Monday, August 28, 2006
Within my brain, foun-tains over rugged moun-tains of my terrain, diggit I came too far to front. So I'm meditatin on how to maintain. Stepped off at City Hall into the rain... - Black Thought
That ain't a igloo, that's my watch. And that ain't snow, baby that's my chain. That's not an ice tray, that's my teeth. And that's not a snowcone, that's my ring. - Paul Wall
Two very important papers came out yesterday in Science. One (from the Bear Lab at Picower) demonstrates that inhibitory avoidance learning can induce LTP (long-term potentiation) in the hippocampus, and the other (from the Sacktor and Fenton labs at SUNY-Downstate) demonstrates that a particular variant of Protein Kinase C (PKC) known as PKM-zeta (that is necessary and sufficient for LTP maintenance) is necessary in the hippocampus for long-term maintenance of a spatial avoidance memory. Together, these papers make one much more confident about making predictions about the cellular mechanisms of memory based on electrophyz studies in the hippocampal slice. The Scientist has a nice summary of these two, and I suggest reading it. I think the two papers could be even further integrated by taking into account a recent discovery by the Malinow group concerning PKC's role in AMPA receptor trafficking.
By the way, as I get deeper into this post I realize that it is going to make no sense if you don't understand at least the basics of neurotransmission, so if you aren't comfortable with that try a little of this before you read on.
Update:From Todd Sacktor in the comments:
ZIP is quite specific to reversing late-LTP. We showed in Serrano et al. and Frey's lab showed in Sajikumar et al. (both in J. Neuroscience) that it did not affect baseline synaptic transmission or reverse early LTP. We are currently looking at other types of behavioral memory (including the inhibitory avoidance used in the Bear study).
If all that above seemed like an alien language, just give me a second and I'll unpack it. First off, the behavioral paradigms. Inhibitory avoidance is sometimes known as passive avoidance. All the rat has to do to avoid a shock is to sit still, but the rat is placed on an inch-high platform in the light and the shock grid is in a dark area of the chamber. Rats would much rather be in this dark area, so they have a tendency to step down. The first time they do this, they get a mild foot-shock. Learning is measured by how long it takes them to step down a second time, but this paradigm is so widely used and consistent that we can assume the rat learned the association without even bothering with a test trial. It is not entirely clear just what it is that rats learn during inhibitory avoidance training. It seems that they should be learning an association between the stepping behavior and the shock. On the other hand, inhibitory avoidance could just be a fancy version of contextual fear conditioning (CFC). In CFC rats are simply placed in a box, given a little time to explore, and then administered foot-shock. The indication of learning is a stereotyped behavior known as freezing when they are placed in the same context. This is an adaptive response to fear for rats at least in part because predators (like hawks and whatnot) have poor foveal vision, so they do a lot better if they have a moving target. Point is, rats stand still when they are scared. So it is not clear in inhibitory avoidance whether rats are scared of the step down or are more generally afraid of the apparatus leading to freezing. The reason I'm bothering to make all this distinction is that the association between a cue and a shock has been traced to the amygdala rather than the hippocampus. The hippocampus is thought to encode the context by binding its features up as a single index cue that can be associated with shock in the amygdala. It is possible that the hippocampus constantly takes snapshots of the various aspects of the context together and that stringing these together makes up the film of vivid episodic memory that some refer to as 'mental time travel'. This makes it curious that the LTP-like changes in the Bear Lab paper are found in the hippocampus when there are "walk-through" controls that get all the same contextual/episodic information except for the shock. My tentative take is that trained and control animals both encode the contextual information, but that shock provides a modulatory input that says, "hey, why don't you go ahead and lay the memory of this context down a little more permanently.. i have a feeling it's going to be important later.." This sort of input could come from the amygdala.
In the PKM-zeta study, the behavioral paradigm is Active Spatial Avoidance. The rats are placed on a slowly spinning circle. One portion of space is declared the "shock zone" and remains stationary while the rest of the world spins, so the rat has to keep moving to keep from being eventually moving into the shock zone due to the platform's rotation. It's like being a vampire who has to remain outdoors and near the equator. This task could again be separated into two separate types of learning. The rat needs to encode the overall spatial scheme, which is something we expect the hippocampus to be good at, and it has to associate a particular portion of the map with shock. It should be noted that use of this behavioral paradigm is not nearly so widespread as inhibitory avoidance, so less is known about the neural substrates, but the hippocampus appears to be involved. A good way to parse the various forms of learning necessary for the task would be to allow the animal access to the spatial environment without the shock reinforcer, so it can learn the spatial information first. You might expect a more rapid acquisition curve once the shock is turned on, and this improved acquisition might be affected by hippocampal manipulations. The drug manipulations in this study were in the hippocampus, and they knocked the memory out, so something that the rat needs to know to perform this task is encoded in the hippocampus. I'm just not sure if it is a cognitive map or some "episodic" memory like "Remember that time when i was in this spatial area and I got shocked? I better get outta here." One is remembering a space and one is remembering what happened in a space.
Now let's take it down a level or two. What is LTP? It is a model for memory at the cellular/synaptic level. A long time ago, Donald Hebb suggested that information in the nervous system was probably stored as changes in the synaptic connection between neurons. In particular, he suggested that they synaptic strength or weight would be increased when the pre-synaptic neuron and the post-synaptic neuron fired coincidentally. I dunno if it was Hebb who came up with the paraphrase, but everyone paraphrases this principle as "Cells that fire together wire together." In 1973, Bliss and Lomo were able to demonstrate a cellular phenomenon that did just what Hebb has imagined. They found that high-frequency stimulation of a set of synapses led to enhanced strength. The strength is ascertained by giving baseline pulses of fairly weak stimulation and measuring the post-synaptic change in electrical potential. After high-frequency (tetanic) stimulation, the electrical response to baseline stimulation is potentiated. The emerging consensus is that synaptic strength is determined by the amount of ion flow through AMPA-type glutamate receptors in response to pre-synaptic glutamate release. Thus, you could enhance synaptic strength either by making existing AMPA receptors more permeable or by putting more AMPA receptors in the synapse. This latter is known as AMPAR trafficking. Modifications to AMPA receptors accompany LTP induction and memory acquisition. In particular, one portion of the AMPA receptor (the c-terminal tail of the GluR1 subunit) has a bunch of little spots on it where it can be accessorized. It's like, does Paul Wall really wanna roll out with his neck, wrist, and grill iced up tonite or is he far from home and might get his chain snatched if he comes too flashy? Then he might only wanna display his Balla Status with his neckpiece.
Three of these spots (S818, S831, and S845) are partially understood. The general idea is that more accesories = stronger synapses. S831 can be modified by CaMKII and PKC, but this doesn't seem to be sufficient to drive trafficking. S845 is modified by PKA, but this isn't enough either. In fact, it seems more like the S831 modification controls ion flow (channel conductance) more than it does trafficking. The Malinow group at Cold Spring Harbor just reported a month ago now that they had characterized S818. S818 is modified by only by PKC isoforms. There are several versions of PKC (11, I think), and they have special expression patterns and activation requirements. S818 is in a part of the AMPA receptor that is close to the cellular membrane and is packed full of positive charges. The membrane is made of lipids and may have trouble playing with charged molecules. PKC adds a bunch of negative charge to S818 and helps neutralize the membrane-proximal region of the AMPA receptor allowing the AMPA receptor to fuse into the membrane and start contributing to synaptic strength. So PKC-mediated phosphorylation of the GluR1 subunit of AMPA receptors may be a central mechanism in enhancement of synaptic strength (LTP). This was not known until a month and a half ago which explains why the Bear lab only checked on the status of S831 and S845 after inhibitory avoidance training.
The Bear lab attempted to show that learning induces LTP with four results: 1) After training, S831 is modified the same way it is after LTP. 2) After training, there are more AMPA receptor subunits in or near synapses. 3) After training, baseline stimulation of the synapses produces greater post-synaptic potentials. 4) Synapses that show this enhancement are harder to potentiate with high-frequency stimulation (i.e. the normal way you induce LTP). The third and fourth are really the money. They were made possible by the use of multielectrode recording arrays. Learning doesn't just globally increase synaptic weights. That would be silly. The changes have to be fairly synapse-specific. So if you just stuck one recording electrode in the hippocampus you might miss the change. Bear and co were able to monitor the status of several recording sites at once. After training, some small portion of synapses were enhanced, while the others seemed to drop off slightly. This drop off is really interesting to me because it looks a lot like some sort of homeostatic signal that might keep overall excitability in the right range for a good signal-to-noise ratio. That's the problem with just using LTP in your model. Eventually you would get everything strengthened to the max, and you couldn't tell one piece of data from the other. The paper is more bland and probably more reasonable. "We interpret the coherent decreases in fEPSP slope as reflecting changes in the behavioral state of the animals over the duration of the recording experiments."
There is one little disconnect in these observations though. The AMPA receptor changes are fairly short-lived. They are up a half-hour after training, but back down after 1-2 hours. The synaptic strength changes, on the other hand, can last over 3 hours, and we know that the memories last much longer than even that. The authors suggest that only a few of the initially enhanced synapses actually stick it out and contribute to the long-lasting potentiation. This would drop the number of synapses displaying the AMPAR-related biochemical markers so far down that the assays just aren't sensitive enough to detect them anymore. Perhaps another issue is that in order to detect the changes in amount of synaptic AMPA receptors they normalize to actin levels. I haven't had anyone explain to me yet why you would expect actin levels to remain constant during synaptic modification. Go back and read here and here. Actin is a cytoskeletal protein that is associated with dendritic spine morphology and is dynamically regulated in response to LTP induction. One could easily imagine an initial burst of AMPA receptor insertion followed by changes in the actin cytoskeleton to produce larger synapses and accomodate the new strength setpoint. The drop off of the AMPA receptor signal back to baseline could really reflect a slower rise in synaptic actin. Also, there is another potentiality that can't explain the overall levels data but could explain the rapid fall off of the S831 modification. The S831 modification may be a simple, fast response to do the trick of enhancing synaptic responses quickly while the slower-reacting process of AMPAR insertion is taking place. S831 modification, remember, allows more ion flow, but doesn't seem to affect receptor insertion. I could see S831 doing its duty to begin with, but then handing off responsibility to a more permanent change effected by S818 modification and an increase in total number of receptors in the synapse. So the mechanism of synaptic strength enhancement would evolve over its lifespan. There would be mechanisms for acquisition that were separate from those involved in maintenance.
Maintenance. What a memory does day-to-day when you aren't recalling it or forgetting it. All those phone numbers you know are sitting there as configurations of synaptic weights that have to stay at the right weight as they receive noisy input and all the consituent synaptic molecules are degraded and replaced. The Sacktor lab has made a very nice case for a special role for PKM-zeta in memory maintenance. PKM-zeta is an atypical isoform of Protein Kinase C. Most PKC's are activated in response to increases in intracellular signaling molecules (such as calcium and diacylglycerol (DAG)), but PKM-zeta doesn't need any of that. It just goes. You make PKM-zeta and it starts doing its job. It is thus referred to as a constitutively active protein kinase. PKM-zeta isn't required for LTP induction. It is required for LTP maintenance. In fact, just washing PKM-zeta onto a hippocampal slice will cause increases in synaptic strength. This increase is probably due to increased AMPA receptor insertion. Sacktor and co have a molecule called Zeta Inhibitor Peptide (ZIP) that specifically inhibits PKM-zeta and no other PKCs or other kinases. They showed that ZIP can return potentiated synapses to baseline even after they have been potentiated for 22 hours. So we've got a drug that specifically affects LTP maintenance and we think LTP = memory, so let's put the drug in and see if it affects memory. Sho' nuff. If you drop ZIP into the hippocampus 22 hours after active spatial avoidance training the animal drops back down to pretraining peformance levels. It forgets everything. This isn't a temporary retrieval impairment. The memory is still gone a week later. That, my friends, is money. Some eternal sunshine type isht. The whole hippocampus isn't messed up cos the rats can acquire new memories even under the influence of the drug. It only screws up fairly new memories that have been stored.
There are a couple of things to note about this discovery. The manipulation is still effective on month-old memories. If you've read any accounts of memory research you will be familiar with the case of H.M. He is commonly used to illustrate the principle of systems consolidation. Some people believe that the hippocampus is only a temporary memory storage site and that (maybe during sleep) it trains up other permanent storage sites in the neocortex. For instance, H.M. basically had his hippocampus removed, and he still seemed good at remembering stuff from his childhood (which was presumably consolidated) but wasn't so hot at memories acquired shortly before his surgery. People have extended this observation to rats and showed that lesioning the hippocampus has less effect on month-old memories than on day-old memories. This whole area is very controversial though and there is growing support for the notion that the hippocampus never stops playing a role. This set of experiments speaks to the issue of systems consolidation by showing that month-old memories are abolished by a drug manipulation targeted only to the hippocampus. There are caveats and complexities, but it certainly looks on the surface like synaptic changes in the hippocampus are still housing the memory. But this brings me to my other point. What is really happening when they drop ZIP into the hippocampus? Is every synaptic weight getting dropped down to its lowest possible setting? Do the rats forget everything that the hippocampal memory system was responsible for? There are control experiments where ZIP is shown not to affect baseline synaptic responses, but why shouldn't it? It seems like the rats should have some information stored in their hippocampi already. Surely some of the weights are already set at high levels. Why wouldn't ZIP knock those off? I'll get back to you when I figure all that out. Perhaps memories really are consolidated, but it just takes longer than a month. That doesn't mean they have to leave the hippocampus, but perhaps the maintenance mechanism changes such that it isn't reliant on PKM-zeta activity anymore.
I don't know if the anyone but the Malinow lab is capable of assaying the state of S818, but I think it would be really interesting to take a look at the correspondence between PKM-zeta levels and S818 status. You could directly control the weight of a given synapse by modulating the local amount of PKM-zeta to determine what percentage of newly generated AMPA receptor subunits get modified at S818 and inserted into synapses. I should say that I am not implying that PKM-zeta is the only determinant of synaptic strength. One should be skeptical about 'magic bullet' hypotheses. No single molecule is going to do the trick. Also, while these studies make a strong case that LTP may be more than just a model of memory, the link still is not complete until we can perform an engineering feat using the theory. Can we actually implant a memory in an animal by inducing LTP at the proper synapses? From a 2003 paper by Richard Morris (that you can download for free if you want):
The fourth criterion, surely not yet met, is mimicry: were it feasible to alter the pattern of synaptic weights in a network in an appropriate manner, the animal should behave as if it remembered something that, in practice, had not happened. Tim Bliss calls this the 'Marilyn Monroe' criterion. This weakness of the available data apart, a rich array of physiological, pharmacological, molecular engineering and other techniques, allied to behavioural studies, have now tightened up the link between activity-dependent synaptic plasticity and memory to a point where it is reasonable to set aside a scientist's natural scepticism about the central principle.