AI guided engineering of pH responsive antibodies enables tumor selective targeting and improves the therapeutic index

Tumor tissues exhibit a mildly acidic extracellular microenvironment (pH 6.0-6.8), in contrast to normal tissues at physiological pH (~7.4), and exploiting this difference could enable tumor-selective antibody targeting and reduce systemic toxicity; however, rationally engineering pH-dependent binding into antibodies remains challenging. We developed an deep learning model that predicts how mutations in antibody complementarity-determining regions (CDRs) modulate antigen binding as a function of pH and applied it to redesign a B7-H3-targeting antibody to preferentially bind under acidic conditions. The resulting antibodies showed strong pH-dependent binding, achieving affinity ratios (pH 6.0 / pH 7.4) exceeding 100-fold, with high affinity maintained in acidic conditions representative of the tumor microenvironment and markedly reduced binding at physiological pH. When reformatted as antibody-drug conjugates (ADCs), these pH-responsive antibodies exhibited improved selectivity and a threefold expansion of the drug administration window compared with the parent antibody. These results demonstrate that AI-driven CDR engineering enables systematic design of microenvironment-responsive antibodies and offers a generalizable strategy to enhance the therapeutic index of antibody-based cancer therapeutics.