Wendy Suzuki, neurons driving memory

Associate professor Wendi Suzuki, Center for Neural Science, New York University

Interview 3. juni 2003 regarding results presented in Science (2003), 6. June, 1578-1581.

Copyright © Rasmus Kragh Jakobsen

Hello, this is Wendy Suzuki

- This i s Rasmus Kragh Jakobsen calling about the interview.

Yes, go ahead I am free.

I would very much like to know something about memory and the perspectives of the work presented in Science. Am I shooting too high by asking if this study will lead to an understanding of the neural circuits underlying memory?

Yeah, no no that is exactly what we are trying to get at with these studies, I mean it is a first step towards answering those larger questions, but that is the goal of the studies really.

- perhaps you could tell me a little about how it is actually done.

So basically we are trying to understand how these structures in the medial temporal lobe (MTL) allow us to form and retain new long term memories for facts and events.
So memories for facts and events have been termed declarative memories, and they are very important because our memories for facts and event really makes us who we are. They kind of make up our individual personalities.
So fx damage to the MTL in humans causes severe declarative memory impairment, so cannot form new declarative memories and you end up basically living in the present.
And also MTL as you probably know is highly vulnerable in Alzheimers disease.
Which is a clinical application of this work. You need first to understand what the NORMAL function of the MTL.
You need to have an understanding of what the normal brain is doing and how the normal brain allows you to form and retain memories.
Thats what our study have gone one step closer to.

- And the MTL is that in the Hippocampus (H) or?

The H is part of the MTL.
That general region is the region where several structures that we know are critical for declarative memory function. H is one, but there is three other areas that are all critical for declarative memory.
But this study focus om H itself.
So what we did.
It is a simple question.
We wanted to know what your braincells are doing as you learn something new, as you form a new memory.
Why were we interested in the H?
Well we know the H is important for the ability to form these kind of memories. If you damage, take it out in humans or animals you no longer have that capacity to form new memories, you fail in these tests.
So we have known for a long long time that the H is important for memory because if you damage you longer have that ability. But that is not the same thing as knowing what the normal H does.
So thats what we were trying to understand. Record the activity of individual cells as animals are performing this memory task.
What is it really doing is it doing something critical do we see something interesting.
And thats what we saw :-)
We saw very dramatic plastic changes in H cells during the learning process. So we could measure behavioural learning. Animals are sitting there, they doing the task, they are learning new associations.
Maybe I should tell you a little bit more of the tasks they were actually doing?

- Right, but first just clear this. When you measure on the individual cells, how do you do that and specifically do you know exactly what cells you are measuring?

Ok, so how we measure is that we insert a very thin wire electrode into the bran so the tip can measure the electrical activity of the individual cells. A standard technique.
Recording individual neurons. Because I have an MRI I can tell for certain that I am in the H.

- There were quite a lot of electrodes in each monkey?

No we used single electrodes but multiple times. Every day the animal would come and we would record from a single electrode, and every day the animal would learn new associations learn something new. And we would record and monitor what the H cell that we happened to find that day were doing.

- So the animal actually had this scraped off area on the back of its head or something where you could easily insert an electrode.

So let me just briefly describe what we have the animals do.
They are basically playing video games. Memory games. So they are sitting in front of a computer monitor and the memory game goes like this:
We show them a big complex picture, of lets say the central park in New York. And superimposed on the picture are four identical targets – little white squares – one north, east, west and south. On the picture. So they look at this picture and then there is a delay interval, where the picture disappears and the targets remain. And then at the end of the trial the monkeys are given the opportunity to make an eyemovement towards one of the targets.
Now the rule of the game is that only one of the four targets will give them a fruitjuice reward, which they want because they are thirsty, but at the beginning of the trainning session they dont know which one of the targets will give them a fruit juice reward aasociated with that central park picture. So they see it multiple times during the training session and with trial and error they get a lot wrong but then they finally get it right by chance. And then they eventually learn: Oh it is the north target that gives me the fruit juice reward, when ever I see the central park scene.

- What would the target be?

The target is literally a little white square. They just move their eyes towards it in the north for example.
That is why we call it a location scene association task. They have to form an association between a particular rewarded locaton, that is the target, associated with a particular complex scene, say the picture of central park.
and so they are seeing of not only central park with the four targets, but they see pictures of – I dont know – a picture of denmark, picture of paris and all these new pictures they have never seen before.
And with trial and error over the session – they get these pictures randomly intermixed – they eventually learn, Oh, I get it, you know central park I have to move my eyes north, denmark south, paris west.
And so on. That is what they are learning, thats the game they play every day.

And in terms of what does that relate to in human memory? This form of memory is called associative memory, it is a critical part of memory, we use it every day. It is for example the same form of memory that we use to remember a name of somebody new that we met. It is a name face association. It is something we do every day, and we are not that good at it actually. But that is also dependent on the H.

So animals now have electrodes in the H, it is sitting there, he is playing his memory game – location scene association task game – and we can show behaviourally by monitoring correct and error trials that he get, that he has learned these new location scene assocation. And so he is learning them behaviourally, and we have – monitoring the individual electrical cell activity in the H.
And so the critical question is: Ok, does the activity of these H cells relate in any way to the animals behaviour, does it signal fx when learning occurs.
And the answer was YES. That is the big finding that we reported.
We see many cells in the H, that change their activity strongly correlated with the animal behavioural learning curve.
So the learning curve goes from 25% – which is chance performance because he has four targets – it is steep all the way to 100%. Both learn them very very well.
So what we saw was parallel increases in cell activity from a very low base line rates to very high rates at the same time that the animals learned a particular new association.

- that wouldnt be linear I suppose?

No it is a sigmoid shape.

- In your fig 2

Yes, exactly fig 2 A
So what does that mean. Fig 2 A shows this nice correlation between the s-shaped learning curve and the squiggly neural activity curve that also goes up.
What we can say from this example is that the change in neural activity seems to be signaling when learning occurs.
BUT a critical figure is 2 F, we show that for all the cells that are changing how many of those cells are changing before the animal actually learns the association versus after the animals learns the association.
Now why is that important?
Well, if all the neurons change just a little bit after the learning – so the animal learned it, aha, I understand the association I remember the association, well that means the H cant be involed in the actual formation of the association if the behaviour changes first and then the neuron acitivity changes.
But we found half the H neurons changed several trials before learning occurred. That suggests that these changes in H neurons are signals in learning. That they are signalling when learning is about to occur. Those neurons may be driving the actual behaviour.
Those that change before are the learning neuron.

- A little to actually describing memory in terms of networks and circuits. How much can you say at this point.

This paper focused on the contribution of H to the early formation of new associations.
We actually have a paper that we are about to submit, looking at how the H signals very very well learned associations. And it is a little bit tricky because that is not waht is in the Science paper, but I can say that we have data, that suggest that the H does play a role in signaling long term memory. And direction of the work is to record not only the H but also these other MTL structures that I mentioned during both the formation of new associations as well as the retrieval of well-learned associations to understtand how each of these structures are contributing to the circuit.
Thats where we are going. The Science paper in focus here does not adress those questions.
You have identified the critical questions, and those are the studies we are doing right now, actually.

- how about relating the cells in the Science paper to each other, what are the circuits who is connected to who?

Well, thats difficult to do here. What we did identify two different subcategories of changing cells, fig 2A is an example of changing cells.
If you compare with 2C you see the 2 different categories.

- So one goes up in a sigma curve and the other goes down?

Exactly. But the one that goes down is also significantly negatively correlated with the behavioural learning.
So how we think about this is: Both these categories of cell – sustained changing cell and baseline sustained changing cell – both these categories are united in signalling when learning occurs. One goes up and the other goes down, but thats still a signal.
Something has changed in the brain.
2C really dont seem to be able to convey any more information after the point learning.
They go back to baseline rate, and are not responding more. Their big event is right at the point of learning, when they change their activity and go back down to baseline.
But the cells we are particularly interested in are the sustained changing cells, 2A, because those not signals when learning occurs but that signal is maintained for as long as we are able hold the cell. At the end of the day the electrode is taken out and the animals goes back to its cage.
So we hypothesize that these are the cells that may be participating in more long lasting – long term trace for this particular association. Thats a speculation. And I still havent told you about how they connect because I dont know that. But it is a little bit more about the circuits, the different types of cells at least contributing to this learning process in the H.

- Maybe I could ask you to speculate a bit more, since they are longlasting cells having activity, do you think that this plasticity changes until something is learned and then we have a locked network or circuit.

Yeah, so we know something about this. Not speculation.
We know the H is needed at the time you are learning something, but also during a consolidation period. Which unless I have more practice in remembering fx your name I will forget in a short while. Now if I have more practice in seeing your and saying your name, say 5-6 times it is pretty hardwired. And I need my H during all those 6-8 times early on before it formed a long term memory.
And thats what I think these sustained changing cells – that process the consolidation proces – is what I hypothesise they are participating in.
Now what happens after it is formed in long term memory, you dont need your H more for wellconsolidated memory.
FX if my H was damaged right now I wouldnt be able to remember this conversation but I would remember my childhood, highschool even graduate school memories because those are well consolidated strong long term memory. Those memories are not dependent on my H, but more on the neo-cortex, which is presumed to be the storage the final kind of repository of these long term memories.
So we think in our studies of the H, that we are focusing mainly on these areas important for initial learning and consolidation of long term declarative memory. And that eventually these memories become independent of the MTl and more dependent on the reside in the neo cortex.

- Do you then have any idea on how this i transferred?

No, thats we are interested in looking at that, but the way that it is transferred is still mysterious. As mentioned a lot this data is from lesion work, which is fine, but because you whats gone you dont know what works in the first place. U know what that area is important for, but it still doesnt tell you how that area does that function.

- You call it hardwiring, that is strong metaphor for what you suspect is happening, but is it really something like hardwiring - 2, 40 or 1.000 neurons are hardwired together and thats where your memory lies?

I wouldnt call it hardwiring, there is certainly evidence for large and widespread networks of the cortical neurons that are important for long term signalling, because if you damage them you damge long term memory. but that doesnt say they are hardwired, that just says they are involved.
The evidence is not that it is hardwired, but that it is in the cortex and that they are involved somehow, but no direct evidence in supprot of hardwiring.

- But you DO believe that new wiring is going on as these H cells are changing their activity?

Yeah, I think another important point of this paper is that it shows that there is strong learning related palsticity in the H. this we have expected for a very long time but the novelty of this finding is that it is probably one the strongest direct pieces of evidence showing Fig 2A, learning related plasticity that one can demonstrate and analyse in this way.
It is very different from saying, ok I am just going to remove the H and he cant learn and therefore the H is plastic. Here we are showing how the H is plastic, what is the timecourse and how selective is that plasticity for particular pieces of learned information.

- Plasticity, do you mean change in connection or change in activity?

Change in activity, I can say nothing whether synapses are forming, I dont think it is possible for synapses to form this fast. No evidence for growth of synapses at this fast timecourse.

- You need longer time for growth of synapses?


- Anybody going to look at which molecules or proteins are needed for this, or is that a completely different field?

Yeah, that is a different but related field and colleagues of mine that are molecular biologists are very interested in identifying the genes and proteins involved in this kind of process.
People like Eric Kandel went down in organism to Applesia californica, a sea slug, to study the molecular mechanisms of this exact kind of memory.

- Lower than Drosophila?

I dont know, it is bigger.

- I mentioned Drosophila because I would have thought you would go to a strong genetic animal.

Yes, a lot of work have in fact been done in Drosophila.

- You yourself will concentrate on long term memory?

Yes, we are in the middle of it. The next obvious question is to use the same method on the adjacent areas to the H.

- One electrode at a time?

We are moving to multielectrode product. To speed up the workprocess.

- I am out of questions, but if you would like to emphasize something that I missed or...

Yeah I guess I would say the novelty of this finding is really that these findings provide some of the strongest direct evidence for learning related plasticity in the Hippocampus, I already said that, but I just want to emphasize that, because it really is a novel finding. It is for once direct evidence instead of inderect evidence that you get from lession studies, that is what I think is the novelty of this report.

- Thank you very much.


Copyright © Rasmus Kragh Jakobsen