Temporal Integration of Serum Proteomics, Metabolomics and MRI Tumor Volumetrics via Deep Learning Identifies Systemic Mediators of Glioblastoma Response to Chemoradiotherapy

TL;DR

This study uses deep learning to combine MRI scans and blood tests to find biomarkers that predict how glioblastoma patients respond to treatment.

Contribution

The novel integration of serum proteomics, metabolomics, and AI-segmented MRI data identifies systemic metabolic pathways linked to GBM treatment response.

Findings

AI-derived CE volumes decreased early post-CRT while edema increased later.

Low-survival clusters showed greater CE alterations and downregulated metabolic pathways.

Key metabolites and proteins were linked to GBM biology and treatment response.

Abstract

Glioblastomas (GBM) are highly aggressive, treatment-resistant brain tumors lacking clinically actionable, noninvasive prognostic biomarkers. Tumor response after standard-of-care chemoradiation (CRT) is difficult to interpret on imaging, and post-CRT MRI changes have not been well linked to molecular features or potential biomarkers.

We evaluated differential proteomic and metabolomic expression in patient serum in relation to AI-segmented MRI volume changes after CRT to integrate clinical, molecular, and imaging data with patient outcomes.

Fifty- five clinically annotated GBM patients provided serum samples pre- and post-CRT, analyzed using the SomaScan® proteomic platform and SECIM metabolomic assay. Pathway signatures were derived from pre- vs. post-CRT differential expression. MRI scans underwent AI segmentation to quantify contrast-enhancing (CE), non-enhancing (NE), and edema volumes. We assessed correlations between early (immediately post-CRT) and late (six months post-CRT) imaging changes and molecular alterations. Integrated multiomic and imaging features were used for unsupervised clustering to identify survival-associated patient groups, followed by pathway re-identification.

AI-derived CE volumes decreased significantly during the early period, while edema increased significantly during the late period. CE changes were associated with metabolic pathways relevant to GBM biology, including epithelial–mesenchymal transition, in ammatory response, coagulation, and interferon-γ signaling. Clustering revealed two groups with distinct survival outcomes; CE alterations were significantly greater in the low-survival cluster (p = 0.02). Multiomic analysis (MOGSA) showed downregulation of key metabolic pathways in the low-survival group, including the citric acid cycle, Warburg effect, amino acid metabolism, oncogenic 2-hydroxyglutarate activity, and purine metabolism. Contributing metabolites included fumarate, succinate, citrate, and 2-hydroxyglutarate, while major proteomic contributors included MPC1, PDHB, DLAT, DLST, IDH3, SDHB, and FH.

AI-derived MRI tumor-volume changes after CRT correspond to specific serum proteomic and metabolomic alterations, highlighting metabolic pathways linked to contrast-enhancing tissue dynamics in GBM.

Figures