TL;DR
This paper challenges the long-held assumption that grid cells have uniform firing rates across their spatial fields. By analyzing real neural data, the authors demonstrate significant field-to-field variability in grid cell firing. They then propose a computational model showing how this variability, combined with specific network connectivity, can explain the irregular firing patterns observed in inhibitory interneurons, a key puzzle in grid cell research.
What Problem Does This Paper Address?
Grid cells, found in the medial entorhinal cortex, are crucial for spatial navigation. They fire in a hexagonal grid pattern across an environment, with each "field" representing a specific location. For years, theoretical models of grid cells have largely assumed that the firing rate (or "amplitude") of each of these fields is the same for a given cell. This assumption, however, has led to a contradiction: these models predict that inhibitory interneurons in the same brain region should also exhibit regular, periodic firing patterns, which is not what experimental recordings show. Inhibitory interneurons in the MEC are observed to have much more irregular, aperiodic firing. This discrepancy between model predictions and experimental observations is the central problem this paper aims to solve.
Methods
The authors employed a multi-pronged approach:
- Experimental Data Analysis: They analyzed recordings from 373 grid cells in rats navigating open-field environments. They specifically investigated the firing rate of each individual field within a cell's rate map.
- Statistical Analysis: To quantify field-to-field variability, they compared the firing rates of real grid cell fields to those generated by synthetic spike trains based on idealized, uniform firing fields. They used statistical tests (like the F-statistic) to determine if the observed variability was significantly greater than what would be expected from random fluctuations or experimental noise.
- Computational Modeling: They built a two-population (excitatory-inhibitory) recurrent neural network model. This model incorporated:
- Sparse, Spatially Selective Input: They hypothesized that input to the excitatory grid cells, potentially from other brain areas like the hippocampus, is not uniform but sparse and spatially organized. This input was modeled as "place-like."
- Field-to-Field Variability Generation: They showed that this sparse, place-like input, when combined with appropriate network connectivity, naturally leads to variations in the firing field amplitudes of the simulated grid cells.
- Connectivity Schemes: They explored two main connectivity strategies: a "fully factorized" model with patterned connections and a "partially factorized" model with more random connections between grid cells and interneurons.
- Path Integration: They also tested the models' ability to integrate velocity information for navigation, a key function of grid cells.
Datasets
The study utilized a dataset of 373 experimentally recorded grid cells from rats performing free navigation in square open-field environments (100x100 cm or 150x150 cm). These recordings were previously published by Stensola et al. [30]. No new datasets were introduced.
Key Insights and Ideas
- Grid cells are not uniform: The most striking finding is that real grid cells exhibit significant and stable variations in firing rate across their different spatial fields. This directly contradicts the long-standing assumption of uniform field amplitudes.
- Variability explains interneuron firing: The authors demonstrate that this field-to-field variability in grid cells is crucial for explaining the aperiodic firing patterns observed in inhibitory interneurons. When grid cells have variable field strengths, their combined input to interneurons becomes less regular.
- Sparse, place-like input is key: The model suggests that the source of this variability is likely sparse, spatially selective input to grid cells, possibly originating from hippocampal place cells or other spatially tuned inputs.
- Network architecture matters: The study highlights the importance of the specific connectivity between excitatory grid cells and inhibitory interneurons. A "fully factorized" model with more structured connectivity was better at generating realistic grid cell and interneuron activity, particularly for path integration.
- Remapping potential: The variability in grid cell field amplitudes, even without changing field locations, can lead to remapping in downstream hippocampal place cells, suggesting a broader role for grid cells in encoding contextual information.
Key Results
- Experimental Validation: Statistical analysis of 373 grid cells confirmed significant field-to-field variability in firing rate amplitudes, which was stable and reproducible. This variability was substantially larger than expected from temporal fluctuations.
- Model Success: A two-population excitatory-inhibitory network model successfully reproduced grid cell firing with field-to-field amplitude variability.
- Interneuron Aperiodicity: The model showed that this grid cell variability, coupled with appropriate connectivity, resulted in inhibitory interneurons with aperiodic firing patterns, matching experimental observations.
- Path Integration: The "fully factorized" model demonstrated stable path integration using velocity input, even with a small number of interneurons. The "partially factorized" model struggled with path integration, showing significant drift.
- Remapping: Simulations showed that varying grid cell field amplitudes could induce remapping in a downstream place cell model, even if the grid cell's spatial phase remained unchanged.
Limitations and Future Work
- Source of Input: While the paper proposes sparse, place-like input as a source of variability, the exact biological origin and mechanism of this input are not fully detailed. Future work could explore how this input is generated and how it interacts with the grid cell network.
- Connectivity Generation: The study used two methods (fully and partially factorized) to generate connectivity. Further investigation into the biological plausibility and functional consequences of different connectivity schemes could be valuable.
- Experimental Validation of Model Predictions: The paper suggests specific experimental tests, such as investigating the role of interneurons that project back to grid cells and their firing patterns. Future experimental work could directly test these predictions.
- Adaptation and Learning: The authors briefly mention adaptation models where grid patterns learn and adapt to environments. Integrating these learning mechanisms into their current model could provide a more complete picture of grid cell function.
Summary updated