The Moving Target: How Neurons Rewrite Their Own Finish Line
Here's a strange fact about how neurons decide to fire: the finish line isn't fixed. It moves.
The textbook version of a neuron is simple and satisfying: voltage builds up as excitatory inputs arrive, and the moment it crosses some threshold, the neuron fires an action potential and resets. Clean, mechanical, like a bucket filling with water until it tips. This is the "integrate-and-fire" model, and it's been the workhorse of theoretical neuroscience for decades because it's tractable and mostly right.
But real neurons cheat. Their threshold isn't a fixed line — it adapts. Fire a lot, and the threshold creeps upward, making the next spike harder to trigger (a kind of built-in fatigue). Depolarize toward threshold, and in some neurons the threshold rises to meet you, staying just out of reach. This paper works through the mathematics of what happens to the statistics of firing — not just when a neuron fires, but how reliably it fires — once you let the threshold move.
The first result is almost intuitive once you hear it: a threshold that rises with the membrane potential (a "depolarizing adaptive threshold") makes firing noisier. Compare two neurons with identical average firing rates — one with a fixed threshold, one where the threshold chases the voltage upward — and the adaptive one shows more variability in the gaps between spikes. The finish line dodging away from you doesn't just delay the race; it makes the race less predictable. The paper quantifies this with the coefficient of variation of the inter-spike interval, and shows it depends in a fairly intricate way on the balance of excitatory and inhibitory input the neuron is receiving — sometimes producing spike trains more regular than random chance (hypo-exponential), sometimes less (hyper-exponential), depending on that balance.
The second result is the one worth sitting with, because it's genuinely counterintuitive.
Consider the opposite kind of adaptive threshold — one that hyperpolarizes, dropping down when the membrane potential drops. Now imagine a neuron that receives only inhibitory input. No excitation at all. Every signal it gets is telling it, in effect, to be quieter, to move its voltage further from firing.
Under a fixed threshold, this neuron would sit silent forever — you can't cross a stationary finish line by moving away from it. But with a hyperpolarizing adaptive threshold, the finish line moves too. As inhibition pushes the voltage down, the threshold drops right along with it, chasing the voltage into negative territory. And because the voltage is a noisy, fluctuating thing — buffeted by the randomness of individual synaptic events — it doesn't move in lockstep with the threshold. Sometimes the voltage's random jitter causes it to catch up to, and cross, the threshold that's supposedly fleeing from it.
The upshot: a neuron can spike from inhibition alone. Not despite the inhibitory input, but because of the way that input reshapes its own excitability. Silence-inducing signals, filtered through an adaptive threshold, can produce sound.
This matters beyond neat math. The standard story of neural computation treats excitation and inhibition as opposing forces — a push and a pull that a neuron simply adds up before deciding whether to fire. This work suggests something subtler: the threshold itself is part of the computation, not just a passive backstop. Change how the threshold adapts, and you change not just when neurons fire but how consistently — and even change which combinations of inputs are capable of producing a spike at all. A circuit that looks purely inhibitory on paper might still be able to talk, simply because the neuron receiving its signals has learned to move its own goalposts.
Distilled from arXiv Neuroscience