Learning place cells, grid cells and invariances: A unifying model

Simon N. Weber, Henning SprekelerbioRxiv 2017 · DOI

TL;DR

This paper proposes a unified computational model where the interaction of excitatory and inhibitory synaptic plasticity, driven by the statistical properties of incoming signals, can explain the diverse spatial tuning patterns observed in the brain, including place cells, grid cells, and head-direction cells. This mechanism offers a parsimonious explanation for how the brain forms stable spatial representations.

What Problem Does This Paper Address?

The brain, particularly areas like the hippocampus and entorhinal cortex, exhibits a remarkable variety of neurons with specific spatial tuning properties. Some neurons, like place cells, are highly selective to a particular location but invariant to head direction, while others, like head-direction cells, are tuned to direction but not location. Grid cells, on the other hand, show periodic firing patterns across an environment. Understanding the underlying neural mechanisms that generate these distinct yet related representations has been a long-standing challenge. Specifically, it's unclear how a single mechanism could give rise to such diverse tuning profiles, especially the seemingly contradictory combination of selectivity and invariance.

Methods

The authors developed a computational model of a feedforward neural network. A single output neuron receives input from both excitatory and inhibitory neurons. The key innovation lies in the plasticity rules applied to the synaptic weights:

  • Excitatory Synaptic Plasticity: A standard Hebbian learning rule, where synaptic weights increase with simultaneous pre- and postsynaptic activity.
  • Inhibitory Synaptic Plasticity: A homeostatic plasticity rule that adjusts inhibitory weights based on presynaptic activity and the difference between postsynaptic activity and a target firing rate (1 Hz). This rule helps balance excitation and inhibition.

The model simulates an animal navigating an environment, and the synaptic inputs to the output neuron are modulated by the animal's location and/or head direction. The statistical properties of these input modulations (specifically, their spatial smoothness) are systematically varied to see how they influence the emergent firing patterns of the output neuron. They used both simulations and mathematical analysis to explore these dynamics.

Datasets

Not discussed in the paper. The paper relies on simulated data generated by the model.

Key Insights and Ideas

  • Smoothness is Key: The relative spatial smoothness of excitatory versus inhibitory inputs is the critical factor determining the type of spatial representation that emerges.
  • Unified Mechanism: A single plasticity mechanism (interacting excitatory and inhibitory plasticity) can generate a wide spectrum of spatial tuning patterns, from localized place fields to periodic grid-like activity and invariant representations.
  • Inhibition Shapes Space: The inhibitory system plays a crucial role in shaping the spatial organization of neural activity. Smoother inhibitory inputs lead to more periodic patterns (like grid cells), while less smooth or untuned inhibition can lead to localized fields (place cells) or invariance.
  • Combinatorial Variety: By varying the input statistics along different dimensions (e.g., spatial location and head direction), the model can reproduce complex conjunctive cells and head-direction cells.

Key Results

  • Grid Cells: When inhibitory input is smoother than excitatory input, the model generates periodic firing patterns resembling grid cells. The spacing of these grid fields scales with the smoothness of the inhibitory input.
  • Place Cells: When inhibitory input is very smooth or spatially untuned, the model produces single firing fields, analogous to place cells.
  • Invariance: When excitatory input is smoother than inhibitory input, the output neuron exhibits spatially invariant firing.
  • Band Cells and Stretched Grids: By introducing anisotropic smoothness in the inhibitory inputs (different smoothness along different spatial directions), the model can generate band cells (periodic in one direction, invariant in another) and stretched grids.
  • Head Direction Cells: By incorporating head direction tuning in the inputs, the model can reproduce pure head direction cells, conjunctive cells (spatial and directional tuning), and grid cells with head direction modulation.
  • Rapid Emergence: The model demonstrates that these spatial patterns can emerge rapidly, within minutes of exploration, aligning with experimental observations.

Limitations and Future Work

  • Input Origin: The paper acknowledges that the precise origin and nature of the spatially tuned input signals are still debated and could stem from various brain regions.
  • Recurrent Connections: The current model is feedforward. Future work could explore how recurrent connections between output cells might further refine spatial representations or explain phenomena like the coordination of neighboring grid cells.
  • Experimental Predictions: The authors suggest that investigating the tuning of individual cells and synapses projecting to spatially tuned cells could serve as a crucial test for their proposed mechanism.
  • Environmental Geometry: While the model shows some influence of environmental boundaries, further exploration of how complex geometries affect pattern formation could be valuable.

Summary updated

Abstract

Abstract Neurons in the hippocampus and adjacent brain areas show a large diversity in their tuning to location and head direction. The underlying circuit mechanisms are not fully resolved. In particular, it is unclear why certain cell types are selective to one spatial variable, but invariant to another. For example, a place cell is highly selective to location, but typically invariant to head direction. Here, we propose that all observed spatial tuning patterns – in both their selectivity and their invariance – are a consequence of the same mechanism: Excitatory and inhibitory synaptic plasticity that is driven by the spatial tuning statistics of synaptic inputs. Using simulations and a mathematical analysis, we show that combined excitatory and inhibitory plasticity can lead to localized, grid-like or invariant activity. Combinations of different input statistics along different spatial dimensions reproduce all major spatial tuning patterns observed in rodents. The model is robust to changes in parameters, develops patterns on behavioral time scales and makes distinctive experimental predictions. Our results suggest that the interaction of excitatory and inhibitory plasticity is a general principle for the formation of neural representations.
BibTeX
@misc{weber2017learning,
  author = {Simon N. Weber and Henning Sprekeler},
  title = {{Learning place cells, grid cells and invariances: A unifying model}},
  year = {2017},
  doi = {10.1101/102525},
  howpublished = {bioRxiv},
}
Figure