Oscillatory control over representational geometry of sequence working memory in macaque frontal cortex

To process sequential streams of information, e.g., language, the brain must encode multiple items in sequence working memory (SWM) according to their ordinal relationship. While the geometry of neural states could represent sequential events in the frontal cortex, the control mechanism over these neural states remains unclear. Using high-throughput electrophysiology recording in the macaque frontal cortex, we observed widespread theta responses after each stimulus entry. Crucially, by applying targeted dimensionality reduction to extract task-relevant neural subspaces from both LFP and spike data, we found that theta power transiently encoded each sequentially presented stimulus regardless of its order. At the same time, theta-spike interaction was rank-selectively associated with memory subspaces, thereby potentially supporting the binding of items to appropriate ranks. Furthermore, this putative theta control can generalize to length-variable and error sequences, predicting behavior. Thus, decomposed entry/rank-WM subspaces and theta-spike interactions may underlie the control of SWM.

为处理诸如语言在内的信息序列流，大脑需依据序数关系，在序列工作记忆（sequential working memory，SWM）中对多个序列项目进行编码。尽管神经状态的几何结构可在前额叶皮层中表征序列事件，但其调控机制仍未明晰。本研究通过在猕猴前额叶皮层开展高通量电生理记录，于单次刺激呈现后观测到广泛分布的θ节律响应。尤为关键的是，通过靶向降维方法从局部场电位（Local Field Potential，LFP）与锋电位数据中提取任务相关神经子空间后，我们发现θ功率可瞬时编码每一个依次呈现的刺激，无论其呈现顺序如何。与此同时，θ-锋电位相互作用与记忆子空间呈等级选择性关联，或可支持将记忆项目绑定至恰当的序数位置。此外，该推测性的θ节律调控机制可推广至可变长度序列与错误序列，并能预测行为表现。综上，分解得到的刺激输入/序数工作记忆子空间与θ-锋电位相互作用，或为序列工作记忆调控的核心基础。