Immune checkpoint blockade (ICB) therapy has transformed the clinical care of cancer, yet the majority of patients do not derive clinical benefit and responders can acquire resistance to therapy. Noninvasive biomarkers to indicate early on-treatment response and resistance mechanisms are needed to improve patient management. We engineer activity-based synthetic biomarkers called immune sensors for monitoring checkpoint blockade therapy (INSIGHT), which comprise a library of mass-barcoded peptides conjugated to ICB antibodies (e.g., αPD1). INSIGHT allows detection of in vivo T cell and tumor protease activity by quantification of cleaved peptide fragments that have cleared into urine. αPD1-sensor conjugates monitoring the T cell protease granzyme B (GzmB) retained target binding and were capable of sensing T cell killing of cancer cells. In syngeneic tumors, systemic administration of these conjugates resulted in therapeutic efficacy comparable to unconjugated antibodies and produced elevated reporter signals in urine indicative of tumor responses by the second dose. To differentiate resistant tumors, we analyzed the transcriptomes of ICB-treated tumors for protease signatures of response and resistance and developed a multiplexed library of mass-barcoded protease sensors. This library enabled us to build machine learning classifiers based on urine signals that detected and stratified two mechanisms of resistance, B2m and Jak1 loss-of-function mutations. Our data demonstrates the potential of INSIGHT for early on-treatment response assessment and classification of refractory tumors based on resistance mechanisms.

Classifying the mechanisms of antibiotic failure has led to the development of new treatment strategies for killing bacteria. Among the currently described mechanismswhich include resistance, persistence and tolerancewe propose defiance as a subclass of antibiotic failure specific to prodrugs. Using locked antimicrobial peptides (AMP) that are activated by bacterial 5 proteases as a prototypic prodrug, we observe that although treatment eliminates bacteria across the vast majority of environmental conditions (e.g., temperature, concentration of growth nutrients), bacteria spontaneously switch from susceptibility to defiance under conditions that alter the competing rates between bacterial proliferation and prodrug activation. To identify the determinants of this switch-like behavior, we model bacteria-prodrug dynamics as a multi-rate 10 feedback system and identify a dimensionless quantity we call the Bacterial Advantage Heuristic (BAH) that perfectly classifies bacteria as either defiant or susceptible across a broad range of treatment conditions. By recognizing that the bacterial switch from susceptibility to defiance behaves analogously to electronic transistors, we construct prodrug logic gates (e.g., AND, OR, NOT, etc.) to allow assembly of an integrated 3-bit multi-prodrug circuit that kills defiant bacteria 15 under all possible combinations of BAH values (i.e., 000, 001, …, 111) that represent a broad range of possible treatment conditions. Our study identifies a form of bacterial resistance specific to prodrugs that is described by a predictive dimensionless constant to reveal logic-based treatment strategies using multi-prodrug biological circuits.