Pheno-MYCN maps the morphological footprint of MYCN amplification in paediatric neuroblastoma

Abstract

MYCN amplification has long been a prognostic marker in paediatric neuroblastoma, yet is typically assayed in bulk, alongside rather than within the heterogeneous tissue architecture pathologists assess. This leaves a gap: MYCN status alone cannot localise MYCN-associated biology, while morphology alone cannot assign molecular risk. Motivated by our finding that the two together identify high-risk cases missed by either, we developed Pheno-MYCN, a weakly supervised framework linking slide-level MYCN prediction to interpretable morphological sub-populations on routine H&E whole-slide images. The aim is not a stronger classifier: prediction probes what MYCN amplification does to the tissue, its evidence open to pathological scrutiny. Across 189 slides, Pheno-MYCN resolved each into phenotypic clusters that expert review mapped to neuroblastoma morphologies. Cell-level profiling revealed MYCN amplification "marked" every sub-population, through a different feature in each: densely cellular yet disorganised tumour with sparser, less diverse networks; chiefly abundance in necrotic and haemorrhagic regions. MYCN-amplified-like tissue was identifiable per slide from these features alone (AUC 0.93-1.00, leave-one-slide-out) and traced as a continuous gradient within tumours. Thus MYCN amplification leaves a concrete, interpretable footprint that can be read and localised on routine H&E, offering a low-cost means to flag and map it where molecular testing is limited.