Journal: bioRxiv
Article Title: AI-Guided CAR Designs and AKT3 Degradation Synergize to Enhance Bispecific and Trispecific CAR-T Cell Persistence and Overcome Antigen Escape
doi: 10.1101/2025.06.12.658477
Figure Lengend Snippet: (A) In-silico analysis of CAR-T cell-treated patients (n=4,219) revealed a high relapse rate, with 42.11% (n=216 of n=513 overall relapse patients) experiencing CD19-negative recurrence after monospecific CAR-Therapy (n=2,916). (B) Schematic overview of the CAR design strategy showing mono, bi, and trispecific constructs targeting CD19, CD20, and CD22. (C) Experimental workflow illustrating CAR screening: 1,452 CARs were transduced into primary T cells and analyzed for signal-1 (activation), signal-2 (exhaustion), and signal-3 (cell death). (D) Categorization of CARs into low (L), medium (M3), and high (H) levels based on fluorescence intensity cutoffs determined by monospecific CD19 CARs. (E) Distribution of 1,452 screened CARs across L-, M-, and H-CARMSeD categories using the CAR-Mediated Self-Destruction (CARMSeD) scoring system. (F) AI model development pipeline for CAR dysfunction risk prediction, based on 1,452 CAR constructs with an 80:20 split for training and testing. (G–J) Performance metrics of AI model predicting CAR-Mediated Self-Destruction (CAR-MSED) scores using 1,452 CAR constructs (G) Model accuracy over 50 epochs, achieving a training accuracy of 0.98 and validation accuracy of 0.95. (H) Scatter plot comparing measured versus predicted CAR-MSED scores for training (R 2 = 0.87) and validation (R 2 = 0.83) sets. (I) Predicted versus measured CAR-MSED scores on the validation set, categorized into low (L-CARMSED, blue), medium (M-CARMSED, orange), and high (H-CARMSED, green) groups. (J) Box plot of predicted signal scores for 9,372 unknown sequences, classified as L-CARMSED (2,749 sequences), M-CARMSED (1,468 sequences), and H-CARMSED (5,155 sequences). (K) Molecular dynamics simulation of CAR constructs with varying linker lengths, assessing CAR-CAR interaction. Structural conformations at 0 ns, 50 ns and 200 ns for different CAR scFv arrangements highlighting CDR regions (surface transparency 30%), Root Mean Square Deviation (RMSD) plots over 200 ns for the both constructs, respectively, indicating structural stability and conformational changes. (L) In vitro receptor binding affinity validation for top humanized scFvs of CD19, CD20, and CD22 CARs (n=6).
Article Snippet: Briefly, CD19, CD22 CAR expression was evaluated using CD19 and CD20 CAR detection antibodies and CD22 CAR expression (Miltenyi Biotec) was evaluted using Protein L-APC (Cell signaling) followed by PE-conjugated anti-biotin secondary antibodies (Miltenyi Biotec).
Techniques: In Silico, Construct, Activation Assay, Fluorescence, Biomarker Discovery, In Vitro, Binding Assay