Source-linked AI summary
Information capacity and transmission are maximized in balanced cortical networks with neuronal avalanches
Woodrow L. Shew, Hongdian Yang, Shan Yu, Rajarshi Roy, Dietmar Plenz
TL;DR
The paper examines how excitation/inhibition balance relates to cortical networks’ information capacity and transmission. Using multisite LFP recordings, experiments, and a computational model, it finds both are maximized near κ=1, where neuronal avalanches emerge.
Problem
Earlier studies addressed entropy maximization, neuronal avalanches, and excitation/inhibition balance separately; this work examines how they converge in cortical dynamics.
Method
The authors analyzed ongoing and stimulus-evoked multisite LFP activity across cortical preparations while varying E/I in vitro and in a computational model.
Results
Entropy and information transmission are maximized at κ=1, the E/I condition under which neuronal avalanches emerge.
Takeaways & Limitations
Operating near κ≈1 may allow cortex to maintain moderate network activity and interactions that maximize information capacity and transmission.
Takeaways & Limitations
Measured entropy remains subject to sample-size effects and spatial structure, although reported sample-size differences were small relative to experiment-to-experiment variability.
Abstract
from arXiv · showhide
The repertoire of neural activity patterns that a cortical network can produce constrains the network's ability to transfer and process information. Here, we measured activity patterns obtained from multi-site local field potential (LFP) recordings in cortex cultures, urethane anesthetized rats, and awake macaque monkeys. First, we quantified the information capacity of the pattern repertoire of ongoing and stimulus-evoked activity using Shannon entropy. Next, we quantified the efficacy of information transmission between stimulus and response using mutual information. By systematically changing the ratio of excitation/inhibition (E/I) in vitro and in a network model, we discovered that both information capacity and information transmission are maximized at a particular intermediate E/I, at which ongoing activity emerges as neuronal avalanches. Next, we used our in vitro and model results to correctly predict in vivo information capacity and interactions between neuronal groups during ongoing activity. Close agreement between our experiments and model suggest that neuronal avalanches and peak information capacity arise due to criticality and are general properties of cortical networks with balanced E/I.
MATERIAL AND METHODS
The study combined multisite LFP recordings, pharmacological E/I manipulation, and computational modeling to quantify cortical activity patterns, entropy, information transmission, and neuronal avalanches.
- Recordings: Multisite MEA recordings measured LFP activity in cortex cultures, anesthetized rats, and awake monkeys using preparation-specific electrode arrays and recording protocols.Culture recordings used a 60-electrode 8x8 array; monkey recordings used a 96-channel 10x10 array; rat recordings used an 8x4 array.
- Event detection: Population events were detected from large negative LFP fluctuations, with nearby negative peaks grouped when their inter-event interval fell below a data-derived threshold.Each event was characterized by peak time, amplitude, and electrode location.
- Avalanche analysis: Neuronal avalanches were assessed by comparing event-size distributions with a reference power-law distribution using the graded statistic κ.κ≈1 indicated agreement with the avalanche reference distribution, whereas positive or negative deviations indicated hyper- or hypo-excitability.
- Pattern representation and entropy: Each population event was encoded as an 8x8 binary pattern, with each bit indicating whether its corresponding electrode was active.Entropy was computed from the probabilities of unique patterns, with additional coarse-binned 4x4 and spatial-subregion analyses.
- Pattern representation and entropy: Entropy H was estimated from observed pattern probabilities and corrected for undersampling using quadratic extrapolation when recording samples were limited.The corrected fit parameter H0 was reported in the results.
- Information measures: Mutual information quantified either information shared between recording sites or information transmission from stimulus S to response R.Site-to-site MI was averaged across all site pairs, while stimulus-response MI was computed as H(R) − H(R|S).
- Interaction analyses: Site participation likelihoods and event shuffling were used to analyze entropy bounds and the contribution of interactions between recording sites.Shuffling randomized each participation vector while preserving each site's likelihood of participation and the number of events.
- Computational model: The computational model represented 16 binary sites whose activation probabilities depended on interactions between sites across successive time steps.Each site modeled activity recorded at an electrode-scale neuronal population.
RESULTS
Across cortical cultures, ongoing and stimulus-evoked information measures peaked at an intermediate E/I condition near κ≈1, where neuronal avalanches emerged and interactions remained moderate. This peak was robust across recording scales, reproduced by a network model, and matched properties observed in rats and awake monkeys.
- Pattern measurement: Large negative LFP deflections were used to identify active recording sites and construct binary cortical population-activity patterns.These patterns formed the basis for measuring repertoire entropy and population events across cultures, rats, and monkeys.
- Ongoing activity: Entropy H peaked near κ≈1, indicating maximal information capacity at an intermediate E/I ratio.The peak was statistically significant and persisted across the tested pharmacological conditions.
- E/I and neuronal avalanches: κ<1 indicated reduced E/I, whereas κ>1 indicated increased E/I; κ≈1 corresponded to neuronal avalanches with event-size distributions near a -1.5 power law.κ provided a continuous network-level measure of E/I and the statistical character of ongoing population dynamics.
- Robustness: Entropy remained near its peak when spatial resolution was halved, recorded area was reduced by 75%, or recording duration was shortened to 12 minutes.The peak occurred at κ*=1.01±0.02 after coarse-graining and was robust to these changes in measurement scale.
- Stimulus-evoked activity: Stimulus-response mutual information also peaked near κ≈1, showing maximal information transmission at the same intermediate E/I condition.Responses to 10 stimulus amplitudes were evaluated at fine and coarse spatial resolutions.
- Mechanism and in vivo agreement: At the entropy maximum, interactions were moderate (MI≈0.2) and site participation was not too depressed (L≈0.25), while model results reproduced the experimental peak.Shuffling removed interaction effects and brought entropy toward bounds set by event counts and participation likelihood; in vivo measurements matched the predicted regime across rats and monkeys.
DISCUSSION
The study links balanced excitation and inhibition, neuronal avalanches, and criticality with maximal entropy and information transmission in cortical networks. Its discussion places these findings within prior work while emphasizing that the relevant result is the peak near κ≈1, not an absolute entropy value.
- Main finding: Entropy and information transmission are maximized at the E/I condition specified by κ=1, where neuronal avalanches emerge.This experimentally connects peak information measures with critical dynamics in cortical networks.
- Scope and interpretation: Absolute entropy values should not be treated as a universal information cap or directly compared across studies because they depend on measurement and analysis choices.The discussion identifies the relative change in entropy and its peak near κ≈1 as the meaningful result.
- Conceptual distinction: The study distinguishes comparing entropy across experimental conditions from maximum-entropy modeling, where entropy is used as a modeling constraint.Here, entropy measurements are outcomes used to identify conditions producing maximal entropy.
- Relation to prior work: Prior theory and modeling predict criticality-related optimization of event-size entropy, activation-pattern repetition, and mutual information, consistent with the present findings.The discussion cites balanced propagation and interactions between excitatory and inhibitory neurons as relevant to critical dynamics.
- Novelty: The work is presented as the first experimental demonstration that cortical entropy and information transmission peak in relation to criticality.Earlier studies addressed entropy, avalanches, or E/I balance separately, whereas this study reports their convergence in cortical dynamics.
- Mechanistic interpretation: The authors interpret κ≈1 as a balanced E/I regime with moderate network activity and interactions that supports information capacity and transmission.They connect this regime to criticality and neuronal avalanches rather than to highly excitable or weakly active states.