Abstract
Parietal cortical areas have been implicated as a critical neural substrate for visual working memory. Human fMRI and ERP studies have revealed persistent parietal activation during delay periods of visual working memory tasks and shown that such activation scales with the capacity limit of visual working memory. A second line of evidence has been provided by neural recordings in primates performing memory-guided saccades. Neurons in parietal cortical areas, such as the lateral intraparietal area (LIP), have been shown to exhibit persistent activity during the memory delay of this task, but the contribution of LIP persistent activity to mnemonic processes remains poorly understood. Specifically, it is unclear whether persistent activity carries a retrospective visual or prospective saccade-related representation, or how it could be related to the capacity limit of visual working memory. To address this, we recorded the activity of single LIP neurons in three monkeys while they performed memory-guided saccades and carried out two sets of analyses. We first compared the visual and motor responses of each neuron with persistent delay period activity. LIP neurons exhibited a pattern of activity consistent with a noisy retrospective code: most neurons had greater visual than saccade-related activity, and persistent activity was most strongly linked with the preferred direction of each neuron, but highly variable. We then investigated how persistent activity could relate to visual working memory capacity. We derived estimates of baseline and persistent activity from our sample of LIP neurons, and used these values to compute the theoretical number of discriminable representations that could be carried over a range of realistic simulated activity distributions using an ROC analysis. This number ranged from one to approximately four. These data show that LIP persistent activity is visually biased and suggest a neural basis for the capacity limit of visual working memory.