Different Origins of Gamma Rhythm and High-Gamma Activity in Macaque Visual Cortex

When you record electrical activity from the brain, how do you know whether what you're seeing is a rhythm — neurons genuinely oscillating together — or just the smeared electrical shadow of individual cells firing? For years, neuroscientists couldn't tell. The work of Ray and Maunsell in the macaque visual cortex is the experiment that pulled those two things apart. Here's the setup. You stick an electrode into the brain and record what's called a local field potential or LFP — the summed electrical activity of thousands of nearby neurons. Analyzing that signal, you reliably see power increases in two frequency ranges. One is the gamma band, roughly thirty to eighty hertz, a band-limited oscillation long hypothesized to coordinate communication between cortical areas. The other is a broad swath of elevated power above about eighty hertz, called high-gamma. The default assumption was that high-gamma was just gamma running faster — the same kind of synchronized oscillation at a higher frequency, implying an additional, faster channel for neural coordination. But there was a rival idea: high-gamma power might not be an oscillation at all. It might be the electrical fingerprint of nearby neurons spiking — action potentials bleeding into the field potential. The problem was that under most conditions, gamma power and spiking move together. When you pay attention to something, both go up. When contrast drops, both go down. So, simple correlation between high-gamma and spiking can't tell you whether they share an origin or just share a stimulus. Ray and Maunsell needed a condition where gamma and spiking went in opposite directions. They found one: stimulus size. The experiment was straightforward in design. Ray and Maunsell implanted a ninety-six electrode array in primary visual cortex, area V1, of two awake rhesus monkeys. While the monkeys held fixation, gratings of six different radii — ranging from zero point three to two point four degrees — were flashed into the receptive fields of the recorded neurons. The key is what happens in V1 as a grating gets larger. Larger stimuli engage a mechanism called surround suppression: activity in the region surrounding a neuron's receptive field inhibits that neuron's response. So, firing rates go down as size increases. But the gamma rhythm does the opposite — larger stimuli actually drive stronger gamma oscillations. Size pushes spiking and gamma in opposing directions. That's the dissociation. If high-gamma were simply fast gamma, it should track the gamma rhythm and rise with stimulus size. If high-gamma reflected local spiking, it should fall. The result was unambiguous. As stimulus size increased, gamma power rose and high-gamma power fell, tracking the suppressed firing rate. The two bands became anti-correlated. Gamma and high-gamma were not the same phenomenon. They had different origins. That's the central finding. Now for the mechanism. To understand what high-gamma actually is, Ray and Maunsell used a signal-processing technique called Matching Pursuit. Standard spectral methods — Fourier transforms, multitaper analyses — work by projecting a signal onto sine waves. They measure oscillations. That's fine when the signal is built from oscillations. But action potentials are sharp, brief transients in the extracellular potential. They are not oscillations. When you force a transient through a Fourier decomposition, it gets smeared across many frequency bins, appearing as elevated power across a wide range — including the gamma and high-gamma range — even when no true sustained oscillation exists. Matching Pursuit avoids this by using a dictionary of both oscillatory and transient functions, picking whichever best matches the actual signal. When Ray and Maunsell applied Matching Pursuit to the LFP and computed spike-triggered time-frequency averages — essentially asking what happens in the field potential at the precise moment a neuron fires — they found a clear answer. Each spike is associated with a sharp, brief negative deflection in the LFP. That deflection is broadband: its spectral energy is distributed across a wide range of frequencies, visibly detectable down to roughly fifty hertz. Because spikes are frequent and each one deposits broadband energy, their cumulative effect shows up in conventional spectral analyses as elevated power in the high-gamma range. High-gamma activity is largely the accumulated electrical footprints of nearby neurons firing. The tight quantitative link between high-gamma and spiking runs through multiple analyses. In a separate experiment varying the temporal frequency of the stimulus, Spearman correlations between firing rate and LFP power above the gamma range exceeded zero point eighty five. In trial-by-trial correlations across the size experiment, the relationship was consistently stronger at higher frequencies: median correlations between firing rate and power in the two hundred fifty to five hundred hertz range were zero point seventy-four and zero point seventy-two in the two monkeys; for the high-gamma band they were zero point sixty-seven and zero point sixty-one; for the gamma band they were lower still, at zero point fifty-six and zero point forty-seven. The higher you go in frequency, the more the LFP tracks the spikes. That gradient makes sense if the signal you're seeing is the broadband transient of an action potential — more of its energy is visible at high frequencies where the LFP's natural background is quieter. That background issue is worth dwelling on, because it explains why this mistake was so easy to make. The LFP has a natural spectrum that falls off with frequency — the so-called one over f structure, meaning there's more power at low frequencies and progressively less at high ones. Spike-related broadband energy exists at low frequencies too, but down there it's buried under the one over f background noise. It only rises above the background at higher frequencies. Ray and Maunsell calculated the cutoff frequency — the point above which spike-related power becomes consistently detectable — as roughly fifty-two hertz in one monkey and forty-eight hertz in the other under baseline conditions. During stimulus periods, those cutoffs shifted upward, ranging from about eighty to one hundred eighty-five hertz. The practical effect: elevated power in the high-gamma range looks like a genuine feature of the signal, and traditional spectral methods, which can't distinguish a true oscillation from a broadband transient, represent it as one. This is the methodological warning at the core of the paper. If you run standard spectral analysis on an LFP and see elevated high-gamma power, you cannot conclude there's a high-frequency oscillation happening. You might simply be seeing more neurons fire. To distinguish them, you need either an experimental manipulation that separates them — as stimulus size does here — or a decomposition method like Matching Pursuit that can capture transients without forcing them into sinusoidal components. So what does this mean going forward? On the practical side, high-gamma broadband power turns out to be a genuinely useful tool — just for a different purpose than previously assumed. Because it tracks local spiking reliably across a wide range of stimulus conditions and cognitive states, it can serve as a proxy for neuronal firing near the electrode. Ray and Maunsell note that high-gamma increases have been observed in human electrocorticography recordings and in LFP and MEG data across diverse tasks. For brain-computer interfaces and for human recordings where you cannot isolate individual spikes, high-gamma is a practical index of local population activity. On the interpretive side, the caution is equally clear. When researchers report high-gamma increases during attention, memory, or decision-making, those increases should not automatically be read as evidence of high-frequency oscillatory coordination. In Ray and Maunsell's data, broadband high-gamma largely reflected spike-locked transients — not sustained narrowband oscillations — and depending on stimulus conditions, it was actually anti-correlated with the true gamma rhythm. Gamma rhythms and high-gamma power tell you fundamentally different things. Treating them as variations on the same phenomenon leads to conclusions the data do not support. The broader lesson here runs deeper than any single frequency band. Two signals can sit near each other on a spectrum and have entirely different biological origins. Standard tools will represent them both as oscillations. Pulling them apart required not better equipment — the electrodes and recordings were standard — but a smarter experimental question: find the condition where the two things you want to distinguish actually diverge. In this case, that was a grating getting bigger while a monkey held still. Simple stimulus, sharp answer. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.