Abstract
Background:
The roles of stem cells in lung adenocarcinoma (LUAD) progression and therapeutic resistance have been recognized, yet their impact on patient prognosis and immunotherapy response remains unclear.
Methods:
Single-cell RNA sequencing was performed to identify stem cell populations characterized by high expression of MKI67 and STMN1. Key marker genes were identified using the FindAllMarkers function, and these genes were subsequently analyzed for mutations, copy number variations, and prognostic significance in LUAD patients. Multiple machine learning algorithms were systematically compared in order to develop an optimal prognostic model. The predictive performance of the model was validated across seven independent LUAD cohorts and immunotherapy datasets. Patterns of immune infiltration were assessed using various computational approaches and were further validated in an internal hospital cohort.
Results:
Through comprehensive machine learning optimization, CoxBoost+Enet (alpha=0.7) was identified as the optimal model, incorporating seven key stem cell–related genes and designated as the Stem Cell Prognostic Model (SCPM). Patients were consistently stratified into high- and low-SCPM groups across all seven validation cohorts, with poorer overall survival observed in the high-SCPM group. Predictive accuracy was demonstrated by ROC analysis (AUC > 0.65), while clear group separation was confirmed through PCA based on the seven-gene signature. Notably, immunotherapy response was also predicted by SCPM, with inferior outcomes observed in high-SCPM patients following treatment with immune checkpoint inhibitors. Significantly lower immune cell infiltration, characteristic of “cold” tumors, was detected in high-SCPM patients by multiple immune infiltration algorithms. These findings were further validated in the internal cohort, where reduced CD8+ T cell infiltration was observed in high-SCPM patients.
Conclusion:
A stem cell–based prognostic model (SCPM) was constructed and validated, enabling accurate prediction of survival and immunotherapy response in LUAD patients. Patients with immunologically “cold” tumors, as identified by the SCPM, may benefit from alternative therapeutic strategies.
Introduction
Lung cancer is the leading cause of cancer-related morbidity and mortality worldwide, with over 2.2 million new cases and approximately 1.8 million deaths annually according to the World Health Organization (1, 2). Lung adenocarcinoma (LUAD), as the predominant histological subtype of non-small cell lung cancer, accounts for 40-50% of all lung cancers (3, 4). Despite significant advances in molecular targeted therapy and immunotherapy in recent years, the 5-year survival rate for LUAD patients remains low (5, 6). Tumor heterogeneity, early metastatic tendency, treatment resistance, and disease recurrence continue to be key factors limiting long-term patient survival, creating an urgent need for novel biomarkers to guide precision treatment strategies.
Cancer stem cell (CSC) theory provides a new perspective for understanding tumor heterogeneity and treatment resistance (7–10). Cancer stem cells are a subset of tumor cells with self-renewal capacity and multipotent differentiation potential. They play crucial roles not only in tumor initiation, progression, metastasis, and recurrence but also serve as an important source of tumor heterogeneity (11). In lung cancer, cancer stem cells typically exhibit resistance to chemotherapy and radiotherapy, enabling survival under treatment pressure and causing disease recurrence (12). MKI67, as a classic marker of cell proliferation, has expression levels closely associated with tumor aggressiveness and patient prognosis. STMN1, as an important microtubule-regulating protein, participates in cell cycle regulation and anti-apoptotic processes through microtubule dynamics control, playing a key role in maintaining cancer stem cell properties. The rapid development of single-cell RNA sequencing technology provides powerful tools for in-depth investigation of tumor heterogeneity and cancer stem cell characteristics, enabling researchers to dissect the complexity of the tumor microenvironment at the single-cell level (13).
Immunotherapy, particularly the application of immune checkpoint inhibitors, has brought revolutionary changes to lung cancer treatment (14). Immune checkpoint inhibitors such as PD-1/PD-L1 and CTLA-4 antibodies reactivate the body’s anti-tumor immune response by relieving tumor cell suppression of the immune system (15, 16). However, immunotherapy also faces significant challenges: only approximately 20-30% of LUAD patients achieve durable clinical benefit from immunotherapy, while the majority exhibit primary or acquired resistance (17). The immune cell infiltration status in the tumor microenvironment, particularly the infiltration levels of effector immune cells such as CD8+ T cells and CD4+ T cells, is considered a key factor affecting immunotherapy efficacy. “Hot tumors” typically have abundant immune cell infiltration and respond well to immunotherapy, while “cold tumors” are characterized by sparse immune cell infiltration, leading to poor immunotherapy outcomes. Studies suggest that cancer stem cells may influence the immune status of the tumor microenvironment through mechanisms such as secreting immunosuppressive factors and recruiting regulatory T cells (18).
Based on the above background, this study aims to systematically identify cancer stem cell-like subpopulations in LUAD and deeply analyze their molecular characteristics and biological functions by integrating single-cell RNA sequencing data with large-scale clinical cohort data. We employ advanced machine learning algorithms to construct a Stem Cell Prognostic Model (SCPM) based on stem cell-related genes and validate its predictive performance across multiple independent cohorts. Additionally, we will explore the potential value of this model in predicting immunotherapy response and analyze the relationship between cancer stem cell characteristics and the tumor immune microenvironment. We hypothesize that tumors with high cancer stem cell features exhibit “cold tumor” characteristics and show poor responsiveness to immunotherapy. The findings of this study will provide novel biomarkers and theoretical foundations for precision diagnosis and treatment of LUAD.
Method
Dataset source
LUAD gene expression profiles, somatic single nucleotide variants (SNVs), and copy number alterations (CNAs) were obtained from The Cancer Genome Atlas (TCGA) repository, while normal lung tissue expression profiles were retrieved from the Genotype-Tissue Expression (GTEx) database for comparative analysis. Recurrent genomic amplifications and deletions were identified through the application of GISTIC 2.0 algorithm (https://software.broadinstitute.org) to TCGA-derived CNA data. Six independent LUAD datasets were consolidated from the Gene Expression Omnibus (GEO) repository: GSE13213 (19) (n=117), GSE26939 (20) (n=115), GSE29016 (21) (n=39), GSE30219 (22) (n=85), GSE31210 (23) (n=226), and GSE42127 (24) (n=133). Inter-cohort batch effects were eliminated using the ComBat normalization method (25), followed by standard data preprocessing procedures. To assess the model’s performance in immunotherapy contexts, three immunotherapy-associated NSCLC datasets were compiled: POPLAR (26) (n=59), OAK (26) (n=257), SU2C (27)(n=130).
scRNA-seq data processing
Single-cell transcriptomic data were sourced from GEO: GSE241934. Raw transcriptomic profiles were preprocessed using the Seurat R package (28) (version 4.2.0). Analytical inclusion criteria were established with a minimum expression threshold requiring each gene to be detected in at least 10 cells per sample. Cell quality filtering was performed according to predefined parameters: cells were excluded if they exhibited expression of more than 5000 or fewer than 200 genes, or if mitochondrial genome-derived unique molecular identifiers (UMIs) comprised more than 10% of total counts. Sample integration was achieved through the harmony R package implementation. The analytical workflow encompassed highly variable gene identification for principal component analysis (PCA), followed by dimensionality reduction employing the top 30 principal components via t-distributed Stochastic Neighbor Embedding (t-SNE) methodology. Subpopulation-specific expression signatures were characterized using the “FindAllMarkers” function, with cellular phenotype annotation performed based on established lineage-specific markers from previous studies (29).
Development of cancer stem cell-derived prognostic signature
Through single-cell transcriptomic profiling, MKI67+STMN1+ cancer stem cell subsets were characterized, with downstream investigations performed utilizing their distinctive molecular markers. Gene expression differences were assessed via the limma software package (30), applying cutoff criteria of FDR < 0.05 and |log2FC| > 1. Mutational profiles of target genes were systematically examined through maftools package implementation. Visual representation of frequently altered stem cell-related genes was achieved using Oncoplot methodology, whereas prevalent mutational signatures were uncovered via signature profiling techniques. Machine learning methodologies serve essential functions in prognostic biomarker construction (31, 32). For developing a reliable Stem Cell Prognostic Model (SCPM), preliminary univariate Cox proportional hazards analysis was executed to detect survival-related genetic elements. The refinement process encompassed thorough assessment of diverse algorithmic strategies via 10-fold cross-validation techniques (33). Multiple computational methodologies were integrated within the analytical framework: sequential Cox modeling, Lasso constraint methods, Ridge penalization, Cox-adapted partial least squares regression (plsRcox), CoxBoost algorithms, random survival forests (RSF), gradient boosting machines (GBM), elastic net (Enet) constraints, supervised principal component (SuperPC) techniques, and survival-oriented support vector machines (survival-SVM). This comprehensive assessment was structured to determine the most effective SCPM architecture, employing concordance index (C-index) as the principal evaluation criterion. Performance validation of the constructed SCPM was accomplished using temporal ROC curve assessment, Kaplan-Meier survival estimation, and principal component analysis (PCA) techniques.
Immunological profile assessment
Assessment of immunotherapy sensitivity in LUAD patients was conducted using immunophenoscore (IPS) calculations through The Cancer Immunome Atlas database (https://tcia.at/home) (34). Characterization of the tumor immune microenvironment was performed via ssGSEA methodology, which measured immune cell penetration dynamics and immunological pathway activity levels in malignant tissue samples. Detailed immune cell infiltration data from TCGA cohorts were obtained using the TIMER2.0 resource (35), which integrated results from various analytical algorithms.
Clinical sample acquisition and transcriptomic analysis
Tissue specimen procurement was granted ethical clearance by the Medical Ethics Board of the First Affiliated Hospital of Nanjing Medical University. These specimens were verified as pulmonary adenocarcinoma by pathological specialists, retrieved during surgical procedures, and subsequently transported to Oncocare Inc. (Suzhou, China) for transcriptomic sequencing analysis.
Study population and specimen acquisition
Tissue specimens consisted of formalin-fixed paraffin-embedded sections obtained from the Pathology Department of The First Affiliated Hospital of Nanjing Medical University. Histopathological confirmation of LUAD diagnosis was established for all samples, with inclusion limited to therapy-naive patients prior to operative procedures. Ethical approval for the investigation was granted by the Institutional Review Board of The First Affiliated Hospital of Nanjing Medical University, and informed consent documentation was secured from each participant. These histological sections were processed for multiplex immunohistochemical examination.
Multi-marker immunohistochemical staining
The correlation between prognostic scores and immunocyte penetration was assessed through multiplex fluorescent immunolabeling techniques. Specimen preparation commenced with dewaxing in xylene solution, subsequently followed by tissue rehydration using graduated alcohol concentrations. Following epitope unmasking and protein blocking with 5% caprine serum, tissue sections were subjected to consecutive primary antibody incubations: CD4 (dilution 1:500, Cat# ab133616), CD8 (dilution 1:2000, Cat# ab217344), and CD20 (dilution 1:100, Cat# ab64088), after which fluorescent-labeled secondary antibodies were applied. Cell nuclei visualization was accomplished using DAPI counterstain.
Statistical methods
Data analysis was conducted using R software version 4.2.0. Two-group comparisons were analyzed through unpaired Student’s t-test for parametric data, whereas Mann-Whitney U test was utilized for non-parametric distributions. Multi-group analyses were executed using one-way ANOVA with subsequent Tukey’s multiple comparison testing for parametric data, while Kruskal-Wallis analysis followed by Dunn’s multiple comparison procedure was applied for non-parametric datasets. Pearson’s correlation coefficients were calculated to assess variable relationships. Results were expressed as mean ± standard deviation (SD). The threshold for statistical significance was established at P < 0.05 (*P < 0.05, **P < 0.01, ***P < 0.001).
Result
Single-cell RNA sequencing analysis reveals expression patterns and mutational characteristics of stem cell marker genes
To deeply explore the role of stem cells in LUAD development and progression, we first analyzed the distribution characteristics of various cell types in the tumor microenvironment using single-cell RNA sequencing technology. Through t-SNE dimensionality reduction visualization (Figure 1A), we systematically examined the expression distribution of multiple cell type–specific markers, including B cells (CD79A), T cells (CD3D), myeloid cells (LYZ), epithelial cells (EPCAM), endothelial cells (CLDN5), fibroblasts (COL1A1), mast cells (MS4A2), natural killer cells (NKG7), as well as proliferation-associated stem cell marker genes MKI67 and STMN1. Based on the expression profiles of these marker genes, we successfully identified the major cell types within the tumor microenvironment (Figure 1B), providing a foundation for further investigation of stem cell roles in the complex tumor microenvironment. It should be noted that the “proliferative stem cells” defined in this study are a population selected based on high expression of broad proliferation-related markers (MKI67 and STMN1), and actually comprise a mixture of cell subtypes with high proliferative activity, rather than strictly corresponding to classical definitions of cancer stem cells or mesenchymal stem cells. Building upon the comprehensive characterization of the tumor microenvironment’s cellular composition, we further focused on analyzing the mutational characteristics of stem cell marker genes (Figure 1C). For the selection of marker genes included in the mutation analysis, we applied the following parameters: only.pos = TRUE, min.pct = 0.25, and logfc.threshold = 0.25. Mutational spectrum analysis targeting stem cell-related genes revealed that missense mutations were the predominant variant type, suggesting these mutations may directly affect stem cell biological properties by altering protein function. Variant type analysis showed that single nucleotide polymorphisms (SNPs) dominated among stem cell marker genes, while insertions (INS) and deletions (DEL) were relatively rare. Single nucleotide variant (SNV) analysis revealed that C>A and C>T transitions were the most common mutation types in stem cell marker genes, potentially related to defects in DNA damage repair mechanisms in stem cells. Analysis of mutational burden distribution across samples revealed significant inter-patient variability in the mutation frequency of these genes. Finally, we constructed a comprehensive mutational landscape of stem cell marker genes (Figure 1D), specifically displaying the top 30 most frequently mutated stem cell marker genes across 522 LUAD samples, with 405 samples (77.59%) harboring at least one stem cell-related gene alteration. Among these highly mutated stem cell marker genes, KRAS showed the highest mutation rate (29%). As a key signaling pathway gene regulating stem cell self-renewal and proliferation, KRAS mutations may directly activate the oncogenic transformation capacity of stem cells. KEAP1, as the second most frequently mutated gene, participates in oxidative stress response regulation, and its mutations may enhance the antioxidant capacity and survival advantage of stem cells. SYNE2, the third most frequently mutated gene, primarily participates in nuclear envelope structure maintenance and nuclear morphology regulation, and its mutations may affect nucleocytoplasmic interactions and gene expression regulatory mechanisms in stem cells. This specialized mutational landscape analysis of stem cell marker genes provides important molecular foundations for understanding the molecular aberrant mechanisms of tumor stem cells and developing precise stem cell-targeted therapeutic strategies.
Figure 1
Comprehensive analysis of stem cell marker gene expression alterations and associated biological pathways
To gain deeper insights into the expression changes of stem cell marker genes in LUAD, we performed systematic differential expression analysis of stem cell-related genes between GTEx normal lung tissues and TCGA LUAD tissues (Figure 2A). The heatmap results revealed complex expression regulatory patterns of stem cell marker genes in tumor tissues. Among these, key genes including TMEM106C, ECT2, PSMD2, STIL, and TTK showed significant upregulation in tumor tissues, and their high expression may be closely associated with abnormal activation of tumor stem cells, cell cycle regulation, and enhanced self-renewal capacity. Notably, the ALDOA gene exhibited downregulation in tumor tissues, suggesting that different stem cell marker genes may play distinct regulatory roles during tumorigenesis. Further analysis of the chromosomal localization characteristics of these differentially expressed stem cell marker genes (Figure 2B) revealed that these genes are widely distributed across various chromosomes, with genes significantly highly expressed in tumors (p<0.05, marked by red dots) mainly concentrated in specific chromosomal regions. The central PCA analysis plot displays the sample distribution pattern from seven LUAD transcriptomic datasets after batch effect correction, demonstrating good consistency in stem cell marker gene expression patterns across different datasets, providing a reliable data foundation for subsequent functional analysis. To elucidate the biological functions of these differentially expressed stem cell marker genes, we performed KEGG pathway enrichment analysis (Figure 2C). The results showed that differentially expressed genes were mainly enriched in core pathways related to cell proliferation regulation, including cell cycle, DNA replication, and p53 signaling pathway, while also being significantly enriched in metabolic reprogramming-related pathways such as glycolysis/gluconeogenesis and HIF-1 signaling pathway. The enrichment of these pathways suggests that abnormal expression of stem cell marker genes may affect tumor stem cell function maintenance by simultaneously regulating cell proliferation and metabolic reprogramming. GO functional enrichment analysis further revealed the specific functional characteristics of these genes (Figure 2D). In terms of biological processes, differentially expressed genes mainly participate in nuclear division, cell division, and other cell proliferation-related processes; regarding cellular components, these genes are primarily localized to chromosomes, centromeric regions, and other cellular structures closely related to cell division; in terms of molecular functions, differentially expressed genes mainly possess ATP-dependent DNA activity, microtubule motor activity, and other functions. These functional enrichment results further confirm the important role of stem cell marker genes in regulating cell division and proliferation, which are core stem cell functions.
Figure 2
Construction and validation of stem cell prognostic model
To deeply explore the genomic characteristics of stem cell marker genes and construct a reliable prognostic prediction model, we first analyzed the chromosomal distribution and copy number variation characteristics of these differentially expressed genes (Figure 3A). The results revealed that stem cell marker genes are widely distributed across various chromosomes, with amplification (AMP, orange) and deletion (DEL, blue) events showing specific patterns in different chromosomal regions. Notably, multiple key genes such as S100A10, UBE2T, and ANLN predominantly exhibited gene amplification, while some genes tended to undergo deletion events. This copy number variation pattern may be closely related to the expression changes and functional abnormalities of these genes in tumors. Through univariate Cox regression analysis in the TCGA cohort (Figure 3B), we found that all stem cell marker genes had hazard ratios (HR) greater than 1, indicating that high expression of these genes was consistently associated with poor prognosis in LUAD patients, providing a foundation for prognostic model construction. Based on these prognostically valuable stem cell marker genes, we employed a random combination strategy of multiple machine learning algorithms to construct prognostic prediction models (Figure 3C). By systematically comparing the performance of different algorithm combinations in the TCGA training cohort and six GEO validation cohorts, we used the C-index as an evaluation metric to measure model prediction accuracy. The heatmap displays C-index values of various algorithm combinations across different cohorts, with colors ranging from blue to red representing C-index values from low to high. Considering both the number of genes included in the model and the C-index performance comprehensively, we ultimately selected the CoxBoost+Enet[alpha=0.7] combination as the optimal model and named it SCPM (Stem Cell Prognostic Model). This model not only demonstrated good predictive performance in the training cohort but also showed stable prognostic prediction capability across multiple independent validation cohorts.
Figure 3
Validation of SCPM prognostic performance and immunotherapy prediction value
To validate the robustness and generalizability of the SCPM model, we first confirmed its performance in the TCGA training cohort (Supplementary Figure S1). Patients were stratified into high-SCPM and low-SCPM groups based on SCPM scores, demonstrating excellent prognostic stratification with significant survival differences (Log-rank p < 0.0001) (Supplementary Figure S1A). Time-dependent ROC analysis showed good predictive accuracy with AUC values of 0.68, 0.68, and 0.64 for 1-, 3-, and 5-year survival predictions, respectively (Supplementary Figure S1B). Principal component analysis based on SCPM signature genes revealed distinct clustering patterns between high-SCPM and low-SCPM groups (Supplementary Figure S1C), indicating that the model effectively captures biological differences related to stem cell characteristics. Subsequently, we systematically evaluated SCPM prognostic performance across six independent GEO validation cohorts (Figure 4). Kaplan-Meier survival analysis consistently demonstrated significant prognostic stratification capability across all validation datasets (Figure 4A). SCPM score-based stratification effectively distinguished patients into prognostically distinct subgroups, further confirming the robust discriminatory power of the SCPM scoring system. Time-dependent ROC analysis (Figure 4B) further validated the predictive accuracy of SCPM across multiple time points, demonstrating good predictive performance in all validation cohorts. It should be noted that, since most patients in the GSE31210 cohort had survival times exceeding one year, it was not possible to calculate the ROC curve and AUC value at the 1-year time point for this cohort. To gain deeper insights into the biological foundation of the SCPM model, we performed principal component analysis based on the expression profiles of 7 SCPM signature genes across six validation cohorts (Figures 5A–F). Results showed that patient samples could be clearly separated into two distinct clusters based on SCPM signature gene expression patterns across all validation cohorts, with high-SCPM (orange) and low-SCPM (blue) groups showing clear separation in principal component space, further validating the molecular biological rationale and cross-cohort consistency of the SCPM model.
Figure 4
Figure 5
Considering the close relationship between stem cell characteristics and the tumor immune microenvironment, we further analyzed the association between SCPM scores and immunotherapy sensitivity (Figures 5G–J). Using The Cancer Immunome Atlas (TCIA) scoring system, we systematically evaluated the predicted responsiveness of different SCPM groups to immune checkpoint inhibitors across multiple immunotherapy subtypes. The results showed that the low-SCPM group demonstrated better performance in several immunotherapy response prediction metrics. Further subgroup analysis revealed that this benefit was mainly observed in patients receiving CTLA-4 inhibitors or double-negative immunotherapy regimens, while no statistically significant differences were found in other immunotherapy subtypes. These findings suggest the potential value and limitations of the SCPM model in guiding precision immunotherapy, indicating that patients in the low-SCPM group may be more suitable for specific types of immunotherapeutic strategies.
Validation of SCPM model in immunotherapy cohorts
To further validate the clinical utility of SCPM in the context of immunotherapy, we evaluated its prognostic performance across three independent immunotherapy cohorts (Figure 6, Supplementary Figure S2). In the OAK and POPLAR clinical trial cohorts, SCPM demonstrated good prognostic stratification capability. In the OAK cohort, the high-SCPM group showed significantly worse outcomes compared to the low-SCPM group for both OS and PFS (Log-rank p = 8e-04 and p = 0.024, respectively) (Figures 6A, B). In the POPLAR cohort, OS difference approached statistical significance (Log-rank p = 0.067), while PFS difference reached statistical significance (Log-rank p = 0.038) (Figures 6C, D). Additional validation was performed in the SU2C immunotherapy cohort (Supplementary Figure S2). The SCPM model maintained its prognostic capability with significant OS stratification (Log-rank p = 0.028) (Supplementary Figure S2A), while PFS stratification showed a similar trend but did not reach statistical significance (Log-rank p = 0.25) (Supplementary Figure S2B). These results across multiple independent immunotherapy cohorts consistently support the prognostic value of SCPM and its potential utility in guiding immunotherapy treatment decisions.
Figure 6
SCPM-associated immune microenvironment characteristics
To elucidate the underlying mechanisms of SCPM’s prognostic and immunotherapeutic predictive value, we comprehensively analyzed the immune microenvironment characteristics between high-SCPM and low-SCPM groups. Analysis of immune-related gene expression revealed distinct patterns between the two groups (Figure 7A). Notably, the high-SCPM group showed reduced expression of HLA class genes, particularly MHC class I and II molecules, indicating potential immune evasion mechanisms. Conversely, the high-SCPM group exhibited elevated expression of immune checkpoint molecules including PDCD1, CD274, and TIGIT, suggesting an immunosuppressive microenvironment despite the presence of checkpoint molecules. To further characterize immune infiltration patterns, we employed seven different computational algorithms to evaluate immune cell infiltration levels across SCPM groups (Figure 7B). Consistent across multiple algorithms including CIBERSORT-ABS, CIBERSORT, EPIC, XCELL, MCP-COUNTER, TIMER, and QUANTISEQ, the low-SCPM group demonstrated significantly higher levels of immune cell infiltration compared to the high-SCPM group. This enhanced immune infiltration in low-SCPM tumors suggests a more immunologically “hot” tumor microenvironment that may be more responsive to immunotherapy. To validate these computational findings, we performed multiplex immunofluorescence staining on clinical samples from our institutional cohort (Figure 7C). RNA-seq data from our internal cohort was used to stratify patients into high-SCPM and low-SCPM groups, followed by immunofluorescence validation using CD4, CD8, and CD20 markers. The results confirmed our computational predictions, showing that low-SCPM tumors had significantly enhanced infiltration of CD8+ T cells and CD20+ B cells compared to high-SCPM tumors. This experimental validation strongly supports the association between low SCPM scores and increased immune cell infiltration, providing a mechanistic basis for the superior immunotherapy responsiveness observed in low-SCPM patients.
Figure 7
Discussion
In this comprehensive study, we developed and validated SCPM, a novel stem cell-based prognostic model for LUAD that demonstrates robust predictive capability across multiple independent cohorts and provides valuable insights for immunotherapy decision-making. Our findings reveal the critical role of stem cell characteristics in determining patient prognosis and treatment response, offering new perspectives for precision medicine in lung cancer. Single-cell RNA sequencing analysis revealed the complex cellular ecosystem within the LUAD microenvironment, with stem cell marker genes showing distinct expression patterns across different cell types. We identified widespread mutations in stem cell-related genes, particularly the high frequency of KRAS mutations (29%), which aligns with previous studies emphasizing the central role of KRAS in cancer stem cell biology (36, 37). The prevalence of KRAS mutations in our cohort is consistent with its established function in maintaining stem cell self-renewal and promoting oncogenic transformation. Similarly, the frequent mutations in KEAP1 suggest that oxidative stress response is a critical pathway disrupted in LUAD stem cells (38).
Comprehensive differential expression analysis between normal and tumor tissues revealed complex regulatory patterns of stem cell marker genes, with key genes such as TMEM106C, ECT2, PSMD2, STIL, and TTK showing significant upregulation in tumor tissues. These findings are consistent with the cancer stem cell hypothesis, which posits that a small subset of cells with stem-like properties drives tumor initiation, progression, and therapeutic resistance (39). The enrichment of differentially expressed genes in cell cycle regulation, DNA replication, and metabolic reprogramming pathways suggests that stem cell characteristics are intimately linked to fundamental cellular processes that drive tumorigenesis. Notably, the downregulation of ALDOA in tumor tissues highlights the metabolic reprogramming that occurs in cancer stem cells, potentially reflecting their unique metabolic dependencies compared to bulk tumor cells. Chromosomal distribution analysis revealed that stem cell marker genes are subject to widespread copy number alterations, with amplification events predominating in genes such as S100A10, UBE2T, and ANLN, suggesting that genomic instability contributes to the dysregulation of stem cell programs in LUAD.
The application of multiple machine learning algorithms in cancer biomarker development has advanced rapidly (40, 41). The consistent performance of SCPM across seven independent cohorts, including both transcriptomic datasets and immunotherapy cohorts, demonstrates its robustness and clinical applicability. The superior prognostic stratification achieved by SCPM, particularly in the context of immunotherapy, suggests that stem cell characteristics serve as important determinants of treatment response. The association between low SCPM scores and improved immunotherapy outcomes may be explained by the enhanced immune cell infiltration observed in low-SCPM tumors. We found that low-SCPM tumors exhibit higher levels of CD8+ T cells and CD20+ B cells, confirmed by both computational analysis and experimental validation, providing mechanistic insights into why these patients respond better to immune checkpoint inhibitors. This is consistent with emerging evidence that cancer stem cells can modulate the immune microenvironment and influence immunotherapy efficacy (42).
Immune microenvironment analysis revealed striking differences between high-SCPM and low-SCPM groups, with high-SCPM tumors characterized by reduced HLA expression and elevated immune checkpoint molecule expression, suggesting an immunosuppressive phenotype despite the presence of immune activation markers. This paradoxical pattern may reflect the complex interplay between stem cell characteristics and immune evasion mechanisms. The reduced expression of MHC class I and II molecules in high-SCPM tumors indicates impaired antigen presentation capacity, which may limit the effectiveness of T cell-mediated immune responses. The clinical implications of our findings extend beyond prognostic stratification to treatment selection, as the SCPM model may help identify patients who are most likely to benefit from immunotherapy. Future studies should focus on validating these findings in prospective clinical trials and exploring the potential for targeting stem cell pathways in combination with immunotherapy to improve outcomes for high-SCPM patients.
In this study, we successfully developed and validated SCPM, a novel prognostic prediction tool for LUAD based on stem cell marker genes. Through integrated multi-omics analysis and advanced machine learning algorithms, we demonstrated that SCPM provides robust prognostic stratification across multiple independent cohorts, with particularly strong performance in predicting responses to immunotherapy. Our findings reveal that patients with low SCPM scores exhibit higher levels of immune cell infiltration and respond more favorably to immunotherapy, offering valuable prognostic assessment and guidance for clinical decision-making. The SCPM model holds significant promise for clinical application, enabling the identification of patient subgroups most likely to benefit from immunotherapy and advancing the precision treatment of LUAD.
Statements
Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.
Ethics statement
The studies involving humans were approved by the Ethics Board of The First Affiliated Hospital with Nanjing Medical University (Approval No. 2019-SR-156). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.
Author contributions
JZ: Methodology, Conceptualization, Writing – review & editing, Writing – original draft. HL: Data curation, Writing – review & editing, Methodology, Writing – original draft. WY: Project administration, Writing – review & editing, Methodology, Writing – original draft. MD: Writing – original draft, Project administration, Writing – review & editing. CH: Methodology, Writing – original draft, Software. JF: Project administration, Writing – review & editing, Writing – original draft, Investigation.
Funding
The author(s) declare that no financial support was received for the research and/or publication of this article.
Acknowledgments
The authors express their gratitude for the provision of data by databases such as TCGA and GEO. Sincere appreciation is extended to the reviewers and editors for their valuable comments.
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declare that no Generative AI was used in the creation of this manuscript.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2025.1634830/full#supplementary-material
Supplementary Figure 1SCPM performance in TCGA training cohort. (A) Kaplan-Meier survival curve showing significant prognostic stratification between high-SCPM and low-SCPM groups (Log-rank p < 0.0001). (B) Time-dependent ROC curves for 1-, 3-, and 5-year survival predictions with AUC values of 0.68, 0.68, and 0.64, respectively. (C) Principal component analysis showing distinct clustering patterns between high-SCPM (orange) and low-SCPM (blue) groups based on SCPM signature genes.
Supplementary Figure 2SCPM prognostic performance in SU2C immunotherapy cohort. (A, B) Kaplan-Meier survival curves showing significant OS stratification (Log-rank p = 0.028) and PFS trend (Log-rank p = 0.25) between high-SCPM (red) and low-SCPM (green) groups in the SU2C immunotherapy cohort.
References
1
DenisenkoTVBudkevichINZhivotovskyB. Cell death-based treatment of lung adenocarcinoma. Cell Death Dis. (2018) 9:117. doi: 10.1038/s41419-017-0063-y
2
FengJZhangPWangDLiYTanJ. New strategies for lung cancer diagnosis and treatment: applications and advances in nanotechnology. biomark Res. (2024) 12:136. doi: 10.1186/s40364-024-00686-7
3
XuYLiJZhuKZengYChenJDongXet al. FIBP interacts with transcription factor STAT3 to induce EME1 expression and drive radio resistance in lung adenocarcinoma. Int J Biol Sci. (2023) 19:3816–29. doi: 10.7150/ijbs.83134
4
ShenJFengSZhangPQiCLiuZFengYet al. Evaluating generative AI models for explainable pathological feature extraction in lung adenocarcinoma grading assessment and prognostic model construction. Int J Surg. (2025).
5
JhalaHHarlingLRodrigoANonakaDMcLeanENgWet al. Clinicopathological predictors of survival in resected primary lung adenocarcinoma. J Clin Pathol. (2022) 75:310–5. doi: 10.1136/jclinpath-2021-207388
6
SeguinLDurandyMFeralCC. Lung adenocarcinoma tumor origin: A guide for personalized medicine. Cancers (Basel). (2022) 14. doi: 10.3390/cancers14071759
7
ChengHShangDZhouR. Germline stem cells in human. Signal Transduct Target Ther. (2022) 7:345. doi: 10.1038/s41392-022-01197-3
8
LiPOuQShiSShaoC. Immunomodulatory properties of mesenchymal stem cells/dental stem cells and their therapeutic applications. Cell Mol Immunol. (2023) 20:558–69. doi: 10.1038/s41423-023-00998-y
9
WangWDGuoYYYangZLSuGLSunZJ. Sniping cancer stem cells with nanomaterials. ACS Nano. (2023) 17:23262–98. doi: 10.1021/acsnano.3c07828
10
YeBJiHZhuMWangATangJLiangYet al. Single-cell sequencing reveals novel proliferative cell type: a key player in renal cell carcinoma prognosis and therapeutic response. Clin Exp Med. (2024) 24:167. doi: 10.1007/s10238-024-01424-x
11
RowbothamSPGoruganthuMULArasadaRRWangWZCarboneDPKimCF. Lung cancer stem cells and their clinical implications. Cold Spring Harb Perspect Med. (2022) 12.
12
MaiuthedAChantarawongWChanvorachoteP. Lung cancer stem cells and cancer stem cell-targeting natural compounds. Anticancer Res. (2018) 38:3797–809. doi: 10.21873/anticanres.12663
13
SunDGuanXMoranAEWuLYQianDZSchedinPet al. Identifying phenotype-associated subpopulations by integrating bulk and single-cell sequencing data. Nat Biotechnol. (2022) 40:527–38. doi: 10.1038/s41587-021-01091-3
14
TopalianSLHodiFSBrahmerJRGettingerSNSmithDCMcDermottDFet al. Five-year survival and correlates among patients with advanced melanoma, renal cell carcinoma, or non–small cell lung cancer treated with nivolumab. JAMA Oncol. (2019) 5:1411–20. doi: 10.1001/jamaoncol.2019.2187
15
SkoulidisFAraujoHADoMTQianYSunXCoboAGet al. CTLA4 blockade abrogates KEAP1/STK11-related resistance to PD-(L)1 inhibitors. Nature. (2024) 635:462–71.
16
SituYZhangPZhangCJiangAZhangNZhuLet al. The metabolic dialogue between intratumoral microbes and cancer: implications for immunotherapy. EBioMedicine. (2025) 115:105708. doi: 10.1016/j.ebiom.2025.105708
17
AttiliIPassaroACorvajaCTrillo AliagaPDel SignoreESpitaleriGet al. Immune checkpoint inhibitors in EGFR-mutant non-small cell lung cancer: A systematic review. Cancer Treat Rev. (2023) 119. doi: 10.1016/j.ctrv.2023.102602
18
WangHYaoLChenJLiYSuZLiuYet al. The dual role of POSTN in maintaining glioblastoma stem cells and the immunosuppressive phenotype of microglia in glioblastoma. J Exp Clin Cancer Res. (2024) 43:252. doi: 10.1186/s13046-024-03175-9
19
TomidaSTakeuchiTShimadaYArimaCMatsuoKMitsudomiTet al. Relapse-related molecular signature in lung adenocarcinomas identifies patients with dismal prognosis. J Clin Oncol. (2009) 27:2793–9. doi: 10.1200/jco.2008.19.7053
20
WilkersonMDYinXWalterVZhaoNCabanskiCRHaywardMCet al. Differential pathogenesis of lung adenocarcinoma subtypes involving sequence mutations, copy number, chromosomal instability, and methylation. PloS One. (2012) 7:e36530. doi: 10.1371/journal.pone.0036530
21
StaafJJönssonGJönssonMKarlssonAIsakssonSSalomonssonAet al. Relation between smoking history and gene expression profiles in lung adenocarcinomas. BMC Med Genomics. (2012) 5:22. doi: 10.1186/1755-8794-5-22
22
RousseauxSDebernardiAJacquiauBVitteALVesinANagy-MignotteHet al. Ectopic activation of germline and placental genes identifies aggressive metastasis-prone lung cancers. Sci Transl Med. (2013) 5:186ra66. doi: 10.1126/scitranslmed.3005723
23
YamauchiMYamaguchiRNakataAKohnoTNagasakiMShimamuraTet al. Epidermal growth factor receptor tyrosine kinase defines critical prognostic genes of stage I lung adenocarcinoma. PloS One. (2012) 7:e43923. doi: 10.1371/journal.pone.0043923
24
TangHXiaoGBehrensCSchillerJAllenJChowCWet al. A 12-gene set predicts survival benefits from adjuvant chemotherapy in non-small cell lung cancer patients. Clin Cancer Res. (2013) 19:1577–86. doi: 10.1158/1078-0432.ccr-12-2321
25
ZhangYParmigianiGJohnsonWE. ComBat-seq: batch effect adjustment for RNA-seq count data. NAR Genom Bioinform. (2020) 2:lqaa078. doi: 10.1093/nargab/lqaa078
26
PatilNSNabetBYMüllerSKoeppenHZouWGiltnaneJet al. Intratumoral plasma cells predict outcomes to PD-L1 blockade in non-small cell lung cancer. Cancer Cell. (2022) 40:289–300.e4. doi: 10.1016/j.ccell.2022.02.002
27
RaviAHellmannMDArniellaMBHoltonMFreemanSSNaranbhaiVet al. Genomic and transcriptomic analysis of checkpoint blockade response in advanced non-small cell lung cancer. Nat Genet. (2023) 55:807–19. doi: 10.1038/s41588-023-01355-5
28
CaoYFuLWuJPengQNieQZhangJet al. Integrated analysis of multimodal single-cell data with structural similarity. Nucleic Acids Res. (2022) 50:e121. doi: 10.1093/nar/gkac781
29
ZhangPZhangXCuiYGongZWangWLinS. Revealing the role of regulatory T cells in the tumor microenvironment of lung adenocarcinoma: a novel prognostic and immunotherapeutic signature. Front Immunol. (2023) 14:1244144. doi: 10.3389/fimmu.2023.1244144
30
ChenYChenSLeiEP. DiffChIPL: a differential peak analysis method for high-throughput sequencing data with biological replicates based on limma. Bioinformatics. (2022) 38:4062–9. doi: 10.1093/bioinformatics/btac498
31
LuoKQianZJiangYLvDZhuKShaoJet al. Characterization of the metabolic alteration-modulated tumor microenvironment mediated by TP53 mutation and hypoxia. Comput Biol Med. (2023) 163:107078. doi: 10.1016/j.compbiomed.2023.107078
32
ZhangLCuiYZhouGZhangZZhangP. Leveraging mitochondrial-programmed cell death dynamics to enhance prognostic accuracy and immunotherapy efficacy in lung adenocarcinoma. J Immunother Cancer. (2024) 12. doi: 10.1136/jitc-2024-010008
33
ZhangPWangDZhouGJiangSZhangGZhangLet al. Novel post-translational modification learning signature reveals B4GALT2 as an immune exclusion regulator in lung adenocarcinoma. J Immunother Cancer. (2025) 13. doi: 10.1136/jitc-2024-010787
34
Van AllenEMMiaoDSchillingBShuklaSABlankCZimmerLet al. Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science. (2015) 350:207–11. doi: 10.1126/science.aad0095
35
LiTFanJWangBTraughNChenQLiuJSet al. TIMER: A web server for comprehensive analysis of tumor-infiltrating immune cells. Cancer Res. (2017) 77:e108–e10. doi: 10.1158/0008-5472.can-17-0307
36
JuulNHYoonJKMartinezMCRishiNKazadaevaYIMorriMet al. KRAS(G12D) drives lepidic adenocarcinoma through stem-cell reprogramming. Nature. (2023) 619:860–7. doi: 10.1038/s41586-023-06324-w
37
DuanXZhangTFengLde SilvaNGreenspunBWangXet al. A pancreatic cancer organoid platform identifies an inhibitor specific to mutant KRAS. Cell Stem Cell. (2024) 31:71–88.e8. doi: 10.1016/j.stem.2023.11.011
38
LuoMBaoLXueYZhuMKumarAXingCet al. ZMYND8 protects breast cancer stem cells against oxidative stress and ferroptosis through activation of NRF2. J Clin Invest. (2024) 134. doi: 10.1172/jci171166
39
GalassiCEstellerMVitaleIGalluzziL. Epigenetic control of immunoevasion in cancer stem cells. Trends Cancer. (2024) 10:1052–71. doi: 10.1016/j.trecan.2024.08.004
40
YeBFanJXueLZhuangYLuoPJiangAet al. iMLGAM: Integrated Machine Learning and Genetic Algorithm-driven Multiomics analysis for pan-cancer immunotherapy response prediction. Imeta. (2025) 4:e70011. doi: 10.1002/imt2.70011
41
YeBJiangALiangFWangCLiangXZhangP. Navigating the immune landscape with plasma cells: A pan-cancer signature for precision immunotherapy. Biofactors. (2025) 51:e2142. doi: 10.1002/biof.2142
42
SawPELiuQWongPPSongE. Cancer stem cell mimicry for immune evasion and therapeutic resistance. Cell Stem Cell. (2024) 31:1101–12. doi: 10.1016/j.stem.2024.06.003
Summary
Keywords
lung adenocarcinoma, single-cell sequencing, stem cells, prognostic model, immunotherapy
Citation
Zheng J, Lin H, Ye W, Du M, Huang C and Fan J (2025) Single-cell and multi-omics analysis reveals the role of stem cells in prognosis and immunotherapy of lung adenocarcinoma patients. Front. Immunol. 16:1634830. doi: 10.3389/fimmu.2025.1634830
Received
25 May 2025
Accepted
07 July 2025
Published
22 July 2025
Volume
16 - 2025
Edited by
Zhijie Zhao, Shanghai Jiao Tong University, China
Reviewed by
Jiaxiong Tan, Tianjin Medical University Cancer Institute and Hospital, China
Zhenyu Wu, First People’s Hospital of Foshan, China
Han Zhang, Tianjin University, China
Updates
Copyright
© 2025 Zheng, Lin, Ye, Du, Huang and Fan.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Jun Fan, fanjun@jsph.org.cn; Chenjun Huang, huangchenjun@jsph.org.cn; Mingjun Du, dumingjun1989@163.com
†These authors have contributed equally to this work and share first authorship
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.