Structural Determinants of Conformational Diversity in OXA Carbapenemases

Carbapenem-resistant Acinetobacter baumannii (CRAb) is a global health threat with limited therapeutic options. Antibiotic resistance is primarily driven by the expression of carbapenem-hydrolyzing class D β-lactamases (CHDLs), with clinically prevalent variants belonging to the OXA-23-like (OXA-23, OXA-146, OXA-225, and OXA-239), OXA-24-like (OXA-24 and OXA-160), and OXA-58-like (OXA-58) subfamilies. To explore the structural dynamics within and across these subfamilies, we employed reinforcement learning-driven enhanced-sampling molecular dynamics simulations integrated with convolutional variational autoencoders (CVAE) and BindSiteS-CNN model. Our CVAE-based deep learning analysis reveals that variants within the same subfamily exhibit distinct dynamic behavior, particularly in the B5B6-Loop, which are consistent with experimentally reported differences in substrate profiles and catalytic efficiency. The BindSiteS-CNN model further resolves the local active-site dynamics across the parent enzymes, offering insights into the structural features that distinguish these CHDL subfamilies. The results presented here highlight how global and local conformational dynamics differ across major CHDL families and provide structural insights for interpreting experimentally observed differences in substrate specificity and catalytic behavior in CRAb.