inflammation-related pyroptosis, a novel programmed cell death pathway, and its crosstalk with immune therapy in cancer treatment. Theranostics 2021;11:8813-35. [Crossref] [PubMed]
44. Riley RS, June CH, Langer R, et al. Delivery technologies for cancer immunotherapy. Nat Rev Drug Discov 2019;18:175-96. [Crossref] [PubMed]
45. Mamat S, et al. Prognostic significance of CUB domain containing protein expression in endometrioid adenocarcinoma. Oncol Rep 2010;23:1221-1227. [Crossref]
46. Dagnino S, Bodinier B, Guida F, et al. Prospective Identification of Elevated Circulating CDCP1 in Patients Years before Onset of Lung Cancer. Cancer Res 2021;81:3738-48. [Crossref] [PubMed]
47. Lin CY, Chen HJ, Huang CC, et al. ADAM9 promotes lung cancer metastases to brain by a plasminogen activator-based pathway. Cancer Res 2014;74:5229-43. [Crossref] [PubMed]
48. Yang X, Wang Y, Lu P, et al. PEBP1 suppresses HIV transcription and induces latency by inactivating MAPK/NF-κB signaling. EMBO Rep 2020;21:e49305. [Crossref] [PubMed]
49. Escara-Wilke J, Yeung K, Keller ET. Raf kinase inhibitor protein (RKIP) in cancer. Cancer Metastasis Rev 2012;31:615-20. [Crossref] [PubMed]
50. Zhu C, Wang Q, Xie J, et al. Expression and significance of RKIP and E-cadherin in lung squamous cell carcinoma. Pathol Oncol Res 2013;19:19-26. [Crossref] [PubMed]
51. Xu YF, Yi Y, Qiu SJ, et al. PEBP1 downregulation is associated to poor prognosis in HCC related to hepatitis B infection. J Hepatol 2010;53:872-9. [Crossref] [PubMed]
52. Floris A, Luo J, Frank J, et al. Star-related lipid transfer protein 10 (STARD10): a novel key player in alcohol-induced breast cancer progression. J Exp Clin Cancer Res 2019;38:4. [Crossref] [PubMed]
53. Datar I, Feng J, Qiu X, et al. RKIP Inhibits Local Breast Cancer Invasion by Antagonizing the Transcriptional Activation of MMP13. PLoS One 2015;10:e0134494. [Crossref] [PubMed]
54. Fu Z, Smith PC, Zhang L, et al. Effects of raf kinase inhibitor protein expression on suppression of prostate cancer metastasis. J Natl Cancer Inst 2003;95:878-89. [Crossref] [PubMed]
55. Burger JA, Wiestner A. Targeting B cell receptor signalling in cancer: preclinical and clinical advances. Nat Rev Cancer 2018;18:148-67. [Crossref] [PubMed]
56. Shadman M. Diagnosis and Treatment of Chronic Lymphocytic Leukemia: A Review. JAMA 2023;329:918-32. [Crossref]