Abstract
A large body of neuroimaging and brain stimulation studies have investigated the neural mechanisms of visual working memory (VWM). However, existing results are highly controversial. In particular, the causal links between regional activity and cortical computation are yet to be fully established. In this study, we combined computational modeling and continuous theta burst repetitive transcranial magnetic stimulation (rTMS) to investigate the causal contributions of V1, intraparietal sulcus (IPS), and dorsolateral prefrontal cortex (DLPFC) in the classical delay-estimation VWM task. 20 subjects completed the study in four days each for one of four experimental conditions (V1/IPS/DLPFC/sham) with at least a 48-hour interval between two real rTMS sessions. On each day, we delivered 600 stimulation pulses over the region-of-interest of that day before a subject performed the task. During the task, the subject was required to memorize a set of color squares (set size 2/4/6) for a short period (1000ms), and then choose the color of a cued square on a colorwheel. Most importantly, we fitted and compared seven mainstream models of VWM to uncover the computational underpinnings of perceptual consequences induced by rTMS. We found that rTMS over V1, compared with the sham condition, impaired the behavioral performance at set size 2 but not higher set size levels. rTMS over IPS and DLPFC had no significant effects on VWM performance. Computational modeling revealed that the variable precision (VP) (Van den Berg et al., 2012) model was the best-fitting model in all four conditions, suggesting that rTMS did not qualitatively change the computational strategy underlying VWM. However, according to the VP model, rTMS over V1 significantly reduced the amount of VWM resources at low set size levels. Taken together, our results support the sensory account of VWM and highlight the causal role of V1 in controlling the amount of VWM resources.