Flexible Control of Sequence Working Memory in the Macaque Frontal Cortex
Research Background
In our daily lives, Sequence Working Memory (SWM) is crucial; for example, when filling in a birth date, the year, month, and day need to be recalled and arranged in a specific order. However, how the brain controls information in sequential memory and flexibly orders information under different task demands remains an unsolved mystery in neuroscience. To explore this process in depth, Jingwen Chen and other scholars published “Flexible Control of Sequence Working Memory in the Macaque Frontal Cortex” in the journal Neuron. The research was conducted by researchers from the Institute of Neuroscience, Chinese Academy of Sciences, the Department of Psychology at New York University, Shanghai Lingang Laboratory, and other institutions. They studied the neural dynamics and sequence memory control mechanisms in macaques by recording the electrophysiological activity in the frontal cortex of the monkeys as they performed forward and reverse memory tasks.
Research Purpose
The research aims to reveal the organization and dynamic properties of neural activity in the prefrontal cortex during sequence working memory tasks and to explore if there are independent low-dimensional sensory and memory subspaces that support flexible memory control. Additionally, the study focused on neural activity during single-error trials, hoping to further understand the execution process of the brain in sequence memory control by comparing correct and incorrect neural responses.
Research Methods and Procedure
The research team implanted a 157-channel microdrive electrode array in the prefrontal and premotor cortex regions of macaques to record neural activity during sequence memory tasks. The specific experimental flow is as follows:
Experimental Design and Task Setup
The experiment trained three macaques (coded O, G, L) to learn and complete a delayed sequential sorting task. The task is divided into “forward sorting” and “reverse sorting”: in “forward sorting,” macaques need to touch the corresponding location on the screen in the order of presentation; in “reverse sorting,” macaques need to touch the presented items in reverse order. The researchers randomly displayed items in 1 to 3 different spatial locations in the task, and the macaques had to recite the presentation sequence after a short delay according to the task requirements.
Neural Recording and Single Neuron Response
In the electrophysiological data recordings, a total of 6,790 neurons were recorded. The research team analyzed the response characteristics of single neurons and found that neuron activity is divided into “stimulus-triggered” and “delay-maintained” categories. Some neurons showed brief encoding of specific items upon stimulus presentation, while others showed sustained activity during the delay period, integrating spatial and sequential information with joint coding features. Further research found that these two types of neuronal activities are not independent but are deeply intertwined at the single-neuron level.
Prefrontal Cortical Subspace Dynamics in Forward Tasks
In the forward task, researchers used low-dimensional decomposition techniques to identify one “sensory subspace” and three “memory subspaces,” corresponding to different memory levels. In the sensory subspace, item information decayed rapidly within 300 milliseconds after the stimulus ended, while in the memory subspace, item information was continuously maintained, forming a stable ring structure. This indicates that different memory contents are orderly and independently represented in the brain’s neural state space.
Flexible Control Mechanisms in Reverse Tasks
In reverse tasks, macaques need to recall items in reverse order and enter memory subspaces sequentially. Researchers found that the sequence control mechanism in the reverse task is more complex, requiring flexible control of memory subspace selection. Unlike forward tasks, in reverse tasks, item information gradually enters the corresponding memory subspace upon task completion, with varying times based on task length and sequence requirements. This process reflects the presence of flexible control mechanisms in the brain, capable of adjusting the sequence and timing of information flow according to task demands.
Research Results
Identification and Independence of Neural Subspaces
The research shows that during sequence memory tasks in the prefrontal cortex, sensory information and memory information can be distributed in different low-dimensional subspaces, achieving flexible control and maintenance of information by separately controlling the activation states of these subspaces. Neurons in the prefrontal cortex can exhibit separation of sensory and memory information within the same neural group and demonstrate independent, nearly orthogonal low-dimensional subspaces. The dynamic changes in these subspaces are closely related to task demands.
Neural Prediction in Single Trials
By analyzing neural activity in error trials, researchers found that even in cases of errors, neural dynamics can accurately predict the type of behavior (sequential error or item error) in macaques. For instance, when monkeys confused item sequences during a sequence task, item information would be recorded in the wrong memory subspace, displaying operational errors in the brain control system. Furthermore, it was discovered that when incorrect items were spatially close to target items, interference or competition phenomena existed between target and incorrect items during the delay period. This finding validates that subspace control mechanisms can simulate correct operations during errors, revealing the neural dynamic processes of sequence memory.
Flexibility of Neural Control
The study also conducted cross-task generalization tests to explore the extent of shared neural subspaces between forward and reverse tasks. It was found that the prefrontal neural activity of macaques could exhibit subspace sharing between the two tasks, and these subspaces could flexibly adjust and combine according to task requirements. This indicates that the brain’s control system has highly flexible abstract control capabilities, capable of adjusting the flow of information to meet the diverse demands of memory sorting tasks under different task requirements.
Research Conclusions and Significance
This research reveals the dynamic organization of sequence working memory control in the prefrontal neural group of macaques and proposes a general control model that can support flexible sequence working memory processes. It shows that the brain organizes low-dimensional subspaces to separately store sensory information and sequence information of different memory levels and flexibly regulates them within the prefrontal neural groups to support flexible sorting tasks. This mechanism is of significant importance in cognitive neuroscience, providing a new perspective for understanding the neural mechanisms of working memory.
Research Highlights
Innovative Subspace Decomposition Method: This study used low-dimensional subspace decomposition techniques to demonstrate independent subspace representation of sensory and memory information in sequence memory.
Flexible Memory Control Mechanism: The brain can flexibly sort and select memory information under different task demands, showing the flexibility of prefrontal neural groups in cognitive control.
Neural Precision in Single Trial Prediction: The research shows that even under error conditions, neural dynamics can accurately predict the type of error, supporting the understanding of neural encoding in memory control processes.
Application Prospects
This research has important application value in the fields of neuroscience and cognitive control. The exploration of the brain’s control of information flow flexibility and the organizational mode of sequence memory can provide insights for artificial intelligence, brain-computer interfaces, and treatment strategies for cognitive disorders.