A clinically relevant SLC2A1-associated malignant epithelial cell state predicts prognosis and immunotherapy response in lung adenocarcinoma

A clinically relevant SLC2A1-associated malignant epithelial cell state predicts prognosis and immunotherapy response in lung adenocarcinoma | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A clinically relevant SLC2A1-associated malignant epithelial cell state predicts prognosis and immunotherapy response in lung adenocarcinoma Linqian Song, Shiyun Xing, Peijie Li, Yunliang Cao, Hu Ma This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9298986/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Background Lung adenocarcinoma (LUAD) remains a leading cause of cancer-related mortality, with therapeutic resistance largely driven by unresolved malignant epithelial heterogeneity within the tumor microenvironment. However, the epithelial cell states that underlie poor prognosis and immunotherapy resistance remain incompletely defined. Methods We performed an integrative multi-omics analysis combining large-scale single-cell RNA sequencing, spatial transcriptomics, and bulk transcriptomic data with clinical outcomes. The Scissor algorithm was applied to identify prognosis-associated epithelial cell states, followed by construction of a risk score model. External validation was conducted across multiple independent cohorts, including immunotherapy-treated datasets. Results We identified a prognostically relevant Scissor⁺ malignant epithelial cell state associated with adverse survival. This state was characterized by activation of MYC, epithelial–mesenchymal transition, hypoxia, and NF-κB signaling, and was linked to an immunosuppressive tumor microenvironment. Based on this state, we developed a Scissor⁺ epithelial cell–derived risk score (SERS), which demonstrated robust and reproducible prognostic performance across multiple cohorts and was associated with reduced responsiveness to immunotherapy. Further analyses identified SLC2A1 as a key gene associated with this malignant epithelial state. Functional experiments confirmed that SLC2A1 promotes tumor cell proliferation, migration, and invasion. In addition, cell–cell communication analysis suggested a potential SLC2A1–CAF–collagen signaling axis linking epithelial cell states with stromal interactions. Conclusion This study defines a clinically relevant malignant epithelial cell state in LUAD and establishes a framework linking cell states, molecular features, and microenvironmental interactions. These findings provide potential biomarkers for prognostic stratification and immunotherapy response prediction in LUAD. Lung adenocarcinoma Single-cell RNA sequencing Malignant epithelial cell states Prognostic risk model Tumor microenvironment SLC2A1 Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 1. Introduction Lung cancer remains the leading cause of cancer-related death worldwide(Siegel et al. 2025 ). Lung adenocarcinoma (LUAD), the most common histological subtype, is characterized by pronounced heterogeneity and complex molecular features(Shen et al. 2025 ; Zhang et al. 2025 ). Although systemic treatment for LUAD has expanded from conventional chemotherapy to include targeted therapy and immunotherapy, overall survival outcomes have improved only modestly(Herbst et al. 2018 ; Wang et al. 2025 ). A major reason for the limited clinical benefit lies in tumor heterogeneity: chemotherapy-related toxicity, the low prevalence of actionable driver mutations, and the unpredictable response to immunotherapy all restrict the effectiveness of current treatment strategies(Bortolot et al. 2025 ; Reck et al. 2025 ; Zhao et al. 2026 ). Therefore, identifying molecular markers that reflect key tumor cell states and predict therapeutic response is critical for advancing precision treatment in LUAD. The tumor immune microenvironment (TIME) plays a central role in LUAD progression, immune evasion, and treatment response(Xu et al. 2022 ). Dynamic interactions between tumor cells, immune cells, fibroblasts, and vascular-associated cells collectively shape the tumor ecosystem and its immunoregulatory landscape(Sun et al. 2025 ). Increasing evidence indicates that the TIME is not static but undergoes continuous remodeling during tumor evolution(Keenan et al. 2025 ). For example, the accumulation of immunosuppressive macrophages, regulatory T cells, and cancer-associated fibroblasts (CAFs) is frequently associated with immune escape and poor clinical outcomes(Huang et al. 2025 ). However, which specific cellular populations drive malignant progression in LUAD and influence patient prognosis through microenvironmental interactions remains incompletely understood. The development of single-cell RNA sequencing (scRNA-seq) has enabled us to re-understand the heterogeneity of LUAD tumor tissue at the cellular level(Boxer et al. 2025 ). In particular, within tumor epithelial cells, different malignant states often correspond to distinct potentials for progression and interactions with the microenvironment(Andreatta et al. 2025 ). By leveraging scRNA-seq, not only can we identify key cell subpopulations associated with tumor progression, but we can also further dissect their immune regulatory features and cell communication networks, providing new biological foundations for prognostic assessment and treatment stratification(Tirosh and Suva 2024 ). Using the Scissor algorithm to integrate single-cell transcriptomics with clinical survival data, we have been able to pinpoint epithelial cell states directly linked to poor prognosis in LUAD. However, despite increasing recognition of epithelial heterogeneity, a key gap remains: the malignant epithelial cell states that directly drive poor prognosis, their molecular regulators, and their relationship with immunotherapy response have not been systematically defined. In particular, how these cell states can be translated into clinically applicable biomarkers remains unclear. Based on this rationale, we integrated large-scale single-cell transcriptomic data with clinical and spatial datasets to systematically characterize malignant epithelial heterogeneity in LUAD. We identified a prognostically relevant epithelial cell state and developed a corresponding risk score model (SERS), which was validated across independent cohorts and immunotherapy-treated samples. Furthermore, we identified SLC2A1 as a key regulator of this malignant state and explored its functional and microenvironmental roles. Together, this study defines a clinically relevant malignant epithelial cell state in LUAD and establishes a framework linking cell states, molecular drivers, and microenvironmental interactions, providing new insights into tumor progression and precision therapy. The overall study design and analytical workflow are illustrated in Fig. 1 . 2. Methods 2.1 Data sources and preprocessing. We used the public single-cell RNA-seq dataset GSE308103, which includes 16 paired normal lung and lung adenocarcinoma (LUAD) samples. Raw expression matrices and sample annotations were downloaded, and gene symbols were harmonized to HGNC official symbols with a consistent matrix format. Bulk RNA-seq expression data and clinical follow-up information for prognostic modeling were obtained from the TCGA-LUAD cohort. External validation cohorts included GSE31210, GSE50081, and GSE72094. These GEO cohorts were also merged into a Meta cohort to support robustness analyses. Immunotherapy-related cohorts included POPLAR, OAK, and GSE135222. Spatial transcriptomics data were taken from E-MTAB-13530, which contains both tumor and normal regions. All datasets used in this study are summarized in Supplementary Table S3. Unless otherwise stated, all analyses were performed in R (v4.2.1). Single-cell analyses were conducted mainly with Seurat (v4.3.1), batch correction with harmony (v0.1.1), cell–cell communication analysis with CellChat (v1.1.1), and prognosis-linked subpopulation identification with Scissor (v1.0.0). Ligand–target inference was performed using nichenetr (v1.0.0). Immune infiltration and immune signature scoring were carried out using IOBR (v0.99.9) and GSVA (v1.46.0). Differential expression and enrichment analyses used clusterProfiler (v4.6.2) and fgsea (v1.24.0). Survival and model evaluation were performed with survival (v3.3-1), survminer (v0.4.9), and timeROC (v0.4). Batch-effect adjustment used sva (v3.44.0). Figures were generated mainly with ggplot2 (v3.4.4) and ComplexHeatmap (v2.14.0). 2.2 scRNA-seq data processing. Single-cell RNA-seq data were processed using a standard Seurat (v4.3.1) workflow. Expression matrices were imported for each sample, and per-cell quality metrics were calculated, including the number of detected genes (nFeature_RNA), total UMI counts (nCount_RNA), and the percentage of mitochondrial transcripts (percent.mt). Cells were retained if nFeature_RNA was between 200 and 8000, and cells with percent.mt ≥ 15% were removed. To reduce the impact of potential doublets and unusually deep sequencing, we also excluded cells with extremely high nCount_RNA within each sample (defined as values above the 99th percentile). Doublets were then identified and removed using DoubletFinder. For each sample, we built a PCA space and performed parameter sweeps to select the optimal pK. The expected doublet rate was set based on the number of cells in each sample. Cells predicted as doublets were excluded. After quality control and doublet removal, we retained 282,289 high-quality cells, including 110,015 cells from normal lung tissues and 173,374 cells from LUAD tumors. Because our analysis was based on multiple samples within a single dataset (GSE308103), we performed within-dataset integration to minimize sample-to-sample technical variation. Each sample was normalized (NormalizeData), highly variable genes were identified (FindVariableFeatures), and data were scaled (ScaleData) before running PCA (RunPCA). We then used Harmony to align samples, with sample/patient origin as the batch variable. Downstream analyses were performed on the integrated low-dimensional space. A KNN graph was built using the top 30 principal components (FindNeighbors), unsupervised clustering was performed with the Louvain algorithm (FindClusters, resolution = 0.6), and UMAP was used for visualization (RunUMAP). Cell-type annotation was performed on the integrated dataset. We identified cluster markers using FindAllMarkers (Wilcoxon rank-sum test with Benjamini–Hochberg correction; FDR 0.25). Major cell types were assigned based on canonical markers: T cells ( CD3D/E, TRAC ), B cells ( MS4A1, CD79A ), plasma cells ( MZB1, JCHAIN ), myeloid cells ( LYZ, CD14, SPI1 ), fibroblasts ( COL1A1/2, DCN ), endothelial cells ( PECAM1, VWF ), and epithelial cells ( EPCAM, KRT7/8 ). Marker patterns were checked using dot plots, violin plots, and feature plots. Cell-type proportions were summarized at both the group level (normal vs LUAD) and the sample level and were visualized accordingly. To increase resolution within key compartments, we performed subset re-clustering of T cells, myeloid cells, and epithelial cells. Subsets were extracted based on the global annotations (subset function in Seurat) and reprocessed using the same workflow (NormalizeData, FindVariableFeatures, ScaleData, RunPCA, FindNeighbors, FindClusters at resolution = 0.6, and RunUMAP). T-cell subsets were annotated using representative markers, including Tcm ( TCF7, IL7R ), Tem ( GZMK) , Tfh ( CXCL13 ), Tpro ( MKI67 ), Trm ( ZNF683 ), Treg ( FOXP3 ), and Tcyto ( NKG7 ). Myeloid subsets were distinguished using markers such as S 100A8, FCGR1A, C1QA , and LST1 to define monocytes, macrophages, myeloid dendritic cells, and neutrophils. Epithelial cells were further divided into subclusters (Epi_0–Epi_8), which were used as input for downstream inferCNV analysis and CellChat cell–cell communication analyses. 2.3 Differential expression analysis. Differential expression analysis was performed in Seurat using the default Wilcoxon rank-sum test to compare gene expression between specified groups or subpopulations. Multiple testing was adjusted with the Benjamini–Hochberg method. Differentially expressed genes were defined using FDR 0.25. 2.4 Functional enrichment analysis. Pathway analysis was based on MSigDB Hallmark gene sets. We used fgsea to run gene set enrichment analysis (GSEA) on a ranked list of all genes ordered by the differential expression statistic. Pathway significance was assessed by permutation testing and adjusted for multiple comparisons to control false positives. 2.5 Module scoring. To describe functional states in myeloid cells, we defined gene modules related to cytokine signaling, inflammatory response, M2-like macrophage features, and NF-κB signaling. Module scores were calculated at the single-cell level using Seurat’s AddModuleScore. Scores were compared across myeloid subtypes and tissue origins, and were also used in downstream trajectory analyses. 2.6 Trajectory inference . Trajectory inference was performed with Slingshot. We used the low-dimensional embedding (PCA/UMAP) of the myeloid subset together with cluster labels to infer lineages and assign a pseudotime value to each cell. We then examined how module scores changed along pseudotime to capture dynamic functional shifts during state transitions. 2.7 CNV inference. Copy number variation (CNV) signals in epithelial cells were inferred using inferCNV. Annotated non-epithelial cell populations (immune cells, endothelial cells, and fibroblasts) were used as the reference cell set. CNV inference and smoothing were performed on the single-cell expression matrix following the recommended inferCNV workflow, and a cell-level CNV intensity metric (CNV score) was calculated. CNV score distributions were then compared across epithelial subclusters within the epithelial subset to assess intra-epithelial CNV heterogeneity. 2.8 Intercellular communication analysis. Intercellular communication was analyzed using CellChat. Annotated cell types/subclusters were defined as communication nodes. Cell–cell signaling was inferred based on the CellChatDB.human ligand–receptor database, restricting analysis to the Secreted Signaling category (all other parameters were kept at default settings). Following the standard CellChat pipeline, putative interactions were computed and filtered using computeCommunProb, filterCommunication, and computeCommunProbPathway, and the overall communication network was summarized using aggregateNet. Pathway-level interaction patterns were visualized using functions such as netVisual_heatmap. Within the epithelial subset, we further focused on collagen-related signaling pathways: communication probabilities of collagen-related ligand–receptor pairs among epithelial subclusters were compared, and network centrality/role analyses were performed to characterize sender, receiver, and mediator (regulatory) features of each subcluster within the collagen signaling network. 2.9 Scissor integrative analysis and definition of signature genes. To identify epithelial cell states associated with clinical outcomes, we used Scissor to jointly model the single-cell epithelial expression matrix with TCGA-LUAD bulk RNA-seq profiles and survival outcomes. Scissor first computed an expression similarity matrix between each single cell and each bulk sample using Pearson correlation. It then fitted a regularized Cox regression model in a survival analysis framework to link the similarity features to clinical outcomes. Cells were classified based on whether their regression coefficients were non-zero: cells with positive coefficients were defined as Scissor⁺, cells with negative coefficients were defined as Scissor⁻, and the remaining cells were treated as background controls. We next performed differential expression analysis between Scissor⁺ and Scissor⁻ epithelial cells (Seurat/Wilcoxon test with Benjamini–Hochberg correction). Pathway enrichment was assessed using GSEA/fgsea with MSigDB Hallmark gene sets to describe functional features associated with Scissor⁺ cells, including MYC targets, TNFα/NF-κB, EMT, G2/M checkpoint, and Hypoxia. To quantify how strongly Scissor⁺ cells were enriched across epithelial subclusters, we calculated observed/expected (Ro/e) enrichment for each subcluster in Scissor⁺ and Scissor⁻ cells(Sun et al. 2022). To evaluate the clinical relevance of the Scissor⁺ transcriptional signature, we built a gene signature from Scissor⁺ differential genes. We then computed signature scores at the sample level in TCGA-LUAD using ssGSEA (implemented in the GSVA package). Samples were stratified by signature score, followed by Kaplan–Meier analysis and Cox regression. To infer upstream cell–cell signals that may drive the Scissor⁺ program, we applied NicheNet within a predefined sender–receiver setting to predict candidate ligands. Ligands were ranked by regulatory potential (aupr_corrected), and top-ranked ligands, their receptors, and putative target gene sets were summarized and visualized. Finally, we intersected Scissor⁺ differential genes with genes upregulated in TCGA-LUAD tumors versus normal tissues, yielding 39 candidate genes for downstream machine-learning–based prognostic model development. 2.10 Prognostic model construction and multi-cohort validation. We built a large-scale machine-learning benchmarking framework based on the 39 candidate genes. To improve comparability across cohorts, expression matrices from each bulk cohort were Z-score standardized. TCGA-LUAD was used as the training set, and independent validation was performed in external cohorts (GSE31210, GSE50081, GSE72094, and a meta-cohort). For the combined cohorts, batch effects were corrected using ComBat (sva package). We systematically evaluated multiple survival modeling methods and their combinations (a total of 117 model settings). Model performance was mainly assessed using Harrell’s C-index, and the final model was chosen based on both predictive accuracy and cross-cohort stability. We ultimately used LASSO plus Random Survival Forest (RSF) to build a risk score model, termed SERS (Scissor⁺-derived epithelial risk score). For each patient, SERS was calculated in both the training and validation sets. Patients were split into high- and low-risk groups using the median SERS as the cutoff. Group differences were tested using Kaplan–Meier survival curves and the log-rank test. The prognostic value of SERS was further assessed by Cox regression (as both a continuous and a categorical variable). Time-dependent ROC curves and AUC were evaluated using tools such as timeROC. 2.11 Immune microenvironment and immunotherapy analyses. To evaluate the association between SERS and the tumor immune microenvironment, we used the IOBR framework to infer immune infiltration from bulk expression profiles in TCGA-LUAD and external cohorts. Immune and stromal cell abundances were jointly estimated using multiple methods, including EPIC, MCPcounter, and xCell. We then compared the estimated abundances of different immune/stromal cell types between the high- and low-risk SERS groups. Immune activation- and immunosuppression-related features were quantified using predefined gene signature sets. Signature scores were computed at the sample level using ssGSEA (implemented in the GSVA package) to measure immune functional states. Tumor mutational burden (TMB) and tumor neoantigen burden (TNB) were included in integrative analyses. TMB was calculated from somatic mutation data as the number of nonsynonymous mutations per Mb. TNB was obtained either from cohort-provided annotations or inferred using a standard neoantigen prediction workflow, depending on data availability. We further performed joint stratification using SERS, TMB/TNB, and immunosuppression signatures, and assessed survival differences across subgroups using Kaplan–Meier analysis and Cox regression. In immunotherapy cohorts (POPLAR, OAK, and GSE135222), SERS was computed for each patient using the same formula as in the training set. Survival analyses were performed for both overall survival (OS) and progression-free survival (PFS) to evaluate the association between SERS and immunotherapy outcomes. 2.12 Spatial transcriptomics analysis. Spatial transcriptomics data (E-MTAB-13530) were processed using a standard Seurat workflow. After spot-level quality control, data were normalized and variance-stabilized using SCTransform. SERS and related gene signature scores were then calculated at the spot level (using AddModuleScore or ssGSEA). SpatialFeaturePlot was used to visualize the spatial distribution of risk-related signals across tissue sections, and scores were compared between tumor and normal regions. 2.13 Causal inference of key genes. To identify key genes with potential causal effects, we performed Mendelian randomization (MR) analysis on the candidate genes. Instrumental variables (IVs) were selected as SNPs significantly associated with the exposure (p < 5×10⁻⁸). We further applied LD clumping to obtain independent variants (r² < 0.005, window size 10,000 kb). The causal effect was mainly estimated using the inverse-variance weighted (IVW) method. MR-Egger and weighted median approaches were used as sensitivity analyses. Heterogeneity was assessed using Cochran’s Q test, and robustness to individual SNPs was evaluated using leave-one-out analysis. 2.14 Functional perturbation analysis. At the single-cell level, we examined the expression of key genes (SLC2A1) across cell types and tested for expression differences within epithelial cells. Virtual knockout analysis was performed using scTenifoldKnk, a perturbation framework based on single-cell gene regulatory network inference. We computationally simulated knockout of the target gene and assessed changes in network structure and downstream transcriptional programs. Differentially perturbed genes were then subjected to pathway enrichment analysis using MSigDB Hallmark gene sets to evaluate potential functional impacts. 2.15 Microenvironment interactions and communication analysis related to key genes. To characterize tumor microenvironment interaction patterns associated with key genes, epithelial cells were grouped by SLC2A1 expression status (SLC2A1⁺ vs. SLC2A1⁻), using a threshold of expression > 0. We then applied CellChat to infer ligand–receptor interaction networks between each epithelial group and other cell types, based on CellChatDB. We compared the communication strength and pathway activity between SLC2A1⁺ and SLC2A1⁻ epithelial cells, with a focus on interactions with fibroblasts/CAFs. We further zoomed in on collagen-related pathways to evaluate changes at the level of specific ligand–receptor pairs. To refine CAF heterogeneity, fibroblasts were re-clustered and CAF subtypes were annotated using marker genes (including POSTN⁺ CAFs). We then assessed the relative contribution of each CAF subtype to collagen-related signaling pathways at the subtype level. 2.16 Cell culture Human lung cancer cell lines H1299 and H1975 were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS). A549 and PC9 cells were maintained in DMEM containing 10% FBS. All cell lines were incubated at 37°C in a humidified atmosphere with 5% CO₂. 2.17 RT–qPCR Total RNA was extracted from cells using TRIzol reagent (Invitrogen, USA) according to the manufacturer’s instructions. Complementary DNA (cDNA) was synthesized using the PrimeScript™ RT reagent kit (RR037A, Takara, Japan). Quantitative real-time PCR was performed using TB Green® Premix Ex Taq™ II (RR820A, Takara, Japan) on a real-time PCR system. The primer sequences for SLC2A1 were as follows: SLC2A1 Forward (F): TGGCATCAACGCTGTCTTCT SLC2A1 Reverse (R): AACAGCGACACGACAGTGAA Primers were synthesized by Sangon Biotech (Shanghai, China). Relative gene expression levels were calculated using the 2^−ΔΔCt method and normalized to GAPDH. All primers used in this study are summarized in Supplementary Table S2. 2.18 Western blot Total protein was extracted from cells using RIPA lysis buffer supplemented with protease inhibitors. Protein concentrations were determined using a BCA protein assay kit. Equal amounts of protein were separated by SDS-PAGE and transferred onto PVDF membranes. The membranes were incubated with primary antibodies against SLC2A1 (1:1000; A6982, ABclonal) and GAPDH (1:5000; ab128915, Abcam), followed by incubation with appropriate secondary antibodies. Protein bands were visualized using enhanced chemiluminescence (ECL). All experiments were performed with at least three independent biological replicates. 2.19 SLC2A1 protein expression in LUAD The protein expression of SLC2A1 in LUAD tissues was evaluated using immunohistochemical (IHC) staining data obtained from the Human Protein Atlas (HPA, https://www.proteinatlas.org/). Representative images from normal lung and tumor tissues were analyzed to assess differential expression patterns. 2.20 Small interfering RNA (siRNA) transfection Small interfering RNAs (siRNAs) targeting SLC2A1 and corresponding negative control siRNA were synthesized by Hanheng Biotechnology (China). Cells were seeded in antibiotic-free medium and transfected with siRNAs using a lipid-based transfection reagent according to the manufacturer’s instructions. After 24–48 h of transfection, cells were harvested for subsequent experiments. Knockdown efficiency was confirmed by RT–qPCR and Western blot analysis. All experiments were performed in at least three independent biological replicates. The sequences of siRNAs are provided in Supplementary Table S2. 2.21 Cell viability assay (CCK-8) Cell viability was assessed using a Cell Counting Kit-8 (CCK-8; MCE, China). Transfected cells were seeded into 96-well plates at an appropriate density. At indicated time points, CCK-8 reagent was added to each well and incubated at 37°C. Absorbance was measured at 450 nm using a microplate reader. Cell viability was calculated relative to the control group. 2.22 Wound-healing assay Cells were seeded into 6-well plates and cultured until reaching near confluence. A straight scratch was created using a sterile pipette tip. Cells were washed with phosphate-buffered saline (PBS) to remove debris and then cultured in medium containing reduced serum. Images were captured at 0 h and 24 h using an inverted microscope. The wound area was quantified using ImageJ software, and the migration rate was calculated based on wound closure. 2.23 Transwell migration and invasion assays Cell migration and invasion assays were performed using Transwell chambers with 8-µm pore membranes (Corning, USA). For invasion assays, the upper chamber was pre-coated with Matrigel. Transfected cells suspended in serum-free medium were seeded into the upper chamber, while medium containing 10% FBS was added to the lower chamber as a chemoattractant. After incubation, cells on the upper surface were removed, and cells that migrated or invaded to the lower surface were fixed, stained with crystal violet, and counted under a microscope. Quantification was performed using ImageJ software. 2.24 Statistical analysis. All statistical analyses were performed in R. For two-group comparisons, Student’s t-test was used for normally distributed data, and the Wilcoxon rank-sum test was used for non-normally distributed data. For comparisons among multiple groups, one-way ANOVA (parametric) or the Kruskal–Wallis test (non-parametric) was applied. Correlations were assessed using Spearman’s correlation. Survival analyses were conducted using Kaplan–Meier curves with the log-rank test, and Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Multiple testing was controlled using the Benjamini–Hochberg method to control the false discovery rate (FDR). Unless otherwise specified, all tests were two-sided, and p < 0.05 was considered statistically significant. 3. Results Single-cell landscape of LUAD and normal lung tissues reveals altered cellular composition . Using the publicly available single-cell RNA sequencing dataset GSE308103, we performed an integrated analysis of 16 paired normal lung and LUAD samples. After quality control and doublet removal, approximately 300,000 cells were retained for downstream analyses. Unsupervised clustering followed by UMAP dimensionality reduction revealed a well-defined cellular landscape (Fig. 2 A). Based on canonical marker gene expression, seven major cell populations were identified, including T cells, B cells, plasma cells, myeloid cells, fibroblasts, endothelial cells, and epithelial cells (Fig. 2 D; Table S1 ). Visualization of normal and LUAD samples separately demonstrated differences in both cellular distribution and relative cell-type proportions between the two conditions, while each population exhibited consistent marker gene expression patterns across samples (Figs. 2 B–D). Quantitative analysis of cell-type composition further showed that the relative abundance of multiple cell populations was reshaped in LUAD tissues compared with normal lung (Figs. 2 E and 2 H). Specifically, LUAD samples displayed an increased proportion of epithelial cells accompanied by a reduced proportion of endothelial cells. In addition, immune cell composition differed between the two groups, with T cells and plasma cells relatively enriched in LUAD, whereas B cells were relatively decreased (Figs. 2 H and 2 I). These compositional changes were consistently observed across samples, as illustrated by stacked bar plots and radar plots, despite the presence of inter-sample heterogeneity (Figs. 2 F and 2 G). Collectively, this single-cell transcriptomic analysis delineates a comprehensive cellular landscape of LUAD and normal lung tissues and highlights a reshaping of cellular composition in the tumor context, providing a foundation for subsequent analyses of specific tumor-associated cell populations. Single-cell analysis reveals immunosuppressive remodeling of the immune microenvironment in LUAD To characterize immune alterations in LUAD at the single-cell level, we further examined T cell and myeloid populations based on the global single-cell atlas. Both cell types showed clear subpopulation structures and notable heterogeneity across normal and tumor tissues (Supplementary Figs. S1–S2). Importantly, LUAD samples exhibited a consistent shift in immune composition toward regulatory and suppressive cell populations. Within the T cell population, LUAD tissues were characterized by increased proportions of regulatory T cells (Treg) and follicular helper T cells (Tfh), along with a relative expansion of central memory T cells (Tcm). In contrast, cytotoxic T cells (Tcyto) and tissue-resident memory T cells (Trm), which are key mediators of anti-tumor immunity, were reduced (Supplementary Fig. S1 ; Table S1 ). This pattern suggests a weakened cytotoxic response accompanied by enhanced immune regulation. Similarly, in the myeloid population, LUAD samples showed an increased abundance of macrophages, with gene expression patterns consistent with M2-like polarization. These macrophages displayed features associated with immunosuppressive activity, including activation of inflammatory signaling pathways, metabolic reprogramming, and NF-κB–related responses (Supplementary Fig. S2 ). Together, these results indicate that the LUAD microenvironment is characterized by a coordinated shift toward immune suppression, providing an important context for understanding tumor cell–intrinsic epithelial states in subsequent analyses. Heterogeneity of epithelial cell subpopulations and their malignant features and collagen signaling communication. We further performed subclustering analysis of epithelial cells. UMAP dimensionality reduction showed that epithelial cells could be divided into nine transcriptionally distinct subpopulations (Epi_0–Epi_8) (Fig. 3 A). The relative proportions of these subpopulations differed between normal lung and LUAD tissues (Figs. 3 B and 3 C), indicating a redistribution of epithelial cell composition in tumor tissues. Analysis of marker gene expression showed clear differences in the expression levels of multiple genes among epithelial cell subpopulations (Fig. 3 D), supporting the validity of this classification. We then applied inferCNV to assess copy number variation (CNV) features and observed pronounced spatial heterogeneity of CNV scores across epithelial cells (Fig. 3 E). CNV score distributions varied among different epithelial subpopulations (Fig. 3 F). Consistent with this, malignant cell proportions were relatively higher in the Epi_5, Epi_6, and Epi_7 subpopulations, whereas other subpopulations were mainly composed of non-malignant cells (Fig. 3 G), suggesting differences in malignant features among epithelial cell subpopulations. To further explore interactions among epithelial cell subpopulations, we performed cell–cell communication analysis. The number and strength of interactions were not uniform across epithelial subpopulations: some subpopulations were more active as signal senders or receivers, whereas others showed lower overall interaction levels (Fig. 3 H). CellChat-based ligand–receptor analysis further revealed that the collagen signaling pathway represented a major component of communication among epithelial cell subpopulations, with notable differences in communication probabilities across subpopulations (Fig. 3 I). Network role analysis indicated that epithelial subpopulations played distinct roles within the collagen signaling communication network, with the Epi_0 subpopulation showing a relatively more prominent role in signal sending, receiving, or regulation (Fig. 3 J). Taken together, single-cell transcriptomic analysis reveals marked heterogeneity among epithelial cell subpopulations in LUAD with respect to transcriptional features, CNV-associated malignant characteristics, and collagen signaling–mediated cell–cell communication, providing a basis for further investigation of key epithelial cell subpopulations and their interactions with the tumor microenvironment. Prognostically relevant Scissor⁺ epithelial cells and their transcriptomic features. We applied the Scissor method to identify epithelial cell states associated with clinical outcomes. Scissor⁺ cells were unevenly distributed across epithelial subpopulations, with relatively higher proportions observed in the Epi_5 and Epi_6 subclusters (Fig. 4 A). UMAP visualization further showed that Scissor⁺ cells were spatially clustered and could be clearly distinguished from Scissor⁻ cells (Fig. 4 B). Differential expression analysis revealed marked differences in gene expression between Scissor⁺ and Scissor⁻ epithelial cells (Fig. 4 C), indicating that Scissor⁺ cells exhibit distinct transcriptomic features. Hallmark gene set enrichment analysis showed that Scissor⁺ epithelial cells were significantly enriched in multiple pathways related to tumor progression, including MYC targets, TNFα/NF-κB signaling, EMT, hypoxia, and cell cycle–related pathways (Fig. 4 D). Among these pathways, EMT showed strong enrichment in Scissor⁺ cells, which was further confirmed by GSEA analysis demonstrating an overall upregulation of EMT-related genes (Fig. 4 E). Comparison of the distribution of Scissor⁺ and Scissor⁻ cells across epithelial subpopulations using Ro/e (observed/expected) analysis showed that the Epi_5 and Epi_6 subclusters were relatively enriched in Scissor⁺ cells, whereas the Epi_7 subcluster was relatively enriched in Scissor⁻ cells (Fig. 4 F). These findings indicate that Scissor⁺ epithelial cells are preferentially enriched in specific subpopulations (Epi_5 and Epi_6), defining a distinct malignant epithelial cell state associated with adverse clinical outcomes. To assess the clinical impact of this malignant epithelial state, survival analysis was performed in the TCGA-LUAD cohort. Patients with high ssGSEA scores reflecting the activity of the Scissor⁺ epithelial state exhibited significantly worse overall survival compared with those with low activity (Fig. 4 G), whereas Scissor⁻ state activity showed the opposite trend. Consistently, gene signatures derived from Scissor⁺-enriched subclusters (particularly Epi_5 and Epi_6) were also associated with poor prognosis, whereas non-enriched subclusters showed no significant association. NicheNet analysis further identified several candidate ligands with high regulatory potential in Scissor⁺ epithelial cells. Among these, TGFB1 was markedly upregulated and ranked among the top predicted regulators (Fig. 4 H), supporting its potential role in mediating Scissor⁺-associated cell–cell communication. Finally, comparison of Scissor⁺-associated genes with genes globally upregulated in LUAD identified 39 overlapping genes (Fig. 4 I). This gene set therefore represents the core transcriptional program of the Scissor⁺ malignant epithelial cell state. Development of a Scissor⁺ epithelial state–derived risk score (SERS) for prognostic stratification in LUAD To translate the Scissor⁺ malignant epithelial cell state into a clinically measurable metric, we constructed a risk score model based on its core transcriptional program. Based on the 39 candidate genes, we implemented a systematic model selection framework and evaluated a total of 117 machine learning algorithms and their combinations across multiple cohorts, including TCGA, GSE31210, GSE50081, GSE72094, and a meta-cohort (Fig. 5 A). Among all evaluated models, the combination of LASSO regression and random survival forest (LASSO + RSF) demonstrated the most stable and best prognostic performance in terms of cross-cohort C-index. Using genes selected by LASSO (Fig. 5 B), we established a Scissor⁺ epithelial risk score (SERS), which represents a quantitative surrogate of the Scissor⁺ malignant epithelial state at the bulk transcriptomic level. The SERS model achieved robust prognostic performance in the TCGA cohort and maintained consistent predictive accuracy across multiple independent validation cohorts (Fig. 5 C), indicating strong generalizability. Patients stratified by SERS showed markedly different survival outcomes, with the high-SERS group exhibiting significantly worse overall survival across all cohorts (Fig. 5 D). In addition, SERS demonstrated favorable time-dependent AUCs and remained significantly associated with overall survival in Cox regression analyses, both as a categorical and continuous variable (Fig. 5 E–G). Collectively, these results demonstrate that SERS not only provides stable prognostic stratification but also effectively captures the activity of a high-risk malignant epithelial cell state, enabling its translation into clinically relevant applications in LUAD. SERS-defined malignant epithelial state is linked to immunosuppressive remodeling of the tumor microenvironment Given that the Scissor⁺ malignant epithelial state is characterized by activation of pathways such as EMT, hypoxia, and NF-κB signaling, which are known to influence immune regulation, we next investigated its relationship with the tumor immune microenvironment. Analyses using multiple immune infiltration algorithms revealed that the tumor immune landscape differed significantly between the SERS high- and low-risk groups. Compared with the SERS-Low group, the SERS-High group showed significant enrichment of immunosuppressive cell types, including M2 macrophages, monocyte-derived cells, and fibroblasts, whereas immune cell populations associated with anti-tumor responses were relatively enriched in the low-risk group (Fig. 6 A). These patterns were consistent across different immune estimation methods. Further analyses demonstrated that the SERS-High group exhibited higher scores for immunosuppressive features, while immune activation-related signatures were more prominent in the SERS-Low group (Fig. 6 B). Notably, cancer-associated fibroblasts (CAFs), estimated by the EPIC method, were significantly increased in the SERS-High group (Fig. 6 C–D), consistent with enhanced stromal interactions associated with Scissor⁺ epithelial cells. In addition, patients in the SERS-High group exhibited significantly higher tumor mutational burden (TMB) and tumor neoantigen burden (TNB) (Fig. 6 E–F). Despite this increased immunogenic potential, the SERS-High group maintained a strongly immunosuppressive microenvironment, suggesting that tumor-intrinsic epithelial states may override neoantigen-driven immune activation. Combined stratification analyses further demonstrated that integrating SERS with TMB, TNB, and immune suppressive features effectively distinguished patient subgroups with markedly different survival outcomes (Fig. 6 G). Collectively, these findings indicate that the Scissor⁺ epithelial state is linked to immunosuppressive remodeling of the tumor microenvironment, potentially contributing to poor prognosis and reduced immunotherapy responsiveness in LUAD. The Scissor⁺ epithelial state predicts immunotherapy outcomes and shows spatial enrichment in tumor regions To evaluate the clinical relevance of the Scissor⁺ malignant epithelial state in immunotherapy settings, we analyzed multiple independent cohorts treated with immune checkpoint inhibitors. SERS, as a quantitative surrogate of this epithelial state, showed a stable association with patient survival outcomes. Kaplan–Meier analyses revealed that, in both the POPLAR and OAK cohorts, patients in the SERS high-risk group had significantly worse overall survival (OS) and progression-free survival (PFS) than those in the low-risk group (Fig. 7 B–E). In the GSE135222 cohort, although the association with PFS did not reach statistical significance, a consistent trend toward poorer outcomes was observed (Fig. 7 A). To further investigate the spatial characteristics of this epithelial state, we analyzed spatial transcriptomic data from E-MTAB-13530. Tumor regions exhibited higher SERS scores and a markedly increased proportion of SERS-positive spatial spots compared with matched normal regions (B1 vs T1). This pattern was consistently observed across multiple samples (Fig. 7 F–J), indicating that the Scissor⁺ epithelial state is preferentially enriched within malignant regions. Collectively, these findings demonstrate that the Scissor⁺ epithelial state is not only associated with clinical outcomes in immunotherapy-treated patients but also exhibits a tumor-specific spatial distribution, further supporting its biological and clinical relevance in LUAD. SLC2A1 is a key gene associated with the Scissor⁺ malignant epithelial state We next sought to identify key genes underlying the Scissor⁺ malignant epithelial state using Mendelian randomization (MR) analyses. The results showed a significant causal association between SLC2A1 and LUAD risk (Fig. 8 A). Leave-one-out sensitivity analysis indicated that this association was not driven by any single SNP (Fig. 8 B), and consistent directions of effect were observed across different MR methods, including inverse variance weighted (IVW) and MR-Egger analyses (Fig. 8 C–D), supporting the robustness of this finding. We next examined the cell type–specific expression pattern of SLC2A1 using single-cell transcriptomic data. SLC2A1 expression was predominantly enriched in epithelial cell populations (Fig. 8 E–F), and was significantly upregulated in tumor epithelial cells compared with normal tissues (Fig. 8 G). This epithelial-specific expression pattern is consistent with its association with the Scissor⁺ malignant epithelial state. To further explore the functional relevance of SLC2A1, we performed virtual knockout analyses. Virtual deletion of SLC2A1 resulted in widespread transcriptional changes and affected multiple cancer-related pathways, including cell proliferation, metabolic reprogramming, and epithelial–mesenchymal transition (EMT) (Fig. 8 H–K). Notably, these pathways are also enriched in the Scissor⁺ epithelial state, suggesting that SLC2A1 may contribute to the functional characteristics of this state. Collectively, these results suggest that SLC2A1 represents a key molecular component of the Scissor⁺ malignant epithelial state, linking this state to tumor progression and malignant phenotypes in LUAD. SLC2A1 expression and genomic alterations support its role in the Scissor⁺ epithelial state Building on these findings, we examined the clinical and molecular features of SLC2A1 in LUAD. In the TCGA cohort, SLC2A1 mRNA expression was significantly higher in LUAD tissues than in normal lung tissues (Fig. 9 A), consistent with its enrichment in malignant epithelial cells identified in the Scissor⁺ state. Survival analyses showed that patients with high SLC2A1 expression had significantly worse overall survival (OS) and progression-free survival (PFS) (Fig. 9 B–C), suggesting that activation of SLC2A1 is associated with high-risk epithelial states. At the protein level, immunohistochemical staining demonstrated markedly increased SLC2A1 expression in LUAD tissues compared with normal lung tissues (Fig. 9 D–E), further supporting its tumor-specific upregulation. From a genomic perspective, SLC2A1 copy number alterations showed a weak but significant positive correlation with its mRNA expression (Fig. 9 F–G), indicating that genomic amplification may contribute to its activation in malignant epithelial cells. In addition, higher SLC2A1 expression was observed in TP53-mutant samples (Fig. 9 H), suggesting a potential link between genomic instability and activation of this epithelial program. Collectively, these findings further support that SLC2A1 is a key molecular feature of the Scissor⁺ malignant epithelial state, linking this state to adverse clinical outcomes and underlying genomic alterations in LUAD. SLC2A1 knockdown suppresses proliferation, migration, and invasion in LUAD cells Given the strong association of SLC2A1 with adverse clinical outcomes and malignant epithelial features, we next examined its functional role in LUAD cells. SLC2A1 expression was first examined in lung adenocarcinoma cell lines and normal bronchial epithelial cells. qRT-PCR analysis showed that SLC2A1 mRNA levels were higher in A549 and H1975 cells compared with BEAS-2B cells, while relatively lower expression was observed in PC9 and H1299 cells (Fig. 10 A). Western blot analysis showed a consistent pattern at the protein level, confirming elevated SLC2A1 expression in A549 and H1975 cells (Fig. 10 B). Based on these results, A549 and H1975 cells were selected for subsequent functional assays. Efficient knockdown of SLC2A1 was achieved using siRNA, as confirmed by qRT-PCR and Western blot analyses in both cell lines (Fig. 10 C–D). To evaluate the effect of SLC2A1 on cell proliferation, CCK-8 assays were performed. Silencing of SLC2A1 significantly inhibited the proliferation of A549 and H1975 cells over time compared with control cells (Fig. 10 E), indicating that SLC2A1 promotes LUAD cell growth. The role of SLC2A1 in cell migration was assessed using wound-healing assays. Knockdown of SLC2A1 markedly reduced wound closure in both A549 and H1975 cells at 48 hours compared with the control group (Fig. 10 F–G), suggesting impaired migratory capacity. In addition, Transwell assays were performed to further examine cell migration and invasion. SLC2A1 silencing significantly decreased the number of migrated and invaded cells in both A549 and H1975 cell lines (Fig. 10 H–I). Quantitative analysis consistently confirmed a significant reduction in both migration and invasion following SLC2A1 knockdown. Taken together, these results demonstrate that SLC2A1 promotes proliferation, migration, and invasion of LUAD cells, supporting its role as a functional mediator of the Scissor⁺ malignant epithelial state. Potential interactions between SLC2A1-positive epithelial cells and CAFs in LUAD Given that SLC2A1 promotes malignant phenotypes such as proliferation and migration, we next explored potential mechanisms underlying its role in the tumor microenvironment. In particular, we focused on interactions between SLC2A1-positive epithelial cells and cancer-associated fibroblasts (CAFs). Global network analysis suggested that interactions between SLC2A1-positive epithelial cells and fibroblasts were increased in both interaction number and strength, pointing to enhanced communication between these two cell types (Fig. 11 A–B). Further analysis of outgoing and incoming signaling showed that SLC2A1-positive epithelial cells exhibited increased reception of extracellular matrix–related signals, while fibroblasts acted as the primary signal sources (Fig. 11 C–D). Differential signaling analysis further suggested that fibroblast-to–SLC2A1-positive epithelial signaling was elevated, with collagen-related pathways showing the most prominent enrichment (Fig. 11 E). At the pathway level, communication network analysis indicated strong interactions between fibroblasts and SLC2A1-positive epithelial cells within collagen signaling pathways (Fig. 11 F). Ligand–receptor analysis further suggested that multiple collagen ligands and their corresponding receptors were enriched in signaling from fibroblasts to SLC2A1-positive epithelial cells, whereas these interactions were less pronounced in SLC2A1-negative epithelial cells (Fig. 11 G–H). Given the heterogeneity of fibroblasts in the tumor microenvironment, CAFs were further subdivided into distinct subpopulations. UMAP visualization identified multiple CAF subtypes within tumor tissues (Fig. 11 I). Communication analysis suggested that different CAF subpopulations exhibited varying interaction strengths with SLC2A1-positive epithelial cells, with the POSTN⁺ CAF subset showing the strongest contribution to collagen-mediated signaling (Fig. 11 J–M). Together, these findings suggest a potential interaction axis between SLC2A1-positive epithelial cells and CAFs, primarily mediated through collagen-related signaling pathways, which may contribute to the maintenance of the malignant epithelial state in LUAD. 4. Discussion Tumor development and progression is a dynamic process shaped by interactions between diverse cell states within a complex microenvironment(Tufail et al. 2026 ; Wang et al. 2022 ). With the advent of single-cell RNA sequencing and spatial transcriptomics, increasing evidence has revealed substantial functional heterogeneity across tumor-associated cell populations(Jia et al. 2025 ; Liu et al. 2026 ). Compared with traditional cell-type classifications, these fine-grained cellular states provide a more accurate representation of tumor biology(Quintero et al. 2023 ). However, in lung adenocarcinoma (LUAD), previous studies have reported significant alterations in cellular composition, including expansion of malignant epithelial cells, vascular abnormalities, and reshaped immune infiltration patterns(Liu et al. 2026 ; Shi et al. 2025 ; Tan et al. 2025 ; Wang et al. 2025 ). Most studies have focused on cell-type proportions or overall infiltration levels(Barjesteh van Waalwijk van Doorn-Khosrovani et al. 2024 ), while fewer have systematically characterized functional cell states, their dynamic transitions, and their roles within the tumor microenvironment(Jongbloed et al. 2025 ). In particular, integrative analyses linking cell states, cell–cell interactions, disease progression, and clinical outcomes remain limited(Chen and Liu 2025 ; Sun et al. 2023 ). As a result, the key cell states that drive LUAD progression, and the mechanisms by which microenvironmental signals shape these states and influence clinical outcomes, are still not fully understood. In this study, we identified a prognostically relevant malignant epithelial cell state in LUAD by integrating single-cell and bulk transcriptomic data. Defined by the Scissor framework, this state captures a functional program associated with tumor progression rather than a conventional cell-type classification. Building on this finding, we further translated this malignant epithelial state into a clinically applicable risk model (SERS), enabling a direct link between tumor cell states and patient outcomes. Single-cell analysis further revealed a marked reshaping of the LUAD microenvironment, characterized by expansion of epithelial cells and substantial changes in immune composition. Within the T cell compartment, LUAD tissues exhibited increased proportions of Treg, Tcm, and Tfh cells, along with a reduction in cytotoxic and tissue-resident memory T cells. Rather than representing isolated compositional changes, this pattern reflects a coordinated shift toward immune suppression, which may facilitate immune escape and support the emergence of aggressive tumor cell states. Consistent with these changes, myeloid cells also showed clear functional remodeling(Wang et al. 2026 ). Tumor-associated macrophages were enriched in LUAD and exhibited transcriptional features consistent with M2-like polarization, including increased metabolic activity and inflammatory signaling. Previous studies have shown that M2-like macrophages suppress immune responses and promote tumor progression through cytokine secretion and extracellular matrix remodeling(Locati et al. 2020 ; Nasir et al. 2023 ; Shao et al. 2025 ). Our findings further support a dynamic transition toward an immunosuppressive state in LUAD. Within this microenvironmental context, epithelial cells displayed pronounced heterogeneity(Fiore et al. 2025 ; Sheng et al. 2025 ). Distinct epithelial subclusters showed differences in CNV levels and malignant features(Dear 2009 ; Ma et al. 2026 ; Meng et al. 2025 ), suggesting the coexistence of multiple functional states during tumor progression. Importantly, cell–cell communication analysis indicated that epithelial cell behavior is influenced not only by intrinsic alterations but also by signals from the microenvironment. In particular, extracellular matrix–related signaling, especially collagen pathways, was prominently enriched, suggesting that interactions with stromal cells may support tumor growth and invasion. To link single-cell states more directly to clinical outcomes, we used the Scissor method to integrate single-cell transcriptomes with survival data from TCGA-LUAD. This analysis identified a group of Scissor⁺ epithelial cells that was strongly associated with poor prognosis. Notably, Scissor⁺ cells were not limited to one single epithelial subcluster. Instead, they were significantly enriched in specific subclusters. This suggests that Scissor⁺ cells may represent a “functional malignant state” related to tumor progression, rather than a traditional cell-type label. Consistent with this idea, differential expression and pathway enrichment analyses showed that Scissor⁺ epithelial cells had strong activation of cell-cycle programs, the G2/M checkpoint, MYC target genes, and EMT. These pathways have been widely linked to sustained proliferation, increased invasiveness, treatment resistance, and poor clinical outcomes(Jakobsen and Siersbæk 2025 ; Löbrich and Jeggo 2007 ; Zhang et al. 2025 ). Therefore, Scissor⁺ cells likely reflect more than a single molecular abnormality. They may capture a stable functional state that emerges during LUAD progression. Based on the Scissor⁺ epithelial state, we developed a prognostic model (SERS) that captures a biologically defined malignant program. Unlike conventional bulk-based models, SERS directly reflects a cell state associated with tumor progression, resulting in stable survival stratification across multiple cohorts. Notably, high SERS scores were consistently associated with an immunosuppressive microenvironment, characterized by enrichment of M2-like macrophages and activated CAFs, suggesting that this cell state is closely linked to immune evasion. To further identify key regulators of this malignant state, we applied Mendelian randomization and identified SLC2A1 as a potential causal driver. Consistent with this, SLC2A1 was highly expressed in tumor epithelial cells and associated with poor prognosis. In silico perturbation further indicated its involvement in pathways related to metabolism, proliferation, and EMT, supporting its central role in maintaining malignant phenotypes. Importantly, functional experiments confirmed that SLC2A1 promotes proliferation, migration, and invasion of LUAD cells, providing direct evidence that it actively drives tumor cell behavior rather than serving as a passive marker. These findings link the Scissor⁺ epithelial state to a specific molecular driver with functional relevance. Further analysis suggested a potential interaction axis between SLC2A1-positive epithelial cells and CAFs, mediated by collagen-related signaling. This suggests a potential model in which tumor-intrinsic metabolic activity and stromal extracellular matrix signals may cooperate to support tumor progression. Previous studies have extensively characterized the LUAD tumor microenvironment using single-cell transcriptomics, revealing substantial heterogeneity across cell types and functional states. In parallel, numerous prognostic models have been developed based on bulk transcriptomic data, identifying gene signatures associated with patient outcomes. In addition, key regulators such as SLC2A1 have been implicated in tumor metabolism and progression. However, these lines of research are often conducted independently, with limited integration across cellular states, molecular drivers, and clinical phenotypes. In this study, we build upon these prior efforts by integrating single-cell-defined cell states with bulk transcriptomic and clinical data. Rather than focusing solely on cell-type composition or gene-level associations, we identify a prognostically relevant malignant epithelial state (Scissor⁺) and systematically trace its impact across multiple levels. Specifically, we translate this cell state into a clinically applicable model (SERS), identify a key regulatory gene (SLC2A1) with causal and functional relevance, and further explore a potential mechanism involving stromal interactions. Importantly, this study provides several advances over previous work. First, we identify a prognostically relevant malignant epithelial cell state rather than relying on conventional cell-type classifications. Second, we translate this cell state into a clinically applicable risk model (SERS) with stable performance across multiple cohorts. Third, by integrating Mendelian randomization, single-cell expression, and functional experiments, we identify SLC2A1 as a potential driver linking tumor cell states to malignant phenotypes. Unlike previous studies that focus on either cell-type composition or bulk-based signatures, our study integrates cell-state identification, causal inference, and functional validation within a single framework. Together, these findings establish a unified framework connecting cell states, molecular drivers, functional phenotypes, and microenvironmental interactions in LUAD, providing new insights into tumor progression and therapeutic response. Although this study used multi-omics integration to reveal multi-level features of the LUAD microenvironment and key cell states, it has several limitations. First, our work is mainly based on bioinformatic analyses. While some key genes were supported at the transcript level, further validation is needed at the protein level. Functional and mechanistic experiments are also required. Second, the detailed molecular mechanism of the SLC2A1–CAF–collagen axis still needs to be tested in vitro and in vivo. Third, technical differences across datasets and sample heterogeneity may affect the results. Future studies should validate the clinical utility of SERS in larger, multi-center, prospective cohorts. In addition, although functional experiments support the role of SLC2A1, further studies are required to establish its direct mechanistic links with microenvironmental interactions in vivo. In this study, we systematically characterized the tumor microenvironment of LUAD at the single-cell level and identified a malignant epithelial cell state associated with poor prognosis. By integrating single-cell and bulk transcriptomic data, we further translated this high-risk cell state into a clinically applicable risk model (SERS), thereby establishing a direct link between tumor cell states and patient outcomes. Building on this framework, we identified SLC2A1 as a key gene associated with the Scissor⁺ malignant epithelial state through integrative analyses, including Mendelian randomization, single-cell expression profiling, and in silico perturbation. Importantly, functional experiments demonstrated that SLC2A1 promotes proliferation, migration, and invasion of LUAD cells, providing direct evidence that this gene is not only associated with malignant epithelial states but also actively drives tumor cell behavior. Furthermore, cell–cell communication analysis suggested a potential interaction axis between SLC2A1-positive epithelial cells and cancer-associated fibroblasts (CAFs), primarily mediated by collagen-related signaling. This finding points to a possible mechanism by which tumor-intrinsic metabolic programs may interact with stromal signals to support malignant phenotypes. Although further experimental validation is required, this SLC2A1–CAF–collagen axis provides a biologically plausible model linking tumor cell states with microenvironmental regulation. Consistent with these findings, the cell state captured by SERS was also associated with patient outcomes in immunotherapy-treated cohorts, suggesting that tumor-intrinsic epithelial programs, together with their microenvironmental interactions, may influence treatment response. Overall, our study establishes a framework that connects malignant epithelial cell states, key regulatory genes, functional tumor phenotypes, and microenvironmental interactions in LUAD. These findings not only improve our understanding of tumor progression but also provide potential targets and strategies for precision therapy. Declarations Ethics approval and consent to participate As all data were obtained from publicly available databases and no identifiable human data were used, ethical approval and informed consent were not required. Consent for publication Not applicable. Competing interests The authors declare that they have no competing financial or non-financial interests. Funding This work was supported by Guizhou Provincial Administration of Traditional Chinese Medicine's Traditional Chinese Medicine and Ethnic Medicine Science and Technology Research Project (No. QZYY-2023-108); Noncommunicable Chronic Diseases – National Science and Technology Major Project (No. 2023ZD0502105); the National Natural Science Foundation of China (No. 82504050); Zunyi Municipal Science and Technology Cooperation Project (No. HZ [2025] 129). Author Contribution LS: Conceptualization, Data curation, Methodology, Software, Formal Analysis, Writing – original draft. SX: Conceptualization, Methodology. YC: Validation, Formal Analysis, Software. PL: Formal Analysis, Methodology. HM: Data curation, Validation, Writing – review & editing, Funding acquisition. All authors discussed and approved the final manuscript. Acknowledgements The authors thank TCGA, GEO, ArrayExpress, and the Human Protein Atlas for providing access to the datasets and protein expression resources used in this study. Data Availability The datasets generated and/or analyzed during the current study are publicly available in the GEO, TCGA, and ArrayExpress repositories. Accession numbers include GSE308103, GSE31210, GSE50081, GSE72094, GSE135222, and E-MTAB-13530. Additional data supporting the findings of this study are included within the article and its supplementary materials. References Andreatta M, Garnica J, Carmona SJ (2025) Identification of malignant cells in single-cell transcriptomics data. Commun Biol 8:1264 Barjesteh van Waalwijk, van Doorn-Khosrovani S, Van Kholmanskikh O, Koole S, Thomas DM, Gelderblom H (2024) Testing dilemmas in the clinic: Lessons learned from biomarker-based drug development. Cancer Cell 42:923–929. 10.1016/j.ccell.2024.05.014 Bortolot M, Remon J, Bironzo P, Cortiula F, Menis J, Chan SW, van Geel R, Reguart N, Arrieta O, Mountzios G, Dingemans AC, Besse B, Hendriks LEL (2025) De-escalation strategies with targeted therapies in non-small cell lung cancer. Cancer Treat Rev 139:102995. 10.1016/j.ctrv.2025.102995 Boxer E, Feigin N, Tschernichovsky R, Darnell NG, Greenwald AR, Hoefflin R, Kovarsky D, Simkin D, Turgeman S, Zhang L, Tirosh I (2025) Emerging clinical applications of single-cell RNA sequencing in oncology. Nat Rev Clin Oncol 22:315–326. 10.1038/s41571-025-01003-3 Chen T, Liu F (2025) Neoadjuvant immunotherapy in early-stage NSCLC: navigating biomarker dilemmas and special population challenges. Lung Cancer 204:108588. 10.1016/j.lungcan.2025.108588 Dear PH (2009) Copy-number variation: the end of the human genome? Trends Biotechnol 27:448–454. 10.1016/j.tibtech.2009.05.003 Fiore VF, Almagro J, Fuchs E (2025) Shaping epithelial tissues by stem cell mechanics in development and cancer. Nat Rev Mol Cell Biol 26:442–455 Herbst RS, Morgensztern D, Boshoff C (2018) The biology and management of non-small cell lung cancer. Nature 553:446–454. 10.1038/nature25183 Huang Z, Chen J, Zhu T, Li J, Ng HY, Zhou Y, Gu X, Xu S, Jia R (2025) Cancer-associated fibroblasts in the tumor microenvironment: heterogeneity, crosstalk mechanisms, and therapeutic implications. Mol Cancer 25:19. 10.1186/s12943-025-02533-1 Jakobsen ST, Siersbæk R (2025) Transcriptional regulation by MYC: an emerging new model. Oncogene 44:1–7. 10.1038/s41388-024-03174-2 Jia H, Chen X, Zhang L, Chen M (2025) Cancer associated fibroblasts in cancer development and therapy. J Hematol Oncol 18:36 Jongbloed M, Bortolot M, Willmann J, Bartolomeo V, Novoa NM, De Ruysscher DKM, Hendriks LEL (2025) Current Controversies and Challenges in Non-Oncogene-Addicted Synchronous Oligometastatic Non-Small Cell Lung Cancer: A Review. JAMA Oncol 11:1385–1392. 10.1001/jamaoncol.2025.2891 Keenan BP, Yadav M, Ansstas G, Fabrizio D, Murugesan K, Montesion M, Guha Niyogi D, Mellman I, Melero I (2025) Intratumoral heterogeneity and immunotherapy resistance: clinical implications. Ann Oncol DOI. 10.1016/j.annonc.2025.10.1239 Liu Y, Dai Y, Wang L (2026) Spatial omics at the forefront: emerging technologies, analytical innovations, and clinical applications. Cancer Cell 44:24–49. 10.1016/j.ccell.2025.12.009 Liu Y, Zhou C, Tang Y, Lei H, Aihemaiti A, Liu H, Zou P, Xie J, Guo X, Xia R, Han BH, Chen H, Zhu L (2026) Targeting AKR1B1 reprograms tumor-associated macrophages to enhance antitumor immunity. J Immunother Cancer 14. 10.1136/jitc-2025-014043 Löbrich M, Jeggo PA (2007) The impact of a negligent G2/M checkpoint on genomic instability and cancer induction. Nat Rev Cancer 7:861–869. 10.1038/nrc2248 Locati M, Curtale G, Mantovani A (2020) Diversity, Mechanisms, and Significance of Macrophage Plasticity. Annu Rev Pathol 15:123–147 Ma L, Xiong B, Liu M, Tan K (2026) Cellular neighborhoods in cancer. Nat Cancer. 10.1038/s43018-025-01107-w Meng F, Li J, Xia Z, Wang Q, Sun Q, Wang S, Xu L, Yin R (2025) Persistent lineage plasticity driving lung cancer development and progression. Clin Transl Med 15:e70458 Nasir I, McGuinness C, Poh AR, Ernst M, Darcy PK, Britt KL (2023) Tumor macrophage functional heterogeneity can inform the development of novel cancer therapies. Trends Immunol 44:971–985. 10.1016/j.it.2023.10.007 Quintero JC, Díaz NF, Rodríguez-Dorantes M, Camacho-Arroyo I (2023) Cancer Stem Cells and Androgen Receptor Signaling: Partners in Disease Progression. Int J Mol Sci 24 Reck M, Frost N, Peters S, Fox BA, Ferrara R, Savai R, Barlesi F (2025) Treatment of NSCLC after chemoimmunotherapy - are we making headway? Nat Rev Clin Oncol 22:806–830. 10.1038/s41571-025-01061-7 Shao N, Qiu H, Liu J, Xiao D, Zhao J, Chen C, Wan J, Guo M, Liang G, Zhao X, Xu L (2025) Targeting lipid metabolism of macrophages: A new strategy for tumor therapy. J Adv Res 68:99–114 Shen Y, Chen JQ, Li XP (2025) Differences between lung adenocarcinoma and lung squamous cell carcinoma: Driver genes, therapeutic targets, and clinical efficacy. Genes Dis 12:101374 Sheng R, Yin Y, Wang X (2025) Mesothelial and immune cells interplay in the tumor microenvironment. Trends Mol Med 31:895–908. 10.1016/j.molmed.2025.03.014 Shi X, Hu L, Huang Y, Zhang MF, Ying X, Yuan X, Wen N, Lu J, Zou H, Tan X, He QY, Wang F, Yang H, Zhang CZ (2025) Integrative proteomic characterization of human lung adenocarcinoma with KRAS G12 mutations reveals molecular pathogenesis. J Adv Res DOI. 10.1016/j.jare.2025.09.014 Siegel RL, Kratzer TB, Giaquinto AN, Sung H, Jemal A (2025) Cancer statistics, 2025. CA Cancer J Clin 75:10–45 Sun D, Guan X, Moran AE, Wu LY, Qian DZ, Schedin P, Dai MS, Danilov AV, Alumkal JJ, Adey AC, Spellman PT, Xia Z (2022) Identifying phenotype-associated subpopulations by integrating bulk and single-cell sequencing data. Nat Biotechnol 40:527–538 Sun Q, Hong Z, Zhang C, Wang L, Han Z, Ma D (2023) Immune checkpoint therapy for solid tumours: clinical dilemmas and future trends. Signal Transduct Target Ther 8:320 Sun Y, Shao C, Duan H, Wang Z, Xu S, Wang C, Xiu J, Liu J, Wang X, Yao X, Gao Y, Yan X (2025) Dynamic Evolution of the Tumor Immune Microenvironment in Malignant Tumors and Emerging Therapeutic Paradigms. MedComm (2020) 6:e70496 Tan Y, Tan W, Liang Y, Long Y, Chen S, Hu Q, Ou Y, Fu J, Chen H, Ren F, Ye J, Zhou Q, Li S, He X, Wang Q, Shen Y, Lu H, Wu D, Gao A, Chen X, Li Y (2025) Machine learning-enabled spatial multi-omics uncovers lactate-driven targets and tumor microenvironmental reprogramming in cancer. NPJ Digit Med. 10.1038/s41746-025-02286-7 Tirosh I, Suva ML (2024) Cancer cell states: Lessons from ten years of single-cell RNA-sequencing of human tumors. Cancer Cell 42:1497–1506. 10.1016/j.ccell.2024.08.005 Tufail M, Gong K, Ijaz B, Patel H, Lui WO, Wang X, Li J (2026) The hallmarks of oncogenic signaling: From pathways to resistance in cancer therapy. Drug Resist Updat 85:101355. 10.1016/j.drup.2026.101355 Wang H, Niu X, Jin Z, Zhang S, Fan R, Xiao H, Hu SS (2025) Immunotherapy resistance in non-small cell lung cancer: from mechanisms to therapeutic opportunities. J Exp Clin Cancer Res 44:250 Wang S, Hu D, Wang R, Huang J, Wang B (2025) Integrative multi-omics and machine learning reveal critical functions of proliferating cells in prognosis and personalized treatment of lung adenocarcinoma. NPJ Precis Oncol 9:243 Wang X, Luo X, Xiao R, Liu X, Zhou F, Jiang D, Bai J, Cui M, You L, Zhao Y (2026) Targeting metabolic-epigenetic-immune axis in cancer: molecular mechanisms and therapeutic implications. Signal Transduct Target Ther 11:28 Wang Z, Xing Y, Li B, Li X, Liu B, Wang Y (2022) Molecular pathways, resistance mechanisms and targeted interventions in non-small-cell lung cancer. Mol Biomed 3:42 Xu L, Zou C, Zhang S, Chu TSM, Zhang Y, Chen W, Zhao C, Yang L, Xu Z, Dong S, Yu H, Li B, Guan X, Hou Y, Kong FM (2022) Reshaping the systemic tumor immune environment (STIE) and tumor immune microenvironment (TIME) to enhance immunotherapy efficacy in solid tumors. J Hematol Oncol 15:87 Zhang CX, Huang RY, Sheng G, Thiery JP (2025) Epithelial-mesenchymal transition. Cell 188:5436–5486. 10.1016/j.cell.2025.08.033 Zhang Y, Chen M, Fang X, Han Y, Li Y (2025) Progression and Metastasis of Lung Cancer: Clinical Features, Molecular Mechanisms, and Clinical Managements. MedComm (2020) 6:e70477 Zhao J, Xu W, Zhou F, Zhang X, Zhou M, Miao D, Yu L, Zhang Y, Fan J, Zhou C, Li W, Mok T, Le X, Li M, Xia Y (2026) Navigating the landscape of EGFR TKI resistance in EGFR-mutant NSCLC - mechanisms and evolving treatment approaches. Nat Rev Clin Oncol 23:63–83. 10.1038/s41571-025-01085-z Additional Declarations No competing interests reported. Supplementary Files figureS1.jpg figureS2.jpg figureS3.jpg figureS4.jpg figureS5.jpg TableS1.xlsx TableS3.xlsx TableS2.xlsx SupplementaryMaterials.pdf Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 08 May, 2026 Reviews received at journal 07 May, 2026 Reviews received at journal 06 May, 2026 Reviewers agreed at journal 06 May, 2026 Reviewers agreed at journal 06 May, 2026 Reviewers agreed at journal 06 May, 2026 Reviewers invited by journal 06 May, 2026 Editor assigned by journal 07 Apr, 2026 Submission checks completed at journal 07 Apr, 2026 First submitted to journal 02 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9298986","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":616363134,"identity":"21de08cb-bd1f-4d47-96ef-b99834bf8f15","order_by":0,"name":"Linqian Song","email":"","orcid":"","institution":"The Second Affiliated Hospital of Zunyi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Linqian","middleName":"","lastName":"Song","suffix":""},{"id":616363135,"identity":"ac458ffd-b999-4bc1-b790-5b6a28c6b236","order_by":1,"name":"Shiyun Xing","email":"","orcid":"","institution":"The Second Affiliated Hospital of Zunyi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shiyun","middleName":"","lastName":"Xing","suffix":""},{"id":616363136,"identity":"6d159b50-eb22-4709-9110-1771cd3abb2e","order_by":2,"name":"Peijie Li","email":"","orcid":"","institution":"The Second Affiliated Hospital of Zunyi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Peijie","middleName":"","lastName":"Li","suffix":""},{"id":616363138,"identity":"bb46c2c4-6683-473e-af7e-95f16bf13ddc","order_by":3,"name":"Yunliang Cao","email":"","orcid":"","institution":"The Second Affiliated Hospital of Zunyi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yunliang","middleName":"","lastName":"Cao","suffix":""},{"id":616363139,"identity":"401cd6b3-081f-4be9-aed7-08c8ffbeace9","order_by":4,"name":"Hu Ma","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAqUlEQVRIiWNgGAWjYFACHgYGxgYbHn7+BtK0pMlIzjhAmpbDNgYNCURq0G0/e0zi547zPAYMBxg/fMwhQovZmbw0yd4zt3nMmRuYJWduI0bLgRwzCd622zyWDQfYmHmJ0nL+jZnk37ZzPAYHEojVciPHTJq37QBJWt4YW8ueSeaRnHGwmUi/nM8xvPl2h509P3/zwQ8fidGCBBgbSFM/CkbBKBgFowA3AAAjjTav9srR2wAAAABJRU5ErkJggg==","orcid":"","institution":"The Second Affiliated Hospital of Zunyi Medical University","correspondingAuthor":true,"prefix":"","firstName":"Hu","middleName":"","lastName":"Ma","suffix":""}],"badges":[],"createdAt":"2026-04-02 06:38:44","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9298986/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9298986/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106094579,"identity":"bd9ca958-91ab-43ae-9910-88cdfd2c0434","added_by":"auto","created_at":"2026-04-03 11:42:54","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":8867457,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverview of the study workflow.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/fefc51cc3826353d2511d2fe.jpg"},{"id":106071672,"identity":"c61c123b-b48e-46b5-a123-75773bb60730","added_by":"auto","created_at":"2026-04-03 06:42:22","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1460623,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSingle-cell transcriptomic landscape of LUAD and normal tissues. (A) \u003c/strong\u003eUMAP visualization of seven major cell populations identified in the GSE308103. \u003cstrong\u003e(B–C) \u003c/strong\u003eUMAP distribution of cells from Normal and LUAD. \u003cstrong\u003e(D)\u003c/strong\u003e Dot plot showing the expression of representative marker genes for each cell population. \u003cstrong\u003e(E)\u003c/strong\u003eOverall proportions of major cell populations in the Normal and LUAD groups. \u003cstrong\u003e(F) \u003c/strong\u003eDistribution of cell-type compositions across individual samples. \u003cstrong\u003e(G) \u003c/strong\u003eComparison of the relative abundance of each cell population between Normal and LUAD tissues. \u003cstrong\u003e(H) \u003c/strong\u003ePie charts showing the composition of major cell populations in Normal and LUAD tissues. \u003cstrong\u003e(I) \u003c/strong\u003eBar plots showing the average proportions of different cell populations in the Normal and LUAD groups.\u003c/p\u003e","description":"","filename":"figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/f7da0a779f14e2f225b5a590.jpg"},{"id":106094926,"identity":"31a7b87c-1008-4edf-91e7-dd93e5b3cc7a","added_by":"auto","created_at":"2026-04-03 11:43:38","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1570512,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHeterogeneity and transcriptional features of epithelial cells in LUAD.\u003c/strong\u003e \u003cstrong\u003e(A)\u003c/strong\u003e UMAP visualization of epithelial cell subpopulations (Epi_0–Epi_8). \u003cstrong\u003e(B) \u003c/strong\u003eProportions of different epithelial cell subpopulations in the Normal and LUAD groups. \u003cstrong\u003e(C) \u003c/strong\u003eDistribution of epithelial cell subpopulations between the Normal and LUAD groups. \u003cstrong\u003e(D) \u003c/strong\u003eHeatmap showing the expression of representative marker genes across epithelial cell subpopulations. \u003cstrong\u003e(E)\u003c/strong\u003e UMAP visualization of copy number variation (CNV) scores in epithelial cells. \u003cstrong\u003e(F) \u003c/strong\u003eViolin plots showing CNV scores of epithelial cell subpopulations calculated using inferCNV. \u003cstrong\u003e(G) \u003c/strong\u003eComparison of the proportions of malignant and non-malignant cells across different epithelial cell subpopulations. \u003cstrong\u003e(H) \u003c/strong\u003eHeatmaps showing the number and strength of cell–cell communication interactions among epithelial cell subpopulations. \u003cstrong\u003e(I) \u003c/strong\u003eBubble plot showing CellChat-based cell–cell communication analysis of ligand–receptor interactions related to the collagen (Collagen) signaling pathway. \u003cstrong\u003e(J) \u003c/strong\u003eRole importance analysis of epithelial cell subpopulations in the cell–cell communication network based on the collagen signaling pathway.\u003c/p\u003e","description":"","filename":"figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/26faaeab9f8ad59030799dfc.jpg"},{"id":106095013,"identity":"f2025dad-38f9-403c-bbe3-297132b9b6af","added_by":"auto","created_at":"2026-04-03 11:43:57","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2042135,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentification of prognostically relevant Scissor⁺ epithelial cells in LUAD. (A)\u003c/strong\u003e Dot plot showing the abundance and proportion of Scissor−, Scissor+ and Background cells across epithelial subclusters (Epi_0–Epi_8). \u003cstrong\u003e(B)\u003c/strong\u003eUMAP visualization of cells colored by Scissor classification. \u003cstrong\u003e(C)\u003c/strong\u003eVolcano plot of differentially expressed genes between Scissor+ and Scissor− epithelial cells. \u003cstrong\u003e(D)\u003c/strong\u003e Hallmark gene set enrichment analysis showing pathways significantly enriched in Scissor+ cells. \u003cstrong\u003e(E)\u003c/strong\u003e GSEA enrichment of the epithelial–mesenchymal transition (EMT) pathway. \u003cstrong\u003e(F)\u003c/strong\u003e Ro/e (observed/expected) heatmap showing the relative enrichment or depletion of epithelial subclusters in Scissor+ versus Scissor− cells. \u003cstrong\u003e(G)\u003c/strong\u003eKaplan–Meier survival analysis in the TCGA-LUAD cohort stratified by ssGSEA scores derived from Scissor+, Scissor−, and Scissor+-enriched epithelial subcluster (Epi_5, Epi_6 and Epi_7) gene signatures. \u003cstrong\u003e(H)\u003c/strong\u003e Heatmap of candidate ligands predicted by NicheNet and their regulatory potential in Scissor+ epithelial cells. \u003cstrong\u003e(I)\u003c/strong\u003e Venn diagram showing the overlap between LUAD upregulated genes and Scissor+ epithelial cell–associated genes.\u003c/p\u003e","description":"","filename":"figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/407304e72f0e130e4745789b.jpg"},{"id":106095031,"identity":"2e83411d-a083-4d8e-af50-1e1fcebf1c96","added_by":"auto","created_at":"2026-04-03 11:44:04","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":2970459,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConstruction and validation of the SERS prognostic model. (A)\u003c/strong\u003e Heatmap showing the C-index values of 117 machine learning methods and their combinations across multiple cohorts. \u003cstrong\u003e(B)\u003c/strong\u003e Model feature genes selected by LASSO regression. \u003cstrong\u003e(C)\u003c/strong\u003e C-index performance of the LASSO + RSF model in the TCGA, GSE31210, GSE50081, GSE72094, and Meta cohorts. \u003cstrong\u003e(D)\u003c/strong\u003e Kaplan–Meier survival curves based on the risk score, showing overall survival of high- and low-risk groups across different cohorts. \u003cstrong\u003e(E)\u003c/strong\u003e One-year AUC values based on the risk score in each cohort. \u003cstrong\u003e(F)\u003c/strong\u003e Forest plot of univariate Cox regression analysis based on the risk score (categorical variable). \u003cstrong\u003e(G)\u003c/strong\u003eForest plot of univariate Cox regression analysis based on the risk score (continuous variable). *P\u0026lt;0.05, **P\u0026lt;0.01 and ***P\u0026lt;0.001.\u003c/p\u003e","description":"","filename":"figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/eaebacfde85b6ce65451bd02.jpg"},{"id":106095068,"identity":"bbfafca3-0b91-4c1c-9a23-048eef823f6e","added_by":"auto","created_at":"2026-04-03 11:44:09","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1601193,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAssociation between SERS and the tumor immune microenvironment in LUAD. (A)\u003c/strong\u003e Heatmap of tumor microenvironment cell types estimated by multiple immune infiltration algorithms (SERS-High vs. SERS-Low). \u003cstrong\u003e(B)\u003c/strong\u003e Heatmap of tumor microenvironment features classified according to immune suppression, immune activation, and immune-related biomarkers. \u003cstrong\u003e(C)\u003c/strong\u003e Comparison of MDSC-like cell abundance between the SERS-High and SERS-Low groups. \u003cstrong\u003e(D)\u003c/strong\u003e Comparison of CAF abundance (estimated by the EPIC method) between the SERS-High and SERS-Low groups. \u003cstrong\u003e(E)\u003c/strong\u003e Group-wise distribution of tumor mutational burden (TMB). \u003cstrong\u003e(F)\u003c/strong\u003e Group-wise distribution of tumor neoantigen burden (TNB). \u003cstrong\u003e(G)\u003c/strong\u003eKaplan–Meier survival curves based on combined stratification of SERS, TMB, TNB, CAFs (EPIC), and MDSC-like cell abundance. *P\u0026lt;0.05, **P\u0026lt;0.01 and ***P\u0026lt;0.001.\u003c/p\u003e","description":"","filename":"figure6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/b0243cc17e2d62c223225d54.jpg"},{"id":106094447,"identity":"9b3b21e9-ded9-4290-9ea2-df45c83a24db","added_by":"auto","created_at":"2026-04-03 11:42:37","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":2838976,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClinical relevance of SERS in immunotherapy-treated cohorts and spatial transcriptomic samples. \u003c/strong\u003e(A–E) Kaplan–Meier survival curves according to SERS score in immunotherapy-treated cohorts. A) GSE135222, PFS; B) POPLAR, OS; C) POPLAR, PFS; D) OAK, OS; E) OAK, PFS. (F–J) Spatial transcriptomic visualization of SERS score. F) UMAP of spatial transcriptomic spots; G) P10 (B1, normal; T1, tumor); H) P11 (B1, normal; T1, tumor); I) P15 (B1, normal; T1, tumor); J) P25 (B1, normal; T1, tumor).\u003c/p\u003e","description":"","filename":"figure7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/278ef000cf92c0602f7cb701.jpg"},{"id":106095042,"identity":"1e991daf-b212-4e7e-a27c-e106b38618d6","added_by":"auto","created_at":"2026-04-03 11:44:06","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":937954,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCausal inference and functional prediction of SLC2A1 in LUAD. (A)\u003c/strong\u003e Mendelian randomization (MR) analysis assessing the causal effect of SLC2A1 on lung adenocarcinoma (LUAD). \u003cstrong\u003e(B)\u003c/strong\u003e Leave-one-out sensitivity analysis evaluating the influence of individual SNPs on the MR estimates for SLC2A1 and LUAD. \u003cstrong\u003e(C)\u003c/strong\u003eCausal effect estimates of SLC2A1 obtained using different MR methods (IVW and MR-Egger). \u003cstrong\u003e(D)\u003c/strong\u003e Scatter plot showing the relationships between SNP effects on SLC2A1 and LUAD risk.\u003cstrong\u003e (E)\u003c/strong\u003e UMAP visualization of SLC2A1 expression in single-cell transcriptomic data. \u003cstrong\u003e(F)\u003c/strong\u003e Proportion of SLC2A1-positive cells across different cell types. \u003cstrong\u003e(G)\u003c/strong\u003e Comparison of SLC2A1 expression levels in epithelial cells between normal and LUAD samples.\u003cstrong\u003e (H)\u003c/strong\u003e Ranking of differentially regulated genes following virtual knockout of SLC2A1. \u003cstrong\u003e(I)\u003c/strong\u003eDistribution of network Z-scores in the virtual knockout analysis. \u003cstrong\u003e(J)\u003c/strong\u003eHallmark pathways affected by virtual knockout of SLC2A1. \u003cstrong\u003e(K)\u003c/strong\u003e Hallmark gene set enrichment analysis (GSEA) based on genes associated with virtual knockout of SLC2A1.\u003c/p\u003e","description":"","filename":"figure8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/6feec02674a11d9e702e3983.jpg"},{"id":106096014,"identity":"f509d7c1-b8cd-42b4-85db-6e8ea68ae517","added_by":"auto","created_at":"2026-04-03 11:52:13","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":2182637,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExpression patterns and clinical relevance of SLC2A1 in LUAD. (A) \u003c/strong\u003eComparison of SLC2A1 mRNA expression between normal lung tissues and lung adenocarcinoma (LUAD) tissues in the TCGA cohort. \u003cstrong\u003e(B)\u003c/strong\u003e Kaplan–Meier overall survival (OS) curves stratified by SLC2A1 expression levels. \u003cstrong\u003e(C) \u003c/strong\u003eKaplan–Meier progression-free survival (PFS) curves stratified by SLC2A1 expression levels. \u003cstrong\u003e(D) \u003c/strong\u003eRepresentative immunohistochemistry (IHC) images of SLC2A1 protein in normal lung and LUAD tissues. \u003cstrong\u003e(E) \u003c/strong\u003eComparison of SLC2A1 protein IHC scores between normal lung and LUAD tissues. \u003cstrong\u003e(F) \u003c/strong\u003eCorrelation analysis between SLC2A1 copy number alterations and its mRNA expression. \u003cstrong\u003e(G) \u003c/strong\u003eDistribution of SLC2A1 expression across different copy number alteration groups (shallow deletion, diploid, gain, and amplification). \u003cstrong\u003e(H) \u003c/strong\u003eHeatmap showing the relationship between SLC2A1 expression levels and common gene mutation profiles.\u003cstrong\u003e (I) \u003c/strong\u003eComparison of SLC2A1 expression levels according to TP53 and TTN mutation status.\u003c/p\u003e","description":"","filename":"figure9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/277c97daa9c91f4c17c89137.jpg"},{"id":106071679,"identity":"d03375b9-734f-442b-bd50-ad7b1519e655","added_by":"auto","created_at":"2026-04-03 06:42:22","extension":"jpg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":5485312,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSLC2A1 promotes proliferation, migration, and invasion of LUAD cells. (A)\u003c/strong\u003e Relative mRNA expression levels of SLC2A1 in LUAD cell lines (A549, PC9, H1299, H1975) and normal bronchial epithelial cells (BEAS-2B) measured by qRT-PCR. \u003cstrong\u003e(B)\u003c/strong\u003eProtein expression levels of SLC2A1 in the indicated cell lines determined by Western blotting. \u003cstrong\u003e(C–D)\u003c/strong\u003e Knockdown efficiency of SLC2A1 in A549 and H1975 cells validated by qRT-PCR (C) and Western blotting (D) after siRNA transfection. \u003cstrong\u003e(E) \u003c/strong\u003eCell proliferation assessed by CCK-8 assay in A549 and H1975 cells following SLC2A1 silencing. \u003cstrong\u003e(F–G) \u003c/strong\u003eWound-healing assays showing reduced migration ability in SLC2A1-knockdown cells compared with control cells. \u003cstrong\u003e(H–I)\u003c/strong\u003e Transwell assays demonstrating decreased migration and invasion capacities in A549 and H1975 cells after SLC2A1 knockdown. Data are presented as mean ± SD from three independent experiments. *P \u0026lt; 0.05, **P \u0026lt; 0.01, ***P \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"Figure10.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/6e41cc91632a48f0d544a52a.jpg"},{"id":106071681,"identity":"92a923e8-42f1-4ce0-8cb9-d0f77379e25d","added_by":"auto","created_at":"2026-04-03 06:42:22","extension":"jpg","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":1664383,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInteraction between SLC2A1-positive epithelial cells and CAFs in LUAD. (A–B)\u003c/strong\u003e Network diagrams showing intercellular interactions among different cell types in tumor samples. \u003cstrong\u003eA)\u003c/strong\u003e number of interactions; \u003cstrong\u003eB)\u003c/strong\u003e interaction strength. \u003cstrong\u003e(C)\u003c/strong\u003eHeatmap of outgoing signaling patterns across different cell types. \u003cstrong\u003e(D)\u003c/strong\u003eHeatmap of incoming signaling patterns across different cell types.\u003cstrong\u003e (E)\u003c/strong\u003eAnalysis of differential signaling pathways transmitted from fibroblasts to SLC2A1-positive and SLC2A1-negative epithelial cells. \u003cstrong\u003e(F)\u003c/strong\u003e Intercellular communication network of the COLLAGEN signaling pathway.\u003cstrong\u003e (G)\u003c/strong\u003eCOLLAGEN–receptor interaction pairs transmitted from fibroblasts to SLC2A1-positive and SLC2A1-negative epithelial cells. \u003cstrong\u003e(H)\u003c/strong\u003e Expression distribution of COLLAGEN-related genes across different cell types.\u003cstrong\u003e (I)\u003c/strong\u003eUMAP visualization of fibroblast subpopulations.\u003cstrong\u003e (J)\u003c/strong\u003e Network diagram showing interaction strength between fibroblast subpopulations and SLC2A1-positive/negative epithelial cells. \u003cstrong\u003e(K)\u003c/strong\u003e Heatmap of interaction strength between different fibroblast subpopulations and epithelial cells. \u003cstrong\u003e(L)\u003c/strong\u003eCommunication strength between fibroblast subpopulations and SLC2A1-positive epithelial cells. \u003cstrong\u003e(M)\u003c/strong\u003e Comparison of COLLAGEN signaling pathway communication strength among different fibroblast subpopulations.\u003c/p\u003e","description":"","filename":"figure11.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/8f03c784f3c2d63bf60ada9e.jpg"},{"id":106414676,"identity":"9a173269-90be-4e16-8154-6b75f5a0e952","added_by":"auto","created_at":"2026-04-08 10:21:49","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":33427573,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/087b414c-b023-4b60-9f77-6e4458c972f7.pdf"},{"id":106071666,"identity":"63e57e0c-2923-4d62-a548-599ef21a49ce","added_by":"auto","created_at":"2026-04-03 06:42:22","extension":"jpg","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":1313916,"visible":true,"origin":"","legend":"","description":"","filename":"figureS1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/fa7a781319fd60bc86f5029f.jpg"},{"id":106094569,"identity":"3b6b56ba-5117-4cd8-b18f-ccac2fed17db","added_by":"auto","created_at":"2026-04-03 11:42:52","extension":"jpg","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1532692,"visible":true,"origin":"","legend":"","description":"","filename":"figureS2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/7aa3c970c481b027b22fcbd8.jpg"},{"id":106071669,"identity":"a04b7dd8-a8d8-44ec-9bca-03aecd24c3a7","added_by":"auto","created_at":"2026-04-03 06:42:22","extension":"jpg","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":1455189,"visible":true,"origin":"","legend":"","description":"","filename":"figureS3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/7fab82ca7ea59b68b050d0ce.jpg"},{"id":106094401,"identity":"8a92939c-a7c1-4324-babd-0181f62ae719","added_by":"auto","created_at":"2026-04-03 11:42:26","extension":"jpg","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":1385676,"visible":true,"origin":"","legend":"","description":"","filename":"figureS4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/149ba2a322d68d6dc45f7aac.jpg"},{"id":106094908,"identity":"0a7247d0-d7f5-4cef-83c0-f58c84c4e1d0","added_by":"auto","created_at":"2026-04-03 11:43:37","extension":"jpg","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":1414063,"visible":true,"origin":"","legend":"","description":"","filename":"figureS5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/fcb6ab04de9c0605c9137b1a.jpg"},{"id":106094516,"identity":"0266c5be-f9f2-4a82-9836-d3883032a2bd","added_by":"auto","created_at":"2026-04-03 11:42:48","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":11140,"visible":true,"origin":"","legend":"","description":"","filename":"TableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/4181868a64a720ccc27e0eb3.xlsx"},{"id":106094661,"identity":"a55145a5-277a-4efb-860b-b9f96bcffe90","added_by":"auto","created_at":"2026-04-03 11:43:04","extension":"xlsx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":11109,"visible":true,"origin":"","legend":"","description":"","filename":"TableS3.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/97fdd809e1468896f05be7a2.xlsx"},{"id":106071674,"identity":"6ef0abf4-4659-4a4c-8ff2-130ac669e9b1","added_by":"auto","created_at":"2026-04-03 06:42:22","extension":"xlsx","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":10330,"visible":true,"origin":"","legend":"","description":"","filename":"TableS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/b68240c6c691605aff906f5f.xlsx"},{"id":106094485,"identity":"d71b122e-0c02-4276-8ad9-f65bda114bc1","added_by":"auto","created_at":"2026-04-03 11:42:43","extension":"pdf","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":1744357,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterials.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9298986/v1/8ebc1e0d05cbb5aa0c23057b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A clinically relevant SLC2A1-associated malignant epithelial cell state predicts prognosis and immunotherapy response in lung adenocarcinoma","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eLung cancer remains the leading cause of cancer-related death worldwide(Siegel et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Lung adenocarcinoma (LUAD), the most common histological subtype, is characterized by pronounced heterogeneity and complex molecular features(Shen et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhang et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Although systemic treatment for LUAD has expanded from conventional chemotherapy to include targeted therapy and immunotherapy, overall survival outcomes have improved only modestly(Herbst et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Wang et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). A major reason for the limited clinical benefit lies in tumor heterogeneity: chemotherapy-related toxicity, the low prevalence of actionable driver mutations, and the unpredictable response to immunotherapy all restrict the effectiveness of current treatment strategies(Bortolot et al. \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Reck et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhao et al. \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Therefore, identifying molecular markers that reflect key tumor cell states and predict therapeutic response is critical for advancing precision treatment in LUAD.\u003c/p\u003e \u003cp\u003eThe tumor immune microenvironment (TIME) plays a central role in LUAD progression, immune evasion, and treatment response(Xu et al. \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Dynamic interactions between tumor cells, immune cells, fibroblasts, and vascular-associated cells collectively shape the tumor ecosystem and its immunoregulatory landscape(Sun et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Increasing evidence indicates that the TIME is not static but undergoes continuous remodeling during tumor evolution(Keenan et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). For example, the accumulation of immunosuppressive macrophages, regulatory T cells, and cancer-associated fibroblasts (CAFs) is frequently associated with immune escape and poor clinical outcomes(Huang et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). However, which specific cellular populations drive malignant progression in LUAD and influence patient prognosis through microenvironmental interactions remains incompletely understood.\u003c/p\u003e \u003cp\u003eThe development of single-cell RNA sequencing (scRNA-seq) has enabled us to re-understand the heterogeneity of LUAD tumor tissue at the cellular level(Boxer et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). In particular, within tumor epithelial cells, different malignant states often correspond to distinct potentials for progression and interactions with the microenvironment(Andreatta et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). By leveraging scRNA-seq, not only can we identify key cell subpopulations associated with tumor progression, but we can also further dissect their immune regulatory features and cell communication networks, providing new biological foundations for prognostic assessment and treatment stratification(Tirosh and Suva \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Using the Scissor algorithm to integrate single-cell transcriptomics with clinical survival data, we have been able to pinpoint epithelial cell states directly linked to poor prognosis in LUAD.\u003c/p\u003e \u003cp\u003eHowever, despite increasing recognition of epithelial heterogeneity, a key gap remains: the malignant epithelial cell states that directly drive poor prognosis, their molecular regulators, and their relationship with immunotherapy response have not been systematically defined. In particular, how these cell states can be translated into clinically applicable biomarkers remains unclear.\u003c/p\u003e \u003cp\u003eBased on this rationale, we integrated large-scale single-cell transcriptomic data with clinical and spatial datasets to systematically characterize malignant epithelial heterogeneity in LUAD. We identified a prognostically relevant epithelial cell state and developed a corresponding risk score model (SERS), which was validated across independent cohorts and immunotherapy-treated samples. Furthermore, we identified SLC2A1 as a key regulator of this malignant state and explored its functional and microenvironmental roles. Together, this study defines a clinically relevant malignant epithelial cell state in LUAD and establishes a framework linking cell states, molecular drivers, and microenvironmental interactions, providing new insights into tumor progression and precision therapy. The overall study design and analytical workflow are illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cp\u003e\u003cstrong\u003e2.1 Data sources and preprocessing.\u003c/strong\u003e We used the public single-cell RNA-seq dataset GSE308103, which includes 16 paired normal lung and lung adenocarcinoma (LUAD) samples. Raw expression matrices and sample annotations were downloaded, and gene symbols were harmonized to HGNC official symbols with a consistent matrix format. Bulk RNA-seq expression data and clinical follow-up information for prognostic modeling were obtained from the TCGA-LUAD cohort. External validation cohorts included GSE31210, GSE50081, and GSE72094. These GEO cohorts were also merged into a Meta cohort to support robustness analyses. Immunotherapy-related cohorts included POPLAR, OAK, and GSE135222. Spatial transcriptomics data were taken from E-MTAB-13530, which contains both tumor and normal regions. All datasets used in this study are summarized in Supplementary Table S3.\u003c/p\u003e\n\u003cp\u003eUnless otherwise stated, all analyses were performed in R (v4.2.1). Single-cell analyses were conducted mainly with Seurat (v4.3.1), batch correction with harmony (v0.1.1), cell\u0026ndash;cell communication analysis with CellChat (v1.1.1), and prognosis-linked subpopulation identification with Scissor (v1.0.0). Ligand\u0026ndash;target inference was performed using nichenetr (v1.0.0). Immune infiltration and immune signature scoring were carried out using IOBR (v0.99.9) and GSVA (v1.46.0). Differential expression and enrichment analyses used clusterProfiler (v4.6.2) and fgsea (v1.24.0). Survival and model evaluation were performed with survival (v3.3-1), survminer (v0.4.9), and timeROC (v0.4). Batch-effect adjustment used sva (v3.44.0). Figures were generated mainly with ggplot2 (v3.4.4) and ComplexHeatmap (v2.14.0).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 scRNA-seq data processing.\u003c/strong\u003e Single-cell RNA-seq data were processed using a standard Seurat (v4.3.1) workflow. Expression matrices were imported for each sample, and per-cell quality metrics were calculated, including the number of detected genes (nFeature_RNA), total UMI counts (nCount_RNA), and the percentage of mitochondrial transcripts (percent.mt). Cells were retained if nFeature_RNA was between 200 and 8000, and cells with percent.mt\u0026thinsp;\u0026ge;\u0026thinsp;15% were removed. To reduce the impact of potential doublets and unusually deep sequencing, we also excluded cells with extremely high nCount_RNA within each sample (defined as values above the 99th percentile). Doublets were then identified and removed using DoubletFinder. For each sample, we built a PCA space and performed parameter sweeps to select the optimal pK. The expected doublet rate was set based on the number of cells in each sample. Cells predicted as doublets were excluded. After quality control and doublet removal, we retained 282,289 high-quality cells, including 110,015 cells from normal lung tissues and 173,374 cells from LUAD tumors.\u003c/p\u003e\n\u003cp\u003eBecause our analysis was based on multiple samples within a single dataset (GSE308103), we performed within-dataset integration to minimize sample-to-sample technical variation. Each sample was normalized (NormalizeData), highly variable genes were identified (FindVariableFeatures), and data were scaled (ScaleData) before running PCA (RunPCA). We then used Harmony to align samples, with sample/patient origin as the batch variable. Downstream analyses were performed on the integrated low-dimensional space. A KNN graph was built using the top 30 principal components (FindNeighbors), unsupervised clustering was performed with the Louvain algorithm (FindClusters, resolution\u0026thinsp;=\u0026thinsp;0.6), and UMAP was used for visualization (RunUMAP).\u003c/p\u003e\n\u003cp\u003eCell-type annotation was performed on the integrated dataset. We identified cluster markers using FindAllMarkers (Wilcoxon rank-sum test with Benjamini\u0026ndash;Hochberg correction; FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and |log2FC| \u0026gt; 0.25). Major cell types were assigned based on canonical markers: T cells (\u003cem\u003eCD3D/E, TRAC\u003c/em\u003e), B cells (\u003cem\u003eMS4A1, CD79A\u003c/em\u003e), plasma cells (\u003cem\u003eMZB1, JCHAIN\u003c/em\u003e), myeloid cells (\u003cem\u003eLYZ, CD14, SPI1\u003c/em\u003e), fibroblasts (\u003cem\u003eCOL1A1/2, DCN\u003c/em\u003e), endothelial cells (\u003cem\u003ePECAM1, VWF\u003c/em\u003e), and epithelial cells (\u003cem\u003eEPCAM, KRT7/8\u003c/em\u003e). Marker patterns were checked using dot plots, violin plots, and feature plots. Cell-type proportions were summarized at both the group level (normal vs LUAD) and the sample level and were visualized accordingly.\u003c/p\u003e\n\u003cp\u003eTo increase resolution within key compartments, we performed subset re-clustering of T cells, myeloid cells, and epithelial cells. Subsets were extracted based on the global annotations (subset function in Seurat) and reprocessed using the same workflow (NormalizeData, FindVariableFeatures, ScaleData, RunPCA, FindNeighbors, FindClusters at resolution\u0026thinsp;=\u0026thinsp;0.6, and RunUMAP). T-cell subsets were annotated using representative markers, including Tcm (\u003cem\u003eTCF7, IL7R\u003c/em\u003e), Tem (\u003cem\u003eGZMK)\u003c/em\u003e, Tfh (\u003cem\u003eCXCL13\u003c/em\u003e), Tpro (\u003cem\u003eMKI67\u003c/em\u003e), Trm (\u003cem\u003eZNF683\u003c/em\u003e), Treg (\u003cem\u003eFOXP3\u003c/em\u003e), and Tcyto (\u003cem\u003eNKG7\u003c/em\u003e). Myeloid subsets were distinguished using markers such as S\u003cem\u003e100A8, FCGR1A, C1QA\u003c/em\u003e, and \u003cem\u003eLST1\u003c/em\u003e to define monocytes, macrophages, myeloid dendritic cells, and neutrophils. Epithelial cells were further divided into subclusters (Epi_0\u0026ndash;Epi_8), which were used as input for downstream inferCNV analysis and CellChat cell\u0026ndash;cell communication analyses.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Differential expression analysis.\u003c/strong\u003e Differential expression analysis was performed in Seurat using the default Wilcoxon rank-sum test to compare gene expression between specified groups or subpopulations. Multiple testing was adjusted with the Benjamini\u0026ndash;Hochberg method. Differentially expressed genes were defined using FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and |log2FC| \u0026gt; 0.25.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4 Functional enrichment analysis.\u003c/strong\u003e Pathway analysis was based on MSigDB Hallmark gene sets. We used fgsea to run gene set enrichment analysis (GSEA) on a ranked list of all genes ordered by the differential expression statistic. Pathway significance was assessed by permutation testing and adjusted for multiple comparisons to control false positives.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 Module scoring.\u003c/strong\u003e To describe functional states in myeloid cells, we defined gene modules related to cytokine signaling, inflammatory response, M2-like macrophage features, and NF-\u0026kappa;B signaling. Module scores were calculated at the single-cell level using Seurat\u0026rsquo;s AddModuleScore. Scores were compared across myeloid subtypes and tissue origins, and were also used in downstream trajectory analyses.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.6 Trajectory inference\u003c/strong\u003e. Trajectory inference was performed with Slingshot. We used the low-dimensional embedding (PCA/UMAP) of the myeloid subset together with cluster labels to infer lineages and assign a pseudotime value to each cell. We then examined how module scores changed along pseudotime to capture dynamic functional shifts during state transitions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.7 CNV inference.\u003c/strong\u003e Copy number variation (CNV) signals in epithelial cells were inferred using inferCNV. Annotated non-epithelial cell populations (immune cells, endothelial cells, and fibroblasts) were used as the reference cell set. CNV inference and smoothing were performed on the single-cell expression matrix following the recommended inferCNV workflow, and a cell-level CNV intensity metric (CNV score) was calculated. CNV score distributions were then compared across epithelial subclusters within the epithelial subset to assess intra-epithelial CNV heterogeneity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.8 Intercellular communication analysis.\u003c/strong\u003e Intercellular communication was analyzed using CellChat. Annotated cell types/subclusters were defined as communication nodes. Cell\u0026ndash;cell signaling was inferred based on the CellChatDB.human ligand\u0026ndash;receptor database, restricting analysis to the Secreted Signaling category (all other parameters were kept at default settings). Following the standard CellChat pipeline, putative interactions were computed and filtered using computeCommunProb, filterCommunication, and computeCommunProbPathway, and the overall communication network was summarized using aggregateNet. Pathway-level interaction patterns were visualized using functions such as netVisual_heatmap. Within the epithelial subset, we further focused on collagen-related signaling pathways: communication probabilities of collagen-related ligand\u0026ndash;receptor pairs among epithelial subclusters were compared, and network centrality/role analyses were performed to characterize sender, receiver, and mediator (regulatory) features of each subcluster within the collagen signaling network.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.9 Scissor integrative analysis and definition of signature genes.\u003c/strong\u003e To identify epithelial cell states associated with clinical outcomes, we used Scissor to jointly model the single-cell epithelial expression matrix with TCGA-LUAD bulk RNA-seq profiles and survival outcomes. Scissor first computed an expression similarity matrix between each single cell and each bulk sample using Pearson correlation. It then fitted a regularized Cox regression model in a survival analysis framework to link the similarity features to clinical outcomes. Cells were classified based on whether their regression coefficients were non-zero: cells with positive coefficients were defined as Scissor⁺, cells with negative coefficients were defined as Scissor⁻, and the remaining cells were treated as background controls. We next performed differential expression analysis between Scissor⁺ and Scissor⁻ epithelial cells (Seurat/Wilcoxon test with Benjamini\u0026ndash;Hochberg correction). Pathway enrichment was assessed using GSEA/fgsea with MSigDB Hallmark gene sets to describe functional features associated with Scissor⁺ cells, including MYC targets, TNF\u0026alpha;/NF-\u0026kappa;B, EMT, G2/M checkpoint, and Hypoxia. To quantify how strongly Scissor⁺ cells were enriched across epithelial subclusters, we calculated observed/expected (Ro/e) enrichment for each subcluster in Scissor⁺ and Scissor⁻ cells(Sun et al. 2022).\u003c/p\u003e\n\u003cp\u003eTo evaluate the clinical relevance of the Scissor⁺ transcriptional signature, we built a gene signature from Scissor⁺ differential genes. We then computed signature scores at the sample level in TCGA-LUAD using ssGSEA (implemented in the GSVA package). Samples were stratified by signature score, followed by Kaplan\u0026ndash;Meier analysis and Cox regression. To infer upstream cell\u0026ndash;cell signals that may drive the Scissor⁺ program, we applied NicheNet within a predefined sender\u0026ndash;receiver setting to predict candidate ligands. Ligands were ranked by regulatory potential (aupr_corrected), and top-ranked ligands, their receptors, and putative target gene sets were summarized and visualized. Finally, we intersected Scissor⁺ differential genes with genes upregulated in TCGA-LUAD tumors versus normal tissues, yielding 39 candidate genes for downstream machine-learning\u0026ndash;based prognostic model development.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.10 Prognostic model construction and multi-cohort validation.\u003c/strong\u003e We built a large-scale machine-learning benchmarking framework based on the 39 candidate genes. To improve comparability across cohorts, expression matrices from each bulk cohort were Z-score standardized. TCGA-LUAD was used as the training set, and independent validation was performed in external cohorts (GSE31210, GSE50081, GSE72094, and a meta-cohort). For the combined cohorts, batch effects were corrected using ComBat (sva package).\u003c/p\u003e\n\u003cp\u003eWe systematically evaluated multiple survival modeling methods and their combinations (a total of 117 model settings). Model performance was mainly assessed using Harrell\u0026rsquo;s C-index, and the final model was chosen based on both predictive accuracy and cross-cohort stability. We ultimately used LASSO plus Random Survival Forest (RSF) to build a risk score model, termed SERS (Scissor⁺-derived epithelial risk score). For each patient, SERS was calculated in both the training and validation sets. Patients were split into high- and low-risk groups using the median SERS as the cutoff. Group differences were tested using Kaplan\u0026ndash;Meier survival curves and the log-rank test. The prognostic value of SERS was further assessed by Cox regression (as both a continuous and a categorical variable). Time-dependent ROC curves and AUC were evaluated using tools such as timeROC.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.11 Immune microenvironment and immunotherapy analyses.\u003c/strong\u003e To evaluate the association between SERS and the tumor immune microenvironment, we used the IOBR framework to infer immune infiltration from bulk expression profiles in TCGA-LUAD and external cohorts. Immune and stromal cell abundances were jointly estimated using multiple methods, including EPIC, MCPcounter, and xCell. We then compared the estimated abundances of different immune/stromal cell types between the high- and low-risk SERS groups. Immune activation- and immunosuppression-related features were quantified using predefined gene signature sets. Signature scores were computed at the sample level using ssGSEA (implemented in the GSVA package) to measure immune functional states. Tumor mutational burden (TMB) and tumor neoantigen burden (TNB) were included in integrative analyses. TMB was calculated from somatic mutation data as the number of nonsynonymous mutations per Mb. TNB was obtained either from cohort-provided annotations or inferred using a standard neoantigen prediction workflow, depending on data availability. We further performed joint stratification using SERS, TMB/TNB, and immunosuppression signatures, and assessed survival differences across subgroups using Kaplan\u0026ndash;Meier analysis and Cox regression. In immunotherapy cohorts (POPLAR, OAK, and GSE135222), SERS was computed for each patient using the same formula as in the training set. Survival analyses were performed for both overall survival (OS) and progression-free survival (PFS) to evaluate the association between SERS and immunotherapy outcomes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.12 Spatial transcriptomics analysis.\u003c/strong\u003e Spatial transcriptomics data (E-MTAB-13530) were processed using a standard Seurat workflow. After spot-level quality control, data were normalized and variance-stabilized using SCTransform. SERS and related gene signature scores were then calculated at the spot level (using AddModuleScore or ssGSEA). SpatialFeaturePlot was used to visualize the spatial distribution of risk-related signals across tissue sections, and scores were compared between tumor and normal regions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.13 Causal inference of key genes.\u003c/strong\u003e To identify key genes with potential causal effects, we performed Mendelian randomization (MR) analysis on the candidate genes. Instrumental variables (IVs) were selected as SNPs significantly associated with the exposure (p\u0026thinsp;\u0026lt;\u0026thinsp;5\u0026times;10⁻⁸). We further applied LD clumping to obtain independent variants (r\u0026sup2; \u0026lt; 0.005, window size 10,000 kb). The causal effect was mainly estimated using the inverse-variance weighted (IVW) method. MR-Egger and weighted median approaches were used as sensitivity analyses. Heterogeneity was assessed using Cochran\u0026rsquo;s Q test, and robustness to individual SNPs was evaluated using leave-one-out analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.14 Functional perturbation analysis.\u003c/strong\u003e At the single-cell level, we examined the expression of key genes (SLC2A1) across cell types and tested for expression differences within epithelial cells. Virtual knockout analysis was performed using scTenifoldKnk, a perturbation framework based on single-cell gene regulatory network inference. We computationally simulated knockout of the target gene and assessed changes in network structure and downstream transcriptional programs. Differentially perturbed genes were then subjected to pathway enrichment analysis using MSigDB Hallmark gene sets to evaluate potential functional impacts.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.15 Microenvironment interactions and communication analysis related to key genes.\u003c/strong\u003e To characterize tumor microenvironment interaction patterns associated with key genes, epithelial cells were grouped by SLC2A1 expression status (SLC2A1⁺ vs. SLC2A1⁻), using a threshold of expression\u0026thinsp;\u0026gt;\u0026thinsp;0. We then applied CellChat to infer ligand\u0026ndash;receptor interaction networks between each epithelial group and other cell types, based on CellChatDB. We compared the communication strength and pathway activity between SLC2A1⁺ and SLC2A1⁻ epithelial cells, with a focus on interactions with fibroblasts/CAFs. We further zoomed in on collagen-related pathways to evaluate changes at the level of specific ligand\u0026ndash;receptor pairs. To refine CAF heterogeneity, fibroblasts were re-clustered and CAF subtypes were annotated using marker genes (including POSTN⁺ CAFs). We then assessed the relative contribution of each CAF subtype to collagen-related signaling pathways at the subtype level.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.16 Cell culture\u003c/strong\u003e Human lung cancer cell lines H1299 and H1975 were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS). A549 and PC9 cells were maintained in DMEM containing 10% FBS. All cell lines were incubated at 37\u0026deg;C in a humidified atmosphere with 5% CO₂.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.17 RT\u0026ndash;qPCR\u003c/strong\u003e Total RNA was extracted from cells using TRIzol reagent (Invitrogen, USA) according to the manufacturer\u0026rsquo;s instructions. Complementary DNA (cDNA) was synthesized using the PrimeScript\u0026trade; RT reagent kit (RR037A, Takara, Japan). Quantitative real-time PCR was performed using TB Green\u0026reg; Premix Ex Taq\u0026trade; II (RR820A, Takara, Japan) on a real-time PCR system.\u003c/p\u003e\n\u003cp\u003eThe primer sequences for SLC2A1 were as follows:\u003c/p\u003e\n\u003cp\u003eSLC2A1 Forward (F): TGGCATCAACGCTGTCTTCT\u003c/p\u003e\n\u003cp\u003eSLC2A1 Reverse (R): AACAGCGACACGACAGTGAA\u003c/p\u003e\n\u003cp\u003ePrimers were synthesized by Sangon Biotech (Shanghai, China). Relative gene expression levels were calculated using the 2^\u0026minus;\u0026Delta;\u0026Delta;Ct method and normalized to GAPDH. All primers used in this study are summarized in Supplementary Table S2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.18 Western blot\u003c/strong\u003e Total protein was extracted from cells using RIPA lysis buffer supplemented with protease inhibitors. Protein concentrations were determined using a BCA protein assay kit. Equal amounts of protein were separated by SDS-PAGE and transferred onto PVDF membranes. The membranes were incubated with primary antibodies against SLC2A1 (1:1000; A6982, ABclonal) and GAPDH (1:5000; ab128915, Abcam), followed by incubation with appropriate secondary antibodies. Protein bands were visualized using enhanced chemiluminescence (ECL). All experiments were performed with at least three independent biological replicates.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.19 SLC2A1 protein expression in LUAD\u003c/strong\u003e The protein expression of SLC2A1 in LUAD tissues was evaluated using immunohistochemical (IHC) staining data obtained from the Human Protein Atlas (HPA, https://www.proteinatlas.org/). Representative images from normal lung and tumor tissues were analyzed to assess differential expression patterns.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.20 Small interfering RNA (siRNA) transfection\u003c/strong\u003e Small interfering RNAs (siRNAs) targeting SLC2A1 and corresponding negative control siRNA were synthesized by Hanheng Biotechnology (China). Cells were seeded in antibiotic-free medium and transfected with siRNAs using a lipid-based transfection reagent according to the manufacturer\u0026rsquo;s instructions. After 24\u0026ndash;48 h of transfection, cells were harvested for subsequent experiments. Knockdown efficiency was confirmed by RT\u0026ndash;qPCR and Western blot analysis. All experiments were performed in at least three independent biological replicates. The sequences of siRNAs are provided in Supplementary Table S2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.21 Cell viability assay (CCK-8)\u003c/strong\u003e Cell viability was assessed using a Cell Counting Kit-8 (CCK-8; MCE, China). Transfected cells were seeded into 96-well plates at an appropriate density. At indicated time points, CCK-8 reagent was added to each well and incubated at 37\u0026deg;C. Absorbance was measured at 450 nm using a microplate reader. Cell viability was calculated relative to the control group.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.22 Wound-healing assay\u003c/strong\u003e Cells were seeded into 6-well plates and cultured until reaching near confluence. A straight scratch was created using a sterile pipette tip. Cells were washed with phosphate-buffered saline (PBS) to remove debris and then cultured in medium containing reduced serum. Images were captured at 0 h and 24 h using an inverted microscope. The wound area was quantified using ImageJ software, and the migration rate was calculated based on wound closure.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.23 Transwell migration and invasion assays\u003c/strong\u003e Cell migration and invasion assays were performed using Transwell chambers with 8-\u0026micro;m pore membranes (Corning, USA). For invasion assays, the upper chamber was pre-coated with Matrigel. Transfected cells suspended in serum-free medium were seeded into the upper chamber, while medium containing 10% FBS was added to the lower chamber as a chemoattractant. After incubation, cells on the upper surface were removed, and cells that migrated or invaded to the lower surface were fixed, stained with crystal violet, and counted under a microscope. Quantification was performed using ImageJ software.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.24 Statistical analysis.\u003c/strong\u003e All statistical analyses were performed in R. For two-group comparisons, Student\u0026rsquo;s t-test was used for normally distributed data, and the Wilcoxon rank-sum test was used for non-normally distributed data. For comparisons among multiple groups, one-way ANOVA (parametric) or the Kruskal\u0026ndash;Wallis test (non-parametric) was applied. Correlations were assessed using Spearman\u0026rsquo;s correlation. Survival analyses were conducted using Kaplan\u0026ndash;Meier curves with the log-rank test, and Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Multiple testing was controlled using the Benjamini\u0026ndash;Hochberg method to control the false discovery rate (FDR). Unless otherwise specified, all tests were two-sided, and p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e \u003cb\u003eSingle-cell landscape of LUAD and normal lung tissues reveals altered cellular composition\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eUsing the publicly available single-cell RNA sequencing dataset GSE308103, we performed an integrated analysis of 16 paired normal lung and LUAD samples. After quality control and doublet removal, approximately 300,000 cells were retained for downstream analyses. Unsupervised clustering followed by UMAP dimensionality reduction revealed a well-defined cellular landscape (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). Based on canonical marker gene expression, seven major cell populations were identified, including T cells, B cells, plasma cells, myeloid cells, fibroblasts, endothelial cells, and epithelial cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD; Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). Visualization of normal and LUAD samples separately demonstrated differences in both cellular distribution and relative cell-type proportions between the two conditions, while each population exhibited consistent marker gene expression patterns across samples (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB\u0026ndash;D). Quantitative analysis of cell-type composition further showed that the relative abundance of multiple cell populations was reshaped in LUAD tissues compared with normal lung (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eE and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eH). Specifically, LUAD samples displayed an increased proportion of epithelial cells accompanied by a reduced proportion of endothelial cells. In addition, immune cell composition differed between the two groups, with T cells and plasma cells relatively enriched in LUAD, whereas B cells were relatively decreased (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eH and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eI). These compositional changes were consistently observed across samples, as illustrated by stacked bar plots and radar plots, despite the presence of inter-sample heterogeneity (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eF and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eG).\u003c/p\u003e \u003cp\u003eCollectively, this single-cell transcriptomic analysis delineates a comprehensive cellular landscape of LUAD and normal lung tissues and highlights a reshaping of cellular composition in the tumor context, providing a foundation for subsequent analyses of specific tumor-associated cell populations.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSingle-cell analysis reveals immunosuppressive remodeling of the immune microenvironment in LUAD\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo characterize immune alterations in LUAD at the single-cell level, we further examined T cell and myeloid populations based on the global single-cell atlas. Both cell types showed clear subpopulation structures and notable heterogeneity across normal and tumor tissues (Supplementary Figs. S1\u0026ndash;S2). Importantly, LUAD samples exhibited a consistent shift in immune composition toward regulatory and suppressive cell populations. Within the T cell population, LUAD tissues were characterized by increased proportions of regulatory T cells (Treg) and follicular helper T cells (Tfh), along with a relative expansion of central memory T cells (Tcm). In contrast, cytotoxic T cells (Tcyto) and tissue-resident memory T cells (Trm), which are key mediators of anti-tumor immunity, were reduced (Supplementary Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e; Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). This pattern suggests a weakened cytotoxic response accompanied by enhanced immune regulation. Similarly, in the myeloid population, LUAD samples showed an increased abundance of macrophages, with gene expression patterns consistent with M2-like polarization. These macrophages displayed features associated with immunosuppressive activity, including activation of inflammatory signaling pathways, metabolic reprogramming, and NF-κB\u0026ndash;related responses (Supplementary Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e). Together, these results indicate that the LUAD microenvironment is characterized by a coordinated shift toward immune suppression, providing an important context for understanding tumor cell\u0026ndash;intrinsic epithelial states in subsequent analyses.\u003c/p\u003e \u003cp\u003e \u003cb\u003eHeterogeneity of epithelial cell subpopulations and their malignant features and collagen signaling communication.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe further performed subclustering analysis of epithelial cells. UMAP dimensionality reduction showed that epithelial cells could be divided into nine transcriptionally distinct subpopulations (Epi_0\u0026ndash;Epi_8) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). The relative proportions of these subpopulations differed between normal lung and LUAD tissues (Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC), indicating a redistribution of epithelial cell composition in tumor tissues. Analysis of marker gene expression showed clear differences in the expression levels of multiple genes among epithelial cell subpopulations (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD), supporting the validity of this classification. We then applied inferCNV to assess copy number variation (CNV) features and observed pronounced spatial heterogeneity of CNV scores across epithelial cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE). CNV score distributions varied among different epithelial subpopulations (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF). Consistent with this, malignant cell proportions were relatively higher in the Epi_5, Epi_6, and Epi_7 subpopulations, whereas other subpopulations were mainly composed of non-malignant cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG), suggesting differences in malignant features among epithelial cell subpopulations.\u003c/p\u003e \u003cp\u003eTo further explore interactions among epithelial cell subpopulations, we performed cell\u0026ndash;cell communication analysis. The number and strength of interactions were not uniform across epithelial subpopulations: some subpopulations were more active as signal senders or receivers, whereas others showed lower overall interaction levels (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eH). CellChat-based ligand\u0026ndash;receptor analysis further revealed that the collagen signaling pathway represented a major component of communication among epithelial cell subpopulations, with notable differences in communication probabilities across subpopulations (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eI). Network role analysis indicated that epithelial subpopulations played distinct roles within the collagen signaling communication network, with the Epi_0 subpopulation showing a relatively more prominent role in signal sending, receiving, or regulation (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eJ). Taken together, single-cell transcriptomic analysis reveals marked heterogeneity among epithelial cell subpopulations in LUAD with respect to transcriptional features, CNV-associated malignant characteristics, and collagen signaling\u0026ndash;mediated cell\u0026ndash;cell communication, providing a basis for further investigation of key epithelial cell subpopulations and their interactions with the tumor microenvironment.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePrognostically relevant Scissor⁺ epithelial cells and their transcriptomic features.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe applied the Scissor method to identify epithelial cell states associated with clinical outcomes. Scissor⁺ cells were unevenly distributed across epithelial subpopulations, with relatively higher proportions observed in the Epi_5 and Epi_6 subclusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). UMAP visualization further showed that Scissor⁺ cells were spatially clustered and could be clearly distinguished from Scissor⁻ cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). Differential expression analysis revealed marked differences in gene expression between Scissor⁺ and Scissor⁻ epithelial cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC), indicating that Scissor⁺ cells exhibit distinct transcriptomic features. Hallmark gene set enrichment analysis showed that Scissor⁺ epithelial cells were significantly enriched in multiple pathways related to tumor progression, including MYC targets, TNFα/NF-κB signaling, EMT, hypoxia, and cell cycle\u0026ndash;related pathways (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD). Among these pathways, EMT showed strong enrichment in Scissor⁺ cells, which was further confirmed by GSEA analysis demonstrating an overall upregulation of EMT-related genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003eComparison of the distribution of Scissor⁺ and Scissor⁻ cells across epithelial subpopulations using Ro/e (observed/expected) analysis showed that the Epi_5 and Epi_6 subclusters were relatively enriched in Scissor⁺ cells, whereas the Epi_7 subcluster was relatively enriched in Scissor⁻ cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eF). These findings indicate that Scissor⁺ epithelial cells are preferentially enriched in specific subpopulations (Epi_5 and Epi_6), defining a distinct malignant epithelial cell state associated with adverse clinical outcomes. To assess the clinical impact of this malignant epithelial state, survival analysis was performed in the TCGA-LUAD cohort. Patients with high ssGSEA scores reflecting the activity of the Scissor⁺ epithelial state exhibited significantly worse overall survival compared with those with low activity (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eG), whereas Scissor⁻ state activity showed the opposite trend. Consistently, gene signatures derived from Scissor⁺-enriched subclusters (particularly Epi_5 and Epi_6) were also associated with poor prognosis, whereas non-enriched subclusters showed no significant association. NicheNet analysis further identified several candidate ligands with high regulatory potential in Scissor⁺ epithelial cells. Among these, TGFB1 was markedly upregulated and ranked among the top predicted regulators (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eH), supporting its potential role in mediating Scissor⁺-associated cell\u0026ndash;cell communication. Finally, comparison of Scissor⁺-associated genes with genes globally upregulated in LUAD identified 39 overlapping genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eI). This gene set therefore represents the core transcriptional program of the Scissor⁺ malignant epithelial cell state.\u003c/p\u003e \u003cp\u003e \u003cb\u003eDevelopment of a Scissor⁺ epithelial state\u0026ndash;derived risk score (SERS) for prognostic stratification in LUAD\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo translate the Scissor⁺ malignant epithelial cell state into a clinically measurable metric, we constructed a risk score model based on its core transcriptional program. Based on the 39 candidate genes, we implemented a systematic model selection framework and evaluated a total of 117 machine learning algorithms and their combinations across multiple cohorts, including TCGA, GSE31210, GSE50081, GSE72094, and a meta-cohort (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA). Among all evaluated models, the combination of LASSO regression and random survival forest (LASSO\u0026thinsp;+\u0026thinsp;RSF) demonstrated the most stable and best prognostic performance in terms of cross-cohort C-index. Using genes selected by LASSO (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB), we established a Scissor⁺ epithelial risk score (SERS), which represents a quantitative surrogate of the Scissor⁺ malignant epithelial state at the bulk transcriptomic level.\u003c/p\u003e \u003cp\u003eThe SERS model achieved robust prognostic performance in the TCGA cohort and maintained consistent predictive accuracy across multiple independent validation cohorts (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC), indicating strong generalizability. Patients stratified by SERS showed markedly different survival outcomes, with the high-SERS group exhibiting significantly worse overall survival across all cohorts (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD). In addition, SERS demonstrated favorable time-dependent AUCs and remained significantly associated with overall survival in Cox regression analyses, both as a categorical and continuous variable (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE\u0026ndash;G). Collectively, these results demonstrate that SERS not only provides stable prognostic stratification but also effectively captures the activity of a high-risk malignant epithelial cell state, enabling its translation into clinically relevant applications in LUAD.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSERS-defined malignant epithelial state is linked to immunosuppressive remodeling of the tumor microenvironment\u003c/b\u003e \u003c/p\u003e \u003cp\u003eGiven that the Scissor⁺ malignant epithelial state is characterized by activation of pathways such as EMT, hypoxia, and NF-κB signaling, which are known to influence immune regulation, we next investigated its relationship with the tumor immune microenvironment. Analyses using multiple immune infiltration algorithms revealed that the tumor immune landscape differed significantly between the SERS high- and low-risk groups. Compared with the SERS-Low group, the SERS-High group showed significant enrichment of immunosuppressive cell types, including M2 macrophages, monocyte-derived cells, and fibroblasts, whereas immune cell populations associated with anti-tumor responses were relatively enriched in the low-risk group (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA). These patterns were consistent across different immune estimation methods. Further analyses demonstrated that the SERS-High group exhibited higher scores for immunosuppressive features, while immune activation-related signatures were more prominent in the SERS-Low group (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB). Notably, cancer-associated fibroblasts (CAFs), estimated by the EPIC method, were significantly increased in the SERS-High group (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eC\u0026ndash;D), consistent with enhanced stromal interactions associated with Scissor⁺ epithelial cells.\u003c/p\u003e \u003cp\u003eIn addition, patients in the SERS-High group exhibited significantly higher tumor mutational burden (TMB) and tumor neoantigen burden (TNB) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE\u0026ndash;F). Despite this increased immunogenic potential, the SERS-High group maintained a strongly immunosuppressive microenvironment, suggesting that tumor-intrinsic epithelial states may override neoantigen-driven immune activation. Combined stratification analyses further demonstrated that integrating SERS with TMB, TNB, and immune suppressive features effectively distinguished patient subgroups with markedly different survival outcomes (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eG). Collectively, these findings indicate that the Scissor⁺ epithelial state is linked to immunosuppressive remodeling of the tumor microenvironment, potentially contributing to poor prognosis and reduced immunotherapy responsiveness in LUAD.\u003c/p\u003e \u003cp\u003e \u003cb\u003eThe Scissor⁺ epithelial state predicts immunotherapy outcomes and shows spatial enrichment in tumor regions\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo evaluate the clinical relevance of the Scissor⁺ malignant epithelial state in immunotherapy settings, we analyzed multiple independent cohorts treated with immune checkpoint inhibitors. SERS, as a quantitative surrogate of this epithelial state, showed a stable association with patient survival outcomes. Kaplan\u0026ndash;Meier analyses revealed that, in both the POPLAR and OAK cohorts, patients in the SERS high-risk group had significantly worse overall survival (OS) and progression-free survival (PFS) than those in the low-risk group (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB\u0026ndash;E). In the GSE135222 cohort, although the association with PFS did not reach statistical significance, a consistent trend toward poorer outcomes was observed (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA). To further investigate the spatial characteristics of this epithelial state, we analyzed spatial transcriptomic data from E-MTAB-13530. Tumor regions exhibited higher SERS scores and a markedly increased proportion of SERS-positive spatial spots compared with matched normal regions (B1 vs T1). This pattern was consistently observed across multiple samples (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eF\u0026ndash;J), indicating that the Scissor⁺ epithelial state is preferentially enriched within malignant regions. Collectively, these findings demonstrate that the Scissor⁺ epithelial state is not only associated with clinical outcomes in immunotherapy-treated patients but also exhibits a tumor-specific spatial distribution, further supporting its biological and clinical relevance in LUAD.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSLC2A1 is a key gene associated with the Scissor⁺ malignant epithelial state\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe next sought to identify key genes underlying the Scissor⁺ malignant epithelial state using Mendelian randomization (MR) analyses. The results showed a significant causal association between SLC2A1 and LUAD risk (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA). Leave-one-out sensitivity analysis indicated that this association was not driven by any single SNP (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eB), and consistent directions of effect were observed across different MR methods, including inverse variance weighted (IVW) and MR-Egger analyses (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eC\u0026ndash;D), supporting the robustness of this finding.\u003c/p\u003e \u003cp\u003eWe next examined the cell type\u0026ndash;specific expression pattern of SLC2A1 using single-cell transcriptomic data. SLC2A1 expression was predominantly enriched in epithelial cell populations (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eE\u0026ndash;F), and was significantly upregulated in tumor epithelial cells compared with normal tissues (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eG). This epithelial-specific expression pattern is consistent with its association with the Scissor⁺ malignant epithelial state. To further explore the functional relevance of SLC2A1, we performed virtual knockout analyses. Virtual deletion of SLC2A1 resulted in widespread transcriptional changes and affected multiple cancer-related pathways, including cell proliferation, metabolic reprogramming, and epithelial\u0026ndash;mesenchymal transition (EMT) (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eH\u0026ndash;K). Notably, these pathways are also enriched in the Scissor⁺ epithelial state, suggesting that SLC2A1 may contribute to the functional characteristics of this state. Collectively, these results suggest that SLC2A1 represents a key molecular component of the Scissor⁺ malignant epithelial state, linking this state to tumor progression and malignant phenotypes in LUAD.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSLC2A1 expression and genomic alterations support its role in the Scissor⁺ epithelial state\u003c/b\u003e \u003c/p\u003e \u003cp\u003eBuilding on these findings, we examined the clinical and molecular features of SLC2A1 in LUAD. In the TCGA cohort, SLC2A1 mRNA expression was significantly higher in LUAD tissues than in normal lung tissues (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eA), consistent with its enrichment in malignant epithelial cells identified in the Scissor⁺ state. Survival analyses showed that patients with high SLC2A1 expression had significantly worse overall survival (OS) and progression-free survival (PFS) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eB\u0026ndash;C), suggesting that activation of SLC2A1 is associated with high-risk epithelial states. At the protein level, immunohistochemical staining demonstrated markedly increased SLC2A1 expression in LUAD tissues compared with normal lung tissues (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eD\u0026ndash;E), further supporting its tumor-specific upregulation. From a genomic perspective, SLC2A1 copy number alterations showed a weak but significant positive correlation with its mRNA expression (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eF\u0026ndash;G), indicating that genomic amplification may contribute to its activation in malignant epithelial cells. In addition, higher SLC2A1 expression was observed in TP53-mutant samples (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eH), suggesting a potential link between genomic instability and activation of this epithelial program. Collectively, these findings further support that SLC2A1 is a key molecular feature of the Scissor⁺ malignant epithelial state, linking this state to adverse clinical outcomes and underlying genomic alterations in LUAD.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSLC2A1 knockdown suppresses proliferation, migration, and invasion in LUAD cells\u003c/b\u003e \u003c/p\u003e \u003cp\u003eGiven the strong association of SLC2A1 with adverse clinical outcomes and malignant epithelial features, we next examined its functional role in LUAD cells. SLC2A1 expression was first examined in lung adenocarcinoma cell lines and normal bronchial epithelial cells. qRT-PCR analysis showed that SLC2A1 mRNA levels were higher in A549 and H1975 cells compared with BEAS-2B cells, while relatively lower expression was observed in PC9 and H1299 cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eA). Western blot analysis showed a consistent pattern at the protein level, confirming elevated SLC2A1 expression in A549 and H1975 cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eB). Based on these results, A549 and H1975 cells were selected for subsequent functional assays.\u003c/p\u003e \u003cp\u003eEfficient knockdown of SLC2A1 was achieved using siRNA, as confirmed by qRT-PCR and Western blot analyses in both cell lines (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eC\u0026ndash;D). To evaluate the effect of SLC2A1 on cell proliferation, CCK-8 assays were performed. Silencing of SLC2A1 significantly inhibited the proliferation of A549 and H1975 cells over time compared with control cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eE), indicating that SLC2A1 promotes LUAD cell growth. The role of SLC2A1 in cell migration was assessed using wound-healing assays. Knockdown of SLC2A1 markedly reduced wound closure in both A549 and H1975 cells at 48 hours compared with the control group (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eF\u0026ndash;G), suggesting impaired migratory capacity.\u003c/p\u003e \u003cp\u003eIn addition, Transwell assays were performed to further examine cell migration and invasion. SLC2A1 silencing significantly decreased the number of migrated and invaded cells in both A549 and H1975 cell lines (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eH\u0026ndash;I). Quantitative analysis consistently confirmed a significant reduction in both migration and invasion following SLC2A1 knockdown. Taken together, these results demonstrate that SLC2A1 promotes proliferation, migration, and invasion of LUAD cells, supporting its role as a functional mediator of the Scissor⁺ malignant epithelial state.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePotential interactions between SLC2A1-positive epithelial cells and CAFs in LUAD\u003c/b\u003e \u003c/p\u003e \u003cp\u003eGiven that SLC2A1 promotes malignant phenotypes such as proliferation and migration, we next explored potential mechanisms underlying its role in the tumor microenvironment. In particular, we focused on interactions between SLC2A1-positive epithelial cells and cancer-associated fibroblasts (CAFs). Global network analysis suggested that interactions between SLC2A1-positive epithelial cells and fibroblasts were increased in both interaction number and strength, pointing to enhanced communication between these two cell types (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eA\u0026ndash;B). Further analysis of outgoing and incoming signaling showed that SLC2A1-positive epithelial cells exhibited increased reception of extracellular matrix\u0026ndash;related signals, while fibroblasts acted as the primary signal sources (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eC\u0026ndash;D). Differential signaling analysis further suggested that fibroblast-to\u0026ndash;SLC2A1-positive epithelial signaling was elevated, with collagen-related pathways showing the most prominent enrichment (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003eAt the pathway level, communication network analysis indicated strong interactions between fibroblasts and SLC2A1-positive epithelial cells within collagen signaling pathways (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eF). Ligand\u0026ndash;receptor analysis further suggested that multiple collagen ligands and their corresponding receptors were enriched in signaling from fibroblasts to SLC2A1-positive epithelial cells, whereas these interactions were less pronounced in SLC2A1-negative epithelial cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eG\u0026ndash;H). Given the heterogeneity of fibroblasts in the tumor microenvironment, CAFs were further subdivided into distinct subpopulations. UMAP visualization identified multiple CAF subtypes within tumor tissues (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eI). Communication analysis suggested that different CAF subpopulations exhibited varying interaction strengths with SLC2A1-positive epithelial cells, with the POSTN⁺ CAF subset showing the strongest contribution to collagen-mediated signaling (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003eJ\u0026ndash;M). Together, these findings suggest a potential interaction axis between SLC2A1-positive epithelial cells and CAFs, primarily mediated through collagen-related signaling pathways, which may contribute to the maintenance of the malignant epithelial state in LUAD.\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eTumor development and progression is a dynamic process shaped by interactions between diverse cell states within a complex microenvironment(Tufail et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2026\u003c/span\u003e; Wang et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). With the advent of single-cell RNA sequencing and spatial transcriptomics, increasing evidence has revealed substantial functional heterogeneity across tumor-associated cell populations(Jia et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Liu et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Compared with traditional cell-type classifications, these fine-grained cellular states provide a more accurate representation of tumor biology(Quintero et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, in lung adenocarcinoma (LUAD), previous studies have reported significant alterations in cellular composition, including expansion of malignant epithelial cells, vascular abnormalities, and reshaped immune infiltration patterns(Liu et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2026\u003c/span\u003e; Shi et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Tan et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Wang et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Most studies have focused on cell-type proportions or overall infiltration levels(Barjesteh van Waalwijk van Doorn-Khosrovani et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), while fewer have systematically characterized functional cell states, their dynamic transitions, and their roles within the tumor microenvironment(Jongbloed et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). In particular, integrative analyses linking cell states, cell\u0026ndash;cell interactions, disease progression, and clinical outcomes remain limited(Chen and Liu \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Sun et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). As a result, the key cell states that drive LUAD progression, and the mechanisms by which microenvironmental signals shape these states and influence clinical outcomes, are still not fully understood.\u003c/p\u003e \u003cp\u003eIn this study, we identified a prognostically relevant malignant epithelial cell state in LUAD by integrating single-cell and bulk transcriptomic data. Defined by the Scissor framework, this state captures a functional program associated with tumor progression rather than a conventional cell-type classification. Building on this finding, we further translated this malignant epithelial state into a clinically applicable risk model (SERS), enabling a direct link between tumor cell states and patient outcomes.\u003c/p\u003e \u003cp\u003eSingle-cell analysis further revealed a marked reshaping of the LUAD microenvironment, characterized by expansion of epithelial cells and substantial changes in immune composition. Within the T cell compartment, LUAD tissues exhibited increased proportions of Treg, Tcm, and Tfh cells, along with a reduction in cytotoxic and tissue-resident memory T cells. Rather than representing isolated compositional changes, this pattern reflects a coordinated shift toward immune suppression, which may facilitate immune escape and support the emergence of aggressive tumor cell states.\u003c/p\u003e \u003cp\u003eConsistent with these changes, myeloid cells also showed clear functional remodeling(Wang et al. \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Tumor-associated macrophages were enriched in LUAD and exhibited transcriptional features consistent with M2-like polarization, including increased metabolic activity and inflammatory signaling. Previous studies have shown that M2-like macrophages suppress immune responses and promote tumor progression through cytokine secretion and extracellular matrix remodeling(Locati et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Nasir et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Shao et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Our findings further support a dynamic transition toward an immunosuppressive state in LUAD.\u003c/p\u003e \u003cp\u003eWithin this microenvironmental context, epithelial cells displayed pronounced heterogeneity(Fiore et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Sheng et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Distinct epithelial subclusters showed differences in CNV levels and malignant features(Dear \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Ma et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2026\u003c/span\u003e; Meng et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), suggesting the coexistence of multiple functional states during tumor progression. Importantly, cell\u0026ndash;cell communication analysis indicated that epithelial cell behavior is influenced not only by intrinsic alterations but also by signals from the microenvironment. In particular, extracellular matrix\u0026ndash;related signaling, especially collagen pathways, was prominently enriched, suggesting that interactions with stromal cells may support tumor growth and invasion.\u003c/p\u003e \u003cp\u003eTo link single-cell states more directly to clinical outcomes, we used the Scissor method to integrate single-cell transcriptomes with survival data from TCGA-LUAD. This analysis identified a group of Scissor⁺ epithelial cells that was strongly associated with poor prognosis. Notably, Scissor⁺ cells were not limited to one single epithelial subcluster. Instead, they were significantly enriched in specific subclusters. This suggests that Scissor⁺ cells may represent a \u0026ldquo;functional malignant state\u0026rdquo; related to tumor progression, rather than a traditional cell-type label. Consistent with this idea, differential expression and pathway enrichment analyses showed that Scissor⁺ epithelial cells had strong activation of cell-cycle programs, the G2/M checkpoint, MYC target genes, and EMT. These pathways have been widely linked to sustained proliferation, increased invasiveness, treatment resistance, and poor clinical outcomes(Jakobsen and Siersb\u0026aelig;k \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; L\u0026ouml;brich and Jeggo \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Zhang et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Therefore, Scissor⁺ cells likely reflect more than a single molecular abnormality. They may capture a stable functional state that emerges during LUAD progression.\u003c/p\u003e \u003cp\u003eBased on the Scissor⁺ epithelial state, we developed a prognostic model (SERS) that captures a biologically defined malignant program. Unlike conventional bulk-based models, SERS directly reflects a cell state associated with tumor progression, resulting in stable survival stratification across multiple cohorts. Notably, high SERS scores were consistently associated with an immunosuppressive microenvironment, characterized by enrichment of M2-like macrophages and activated CAFs, suggesting that this cell state is closely linked to immune evasion.\u003c/p\u003e \u003cp\u003eTo further identify key regulators of this malignant state, we applied Mendelian randomization and identified SLC2A1 as a potential causal driver. Consistent with this, SLC2A1 was highly expressed in tumor epithelial cells and associated with poor prognosis. In silico perturbation further indicated its involvement in pathways related to metabolism, proliferation, and EMT, supporting its central role in maintaining malignant phenotypes.\u003c/p\u003e \u003cp\u003eImportantly, functional experiments confirmed that SLC2A1 promotes proliferation, migration, and invasion of LUAD cells, providing direct evidence that it actively drives tumor cell behavior rather than serving as a passive marker. These findings link the Scissor⁺ epithelial state to a specific molecular driver with functional relevance. Further analysis suggested a potential interaction axis between SLC2A1-positive epithelial cells and CAFs, mediated by collagen-related signaling. This suggests a potential model in which tumor-intrinsic metabolic activity and stromal extracellular matrix signals may cooperate to support tumor progression.\u003c/p\u003e \u003cp\u003ePrevious studies have extensively characterized the LUAD tumor microenvironment using single-cell transcriptomics, revealing substantial heterogeneity across cell types and functional states. In parallel, numerous prognostic models have been developed based on bulk transcriptomic data, identifying gene signatures associated with patient outcomes. In addition, key regulators such as SLC2A1 have been implicated in tumor metabolism and progression. However, these lines of research are often conducted independently, with limited integration across cellular states, molecular drivers, and clinical phenotypes. In this study, we build upon these prior efforts by integrating single-cell-defined cell states with bulk transcriptomic and clinical data. Rather than focusing solely on cell-type composition or gene-level associations, we identify a prognostically relevant malignant epithelial state (Scissor⁺) and systematically trace its impact across multiple levels. Specifically, we translate this cell state into a clinically applicable model (SERS), identify a key regulatory gene (SLC2A1) with causal and functional relevance, and further explore a potential mechanism involving stromal interactions.\u003c/p\u003e \u003cp\u003eImportantly, this study provides several advances over previous work. First, we identify a prognostically relevant malignant epithelial cell state rather than relying on conventional cell-type classifications. Second, we translate this cell state into a clinically applicable risk model (SERS) with stable performance across multiple cohorts. Third, by integrating Mendelian randomization, single-cell expression, and functional experiments, we identify SLC2A1 as a potential driver linking tumor cell states to malignant phenotypes. Unlike previous studies that focus on either cell-type composition or bulk-based signatures, our study integrates cell-state identification, causal inference, and functional validation within a single framework. Together, these findings establish a unified framework connecting cell states, molecular drivers, functional phenotypes, and microenvironmental interactions in LUAD, providing new insights into tumor progression and therapeutic response.\u003c/p\u003e \u003cp\u003eAlthough this study used multi-omics integration to reveal multi-level features of the LUAD microenvironment and key cell states, it has several limitations. First, our work is mainly based on bioinformatic analyses. While some key genes were supported at the transcript level, further validation is needed at the protein level. Functional and mechanistic experiments are also required. Second, the detailed molecular mechanism of the SLC2A1\u0026ndash;CAF\u0026ndash;collagen axis still needs to be tested in vitro and in vivo. Third, technical differences across datasets and sample heterogeneity may affect the results. Future studies should validate the clinical utility of SERS in larger, multi-center, prospective cohorts. In addition, although functional experiments support the role of SLC2A1, further studies are required to establish its direct mechanistic links with microenvironmental interactions in vivo.\u003c/p\u003e \u003cp\u003eIn this study, we systematically characterized the tumor microenvironment of LUAD at the single-cell level and identified a malignant epithelial cell state associated with poor prognosis. By integrating single-cell and bulk transcriptomic data, we further translated this high-risk cell state into a clinically applicable risk model (SERS), thereby establishing a direct link between tumor cell states and patient outcomes. Building on this framework, we identified SLC2A1 as a key gene associated with the Scissor⁺ malignant epithelial state through integrative analyses, including Mendelian randomization, single-cell expression profiling, and in silico perturbation. Importantly, functional experiments demonstrated that SLC2A1 promotes proliferation, migration, and invasion of LUAD cells, providing direct evidence that this gene is not only associated with malignant epithelial states but also actively drives tumor cell behavior.\u003c/p\u003e \u003cp\u003eFurthermore, cell\u0026ndash;cell communication analysis suggested a potential interaction axis between SLC2A1-positive epithelial cells and cancer-associated fibroblasts (CAFs), primarily mediated by collagen-related signaling. This finding points to a possible mechanism by which tumor-intrinsic metabolic programs may interact with stromal signals to support malignant phenotypes. Although further experimental validation is required, this SLC2A1\u0026ndash;CAF\u0026ndash;collagen axis provides a biologically plausible model linking tumor cell states with microenvironmental regulation. Consistent with these findings, the cell state captured by SERS was also associated with patient outcomes in immunotherapy-treated cohorts, suggesting that tumor-intrinsic epithelial programs, together with their microenvironmental interactions, may influence treatment response. Overall, our study establishes a framework that connects malignant epithelial cell states, key regulatory genes, functional tumor phenotypes, and microenvironmental interactions in LUAD. These findings not only improve our understanding of tumor progression but also provide potential targets and strategies for precision therapy.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e \u003cp\u003eAs all data were obtained from publicly available databases and no identifiable human data were used, ethical approval and informed consent were not required.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eCompeting interests\u003c/strong\u003e \u003cp\u003eThe authors declare that they have no competing financial or non-financial interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was supported by Guizhou Provincial Administration of Traditional Chinese Medicine's Traditional Chinese Medicine and Ethnic Medicine Science and Technology Research Project (No. QZYY-2023-108); Noncommunicable Chronic Diseases \u0026ndash; National Science and Technology Major Project (No. 2023ZD0502105); the National Natural Science Foundation of China (No. 82504050); Zunyi Municipal Science and Technology Cooperation Project (No. HZ [2025] 129).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eLS: Conceptualization, Data curation, Methodology, Software, Formal Analysis, Writing \u0026ndash; original draft. SX: Conceptualization, Methodology. YC: Validation, Formal Analysis, Software. PL: Formal Analysis, Methodology. HM: Data curation, Validation, Writing \u0026ndash; review \u0026amp; editing, Funding acquisition. All authors discussed and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eThe authors thank TCGA, GEO, ArrayExpress, and the Human Protein Atlas for providing access to the datasets and protein expression resources used in this study.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated and/or analyzed during the current study are publicly available in the GEO, TCGA, and ArrayExpress repositories. Accession numbers include GSE308103, GSE31210, GSE50081, GSE72094, GSE135222, and E-MTAB-13530. Additional data supporting the findings of this study are included within the article and its supplementary materials.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAndreatta M, Garnica J, Carmona SJ (2025) Identification of malignant cells in single-cell transcriptomics data. Commun Biol 8:1264\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarjesteh van Waalwijk, van Doorn-Khosrovani S, Van Kholmanskikh O, Koole S, Thomas DM, Gelderblom H (2024) Testing dilemmas in the clinic: Lessons learned from biomarker-based drug development. Cancer Cell 42:923\u0026ndash;929. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ccell.2024.05.014\u003c/span\u003e\u003cspan address=\"10.1016/j.ccell.2024.05.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBortolot M, Remon J, Bironzo P, Cortiula F, Menis J, Chan SW, van Geel R, Reguart N, Arrieta O, Mountzios G, Dingemans AC, Besse B, Hendriks LEL (2025) De-escalation strategies with targeted therapies in non-small cell lung cancer. Cancer Treat Rev 139:102995. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ctrv.2025.102995\u003c/span\u003e\u003cspan address=\"10.1016/j.ctrv.2025.102995\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoxer E, Feigin N, Tschernichovsky R, Darnell NG, Greenwald AR, Hoefflin R, Kovarsky D, Simkin D, Turgeman S, Zhang L, Tirosh I (2025) Emerging clinical applications of single-cell RNA sequencing in oncology. Nat Rev Clin Oncol 22:315\u0026ndash;326. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41571-025-01003-3\u003c/span\u003e\u003cspan address=\"10.1038/s41571-025-01003-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen T, Liu F (2025) Neoadjuvant immunotherapy in early-stage NSCLC: navigating biomarker dilemmas and special population challenges. Lung Cancer 204:108588. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.lungcan.2025.108588\u003c/span\u003e\u003cspan address=\"10.1016/j.lungcan.2025.108588\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDear PH (2009) Copy-number variation: the end of the human genome? Trends Biotechnol 27:448\u0026ndash;454. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.tibtech.2009.05.003\u003c/span\u003e\u003cspan address=\"10.1016/j.tibtech.2009.05.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFiore VF, Almagro J, Fuchs E (2025) Shaping epithelial tissues by stem cell mechanics in development and cancer. Nat Rev Mol Cell Biol 26:442\u0026ndash;455\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHerbst RS, Morgensztern D, Boshoff C (2018) The biology and management of non-small cell lung cancer. Nature 553:446\u0026ndash;454. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nature25183\u003c/span\u003e\u003cspan address=\"10.1038/nature25183\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang Z, Chen J, Zhu T, Li J, Ng HY, Zhou Y, Gu X, Xu S, Jia R (2025) Cancer-associated fibroblasts in the tumor microenvironment: heterogeneity, crosstalk mechanisms, and therapeutic implications. Mol Cancer 25:19. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12943-025-02533-1\u003c/span\u003e\u003cspan address=\"10.1186/s12943-025-02533-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJakobsen ST, Siersb\u0026aelig;k R (2025) Transcriptional regulation by MYC: an emerging new model. Oncogene 44:1\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41388-024-03174-2\u003c/span\u003e\u003cspan address=\"10.1038/s41388-024-03174-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJia H, Chen X, Zhang L, Chen M (2025) Cancer associated fibroblasts in cancer development and therapy. J Hematol Oncol 18:36\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJongbloed M, Bortolot M, Willmann J, Bartolomeo V, Novoa NM, De Ruysscher DKM, Hendriks LEL (2025) Current Controversies and Challenges in Non-Oncogene-Addicted Synchronous Oligometastatic Non-Small Cell Lung Cancer: A Review. JAMA Oncol 11:1385\u0026ndash;1392. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jamaoncol.2025.2891\u003c/span\u003e\u003cspan address=\"10.1001/jamaoncol.2025.2891\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeenan BP, Yadav M, Ansstas G, Fabrizio D, Murugesan K, Montesion M, Guha Niyogi D, Mellman I, Melero I (2025) Intratumoral heterogeneity and immunotherapy resistance: clinical implications. Ann Oncol DOI. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.annonc.2025.10.1239\u003c/span\u003e\u003cspan address=\"10.1016/j.annonc.2025.10.1239\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Y, Dai Y, Wang L (2026) Spatial omics at the forefront: emerging technologies, analytical innovations, and clinical applications. Cancer Cell 44:24\u0026ndash;49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ccell.2025.12.009\u003c/span\u003e\u003cspan address=\"10.1016/j.ccell.2025.12.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Y, Zhou C, Tang Y, Lei H, Aihemaiti A, Liu H, Zou P, Xie J, Guo X, Xia R, Han BH, Chen H, Zhu L (2026) Targeting AKR1B1 reprograms tumor-associated macrophages to enhance antitumor immunity. J Immunother Cancer 14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/jitc-2025-014043\u003c/span\u003e\u003cspan address=\"10.1136/jitc-2025-014043\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eL\u0026ouml;brich M, Jeggo PA (2007) The impact of a negligent G2/M checkpoint on genomic instability and cancer induction. Nat Rev Cancer 7:861\u0026ndash;869. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nrc2248\u003c/span\u003e\u003cspan address=\"10.1038/nrc2248\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLocati M, Curtale G, Mantovani A (2020) Diversity, Mechanisms, and Significance of Macrophage Plasticity. Annu Rev Pathol 15:123\u0026ndash;147\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa L, Xiong B, Liu M, Tan K (2026) Cellular neighborhoods in cancer. Nat Cancer. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s43018-025-01107-w\u003c/span\u003e\u003cspan address=\"10.1038/s43018-025-01107-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeng F, Li J, Xia Z, Wang Q, Sun Q, Wang S, Xu L, Yin R (2025) Persistent lineage plasticity driving lung cancer development and progression. Clin Transl Med 15:e70458\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNasir I, McGuinness C, Poh AR, Ernst M, Darcy PK, Britt KL (2023) Tumor macrophage functional heterogeneity can inform the development of novel cancer therapies. Trends Immunol 44:971\u0026ndash;985. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.it.2023.10.007\u003c/span\u003e\u003cspan address=\"10.1016/j.it.2023.10.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQuintero JC, D\u0026iacute;az NF, Rodr\u0026iacute;guez-Dorantes M, Camacho-Arroyo I (2023) Cancer Stem Cells and Androgen Receptor Signaling: Partners in Disease Progression. Int J Mol Sci 24\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReck M, Frost N, Peters S, Fox BA, Ferrara R, Savai R, Barlesi F (2025) Treatment of NSCLC after chemoimmunotherapy - are we making headway? Nat Rev Clin Oncol 22:806\u0026ndash;830. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41571-025-01061-7\u003c/span\u003e\u003cspan address=\"10.1038/s41571-025-01061-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShao N, Qiu H, Liu J, Xiao D, Zhao J, Chen C, Wan J, Guo M, Liang G, Zhao X, Xu L (2025) Targeting lipid metabolism of macrophages: A new strategy for tumor therapy. J Adv Res 68:99\u0026ndash;114\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShen Y, Chen JQ, Li XP (2025) Differences between lung adenocarcinoma and lung squamous cell carcinoma: Driver genes, therapeutic targets, and clinical efficacy. Genes Dis 12:101374\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSheng R, Yin Y, Wang X (2025) Mesothelial and immune cells interplay in the tumor microenvironment. Trends Mol Med 31:895\u0026ndash;908. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.molmed.2025.03.014\u003c/span\u003e\u003cspan address=\"10.1016/j.molmed.2025.03.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi X, Hu L, Huang Y, Zhang MF, Ying X, Yuan X, Wen N, Lu J, Zou H, Tan X, He QY, Wang F, Yang H, Zhang CZ (2025) Integrative proteomic characterization of human lung adenocarcinoma with KRAS G12 mutations reveals molecular pathogenesis. J Adv Res DOI. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jare.2025.09.014\u003c/span\u003e\u003cspan address=\"10.1016/j.jare.2025.09.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiegel RL, Kratzer TB, Giaquinto AN, Sung H, Jemal A (2025) Cancer statistics, 2025. CA Cancer J Clin 75:10\u0026ndash;45\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun D, Guan X, Moran AE, Wu LY, Qian DZ, Schedin P, Dai MS, Danilov AV, Alumkal JJ, Adey AC, Spellman PT, Xia Z (2022) Identifying phenotype-associated subpopulations by integrating bulk and single-cell sequencing data. Nat Biotechnol 40:527\u0026ndash;538\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun Q, Hong Z, Zhang C, Wang L, Han Z, Ma D (2023) Immune checkpoint therapy for solid tumours: clinical dilemmas and future trends. Signal Transduct Target Ther 8:320\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun Y, Shao C, Duan H, Wang Z, Xu S, Wang C, Xiu J, Liu J, Wang X, Yao X, Gao Y, Yan X (2025) Dynamic Evolution of the Tumor Immune Microenvironment in Malignant Tumors and Emerging Therapeutic Paradigms. MedComm (2020) 6:e70496\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTan Y, Tan W, Liang Y, Long Y, Chen S, Hu Q, Ou Y, Fu J, Chen H, Ren F, Ye J, Zhou Q, Li S, He X, Wang Q, Shen Y, Lu H, Wu D, Gao A, Chen X, Li Y (2025) Machine learning-enabled spatial multi-omics uncovers lactate-driven targets and tumor microenvironmental reprogramming in cancer. NPJ Digit Med. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41746-025-02286-7\u003c/span\u003e\u003cspan address=\"10.1038/s41746-025-02286-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTirosh I, Suva ML (2024) Cancer cell states: Lessons from ten years of single-cell RNA-sequencing of human tumors. Cancer Cell 42:1497\u0026ndash;1506. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ccell.2024.08.005\u003c/span\u003e\u003cspan address=\"10.1016/j.ccell.2024.08.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTufail M, Gong K, Ijaz B, Patel H, Lui WO, Wang X, Li J (2026) The hallmarks of oncogenic signaling: From pathways to resistance in cancer therapy. Drug Resist Updat 85:101355. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.drup.2026.101355\u003c/span\u003e\u003cspan address=\"10.1016/j.drup.2026.101355\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang H, Niu X, Jin Z, Zhang S, Fan R, Xiao H, Hu SS (2025) Immunotherapy resistance in non-small cell lung cancer: from mechanisms to therapeutic opportunities. J Exp Clin Cancer Res 44:250\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S, Hu D, Wang R, Huang J, Wang B (2025) Integrative multi-omics and machine learning reveal critical functions of proliferating cells in prognosis and personalized treatment of lung adenocarcinoma. NPJ Precis Oncol 9:243\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Luo X, Xiao R, Liu X, Zhou F, Jiang D, Bai J, Cui M, You L, Zhao Y (2026) Targeting metabolic-epigenetic-immune axis in cancer: molecular mechanisms and therapeutic implications. Signal Transduct Target Ther 11:28\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Xing Y, Li B, Li X, Liu B, Wang Y (2022) Molecular pathways, resistance mechanisms and targeted interventions in non-small-cell lung cancer. Mol Biomed 3:42\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu L, Zou C, Zhang S, Chu TSM, Zhang Y, Chen W, Zhao C, Yang L, Xu Z, Dong S, Yu H, Li B, Guan X, Hou Y, Kong FM (2022) Reshaping the systemic tumor immune environment (STIE) and tumor immune microenvironment (TIME) to enhance immunotherapy efficacy in solid tumors. J Hematol Oncol 15:87\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang CX, Huang RY, Sheng G, Thiery JP (2025) Epithelial-mesenchymal transition. Cell 188:5436\u0026ndash;5486. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cell.2025.08.033\u003c/span\u003e\u003cspan address=\"10.1016/j.cell.2025.08.033\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Y, Chen M, Fang X, Han Y, Li Y (2025) Progression and Metastasis of Lung Cancer: Clinical Features, Molecular Mechanisms, and Clinical Managements. MedComm (2020) 6:e70477\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao J, Xu W, Zhou F, Zhang X, Zhou M, Miao D, Yu L, Zhang Y, Fan J, Zhou C, Li W, Mok T, Le X, Li M, Xia Y (2026) Navigating the landscape of EGFR TKI resistance in EGFR-mutant NSCLC - mechanisms and evolving treatment approaches. Nat Rev Clin Oncol 23:63\u0026ndash;83. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41571-025-01085-z\u003c/span\u003e\u003cspan address=\"10.1038/s41571-025-01085-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"functional-and-integrative-genomics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"fige","sideBox":"Learn more about [Functional \u0026 Integrative Genomics](http://link.springer.com/journal/10142)","snPcode":"10142","submissionUrl":"https://submission.nature.com/new-submission/10142/3","title":"Functional \u0026 Integrative Genomics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Lung adenocarcinoma, Single-cell RNA sequencing, Malignant epithelial cell states, Prognostic risk model, Tumor microenvironment, SLC2A1","lastPublishedDoi":"10.21203/rs.3.rs-9298986/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9298986/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eLung adenocarcinoma (LUAD) remains a leading cause of cancer-related mortality, with therapeutic resistance largely driven by unresolved malignant epithelial heterogeneity within the tumor microenvironment. However, the epithelial cell states that underlie poor prognosis and immunotherapy resistance remain incompletely defined.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe performed an integrative multi-omics analysis combining large-scale single-cell RNA sequencing, spatial transcriptomics, and bulk transcriptomic data with clinical outcomes. The Scissor algorithm was applied to identify prognosis-associated epithelial cell states, followed by construction of a risk score model. External validation was conducted across multiple independent cohorts, including immunotherapy-treated datasets.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eWe identified a prognostically relevant Scissor⁺ malignant epithelial cell state associated with adverse survival. This state was characterized by activation of MYC, epithelial\u0026ndash;mesenchymal transition, hypoxia, and NF-κB signaling, and was linked to an immunosuppressive tumor microenvironment. Based on this state, we developed a Scissor⁺ epithelial cell\u0026ndash;derived risk score (SERS), which demonstrated robust and reproducible prognostic performance across multiple cohorts and was associated with reduced responsiveness to immunotherapy. Further analyses identified SLC2A1 as a key gene associated with this malignant epithelial state. Functional experiments confirmed that SLC2A1 promotes tumor cell proliferation, migration, and invasion. In addition, cell\u0026ndash;cell communication analysis suggested a potential SLC2A1\u0026ndash;CAF\u0026ndash;collagen signaling axis linking epithelial cell states with stromal interactions.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThis study defines a clinically relevant malignant epithelial cell state in LUAD and establishes a framework linking cell states, molecular features, and microenvironmental interactions. These findings provide potential biomarkers for prognostic stratification and immunotherapy response prediction in LUAD.\u003c/p\u003e","manuscriptTitle":"A clinically relevant SLC2A1-associated malignant epithelial cell state predicts prognosis and immunotherapy response in lung adenocarcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-03 06:42:14","doi":"10.21203/rs.3.rs-9298986/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-08T05:51:59+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-07T13:19:46+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-06T14:09:10+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"131172114248321111694931928942276559761","date":"2026-05-06T12:18:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"124700093649292799610510197852686924841","date":"2026-05-06T10:47:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"69032031936438090189183518134854553297","date":"2026-05-06T08:06:12+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-05-06T07:48:14+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-07T05:33:23+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-07T05:33:14+00:00","index":"","fulltext":""},{"type":"submitted","content":"Functional \u0026 Integrative Genomics","date":"2026-04-02T06:32:56+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"functional-and-integrative-genomics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"fige","sideBox":"Learn more about [Functional \u0026 Integrative Genomics](http://link.springer.com/journal/10142)","snPcode":"10142","submissionUrl":"https://submission.nature.com/new-submission/10142/3","title":"Functional \u0026 Integrative Genomics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"d6fc70da-46c3-4b0c-b38c-df8f8da4fcb4","owner":[],"postedDate":"April 3rd, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-08T05:51:59+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-07T13:19:46+00:00","index":27,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-06T14:09:10+00:00","index":26,"fulltext":""},{"type":"reviewerAgreed","content":"131172114248321111694931928942276559761","date":"2026-05-06T12:18:50+00:00","index":25,"fulltext":""},{"type":"reviewerAgreed","content":"124700093649292799610510197852686924841","date":"2026-05-06T10:47:11+00:00","index":23,"fulltext":""},{"type":"reviewerAgreed","content":"69032031936438090189183518134854553297","date":"2026-05-06T08:06:12+00:00","index":21,"fulltext":""},{"type":"reviewersInvited","content":"9","date":"2026-05-06T07:48:14+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-18T14:08:17+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-03 06:42:14","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9298986","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9298986","identity":"rs-9298986","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}