Iterative Extracellular Vesicle Protein Co-Expression Biomarker Refinement for Preoperative Classification of Histopathological Growth Patterns in Colorectal Liver Metastasis Patients

Abstract

Preoperative triage of colorectal liver metastases (CRCLM) by histopathological growth pattern (HGP)--angiogenesis-dependent desmoplastic (dHGP) and vessel co-opting replacement (rHGP)--could guide anti-angiogenic therapy, yet HGP scoring requires resected tissue. We present an extracellular vesicle (EV) inner and outer protein (EVPio) co-expression assay and iterative biomarker refinement for plasma-based HGP classification. We established a minimal, high-throughput plasma pre-processing workflow (low-speed centrifugation and 0.45 m filtration) with comparable EVPio assay performance to size-exclusion chromatography. We established an EV biomarker selection template with growing cohorts--feasibility (n = 3), pilot (n = 9), discovery (n = 67)--ranking candidate protein pairs by signal quality (SNR, CV), redundancy (inter-correlation/orthogonality), and HGP separation (effect size, significance, ROC). This process reduced an initial 19x18 capture/detection set to a focused 9x9 panel (81 co-expression pairs). In a 58-patient CRCLM subset (22 dHGP, 14 rHGP, 22 mixed), three pairs achieved high signal quality with significant differential expression across HGPs. A three-feature linear discriminant model yielded 75.9% cross-validated accuracy (AUC 0.77) for classifying pure dHGP vs. non-dHGP. The results show that co-expression signatures capture defining features of HGP biology while revealing heterogeneity. The proposed EV biomarker refinement template is generalizable and our results show that co-expression signatures capture defining features of HGP biology supporting efforts towards clinically actionable, HGP-driven therapeutic guidance.