Probable nature of higher-dimensional symmetries underlying mammalian grid-cell activity patterns

Alexander Mathis, Martin Stemmler, Andreas V. M. HerzeLife 2015 · DOI

TL;DR

This paper uses information theory to predict the optimal spatial arrangement of neural firing fields for "grid cells," which are thought to represent an animal's location. They find that hexagonal lattices are optimal for 2D environments, and face-centered cubic (FCC) lattices are optimal for 3D environments, suggesting a fundamental link between efficient packing in nature and neural coding.

What Problem Does This Paper Address?

The brain needs to represent an animal's location in space, and a specific type of neuron called a "grid cell" is believed to play a crucial role in this. Grid cells exhibit a remarkable property: their firing fields (the areas where they become active) form a repeating pattern, often described as a hexagonal lattice. While this hexagonal pattern has been observed in 2D environments (like a flat surface), animals often navigate in three dimensions. This raises a fundamental question: what is the optimal arrangement of grid cell firing fields in 3D and potentially even higher-dimensional spaces to maximize the precision of spatial representation? The authors aim to answer this by applying information theory to determine which lattice structures provide the highest spatial resolution for neural population codes.

Methods

The core of the paper's methodology lies in using information theory, specifically the Fisher Information (FI), to quantify the resolution of a neural population code. They model grid cells as having periodic tuning curves, meaning their firing rate repeats at regular intervals dictated by an underlying lattice.

  1. Lattice Construction: They define a grid cell's tuning curve by periodifying a basic "bump" function (Ω) over a lattice (L). This means the bump function's value at a certain distance from the cell's "center" is repeated across the entire space according to the lattice structure.
  2. Fisher Information Calculation: They derive the Fisher Information matrix for a population of such grid cells. The inverse of the Fisher Information provides a lower bound on the covariance of the estimated stimulus (i.e., the animal's location), thus quantifying the resolution of the code.
  3. Packing Ratio Connection: A key theoretical step is showing that maximizing the Fisher Information for a population of grid cells is equivalent to maximizing the packing ratio of the underlying lattice. The packing ratio measures how densely spheres (representing the effective "coverage" of a firing field) can be packed within a given space.
  4. Dimensionality Analysis: They extend this analysis from 2D to higher dimensions, comparing different lattice structures (e.g., square, hexagonal, face-centered cubic, body-centered cubic) and their corresponding packing ratios.

Datasets

Not discussed in the paper. The study is theoretical and relies on mathematical modeling and simulations rather than experimental datasets.

Key Insights and Ideas

  • Nature's Efficiency Mirrors Brain's: The most striking insight is the direct link between optimal geometric packings found in nature (like how spheres arrange themselves) and the hypothesized neural codes for spatial representation. The brain appears to leverage fundamental principles of efficient space-filling.
  • Lattices are Key to Resolution: The paper rigorously demonstrates that the spatial resolution of a grid cell population code is directly proportional to the packing ratio of the underlying lattice. Denser lattices lead to more precise spatial information.
  • Hexagonal for 2D, FCC for 3D: This is the central prediction. For 2D environments, the hexagonal lattice, known for its optimal packing, is predicted to be the best for grid cells. For 3D, the face-centered cubic (FCC) lattice, which is the densest packing in 3D, is predicted to be optimal.
  • Beyond Physical Space: The authors hypothesize that this principle extends beyond just representing physical space. Other high-dimensional sensory or cognitive variables might also be encoded using similar lattice structures.
  • Nested Codes for Ambiguity Resolution: The paper touches upon how multiple grid modules with different spatial scales (nested codes) can overcome the inherent ambiguity of periodic firing patterns, leading to a more robust and refined representation.

Key Results

  • 2D Optimization: The hexagonal lattice provides approximately 15.5% better spatial resolution than a square lattice for 2D grid cell representations, consistent with their respective packing ratios.
  • 3D Prediction: The face-centered cubic (FCC) lattice is predicted to be the optimal arrangement for grid cell firing fields in 3D environments. This FCC lattice provides about 40% more resolution than a simple cubic lattice and 8.8% more than a body-centered cubic lattice.
  • Non-Lattice Packings: In 3D, highly symmetric non-lattice packings like hexagonal close packing (HCP) can achieve the same optimal resolution as FCC lattices.
  • Simulation Validation: Simulations with a finite number of neurons (M=200) showed that the theoretical predictions held, although with some variability, indicating that the FCC lattice generally outperformed others even in smaller populations.

Limitations and Future Work

  • Isotropic Space Assumption: The study assumes an isotropic and homogeneous space. Future work should explore how complex or anisotropic environments (e.g., caves, burrows) might influence optimal grid cell arrangements.
  • Radially Symmetric Tuning: The analysis often assumes radially symmetric tuning curves. However, the authors acknowledge that other tuning shapes (e.g., elliptical) might outperform radial ones in certain scenarios, and experimentally observed tuning curves are not always isotropic.
  • Decoder's Role: While the paper optimizes for resolution, it doesn't fully address how a biological decoder would actually extract this information. The authors suggest that whether a realistic decoder favors one optimal packing solution over another is an open question.
  • Experimental Verification: The predictions for 3D grid cell arrangements (FCC or HCP) are experimentally testable in animals that navigate in 3D, such as flying bats or climbing animals.
  • Broader Applications: The hypothesis that lattice codes extend to other high-dimensional cognitive spaces warrants further investigation.

Summary updated

Abstract

Lattices abound in nature-from the crystal structure of minerals to the honey-comb organization of ommatidia in the compound eye of insects. These arrangements provide solutions for optimal packings, efficient resource distribution, and cryptographic protocols. Do lattices also play a role in how the brain represents information? We focus on higher-dimensional stimulus domains, with particular emphasis on neural representations of physical space, and derive which neuronal lattice codes maximize spatial resolution. For mammals navigating on a surface, we show that the hexagonal activity patterns of grid cells are optimal. For species that move freely in three dimensions, a face-centered cubic lattice is best. This prediction could be tested experimentally in flying bats, arboreal monkeys, or marine mammals. More generally, our theory suggests that the brain encodes higher-dimensional sensory or cognitive variables with populations of grid-cell-like neurons whose activity patterns exhibit lattice structures at multiple, nested scales.
BibTeX
@inproceedings{mathis2015probable,
  author = {Alexander Mathis and Martin Stemmler and Andreas V. M. Herz},
  title = {{Probable nature of higher-dimensional symmetries underlying mammalian grid-cell activity patterns}},
  booktitle = {eLife},
  year = {2015},
  doi = {10.7554/elife.05979},
}
Figure