Prion-like dynamics of galectin-3 as a driver of colorectal cancer progression: A digital pathology and computational simulation study of supramolecular aggregation.

Abstract

e15092 Background: Modern oncology is identifying "onco-prions"—proteins that regulate malignancy through biophysical behaviors analogous to neurodegenerative prions. Galectin-3 (Gal-3) is a chimeric lectin characterized by an N-terminal intrinsically disordered domain (IDR) that drives Liquid-Liquid Phase Separation (LLPS), forming dynamic biomolecular condensates and extracellular "lattices" that stabilize oncogenic signaling. While Gal-3 expression is a known prognostic factor, the clinical significance of its physical state—transitioning from soluble monomers to supramolecular aggregates—remains poorly defined in colorectal cancer (CRC). Methods: We integrated computational modeling with clinical validation in a retrospective cohort of 94 CRC tissue samples (Stages I–IV), 10 adenomas and 8 normal colon tissue samples. To characterize the "prion-like" potential of Gal-3, we employed the PLAAC (Prion-Like Amino Acid Composition) algorithm to identify prion-like domains (PrLDs) and utilized the Neurosnap platform for deep learning-based structural simulations of phase separation dynamics. Findings were validated using QuPath-based digital image analysis, applying custom segmentation scripts to quantify Gal-3 aggregation density (nuclear "speckles," cytoplasmic clusters, and extracellular lattices) across CCR progression, including normal, adenomas and tumor stages. Results: Bioinformatic analysis with PLAAC confirmed high-probability PrLDs within the Gal-3 N-terminal domain, which Neurosnap simulations identified as the primary driver of multivalent self-association. In clinical samples, QuPath-assisted quantification revealed a significant correlation between Gal-3 aggregation density and advanced tumor stage (p < 0.05). Conclusions: Galectin-3 operates as a functional prion-like agent in CRC, where its supramolecular aggregation state dictates its pathogenic potential. The integration of PLAAC, Neurosnap, and QuPath provides a robust workflow for characterizing these "onco-prion" dynamics. Our results suggest that the physical aggregation of Gal-3 is a superior biomarker for progression compared to simple protein expression levels. These findings support a shift toward "phase engineering" therapies, such as dominant-negative variants (Gal-3C) or allosteric inhibitors, to dismantle the prion-like scaffolds sustaining the malignant phenotype.