The 1000+ mouse project: large-scale spatiotemporal parametrization and modeling of preclinical cancer immunotherapies 2309521

Abstract

Abstract Introduction Preclinical CAR-T studies often monitor luciferase-labeled tumors in mice to gauge cytotoxicity, but analysis is frequently manual and hard to standardize. Here, we introduce maRQup (murine automatic Radiance Quantification and parametrization) for automated, spatiotemporal quantification and modeling of in-vivo tumor imaging. Methods maRQup is a Python pipeline that separates 1—5 mice per IVIS image, aligns anatomy by removing tails and vertically scaling mice, excludes slanted images, and computes average radiance per pixel as tumor burden. We curated >7,500 images from 1,060 mice across 37 experiments, with metadata on tumor dose, CAR targets, constructs, and CAR-T dose. Tumor trajectories were classified into growth/decay/relapse phases using thresholded changes in radiance, then fit with piecewise dynamical models (logistic growth for growth/relapse; exponential decay for decay) to extract kinetic rates. Results Complete tumor control was rare: among mice with initial tumor decay after CAR-T (43.2%), most later relapsed (77.5%). Comparing CD19 CARs with distinct costimulatory domains, CD19.CD28 more often prevented early tumor growth than CD19.4-1BB (29.7% vs 43.2% continued growth) yet relapsed more frequently (79.9% vs 58.6%). Increasing CAR-T dose improved immediate regression (reaching 100% at > 2 × 10^6 cells) but also simulatenously increased relapse frequency. Faster tumor growth was observed during the initial growth phase than during relapse. Anatomical mapping of tumor control revealed substantial spatial heterogenity in growth patterns, with the snout standing out as a particularly prominent relapse location. Conclusion maRQup delivers quantitative, statistically robust, and scalable readouts of preclinical CAR-T efficacy, exposing how construct choice, dose, and anatomical context shape distinct kinetic phases of tumor response, and providing a practical route to optimize immunotherapies. Funding Source National Cancer Institute Intramural funding Topic Categories Computational and Systems Immunology (COMP)