Reduced variability of bursting activity during working memory

Mikael Lundqvist, Jonas Rose, Scott L. Brincat, Melissa R. Warden, Timothy J. Buschman, Pawel Herman, Earl K. MillerScientific 2022 · DOI

TL;DR

This paper challenges the traditional view of working memory relying on persistent neuronal firing. Instead, it provides evidence that intermittent bursts of neural activity, specifically gamma bursts, are key. Crucially, the authors demonstrate that task-related reductions in spike variability are directly linked to, and likely explained by, reductions in the variability of these gamma bursts, offering a new mechanistic understanding of cognitive control.

What Problem Does This Paper Address?

For decades, the prevailing model of working memory (WM) has been that neurons maintain information by firing persistently. However, recent advancements in recording technology have revealed that neuronal activity is often characterized by intermittent bursts rather than continuous firing, especially when examined on a trial-by-trial basis. This observation has led to a debate: if WM relies on intermittent bursts, how can we reconcile this with findings that neuronal spike-time variability decreases during task performance? Some researchers have argued that intermittent bursting should increase variability, posing a challenge to burst-based WM models. This paper aims to resolve this apparent contradiction by investigating whether task-related reductions in spike variability can coexist with intermittent spiking, and if so, what the underlying mechanism might be.

Methods

The researchers analyzed electrophysiological data (Local Field Potentials - LFPs and single-unit spiking) from the prefrontal cortex (PFC) of monkeys performing three distinct working memory tasks.

Key methodological aspects include:

  • Multi-electrode recordings: This allowed for the simultaneous recording of activity from many neurons and LFPs.
  • Single-trial analysis: Crucial for observing the intermittent nature of spiking, which is often obscured by averaging across trials.
  • Fano Factor (FF) calculation: Used to quantify spike-time variability across trials. The FF was calculated for both spiking activity and gamma burst events.
  • Gamma burst identification: LFPs were analyzed to identify bursts of gamma-band power, which were then linked to spiking activity.
  • Synthetic spike train generation: To directly test the hypothesis that gamma burst variability explains spike variability, the researchers created artificial spike trains. These synthetic trains were generated using the recorded gamma burst timings and estimated firing rates, allowing them to isolate the effect of burst timing on variability.

Datasets

The study utilized existing datasets from previous research (Lundqvist et al. and Warden & Miller) involving non-human primates performing three working memory tasks (Task 1, Task 2, and Task 3). These tasks involved remembering sequences of visual stimuli. No new datasets were introduced.

Key Insights and Ideas

  • Intermittent spiking is not at odds with reduced variability: The core finding is that task performance leads to reduced spike-time variability, even though the underlying activity is intermittent and bursty. This directly counters the argument that intermittent bursting inherently increases variability.
  • Gamma bursts are the key modulator: The reduction in spike variability is not random; it is tightly linked to the modulation of gamma bursts. Specifically, when gamma bursts are more consistently timed (i.e., have reduced variability), spike variability also decreases.
  • Task-modulation is crucial: The bursts of spiking and gamma power are not random but are modulated by task events (stimulus presentation, delays, test phases). This task-related modulation is what drives the reduction in variability.
  • Mechanistic link established: The synthetic spike train generation provides strong evidence that the timing of gamma bursts, rather than just the overall rate of spiking, is the primary driver of reduced spike variability.
  • Support for "synaptic attractor" models: These findings bolster models of WM that propose information is held not just by persistent firing but by a combination of brief spiking bursts and short-term synaptic plasticity that bridges these bursts.

Key Results

  • Reduced Fano Factor (FF) during task: Across all three tasks, the Fano Factor for both spiking activity and gamma bursts decreased during task performance compared to the pre-stimulus baseline. This indicates reduced trial-to-trial variability.
  • Correlation between spike and gamma burst FF reduction: The reduction in spike FF was strongly correlated with the reduction in gamma burst FF. This effect was particularly pronounced at sites where gamma bursting was most modulated by the task.
  • Synthetic data mimics real data: Synthetic spike trains generated based on recorded gamma burst timings and firing rates showed a similar reduction in FF as the actual recorded data, demonstrating that gamma burst variability can largely explain the observed decrease in spike variability.
  • Higher gamma bursting, greater variability reduction: Sites exhibiting stronger task-related increases in gamma bursting also showed a greater reduction in both spike and gamma burst variability.

Limitations and Future Work

  • Model assumptions: While the synthetic spike train generation strongly supports the findings, the authors acknowledge that individual neurons have other sources of variability not captured by population bursts (e.g., stimulus preference, non-Poissonian firing).
  • Generalizability: The study was conducted in monkeys' PFC. While this is a key area for WM, future work could explore if these principles apply to other brain regions or species.
  • Future directions: The authors suggest that this framework of discrete, packet-based activity, modulated by rhythmic processes like gamma bursts, could be a fundamental principle of cortical processing, facilitating time multiplexing and inter-areal communication. Further research could explore how these mechanisms contribute to other cognitive functions beyond WM.

Summary updated

Abstract

Working memories have long been thought to be maintained by persistent spiking. However, mounting evidence from multiple-electrode recording (and single-trial analyses) shows that the underlying spiking is better characterized by intermittent bursts of activity. A counterargument suggested this intermittent activity is at odds with observations that spike-time variability reduces during task performance. However, this counterargument rests on assumptions, such as randomness in the timing of the bursts, which may not be correct. Thus, we analyzed spiking and LFPs from monkeys' prefrontal cortex (PFC) to determine if task-related reductions in variability can co-exist with intermittent spiking. We found that it does because both spiking and associated gamma bursts were task-modulated, not random. In fact, the task-related reduction in spike variability could largely be explained by a related reduction in gamma burst variability. Our results provide further support for the intermittent activity models of working memory as well as novel mechanistic insights into how spike variability is reduced during cognitive tasks.
BibTeX
@inproceedings{lundqvist2022reduced,
  author = {Mikael Lundqvist and Jonas Rose and Scott L. Brincat and Melissa R. Warden and Timothy J. Buschman and Pawel Herman and Earl K. Miller},
  title = {{Reduced variability of bursting activity during working memory}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-18577-y},
}
Figure