Identification of high-grade neuroendocrine carcinoma (HGNEC) biology across tumor types through transcriptomic profiling and validation in lung cancer.

3121 Background: HGNECs, including small-cell (SC) and large-cell (LCNEC), are aggressive malignancies frequently misclassified due to subtle morphology and inconsistent use of neuroendocrine (NE) immunohistochemistry in routine pathology workflows. It is unclear if tumors with transcriptomic signatures reflecting HGNEC-like biology have inferior outcomes with conventional treatment. Leveraging a large real-world dataset, this study aims to identify transcriptomically defined HGNEC to improve upon morphology-based classification. Methods: 49,144 specimens underwent DNA and RNA sequencing at Caris Life Sciences. Using a previously reported framework, we developed a weighted, z-scored based HGNEC activity score (HAS) from 10,137 pathologically defined HG- and LG- NEC specimens and applied it to characterize NSCLC samples (without any pathologist assigned HGNEC diagnosis) further stratified by histology (n=39,007). Overall survival (OS) was obtained from insurance claims data and calculated from specimen biopsy to last contact using Kaplan-Meier and Cox proportional hazards estimates. Results: To identify a high-confidence HGNEC-like subpopulation, a HAS threshold (z-score>=3) was applied, classifying 0.2% squamous (sq) and 2.3% non-squamous (nsq) as HGNEC-like; this analysis is focused in nsq. Patient demographics, including age, gender, race, ethnicity, and smoking history, were similar between the HGNEC-like and non-HGNEC-like nsq cohorts. However, biopsies from lymph nodes (26 vs 11%) and brain (9.1 vs 5.6%) were more frequent, and lung (39 vs 56%) less frequent in HGNEC-like tumors (all p-adj<0.05). HGNEC-like tumors were associated with poor OS (HR:1.18, 95% CI:1.07–1.31, p<0.001). HGNEC-like tumors were more frequently PD-L1 negative (64% vs 47%), and were enriched for mutations in TP53 , PTEN and RB1 (OR: 1.7-14.1), as well as amplifications in MYC and CCNE (OR: 2.2-3.5), with lower odds of KRAS and EGFR mutations (OR:0.4-0.5, p-adj<0.05). Consistent with a highly proliferative phenotype, HGNEC-like tumors were enriched for MKI67 expression (1.4-fold) and higher cell-cycle and replication stress scores (1.5-fold) (all p-adj<0.05) . Gene set enrichment analysis revealed significant activation of neurodevelopmental programs (NES: 2.0-3.0, all p-adj<0.05) and negative enrichment of immune pathways (e.g antigen presentation) (NES: -3.0 to -1.5, all p-adj<0.05). Finally, transcriptomic expression of potentially actionable SCLC-directed targets, including DLL3 (10-fold), SEZ6 (3.7-fold), and SSTR2 (10-fold), were all higher in HGNEC-like tumors (all p-adj<0.05). Conclusions: We identify a transcriptomically defined HGNEC-like subset of NSCLC with inferior survival and targetable NEC biology, supporting transcriptomic stratification to inform clinical trial design and selection of SCLC-directed therapies.