A parietal grid-like code rotates with cognitive maps but lags rapid behavioral transfer

The neural grid code has been proposed to provide a mechanism for generalization and transfer of relational knowledge between situations enabling rapid adaptation of behavior in novel circumstances. However, to date, very little is known about the dynamics with which grid representations change at context transitions, or how such dynamics relate to downstream behavioral adaptation. Here we tested whether grid representations measured with fMRI rotate to match behavioral goals at context transitions and whether such rotations underlie knowledge transfer. Human participants (25 female and 23 males) performed a task that included unsignaled state changes at which the position of multiple target locations abruptly and synchronously rotated by the same degree. After state changes, participants were able to leverage the relative positions of the targets to rapidly infer locations, even novel ones, constituting a form of zero-shot transfer. We observed a cognitive grid-like code in the right posterior parietal cortex with a consistent phase angle that rotated with the relative positions of the targets. However, this rotation was too slow to account for rapid improvements in performance after a state change, and instead these improvements were more closely related to representations of the identity and location of spatial targets in the frontoparietal and orbitofrontal cortex, respectively. Our results highlight the ability of humans to rapidly transfer knowledge and demonstrate that a parietal grid-like code rotates into behaviorally relevant reference frames, but indicate that alternate neural mechanisms support rapid knowledge transfer.Significance Statement Behavioral flexibility depends on quickly transferring behaviors learned from past experiences to novel situations. Recent prominent theoretical accounts have argued that transfer relies on an internal cognitive map of the environment maintained by the neuronal grid code, but few experiments have examined its role in a task requiring behavioral flexibility. In a spatial predictive inference task, we observed a parietal grid-like code that adapted to task changes, but did so too slowly to account for participants' rapid behavioral transfer. Our results instead point to task representations in the frontoparietal and orbitofrontal cortices associated with this behavior, providing new insights into the neural mechanisms underlying behavioral transfer.