Deep molecular profiling of lung neuroendocrine tumours and supra-carcinoids.

BACKGROUND: Lung neuroendocrine tumours (NETs, also known as carcinoids) are rapidly rising in incidence worldwide but have unknown aetiology and limited therapeutic options beyond surgery. The current WHO classification, based on mitotic count and presence or absence of necrosis, divides lung NETs into grade-1 typical, and grade-2 atypical tumours. This dichotomous classification however does not account for recently described molecular entities nor is it sufficient for clinical management. METHODS: Here we conducted integrative multi-omic analyses on over 300 lung NETs including whole-genome sequencing, transcriptome profiling, and DNA methylation arrays, followed by archetype analysis, to identify and characterise molecular groups. We further investigated molecular groups using spatial RNA sequencing and proteomics, and deep learning analysis of whole slide images. RESULTS: The integration of multi-omic data provided definitive proof of the existence of four strikingly different molecular groups that vary in patient characteristics, genomic and transcriptomic profiles, microenvironment, and morphology. Among these, we identified a new molecular group, enriched for highly aggressive supra-carcinoids that displayed an immune-rich microenvironment linked to tumour-macrophage crosstalk. We uncovered an undifferentiated cell population within supra-carcinoids and show the transcriptomic similarities between supra-carcinoids and the recently identified atypical small cell lung cancer tumours, further demonstrating their molecular link to high-grade lung neuroendocrine carcinomas. Multi-regional genomic analyses identified distinct evolutionary trajectories, suggesting that molecular groups are determined early in tumourigenesis by genomic events, and that transitions between groups, though infrequent, are possible for supra-carcinoids. Deep learning models accurately identified these groups based on morphology alone, outperforming current histological criteria. Together with the validation of a panel of immunohistochemistry markers, we demonstrated that these molecular groups can be accurately identified based on morphological features, facilitating their future implementation in the clinical setting. Our proposed morpho-molecular classification highlights potential group-specific therapeutic opportunities, with differences in expression to DLL3, EGFR, FGFR and TERT inhibitor targets. CONCLUSIONS: Overall, our findings unify previously proposed molecular classifications and refine the lung cancer map by revealing novel tumour phenotypes with potential implications for prognosis and therapeutic management.