Abstract
The prefrontal cortex (PFC) is critical for our ability to rapidly and flexibly adapt our behavior in new environments based on our previous experience. Despite its importance, the neural substrates and mechanisms by which the PFC supports this function have long remained enigmatic. Recent advances, however, have begun to change this. An increasingly large body of work suggests the PFC represents structured relationships-both among states of the outside world and between internally generated actions. In this review, we describe work from rodents, nonhuman primates, and humans to draw attention to the breadth of such representations and how they support flexible behavior. Across species, the PFC appears to represent the relational structure of problems: how stimuli relate to one another in cognitive maps or how different behaviors relate to one another when pursuing a goal. These results have started to reveal shared computational principles for PFC that generalize from rodents to humans and have inspired formal computational models and simulations. By reviewing experimental work showing both correlation and causation through invasive and noninvasive methods, along with theoretical work using artificial neural networks, we aim to highlight similarities and differences between species and models to provide a common language for interpreting findings in PFC. This will move us closer to a mechanistic understanding of the PFC that scales across tasks and species.