Yueying Dong¹, Yun-Chen Hung¹, Connie Xie¹, Anastasia Kiyonaga¹; ¹UC San Diego

Prominent working memory (WM) models posit that transient goal-relevant information is distributed across activations in sensorimotor cortical regions. Newer evidence suggests that even peripheral physiology like the pupils can contain WM content information. For instance, darker WM items elicit greater delay-spanning pupil dilation than brighter items, even when sensory input is matched between conditions. This pupillary WM response is sensitive to the behavioral priority of the WM content and is greater for individuals with stronger self-reported visual imagery. This suggests that pupillary WM effects may be both flexible to behavioral goals and magnified when relying on more visual maintenance strategies. Here, we manipulate WM probe demands, during continuous eye-tracking, to test whether WM task strategy modulates the WM pupil effect. Participants completed a WM task for naturalistic scene images and were probed on either the precise visual features or the semantic category of the WM content. Participants encoded both daytime and nighttime WM sample images, followed by a retrocue indicating which item would be tested. The retrocue color also varied, to signal whether the visual or semantic item features would be probed. Pupil dilation was greater during the delay when a nighttime scene was prioritized (vs. daytime), corroborating prior work that used oriented Gabors of varying luminance as stimuli. Moreover, the WM-pupil effect here was larger when the task probed precise visual information. The effect diminished when the task probed higher level semantic categories, resembling the previously observed diminished response in individuals with weaker visual imagery. These findings demonstrate that peripheral sensorimotor responses can carry information not only about the visual properties of WM content, but also how it will be used. Pupillary WM responses are modulated by the upcoming WM task, pointing to a potential role for the earliest stages of visual processing in strategic and flexible WM maintenance.