LTP in the Hippocampus: A Mechanism for Long-term Memory Storage?

The neural basis for learning is debatably one of the interesting pursuits currently facing neuroscience. An early conceptual scheme describing how this physiology might occur in the brain was contributed by Donald Hebb (Hebb, 1949). He hypothesized that memories were formed by a distribution of synaptic weights over a neural network. The weighting of particular synaptic connections creates a situation where, when one neuron among a network fires, the probability of other neurons in the network depends upon the weights associated with each connection. In this way, he hypothesized that memories are stored in an inactive form, which has been more recently come to be called a dormant engram. Hebb further proposed that the enhancement of such weights in a network could be brought about by "repeated stimulation of particular receptors" leading to an "assembly of association area cells which can act briefly as a closed system after stimulation has ceased."

In 1973, Bliss and Lomo uncovered a phenomenon that closely paralleled the mechanism Hebb proposed to underlie memory formation, which has now become known as long-term potentiation (LTP) (Bliss and Lomo, 1973). The basic idea behind the LTP mechanism is that when a particular synapse receives strong and sustained or rhythmic input, the synaptic physiology will be altered such that subsequent inputs to the presynaptic cell will result in significantly increased responses from the postsynaptic cell. This enhanced responding has been shown, in some cases, to last more than a month. One might ask what exactly the mechanism is that underlies this enhanced firing? Is it an increase in NT release by the presynaptic cell into the synaptic cleft (increased quantal content) or is it an increased amplitude of response by the postsynaptic cell (increased quantal size)?

Since it's discovery, the mechanisms of LTP have been brought to some light. It is proposed that on the postsynaptic cell, there exist two specific types of glutamate receptors, NMDA and AMPA, which are so named after compounds to which they respond selectively. The AMPA receptor is a fast acting ionotropic receptor, which causes short lasting depolarization of the postsynaptic cell by entry of sodium, while the NMDA receptor is slower acting and permits the entry of calcium. The NMDA receptor's response is more complex than that of the AMPA receptor as its channel only permits the entry of Ca2+ if the receptor is activated while the cell is depolarized. This mechanism is proposed to be mediated by an Mg2+ molecule located within the receptor. If the receptor opens under circumstances where the inside of the cell has a negative potential gradient with respect to the outside of the cell, the magnesium ion remains trapped in the channel of the receptor, blocking calcium entry into the cell. However, if the cell is depolarized, the magnesium will move out of the channel into the extracellular space, permitting the entry of calcium into the cell. This calcium entry has been shown to cause stimulation of a variety of calcium dependent enzymes as PLA2, calpain, and kinases.

These intracellular changes, then make at least three changes that probably affect the functioning of the synapse. 1) The distribution of phospholipids across the cell membrane, leading to a change in the AMPA receptor from a low affinity glutamate binding state to a high affinity state, 2) The addition of more AMPA receptors into the cell membrane, and 3) the release of nitric oxide and possibly certain amino acids into the synaptic cleft to alter the NT release from the presynaptic terminal. The former two cause changes in the quantal size, while the latter causes a change in the quantal content. In addition to these cellular properties, a postsynaptic neuron involved in LTP, synapses on other neurons which then provide negative feedback to the postsynaptic neuron (as well as neurons "parallel" to the postsynaptic neuron). This negative feedback is rather fast acting and is what makes it somewhat difficult to induce LTP. This is because when the cell is excited with glutamate, the fast acting AMPA receptors cause the cell to fire which in turn leads hyperpolarization of the cell, via the inhibitory feedback, before the NMDA receptors can activate. Due to this, induction of LTP is best performed by rhythmic firing with bursts occurring at the so-called theta rhythm of 5-7 Hz. This is presumably determined by the length of time for the NMDA receptor to activate and the time it takes for the inhibitory synapse to quiet down.

If one wishes to find a single brain area which serves to form associations, the hippocampus is a natural choice because of the wide convergence of brain areas synapsing onto its "afferent end" as well as its widely distributed projections out of its "efferent end." Afferent projections to the hippocampus converge greatly to the entorhinal cortex (ENT). If fact about 90% of the projections to the hippocampus synapse on the ENT. The ENT then projects to the dentate gyrus (DG) via the perforant path (PP). The principle cells of the DG, the granule cells, then project to the CA3 region. The pyramidal cells of the CA3 then project to the CA1 region. The CA1 then makes some projections to the subiculum and other areas, which then project to distributed brain areas, and some other projections back the ENT.

The role of the hippocampus in memory has been implicated since the late 50's when a patient suffering from severe epilepsy underwent a surgery which effectively removed the majority of the his hippocampal formations (Scoville and Milner, 1957). This patient, known to the scientific community as H.M., exhibited very severe and selective memory impairment. His memory deficit was selective in the sense that he had perfectly intact and retrievable memories from events before the surgery; it was only the ability to form new memories that was destroyed with his hippocampus. This anterograde, but not retrograde amnesia, implies that the hippocampus is involved in the storage of long-term memories, but that the memories are not stored in the hippocampus and the hippocampus is not necessary for the retrieval of previously stored memories. To be clear, H.M. was normal in most all other cognitive processing. In fact, in the literature he is described as being able carry on a normal conversation (as long as it doesn't require his memory of earlier bits of the conversation), which describes that he even has normal short-term or working memory, but clearly lacks the transfer mechanism for long-term storage. This provides a nice window into the role the hippocampus plays in memory formation, though it is important to keep in mind that it is possible that other cortical and subcortical areas besides the hippocampus could have been damaged in the surgery.

Studies of the rat hippocampus have been widely used to model memory storage. Rats, have particularly good spatial navigation skills and accordingly, neurons in the rat hippocampus have been shown to fire selectively to when the rat is at a particular location in a known environment (O'Keefe and Nadel, 1978). Because of this fact, memory guided spatial navigation tasks have been a widely employed in rats to answer questions about memory. In determining the role of the rat hippocampus in spatial learning, Morris et al. (1982) developed a milk maze where a rat is put in a circular pool of opaque water which has a platform hidden just below the surface of the water. The only escape from the water is to find the submerged platform and climb onto it. Run over a series of trials, rats quickly learn to swim directly to the platform with no direct visual, auditory, or olfactory cues from the platform itself. The only cues available to the rat are the spatial cues created by the objects in the surrounding room.

I'm now going to describe in some detail an early experiment run by Morris et al. (1982), because this learning paradigm is used many times in the literature and because it sets up a nice foundation for the following discussion. In the study, Morris bilaterally lesioned the hippocampal formation of one group of rats and performed sham operations on two other groups of rats: one lesioning the small portion of cortex overlying the hippocampus and the other drilling holes in the heads of the animals, but causing no cortical damage. After sufficient recovery, on what we'll call day one, he allowed the rats to swim freely in the pool for 1 min. On day two, a platform was hidden in one quadrant of the pool and rats were put in the pool at different locations. Over the next 26 trials, normal and cortical lesioned rats learned to escape rapidly from the pool, while hippocampal lesioned rats showed a highly significant impairment. On trials 30-41, platforms which visibly protruded from the water were used. These platforms were made with the edges sticking up beyond the middle of the platform so that 1 cm of water was held on top of the platform to control for the reinforcement value. On these trials, the difference in the performance between the groups vanished. Trials 43-50 used the hidden platforms again and the performance of the hippocampally lesioned rats dropped back to the previous levels. Quantitative behavioral analysis was performed on trials 29 and 42, by placing the rats in the pool with no platform for 60 seconds and measuring the percentage of the time spent in each quadrant. This confirmed the above results by showing the control and cortical lesioned rats spending most of the time in the platform quadrant, while the hippocampally lesioned rats distributing the time evenly over the four quadrants on both trials. One might argue that the hippocampally lesioned rats do better on the visible platform task because it is easier rather than because of the spatial redundancy, but in this case at least some of the spatial bias should transfer to trial 42 and it does not.

All in all this experiment gives some indication that the hippocampus is necessary for rat spatial navigation. There are some variables of course which are not addressed by the experiment, namely those involving drive or motivation to complete the task. If the hippocampal lesion reduced the drive state of the animal, it could be that the animal just didn't have enough motivation to find the platform if it couldn't see the platform. Another concern is the relevance that this spatial learning in rats to more global learning in humans. To attenuate some of this concern, others have found learning deficits in other modalities due to hippocampal lesioning (Eichenbaum et al., 1989), leaving us with the usual concern in any animal model.

With all the above in mind, what predictions can we posit that, if confirmed, would build a case for LTP as a mechanism of long-term storage? One general strategy to building such a case is to find correlations between times when learning is occurring and when LTP is occurring. This can be done in a couple of ways. One is by using techniques that are known to either facilitate or impede memory and measure whether they also affect LTP accordingly. Complimentary to this is the reverse method: to disrupt normal LTP induction and observe whether memory formation is impeded. This LTP disruption should only affect memory formation and have no effect on the retrieval of previously learned memories. This is an important control because without it is impossible to tell whether the behavioral deficit is due to a storage failure, a retrieval failure, or possibly neither, but that no conclusive behavior is exhibited due to reduced motivation or other "state" variables. There are two types of LTP disruption that could affect memory. Given our distributed network hypothesis, if LTP were broadly induced in the hippocampus, the weights necessary to form a specific association should be washed out by the noise and thus new learning should be barred. Likewise, if LTP in the hippocampus was blocked, new memory formation, but not memory for previously learned tasks should be impeded.

An interesting report has been given by Bloch and Laroche (1985), showing that stimulation of the reticular formation, which is known to enhance memory, enhances LTP in dentate gyrus of the hippocampus. In this study, a recording electrode was lowered into the rat dentate gyrus and stimulating electrodes were lowered into the perforant path and the medial reticular formation (MRF). Stimulating and recording electrodes were adjusted until maximal positive potentials were encountered. The intensity of the MRF stimulation was determined for each rat by measurement of the arousal threshold using 2 s stimulation on awake and immobile rats. The first indication of arousal was usually indicated by changes in respiration and the experimental intensity was then set at 10% below this threshold. The PP stimulation was set for each animal to the level which produced half of the maximal population spike found in the DG during a control stimulation period. Animals were then subjected to either a control condition or an experimental condition. Each animal was run on both conditions, serving as its own control, after a 15 day rest period in between. In the control condition, LTP was induced by a series of 10 high frequency stimulus trains (400 Hz, 20 ms total duration, eight pulses of 100 µs each, per train), delivered to the PP with 5 min. inter-train intervals. In the experimental condition, the MRF stimulations followed the PP stimulation by 10 s. Post excitatory facilitation in both conditions was measured by a ten second series of PP stimulations taken at varied time intervals. The results of this experiment showed that in the MRF stimulation condition LTP was significantly enhanced for 6 days after induction.

These results are further enhanced by the results of a second experiment. In the second experiment the interval between LTP inducing PP stimulation and MRF stimulation was varied, so the conditions were as follows: -120 s, -10 s, 10 s or 120 s, or no MRF stimulation. This was to explore whether the MRF stimulation only caused significant LTP increase when stimulation was given at a short interval after PP stimulation similar to the way MRF stimulation must be given within the first 90 s of the post-training period to facilitate learning (Laroche and Block, 1982). The results from this experiment showed that MRF stimulation given 10 s after PP stimulation lead to a significant increase in LTP as measured by the increase in population spike and population EPSP over all other conditions. Additionally, there was no significant difference in the other latencies from the control with no MRF stimulation. Histological reports served to verify the locations of the electrodes.

This experiment provides some nice correlative evidence by showing that MRF stimulation at a short latency of 10 s after LTP induction leads to significant LTP increase, while other latencies do not, which correlates with the previous finding that MRF stimulation in the first 90 s of a post-training period facilitates memory. Other correlational evidence for the link between LTP and learning comes from Green et al., showing that an increase in the dentate gyrus EPSP occurs when animals are exposed to complex novel environments (Green et al., 1986). Further, it has been found that the benzodiazapines, which are known for their memory blocking action, block LTP. All of this evidence is entirely correlational, however, and thus one must be clear that other conclusions are immanently possible. For example, MRF stimulation could facilitate learning through a change in reward or in drive, not through memory facilitation at all, which would greatly undermine the support the above experiment provides for the LTP/memory hypothesis. The firing pattern changes due to the novel environment could be explained by increase sensory input or increase motor output and the benzodiazapines could be shutting down large portions of brain function, two of them happening to be memory and LTP in the hippocampus. However taken together, these are interesting and should be noted.

A different line of support comes from "saturating" LTP in the hippocampus and measuring whether or not memory formation is affected. According to our distributed network hypothesis this should disrupt memory formation because synaptic weights will be washed out. Castro et al. (1989) ran an experiment where rats were either given low frequency or high frequency stimulation of the rat perforant path and the animals were tested for the affects in a Morris water maze. High or low frequency stimulation of the perforant path was repeated daily for 19 days and populations spike and population EPSPs were recorded from the fascia dentata. On day 20 through day 34 all animals began receiving low frequency stimulation. The high frequency group showed significantly increased firing through day 26. On days 19 and 34 rats were tested for spatial learning capability in the water maze. The training session consisted of 6 blocks of 2 trials each with each block separated by 2 minutes. Testing was done immediately following training by placing the rat in the pool with the platform removed measuring times spent in each quadrant. Their results show that on day 19, when the high frequency rats showed significantly more LTP saturation in the hippocampus, they spent equal amount of times in each quadrant, while the low frequency rats spent significantly more time in the training quadrant. On day 34 when the firing levels no longer differed between the two groups, the behavior scores also did not differ.

This experiment supports our prediction that while LTP is saturated across hippocampal neurons, synaptic weights cannot be formed and thus memories cannot be formed. This experiment is complimented by another performed by McNaughton et al. (1986) using similar electrophysiological techniques and a different behavioral paradigm. They had one group of rats which learned the task and only subsequently recieved electrical stimulation (either high or low) and one group which received electrical stimulation after each training period. They found that the high frequency rats performed as well as the low frequency group when the training was before stimulation, but much poorer than the low frequency group when the training was concurrent with the high frequency stimulation. This being one of the major short-comings of the Castro et al. experiment, these two add some more support to the LTP-learning hypothesis.

Another test of the LTP-learning hypothesis comes from the infusion of the hippocampus with NMDA receptor blocker. Davis et al. (1992) did a thorough analysis of this procedure by using minipumps to infuse the hippocampus with varying concentrations of AP5 (a selective NMDA receptor blocker), while recording from the DG and performing microdialysis to verify concentration of AP5 and testing the rats for spatial learning in the water maze. In the first experiment, Davis et al. found that the concentration of AP5 related linearly with the performance on the water maze. Namely, in the low concentration, animals performed similar to controls, and as the concentration increased performance dropped. This performance on the behavioral task is paralleled by the finding that at low concentrations of AP5, the firing of neurons in the hippocampus was very similar to that of the control and as concentrations increased, the firing decreased. Davis et al. note that at the highest concentration of AP5, the rats showed signs of sensori-motor disturbances. Although the lower concentrations didn't show any disturbances, the authors ran a second experiment using a visible platform to test whether there could be subtle motor disturbances that account for the data. In this task the animals performed almost identically, supporting that it is spatial memory that is being affected by the AP5. A final experiment in this study showed the rats infused with AP5 performed slightly better than, but not significantly different from hippocampally lesioned rats.

Taken as a whole, these data make a fairly nice story to present for the role of LTP in long term memory. This story though, is not as direct as one might want though, with a lot of correlative evidence and possible alternative interpretations. Because of this, the story is somewhat fragile in the sense that evidence against it must be taken seriously. With that in mind, Bannerman et al. (1995) present some very powerful data in refuting LTP's role in memory. Bannerman et al. employ a paradigm very similar to Davis et al., but they include as another variable whether the animal was spatially pretrained, non-spatially pretrained, or not pretrained as in previous experiments. Spatial pretraining included six trials in a water maze in a different room than later training. Non-spatial pretraining included six trials with a thick curtain pulled around the maze so that outside spatial cues are inaccessible to the rat. After pretraining, animals were fixed with a minipump and physiological recording equipment and run on the water maze task similar to described above. As before, the AP5 almost entirely prevented the induction of LTP in the DG. However, despite the fact that LTP was almost entirely blocked, the spatially pretrained and to a lesser degree, the non-spatially pretrained animals performed significantly above chance in the maze as measured by the percent of the time the animal spent in the correct quadrant. To determine whether spatially pretraining eliminates the need for the hippocampus in spatial learning, the same spatial pretraining was given and hippocampal lesions or a sham surgery or no surgery was given. Hippocampally lesioned animals performed significantly worse than controls. Interestingly, as a brief note, the controls in this second experiment seemed to learn slightly more slowly than the AP5 treated animals in the first experiment. This is kind of weird, but the rats do learn the task rather quickly still and this may not be anything real.

This data presents real trouble to the theory that LTP in the hippocampus is a mechanism for memory. Cain et al. (1996) have replicated these results and followed up with detailed behavioral analyses of the rats performing in the water maze. In this paper, Cain shows that there a number of things which could lead the formerly seen performance deficit in NMDA receptor antagonist treated rats. He noticed that AP5 treated rats sometime bumped into, but did not climb onto the platform. Additionally rats would sometimes actually swim directly over the platform. In fact AP5 treated rats had a significantly higher percent of contacts that were deflections or swimovers. He also noted that when the AP5 treated rats were picked up, they would often continue to make limb movements that resemble swimming movements. These effects were eliminated when the rats were given spatial or nonspatial pretraining. Cain offers that this sensori-motor disturbance observed here may be able to account for the findings from previous studies which show that NMDA receptor antagonists do impede memory formation. This study in conjunction of the Bannerman et al. serves to highlight the need to reevaluate the hypothesis that LTP is necessary for normal memory formation. They show that LTP blockage does not necessarily block memory formation and the previous findings that it may have been due to sensori-motor disturbances. In combination with the fact that the evidence supporting LTP is built up somewhat tentatively, this casts doubt on the hypothesis that LTP underlies memory.

However, in the face of this, posing a small doubt to our doubt, Nicoll and Malenka (1995) made an interesting discovery that on the mossy fiber synapses in CA3, LTP can be induced in a NMDA-receptor independent form. They hypothesize that this form of LTP is mediated via the presynaptic terminal rather than the postsynaptic cell. This might suggest that the NMDA antagonists we are using to eliminate LTP in the hippocampus may not be stopping all of the LTP. One might argue that this is the reason that learning can still be observed under NMDA antagonists. This may make it even harder to say for sure that the differences found above are differences in LTP's role and the hippocampus' role in memory. However, before going too far with this, one must be careful that the current hypothesis for memory formation still includes a distributed network. It seems like it might be a little far fetched to say that all of our proposed weighting is being done in the CA3 while the other areas aren't necessary.

Whishaw et al. (1995) also present an interesting finding that it may be possible to train rats to successfully complete the water maze even after hippocampal lesions. First they replicated the results that rats with fornix lesions were severely impaired when trained normally on a water maze task. Then they followed with an experiment using a more elaborate training procedure in which control and lesioned animals were first, once a day, for six days, were placed on the platform. Following, for four days, rats were placed just next to the platform and allowed to climb on. Following, 10 trials of normal training was performed with the rats. After all of this training rats were tested as normal and it was found that the hippocampally lesioned rats performed at the same level as the control rats.

This somewhat startling evidence indicates that need to reconsider our hypothesis that even the hippocampus is necessary for memory formation. These data certainly suggest that learning can indeed occur without the hippocampus. In discussing their data, Whishaw et al. attempt to make a distinction between two global types of memory. Namely they describe that the rats with the hippocampal lesions may be "getting there" but not "knowing where" that is. They talk about it as though the rats can acquire the skill of getting to the platform, but still do not "know" where the platform is located. This distinction is an interesting one and one that has been described thoroughly in the neuropsychology literature as "knowing how" vs. "knowing what." (Cohen and Squire, 1980). This distinction is one that has actually been around for a while, with its first indication coming from the previously described case of H.M. Even though H.M. could not remember explicitly anything that happened to him after his operation, he could still become better at certain "skill" tasks including reading mirror reflected words or tracing complex geometric shapes. It is possible then that the apparent discrepancy seen between the LTP and memory formation is that there are two (or more) systems by which memories are stored. This possibility is one that is paid very careful attention by some authors and not nearly as careful by others. As Whishaw points out, the milk maze has two components that can contribute to the animal's successful performance of the task, namely the "knowledge" of where the platform lies as well as the "skill" of how to actually get there.

This in mind, it appears that the question -- is LTP in the hippocampus the mechanisms of memory storage? -- is not specific enough. We must move to a more complex model and ask if this mechanism underlies a particular, explicit type of memory. However, how exactly does one go about testing for this type of memory, especially in animals? How does one define explicit memories without describing them as those things which are destroyed when one destroys the hippocampus or when one stops LTP? As one starts to make these fine distinctions, it becomes hard to have entire faith that the differences that are appearing are not merely due to species differences. These are the things that must be attended to in order to investigate the role of LTP in the hippocampus. As it stands now, there is some fragile correlational data on the table pointing indicating that LTP might underlie memory as well as some under-specified data indicating that LTP, if it does have a role in memory, it is not the mechanism. The arguments presented here aren't strong enough to convince me LTP does underlie memory, but the converse data aren't defined well enough to convince me that LTP doesn't underlie some particular part of memory.