Definition
Neuronal avalanches: Neuronal avalanches are cascades of activity that spread through a neural network, with sizes and durations that follow power-law distributions instead of a typical scale. First reported by Beggs and Plenz in 2003, they are read as a sign that a network operates near criticality, a regime linked to wide dynamic range and efficient information transmission.
Also known as: neural avalanches, avalanche dynamics, critical brain dynamics, neural criticality
Neuronal avalanches are cascades of activity that spread through a neural network and then stop, with sizes and durations that follow a power law rather than clustering around a typical value. First reported by Beggs and Plenz in 2003, neuronal avalanches are widely read as a signature that a network operates near criticality, a balanced regime between silence and runaway excitation.
Key takeaways
- Neuronal avalanches are defined statistically: activity is grouped into cascades, and their size and duration distributions are analyzed.
- Beggs and Plenz found avalanche sizes following a power law with exponent near -3/2 and a branching parameter near 1, as predicted for a critical branching process.
- Experiments in cortical slice cultures linked avalanche dynamics to maximum dynamic range, the ability to respond to the widest range of inputs.
- Dissociated cultures and the DishBrain system have shown avalanche-like or near-critical dynamics, which varied with development and with task input.
- For organoid intelligence, avalanche statistics are a candidate readout of whether a living network is in a good state to compute.
What are neuronal avalanches and where did the idea come from?
Physicists had long studied systems such as sandpiles, earthquakes and forest fires, where events of every size occur and the size distribution follows a power law. A key idea was self-organized criticality: some systems naturally settle at a critical point where small triggers can cause cascades of any size.
Beggs and Plenz asked whether cortex behaves the same way. In The Journal of Neuroscience (2003) they recorded spontaneous local field potentials from mature organotypic cultures and acute slices of rat cortex, using a 60-channel microelectrode array. They found that activity propagated in cascades whose sizes followed a power law with an exponent of -3/2, with a branching parameter close to the critical value of 1. Simulations in the same paper suggested that a branching parameter at this value optimizes information transmission in feedforward networks while preventing runaway excitation. They called these events neuronal avalanches and proposed that they may be a generic property of cortical networks.
How are neuronal avalanches measured?
Avalanche analysis turns a multichannel recording into a sequence of events. A typical workflow looks like this:
- Detect events on each electrode, such as negative LFP peaks or sorted spikes, using a threshold.
- Bin time into short windows. The bin width matters and is often chosen relative to the average interval between events.
- Define an avalanche as a run of consecutive bins containing at least one event, bounded by empty bins before and after.
- Measure size and duration: size is the number of events (or electrodes) involved, duration the number of bins.
- Fit distributions of size and duration and test whether a power law fits better than alternatives.
- Estimate the branching ratio, the average number of events in one bin that follow each event in the previous bin. A value near 1 indicates a critical branching process.
| Branching ratio | Regime | What activity looks like |
|---|---|---|
| Less than 1 | Sub-critical | Cascades die out quickly; mostly small, local events |
| About 1 | Critical | Cascades of all sizes; power-law statistics |
| Greater than 1 | Super-critical | Activity tends to spread widely; large, stereotyped events dominate |
Because the outcome depends on thresholds, bin widths and which neurons the electrodes happen to sample, results are only comparable when these settings are reported. Sharing raw data in a standard format such as Neurodata Without Borders makes reanalysis possible.
Why would a brain operate near criticality?
The critical brain hypothesis proposes that cortex benefits from sitting near this balance point. Shew and Plenz (2013, The Neuroscientist) reviewed the proposed functional benefits, which include:
- Maximum dynamic range: the network can distinguish both weak and strong inputs.
- Efficient information transmission: signals propagate without fading out or saturating.
- Large repertoire of states: many different activity patterns are possible, which could support information capacity.
The clearest experimental test came from Shew et al. (2009, The Journal of Neuroscience). Using cortex slice cultures on planar electrode arrays, they changed the balance of excitation and inhibition with drugs and measured responses to stimuli of different strengths. Networks that produced neuronal avalanches had the greatest dynamic range, and moving away from criticality in either direction reduced it, in agreement with their simulations. The authors suggested that balanced excitation and inhibition establishes criticality.
Do cultured neurons and organoids show avalanches?
Pasquale et al. (2008, Neuroscience) analyzed dissociated rat cortical neurons on electrode arrays. These networks also generated neuronal avalanches, but their statistics varied: cultures showed sub-critical, critical or super-critical distributions depending on age and development. Criticality coincided with moderate synchronization among bursts and a high ratio of bursting to random spiking. This links avalanche analysis to network bursts, which dominate many mature cultures.
The most direct connection to biocomputing comes from Habibollahi et al. (2023, Nature Communications), working with the DishBrain system in which cultured neurons played a simplified Pong game. They reported that critical dynamics emerged when networks received task-related structured sensory input, and that better task performance correlated with proximity to criticality. They also reported that criticality alone was insufficient for learning without feedback about the consequences of previous actions.
For brain organoids, avalanche and criticality measures have been explored less systematically than in slices and 2D cultures, and the three-dimensional structure means surface electrodes sample only part of the tissue. Claims about organoid criticality should be read with that sampling limitation in mind.
Neuronal avalanches vs network bursts vs oscillations
| Pattern | Defining feature | Characteristic scale? | Typical analysis |
|---|---|---|---|
| Neuronal avalanche | Cascade bounded by quiet periods | No (power law at criticality) | Size and duration distributions, branching ratio |
| Network burst | Population-wide surge of spikes | Often yes | Threshold detection on spike counts |
| Oscillation | Rhythmic activity | Yes (a frequency) | Spectral analysis of LFP or rates |
These are different lenses on the same recordings rather than mutually exclusive states.
What are the debates and limitations?
- Power laws are not proof of criticality. Critics have noted that power-law-like distributions can arise from non-critical processes, and that fitting power laws over limited ranges is statistically tricky. Stronger tests combine several exponents, scaling relations and shape collapse.
- Subsampling. Electrodes record a tiny fraction of neurons, which can distort avalanche statistics.
- Parameter dependence. Thresholds and bin widths change the results, so conclusions must be robust across reasonable settings.
- Near-critical, not critical. Many researchers now talk about networks operating near, or slightly below, criticality, rather than exactly at it.
- How is it maintained? Ma et al. (2019, Neuron) reported that rat visual cortex returned to near-critical dynamics after sensory deprivation, suggesting a role for homeostatic plasticity, particularly of inhibition.
Why do neuronal avalanches matter for organoid intelligence?
Approaches such as reservoir computing need a network that responds to inputs in rich, input-dependent ways without collapsing into stereotyped bursts or fading to silence. Near-critical dynamics are one proposed description of that sweet spot. In practice, avalanche statistics could serve as:
- a quality check on a culture before an experiment, alongside firing rate and burst metrics;
- a diagnostic during training, to see whether closed-loop stimulation moves a network toward or away from a responsive regime;
- a comparison metric across cultures, lines and protocols, provided analysis parameters are fixed and reported.
The Habibollahi et al. result is a useful caution: being near criticality may be necessary or helpful, but it did not produce learning on its own. The Brainoware and DishBrain work covered in reservoir computing with brain organoids and how neurons learn in a dish shows why both the state of the network and the structure of its feedback matter.
Computing avalanche statistics with fixed, versioned parameters for every recording, so that one culture’s “near-critical” means the same as another’s, is the kind of analysis we want to make routine at Organoid Programming.
Conclusion
Neuronal avalanches are scale-free cascades of neural activity, and since Beggs and Plenz described them in 2003 they have become the main empirical window on the idea that cortex operates near criticality. Evidence from slices links them to maximum dynamic range, and work in cultured networks suggests structured input can push dynamics toward criticality. For living biocomputers, neuronal avalanches offer a measurable, if debated, indicator of whether a network is ready to compute.
Frequently asked questions
What is a neuronal avalanche in simple terms?
A neuronal avalanche is a burst of activity that starts somewhere in a neural network and spreads, like a chain reaction, until it dies out. Most avalanches are small and a few are very large, with no typical size in between. That pattern, described by a power law, resembles earthquakes and sandpile avalanches, which is where the name comes from.
Who discovered neuronal avalanches?
John Beggs and Dietmar Plenz reported neuronal avalanches in The Journal of Neuroscience in 2003. They recorded spontaneous local field potentials from organotypic cultures and acute slices of rat cortex with a 60-channel multielectrode array and found that event sizes followed a power law with an exponent of about -3/2, with a branching parameter close to 1.
What does criticality mean in the brain?
Criticality refers to a regime poised between two phases: one where activity dies out quickly and one where it spreads explosively. At the critical point, each active neuron on average triggers about one more, so cascades of all sizes occur. The critical brain hypothesis proposes that cortex tunes itself near this point because it balances stability with responsiveness.
How are neuronal avalanches different from network bursts?
Network bursts are large, synchronized surges of firing detected with thresholds on spike counts, and they often have a characteristic size. Neuronal avalanches are defined by grouping activity into cascades and analyzing the statistics of their sizes and durations. The same recording can contain both: avalanche analysis asks whether events lack a typical scale, burst analysis describes discrete population events.
Do brain organoids or cultured neurons show neuronal avalanches?
Dissociated cortical cultures on electrode arrays have been reported to produce avalanches, with sub-critical, critical or super-critical statistics depending on age and development (Pasquale et al., 2008). In the DishBrain system, Habibollahi et al. (2023) reported that cultures moved closer to criticality when receiving structured, task-related input. Systematic characterization in brain organoids is still developing.
Why do neuronal avalanches matter for biocomputing?
Theory and slice experiments suggest that networks near criticality respond to the widest range of inputs and transmit information well. If that holds for cultures and organoids, avalanche statistics could serve as a quality check before training and as a diagnostic during it. Results so far suggest criticality may help computation but does not by itself produce learning.
Sources
- Beggs & Plenz (2003). Neuronal avalanches in neocortical circuits. The Journal of Neuroscience.
- Shew, Yang, Petermann, Roy & Plenz (2009). Neuronal avalanches imply maximum dynamic range in cortical networks at criticality. The Journal of Neuroscience.
- Shew & Plenz (2013). The functional benefits of criticality in the cortex. The Neuroscientist.
- Pasquale, Massobrio, Bologna, Chiappalone & Martinoia (2008). Self-organization and neuronal avalanches in networks of dissociated cortical neurons. Neuroscience.
- Habibollahi, Kagan, Burkitt & French (2023). Critical dynamics arise during structured information presentation within embodied in vitro neuronal networks. Nature Communications.
- Ma, Turrigiano, Wessel & Hengen (2019). Cortical circuit dynamics are homeostatically tuned to criticality in vivo. Neuron.
