Engineered gene circuits with reinforcement learning allow bacteria to master gameplaying

Gene circuits enable cells to make decisions by controlling the expression of genes in reaction to specific environmental factors1. These circuits can be designed to encode logical operations2-7, but implementation of more complex algorithms has proved more challenging. Directed evolution optimizes gene circuits8 without the need for design knowledge9, but adjusting multiple genes and conditions10 in genotype searches poses challenges11. Here we show a multicellular sensor system, AdaptoCells, in Escherichia coli, that can evolve complex behavior through an accelerated adaptation to chemical environments. AdaptoCells recognize chemical patterns and act as a decision-making system. Using an iterative improvement method, we demonstrate that the AdaptoCells can evolve to achieve mastery in the game of tic-tac-toe, demonstrating an unprecedented level of complexity for engineered living cells. We provide an effective and straightforward way to encode complexity in gene circuits, allowing for fast adaptation in response to dynamic environments and leading to optimal decisions.