The Importance of Synchrony in the Neural Control of Movement

TL;DR: Only the abstract was available to me, so what follows builds intuition around the claimed result rather than dissecting the data. The claim: using holographic optogenetics to activate the same 50–75 motor-cortex neurons at the same average firing rate, but shuffling only their millisecond-scale timing, changes whether a mouse moves. That is a rare causal handle on one of the oldest debates in neuroscience—does the cortex speak in rates or in spike timing?—and the answer here leans hard toward timing.

The debate this lands in

Since Adrian in the 1920s, the default assumption in systems neuroscience has been a rate code: what a neuron "says" is how fast it fires, averaged over some window of tens of milliseconds. Most of our decoders, most of our tuning curves, most of our population-geometry stories (think of the elegant rotational dynamics people find in motor cortex) are, at bottom, rate stories. Timing between spikes is treated as noise to be averaged away.

The competing idea is a timing code (or synchrony code): the relative timing of spikes across neurons carries information that the rate throws away. Twenty neurons firing within a 5 ms window is a fundamentally different message than the same twenty spikes smeared across 50 ms, even though the rate is identical.

The problem has always been that this is nearly impossible to test causally. In a behaving animal you observe both rate and timing together; when a neuron fires faster it usually also fires more synchronously with its neighbors. Correlation everywhere, causation nowhere. To settle it you would need to grab a fixed set of neurons and independently dial timing while pinning rate. That is exactly the knob holographic optogenetics now gives you.

Why the tool makes this newly possible

Two-photon holographic optogenetics is the enabling technology, so it's worth being concrete about why it matters here. Ordinary optogenetics floods a whole region with light: you drive thousands of opsin-expressing cells, and you have no control over which ones or when each fires. That is a sledgehammer—useless for a coding question, because you cannot hold identity or timing fixed.

Holographic stimulation instead shapes the laser into a custom 3D interference pattern (a hologram) that puts light only on chosen individual neurons, and can do so with millisecond timing. So the experimenter gets three independent dials that are normally welded together in nature:

• which neurons fire (identity),
• how many spikes each fires over the trial (rate),
• when they fire relative to each other (synchrony).

Synchronous timing

spikes land together

Asynchronous timing spikes spread in time

movement little / no movement

What they found, in order

Three claims stack up in the abstract.

First, a small population is enough. Driving just 50–75 layer 2/3 excitatory neurons was sufficient to evoke movement. That is a striking sensitivity figure on its own—L2/3 is not the direct output layer (L5 pyramidal tract neurons project to the spinal cord), so this is a small nudge to an intermediate node, amplified by recurrent circuitry into behavior. It tells you the cortex sits near a threshold where tiny, structured perturbations tip into action.

Second, circuit state dominates identity. Whether the stimulation worked depended strongly on the "state of the local circuit," and only more weakly on which particular neurons were hit. Read that carefully: it partly undercuts a strict labeled-line view where specific neurons are dedicated movement command lines. The population's momentary dynamical context matters more than the roster.

Third—the headline—synchrony drives movement. Holding rate and identity fixed and varying only millisecond timing, they find "an unexpected and strong dependence on inter-cell synchrony." Synchronous activation drove movement; the same spikes desynchronized did not, or did so far less. And crucially they offer a mechanism: synchronous input recruited distinct patterns of recurrent excitation and inhibition in the surrounding network. That is the part that turns a phenomenon into an explanation.

Why synchrony would matter mechanistically

The mechanistic hint is the most satisfying piece, because it explains why rate alone should fail. A cortical neuron is a coincidence detector with a leak. Excitatory postsynaptic potentials decay over a few to tens of milliseconds. If presynaptic spikes arrive spread out, each EPSP mostly decays before the next arrives, and the membrane never reaches threshold. Pack the same spikes into a few milliseconds and they summate, cross threshold, and fire the downstream cell. Same number of spikes, completely different outcome—the neuron is nonlinear in time.

Layered on top is inhibition. Cortical circuits are dominated by fast feedforward and feedback inhibition that arrives shortly after excitation. Asynchronous excitation gives inhibition time to catch up and cancel each contribution—a "balanced" regime that suppresses propagation. Synchronous excitation can outrun inhibition in a brief window and punch through before the brakes engage. The abstract's claim that timing "recruits distinct patterns of recurrent excitation and inhibition" is exactly this signature: synchrony isn't just adding a bit more drive, it flips the network into a qualitatively different E/I regime that amplifies rather than suppresses.

Same total spikes; only timing differs. Fast inhibition (not shown) cancels the slow, spread-out case more effectively.

How strong is this, really

Since I only have the abstract, treat everything below as calibration, not verdict.

What's genuinely compelling about the design is that it addresses the confound that has stalled this debate for decades. Independently setting identity, rate, and timing is the right experiment, and holographic optogenetics is the right tool. If the controls are clean—matched total spike counts, matched cell sets, only the temporal jitter varying—then a movement difference is hard to explain away with rate. That is the whole point.

Where I would push, and where the selection note flags concern:

• Behavioral scope. "Movement" evoked by direct cortical stimulation is not the same as naturalistic, goal-directed movement. It is possible that the synchrony dependence is a property of how you inject activity into a recurrent network from outside, not how the network natively codes during self-generated behavior. The experiment shows synchrony is a sufficient lever; it does not show the brain actually uses that lever during real movement.

• What counts as "rate held constant." Rate over what window? If you match total spikes over a long trial but the synchronous condition concentrates them, then at fine timescales the instantaneous rate differs enormously. Timing vs. rate can become a question of temporal resolution rather than two distinct codes. The distinction may be semantic at the extremes.

• Threshold artifact. Because 50–75 neurons is right at the edge of sufficiency, the system is poised near a nonlinearity, where any manipulation that increases effective drive (and synchrony does) will look potent. That is real and interesting, but it may exaggerate how much timing matters away from threshold, at natural activity levels.

• Number of numbers. The abstract gives us essentially one quantitative anchor (50–75 cells). The strength of the synchrony effect, jitter magnitudes, trial counts, and animal numbers are all in the full text I don't have.

What changes if it holds

If synchrony causally gates motor output the way the abstract claims, then the rate-based decoders and rate-based dynamical-systems accounts of motor cortex are capturing a shadow of the real variable. Population geometry might need a temporal-fine-structure axis we've been smoothing over. For neural interfaces the implication is sharper still: writing information into cortex—for sensory prostheses or closed-loop motor BCIs—may need to control spike timing across electrodes/neurons to millisecond precision, not just set firing rates. Most stimulation hardware today does not think in those terms.

The section to read in the full paper is the one containing the timing manipulation and the recurrent E/I measurements—the third result. That is where the causal claim either survives its controls or doesn't, and where the "distinct patterns of recurrent excitation and inhibition" mechanism gets its evidence. Everything upstream (small populations suffice, state beats identity) is context; the synchrony-plus-mechanism pairing is the paper.