Pre-stimulus phase and amplitude regulation of phase-locked responses are maximized in the critical state

Arthur-Ervin Avrămiea, Richard Hardstone, Jan-Matthis Lueckmann, Jan Bím, Huibert D. Mansvelder, Klaus Linkenkaer‐HanseneLife 2020 · DOI

TL;DR

This paper uses a computational model to show that the brain's ability to flexibly adjust its response to stimuli based on pre-stimulus neural activity is maximized when the network operates in a "critical state." This state, characterized by balanced excitation and inhibition and scale-free fluctuations, allows for optimal processing of a wide range of stimulus intensities and is crucial for both amplitude and phase-based regulation of neural responses.

What Problem Does This Paper Address?

A fundamental question in neuroscience is why identical sensory stimuli can lead to vastly different neural responses and subjective perceptions. While it's known that ongoing neural oscillations, particularly alpha rhythms, influence how incoming signals are processed through their amplitude and phase, it's unclear how these effects are modulated by the brain's overall dynamic state. Specifically, researchers wanted to understand if the brain's ability to regulate stimulus processing based on its state before a stimulus arrives is linked to its global dynamics, particularly if it operates near a critical point. This is important because understanding this link could explain how the brain achieves flexible and adaptive sensory processing.

Methods

The researchers employed a computational model called the Critical Oscillations (CROS) model, which simulates spiking neurons and exhibits dynamics akin to human EEG data, including alpha oscillations and long-range temporal correlations (LRTCs). They systematically varied the balance of excitatory and inhibitory connections within the network to explore different dynamical states: subcritical (stable, predictable), critical (balanced, scale-free), and supercritical (chaotic, unstable).

Key analyses included:

  • Dynamic Range: Measuring how well the network's phase-locking response (PLF) to a stimulus varied across different stimulus sizes.
  • Pre-stimulus Amplitude Regulation: Assessing how the amplitude of ongoing alpha oscillations before a stimulus influenced the subsequent phase-locking response.
  • Pre-stimulus Phase Regulation: Investigating how the phase of ongoing alpha oscillations before a stimulus affected the phase-locking response.
  • Detrended Fluctuation Analysis (DFA): Quantifying the degree of criticality by measuring LRTCs in the oscillatory activity.

They also explored the robustness of their findings by altering synaptic weights, which shifts the critical state within the parameter space.

Datasets

The study utilized a computational model and generated synthetic data. No external datasets were used for training or evaluation. The generated data, including simulation code and scripts, is made publicly available.

Key Insights and Ideas

  • Criticality is Key for Versatility: The most striking finding is that the brain's ability to dynamically adjust its response to stimuli, both based on pre-stimulus amplitude and phase, is significantly enhanced when the network operates in a critical state.
  • Beyond Dynamic Range: While criticality is known to maximize dynamic range in non-oscillatory systems, this study extends this to oscillatory networks, showing it's crucial for phase-locking responses as well.
  • Amplitude and Phase Regulation: Critical networks exhibit robust regulation of stimulus processing based on both the amplitude and phase of pre-stimulus alpha oscillations, allowing for fine-tuning of sensory input. Subcritical and supercritical networks lack this flexible regulation.
  • Stimulus Intensity Matters: Critical networks can differentiate responses across the widest range of stimulus intensities, suggesting an optimal balance for processing varying levels of sensory information.
  • Generic Property: The relationship between criticality and network versatility appears to be a fundamental property, robust to changes in specific model parameters like synaptic weights.

Key Results

  • Maximized Dynamic Range: Critical networks showed the widest dynamic range of phase-locking responses across different stimulus sizes compared to subcritical and supercritical networks.
  • Pre-stimulus Amplitude Regulation: Critical networks demonstrated strong negative pre-stimulus amplitude regulation, meaning higher pre-stimulus alpha amplitude led to weaker stimulus-evoked responses. This effect was significantly correlated with LRTCs (a measure of criticality). Subcritical and supercritical networks showed no significant pre-stimulus amplitude regulation.
  • Pre-stimulus Phase Regulation: Similar to amplitude regulation, pre-stimulus phase dependence of phase-locking response was found to require a balance in E/I connectivity and was significantly correlated with criticality. Critical networks showed this regulation over the widest range of stimulus sizes.
  • Robustness: Modifying synaptic weights to shift the critical line in the parameter space did not alter the fundamental relationship between criticality and enhanced network versatility (dynamic range, amplitude, and phase regulation).

Limitations and Future Work

The authors acknowledge that their computational model is an abstraction of complex biological neuronal mechanisms. While it captures essential features like alpha oscillations and LRTCs, future work could explore more physiologically realistic models, including those with balanced excitatory and inhibitory currents.

They also suggest that future studies should investigate whether oscillations are essential for pre-stimulus phase and amplitude regulation, and explore how different types of critical phase transitions might influence these effects. Furthermore, they propose that the brain might leverage different dynamical states (critical, subcritical, supercritical) in a task-dependent manner, or combine networks operating at different distances from criticality, to achieve even broader processing capabilities. The findings also open avenues for experimental testing using non-invasive techniques like EEG/MEG to examine the criticality of sensory cortices and its link to stimulus processing.

Summary updated

Abstract

Understanding why identical stimuli give differing neuronal responses and percepts is a central challenge in research on attention and consciousness. Ongoing oscillations reflect functional states that bias processing of incoming signals through amplitude and phase. It is not known, however, whether the effect of phase or amplitude on stimulus processing depends on the long-term global dynamics of the networks generating the oscillations. Here, we show, using a computational model, that the ability of networks to regulate stimulus response based on pre-stimulus activity requires near-critical dynamics-a dynamical state that emerges from networks with balanced excitation and inhibition, and that is characterized by scale-free fluctuations. We also find that networks exhibiting critical oscillations produce differing responses to the largest range of stimulus intensities. Thus, the brain may bring its dynamics close to the critical state whenever such network versatility is required.
BibTeX
@inproceedings{avramiea2020pre,
  author = {Arthur-Ervin Avrămiea and Richard Hardstone and Jan-Matthis Lueckmann and Jan Bím and Huibert D. Mansvelder and Klaus Linkenkaer‐Hansen},
  title = {{Pre-stimulus phase and amplitude regulation of phase-locked responses are maximized in the critical state}},
  booktitle = {eLife},
  year = {2020},
  doi = {10.7554/elife.53016},
}
Figure