Rebound Bursts in Real Dopamine Neurons
Part 5 built a rebound burster out of an Izhikevich neuron and then left it sitting there as a piece of geometry. Fair question to ask next: what is the geometry for? Stojanovic et al. (J Neurosci, 2025) give a clean answer. Rebound bursting isn't a party trick of thalamic relay cells. In the midbrain it looks like the mechanism that lets one specific population of dopamine neurons run fast.
The question
Dopamine neurons in substantia nigra are not one population. Split them by where they project and you get subpopulations with different molecular markers and different intrinsic physiology. Downstream, the dorsolateral striatum (DLS) shows the fastest dopamine transients of any striatal territory, fast enough that people have tied it to moment-to-moment vigor and variability in voluntary movement. Nobody had a mechanism for why DLS in particular gets the fast signal. The paper's answer is that DLS-projecting SN neurons carry a biophysical profile that lets them jump their firing rate about tenfold, immediately, at the offset of inhibition. Rebound bursting is the accelerator.
What it looks like in a cell
The phenomenology is exactly what Part 5 predicts. Hold a DLS-projecting cell hyperpolarized, release it, and it doesn't just return to its background pacemaker rate. It overshoots into a burst that is severalfold faster than baseline, then decays back down over the next handful of spikes. Plot rebound frequency against ISI number and you see the first interval is by far the shortest, with each subsequent interval relaxing toward the background band. The rebound delay is the part I found most striking: it sits around a hundred milliseconds and is essentially flat across the minimum voltage reached during the hyperpolarization. How deep you take the cell barely matters. Something is setting the clock independently of how hard you pushed.
 |
| From the paper Figure 2B-2D. Shows a clear Rebound Burst. Plot D measures the interspike interval (ISI), the time between spikes. A clear exponential decay to a steady state in ISI is a tell-tale sine of a burst, and will be used as a metric for rebound burst later in this article. |
Two currents, pulling in opposite directions
T-type calcium (Ca_v_3) is the obvious suspect, and the pharmacology agrees. T-type antagonists severely reduce the rebound burst, though they don't eliminate it. The logic is the standard one: T-type de-inactivates while the cell is held down, activates at low threshold on release, and turns off slowly once the cell is up at spiking voltages, so it supplies depolarizing drive for exactly as long as the burst lasts and then removes itself.
The A-type potassium result is the one worth sitting with. Blocking A-type (K_v_4) also degrades the burst, which is backwards if you think of it as a brake. It works through the resting potential. With A-type gone the cell sits more depolarized, and from there T-type never de-inactivates enough to have anything left to give on release. The two currents aren't opponents in a tug-of-war, they're a sequence: A-type holds the cell in the voltage range where T-type can recharge. The paper sorts this into gain versus timing, with the calcium conductance (together with SK) setting how big the burst gets, and K_v_4 and HCN setting when it arrives. HCN plays a small role here, but a real one.
Synchrony comes along for free
Because the rebound delay is stereotyped and roughly independent of the depth of inhibition, a shared inhibitory input releases a whole subpopulation onto the same clock. That's what the raster plots show: the DLS-projecting population is noticeably more synchronized at the offset than the DMS-projecting one, which has the weaker rebound burst to begin with. Dr. Pan has noted the same behavior in VTA dopamine neurons, with the same channels doing the work, which suggests this isn't a nigral quirk so much as a general feature of dopamine neurons that happen to be equipped for it.
So there are two signals leaving at once. The rate jump against a slow background is one. The transient co-firing across a subpopulation is another, and it's a different message: it says these cells, now, which a downstream integrator can read even if it can't resolve individual rates. Rebound bursting gets you both out of a single released brake.
Back to the cartoon
The Part 5 phase plane still holds up as a picture of this. Baseline equilibrium sitting a hair inside the resting region, complex eigenvalues at −0.09 ± 0.3i pulling the separatrix in tight around it, inhibition dragging the equilibrium down and left until it lands inside the spiking basin of the baseline plane, release, burst. The one honest caveat is the one I flagged before: the Izhikevich model has a single slow variable doing the job that Ca_v_3, K_v_4 and HCN split between them. It reproduces the behavior, it doesn't decompose it. You cannot recover gain-versus-timing from a model with one knob for both.
Deleting spikes instead of injecting current
What I wanted to check was whether the toy model reproduces the timing sensitivity, not just the burst. So rather than a hyperpolarizing step, I ran a tonically firing Izhikevich neuron on noisy input and blanked the input over a short window, a deletion band. A ten millisecond pause is enough. Firing stops, the slow variable relaxes down, and on release you get a rebound burst that looks like the recordings.
 |
| An Izhikevich neuron with 50 random Poisson inputs. There is a inhibitory pause for 10 msec on the top. This causes a rebound burst. Sweeping from 30 ms pause (gold) to a 1ms pause (blue) produces a rebound burst for all but the shortest pauses (sub 5ms pauses). |
Then I swept the band width from zero to thirty milliseconds, fifty trials each, and measured the burst by ISI rather than by eye. The result is that the burst is stereotyped down to about ten milliseconds. The first interval after the band drops well below baseline and relaxes back up over roughly eight to ten spikes, and the shape of that relaxation is about the same whether the band was thirty milliseconds or ten. Below five milliseconds it flips: no burst at all, and the next few spikes are actually slower than baseline. There is a genuine threshold duration, and it lands about where the 1/a argument from Part 5 says it should. The slow variable needs time to get down before releasing it means anything.
Where I'd go next
The stereotypy is the interesting part to me. A burst whose timing barely depends on the depth or duration of the inhibition, above threshold, is a much better synchronizing device than one that scales smoothly, and it's the kind of thing that ought to fall out of the geometry rather than out of parameter tuning. I don't have that argument yet. The deletion-band code is short and I'm happy to share it; push d up, as always, and watch the burst collapse to a single spike.
Author: Alexander White
留言
張貼留言