ID #992 Trajectory-Aware Functional Drug Screening Bridges Preclinical and Clinical Response Metrics in Diffuse Midline Glioma

Abstract

Functional drug screening (FDS) in patient-derived cell lines, generated from biopsy specimens, is increasingly employed to optimize treatments for diffuse midline glioma (DMG). However, a critical gap in translational drug evaluation persists, given differences in response assessment metrics and treatment protocols: While response in DMG patients is assessed longitudinally using MRI, drug screens typically rely on single-timepoint viability assays, often conducted in physiologically infeasible concentrations. We present a longitudinal FDS as dynamic response quantification paralleling patient assessments.

A total of 33 DMG-relevant therapeutics were tested at three concentrations each, in ten patient-derived DMG spheroids cultured for two weeks on an automated live-cell imaging and drug administration platform (3DTwin® Profiler). 1,100 longitudinal response profiles were analysed by an ordinary differential equation (ODE)-framework, to capture spheroid growth, treatment-induced cell kill, and post-treatment dynamics. The Bayesian Information Criterion was used to balance goodness-of-fit and model complexity based on a least-squares fit assessed by coefficients of determination (R2) and root-mean-squared-errors (RMSE). Inferred parameter distributions were analyzed using unsupervised clustering and compared to clusters derived from data-driven time–series–based trajectory analysis. We further demonstrate the same model applied to DMG patient response trajectories (n = 5) as a proof of concept.

We obtained excellent fits across cell lines and compounds (Rcell_line2 =0.969(IQR [0.960, 0.972], RMSEcell_line=3.8x106µm3 (IQR [3.0, 6.9] x106µm3), Rcompound2 = 0.97 (IQR [0.93, 0.98]), RMSEcompound=4.1x 106 (IQR [2.1, 5.3] x106µm3)). Clustering of model parameters identified distinct groups by cell lines and drugs characterized by qualitatively different response dynamics, which aligned with known mechanisms of drug action, such as cytostatic proliferation arrest or rapid cytotoxicity. These mechanistic clusters showed strong concordance with trajectory-based clustering, supporting the robustness of the derived response metrics.

We established a trajectory-aware, mechanistically interpretable metric of drug-response. By aligning preclinical and clinical response paradigms, this framework provides a foundation for improved translational relevance of functional precision oncology approaches in DMG.