A high-grade neuroendocrine tumor. Pulmonary large cell neuroendocrine carcinoma (LCNEC) is a rare and aggressive form of lung cancer classified within the family of high-grade neuroendocrine tumors, alongside small cell lung cancer (SCLC). It accounts for only about 1-3% of all lung cancers but carries a particularly poor prognosis, with five-year survival rates below 15%.
Diagnostic challenges. LCNEC is defined by a combination of large cell morphology (cells that look distinctly different from normal tissue), neuroendocrine features visible under the microscope, and expression of neuroendocrine markers. Diagnosing it reliably requires adequate tissue samples and experienced pathologists, making it one of the most diagnostically challenging lung cancer subtypes.
Treatment uncertainty. Because LCNEC is rare and shares features with both NSCLC and SCLC, it has historically been treated with regimens designed for either type - without a clear evidence base. Whether NSCLC-type regimens (such as platinum-pemetrexed) or SCLC-type regimens (such as platinum-etoposide) are more effective has remained a matter of ongoing debate.
The molecular heterogeneity problem. Several studies have suggested that LCNEC is not one disease but a heterogeneous mix of tumors with different molecular profiles. Understanding this molecular diversity - and its implications for treatment - is the central challenge this study set out to address using the largest integrated dataset assembled to date for this cancer type.
The largest LCNEC dataset assembled. This study combined two independent patient cohorts totaling 590 patients with LCNEC: a discovery cohort of 241 patients from multiple French institutions and a validation cohort of 349 patients from The Cancer Genome Atlas (TCGA). This represents the largest integrated molecular and clinical analysis of LCNEC conducted to date.
Comprehensive molecular profiling. All tumors underwent extensive molecular characterization including DNA sequencing to identify genomic mutations, RNA sequencing to measure gene expression patterns, and immunohistochemistry to assess protein expression of key markers. This multiplatform approach allowed the researchers to define molecular subtypes through multiple independent lenses.
Clinical data integration. In addition to molecular data, the researchers collected detailed treatment information - including which chemotherapy or immunotherapy regimens patients received - and survival outcomes. This allowed direct linkage between molecular subtype, treatment type, and patient outcomes.
Computational and machine learning methods. To handle the complexity of integrating genomic, transcriptomic, and clinical data across two cohorts, the team applied advanced computational tools including unsupervised clustering for subtype discovery, support vector machine (SVM) classification for molecular subtyping, and survival analysis to compare outcomes across groups.
NSCLC-like and SCLC-like LCNECs. Genomic analysis revealed that LCNEC tumors cluster into two main molecular subtypes. NSCLC-like LCNECs are defined by mutations in genes commonly altered in non-small cell lung cancer, particularly KEAP1, KRAS, and STK11. SCLC-like LCNECs carry the hallmark mutations of small cell lung cancer - inactivation of both RB1 and TP53 tumor suppressor genes.
Distinct biological programs. Beyond their mutational profiles, the two subtypes have fundamentally different gene expression patterns. NSCLC-like LCNECs show activation of pathways involved in metabolic stress responses and cell survival. SCLC-like LCNECs are characterized by neuroendocrine differentiation programs and expression of markers associated with the neural lineage, including DLL3 and ASCL1.
A third group: truly unclassified LCNECs. Approximately 20% of LCNECs did not fit cleanly into either the NSCLC-like or SCLC-like category based on their mutational profiles. These tumors were analyzed separately and were found to have mixed features - the SVM machine learning classifier was developed specifically to address this ambiguous group.
Transcriptomic subtypes provide additional resolution. Gene expression analysis identified additional layers of heterogeneity, with ASCL1-high and YAP1-high transcriptomic subtypes partially overlapping with but not identical to the genomic subtypes. This suggests that LCNEC biology is even more complex than its genomic alterations alone reveal.
The challenge of unclassified LCNECs. One in five LCNEC tumors did not carry the defining mutations of either the NSCLC-like or SCLC-like subtype. These tumors created a classification gap that limited the clinical utility of molecular subtyping - if a significant fraction of tumors cannot be classified, the approach cannot guide treatment decisions for all patients.
SVM classifier developed from gene expression data. To address this, the researchers trained a support vector machine (SVM) classifier using gene expression profiles from clearly classified NSCLC-like and SCLC-like tumors. The SVM learned to distinguish the two subtypes based on their transcriptomic signatures and could then be applied to unclassified tumors.
Near-perfect classification accuracy. In internal validation, the SVM classifier achieved an area under the curve (AUC) of 0.98 - indicating exceptional discriminatory power. The classifier successfully assigned the vast majority of previously unclassified LCNECs to one of the two molecular subtypes based on their gene expression patterns.
Practical implications. Because RNA sequencing is increasingly available in clinical practice, this SVM-based approach offers a practical pathway to molecularly classify all LCNEC patients - not just the approximately 80% with clear mutational profiles. This could make subtype-guided treatment selection feasible for essentially the entire LCNEC patient population.
No significant survival difference across treatment types overall. When the entire cohort was analyzed together, there was no statistically significant difference in overall survival between patients treated with NSCLC-type chemotherapy (platinum-pemetrexed), SCLC-type chemotherapy (platinum-etoposide), or chemoimmunotherapy regimens. This finding challenges the assumption that LCNEC patients uniformly benefit from one approach over another.
Molecular subtype influences treatment benefit. When analyzed by molecular subtype, NSCLC-like LCNECs showed a trend toward better outcomes with NSCLC-type regimens, while SCLC-like LCNECs showed a trend toward better outcomes with SCLC-type regimens. While these subgroup analyses did not always reach conventional statistical significance due to smaller sample sizes, the directional finding supports molecularly guided treatment selection.
Immunotherapy response signals. Patients who received chemoimmunotherapy showed outcomes that varied by molecular subtype - suggesting that immune checkpoint inhibitors may benefit LCNEC patients differentially based on their underlying molecular biology. NSCLC-like LCNECs appeared more likely to respond, consistent with their higher tumor mutational burden in some analyses.
Implications for clinical trial design. The absence of a clear overall treatment benefit from any single approach underscores why LCNEC has been so difficult to study. Future clinical trials must stratify patients by molecular subtype to detect real treatment differences that are diluted when all LCNECs are grouped together.
FGL-1: a novel immune checkpoint in NSCLC-like LCNECs. Fibrinogen-like protein 1 (FGL-1) was found to be significantly overexpressed specifically in NSCLC-like LCNECs compared to SCLC-like tumors and other lung cancers. FGL-1 is a ligand for the LAG-3 immune checkpoint receptor and suppresses T-cell activity, providing a potential mechanism for immune evasion in this subtype.
LAG-3 inhibition as a therapeutic strategy. The overexpression of FGL-1 in NSCLC-like LCNECs points directly to LAG-3 inhibition as a potential therapeutic approach. LAG-3 inhibitors are currently in clinical development for multiple cancer types, and this finding suggests that NSCLC-like LCNEC patients may specifically benefit from trials combining LAG-3 inhibitors with existing immunotherapy agents.
SPINK1: a growth factor vulnerability. Serine protease inhibitor Kazal type 1 (SPINK1) was also found to be overexpressed in NSCLC-like LCNECs. SPINK1 promotes cancer cell growth by activating the EGFR signaling pathway, and its overexpression has been associated with resistance to EGFR-targeted therapies in other cancer types - making it both a potential therapeutic target and a resistance biomarker.
DLL3 elevation in SCLC-like LCNECs. Delta-like ligand 3 (DLL3) was significantly elevated specifically in SCLC-like LCNECs. DLL3 is a surface protein with limited expression in normal tissues and is the target of the antibody-drug conjugate rovalpituzumab tesirine, which has been studied in SCLC. Its selective elevation in SCLC-like LCNECs suggests this subtype may specifically benefit from DLL3-directed therapies.
Markedly reduced immune infiltration. Analysis of tumor-infiltrating lymphocytes (TILs) revealed that LCNECs have significantly lower immune cell infiltration compared to both NSCLC and SCLC. This characterizes LCNEC as a predominantly 'immune-cold' tumor type - meaning the immune system is largely excluded from the tumor microenvironment in most cases.
Implications for immunotherapy response. The immune-cold phenotype of LCNEC helps explain why immunotherapy has shown modest and variable results in this tumor type. Effective immune checkpoint inhibition generally requires pre-existing immune activity within tumors; cold tumors that lack this foundation are inherently less likely to respond to checkpoint inhibition alone.
Differential TIL levels between subtypes. SCLC-like LCNECs showed particularly low TIL levels, even lower than NSCLC-like LCNECs. This difference may explain why SCLC-like LCNECs are especially unlikely to benefit from standard immunotherapy approaches, while NSCLC-like LCNECs with their relatively higher immune infiltration may be more responsive.
Strategies to convert cold to hot tumors. The finding that LCNEC is immune-cold raises interest in combinations that can increase immune infiltration - such as combining immunotherapy with chemotherapy, radiation, or targeted agents that promote an inflammatory tumor microenvironment. Such combinations may be required to achieve meaningful immunotherapy benefit in LCNEC.
A framework for precision LCNEC treatment. The integrated findings from this study support a new clinical framework: LCNEC should not be treated as a single disease. Instead, molecular subtyping using genomic or transcriptomic profiling should guide treatment selection, with NSCLC-like tumors directed toward NSCLC-type regimens and targeted immunotherapy approaches, and SCLC-like tumors directed toward SCLC-type chemotherapy and DLL3-directed therapies.
Specific clinical trial recommendations. Based on the identified therapeutic vulnerabilities, the authors advocate for enriching clinical trials with LAG-3 inhibitors specifically in NSCLC-like LCNECs (given FGL-1 overexpression), and DLL3-targeted antibody-drug conjugates specifically in SCLC-like LCNECs (given DLL3 elevation). These biomarker-driven trial designs could finally demonstrate meaningful treatment benefits in this difficult cancer.
The SVM classifier as a practical tool. The near-perfect SVM classifier developed in this study could be implemented in clinical practice to molecularly subtype all LCNEC patients from RNA sequencing data. This would make precision treatment selection feasible for the entire patient population, not just those with clear mutational patterns - addressing a major practical barrier to implementing molecular subtype-guided therapy.
Limitations and future work needed. The retrospective nature of this study, combined with the lack of prospective randomized trials comparing treatment by molecular subtype, means that the clinical recommendations remain hypothesis-generating rather than practice-changing. Prospective clinical trials stratifying LCNEC patients by molecular subtype and testing subtype-specific therapies are urgently needed to validate this framework and potentially transform outcomes for this underserved patient population.