New Model Reveals How Two Interneuron Types Generate Brain Oscillations

Neuroscientists have long puzzled over how the brain generates the rhythmic electrical patterns that seem to underlie thought and perception. These oscillations—waves of coordinated neural firing that ripple across brain tissue—appear in recordings from individual neurons and in the electromagnetic signals that blanket the skull. They matter because they seem to organize how information flows through neural circuits, yet the basic mechanics of how they arise have remained opaque, especially in the face of the brain's bewildering cellular diversity.

A team of computational neuroscientists has now built a mathematical model that cracks open this puzzle, at least for one crucial piece of it. The model focuses on a small but fundamental circuit: excitatory pyramidal cells and two types of inhibitory interneurons called parvalbumin-positive (PV) and somatostatin-positive (SOM) neurons. Using a technique called mean-field modeling—which treats large populations of neurons as unified entities rather than tracking each cell individually—the researchers showed that gamma oscillations, the rapid rhythmic firing in the 30-to-100-hertz range, emerge naturally from the specific wiring pattern between these cell types. The finding is striking because it works without invoking the temporal tricks that neuroscientists have long assumed were necessary: the model generates oscillations even when you strip away synaptic delays and spike-frequency adaptation, the biological mechanisms that have traditionally been credited with creating rhythm.

The researchers built their model on a foundation of biological realism. They simulated a network of 8,000 excitatory neurons and 1,500 inhibitory interneurons—750 PV cells and 750 SOM cells—with sparse, random connectivity matching what anatomists have observed in actual cortex. The key architectural feature was an asymmetry in how the inhibitory neurons connect: SOM cells do not inhibit themselves, and PV cells provide little to no inhibition onto SOM cells. When the team ran simulations of this network, it spontaneously generated oscillations at around 35 hertz, with SOM neurons consistently firing about 8.5 milliseconds after PV neurons—a phase lag that matches what experimentalists have measured in the visual cortex of living animals.

To understand whether this oscillatory behavior was a quirk of their specific network or something more fundamental, the researchers derived a mean-field approximation—a set of equations that describe the population-level dynamics without tracking individual spikes. The mean-field model reproduced the key features of the full network: the oscillation frequency, the phase relationships between cell types, and the mean firing rates all aligned closely with the spiking simulations. Then came the crucial test: they mimicked optogenetic experiments, in which neuroscientists use light to silence specific cell types. When they progressively silenced PV neurons in the model, oscillation frequency dropped and amplitude grew—exactly as observed in experiments. Silencing SOM neurons had the opposite effect: amplitude collapsed while frequency barely budged. This dissociation revealed that PV and SOM interneurons play distinct roles. PV cells act as a frequency governor, while SOM cells sustain the oscillatory rhythm itself.

The model also revealed something unexpected about vulnerability to seizures. When PV silencing exceeded 70 percent, the system underwent a phase transition into a pathological state of hypersynchronous firing at around 500 hertz—the computational signature of an epileptic seizure. This suggests that the balance between PV and SOM populations may be critical for protecting against runaway synchronization. Intriguingly, the researchers found that networks with high ratios of SOM to PV interneurons were more susceptible to this instability when excitatory input transiently increased, while networks dominated by PV cells remained stable. This prediction aligns with anatomical reality: sensory cortices, which are dominated by PV interneurons, tend to be more resistant to seizures than higher-order association areas like the prefrontal cortex, where SOM cells are more abundant.

To isolate the fundamental mechanism, the researchers performed a radical simplification. They removed synaptic delays and adaptation from the model entirely, reducing it to a bare three-population system with only nonlinear transfer functions. Even in this stripped-down form, oscillations persisted. They then went further, deriving a fully linear model—one in which the only parameters were the connection strengths between populations. Using analytical mathematics, they proved that oscillations emerge from the specific connectivity structure alone, independent of any temporal parameters. They showed mathematically that whenever this linear system oscillates, E and PV neurons must lead SOM neurons in phase—a prediction that follows directly from the asymmetric wiring, not from any tuning of delays or time constants.

This work reframes how neuroscientists should think about neural oscillations. Rather than viewing rhythm as a product of carefully balanced temporal dynamics—synaptic delays that create feedback loops, adaptation currents that modulate excitability—the model suggests that rhythm is fundamentally a structural property of the circuit. The specific way inhibitory neurons connect to each other and to excitatory cells determines whether oscillations will emerge and at what frequency. The implications extend beyond basic science. The model predicts that regional variations in the density ratio of SOM to PV interneurons should produce corresponding variations in oscillation frequency: sensory areas with low SOM-to-PV ratios should oscillate faster, while association cortices with high ratios should oscillate slower. This prediction aligns with empirical observations across the cortical hierarchy. The framework opens a path toward building large-scale models of cortical columns and interconnected brain regions that explicitly incorporate cell-type-specific connectivity and can predict how variations in interneuron composition shape the oscillatory landscape of different brain areas.