Recall requirements can drastically modulate working memory representations in human visual cortex

Visual working memory (VWM) allows for temporary storage of relevant information to support adaptive behavior. Prior fMRI studies investigating the mechanisms of passive storage have decoded mnemonic information from activity patterns in early visual cortex (EVC). Given that stored information is ultimately used to accomplish specific tasks, might EVC involvement be invoked by anticipated recall requirements? Here, we manipulated recall requirements during an orientation memory task to investigate how adaptive behavior impacts EVC representations. Specifically, we used four recall conditions based on a method-or-adjustment paradigm, as even a simple task of this sort involves a multitude of processes, and allows us to dissociate the roles of visual input, motor output, and online monitoring. To report a remembered orientation, participants (1) used “closed-loop” button presses to rotate a thin dial on the screen, (2) viewed a “matched replay” of a pre-recorded response, and indicated a clockwise or counter-clockwise offset of the final dial orientation, (3) viewed a “mismatched replay”, i.e. a pre-recorded response to a random orientation, and pressed a button if the dial crossed the currently remembered orientation, and (4) used button presses to rotate an “invisible dial” with an initial orientation shown only briefly before response onset. We uncovered a drastic increase in memory decoding during recall when the remembered orientation matched the final dial orientation (“closed-loop” and “matched replay” conditions). This implies that visual input alone (a rotating dial with final orientation unrelated to the remembered orientation, as in “mismatch replay”), motor output alone (high during “closed-loop”, low during “matched replay”), or online monitoring alone (not required during “exact replay” until the very end of the trial) cannot account for high memory decoding in EVC during recall. Instead, our findings suggest amplified information in EVC when mnemonic contents match sensory input during recall.