(Quick poorly-researched blog post, probably only of interest to neuroscientists.)
There’s a paper “What is reinforced by phasic dopamine signals?” (Redgrave, Gurney, Reynolds 2008). I was interested in it as part of my project to continue checking whether I should really believe all the crap I wrote in Big Picture Of Phasic Dopamine, and trying to fill in the missing pieces. The paper is also related to a different ongoing project of mine, namely to resolve various confusions I have about the superior colliculus.
Anyway, in a classical conditioning experiment (flash of light then reward), when the flash of light happens, there’s a reward-related phasic dopamine signal. The paper points out that this phasic signal is very fast, starting just 70-100ms after the light (in rodents).
Supposedly this is faster than the cortex can do object identification (“signals related to object identity can be detected in inferotemporal cortex within 80-100ms after stimulus onset”), and also faster than the amygdala (“visual response latencies in the amygdala are…ranging between 100 and 200 ms from stimulus onset with mean value of ~150 ms”). But the superior colliculus (SC) is fast, and it’s right next to and connected to the dopamine neurons in VTA / SNc, and there is various other streams of evidence that SC can cause dopamine release. So that’s their theory! “We have proposed the superior colliculus as the primary, if not exclusive source of short-latency visual input to ventral midbrain DA neurons.”
I found this very confusing. I thought the amygdala learns that the flashing light leads to reward! It’s been studied to death, right? That’s what everyone’s always said, I like that story, it fits in beautifully with everything else I think I know about the brain. What the heck??
Anyway, I was relieved to see that Redgrave seems to have changed his mind on this point: a decade later he co-authored Cortical visual processing evokes short-latency reward-predicting cue responses in primate midbrain dopamine neurons, which offered experimental evidence that deactivating (part of) SC does not stop the low-latency phasic dopamine reward signal. I don’t know enough to evaluate how water-tight the experiment was, but I guess if Redgrave signed on, after previously arguing the other side, presumably it’s pretty solid evidence.
(I hope I’m understanding this story correctly, and not putting words in anyone’s mouth.)
But this still leaves the question...
How is the dopamine signal so fast?
My first thought was: C’mon, really, it’s gotta be the amygdala, right? Hmm. The amygdala response is supposedly 100-200 ms … Well that’s not dramatically slower than the 70-100ms phasic dopamine response, right? I did read (I think here?) about a thing where a downstream signal can sometimes be triggered by the leading edge of an upstream signal, such that the downstream can peak sooner. Is that 100-200 ms number the peak, or is the first detectable signs of activity? I didn’t check. Hmm, or maybe there’s faster and slower neurons in the amygdala, and the experimenters happened to be looking at slower ones?
Well, anyway, I’m not quite ready to totally give up on the amygdala, but I have to admit this line of thought is feeling like special pleading.
But then I had another thought: The cerebellum!!! The poor cerebellum is too often forgotten in dopamine discussions, and indeed Redgrave 2008 doesn’t even mention the cerebellum anywhere in the paper. But to me, it would make perfect sense!
Let’s pause for my startlingly arrogant attempt to explain the entire cerebellum in a single sentence: As far as I can tell, the cerebellum is kinda like a giant memoization system: it watches the activity of other parts of the brain (including the neocortex and amygdala, and maybe other things too), it memorizes patterns in what signals those parts of the brain send under different circumstances, and when it learns such a pattern, it starts sending those same signals itself—just faster. (See my post here for more.)
Well, if that’s right … it fits the bill! The story would be:
- First the amygdala learns that when it sees a light, it should send a signal to VTA/SNc. It learns this via supervised-learning dopamine training signals from the hypothalamus / brainstem.
- Shortly thereafter, the cerebellum learns to do the exact same thing. It learns this by watching the amygdala, trying to model its input-output relation, and then just mimicking whatever the amygdala does.
…But the cerebellum is much faster than the amygdala; that’s the whole reason the cerebellum exists.
And wouldn’t you know it, as soon as I looked, I indeed found that there’s a known connection from cerebellum to VTA/SNc.
If this is right, my prediction would be that when you do a CS-US classical conditioning experiment, there’s some period early in training where the amygdala has learned what to do from the hypothalamus / brainstem, but the cerebellum has not yet learned what to do from the amygdala. And during this period we would see light-induced dopamine signals with higher-than-usual latency. Is that right? There must be data. I dunno. I'm publishing this without checking. It's more fun that way. :-P