For those in the know: this is a bold claim! Neuroscience dogma is that memories are stored (somehow) within the synapses between neurons, and that memories become stronger or weaker with the concomitant strengthening and weakening of these synapses. However, there is very little evidence for such an unquestioned theory. We have solid proof that small groups of neurons can form a memory, so the engram must be somewhere within that circuit! But does it really make sense for it to be held in the synapses between those neurons? Charles Gallistel (who admittedly is in the bold idea phase of a legitimately impressive career) was willing to take a stand and say no, from a computational perspective it doesn’t make sense for memories to be stored in synapses. He points out that “No code has been proposed for use with plastic synapses” and cites seminal work from the father of information theory himself in asserting: “We cannot have a computationally relevant understanding of the hypothesized synaptic engram until we have a testable hypothesis about the code by which altered synaptic conductances specify numbers.”
An alternative possibility for how the brain stores memories is within the intrinsic machinery of neurons themselves. After all, we certainly know there are codes within cells capable of storing and reading out huge amounts of information (hint, it starts with a D ‘n ends with ‘n A). But is there any evidence that single neurons can hold memories?
I’m glad you asked! Here I’ll briefly go over the results of Johansson et al (PNAS 2014), which marks a triumphant culmination from over two decades of research from Germund Hesslow studying the formation of conditioned memories in ferret Purkinje neurons. I presented this paper in lab meeting and we found the results hard to question. In fact, considering the iconoclastic nature of the findings, I’m surprised more people don’t know about this paper. It only has 59 citations since its publication in 2014, and literally every neuroscientist I’ve ever brought it up to hasn’t heard of it (admittedly a limited sample size—lets face it I don’t have that many neurofriends). Meanwhile, Gallistel’s “The Coding Question” TICS piece that features Johansson et al has only 7 citations. So I’m not even sure if people take this paper or Gallistel seriously. In fact, I’d love to hear people’s opinions to the contrary! As it is, I’ll summarize where this paper came from and its relatively straightforward results to help people understand the evidence showing that the memory for a conditioned length of time is able to be stored within a single neuron.
First, to understand the results, it’s important to understand conditioned and unconditioned stimuli (CS and US, respectively). I never found these terms particularly intuitive, so it helps to use a real-life example. In previous work, the CS is a touch to the ferret’s paw, while the US is a puff of air on its eye. You then create a temporal memory by doing the CS first (touch!), waiting a couple hundred milliseconds, and then doing the US (puff!). After a number of trials, the ferret implicitly learns to close its eye after the touch (about 50 milliseconds before the expected air puff in anticipation of this negative stimulus).
As a second piece of key background info, let me introduce the key neural signature of this memory. Hesslow did this experiment all the way back in 1994, and found that single Purkinje neurons would silence their action potentials during the exact amount of time from the touch of the ferret’s paw (the CS) to when the ferret would start to close its eye (the US). A typical neuron’s response looks like this:
You can see the clear conditioned response in the Purkinje cell from just after the CS (which begins at the left of the blue bar) until just before the US, as action potential firing is inhibited during this time period. The pause in firing during this key period made researchers think that these neurons were part of the circuit that forms the memory necessary to keep track of this time gap between the CS and US. Further work over the next few decades has helped to confirm the necessity of these Purkinje cells in such conditioned memories, as using either optogenetic or chemical methods to stop the CS-US pause in the Purkinje cell firing coincides with removal of the behavior (in this case, touching the ferret paw would no longer lead to the just-trained eye blink).
Now that we’ve gone over this background: we can talk about the result crazy enough for me to blog about it. After a few decades of experiments, as you can tell by the paper trail of the Hesslow lab, researchers kept trying to create the CS and US by stimulating closer and closer to the actual Purkinje cell itself. Remember, the first experiment involved touching the paw and blowing on the eye. Well if you go a few synapses up from the paw (♫ the paw neuron’s connected to the: spinal neuron; the spinal neuron’s connected to the: brain neuron ♫), the neurons that touch the Purkinje cells on one side are parallel fibers from cerebellar granule cells. And if you go a few synapses up from the eye, the neurons that touch the Purkinje cell on the other side are climbing fibers from the olivary nucleus.
The names and regions aren’t important, but what is important is we now can do three things simultaneously: 1) create the CS by sticking an electrode on one side of a Purkinje cell and zapping it (consider it a fake paw touch), 2) create the US by sticking an electrode on the other side of the cell and zapping it (consider it a fake air puff), and 3) recording the electrical signal from the Purkinje cell itself while this experiment happens. And what do they see? The same pause in firing that was seen when they used the paw and the eye blink as the CS and US:
The blue box represents the expected 200 millisecond pause in firing between the CS and US. You might have noticed that in this case they also injected Gabazine during the experiment. This was to control for one more possible confound: that maybe when they zap the parallel fiber to create the CS that this causes local inhibitory neurons connected to the parallel fiber to then silence the whole region (such autoinhibition of sorts could cause a fake pause). I’ll ignore the remaining details, but this figure shows that even with Gabazine, which was shown to stop this local inhibition, the existence of this key CS-US pause in the Purkinje cell was still there!
Okay, so at this point you might be wondering: soooooooo? Allow me to contextualize. Somehow, these Purkinje cells are remembering a temporal memory. In this case, a 200 ms. pause in firing. Numerous computational neuroscientists have come up with circuit models (involving many neurons) to explain this Purkinje cell temporal response. Instead, it appears they were all overthinking it. The answer was much simpler than a complicated circuit somehow keeping track of time: the neuron itself appears to be keeping track of time. As in, within the messy milieu of the neuron, some sequence of chemical reactions is likely able to form a temporal memory (in fact, Fredrik Johansson from the Hesslow lab, the same author of the paper described here, has since published a paper showing one key receptor protein involved).
The implications for this finding could turn neuroscience on its head (does neuroscience’s head have a brain?). As I explained in the intro, consensus thinking is that memories are stored in the synapses between neurons. The paper presented here says: well in at least one well-studied location in the brain, the memory is intrinsic to the cell. And it seems they’re not the only neuroscientists with results that question this dogma. Now the question becomes: what kinds of memories can be stored inside neurons themselves? Are they limited to short-term conditioned response memories in the cerebellum, or are there more complicated forms of memory that can be stored intrinsically in neurons in other parts of the brain? Charles Gallistel featured this paper—in the opinion piece I detailed above—in support of the latter idea. I’m not sure what to make of his argument that all memories come down to numbers, and that computationally it is unlikely for synapses to contain these numbers and read them out in a rate code. But I am intrigued at the possibility that the storage and retrieval of memories in our brain might be way different than the neuroscience world currently believes.