Wednesday, April 25, 2012

Erasing Memories Cell by Cell

3d glass brain
by Kazuhiko Nakamura
We've discussed recent findings about erasing fears from memories, but today we'll be talking about erasing the fear memory itself. This involves actually inhibiting or killing the individual neurons that encode for a particular memory, so for obvious reasons these experiments are done on mice rather than humans. 

Mice can be trained to associate a mild electrical foot shock with a tone.  The tone plays and then a foot shock is given.  Once the mouse has learned this association, it will freeze in place when the tone is played.  This is called an auditory fear memory. 
Using a fear memory paradigm, Sheena Josselyn in her Toronto lab discovered how to visualize the neurons which are active during fear memory formation. They also developed a way to target and delete them, consequently deleting the memory. 

In Han et al. (2009), some beautiful genetic trickery was used to promote a 'kill switch' only in the neurons which are active during the memory formation.  This kill switch is the diphtheria toxin receptor.  Normally cells do not have this receptor, but when they promote this receptor artificially on the cell surface, an injection of diphtheria toxin will kill that cell, but not neighboring (dtr-free) cells.  The real impressive genetics is in promoting the diphtheria toxin receptor only in neurons active during memory formation.  To do this, the Josselyn lab used a marker for cell activity in amygdala neurons during memory formation, CREB.  Specifically, they used a transgenic mouse that expressed the diphtheria toxin receptor only when CREB activates cre.

So now with the memory encoded and the kill switch in place, they pull the trigger and inject diphtheria toxin into the mice. This kills all the amygdala cells that were active during memory formation (about 250 amygdala cells or so, Han et al., 2009 figure 1B).  They then test the mice again for freezing behavior.

Han et al., 2009 Figure 3

The second set of columns (CREB-cre, DT) is the experiment I have described.  Before any drug is injected the mice freeze in response to the tone, but after the diphtheria toxin (DT) injection, the mice freeze much less in response to the tone. What is really essential to this study is the control experiments that they ran. 

They wanted to make sure that just killing any 250 neurons in the amygdala didn't causes memory loss.  So instead of using the CREB promoter to activate cre (and thus the diphtheria toxin receptor) they used a control promoter (cntrl-cre, DT above) to promote cre in about the same number of neurons, but not dependent on neural activity.  In this case, there is no statistical difference in how much the mouse freezes in response to the tone. (compare the first two columns to each other.) 

Similarly, they wanted to make sure that the diphtheria toxin (DT) itself didn't erase the memories. They injected CREB that did not promote cre, and thus did not cause any diphtheria receptors to be expressed (CREB, DT). In this case, there was again no difference between pre and post DT injection.  Finally, they wanted to make sure it wasn't the CREB-cre construct itself, so they added the CREB-cre like normal, but did not inject the diphtheria toxin, so the receptors were expressed on these cells, but were not activated. In this case again, not difference in the amount of freezing. 

Because none of these control groups showed a difference in freezing, Josselyn could be confident that she had really shown that the specific neurons that encoded the memory were necessary for recalling the memory. 

They are also clear that the amygdala is not seriously damaged in this study, as the mice can re-learn the task after the specific neurons have been deleted.

One particularly interesting aspect of this study, which the authors do not discuss, is the number of neurons necessary for encoding a memory.  They delete hundreds of neurons.  I wonder if deleting half of them or even a quarter would result in the same erasure of the memory. How many neurons does it take to encode a memory?

Recently this concept of targeting proteins to only the active cells has been extended to include channel rhodopsin, the protein which allows cells to be activated by light.  Liu et al., (2012) was able to reactivate the neurons that were specifically active during the learning of a fear response. Stimulating these neurons caused the mouse to freeze, suggesting that stimulating these neurons reactivates the memory. This paper is covered thoroughly by Mo Costandi at Neurophilosophy.

© TheCellularScale

ResearchBlogging.orgHan JH, Kushner SA, Yiu AP, Hsiang HL, Buch T, Waisman A, Bontempi B, Neve RL, Frankland PW, & Josselyn SA (2009). Selective erasure of a fear memory. Science (New York, N.Y.), 323 (5920), 1492-6 PMID: 19286560


  1. I really like this type of research, but for the Han et al. paper you have to keep in mind that this is a somewhat artificial situation because the cells were 'forced' into a memory trace because they are overexpressing CREB. Personally, I like this trick better, since in this paper Koya et al. kill neurons that express endogenous fos.

  2. The number of cells that is caspase-3 and TUNEL positive, that's not an extrapolated number for the whole lateral amygdala, but just the number that's counted in the sections, right? I can't recall. InBabyAttachMode is right, overexpressing CREB is 15-20% of LA neurons is a highly artificial situation.

  3. Great points, thanks. BabyAttach, it is definitely artificial to overexpress CREB, and the controls show that the over expression itself enhances the memory to some degree, but the ability to 'erase it' is only dependent on the deletion of the neurons. So I am convinced they are showing what they say they are showing, but I agree the situation is artificial. And thanks for the link, that Koya paper looks great, they are even in the Nucleus Accumbens!
    Tetyana, I think you are right. I misunderstood the values on their figures as 'total' but they are counting cells in sections to see how many are showing the apoptosis indicators for each group. I suppose this means that if there are about 250 cells in a particular slice that are dying, then many many more are actually being killed by diphtheria toxin. I still would love to see a technique that could randomly kill only half the tagged neurons.

  4. Well, I think the point of that is to show that DT is specific to CREB+ cells, and that most of these are dying (high degree of co-localization with GFP). But the cell numbers, outside of the context of co-localization, do not give a sense of the number of neurons that are actually being killed in the LA (and the number over-expressing CREB). 15-20% of cells dying seems like a large number. I wonder what the minimum number of cells is minimally sufficient to replicate this experiment. I think this is too artificial of a system to show how many neurons are required for the memory, not only because of the over-expression of CREB in 1/5th of the LA, but also because the learning protocol is very suited for this experiment (one-trial learning), what about in the watermaze or conditioned place preference? I think it is great way to show that CREB is important!

    What would you hypothesize if you killed half of the targeted neurons?

  5. I was wondering exactly that. I expect that deleting half the memory-activated neurons might be enough to erase the memory (just a guess, obviously), but I would be interested to see what the minimum number would be.

  6. Yeah, but it would be hard to know, presumably, in a network, not all cells are equal. I would think it would depend on "which half" (if you could know, of course). My other question is, what happens downstream of CREB activation that's so important. We don't really know that, as far as I know, anyway.

    I like your blog, I think you are right about many others focusing on "higher-level" neuroscience. What about cellular and molecular neuroscience in C. elegans or Drosophila? Or small-neuronal networks mediating sensory processing, that's my favourite! (okay, I'll stop.)

  7. Another good point, 'Which half' or could there even be a single essential neuron? The downstream mechanisms are fascinating too, are ALL these neurons undergoing LTP or what? So many interesting questions.
    I am glad you like the blog. I will definitely blog about dros. And C. Elegans too. They both showcase some great cellular neuroscience issues in ways mammalian experiments can't yet. As for small network sensory processing, I don't have any papers in mind. Suggestions?

  8. I'll get back to you on this in a few days and provide some suggestions.