Transcriptomic Profiling Reveals Molecular Differences Across Small Cell Lung Cancer Stages

A comprehensive transcriptomic investigation of 361 patients with histologically confirmed small cell lung cancer has demonstrated that tumors categorized by conventional clinical staging exhibit pronounced molecular and immunological divergence. Small cell lung cancer represents an exceptionally aggressive neuroendocrine malignancy that accounts for approximately 10 to 15 percent of all diagnosed lung cancers, marked pathologically by rapid cellular proliferation, widespread early metastatic dissemination, frequent post-treatment recurrence, and an overall five-year survival rate falling below 7 percent. Clinically, the disease has historically been partitioned into limited-stage and extensive-stage classifications based solely on the anatomical extent of tumor spread and the technical feasibility of administering definitive thoracic radiotherapy. Although standard management traditionally assigns concurrent chemoradiotherapy to limited-stage presentations and systemic platinum-based chemotherapy to extensive-stage disease, clinical outcomes and therapeutic responses display marked heterogeneity both across and within these two broad stages. In this large-scale profiling effort, researchers revealed that limited-stage tumors possess an active, immune-enriched microenvironment dominated by B-cell signaling programs, whereas extensive-stage tumors are characterized by profound metabolic reprogramming featuring elevated oxidative phosphorylation and upregulated MYC target pathways.

The patient cohort was prospectively enrolled at Samsung Medical Center between 2003 and 2023, comprising 100 individuals clinically staged with limited-stage small cell lung cancer and 261 individuals staged with extensive-stage small cell lung cancer. All included participants possessed histologically confirmed disease alongside high-quality bulk RNA-sequencing data derived from tumor specimens obtained concordantly during their defined initial clinical stage. Baseline clinical evaluations revealed a mean patient age of 65.1 years within the limited-stage cohort and 66.9 years within the extensive-stage cohort. Cigarette smoking histories were highly comparable between the two diagnostic classifications, with current or former smokers constituting 94 percent of the limited-stage group and 92 percent of the extensive-stage group, while never-smokers represented 6 percent and 8 percent of the respective stage cohorts. Clinically documented median overall survival reached 24.5 months among patients presenting with limited-stage disease compared with 13.6 months among individuals with extensive-stage disease. Biopsy sites across the extensive-stage group were anatomically diverse and frequently encompassed distant metastatic lesions, yet computational assessment confirmed that anatomical biopsy location accounted for only a negligible fraction of transcriptomic variance and did not separate specimens during unsupervised principal component analyses.

To ensure robust transcriptomic characterization from formalin-fixed paraffin-embedded tissue blocks, tumor areas were microscopically dissected following hematoxylin and eosin staining. Total cellular RNA was extracted at enrollment utilizing either the QIAGEN miRNeasy FFPE Kit or the Promega RNA Mini Kit in strict accordance with manufacturer protocols. Extracted nucleic acid concentrations were quantified using the Quant-iT RiboGreen RNA Assay on a PerkinElmer Victor Nivo Multimode Microplate Reader, while structural integrity was evaluated via Agilent 2100 Bioanalyzer or Agilent 4200 TapeStation instruments. Quality assurance protocols mandated that only specimens satisfying strict thresholds—specifically possessing a minimum total RNA yield of 0.200 micrograms and a DV200 metric equal to or exceeding 50 percent of fragments longer than 200 nucleotides—proceeded to sequencing library preparation using the Illumina TruSeq RNA Access Library Prep Kit. Sequencing was executed across Illumina HiSeq 2500, NextSeq 2000, or NovaSeq 6000 platforms. Raw sequencing reads were mapped to the GRCh38 human reference genome via the STAR aligner, with quantification performed using RSEM. Stringent bioinformatic filtration required total read counts exceeding 20 million, an exonic mapped read ratio above 0.5 via RSeQC, and the exclusion of aberrant profiles using the RNAseqQC package.

Analytical handling of expression data incorporated ComBat-seq to correct for batch effects across distinct sequencing batches, followed by variance-stabilizing transformations for downstream modular and clustering workflows. Differential gene expression analysis conducted with DESeq2 identified prominent transcriptomic divergence between limited-stage and extensive-stage tumors. Limited-stage small cell lung cancer exhibited significant upregulation of key immune-related and B-cell-associated transcripts, specifically highlighting MS4A1, GCH1, CD38, BANK1, CD79A, LIME1, and IFI30. Subsequent gene set enrichment analysis across Hallmark and Gene Ontology Biological Process databases established that interferon-gamma response cascades, B-cell-receptor signaling pathways, and positive regulation of B-cell proliferation were significantly enriched in limited-stage lesions. In stark contrast, genes upregulated in extensive-stage malignancies demonstrated no significant enrichment across Hallmark collections and instead exhibited enrichment restricted to an innate-associated inflammatory program within Gene Ontology biological process terms. To confirm these microenvironmental observations, computational deconvolution was performed using CIBERSORTx referenced against an independent human small cell lung cancer single-cell RNA-sequencing dataset, confirming a markedly higher quantitative abundance of tumor-infiltrating B cells within limited-stage tumors compared with extensive-stage cases.

Weighted gene co-expression network analysis using the WGCNA package further dissected co-expression architectures, applying a soft thresholding power of 14 for signed adjacency matrices and a dynamic tree cut height of 0.25 to resolve 14 distinct co-expression modules. Two modules were found to be significantly upregulated in extensive-stage tumors relative to limited-stage disease, corresponding functionally to MYC target genes and the oxidative phosphorylation respiratory chain pathway. Because underlying clinical variables could potentially confound these stage-associated expression differences, researchers conducted analyses of covariance adjusting for patient age, biological sex, and baseline Eastern Cooperative Oncology Group performance status. Following multivariable adjustment, both the core differentially expressed genes and the stage-specific upregulation of metabolic modules retained robust statistical significance, with the oxidative phosphorylation module yielding an adjusted P-value of 0.00045 and the MYC-target module yielding an adjusted P-value of 0.01999. These findings indicated that extensive-stage disease is transcriptionally driven by metabolic pathways previously implicated in immune evasion, whereas limited-stage disease maintains an immunologically active microenvironment.

Beyond binary anatomical staging, the researchers explored unappreciated biological heterogeneity by conducting unsupervised non-negative matrix factorization clustering using the nmf package and the Brunet factorization algorithm. Based on stability evaluations of the cophenetic correlation coefficient across cluster ranks from two to eight, an optimal rank of four distinct clusters was selected. Integration of curated pathway expression, transcriptomic profiles, and immunohistochemical H-scores of lineage transcription factors allowed the designation of four molecular subtypes: an immune-infiltrated neuroendocrine subtype termed NMF1-INE, a canonical neuroendocrine subtype designated NMF2-NE, a non-neuroendocrine subtype displaying an oxidative phosphorylation signature termed NMF3-OXPHOS, and a non-neuroendocrine subtype enriched for MYC targets termed NMF4-MYC. Immunohistochemical subtyping was successfully executed across 341 of the 361 patients using automated Leica Bond III staining for ASCL1, NEUROD1, and POU2F3, with 20 cases excluded due to tissue limitations. A single specimen displaying a POU2F3 H-score of 160 below the formal 200 threshold was manually designated as SCLC-P based on morphological staining patterns.

Characterization of these non-negative matrix factorization subtypes illustrated that NMF1-INE displayed prominent expression of interferon-gamma and B-cell infiltration signatures alongside elevated antigen-presenting machinery scores, indicating a highly immunogenic state. Lineage marker distributions demonstrated that NMF1-INE and NMF2-NE were enriched for NEUROD1 expression, whereas NMF2-NE and NMF4-MYC exhibited high ASCL1 expression characteristic of classic neuroendocrine differentiation. Conversely, NMF3-OXPHOS exhibited markedly elevated POU2F3 H-scores indicative of a non-neuroendocrine lineage, mirroring standard immunohistochemical categorizations. Both NMF3-OXPHOS and NMF4-MYC showed substantial enrichment for the oxidative phosphorylation and MYC target gene modules characteristic of advanced disease. CIBERSORTx immune deconvolution confirmed that NMF1-INE harbored significantly higher B-cell proportions than the other clusters. Crucially, individual patients across both limited-stage and extensive-stage clinical classifications were distributed throughout all four non-negative matrix factorization clusters, demonstrating that biologically aggressive transcriptional programs can emerge within clinically early presentations, bypassing standard anatomical stage designations.

To model disease biology along a continuous trajectory rather than discrete categories, the investigators established a computational pseudostage rank. Utilizing a Seurat analytical framework, 2,000 highly variable features were subjected to principal component analysis, after which the top 10 principal components served as input for the guided_tscan function in the TSCAN package to order individual tumors along an LS-to-ES transcriptomic spectrum. The resulting pseudostage ranking demonstrated overall alignment with actual clinical stages when examining sample distributions across deciles. Along this continuous trajectory, limited-stage signatures such as B-cell and interferon-gamma pathways exhibited progressive declines, whereas extensive-stage signatures including MYC targets and oxidative phosphorylation displayed marked increases. Pearson correlation analyses confirmed statistically significant associations between pseudostage rank and both metabolic modules. Furthermore, comparisons between the lowest decile and highest decile showed significant divergences in both immune and metabolic gene sets. Decile distributions also confirmed that the immunogenic NMF1-INE subtype was concentrated at lower ranks, whereas subtypes enriched for MYC and oxidative phosphorylation programs predominated across upper ranks.

The methodological validity of the pseudostage trajectory was rigorously interrogated using stage-stratified assessments and repeated train-test split validations. When evaluated independently within the extensive-stage cohort, pseudostage rank retained highly significant correlations with oxidative phosphorylation, MYC targets, B-cell, and interferon-gamma signatures. Within the limited-stage cohort, these trajectories were less pronounced, with only the MYC target module maintaining statistical significance. To ensure that these transcriptomic patterns were not an artifact of deriving rankings and signatures within the same population, the full cohort was repeatedly split into 80 percent training and 20 percent testing subsets stratified by stage across 10 iterations. Projecting held-out test specimens onto the training-derived trajectory demonstrated that extensive-stage cases consistently occupied upper ranks. In the test set, comparisons between the lowest and highest deciles confirmed significant differences for the oxidative phosphorylation module, the MYC target module, and the B-cell-receptor signature, although the interferon-gamma response signature did not reach significance across extreme deciles in the validation split.

Clinical survival analyses established that pseudostage ranking provides decisive prognostic discrimination among patients receiving modern systemic regimens. When evaluating 182 patients with extensive-stage small cell lung cancer categorized into rank tertiles—excluding the intermediate tertile and untreated individuals—patients in the highest tertile treated with chemoimmunotherapy experienced significantly inferior overall survival compared with those in the lowest tertile, exhibiting median survival times of 11.4 months versus 17.2 months, respectively, with a hazard ratio of 0.49 and a P-value of 0.002. Conversely, among patients treated exclusively with platinum chemotherapy without checkpoint inhibitors, overall survival between high and low rank tertiles showed no statistically significant difference, recording median survival times of 13.2 months versus 15.0 months, a hazard ratio of 0.68, and a P-value of 0.1132. In a multivariable Cox proportional hazards model adjusting for baseline brain metastasis, age, sex, performance status, and therapeutic regimen, the low pseudostage rank tertile remained independently associated with improved overall survival relative to the high tertile, demonstrating an adjusted hazard ratio of 0.61 and a P-value of 0.002, while brain metastasis exhibited a hazard ratio of 0.90 with a P-value of 0.48. Study authors noted important limitations, including the necessity of external multicenter cohort validation, the absence of high-resolution spatial profiling or proteomic confirmation, and potential residual treatment confounding resulting from the temporal evolution of clinical management over the multi-decade enrollment period.

Subject of Research: Transcriptomic and immune heterogeneity across clinical stages of small cell lung cancer

Article Title: Molecular and immune heterogeneity across limited-stage and extensive-stage small cell lung cancer

Article References: Kim, J., Lee, K. J., Kim, H., Oh, E., Lee, N., Kim, S., Jeong, H.-O., Kim, J., Park, S., Jung, H. A., Sun, J.-M., Ahn, J. S., Ahn, M.-J., Lee, S., & Lee, S.-H. (2026). Molecular and immune heterogeneity across limited-stage and extensive-stage small cell lung cancer. Experimental & Molecular Medicine. https://doi.org/10.1038/s12276-026-01878-x

Image Credits: AI Generated

DOI: 10.1038/s12276-026-01878-x

Keywords: small cell lung cancer, transcriptomics, tumor microenvironment, chemoimmunotherapy, pseudostage rank, oxidative phosphorylation, MYC targets, molecular subtyping, Molecular, immune, heterogeneity, across