Active dissociation of intracortical spiking and high gamma activity
Article Date: 01 April 2026
Article URL: https://www.nature.com/articles/s41586-026-10331-y
Article Image: (not provided)
Summary
This Nature paper tests the long-standing question of what generates high gamma-band activity (HGA; 70–300 Hz) recorded in cortex. Using an orthogonal neurofeedback brain–machine interface (ONF BMI), the authors trained rhesus macaques to control cursor movement such that local intracortical spike rate and HGA on the same electrode drove orthogonal cursor axes. Monkeys rapidly learned to dissociate the two signals on a single electrode. Analyses showed that HGA on a control electrode correlated poorly with nearby spike leakage but strongly with a low-dimensional, synchronous co-firing pattern distributed across the array. Spike-triggered averages revealed that spikes across the network reliably preceded local HGA by ~10–16 ms, consistent with presynaptic spikes causing postsynaptic potentials (PSPs) that summate to produce HGA. The authors conclude that summed PSPs from synchronised, mesoscale-distributed inputs are the predominant source of intracortical HGA, reconciling prior mixed findings about spike leakage versus synaptic origins.
Key Points
- Monkeys can rapidly and reliably decouple intracortical spike rate and HGA on the same electrode using an orthogonal neurofeedback BMI.
- Correlations between single-electrode spikes and HGA are modest; HGA instead tracks a low-dimensional, synchronous co-firing pattern spread across the electrode array.
- Distance-weighted sums of nearby spikes (a model of spike ‘leakage’) explain little of the variance in local HGA, arguing against volume-conducted spike power as the dominant source.
- Spike-triggered averaged HGA on the control electrode peaks ~10–16 ms after network spikes — timing consistent with presynaptic spike → PSP dynamics.
- Contrastive PCA identifies orthogonal HG+ and SP+ neural subspaces; HG+ activity is lower dimensional and distributed, matching the co-firing picture for HGA generation.
- Implication: HGA is best interpreted as a mesoscale marker of synchronous synaptic input rather than a simple proxy for nearby unit firing, which matters for BMI design and neurophysiological inference.
Content summary
The study implanted 96-electrode Utah arrays in primary motor cortex of three macaques. After initial hand-control tasks, electrodes with good behavioural correlations were chosen as control electrodes (CEs). The ONF BMI mapped unsorted spikes and HGA from a single CE to orthogonal cursor velocities, forcing the animals to modulate spike rate and HGA independently to reach targets.
Monkeys became proficient within a few sessions and showed a clear reduction of spike–HGA correlation during ONF control versus hand control. Lowering spike thresholds into the multiunit ‘hash’ did not prevent decoupling, demonstrating that many nearby neurons contributing to the spike signal could still be separated from HGA. Spatial analyses showed that increased HGA on the CE during HG-target trials was not associated with higher spiking specifically on neighbouring electrodes, ruling out simple local spike leakage as the main cause.
Factor analysis extracted a single co-firing component that correlated much better with CE HGA than with CE spike rates. cPCA revealed HG+ and SP+ subspaces that were near-orthogonal and partly embedded within the intrinsic manifold used during movement. Spike-triggered averaging showed CE HGA reliably followed spikes across the array with latencies matching PSP peaks, and electrodes with larger co-firing weights produced larger spike-triggered HGA responses.
Context and relevance
This paper addresses a fundamental interpretative issue in electrophysiology: whether high-frequency broadband LFP/HGA reflects local spiking (via spectral leakage) or synaptic activity. The results favour a PSP-dominated account driven by synchronised mesoscale inputs. That matters because researchers and engineers often use HGA as a surrogate for local firing in cognitive studies, clinical mapping and brain–machine interfaces. If HGA indexes distributed synchronous input rather than strictly local spikes, it explains why HGA can be robust and decodable even when individual units are unstable, and why HGA may better capture integrative processes such as attention, binding or multisite sensory integration.
Author
Punchy take: this is a neat, causal demonstration that HGA isn’t just spike noise leaked into the LFP. The team forced the brain to show its hand — monkeys separated HGA from spikes on demand, and the data point to synchronised synaptic input as the main driver. If you build BMIs or interpret high-frequency field power in cognitive work, this paper should change how you think about and use HGA.
Why should I read this?
Short version: if you care about BMIs, neural decoding or what local field signals actually mean, read it. The authors ran clever closed-loop experiments that make the causal argument much stronger than previous correlative or modelling studies. It’s a tidy package — method, behaviour, and physiology — that saves you from misreading HGA as just local spike power.