Naturalistic coding of working memory in primate prefrontal cortex

Megan Roussy, Rogelio Luna, Lyndon Duong, Benjamin Corrigan, Roberto A. Gulli, Ramon Nogueira, Rubén Moreno‐Bote, Adam Sachs, Lena Palaniyappan, Julio Martínez-TrujillobioRxiv 2020 · DOI

TL;DR

Primate prefrontal cortex (LPFC) neurons robustly encode working memory representations even in naturalistic, unconstrained environments. Disrupting the balance between excitatory and inhibitory neurons with ketamine impairs working memory performance, highlighting the critical role of this balance for cognitive function.

What Problem Does This Paper Address?

The lateral prefrontal cortex (LPFC) is widely accepted as crucial for working memory (WM), the ability to temporarily hold and manipulate information. While studies have shown LPFC activity encodes WM representations, these experiments often used simplified visual displays and strictly controlled eye movements. This raises a critical question: can LPFC circuits support WM in more naturalistic settings, where behavior is dynamic and gaze is unconstrained? This paper aims to bridge this gap by investigating WM in macaque monkeys performing a task in a life-like virtual environment with free gaze.

Methods

The researchers trained two rhesus monkeys to perform a visuospatial working memory task in a virtual reality environment. This environment mimicked naturalistic scenes, allowing for unconstrained eye movements and navigation via joystick.

Key methodological aspects include:

  • Virtual Reality Environment: A realistic 3D virtual arena was created to simulate naturalistic visual scenes, moving beyond simplified displays.
  • Unconstrained Gaze: Unlike previous studies, monkeys were allowed to freely explore the environment with their gaze, mimicking natural behavior.
  • Working Memory Task: A target was presented, then disappeared, followed by a delay period, after which the monkeys had to navigate to the remembered location.
  • Neural Recordings: High-density microelectrode arrays (Utah Arrays) were implanted in the LPFC to record neuronal activity.
  • Pharmacological Manipulation: Low doses of ketamine, an NMDA receptor antagonist, were administered to investigate its causal role in LPFC function and WM. A control task (ketamine-perception variant) was used to differentiate WM-specific effects from general perceptual or motor impairments.
  • Data Analysis:
    • Behavioral Performance: Measured by percent correct trials and response time.
  • Single Neuron Tuning: Assessed the spatial selectivity of individual neurons.
  • Population Decoding: Used Support Vector Machines (SVM) to decode target locations from neuronal ensemble activity.
  • Gaze Decoding: Analyzed eye movement data to understand its contribution to target representation.
  • Statistical Properties of Neural Decoding: Investigated Population Signal (PS) and Projected Precision (PP) to understand how information is encoded.
  • Cell Type Specific Analysis: Differentiated between narrow spiking (putative inhibitory) and broad spiking (putative excitatory) neurons to understand their distinct roles.

Datasets

The study utilized data collected from two adult male rhesus macaques performing the virtual reality working memory task. No external datasets were used for training or evaluation. The task itself, conducted in a novel virtual environment, can be considered a new benchmark for studying WM in more naturalistic conditions.

Key Insights and Ideas

  • LPFC Excels in Naturalistic Settings: Contrary to concerns that simplified tasks might overestimate LPFC's WM capacity, this study demonstrates that LPFC neurons robustly encode WM representations even with unconstrained gaze in a complex virtual environment. This suggests LPFC's role in WM is more generalizable to real-world behavior than previously thought.
  • Balance is Key: The findings strongly suggest that WM relies on a delicate balance between excitatory and inhibitory neuronal activity. Ketamine's disruption of this balance, rather than simply impairing excitation or inhibition, is central to its effects.
  • Cell-Type Specific Roles: Ketamine differentially affects narrow spiking (inhibitory) and broad spiking (excitatory) neurons. Narrow spiking neurons show a loss of tuning due to decreased firing at preferred locations, while broad spiking neurons show a loss of tuning due to increased firing at non-preferred locations. This highlights distinct contributions of different neuronal populations.
  • Dissociation of Gaze and Neural Codes: While eye position can carry some information about target location, the study shows that neuronal firing rates carry significantly more information. Furthermore, ketamine's impact on WM decoding is not explained by changes in gaze bias, suggesting a dissociation between eye movements and LPFC's internal WM representations.

Key Results

  • Behavioral Impairment by Ketamine: Low doses of ketamine significantly impaired working memory performance (percent correct) while sparing perceptual and motor skills, as evidenced by the ketamine-perception control task.
  • Neuronal Tuning Disruption: Ketamine led to a significant decrease in the proportion of spatially tuned neurons during the early post-injection period, which recovered later.
  • Population Decoding Impairment: Decoding accuracy of remembered locations from neuronal ensembles significantly decreased after ketamine injection, particularly during the delay and response epochs.
  • Cell Type Specific Firing Rate Changes:
    • Narrow spiking neurons showed decreased firing at preferred locations after ketamine.
  • Broad spiking neurons showed increased firing at non-preferred locations after ketamine.
  • PS and PP Modulation: Ketamine significantly decreased Population Signal (PS), indicating reduced tuning and modulation across target locations, while Projected Precision (PP) showed more complex changes.

Limitations and Future Work

  • Limited Number of Subjects: The study involved only two rhesus monkeys. While results were consistent, replication with more subjects would strengthen the findings.
  • Generalizability of Virtual Environment: While more naturalistic than previous setups, the virtual environment is still a controlled setting. Future work could explore WM in even more complex, real-world scenarios.
  • Ketamine's Broader Effects: While the study focused on LPFC, ketamine has widespread effects on the brain. Future research could investigate how these broader effects interact with LPFC function.

The authors suggest that understanding the precise mechanisms by which ketamine disrupts the excitatory-inhibitory balance in LPFC could offer insights into cognitive deficits observed in conditions like schizophrenia, which are characterized by LPFC abnormalities and NMDA receptor hypofunction. Further investigation into how this balance is maintained and how its disruption leads to specific cognitive impairments is a promising avenue for future research.

Summary updated

Abstract

Summary The primate lateral prefrontal cortex (LPFC) is considered fundamental for temporarily maintaining and manipulating mental representations that serve behavior, a cognitive function known as working memory 1 . Studies in non-human primates have shown that LPFC lesions impair working memory 2 and that LPFC neuronal activity encodes working memory representations 3 . However, such studies have used simple displays and constrained gaze while subjects held information in working memory 3 , which put into question their ethological validity 4,5 . Currently, it remains unclear whether LPFC microcircuits can support working memory function during natural behavior. We tested macaque monkeys in a working memory navigation task in a life-like virtual environment while their gaze was unconstrained. We show that LPFC neuronal populations robustly encode working memory representations in these conditions. Furthermore, low doses of the NMDA receptor antagonist, ketamine, impaired working memory performance while sparing perceptual and motor skills. Ketamine decreased the firing of narrow spiking inhibitory interneurons and increased the firing of broad spiking cells reducing population decoding accuracy for remembered locations. Our results show that primate LPFC generates robust neural codes for working memory in naturalistic settings and that such codes rely upon a fine balance between the activation of excitatory and inhibitory neurons.
BibTeX
@misc{roussy2020naturalistic,
  author = {Megan Roussy and Rogelio Luna and Lyndon Duong and Benjamin Corrigan and Roberto A. Gulli and Ramon Nogueira and Rubén Moreno‐Bote and Adam Sachs and Lena Palaniyappan and Julio Martínez-Trujillo},
  title = {{Naturalistic coding of working memory in primate prefrontal cortex}},
  year = {2020},
  doi = {10.1101/2020.06.19.162446},
  howpublished = {bioRxiv},
}
Figure