Flexible coding of memory and space in the primate hippocampus during virtual navigation

Roberto A. Gulli, Lyndon Duong, Benjamin Corrigan, Guillaume Doucet, Sylvain Williams, Stefano Fusi, Julio Martínez-TrujillobioRxiv 2018 · DOI

TL;DR

The primate hippocampus dynamically reconfigures its spatial representations based on task demands, integrating non-spatial sensory and mnemonic information. This suggests the hippocampus acts as a flexible memory system, not just a static spatial map, crucial for learning and adapting to complex environments.

What Problem Does This Paper Address?

The hippocampus is famously known for its role in spatial navigation, with "place cells" firing when an animal is in a specific location. However, it's also critical for memory. How these two functions interact, and how hippocampal representations of space change when the task demands shift, remains unclear. Specifically, do spatial maps remain stable across different cognitive tasks, or do they adapt to incorporate task-relevant information? This paper investigates this by recording hippocampal activity in monkeys performing two distinct tasks in the same virtual environment.

Methods

The researchers recorded the activity of individual hippocampal neurons in two male macaques while they navigated a virtual reality environment (X-Maze). They employed two main tasks:

  • Foraging Task: Monkeys navigated to find a randomly placed reward. This task primarily relied on cue-guided navigation.
  • Associative Memory Task: Monkeys had to learn a context-dependent reward hierarchy involving object-color associations. This task required more complex cognitive processing beyond simple navigation.

Key analytical techniques included:

  • Place Selectivity Analysis: Binning the virtual environment into pixels and analyzing neuron firing rates to identify "place fields."
  • Spatial Information Content (SIC): Quantifying how much information a neuron's firing rate provides about the animal's location.
  • Population Decoding: Using a linear classifier (Support Vector Machine) to predict the monkey's position in the maze based on the collective activity of hippocampal neurons. This was done within each task and also across tasks to assess generalization.
  • Non-spatial Feature Encoding: Regressing neuron firing rates against trial-specific features (context, object color) during different task periods to understand how non-spatial information is encoded.

Datasets

The study used data collected from two male macaques performing tasks in a custom-built virtual reality environment. No new benchmarks or datasets were introduced.

Key Insights and Ideas

  • Task-Dependent Spatial Reorganization: Hippocampal spatial maps are not static. They change significantly between the foraging and associative memory tasks, even within the same environment.
  • Beyond Pure Spatial Mapping: The changes in spatial coding are not solely due to remapping of place fields. Instead, they are driven by neurons encoding non-spatial features relevant to the associative memory task.
  • Sensory and Mnemonic Integration: Hippocampal neurons encode both currently visible (sensory) and remembered (mnemonic) features of the environment and task. This integration is crucial for flexible behavior.
  • Flexible Neural Substrate: The hippocampus appears to provide a single, adaptable neural substrate that supports both spatial navigation and associative memory by dynamically incorporating relevant sensory and mnemonic information.

Key Results

  • Spatial Response Fields Change: While individual neurons had spatially-selective response fields in both tasks, the population code did not generalize across tasks.
  • Increased Spatial Information in Memory Task: 70% of neurons showed spatial response fields in the associative memory task, compared to 55.7% in the foraging task. These fields were also less homogeneously distributed in the memory task.
  • Poor Cross-Task Decoding: A classifier trained to predict position in the foraging task performed poorly when tested on the associative memory task, and vice-versa, indicating a lack of generalized spatial representation.
  • Encoding of Non-Spatial Features: A significant proportion of neurons showed selectivity for non-spatial features (object color, context) during specific periods of the associative memory task, even when these features were not immediately visible (mnemonic encoding).
  • Trial Type Encoding: The population of hippocampal neurons could reliably predict the type of associative memory trial (context and object pair) based on firing rates.

Limitations and Future Work

  • Limited Sample Size: Data was collected from only two monkeys. Future work could expand this to increase statistical power and generalizability.
  • Focus on CA3: Recordings were primarily from the CA3 region of the hippocampus. Investigating other hippocampal subfields could provide a more complete picture.
  • Mechanisms of Remapping: While the study identified changes in spatial coding and linked them to non-spatial feature encoding, the precise mechanisms by which these changes occur and the thresholds for different types of remapping (global, partial, local) remain unclear. Future research could explore these dynamic processes in more detail.
  • Theoretical Framework: The authors suggest their findings align with theoretical models of the hippocampus as a memory system that compresses sensory information and predicts future states, rather than solely a "Global Positional System." Further theoretical development and experimental testing of these models would be valuable.

Summary updated

Abstract

Abstract Hippocampal maps of space change across tasks. The mechanisms of this effect remain unclear. To examine this, we recorded activity of hippocampal neurons in monkeys navigating the same virtual maze during two different tasks: a foraging task requiring only cue guided navigation, and a memory task also requiring context-object association. Within each task, individual neurons had spatially-selective response fields, enabling a linear classifier to decode position in the virtual environment in each task. However, the population code did not generalize across tasks. This was due to sensory and mnemonic coding of non-spatial features and their associations by single neurons during each period of the associative memory task. Thus, sensory and mnemonic representations of non-spatial features shape maps of space in the primate hippocampus during virtual navigation. This may reflect a fundamental role of the hippocampus in compressing information from a variety of sources for efficient memory storage.
BibTeX
@misc{gulli2018flexible,
  author = {Roberto A. Gulli and Lyndon Duong and Benjamin Corrigan and Guillaume Doucet and Sylvain Williams and Stefano Fusi and Julio Martínez-Trujillo},
  title = {{Flexible coding of memory and space in the primate hippocampus during virtual navigation}},
  year = {2018},
  doi = {10.1101/295774},
  howpublished = {bioRxiv},
}
Figure