ABSTRACT Genetically encoded fluorescent biosensors are powerful tools that allow the tracking of chemical events inside living cells, in real time. Even with a detailed understanding of biochemistry, enzymology, regulatory signaling, and genetics, there is no substitute for direct empirical information about the dynamics of chemical processes and signaling in cells. Unlike most biochemical measurements, the biosensors can provide spatial resolution at the level of single cells or parts of cells, and temporal resolution of seconds (or better). Nevertheless, there are major gaps in our ability to follow the details of cell signaling or metabolism using biosensors. For many interesting biochemical processes, we have no biosensors for the key metabolites. And even when a biosensor exists, it may not have the right sensitivity and specificity required for observing the desired process, or it may have sensitivity to pH or other environmental parameters that can mislead the experimenters. Biosensors are constructed by combining a fluorescent protein (like the jellyfish green fluorescent protein, GFP) with a binding protein for the chemical of interest. But finding the right way to combine the proteins is challenging, and even with a well-reasoned design, getting a biosensor with a strong, specific, and robust signal requires a large amount of optimization. This optimization is done by screening targeted random libraries of sensor variants. Current methods are typically limited to processing hundreds of variants per day, usually with just a single pair of measurements to guide selection of a variant for further validation. In the previous grant period, we developed a high-throughput, high-content screening pipeline that can screen thousands to tens of thousands of variants in a day, selecting “winners” based on detailed dose-response and selectivity data. Our approach uses microfluidic encapsulation of both DNA and protein for each variant in a small, semipermeable bead, followed by automated microscope imaging of thousands of beads under a series of conditions (varying [analyte], other test compounds, pH, etc.). This screen will permit thorough optimization of sensors and will allow success in otherwise failed sensor projects. We propose to use the new screening method to optimize some existing sensors (e.g., glucose and ATP:ADP ratio) and sensor prototypes (e.g., lactate and malonyl-CoA). We will also optimize a new general strategy for constructing sensors from dimeric transcription factors (a large family of microbial proteins useful for sensing), and we will exploit the high throughput of the screen in concert with computational methods to change the binding site specificity of existing sensors to produce sensors for important metabolic target molecules. In parallel, we will make improvements in the screening pipeline to expand its reach, with the goals of substan- tially increasing efficiency and throughput, and of recovering genotype information ...