Daniel V. Madison, Ph.D.
Stanford University School of Medicine
We have been working over the last couple of years on electro-physiology of very small populations of synapses, and what we have gleaned about learning mechanisms from those studies. By way of an introduction, I want to give you a very brief and incredibly oversimplified view of how we think memories are made and broken in the central nervous system.
You have a brain, and when objects appear or events appear out in the world you perceive those objects. If they are important enough to you, they can cause the formation of memory trace in your central nervous system. The mechanism, by which this memory trace is thought to be formed, again in a very oversimplified way, is that the sensory input causes a high-frequency activation of neurons in your brain, which lead to the formation of a memory trace. Conversely, if those same inputs receive a sufficient amount of low-frequency activation, that can lead to the loss of the memory trace. We believe that these are models, or at least the beginnings of a model, of how one might go about retraining a memory, storing a memory, or forgetting a memory.
The models on which the people have focused over the past 10 to 20 years are known as long-term potentiation and long-term depression. Long-term potentiation is the persistent increase in synaptic strength in cortical tissues of the central nervous system. Of particular interest is the study of the hippocampus, which is the center for memory consolidation. The strength of the synaptic connection between two hippocampal pyramidal cells is fairly stable over a long period of time. If the input that is producing that synaptic response receives a high-frequency activation,
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 42
The Markey Scholars Conference: Proceedings Electrophysiology of Synapses Daniel V. Madison, Ph.D. Stanford University School of Medicine We have been working over the last couple of years on electro-physiology of very small populations of synapses, and what we have gleaned about learning mechanisms from those studies. By way of an introduction, I want to give you a very brief and incredibly oversimplified view of how we think memories are made and broken in the central nervous system. You have a brain, and when objects appear or events appear out in the world you perceive those objects. If they are important enough to you, they can cause the formation of memory trace in your central nervous system. The mechanism, by which this memory trace is thought to be formed, again in a very oversimplified way, is that the sensory input causes a high-frequency activation of neurons in your brain, which lead to the formation of a memory trace. Conversely, if those same inputs receive a sufficient amount of low-frequency activation, that can lead to the loss of the memory trace. We believe that these are models, or at least the beginnings of a model, of how one might go about retraining a memory, storing a memory, or forgetting a memory. The models on which the people have focused over the past 10 to 20 years are known as long-term potentiation and long-term depression. Long-term potentiation is the persistent increase in synaptic strength in cortical tissues of the central nervous system. Of particular interest is the study of the hippocampus, which is the center for memory consolidation. The strength of the synaptic connection between two hippocampal pyramidal cells is fairly stable over a long period of time. If the input that is producing that synaptic response receives a high-frequency activation,
OCR for page 42
The Markey Scholars Conference: Proceedings the strength of that synapse instantly strengthens, and that strength of synaptic transmission increase is persistent. This is known as long-term potentiation, and is the most compelling model that we have for how memory traces may be stored in the central nervous system at least on a fairly short-term basis of a few hours. Conversely, if you give those same inputs low-frequency activation for a period of several minutes, you can cause a decrease in the strength of synaptic transmission. This essentially erases the memory trace that is being stored in the neurocircuit. Memories are stored by the increase of synaptic transmission in neurocircuits, and perhaps are forgotten by the decrease in transmission in those neurocircuits. Both of these events are induced by activity that arrives at the hippocampus, either sensory activity or the lack of sensory activity. Synapses have two parts, presynaptic and postsynaptic, that are from two separate neurons. The presynaptic neuron puts a synaptic terminal onto a postsynaptic process of another neuron, and these two neurons communicate with each other via the release of a chemical neurotransmitter—excitatory, glutamate-mediated transmission. There are several subtypes of postsynaptic glutamate receptors. One of the subtypes is an AMPA receptor, and another is a NMDA receptor. Both of these receptors are found in the postsynaptic membrane, and both are ionotropic receptors. They open the IM channel when glutamate binds to the extra cellular binding domain. And then current will flow into the postsynaptic cell causing a synaptic current or a synaptic potential. These two receptors differ in two important ways. First, the AMPA receptor is not voltage-dependent. That is, it will open any time glutamate binds to it, regardless of the membrane potential of the postsynaptic cell. The NMDA receptor is voltage-dependent. It will open only when the postsynaptic cell is depolarized. In fact, the NMDA receptor opens, but it is blocked by magnesium ions in a voltage-dependent way. Only if you can depolarize the postsynaptic cells efficiently, can you actually get current flow through that NMDA receptor. In summary, AMPA receptors are not voltage-dependent, and NMDA receptors are voltage-dependent. The second important difference is that the AMPA receptor, when it carries current across the postsynaptic membrane, carries primarily sodium and potassium currents, but no calcium. The NMDA receptor, on the other hand, carries sodium, potassium, and calcium. The triggering factor for both long-term potentiation and long-term depression is the calcium that comes into the NMDA receptor. When a lot of calcium comes through the NMDA receptor the strength of the synapse increases, and, conversely when only a little calcium comes through, the strength of the synapse decreases. Finally, there is a pool of AMPA receptors, which are tethered in some intracellular compartment, which are not on the surface,
OCR for page 42
The Markey Scholars Conference: Proceedings and, therefore, do not detect the glutamate, which is coming from the presynaptic cell. When you subject a cell to high-frequency stimulation, calcium is coming into the NMDA receptor. This calcium causes the untethering of the intracellular AMPA receptors, which then are inserted into the membrane. The population of AMPA receptors in the membrane is now greater. The glutamate, which is released by the presynaptic cell, now is detected more efficaciously, and the strength of the synapse is increased. That is long-term potentiation. And those AMPA receptors dwell persistently in the membrane as long as you do not take them out by some other manipulation. Conversely, if you apply low-frequency stimulation to the synapse, typically at about a rate of about 1 hertz, the AMPA receptors will then come back out of the membrane and go back to their intracellular tether, decreasing the strength of the synapse back to where it was before you potentiated it. This is the basic mechanism of long-term potentiation and long-term depression. The reason why you get a greater calcium influx during high-frequency stimulation is simply that the AMPA receptor is activated repetitively and it summates the depolarization of the postsynaptic cell, therefore letting a lot calcium into the NMDA receptor. With one-hertz stimulation there is far less summation and far less calcium enters through the NMDA receptor, resulting in long-term depression. In the hippocampus, a fairly large proportion of synapses are so-called silent synapses. Silent synapses are basically synapses that do not have any AMPA receptors in their postsynaptic membrane. They have AMPA receptors, but they are all tethered intracellularly. When glutamate is released by a silent synapse there is no response in the resting membrane potential of the cell because the AMPA receptor is not there, and the NMDA receptor will not pass current at the resting potential. You can see a synaptic current at a silent synapse only if you depolarize the postsynaptic cell to unblock the NMDA receptor. It is not exactly silent as it is actually releasing transmitter. It is more like it is deaf. The postsynaptic cell cannot “hear” the release of the transmitter. These silent synapses are really an important theme for the rest of this talk. Long-term potentiation has been studied intensely for about thirty years. There were a number of questions, which have proven to be relatively intractable in terms of trying to figure out the mechanism, e.g., AMPA receptor movement, which is a fairly recent finding. But even more basic questions really were not understood. The reason why it was so difficult to answer these questions was that the techniques that were available depended on the study of large populations of synapses. When people were trying to understand IM channel function with whole cell currents, all they could get was a broad outline of what permeabilities
OCR for page 42
The Markey Scholars Conference: Proceedings were and what kind of things IM channel did. It required single-channel recording, where you could record from a few IM channels at once, to really understand the details of biophysical mechanism. We knew that the way to get at some of these intractable questions was to assess just a few synapses. We knew for a long time how to do this in principle, which was to record from just two neurons at once. The strategy was to do whole cell recording on just two neurons, which were connected synaptically. This way you can get down to the very minimum number of synapses that you could record at once. For some time everybody understood that this was the way to proceed. The problem was that in the brain, even in fields that are meant to be synaptically connected, any two cells are only connected somewhere between one and 5 percent of the time. Consequently, finding these synaptic connections was quite difficult. Postdocs and graduate students are smart even before you teach them anything. If you tell them that there is a great study that is going to solve a lot of problems, but it is going to fail 99 percent of the time, they will be reluctant to conduct the experiment. I do not blame them. We solved this problem by using a cultured hippocampus slice preparation. Hippocampal slices have been in use for a number of years. A hippocampal slice, can be kept in culture for about a week before we record from it. The organotypic slice retains all of the architecture of the hippocampus and it maintains all the appropriate connections that you see in the hippocampus, with one difference: the cells between the neurons in these organotypic slices hyperconnect, that is, they form more synapses than they should. They are appropriate synapses and they go to the right cells, but they make more connections than is usual. The advantage of this is that it increases the connection frequency between two cells to about 50 percent. So now postdocs and graduate students will actually imagine doing the experiment. We developed this technique from the work of Dominic Muller. For this particular application we simply picked two cells in the CA-3 region of hippocampus because these make synapses with each other and we recorded from the two cells simultaneously You can put your electrodes down on any two cells, and see if are they connected, and about half the time they are. If they are connected and if you depolarize one of the cells, it causes an action potential, and if it is synaptically connected it will cause a synaptic current to be produced in the postsynaptic cell. Now these connections are not necessarily unitary. Generally speaking, the number of synapses between one cell and another cell in this preparation is between about one and five. Even though we are using small population of synapses, sometimes
OCR for page 42
The Markey Scholars Conference: Proceedings one, it is difficult to know which is which in any given experiment. We have to assume that there are a small number of multiple synapses when we do these experiments. Now, interestingly, we also found that sometimes it would appear that you did not have a connection. That if you stimulated the presynaptic cells nothing would happen in the postsynaptic cell. However, in about 40 percent of those cases, the cells were actually connected. They were connected by silent synapses. One issue that immediately developed was to identify which synapses would be active and which would be silent. We developed a test to see whether the connections of a group of synapses were mixed—active and silent—or whether they were always all active or whether they were always all silent. Every once in a while Mother Nature does you a favor, and this is one case. There are never many mixed populations in these small numbers of synapses; they are almost either all active or all silent. This means that in experiments we could treat a group of synapses as one, which gave us a great experimental advantage. In the hypothetical case where you have a mixture of silent synapses, those with no surface AMPA receptors, and active synapses, those with AMPA receptors in the cell membrane, the synaptic failure rates differ at different membrane potentials. Each one of these presynaptic synapses will release at some probability less than one. Recording from a group of synapses at once, you will obtain a rate, which represents the probability that all will fail at once. This happens about half the time that you stimulate. In the case of a mixed population, only two synapses will be seeing the glutamate release at a hyperpolarized potential, but all will be seeing the glutamate release at depolarized potential. So if they are mixed, you should see a higher failure rate when the cell is hyperpolarized when you cannot see the silent synapses, and a lower failure rate at depolarized potential when you can see all the synapses. And conversely, if all the synapses are active, then the failure rate should be the same regardless of the membrane potential. The answer, as I already told you, is that the failure rates are not significantly different at depolarizing or hyperpolarized potentials. When you see an active connection, where you actually see a synaptic response, these connections are all active and there are no silent synapses hiding in there. We can find cells with no active synapses. If you take and record from a presynaptic cell and a postsynaptic cell at the same time, and then go about stimulating the presynaptic cell with an action potential repeatedly, in about half the cases you see no response from the postsynaptic cell in trial after trial. About 40 percent of these cases are connected by silent synapses. They can be identified by depolarizing the postsynaptic cell and looking for the NMDA mediated synaptic current. We started
OCR for page 42
The Markey Scholars Conference: Proceedings with a connected pair of cells, but connected by nothing but silent synapses. When this is subjected to this stimulus, which potentiates the synapse, the silent synapses will be awakened and essentially potentiated. When we put a memory-producing stimulus on a pair of cells, you can see they are responding to almost every action potential. You can convert silent synapses to active synapses by providing the proper kind of activity. As a control, in the other 60 percent of connections where you see nothing, you also see no NMDA response, and you cannot awaken them. They are not connected at all. Synaptic connections between pyramidal cells can be made entirely of silent synapses. Long-term potentiation can be expressed via the unsilencing of these synapses. This occurs by the postsynaptic increase in AMPA sensitivity and both silent and active connections can potentiate. Earlier we thought that only silent synapses could be potentiated and that active synapses already had AMPA receptors. It turns out you can insert more AMPA receptors into an already active synapse; you can grade potentiation even in a single synapse. Moreover, we began to understand that there were three different states of the synapse: 1) the silent synapse state, 2) the active synapse state, and 3) the potentiated state. We are trying to understand how these different states are related. Long-term depression is the activity-dependent decrease in synaptic potential. We wondered whether we could see this in our pairs of connected cells. The answer is yes, we can. If you provide a synaptic connection between just two cells with low-frequency stimulation for a period of 10 minutes, you can suppress the strength of that synaptic connection in a permanent way, making long-term depression the state of the synapse. We obtained a depressed state, taking an active synaptic pair and depressing it by providing low-frequency stimulation. This is mediated by activation of the NMDA receptors. In addition, we found that you can actually return synapses to their silent state, if you gave a longer period of low-frequency stimulation—20 minutes instead of 10 minutes—you could drive about half of the synapses back to being completely silent. There are about 125 traces in here after receiving a low-frequency stimulation without a single AMPA response in there. The NMDA response is still there, so we have not killed the synapse, it has just been silenced. Synapse silence was thought to just be a developmental step. When they first formed, synapses were silent, and then at some point became active and were then forever active. However, we have shown here that you can actually move back and forth between the active and the silent stages. There is another form of synaptic depression called depotentiation where instead of simply depressing the synapse from baseline, you potentiate it first and then give a low-frequency stimulation, and return it
OCR for page 42
The Markey Scholars Conference: Proceedings back to its baseline. We raised the question: was this simply the same thing as long-term depression or was it a distinct process? We found that, in one sense, this is a distinct process because it does not depend on the NMDA receptor, as does the long-term depression. If you apply the NMDA receptor antagonist when you try to depotentiate a synapse, it depotentiates just fine. If you apply another kind of receptor antagonist, such as the so-called metabotropic glutamate receptor, it completely blocks the depotentiation of synaptic transmission. So, at least in this sense, there are actually two forms of depression, the long-term depression and the depotentiation. One is mGluR dependent; one is NMDA dependent. We found that even if we waited for some longer period of time before applying the low-frequency stimulation, that this depotentiation was still mGluR-dependent. Even if we waited an hour, which was about as long as we could manage in this experiment before we tried to depress the synapse, they were still mGluR-dependent. Within this time period there was no return to NMDA dependence; it was always mGluR-dependent. We wanted to see what would happen if we came from the silent state of the synapse rather than the active state. We started with synapses which were silent, potentiated them, and then applied low-frequency stimulation. What we found was that you cannot depotentiate a previously silenced synapse; they are protected in some way from being redepressed. This does not last forever. If you wait 30 minutes, activate a silent synapse, and then apply the low-frequency stimulation, it can be depressed. They are initially protected from depotentiation and recover very quickly, and within about 5 minutes, they recover the ability to be depotentiated again. This is interesting because silent synapses represent a reserve pool that can take input and store a memory maybe more effectively than an active synapse because they go from nothing to a very large synaptic response. It is interesting that these silent synapses, which may be preferentially storing new information, protect that information from low-frequency stimulation, which is impinging on neurons all the time. This leads to the suggestion that a new memory trace may be protected in some way for at least some period of time over half an hour, which, interestingly enough, is sort of the time period over which memory consolidation is thought to occur. We knew that this recently silent state, which was protected from depression, was transitioning probably somewhere else in the model. But we did not know where to draw the line: was it a potentiated synapse at this point, or was it an active synapse? Where was the point were it recovered its ability to be depotentiated? We tested for that by looking at the receptor dependency of this recovered depotentiation.
OCR for page 42
The Markey Scholars Conference: Proceedings The answer is that the recovered depotentiation of a recently silenced synapse, unlike normal depotentiation, is actually NMDA dependent, and not mGluR dependent. In addition, the other thing to recall is that if you depotentiate, it does not recover its NMDA dependence even after an hour. However, if you depotentiate an active synapse and we know that this depotentiation is mGluR dependent, it immediately switches back to being an NMDA dependent in terms of any further depression. So basically, we know that any time you can return to this active state, that any depression out of that active state is an NMDA dependent. From those two pieces of data we conclude that the vector from recently silent actually goes to active. We have these different states of the synapse: silent, depressed, active, potentiated, and their interrelationships. Our working model is that active synapses have both NMDA and AMPA receptors, and they can potentiate, where they insert AMPA receptors into the membrane. The depotentiation from that state is now mGluR dependent, even though depression from this state is totally NMDA dependent. It raises an interesting possibility that you may be inserting not only AMPA receptors in this case, but also metabotropic glutamate receptors. Silent synapses have no AMPA receptors in their membrane, and when they are potentiated they receive AMPA receptors, but then they transition back into this active state. They are protected from any kind of depression for the first half an hour of their new life. It is important because we know now that we can transition from this active state back to the silent state, which means you can potentially regenerate the silent synapses during the normal activity that the brain undergoes during both frequency activation or during forgetting. This sort of reloads the brain to store new memories, which can then be initially protected when new information comes in and causes a high-frequency activation of the inputs. We also know that the NMDA receptors are also regulated, which produces a kind of metaplasticity of the whole system. The main conclusion that we have from this research is that both active and silent synapses can potentiate and that long-term potentiation can be graded in a single synapse. So all synapses are basically capable of storing memory traces; but silent synapses do it a little bit better. The synapses can exist in different states of plasticity and how they behave depends on what state they are coming from. The synapses move between these states in an activity-dependent manner, and the history of the synapse actually matters for plasticity, and therefore should matter for the way that memory traces are formed. In closing I want to give credit where credit is due. Much of this research emerged from my laboratory and Johanna Montgomery played a prominent role in the study.