Neuronal Diversity Finds Its 'Goldilocks' Zone in Brain Network Models

Three published studies and unpublished findings presented at the Computational Neuroscience Meeting in Halifax, Nova Scotia, converge on a single conclusion: heterogeneity in neurons' firing properties improves decision making by making networks more flexible — but only up to a point. The work also carries a clinical warning. Analysis of tissue from people with epilepsy suggests that loss of heterogeneity in pyramidal neurons' intrinsic firing properties may itself trigger sudden shifts into seizure-like, hypersynchronous firing.

"The convention in the field is homogeneity," says Scott Rich, assistant professor at the University of Connecticut. Heterogeneity, he says, makes models mathematically complex and unwieldy, so researchers have avoided it. "But it shouldn't be [omitted] if we really want to understand how the brain works."

The workshop, co-organized by Rich and Richard Gast, a postdoctoral researcher in Ann Kennedy's lab at The Scripps Research Institute, gathered evidence that the right amount of diversity — the 'Goldilocks' amount — helps a network switch flexibly between distinct memories or representations. Too much diversity in firing properties can destabilize it.

What did the attractor-network models show?

Attractor networks (models of memory systems whose stable states persist after an input disappears) are often pictured as a hilly landscape where each valley holds a stored memory. Gast and his colleagues found that heterogeneity flattens that landscape. Networks whose neurons differ in how easily they fire hold fewer distinct memories, but they move from one to another more flexibly, and their richer activity is easier for downstream circuits to read out. That result, which improved performance on decoding tasks, appeared in PNAS.

The finding was purely computational. Applying the same principles to real data changed that. The researchers built network models from firing-threshold measurements of mouse cortical neurons. Compared with homogeneous networks, widely varying cortical neurons supported fewer distinct stable states but needed less input to snap into a new one, according to a preprint posted in July and presented at the workshop.

How does heterogeneity shape arousal and task performance?

A study published in December 2025 in Neuron connects heterogeneity to a phenomenon that has been "a mystery for 100 years," in the words of investigator Luca Mazzucato, associate professor at the University of Oregon. The relationship between auditory task performance and arousal follows a famous inverted-U shape: too little attention or too much agitation both degrade performance.

Mazzucato and his colleagues measured how well they could predict which tone a mouse heard by correlating pupil size, a marker of arousal, with the activity of neuronal populations in the auditory cortex. In a network model of the circuit, too little heterogeneity blocked switching between states; too much pushed the system into chaos. An intermediate amount let the network switch cleanly and sharpen its ability to distinguish low- and high-frequency tones. Neuromodulators may be the tuning dial: arousal tracks norepinephrine levels, which can adjust how heterogeneous neurons' excitability is.

A second preprint from Mazzucato's group, published in March and presented at the meeting, examined heterogeneity in connectivity degree — how many neurons connect to a given cell. Using mouse connectomics data, the team found that cortical neurons vary enormously in this measure, and that these variations alter how quickly neurons return to a baseline state. That mix of timescales lets a single input echo through a network at different speeds, says Martin Nawrot, professor of computational systems neuroscience at the University of Cologne, who was not involved in the presented work: some neurons react instantly while others integrate information over time.

Can different kinds of heterogeneity work together?

A June 2025 study involving a Hopfield network (an attractor network in which every neuron connects to all others) showed they can. The researchers removed full connectivity and varied the neurons' coding level — the number of different stimuli a cell responds to. Each heterogeneity alone hurt memory capacity. Combined, the two restored it: when popular, highly connected neurons also carried high coding levels, the network performed roughly as well as a fully homogeneous one.

"It's not just, 'heterogeneity is good' or 'heterogeneity is bad,' but it depends on the computation and how this heterogeneity is structured," says study investigator Gaia Tavoni, assistant professor of neuroscience at Washington University in St. Louis.

What does this mean for epilepsy and clinical neurophysiology?

The most direct clinical signal comes from human epileptic tissue: homogenization of intrinsic firing properties could itself drive the transition into hypersynchronous, seizure-like dynamics. Conversely, unpublished data Rich shared at the workshop indicates that heterogeneity in firing properties prevents runaway synchrony, enabling shorter, weaker and more biologically realistic oscillations rather than the idealized rhythms of computational models.

For reading-room and monitoring-unit practice, the implication is cautious but concrete: trial-to-trial variability and desynchronized background activity in scalp EEG and intracranial recordings may reflect a healthy computational feature, not noise. Nawrot notes that existing attractor models struggle to reproduce such variability without becoming unstable — a problem that clustering excitatory and inhibitory neurons can, at least sometimes, fix.

"Heterogeneity shouldn't be a bug, it should be a feature," Nawrot says.

What comes next?

The apparent contradiction across studies — heterogeneity eroding stability in some contexts and reinforcing it in others — resolves by function, Rich argues. "The brain is complex enough that both answers are probably right, in different scenarios and for different functions." Cells in the thalamus, for example, are largely homogeneous, and its role as an on-off gate may not require cortical-style variability.

Future models, Rich says, should capture heterogeneity as it is actually measured in biological systems. "Even the studies looking at heterogeneity, the heterogeneity is still kind of implemented somewhat ad hoc," he notes. "That's the next step."

By Priya Raman