Geometry of abstract learned knowledge in the hippocampus

Area of Science:

• Neuroscience
• Cognitive Science

Background:

• Hippocampal neurons are known to encode physical variables like space.
• Functional magnetic resonance imaging (fMRI) studies indicate the human hippocampus also encodes abstract, learned variables.
• The integration of these physical and abstract variables within neural representations remains unclear.

Purpose of the Study:

• To investigate how individual hippocampal neurons jointly encode physical and abstract variables.
• To determine if the hippocampus integrates spatial information with abstract learned variables during a decision-making task.

Main Methods:

• Utilized two-photon calcium imaging in mice performing a virtual reality decision-making task.
• Applied nonlinear dimensionality reduction techniques to analyze neural population activity.

Main Results:

• Individual dorsal hippocampus neurons jointly encode accumulated evidence and spatial position.
• Neural population activity is constrained to a low-dimensional manifold (4-6 latent variables).
• Physical and abstract variables are mapped in an orderly, geometric manner within this low-dimensional space, showing consistency across mice.

Conclusions:

• The hippocampus creates conjoined cognitive maps, integrating physical and abstract information.
• This suggests a general hippocampal computation: generating task-specific low-dimensional manifolds for geometric representation of learned knowledge.