Compositional architecture: Orthogonal neural codes for task context and spatial memory in prefrontal cortex.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The prefrontal cortex (PFC) is vital for working memory.
• How the PFC adapts working memory to different tasks is not fully understood.
• Compositional theories suggest reusable neural components, but evidence for task-specific working memory has challenged this.

Purpose of the Study:

• To investigate whether neural representations of working memory in the PFC are shared across tasks with opposing demands.
• To test the predictions of compositional theories of neural computation in the context of working memory.

Main Methods:

• Recorded population neural activity in macaque monkeys.
• Utilized two spatial working memory tasks: a 'look' task (movement toward locations) and a 'no-look' task (avoidance of locations).
• Analyzed neural data using population-based and low-dimensional subspace methods to identify representational geometry.

Main Results:

• Spatial memory representations were largely conserved at the population level across both tasks.
• A common low-dimensional neural subspace encoded spatial memory in both the look and no-look tasks.
• Task identity was represented in a separate, orthogonal neural subspace, independent of spatial memory content.

Conclusions:

• Neural representations in the PFC support a compositional model of working memory.
• Representational geometry allows for the flexible reuse of mnemonic codes across different behavioral contexts.
• The brain maintains independent representations for task context and memory content, enabling cognitive flexibility.