The role of attractor dynamics and feed-forward integration in working memory

描述（由申请人提供）：工作记忆（Working memory， WM）是高级认知功能的一个基本组成部分，它能够在运动规划和决策过程中持续编码信息。WM功能障碍与精神分裂症、注意力缺陷障碍和与年龄相关的认知能力下降有关。在人类和猴子中，前额叶皮层（PFC）是分布式脑区网络的中枢，这些脑区共同支持WM， PFC的损伤完全消除了WM功能。因此，了解WM在PFC中的作用机制可能有助于理解人类WM相关疾病的病因学。灵长类动物的WM功能与PFC神经元有关，PFC神经元的放电率在感觉刺激终止后和运动反应之前的延迟期间持续升高。PFC中持续活动的机制尚不清楚，但理论工作提出了两种相互竞争的假设，即PFC中的网络相互作用如何在WM期间引起持续活动。“吸引子”假说提出，相互兴奋的相互作用产生了一种暂时不变的网络活动模式，这种模式在感觉输入终止后仍然存在，并且在时间上是稳定的。相比之下，“feed”

DESCRIPTION (provided by applicant): Working memory (WM) is a fundamental component of higher cognitive function that enables the persistent encoding of information during movement planning and decision-making. Dysfunction of WM has been linked to schizophrenia, attention deficit disorder, and age-related cognitive decline. In humans and monkeys, the prefrontal cortex (PFC) functions as the hub of a distributed network of brain regions that collectively supports WM, and damage to PFC abolishes WM function entirely. Therefore, understanding the mechanisms of WM function in PFC may be helpful to understanding the etiology of WM-related disorders in humans. Primate WM function has been linked to neurons in PFC whose firing rates remain persistently elevated during a delay period after the termination of a sensory stimulus and preceding a motor response. The mechanisms of persistent activity in PFC remain unknown, but theoretical work presents two competing hypotheses for how network interactions in PFC may give rise to persistent activity during WM. The "attractor" hypothesis proposes that mutually excitatory interactions produce a temporally invariant mode of network activity that persists after sensory input terminates and is stable in time. By contrast, the "feed forward integrator" hypothesis proposes that persistent activity results from the transient flow of sensory input through a sequence of modules. WM function then arises from the summed, time-varying outputs from these modules, approximating a step function. Computational models based on both hypotheses reproduce the firing rate responses of isolated PFC neurons during a spatial WM task. However, the experimental data needed to properly distinguish between these two competing frameworks is currently lacking. This proposal combines multiple-electrode recordings, behavioral manipulations and computational techniques to test three opposing predictions made by attractor and feed-forward models about firing rate dynamics and correlated variability during spatial WM. In all experiments, PFC neurons will be isolated and their responses recorded from an electrode array of up to 32 electrodes in monkeys performing a memory-guided oculomotor delayed response (ODR) task. Aim 1 will test conflicting model predictions about correlated variability among pairs of persistently active neurons during WM, by analyzing spike count correlations in paired single unit recordings during the memory ODR task. Aim 2 will test conflicting model predictions about the response of persistently active neurons to sustained visual stimulation, by analyzing single unit firing rate changes over time during a visually-guided ODR task. Finally, Aim 3 will test conflicting model predictions about the response of persistently active neurons to distracter stimuli during WM, by studying differences between single unit firing rates before and after a briefly flashed distracter stimulus. The result of these experiments will improve our understanding of WM function by supporting an existing model or pointing toward a new model of persistent activity generation in PFC.