Decoding neuroendocrine tumors: a cross-site multiomic approach unveils clinically relevant molecular subtypes

Neuroendocrine tumors (NETs) comprise a heterogeneous family of malignancies of wide anatomic distribution but shared clinical and molecular features. Although tissue-specific biology has been well characterized for some common primary sites, comprehensive analyses capable of defining biologically meaningful, site-independent subtypes are lacking. This study aimed to identify and validate cross-site molecular subtypes of NETs and to evaluate their clinical relevance. Multi-omic factor analysis (MOFA) and unsupervised clustering were used to integrate tumor gene expression and DNA methylation data from 194 patients with lung, stomach, pancreas, small intestine, and colorectal NETs. Latent factors representing biologically meaningful sources of variation were leveraged to generate a robust consensus clustering derived from eight unsupervised algorithms. Each molecular subtype was comprehensively characterized by integrating clinical, pathological and molecular features with survival through multivariable Cox proportional hazards models. Validation was performed in independent external cohorts of lung, small intestine, and pancreatic NETs with publicly available multi-omic datasets. We identified three distinct molecular subtypes (NS1–3). NS1 exhibited a strong immune phenotype and was enriched in metabolic and nutrient-digestion pathways. NS2 showed pronounced neuroendocrine differentiation with enrichment of neuronal and secretory programs. NS3 was characterized by global DNA hypermethylation, enrichment in DNA-replication pathways, and significantly poorer prognosis (10-year OS: 51.9% vs. 83.2–83.9% for NS3 vs NS1/NS2; p = 3 × 10 −4 ). NSs were independent of primary tumor location and retained independent prognostic significance in multivariable models adjusting for age, sex, TNM stage and tumor site. Validation in an external cohort confirmed the robustness of these findings. This new cross-site classification of NETs reveals convergent biological traits and common potentially targetable vulnerabilities, providing a framework for biomarker development and the rational design of novel therapeutic strategies.