Erasing memory, one neuron at a time

Mar 26 2009 Published by under Neuroscience

And now we get to the third paper I could present for Journal Club. This one's on something that Sci doesn't really know as much about. This makes it both more exciting, and slightly more scary. But the science is elegant and the results are amazing. So it might very well be worth it.
Have you ever thought of what happens when you commit something to memory? If you're like Sci, you think of filing it away in little filing cabinets in your brain. But the way that memory encoding is actually done is still something of a mystery. We know that groups of neurons work together to serve as a physical representation of memory, otherwise known as the "memory trace". Groups of neurons acting together have been seen correlated with memories being encoded, and with the expression of memories. But we still don't know whether these disparate groups of neurons "are" memory, whether they are essential for a memory trace.
This group of researchers decided to find out whether specifically activated neurons were essential for memory learning and expression. As you might imagine, what they had to do was pretty insanely complicated, but what they found makes it very well worth it. Seriously. Complicated. My print-out of the paper (I had to print it out because I couldn't just read it, had to make notes), is COVERED IN INK. But I think I've got it, and they've got an awesome message. Now to pass it on! Han, et al. "Selective erasure of a fear memory" Science, 2009.

To start with, it is know that certain neurons in an area of the brain known as the lateral amygdala are activated by auditory fear memory training and testing. The amygdala is a small, almond-shaped area of your brain, located right above and slightly in front of your hippocampus, and deep inside the temporal lobe.

The amygdala is an area involved in the processing and memory of strong emotional experiences, such as experiences involving fear. So it's a very good place to look when you're looking for how fear memories are processed and encoded. In this case, researchers used the paradigm of auditory fear learning in mice to look at activation of neurons in the amygdala.
Auditory fear learning is pretty simple. If you give a mouse an unhappy stimulus (usually a foot shock), and associate it with a tone, the mouse will learn the association. Pretty soon, if they hear the tone and remember that something unpleasant is about to happen, the mouse will freeze up and brace itself in preparation. This is a very simple learning task, and it's very easy to tell when a mouse has learned it, so it makes for a very good paradigm when you're testing something this complicated.
So they're looking at auditory fear learning, and they're looking at a subset of neurons in the lateral amygdala. This subset of neurons shows increased activity during memory training or testing. These neurons ALSO show an increase in a transcription factor known as CREB, which is a protein which play a role in increasing gene expression. The authors of this study were trying to find out whether the expression of CREB in these neurons was critical to the development of memory.
And this is where it gets interesting. Basically, they used a virus to inject CREB DNA (or control DNA which did nothing) into mouse neurons. These weren't ordinary mice. They were mice which contain something called DTR, which a diptheria toxin receptor. Activation of this receptor (by injecting diptheria toxin) will kill the cell. But the DTR isn't active. It's dormant, until you activate it with a sequence called 'cre'. Once the DTR is activated, you can then use diptheria toxin to kill off that cell.
So this virus they used to inject CREB or control ALSO had cre in it. It got into a specific, very small group of neurons, which are now susceptible to diptheria toxin. This means that you have cells that overexpress CREB, and you can knock them out and kill them whenever you want to, without ever affecting the other cells around them. It's complicated, but this is a GREAT model.

To the far left (in white) is your basic mouse which has DTR expressed. The DTR is normally inhibited (you can see the little "stop" in there). After injection of the virus, which has "cre", you can see the green mouse now expresses the DTR (it says "on" and he's got little receptors sticking out the sides). So he's now susceptible to diptheria toxin. Then you give diptheria toxin to the mouse (the blue triangles hitting the purple mouse), and the cells expressing DTR, and ONLY those cells, will die (the mouse in red).
So the researchers used this model to increase CREB expression in a small group of neurons in the lateral amygdala. They then trained the mice using fear conditioning, and looked to see which neurons showed more activity following training. It turned out that neurons which expressed high levels of CREB were much more likely to be activated than those that expressed normal levels of CREB or expressed no CREB at all. This implies that high levels of CREB expression influence which neurons are recruited in memory learning.
Next, they did the same task, on this time using a weak training for fear. Those mice which overexpressed CREB in some neurons learned better than those that did not, and when they gave diptheria toxin and knocked out the CREB, this effect reversed! This means that high expression of CREB can increase memory development, and that knocking down CREB can decrease the effect.
Not only that, since neurons with high expressions of CREB are activated most by memory learning, it seems that high CREB levels will determine which neurons will go into a forming memory trace and which will not. And these neurons with high CREB levels will then be the source of the memory. They found that, when they knocked out the CREB neurons in already trained mice, the mice...couldn't remember. They lost the memory, and it seemed to be permanent. BUT, they were still able to learn other tasks, presumably because other cells could pick up the slack.
What does it all MEAN?! It means that expression of CREB in neurons in the lateral amygdala is critical for the learning and expression of memory in mice (and possibly also in humans). Not only that, it means that the neurons themselves expressing the CREB, are essential for the learning and expression of memory. The neural networks hold the memories themselves, they don't just create an environment in which memories can form.
Mostly this is cool because it expands what we know about memory (and we still don't know a lot). And it's got some cool implications for problems of memory in the clinic. If we know how memories are formed, we are one step closer to finding out how they are lost. I can also see some implications in this for things like Post-traumatic stress disorder. If all memories (especially, in this case, fear-related ones) are formed with specific networks of neurons, then it may be possible to parse out and delete those specific neurons, deleting a traumatic memory that can have severe psychiatric consequences. Of course, I think this sort of thing would be a very long time coming, but it's always cool to think of the possibilities that can come from this basic knowledge of the brain.
Finally, the pros and cons:
This study is beautifully designed and executed. For every thing, there is a control. Gorgeous.
The finding is really novel and adds a lot to the knowledge on the topic.
The finding is not just a cool mouse, it's behavior, neuronal networks, AND a mechanism! AWESOME!
It's INSANELY COMPLICATED and very difficult to explain, especially given the time allotted.
...but it's SUCH a good paper...
Han, J., Kushner, S., Yiu, A., Hsiang, H., Buch, T., Waisman, A., Bontempi, B., Neve, R., Frankland, P., & Josselyn, S. (2009). Selective Erasure of a Fear Memory Science, 323 (5920), 1492-1496 DOI: 10.1126/science.1164139
EDIT: Based on the brilliant model proposed by Neuroskeptic, I made a model to show the main findings! And it's cute. So I'm putting it here:
new memory1.jpg
This is where we start. A new memory can get encoded into any one of those boxes, each of which represents a neuronal network in the lateral amygdala. But it will preferentially go to ones that are red, the ones that are expressing CREB in high levels. So the scientists here painted one of the neuronal networks red, by causing it up over-express CREB:
new memory2.jpg
Then they trained the mice and looked to see which neuronal network became activated, showing where the memory had encoded. Sure enough, the new memory preferentially encoded on the neuronal networks high in CREB:
new memory3.jpg
And when they deleted the neuronal network with high levels of CREB, they ALSO deleted expression of the memory.
new memory4.jpg
But there were still lots of neuronal networks left, so further memories could still be trained, and the animals could still encode them. I shall be suing this in my presentation, I think. Thanks, Neuroskeptic!!

23 responses so far

  • hat_eater says:

    "high levels of CREB expression influence which neurons are recruited in memory learning."
    Am I right that this is the most basic takeaway message from this experiment?

  • Scicurious says:

    "high levels of CREB expression influence which neurons are recruited in memory learning."
    Yes, AND "Recruitd neurons with high CREB expression are necessary (and possibly sufficient, but no proof of that) for the training and expression of memory"

  • leigh says:

    ooh, i love memory papers. i will have to download and read this one later today. meanwhile, here's one of my all time favorite memory papers 🙂

  • hat_eater says:

    Thanks! I'm glad that you managed to get through my thick skull in just two tries - I needed to read your piece twice to make sure I got it all but it was worth it.
    Properly designed experiment has a beauty umatched neither by living thing nor machine, it's an embodiment of thinking that WORKS.
    Hell, even a flawed experiment is beautiful if you can learn something worthwhile from it (as happened to me).

  • Scicurious says:

    hat_eater: next time, I'll add more pics (maybe I'll do that anyway). Now that I really get the figures, they REALLY help. Esp the figures in their supplementary materials. Trust me, I know, I read this thing three times at least!

  • Neuroskeptic says:

    It's not that complicated 🙂 If I were trying to explain it to someone who knew nothing of neuroscience I would put it like this - when you form a memory, the memory gets put into a box [neural array]. There are lots of empty boxes in the brain. The brain chooses which box to put the memory into at random, but it tends to choose boxes painted red [expressing CREB]. They painted a box red [through inserting CREB via a viral vector], then taught the mouse something. The memory got put into the red box, and then they smashed up [killed with Diptheria toxin] the red box. This knocked out the memory but it didn't stop the mouse learning other stuff because it had plenty of other boxes left over.
    The lesson of the study really is that neurons overexpressing CREB tend to be recruited into memory networks; that has big implications for anyone trying to work out how neural networks encode memories - I would guess that neurons somehow "compete" to become part of the array storing a given memory and that over expressing CREB somehow helps them to do this. If I remember my neural network theory classes there are various mathematical models of how that might work.

  • Scicurious says:

    Neuroskeptic FTW!!! That explanation is GREAT. Much better than mine. I'm afraid this whole specific ell expression of stuff when you inject certain viral vectors, etc, hit me in the face. But this explanation is perfect! You totally just made my journal club. 🙂

  • Stephanie Z says:

    Oh, Sci. That paper is a thing of beauty, isn't it? You can't see it, of course, but by the time I was done reading this, I discovered I was leaning toward my monitor. 🙂

  • AK says:

    What I don't see is any indication of testing selectivity of memory. That is, there's no way of knowing whether those cells stored only that memory, or many/most/all other similar memories.
    Given the way synapses are structured, it seems likely to me that the same set of cells would "learn" to react to many different stimuli, depending on which input synapses were stimulated. Thus, killing these cells will not only erase the memory under study, but all (or at least many) other memories of this type. (Or at least the pre-existing ones. Presumably after the destruction of the cells involved, new ones would take over the functions for future memories.)
    In fact, based on my own (amateur intuitive) models of how the neural organization of the brain works, I'll make that a prediction. We can come back in a few years and see whether I'm right.

  • cm says:

    Although this technique seems quite nice, I'm not sure what the novelty is of the paper besides that. That CREB is important to memory formation (including fear memories) is not new at all. CREB is involved with a lot of important processes.
    The authors also state in the abstract:

    These results establish a causal link between a specific neuronal subpopulation and memory expression, thereby identifying critical neurons within the memory trace.

    But do they actually identify the neurons in any circuit-diagram satisfying way? I don't think they do, but I have not read the paper.

  • JLK says:

    Interesting paper. But I do see a couple of problems with it (if I understood it correctly, which I may not have).
    First, could the mice re-learn the fear response? Because if not, it would take a hell of a lot of work to make this useable in any way in humans. Fear is GOOD for us - it keeps us from doing stupid shit. Imagine if they accidentally knocked out the "memory box" for "Do not touch hot things" and we could never re-learn that. We'd be screwed.
    Second, if the "treatment" needs to be started BEFORE the fear memory is created, that would be extremely problematic in humans and likely useless in PTSD or phobias.
    I don't know Sci, when it comes to human memory I feel like cognitive-behavioral therapies and the like are the way to go to treat fear responses. Ever see Eternal Sunshine of the Spotless Mind? Kinda the same idea, I think.

  • Scicurious says:

    JLK: the mice COULD relearn the fear response. So don't worry about that.
    And the treatment in this case was just to test if upregulated CREB was a priority in recruiting neural networks. Neurons naturally upregulate CREB all the time.
    cm: I liked the paper for the really elegant model. I know it's not the most novel finding, but the WAY they did it was quite lovely.
    AK: the selectivity of memory is a very good point! I hadn't thought of that. I imagine specific neurons are involved in more than one neural network, so this is very possible. Interesting point.

  • Tony Jeremiah says:

    An interesting read. However, I'm also skeptical about the novelty of the findings.
    I can also see some implications in this for things like Post-traumatic stress disorder. If all memories (especially, in this case, fear-related ones) are formed with specific networks of neurons, then it may be possible to parse out and delete those specific neurons, deleting a traumatic memory that can have severe psychiatric consequences.
    Propanolol has been shown to reduce fear responses to spiders without eliminating memory traces, and I suspect other antianxiety meds operate in a similar fashion of reducing emotional reactions without eliminating memories. This indicates that emotional and memory states can be separated, as would seem to be suggested by the primary function of the amygdala and the structure attached to it (hippocampus). There are also meditative techniques and (controversial) repression neural mechanisms that may accomplish the same separation.
    The current study results seem consistent with Richard Thompson's (1994, 2000) experiments, whereby he eliminated a conditioned blinking reflex in a rabbit by removing a specific region of the cerebellum (although this particular research doesn't involve emotional memories). It would be interesting to see if the same results (the role of CREB)are obtained with motor memory.
    I'm skeptical that it would be at all possible to eliminate any particular memory by removing specific neurons (or even specific neural networks) with more complicated memories. In particular, Karl Lashley's research showing that regardless of what region of a rat's brain was removed, they were still capable of navigating a maze with little difficulty. One would likely have to remove an entire brain to remove any one memory due to it's rather extensive parallel processing network. But I suppose we can learn a lot about how that happens by understanding Alzheimers.
    There's also a philosophical issue of whether anyone would want to remove bad memories, as memories (good and bad) are most likely at the core of personality differences.

  • leigh says:

    well... i gotta hand it to em for an impressive technique. that's some pretty cool molecular manipulation, but i've got a real soft spot for that stuff.
    they correlated the effects nicely- overexpress creb, those neurons were more likely to express ieg's. cool. take out the overexpressing neurons, memory goes away. the last bit makes the whole story worth reading for me, but i still have my reservations.
    i guess i tend to approach things in a more negative manner- if i BLOCK the phenomenon in question, can i BLOCK the result?

  • Trisha B says:

    Okay the amy g dala,,, interesting thing about fight or flight memory, like this is aggression!!! Who is onto hearing memory and aggression and doing the Lashley thing? Hey guys the forties are long over. B

  • Lord Zero says:

    Fantastic finding, its really awesome.
    Anyway, if a neuroscientist finds it
    complex enough to be difficult to simplify...
    Im not sure if a zoologist like me will be
    able to understand the paper fully.

  • Alex Besogonov says:

    "I shall be _suing_ this in my presentation, I think."
    You are not a lawyer, are you? 🙂

  • Ian says:

    "I shall be suing this in my presentation, I think. Thanks, Neuroskeptic!!"
    Isn't that a little litigious of you?

  • Scicurious says:

    YEESH! I get it people, I get it, I can't type! 🙂

  • Scienginerd says:

    Very cool! This brings up a question for me:
    I know that remembering past experiences is actually a process of reforming and reencoding those memories, and when you block new memory formation in the middle of a remembrance, the old memory is lost (at least in mice). The question is: when reliving a fear memory, can one somehow block the influence of the amygdala, thereby deleting the emotional tag attached to the memory?

  • Stagyar zil Doggo says:

    Very cool study. However, like leigh I would find elimination of normal function more convincing. Pumping neurons with CREB apparently gives the mice super-powers (wrt memory), and subsequently killing those neurons takes them away. I'm not sure that this proves that their ordinary memory works in the same way. Perhaps if they can engineer a link between the normal over-expression of CREB in neurons to a temporary expression of Diprheria toxin receptors, this might be demonstrated?
    The other point I am curious about is how did they restrict all the happenings to the Lateral Amygdala? Isn't is possible that neurons elsewhere were also infected by their virus, recruited into memory traces, yada yada yada. Or does prior work exclude this possibility?
    Incidentally, Ed Yong also blogged about this here.

  • Ewan says:

    I covered this paper last class in my behavioural neuro course; I think I might steal your revised figs as a 'refresher' piece for start of next class. Thanks! 🙂

  • hemineglect says:

    hat_eater and Neuroskeptic:
    If this paper were only about competition between neurons for encoding memories, they would have published:
    Jin-Hee Han et al. (2007) Science 316, 457
    PMID: 17446403
    Equally nice paper and standards. In their recent paper, killing the "biased" neuron is the point.

Leave a Reply to Trisha B Cancel reply