Using Latent Chemical Recognition in an Evolved Periplasmic Binding Protein Family to Diversify Biosensors

Abstract

Genetically encoded fluorescent biosensors have gained traction in neuroscience as continuous, reagentless reporters of cellular activity in situ. These sensors, often soluble, can also provide time-resolution in multiple biosensing form factors: benchtop and wearable devices, lyophilized powder tests, and smartphone-based diagnostic tests. These biosensors often take advantage of naturally occurring conformation-switching but require extensive screening to optimize the linkers to a fluorescent reporter. Each new target ligand often requires its own engineering campaign. We asked whether sensors evolved towards a particular target retain useful recognition scope for others. We screened a family of 18 OpuBC-cpGFP sensors evolved toward nicotinic agonists, SSRIs, opioids, and other neural drugs, against 63 structurally diverse compounds. We found that 24 ligands activated at least one biosensor with {Delta}F/F0 > 0.3, sufficient to begin directed evolution, with 8 of those ligands activating at least one biosensor with {Delta}F/F0 > 1.0, the regime of dynamic range usable in end applications. With 124 ligand-biosensor pairs in total, we found multiple leads suitable for directed evolution. Most notably, ligands participating in hits spanned well beyond neural drugs and included DEHP, ergothioneine, ciprofloxacin, thiamine, betahistine, L-carnitine, and L-thyroxine. Across the biosensor family, mutation distance weakly predicted substrate scope. In particular, we observed sequence-function cliffs that could be exploited for future protein engineering campaigns. Thus, broad screening of performant scaffolds offers rapid bootstrapping in biosensor engineering particularly for exogenous molecules.