Neurons in the cortex fire very sparsely, often less than one spike per sensory window. That makes it hard to use average firing rate to carry information, and instead the brain must rely on the exact timing of spikes. This paper asks a basic question: what determines whether a single neuron responds to one brief, coincident volley of inputs, or to a particular order of inputs that arrive as a short sequence?

The authors propose a simple physical idea called the delay-signature framework. Axons take different amounts of time to carry a spike to a dendritic branch. Those differences are called conduction delays. If the delays happen to cancel the original timing differences between incoming spikes, the inputs arrive together — synchronously — at the branch. The neuron can detect that synchrony with an all-or-none mechanism tied to local calcium plateaus in the dendrite (a calcium plateau is a burst of calcium that can trigger a big change in the neuron). In short: the pattern of delays acts like a key that only lets certain spike sequences line up in time and be noticed.

The authors tested this idea with simulations of an “integrator” neuron model. They report three main results. First, a single number — the dispersion or spread of the set of axonal delays that converge on a branch — controls whether a population of neurons behaves as event detectors or as order-selective sequence detectors. When the spread is narrow, sequence detectors are effectively absent. As the spread increases past a certain point, sequence detectors become common. That transition emerges even with random delays and random connectivity, and the delay spread at which sequence detectors dominate tracks the timing between events with a slope statistically indistinguishable from one.

Second, the same spread of delays sets limits on the code. It fixes the longest time interval a neuron can reliably represent and it sets an absolute timing tolerance of about one millisecond. The model also predicts that inputs arriving slower than expected are tolerated better than inputs arriving faster.