Non-small cell lung cancer (NSCLC) accounts for 85% of all lung cancer cases and continues to carry high mortality rates despite advances in treatment. A major driver of poor outcomes is late-stage diagnosis combined with resistance to conventional therapies, making the identification of underlying molecular mechanisms a priority for researchers.
Programmed cell death (PCD) - including apoptosis and autophagy - plays a central role in whether tumors grow or are contained. Apoptosis is a tightly controlled form of cellular self-destruction that eliminates damaged cells and maintains tissue balance. When this process is disrupted, cancer cells can evade death and proliferate unchecked. Autophagy, a cellular recycling mechanism, can either suppress tumors early on or help them survive under stress conditions like oxygen deprivation.
This study aimed to map how programmed cell death genes behave within NSCLC tumors at the level of individual cells and across physical tissue space. By combining single-cell RNA sequencing, spatial transcriptomics, machine learning, and laboratory experiments, the researchers built a comprehensive picture of how PCD shapes tumor biology and patient outcomes.
A key discovery was the gene SLC7A5, a protein that transports amino acids into cells. This gene emerged as both a prognostic risk factor and a potential therapeutic target. When SLC7A5 was reduced in NSCLC cells, proliferation slowed and apoptosis increased, suggesting it actively helps cancer cells survive and grow.
The study assembled a large and diverse dataset combining single-cell RNA sequencing from five NSCLC patient cohorts with bulk transcriptomic data from TCGA (spanning both lung adenocarcinoma and squamous cell carcinoma). Spatial transcriptomics data were added to map gene expression directly onto tissue sections, preserving the spatial context of gene activity that standard sequencing destroys when cells are dissociated.
Single-cell data were processed and clustered to identify distinct cell populations within NSCLC tumors. After quality control filtering and batch effect correction using the Harmony method, cells were grouped by similarity using the Seurat computational framework. Cell type identities were assigned using marker genes verified against the CellMarker database, enabling the researchers to distinguish cancer cells from immune cells, stromal cells, and others.
A programmed cell death gene panel was constructed by intersecting upregulated genes from five bulk cohorts with 18 curated PCD gene sets. Strict statistical filters (meta-analysis FDR below 0.05 and effect size above 0.5) ensured only robustly and specifically elevated genes were retained. Thirteen different single-cell scoring methods were then applied in parallel to evaluate PCD activity across cell types, with results averaged to reduce method-specific bias.
Spatial transcriptomics analysis added a geographic dimension to the findings. Using robust cell type decomposition (RCTD), the single-cell data were projected onto spatial tissue sections, allowing the team to observe where different cell types physically reside within tumors. Spatial dependency analysis with the MISTy algorithm then measured how the presence of one cell type predicted the presence or behavior of nearby cells, revealing clinically meaningful interaction patterns.
Programmed cell death gene activity was highest in NSCLC epithelial cells compared to all other cell types in the tumor. This was a consistent finding across 13 independent scoring methods. Spatial transcriptomics confirmed that PCD scores were significantly higher in malignant tissue regions than in adjacent non-malignant tissue, validating the specificity of the PCD gene panel for cancerous cells.
Epithelial cells were divided into two groups - high PCD (PCDhighepi) and low PCD (PCDlowepi) - based on their enrichment score. The PCDhighepi cells showed markedly upregulated pathways related to cell cycle control, tumor metastasis, and hypoxia response. Key transcription factors including KLF5, HIF1A, MYC, E2F4, and BACH1 were particularly active in this high-PCD group, connecting programmed cell death to cell cycle regulation and cancer-promoting gene programs.
Pseudotime trajectory analysis revealed that PCDhighepi cells represent the developmental origin of NSCLC epithelial cells, meaning these high-PCD cells appear to be the 'earlier' or more stem-like state from which other tumor cells emerge. As cells progress through development, PCD activity decreases and features of autophagy and metabolic adaptation become more prominent, consistent with tumor cells adapting to survive harsh conditions.
Metabolic analysis showed PCDhighepi cells have enhanced metabolic activity across multiple pathways. The tricarboxylic acid cycle, glycogen synthesis, fatty acid uptake, pyrimidine synthesis, and glycan synthesis were all more active in PCDhighepi compared to PCDlowepi. This heightened metabolism may support the energy demands of both active PCD signaling and the proliferative behavior that characterizes this cell population.
The most striking cell-cell communication in the tumor microenvironment was between PCDhighepi cells and smooth muscle cells. Smooth muscle cells, which normally help regulate blood vessel tone, are found within tumors where they contribute to angiogenesis (new blood vessel formation) and support tumor invasion. Cell communication analysis revealed that PCDhighepi cells send signals to smooth muscle cells through multiple pathways including the Hippo, FoxO, and focal adhesion pathways.
Specific molecular messenger pairs were identified linking these two cell types. From PCDhighepi cells to smooth muscle cells, key signaling pairs included WNT7B-FZD1, GAS6-AXL, LIF-IL6ST, and BMP2-BMPR2. In the reverse direction, smooth muscle cells sent WNT2 signals to cancer cells through LRP6, LRP5, and FZD5 receptors, and also signaled via IGF1-IGF1R, a pathway known to promote cancer cell survival and growth.
Spatial transcriptomics confirmed that PCDhighepi and smooth muscle cells are physically located near each other within tumors. This co-localization was detected across multiple independent tissue samples and validated using MISTy spatial dependency analysis, which showed that the presence of PCDhighepi predicted the presence of smooth muscle cells at close, intermediate, and extended distances within the tumor tissue.
Autophagy plays a key role in mediating this tumor cell-smooth muscle cell crosstalk. Under hypoxic conditions common in tumors, HIF-1 alpha activates autophagy in smooth muscle cells, enhancing their survival and supporting new vessel formation. The tumor cells use these vessels for nutrient supply and metastatic escape, making this interaction a potentially important target for disrupting tumor progression.
To translate the PCD biology into a clinically useful tool, the researchers built a machine learning prognostic model. They first identified characteristic genes of PCDhighepi cells and intersected these with genes showing causal links to lung cancer from Mendelian randomization analysis - a technique that uses genetic variants to test cause-and-effect relationships rather than mere correlations. This intersection produced a high-confidence gene set for model construction.
Ten different machine learning algorithms were compared for their ability to predict patient survival, including random survival forests, LASSO, elastic net, CoxBoost, gradient boosting machine, and others. The algorithm combination Lasso + SuperPC was selected because it achieved one of the highest average C-index scores across all validation datasets while also retaining interpretable gene weights, allowing the model to be understood and potentially translated.
The NSCLCPCD model was tested in the TCGA training set and seven independent validation cohorts. Across these datasets, high-risk patients consistently had significantly worse survival than low-risk patients. ROC analysis showed AUC values ranging from 0.64 to 0.81 for 1-year survival prediction across different cohorts, with particularly strong performance in GSE31210 (AUC 0.81), GSE42127 (AUC 0.73), and GSE72094 (AUC 0.71).
The NSCLCPCD risk score was confirmed as an independent prognostic factor by multivariate Cox regression. After adjusting for tumor stage, grade, and other clinical variables, the model remained significantly associated with overall survival (hazard ratio 1.476, p = 0.03) and disease-specific survival (hazard ratio 1.780, p less than 0.001). High-risk patients also showed poorer responses to immunotherapy but greater sensitivity to several chemotherapy drugs including paclitaxel, docetaxel, erlotinib, and gefitinib.
SLC7A5, which encodes a large neutral amino acid transporter (LAT1), was identified as the single gene that functions as a risk factor in both lung adenocarcinoma and lung squamous cell carcinoma. Its expression was elevated in tumor versus normal tissue across multiple bulk RNA sequencing cohorts and proteomics datasets. Spatial transcriptomics further confirmed higher SLC7A5 expression specifically in tumor regions compared to surrounding normal tissue.
When SLC7A5 was knocked down using RNA interference in H520 and H460 lung cancer cells, proliferation fell significantly. EdU incorporation assays measuring DNA synthesis, CCK-8 viability assays tracking cell numbers over time, and 2D colony formation assays all showed consistent reductions in cell growth following SLC7A5 knockdown. Transwell invasion assays additionally showed reduced migration ability, suggesting SLC7A5 also contributes to the invasive behavior of NSCLC cells.
Knockdown of SLC7A5 substantially increased cancer cell apoptosis. Flow cytometry with Annexin V/PI staining quantified a significant increase in the fraction of apoptotic cells after SLC7A5 reduction. Western blot analysis showed that the anti-apoptotic protein BCL-2 decreased, while pro-apoptotic markers increased: Bax was upregulated and active Cleaved-Caspase-3 and Cleaved-PARP rose, indicating that SLC7A5 normally acts as a brake on the cell death machinery.
These laboratory findings are consistent with clinical data showing SLC7A5 as a poor prognostic marker. Published studies show that 91% of surgically removed NSCLC tumors have high SLC7A5 levels, and that patients with SLC7A5-positive tumors have a 5-year survival rate of only 51.8% compared to 87.8% for SLC7A5-negative cases. A specific SLC7A5 inhibitor, JPH203, has already been developed and may offer a therapeutic path for patients with SLC7A5-driven tumors.
One of the study's nuanced findings is that autophagy - cellular self-recycling - plays opposite roles at different stages of lung cancer. In early tumor development, autophagy helps eliminate damaged components and suppress tumor growth. But as tumors progress into nutrient-poor, oxygen-depleted environments, autophagy switches to a survival mode that helps cancer cells persist and even spread. PCDlowepi cells, the more developmentally advanced tumor cells, rely on autophagy to maintain metabolic balance under stress.
Autophagy also sustains tumor stem cells, which are responsible for tumor initiation, relapse, and treatment resistance. By maintaining Wnt/beta-catenin, Notch, and Hedgehog signaling pathways, autophagy helps stem-like cancer cells self-renew. Importantly, inhibiting autophagy in some cancers can push tumor stem cells toward terminal differentiation, reducing their ability to regenerate tumors - suggesting autophagy inhibitors could complement standard treatments.
The interaction between autophagy and smooth muscle cells also shapes the physical tumor microenvironment. Autophagy in smooth muscle cells regulates matrix metalloproteinase secretion, which remodels the tumor stroma in ways that facilitate cancer cell invasion. Targeting autophagy could therefore interfere not just with cancer cell survival but also with the structural support systems that enable metastasis.
The study acknowledges important limitations that frame its conclusions appropriately. Large cell carcinoma was not included in the analysis, limiting generalizability across all NSCLC subtypes. The NSCLCPCD model performed less consistently in predicting progression-free interval compared to overall survival, suggesting it captures long-term mortality risk better than short-term disease dynamics. The H460 cell line used in experiments is technically derived from large cell carcinoma rather than adenocarcinoma, and future work will validate SLC7A5 findings using adenocarcinoma-specific lines such as A549.
This study provides an integrated molecular portrait of programmed cell death in NSCLC using cutting-edge single-cell and spatial technologies. By combining single-cell RNA sequencing, spatial transcriptomics, Mendelian randomization, and 10 machine learning algorithms, the work moves beyond descriptive gene expression analysis to identify causal relationships and build predictive tools. The multi-layer approach is a significant methodological advance over prior studies relying on single algorithms or bulk sequencing alone.
The NSCLCPCD prognostic model offers a framework for stratifying patients into risk groups that differ in their likely outcomes and treatment sensitivities. Low-risk patients appear to respond better to immunotherapy, while high-risk patients show greater sensitivity to conventional chemotherapy agents and to drugs like leptomycin B and epothilone-B. This actionable information could help guide treatment selection in clinical practice once the model is prospectively validated.
SLC7A5 emerges as a compelling therapeutic target with both biological and clinical evidence supporting its importance. It is highly expressed in the majority of NSCLC tumors, independently predicts poor survival, and functionally drives proliferation and apoptosis evasion. Existing inhibitors like JPH203 are already in development and may be applicable alone or in combination with chemotherapy in SLC7A5-positive NSCLC.
Prospective validation in larger, multicenter cohorts is the essential next step. The current study is limited by its reliance on retrospective public databases and cell line experiments. Confirming the NSCLCPCD model's clinical utility and validating SLC7A5 as a therapeutic target in prospective trials with standardized tissue collection will be necessary before these findings influence routine clinical management of NSCLC patients.