Advances in Neurobiology: “Navigating” Predictable Spontaneous Brain States through Cognitive Maps

Background Introduction

Spontaneous brain activity refers to brain processes that are not constrained by specific inputs or outputs, and these processes play an important role in neural and cognitive variability. Although the specific mechanisms of spontaneous brain activity are not fully understood, recent research has shown that these spontaneous brain states exhibit certain regularities and predictabilities during dynamic reorganization. One of the important theories in this research field is the concept of the “cognitive map,” first proposed by Tolman in 1948, which suggests that our external experiences are encoded as “cognitive maps” within the brain, organizing perceived entities by establishing relationships between mental representations.

Research Source and Author Information

This paper is a research article published in volume 233 of “Progress in Neurobiology” in 2024, titled “Predictable Navigation through Spontaneous Brain States with Cognitive-Map-Like Representations.” The paper is co-authored by Li Siyang, Li Zhipeng, Liu Qiuyi, Ren Peng, Sun Lili from the School of Life Science and Technology at Harbin Institute of Technology; Cui Zaixu from the Chinese Institute for Brain Science; Li Siyang from the Artificial Intelligence Research Institute at Zhejiang Lab; and Liang Xia from the Key Laboratory of Space Environment and Physical Sciences at Harbin Institute of Technology. The paper was published online on January 15, 2024, and is accessible through Elsevier Ltd. under the CC BY-NC license.

Detailed Research Process

Dataset and Experimental Design

The research utilized two datasets. The first is the “Midnight Scan Club (MSC)” dataset, which includes approximately 5 hours of resting-state functional magnetic resonance imaging (fMRI) data from 10 subjects. Secondly, the “Human Connectome Project (HCP)” dataset was used, containing resting-state 7T fMRI data from 174 subjects, with each subject having four 15-minute scans.

Identification of Cognitive Map Brain Networks

Through terminological analysis, we identified 132 terms closely related to “navigation” and generated brain activation maps for these terms. These activation maps were reduced to a 2D space using t-SNE and clustered into seven brain networks through k-means clustering: Hippocampal and Posterior Medial Network (HPC-PMN), Dorsal Visual Stream Network (dVIS), Ventral Visual Stream Network (vVIS), Frontoparietal Network (FPN), Dorsal Attention Network (DAN), Somatosensory Motor Network (SMN), and Auditory Network (AUD).

Individual Navigation Network Decomposition during Resting-State

Next, using Connectivity-Based Decomposition (CBD) algorithm to analyze resting-state data, we found that the navigation networks during resting-state resemble those during task states. Combining these seven networks with the most navigation-related brain regions, a navigation brain mask was generated, and individual-level module decomposition yielded seven networks at the individual level.

Estimation of Dynamic Brain States

Using Hidden Markov Model (HMM), we revealed three recurring brain states in resting-state fMRI, corresponding to discrete brain network activity patterns. State 1 showed higher activation in HPC and Default Mode Network (DMN), while State 2 showed higher activity in the primary sensory cortex and hippocampus.

“Dwelling” and “Transition” in Spontaneous Brain States

To study “dwelling” and “transition” in spontaneous brain states, we used pattern similarity analysis and t-SNE to embed the occurrence frequencies of different brain states into a 2D space. K-means clustering analysis revealed that the main brain states of State 1 and State 2 could be divided into multiple clusters (“dwelling”), with these “dwellings” being relatively stable in 2D space, showing gradual changes and transitions in brain activity during spontaneous states.

Place Cell-Like Representations and Predictive Transitions

Based on the pattern similarity matrix, we calculated each brain state’s successor representation (SR) matrix and analyzed its receptive field. We found that the gradient of this matrix displays a typical “place cell”-like pattern, with a backward skew in the 2D space. Moreover, by analyzing each brain state’s receptive fields, we identified predictive transitions in spontaneous brain states, showing linear trajectories toward the predicted final states in the brain state space.

Correlation between Predictive Transitions and Cognitive and Emotional Abilities

Finally, we explored the relationship between predictive transitions in spontaneous brain states and individual cognitive and emotional abilities. Results showed that in the internally-guided State 1, higher predictability correlated positively with better fluid abilities (e.g., episodic memory and working memory), while in the externally-guided State 2, higher predictability correlated with poorer executive function and inhibitory control. These findings suggest that the predictive characteristics of spontaneous brain activity are closely related to individual cognitive and emotional traits.

Conclusions and Significance

This study reveals cognitive map-like representations for each state in spontaneous brain activity, indicating that the predictive transitions of these states indeed show significant directionality in brain state space. The research not only supports the cognitive significance of spontaneous brain activity but also proposes a new viewpoint of cognitive maps as a unifying framework. This discovery provides a new perspective on understanding the nature of spontaneous brain activity, emphasizing the importance of spontaneous brain states in cognitive and emotional functioning.

Research Highlights

1. Dynamic Brain States: The study uncovers two main dynamic brain states during resting-state, corresponding to internally and externally directed brain activities.
2. Place Cell-Like Representations: It is the first to demonstrate that spontaneous brain states have cognitive map-like “place cell” representations, exhibiting typical gradient fields in 2D space.
3. Predictive Transitions: Significant backward skew was found in spontaneous brain state transitions, displaying characteristics of predictive transitions in time and space.
4. Cognitive and Emotional Correlations: Predictive brain state transitions show significant correlations with individual cognitive and emotional abilities, further supporting the role of spontaneous brain activity in everyday cognitive and emotional regulation.

Future Research Directions

While this study provides new insights into spontaneous brain activity, future research should further combine real-time surveys during resting-state scans to decode the specific psychological content of spontaneous brain states. Additionally, exploring whether similar representational codes exist in other brain regions and how these codes vary across different mental states will be an important direction for future studies.