Development of Non-Spatial Grid-Like Neural Codes Predicts Inference and Intelligence

Yukun Qu, Jianxin Ou, L. Pang, Shuqi Wu, Yuejia Luo, Timothy E.j. Behrens, Yunzhe LiubioRxiv 2024 · DOI

TL;DR

This study reveals that the development of grid-like neural codes in the entorhinal cortex (EC) is crucial for inferential reasoning and knowledge assimilation in humans, from childhood to adulthood. These non-spatial codes, akin to a cognitive map, mature with age and predict intelligence measures, offering a mechanistic link between brain development and cognitive abilities.

What Problem Does This Paper Address?

Piaget's theory suggests that children develop structured knowledge schemas to understand and infer information, but the underlying neural mechanisms have remained elusive. Specifically, how the brain builds and utilizes these "knowledge maps" to support reasoning and learning, especially as we age, is not well understood. This paper investigates the neural basis of this developmental process, focusing on how grid-like neural activity, typically associated with spatial navigation, might be repurposed for abstract, non-spatial reasoning.

Methods

The researchers employed a multi-faceted approach combining behavioral tasks and fMRI.

  • Behavioral Tasks: Participants (aged 8-25) engaged in a series of tasks designed to assess:
  • Piecemeal Learning: Learning one-dimensional relationships between abstract "monsters" with attack and defense power attributes.
  • Map-Based Inference (Game 1): Inferring relationships across two dimensions of these attributes, essentially navigating a 2D knowledge map.
  • Knowledge Assimilation (Game 2): Integrating new "monsters" into the existing knowledge map and making inferences involving these new entities.
  • fMRI Analysis: Advanced fMRI techniques were used to detect grid-like codes in the entorhinal cortex (EC) and distance codes in the medial prefrontal cortex (mPFC). These included:
  • Hexagonal Modulation Analysis: Detecting patterns of neural activity consistent with hexagonal grid firing.
  • Hexagonal Consistency Analysis: Assessing how well neural activity aligns with a consistent grid orientation.
  • Representational Similarity Analysis (RSA): Comparing neural patterns to a model of grid-like representation.
  • Whole-Brain Covariate Analysis: Examining how brain activity relates to age, inference performance, and structural connectivity.
  • Structural MRI: Grey matter volume and white matter connectivity were analyzed to understand the structural basis of cognitive map development.
  • Mediation Analysis: Statistical models were used to determine if neural measures mediated the relationship between age and inference abilities.
  • Canonical Correlation Analysis (CCA): To explore the link between cognitive map measures and standardized intelligence scores (IQ).

Datasets

  • Participant Data: fMRI and behavioral data were collected from 231 participants aged 8-25 years.
  • Publicly Available Datasets: The study utilized existing atlases for brain region definitions (e.g., Desikan-Killiany atlas, NeuroVault masks).

Key Insights and Ideas

  • Grid Cells for Abstract Reasoning: The study provides compelling evidence that grid-like neural codes, traditionally thought to be for spatial navigation, are also fundamental for abstract, non-spatial reasoning and knowledge organization in humans. This suggests a powerful repurposing of fundamental neural mechanisms.
  • Schema as a Foundation: A stable, age-matured grid-like schema in the EC acts as a scaffold for building more concrete, environment-specific maps in the mPFC. This schema facilitates efficient assimilation of new information.
  • Age-Dependent Maturation: The strength and coherence of these grid-like codes mature significantly with age, directly correlating with improved inferential reasoning and intelligence. This highlights a critical developmental window.
  • Structural Underpinnings: The development of cognitive maps is supported by structural changes in the EC and mPFC, as well as the maturation of white matter tracts connecting these regions, particularly the cingulum bundle.

Key Results

  • Grid-like codes in EC: Three independent fMRI methods consistently detected significant grid-like codes in the EC, showing hexagonal symmetry.
  • Age-related improvement: Inferential reasoning performance and knowledge assimilation showed significant age-related improvements, even after controlling for memory capacity.
  • EC grid-like code mediates age effect: The EC grid-like code significantly mediated the age-related increase in inferential performance.
  • mPFC distance code: The mPFC showed significant involvement in encoding the Euclidean distance between objects, and this coding also improved with age.
  • Schema consistency aids assimilation: Maintaining a consistent EC grid-like code across original and new maps facilitated better nonlocal inference.
  • Structural connectivity: Age-related increases in EC-mPFC structural connectivity, particularly via the cingulum bundle, were positively correlated with inference performance.
  • Link to IQ: Cognitive map measures, specifically the EC grid-like code and mPFC distance code, significantly predicted reasoning and comprehension scores on standardized IQ tests.

Limitations and Future Work

  • Causality: The cross-sectional design and use of age as an instrumental variable in mediation analyses suggest association but do not definitively prove a causal link between grid code development and reasoning ability.
  • Learning Process: The study did not record neural activity during the learning process itself, making it difficult to fully understand how learning shapes these representations, especially in children.
  • IQ Scope: Reasoning is only one component of intelligence, and IQ encompasses multiple dimensions beyond the scope of this study.

Future work could involve longitudinal designs to clarify the neural trajectory throughout learning and explore other facets of intelligence. Further investigation into the role of replay mechanisms in children with less coherent grid codes is also warranted.

Summary updated

Abstract

Piaget's theory posits that children develop structured knowledge schemas for inferring and assimilating new information, yet the underlying neural mechanisms remain unclear. In 203 participants aged 8-25 years, we investigated how maturation of a two-dimensional (2D) knowledge map underpins inferential reasoning and knowledge assimilation. Grid-cell-like codes in the entorhinal cortex (EC) strengthened with age, reflecting schema representations in non-spatial conceptual spaces, and were directly linked to improved inferential reasoning. These grid-like codes also supported the medial prefrontal cortex (mPFC) in encoding distance relationships between objects on the 2D map. As participants assimilated new information, they integrated it into existing grid patterns in the EC. Moreover, the maturation of these neural codes predicted real-world intelligence measures, particularly in reasoning abilities. Our findings demonstrate that the development of non-spatial grid-like neural codes offers a mechanistic account of cognitive development, bridging psychological theory with a fundamental cellular representation of cognitive map.
BibTeX
@misc{qu2024development,
  author = {Yukun Qu and Jianxin Ou and L. Pang and Shuqi Wu and Yuejia Luo and Timothy E.J. Behrens and Yunzhe Liu},
  title = {{Development of Non-Spatial Grid-Like Neural Codes Predicts Inference and Intelligence}},
  year = {2024},
  doi = {10.1101/2024.11.20.624569},
  howpublished = {bioRxiv},
}
Figure