|This text was taken from
P.A.Wozniak, Economics of learning, Doctoral
and adapted for publishing as an independent article on the Web. (P.A.Wozniak, Apr 23,
For a more up-to-date text on two-component model of long-term memory see: Hypothetical molecular correlates of the two-component model of long-term memory by E.J.Gorzelanczyk and P.A.Wozniak
By identifying molecular correlates of the attributes of long-term memory manifested at the behavioral level, I would like to strengthen the evidence in favor of the universal nature of repetition spacing in achieving sustained levels of knowledge retention, as well as for the inevitability of its application in modern knowledge access systems that account for imperfectness of human memory and cognition
Item difficulty, expressed by the concept of E-factor, is a property of the item in question, not the memory system. In other words, items will differ in the number of synapses involved in storing memories, structure of inter-neuronal connections, etc. rather than volume or chemical nature of neurotransmitter, conductive properties of the synapse, etc. It is easy to notice that the same intricate items will cause retention problems in both capable and less capable students. On the other hand, all students will experience problems with remembering some items. As a rule, the main factor that determines the difficulty of an item is its complexity. Invariably, simple items sport higher E-factors than items that are enumerative or descriptive in nature. An interesting question arises: what is the biological basis of item difficulty?
As storing information in a synapse may be compared to storing a single bit of data rather than the entire message, items cannot be coded for by a single synapse. It is rather neural cell assemblies, more accurately called synaptic patterns, that will be responsible for storing complex information structures such like spelling of a foreign word, definition of a concept in economics, mathematical formula, etc. Naturally, we might expect that an increasing complexity of an item, associated with an increasing information content, will result in an increasing number of synapses involved in preserving the underlying memory engram. Different items are remembered by different subsets of synapses, with item difficulty reflected in the complexity of the involved neural network. The second question arises: why an increasing number of synapses might cause learning difficulties? After all, during a repetition, all synapses might undergo parallel processes that would result in parallel and symmetrical increase in synaptic strength.
The two following phenomena are likely to play a role here:
- the synaptic stimulation is not uniform over a number of repetitions
- a larger number of synapses increases the probability of inter-item interference
The evidence for non-uniform stimulation of synapses comes from the very simple observation that in case of enumerative items, i.e. items that ask for recalling a set of subitems, the order of recall can be observed as different at different repetitions. Like in slow motion, one can even experience the browsing effort of ones consciousnesss search for the correct components. Not by coincidence, these are enumerations that make up probably the most notoriously intractable class of items. As for interference, its contribution to linking difficulty with the size of the neuronal assembly is certain though it does not seem to be a major factor (note, that in a substantial majority of cases, difficulty is not context dependent, while interference is).
In the subsequent chapter, I will often refer to the concept of synaptic extraction. As I will argue in later section, in the course of repetitions, the number of synapses involved in eliciting the response is gradually reduced. One of possible reasons may be the lack of uniform stimulation of the particular components of the synaptic pattern. In other words, in the course of the learning process, only the synapses that are directly stimulated in each of the repetitions retain memory traces and the increased conductivity. This decline in the number of connections I will call the pattern extraction. In the chapter devoted to knowledge representation issues in business and economic applications, I will try to show a number of techniques by which minimum loss of information is achieved through appropriate representation of the learned knowledge.
In the following sections, I will try to look for possible correlates of behavioral regularities in the process of learning at the molecular level. In particular, I will attempt to make use of the earlier formulated guidelines that should be used to identify the correlates of the two components of long-term memory: retrievability and stability.
Advances in molecular research of memory
Over the last four years, i.e. since I first attempted to formulate a tentative molecular model of long-term memory (Wozniak, 1990), a very dynamic growth of interest in memory and learning has been marked. This resulted in rapid development of new laboratory techniques that can shed more light on those mechanisms that for long had been persistently intractable to study. Indeed, the molecular basis of memory remains one of the very few bastions of molecular biology in which no coherent models of the biochemical and cytophysiological phenomena has as yet been established. The difficulty in studying memory comes from the inherently complex structure of the nervous system and lack of non-invasive techniques that could make it possible to easily monitor the molecular phenomena occurring during the learning process. On the other hand, study of the long-term synaptic potentiation in hippocampal slices that show survivability of just several hours cannot serve as a good substitute for the research on intact animals that could develop memories lasting for months and years. Nevertheless, the recent inflow of new findings has indeed been impressive, and it seems that in a not so distant future, the most basic elements of the molecular mechanisms of memory shall be unraveled. The present status of the research field is still under shaky influence of new finding and shifting focus of attention. A number of labs are forging research paths going in different directions. However, there is a broad consensus that the study of long term potentiation of the synaptic transmission in the cells of the CA1 region of the hippocampus can serve as an excellent source of information for formulating the ultimate model of memory and learning in humans. After the initial flurry of research on NMDA receptor, nitric oxide synthase and protein kinase C, there is an increasing interest in metabotropic glutamate receptor, cAMP, protein kinase A and gene expression in the nerve cells subject to conditioning. One can expect a surge of interest in new molecular factors as soon as new data and observations flow in. Some of the highlights of the most widely established findings are presented in the following section. Though no coherent picture will emerge from the presented facts yet, the progress that resulted from the research over the last four years seems indeed staggering.
General observation in reference to memory and learning
It has been for long a very strong claim that memory mechanisms have probably been strongly conserved in the course of evolution. This led a number of researchers to study memory in invertebrates in order to find inspiration for similar research in mammals. Indeed the research done by Eric Kandel in Aplysia, as well as the research of David Alkon on Hermissenda led to establishing the focus of memory research on protein kinases and protein phosphorylation. These mechanisms will be discussed in later sections.
A great deal of research has been done on establishing the key neurotransmitter involved in memory consolidation. It has been earlier found that hormones such as noradrenaline, acetylcholine, serotonin, dopamine and bradykinins, all improve the performance in learning tasks. This effect has largely been attributed to the modulatory action of the above transmitters. However, there seems to be no doubt that the key role in synaptic transmission subject to potentiation is played by glutamate, a major excitatory neurotransmitter in the hippocampus, and not only. Glutamate and the molecular cascades triggered by its receptors will remain in the focus of attention over the rest of the chapter.
There seems to be no more interest in electrical aspects of memory. The idea of reverberating impulsation in the limbic system has lost most of its followership. The anatomical evidence as well as the finding that the electroconvulsive shock obliterates memories have been the major reason for establishing the reverberating circuit model. However, it has been justly observed that neuronal activity that corresponds with ECS can equally well influence the balance of signal factors in the synapses, and consequently, affect the storage of short-term memories. As in ECS, memory loss occurs also in head injuries, anoxia, hypothermia, anesthesia, administration of ouabain, etc. All these factors can affect the cytochemical status of the synapse.
Direct evidence has recently been provided for the fact that consolidation of memories occurs during sleep [see also: Good sleep, good learning, good life]. An interesting extension of the above is an observation that an increased learning effort sparks greater demand for sleep; a fact of enormous practical consequences for those who want to learn effectively. It has also been confirmed that amino acid uptake increases in the brain during the REM phase of sleep. This additionally supports the important finding that protein synthesis forms a crucial step in the establishment of long-lasting memories. The findings in reference to protein synthesis in memory consolidation will be discussed later, here I would only like to mention that it has been suggested that the size of neuronal nucleus increases as a result of repeated conditioning providing one of many indicators of the central role of gene expression in learning (see later).
In reference to pattern extraction discussed in the chapter devoted to the principles of knowledge representation for the purpose of repetition spacing, it has been found that the number of synapses involved in learning a motor tasks gradually decreases as the learning continues. Similarly, activity of the brain has been found to be higher at the beginning of a learning task that at repetitions. The latter observation might, however, reflect the greater biochemical activity involved in establishing memories that in sustaining memories.
Finally, there has been more research confirming that enriched environment indeed increases the size of dendritic fields and neural branching (Alkon 1989), which every student should interpret as the sure indication that learning, apart from the establishment of specific memories, enhances the brains ability to process information and form new memories.
It has been found that long-term potentiation can be induced in temporal lobe (Squire et al. 1991), striate cortex (Bear 1991), medial septal area, and other structures of the brain. Invariably, however, most of the attention has always been focused on the hippocampus. The attention was brought to this MTL structure upon discovering severe amnesia induced in animals and humans with lesion in the related areas. Additionally, laboratory procedures have been well established for inducing LTP in hippocampal slices in vitro. As a consequence, most of the new molecular data referring to memory comes from labs that chose LTP in the hippocampus as the primary investigative model of memory in humans. For example, eyeblink conditioning has been identified as dependent on potentiation of CA1 neurons of the hippocampus. In eyeblink experiments, animals are learned to associated a sound signal with a puff to the eye. Lesions to the hippocampus abolish eyeblink conditioning in animals (Disterhof 1994). In the nictitating membrane reflex involved in eyeblink conditioning, impulsation from the eye and the ear meet ultimately in CA1 cells of the hippocampus (Alkon 1989). Although a large proportion of cells in the CA1 region show conditioning, it is supposed that specific molecular marking results in strengthening synaptic conductivity in only those dendrites that are responsible for storing particular memories (Alkon 1989). It has been found that learning increases firing rates in the hippocampus just before the appearance of the response (Disterhof 1991). The often quoted hypothesis is that the hippocampus may play a role in associating impulsation coming from different areas of the brain (Eichenbaum et al. 1991). This was concluded among other things because of the observation that the same cells can fire during olfactory and spatial responses. It has even been suggested that memories move from the hippocampus to the cortex during the consolidation phase of sleep. To support such a claim, it has been observed that destruction of the hippocampus does not indeed destroy memories. It rather prevents learning new things (Eichenbaum et al. 1991). However, the mechanisms of mapping hippocampal synaptic patterns to the putative cortical synaptic patterns, as well as the hypothetical transfer mechanism are entirely unclear [for new insights see: Good sleep, good learning, good life]. Although the technique of preserving hippocampal slices in vitro has been well developed and makes it possible to conduct experiments lasting up to several hours, the critical phase of LTP called L-LTP (late LTP) cannot be effectively studied in hippocampal slices (Kandel 1993). Hence the disproportionately large number of studies of short-term potentiation and E-LTP (early LTP) as compared with later stages of LTP dependent on protein synthesis and gene expression.
Although glutamate has been established as the main excitatory neurotransmitter in the hippocampus, cholinergic pathways do also terminate at CA1 cells of the hippocampus. As acetylcholine has been found as a memory enhancing transmitter, considerable amount of attention has been brought to the place of the cholinergic system in development of memories. A cholinergic antagonist scopolamine impairs learning, while physostigmine which blocks acetylcholinesterase improves learning. Similarly, activity of acetylcholinesterase has been found to be lower in the areas of reduced mnemonic capacity. Additionally, enhanced cholinergic activity improves memory, while high affinity choline uptake has been found to correlate positively with the amount of learning (Mishkin et al. 1991). Stimulating GABA-ergic neurons in medial septal area reduces the firing of neurons in the hippocampus, while at the same time cholinergic neurons increase the firing. Interestingly, caffeine and teophiline increase the number of cholinergic receptors in nerve terminals (via inhibiting cAMP phosphodiesterase and increasing the levels of cAMP), and both these stimulants have been found to have positive influence on establishing memories (though this positive effect may be partially occluded by the attention deficit as a result of increased excitability of the cortex and the limbic system). It has been found, finally, that it is the arousal impulsation from the reticular formation that ends up in CA1 cells as acetylcholinergic input (Alkon 1989). The evidence collected until now has led many researchers to believe that though acetylcholine plays an important role as a memory modulator (cf. the arousal impulsation), it does not take part in the process of establishing memories per se. As for the mechanism of cholinergic action, it has been found that the muscarinic receptor activates protein kinase C (Craig 1993). As it will be shown later, protein kinase C has been proposed as one of the central elements of the control system responsible for consolidation of memories.
Development of memory proceeds as a cascade of molecular phenomena. Depending on the methodology, different authors distinguish different phases of synaptic potentiation that are given overlapping and often confusing names. Bliss and Collingridge propose the following division of the potentiation phenomena (Bliss et al. 1993):
- post-tetanic potentiation (PTP)
- short-term potentiation (STP)
- long-term potentiation (LTP)
PTP, which is NMDA-independent, seems to be governed by an alternative mechanism as it can develop even if LTP is at its peak (Bliss et al. 1993). STP can be distinguished from LTP by protein kinase inhibitors (Bliss et al. 1993), which do not affect STP while abolishing the development of LTP. Phosphorylation of protein is a necessary element of the development of STP (Kandel 1982), but it is not affected by protein synthesis inhibitors (Kandel 1982). STP is NMDA-dependent, i.e. blocking glutamate NMDA receptors prevents its establishment, while NMDA itself is able to induce STP (Bliss et al. 1993). Additionally, blocking NMDA receptors makes it impossible to shift from STP to LTP (Bliss et al. 1993). The most often quoted duration span for STP is about one hour, and is correlated with an increased release of the neurotransmitter (Bliss et al. 1993). A very interesting recent report indicates that there are two kinds of short-term memory in the lower temporal lobe. One is automatic, and the other can be maintained by conscious rehearsal(Science 1994). Little is known, however, if distinguishing these two forms might in any way affect the formulation of the universal molecular model of memory.
A train of 100 stimuli at 100 Hz can evoke a long-lasting potentiation of synaptic transmission in neurons of selected structures in the brain. This potentiation is called long-term potentiation, or LTP for short. LTP has been elected as the primary experimental model of memory and learning in vertebrates. During LTP, the resting potential of affected neurons appears to be increased resulting in easier and faster depolarization, and in consequence, in an enhanced synaptic response. It has been found that drugs disrupting LTP impair learning (e.g. in odor discrimination studies). LTP can be induced in all excitatory pathways in the hippocampus, but has also been found in other numerous parts of the brain (Bliss et al. 1993). Naturally induced LTP is behaviorally dependent, and it can be detected only in the wake of a selected group of tasks (Lynch 1984). As for its duration, it has been found that LTP can persist for weeks in CA1 cells in freely moving animals, while presynaptic changes last for at least several hours (Bliss et al. 1993). However, in synapses that are cut off the cell body, LTP lasts only for a couple of hours; the hypothetical reason being the lack of new protein supply from the cell body (Bliss et al. 1993).
The widely accepted division of LTP into temporal stages is as follows (Bliss et al. 1993):
- LTP1, blocked by protein kinase inhibitors (including kinases such as PKA, PKC, and PKG)
- LTP2, blocked by protein synthesis inhibitors
- LTP3, blocked by gene expression inhibitors
LTP1 is also called E-LTP (early LTP), while LTP2 and LTP3 are called L-LTP (late LTP). Blocking depolarization during the tetanus blocks LTP entirely (Bliss et al. 1993). LTP1, depending on the source has been shown to last from one to five hours (Kandel 1993). LTP3, which requires gene expression, cannot be established in anesthetized animals (Bliss et al. 1993). Some authors exclude NMDA-independent LTP from the definition of LTP. NMDA-independent LTP occurs, among others in mossy fibers (Bliss et al. 1993), and is not blocked by NMDA antagonists. A phenomenon analogous to LTP is the long-term synaptic depression (LTD). It has been postulated that memories may result not only from synaptic potentiation, but also from synaptic depression. A hypothetical cause of LTD is an elevated level of postsynaptic calcium (Bear 1991).
NMDA receptor as the central factor in establishing LTP
The involvement of glutamate in memory has been suspected for long. It has been noticed, for example, that glutamatergic action is reduced in Alzheimers disease (Lynch et al. 1984) as well as in Huntingtons disease. There are two categories of glutamate receptors: NMDA receptors, which can be activated by N-methyl-aspartate, and non-NMDA receptors that do not get activated by NMDA. It has been shown that the activity of NMDA is increased during LTP (Bear 1991); hence the particular interest in NMDA receptors in the context of synaptic potentiation.
The attractiveness of NMDA receptors as a possible element involved in the cascade leading to the establishment of long-term memories has greatly been enhanced by the observation that it might provide a molecular mechanism for the so-called coincidence-detection rule proposed by Hebb and Konorski in the late 1940s. The coincidence-detection rule says that the synapse that is activated at the moment of the simultaneous depolarization of the target cell might undergo the process of strengthening its conductivity. This rule has been used as the basis of associative learning in natural neural networks, and NMDA receptor has been put forward as the most prominent candidate for the molecular implementation of coincidence detection.
NMDA receptors are associated with calcium channels, but are usually not active in standard synaptic transmission which is effected by non-NMDA receptors that open sodium and potassium channels. The reason for lack of activity is the block of the calcium channel with magnesium cations. However, if the target cells undergoes depolarization, the magnesium ions are repelled from their site and the influx of calcium may take place as a result of activating NMDA receptors by glutamate. In other words, NMDA receptors act as detectors of the coincidence between the target depolarization and the presence of glutamate in the synaptic cleft. The resulting calcium transients have been show to be able to result in triggering LTP (Bliss et al. 1993). Therefore, detection of coincidental activity can be transformed into lasting potentiation. An interesting property of NMDA receptors is that they release glutamate more slowly than non-NMDA receptors resulting in a long-lasting impact on the chemical processes triggered within the synapse. Interestingly, changes induced by the activation of NMDA receptors are both pre- and post-synaptic (Bliss et al. 1993). The explanation for this puzzling effect came from the discovery of retrograde messengers (see later sections). Strong evidence for the importance of calcium transients in evoking LTP comes also from the observation that the release of calcium from intracellular stores can substitute for NMDA-dependent calcium influx into the cell (Bliss et al. 1993). It is possible, that the inflow of calcium into the cell results in an increased activity of NMDA receptors as a result of phosphorylation (similar phosphorylation has also been reported to affect AMPA glutamate receptors)(Bliss et al. 1993). More evidence for the pivotal role of NMDA receptors in LTP comes from the fact that NMDA antagonists (e.g. AP5, MK-801) have been shown to impair learning (Lynch 1984), and to block LTP (Bliss et al. 1993). However, AP5 does not block shock-avoidance learning (Lynch 1984) nor visual discrimination tasks. As mentioned earlier, not all forms of LTP have been found to depend on the activation of NMDA receptors. Blocking NMDA makes it impossible to shift from STP to LTP (Bliss et al. 1993), but it does not reduce previously established LTP. In more detailed study, it was found that application of NMDA receptor blockers just before training impairs learning, but post-training and pre-test administration have no such effect (Venable et al. 1990). Similarly, chemically induced LTP is also not blocked by NMDA receptor inhibitors (Bliss et al. 1993). The agents of LTP most likely lie downstream the activation cascade; hence the lack of abolition with NMDA antagonists. A very puzzling observation is that inadequate stimulation of NMDA receptors not only does not induce LTP, but can even impair the cells ability to exhibit LTP in subsequent tetanization (Bliss et al. 1993). I mentioned earlier that STP is correlated with increased release of glutamate, this increase may possibly last a few hours into LTP (Bliss et al. 1993). Mutant mice lacking CAM kinase II and unable to induce LTP show normal NMDA currents, which suggest the participation of the enzyme in the activation cascade downstream NMDA receptors (Grant et al. 1994). Two interesting facts have also been observed in reference to drinking alcohol and its influence of memory. First it has been found that alcohol inhibits and reduces the number of NMDA receptors in the brain (NMDA.TXT). Secondly, the number of NMDA receptors in babies of mothers that abuse alcohol is markedly reduced (Queen et al. 1993).
Non-NMDA receptors are permeable for potassium and sodium and take part in normal transmission of impulsation via glutamatergic synapse. This receptors also show the ability to activate phospholipase C via protein G, and consequently, result in the increased levels of calcium in the cell. Some authors suggest that LTP may be associated with an increased AMPA component of synaptic response (Bliss et al. 1993) (AMPA is a kind of ionotropic glutamate receptor). Sensitivity of AMPA in LTP is increased, and this increase is blocked by kinase inhibitors (Bliss et al. 1993). Interestingly, catalytic subunit of PKA can directly increase the activity of AMPA (Bliss et al. 1993), but the significance of this fact for memory consolidation has not been elucidated.
The concept of retrograde messengers has been brought to attention of neuroscientists upon the observation that the activation of postsynaptic NMDA receptors increases the volume of the neurotransmitter released from the presynaptic element. In other words, the long held concept of one-way synaptic transmission, from presynaptic to postsynaptic element has been called in question. Since this observation, several possible retrograde messengers have been proposed such as nitric oxide (NO), arachidonic acid, carbon monoxide (CO), etc. In reference to nitric oxide, whose role has been investigated starting as late as in 1991, it has been found that NMDA increases the release of NO from cultured neurons (Bliss et al. 1993). It has been proposed that NO synthase is probably activated by calcium transients (Bliss et al. 1993). Physiological activity of NO drops to 50% in just four seconds, which suggests that nitric oxide might play the role of a rapid retrograde messenger. Interestingly, nitric oxide does not just affect the presynaptic element, it may diffuse to neighboring synapses and increase their potentiation(Science 1994). This finding, however, calls in question its participation in establishing specific memories, and suggests that nitric oxide might play the role of a growth signal that might intensify metabolic processes in highly active areas of the nervous tissue. In the presynaptic element, nitric oxide activates guanyl cyclase, and this effect is proposed as the main possible reason for potentiating the synapse. However, the nitric oxide effect can be seen only in presynaptic elements that are activated at the moment of the release of the retrograde messenger. Hemoglobin, a nitric oxide scavenger, blocks LTP (Bliss et al. 1993). Similarly, blocking NO synthase impairs learning (but not shock avoidance learning). Protein kinase C, on the other hand, activates the enzyme. The role of nitric oxide as a rapid retrograde messenger has been called in question, however, upon noticing that potentiation it induces is slow to develop (Bliss et al. 1993). Moreover, NO synthase inhibitors have little impact on STP (Bliss et al. 1993). As mentioned earlier, inhibiting NO synthase suppresses LTP, but not always. The result of the inhibition depends on the previous activity in the hippocampus (Bliss et al. 1993). Additionally, NO synthase has not been immunocytochemically found in granule cells (Bliss et al. 1993), though it was found in hippocampal interneurons. A great deal of further research is needed before the true role of nitric oxide in memory consolidation could be determined. The other candidate, arachidonic acid can also induce LTP. Potentiated neurons release more arachidonic acid, which activates NMDA receptors and reduces the uptake of glutamate by glial cells (Bliss et al. 1993). Additionally, blocking phospholipase A2 suppresses LTP (Bliss et al. 1993). However, analogously to nitric oxide, arachidonic acid cannot be considered a rapid retrograde messenger as the potentiation it induces is slow to develop. Finally, it is worth mentioning that some authors proposed the increase in extracellular potassium as a possible retrograde messenger (Bliss et al. 1993). Potassium enhances coupling of mGluR with protein G, and might this way affect the presynaptic element of a recently active synapse.
In conditioned neurons, calcium ions enter the cell via NMDA channels, but it is also be amplified by the efflux of calcium from intracellular stores. Calcium alone has been found to mediate reduced afterhyperpolarization in conditioned CA1 cells (Disterhof 1991), but the most interest has been attracted by its ability to induce LTP. Three seconds of elevated calcium can induce LTP (Bliss et al. 1993), but if it is the sufficient cause of LTP is not clear (Bliss et al. 1993). Blocking the influx of calcium or applying EGTA suppresses the cells ability to manifest LTP in the hippocampus (Squire et al. 1991). Similarly, drugs that block intracellular calcium, block the development of LTP (Bliss et al. 1993). Elevation of calcium in conditioned neurons can persist for minutes (Bliss et al. 1993)! The influx of calcium has been shown to result in the activation of at least three protein kinases. It also stimulates the migration of protein kinase C to the membrane. Additionally, nitric oxide synthase also seems to be activated by calcium transients (Bliss et al. 1993), which also, in conjunction with calmodulin, can activate adenylate cyclase. The activation of protein kinases has been shown to result in phosphorylation of NMDA and AMPA receptors (Bliss et al. 1993). It has been found quite long ago that calcium is able to increase the number of glutamate receptors; the effect that does not result from the synthesis of receptors de novo (Lynch et al. 1984). One of the hypotheses says that this results from activation of the protease calpain that exposes glutamate receptors by degrading a membrane protein fodrin. Influx of calcium into the presynaptic element increases the release of the neurotransmitter (Kandel 1982). This probably results from the fact that calcium helps binding presynaptic vesicles with the membrane (Kandel 1982). There seems to be no doubt that learning reduces calcium-activated potassium currents in neurons (Disterhof 1991). Though the exact path from calcium to potassium currents is not clear, it seems to be crucial for the development of LTP.
One of the first enzymes attracting attention in reference to studying the molecular basis of memory and learning was protein kinase C (PKC). It has been found that slow-learner mice mutants show lower levels of PKC(Science 1994 Genes and behavior). PKC is also lower in Alzheimers patients (Mishkin et al. 1991). The amount of PKC in the membrane of conditioned neurons seems to be directly correlated with the amount of training (Mishkin et al. 1991). The mere injection of PKC into the hippocampus is claimed to have caused potentiation (Craig 1993), though some authors claim it is not enough to evoke LTP (Bliss et al. 1993). Alkon et al. used PKC to mark parts of the brain involved in learning (Craig 1993). Injecting PKC, PKA or PKG inhibitors causes amnesia in 60 minutes (Serrano et al. 1994). Calcium stimulates the migration of PKC to the membrane (Alkon 1989). The same effect is also caused by diacylglycerol whose presence in the cell may result from the activity of phospholipase C that may enhance calcium transients via inositol 3-phosphate (Alkon 1989). Protein kinase C that is the subject of transfer from the cytosol to the membrane is not synthesized de novo, as no increase in mRNA for PKC has been found in conditioning (Craig 1993). It has even been found that mRNA for beta-isoform of PKC is downregulated upon tetanization (Bliss et al. 1993). The soluble fraction of PKC decreases as a result of conditioning (Spieler 1994), which suggests that the increase in the membrane activity comes entirely from the transfer of enzyme molecules from the cytosol. The increase in PKC activity appears first on the surface of cell somata (peak in 24 hours) and shifts within 72 hours to the surface of basiliar dendrites (Mishkin et al. 1991). The transfer of activity of PKC in the membrane may last for days (Alkon 1989) though Bliss claims it can rather be measured in hours (Bliss et al. 1993). As for the phosphorylation target of PKC, there are many possible candidates. One of the hypotheses is that phosphorylation of proteins marks places at which newly synthesized proteins are inserted later on (Kaczmarek 1993). This would ensure dendritic specificity despite the depolarization of the entire neuron. It has been found that PKC appears to be collocated in the brain with protein Go(Gorzelanczyks source). Additionally, a possible phosphorylation target of PKC is a variety of protein G that is homologous with Ras protein (Mishkin et al. 1991). PKC also increases the activity of NMDA (Bliss et al. 1993) probably via phosphorylation. Phosphorylation would reduce the block of NMDA by magnesium and increase the influx of calcium. Finally, the phosphorylation of proteins caused by PKC increases the release of the neurotransmitter (Spieler 1994), and reduces the flow of potassium via potassium channels. Artificially injected mPKC was demonstrated to reduce voltage-dependent potassium currents (Mishkin et al. 1991). At the same time, PKC in the cytosol may even increase the potassium ion flow (Alkon 1989). Among activators of protein kinase C are also adenylate cyclase (unknown mechanism of activation and arachidonic acid (Craig 1993), whose possible role as a retrograde messenger has been discussed earlier.
Other kinases involved in establishing LTP
It has been observed that calcium-calmodulin inhibitors cause amnesia in 15-30 minutes (Serrano et al. 1994). This attracted the attention of researchers to calcium-calmodulin dependent kinases. A very important finding was that mice lacking CAM kinase II cannot induce LTP (Grant et al. 1994). The activity of this enzyme has earlier been found as increased in conditioned neurons (Craig 1993); very much the same way as the activity of protein kinase C. However, other sources report that NMDA does not activate the enzyme. On the other hand, activation of NMDA receptors results in the activation of tyrosine kinase, but the lack of Fyn tyrosine kinase in mutant mice does not have such a dramatic effect on LTP as the lack of CAM kinase II. In a recent study, it has been shown that tetanization increases the level of mRNA for CAM kinase II within 30 minutes to 3 hours since the onset of the stimulus (Bliss et al. 1993). Though protein kinase C has until know attracted more attention, similar results could not be confirmed for the latter (Bliss et al. 1993). An interesting fact is that CAM kinase II has the ability to autophosphorylate in the presence of calcium and, consequently, maintain its increased activity long after the calcium signal wanes (Bliss et al. 1993). This seemed interesting as a form of short lasting memory ability in which the enzyme would remember past calcium transients. The significance of autophosphorylation is, however, entirely unclear. Among other findings, it is worth noting that the development of E-LTP has been postulated to require the activation of serine-threonine protein kinases (Kandel 1993). However, very little data is available in reference to the possible role of these enzyme in establishing long-term potentiation.
New evidence on the role of cAMP in memory and learning
Cyclic adenosine monophosphate (cAMP) is a major second messenger in regulating cellular response to the ligand molecules(terminology!). Its possible involvement in long-term potentiation must have been considered for as long as the interest in molecular aspects of memory. The pivotal role of cAMP has been well-established in seminal research by Dr. Eric Kandel who studied potentiation phenomena in a marine snail Aplysia caliphornica. In a historic article in Science in 1982, Kandel described the molecular mechanisms responsible for sensitization in Aplysia and boldly proposed that similar mechanisms might also be involved in establishing memories in humans. A decade later, there is an increasing body of evidence that his hypotheses were largely justified. The sensitization mechanism in Aplysia involves the neurotransmitter serotonin. Kandel noticed that the level of cAMP remains elevated for 1 hour after an injection of serotonin into the system (Kandel 1982). Interestingly, the facilitation period seemed to be closely correlated with the increased level of cAMP. Subsequently, it was found that cAMP reduces potassium currents through inducing cAMP-dependent kinase that is able to phosphorylates the potassium channel. The natural conclusion was that the reduced potassium currents increase the influx of calcium because of slower repolarization. This mechanism is identical to the one proposed by Alkon in reference to the action of protein kinase C. In his Science article Kandel proposed that cAMP-dependent protein kinase should be considered an important candidate for a long-term memory factor (Kandel 1982). Additionally, Kandel concluded that the enzyme should be membrane-bound in order to ensure dendritic specificity even if the entire cell undergoes depolarization and global increase in cAMP (see similar conclusions drawn from the analysis of the behavior of protein kinase C earlier in the chapter). In the follow-up research several later, Kandel managed to greatly substantiate his hypothesis. First of all, he noticed that L-LTP can be blocked by antagonists of dopamine receptors which activate adenylate cyclase (Kandel 1993). It has also been found that activation of NMDA increases the level of cAMP in the cell (Bliss et al. 1993). Naturally, an interest had to by transferred to protein kinase A. PKA inhibitors have been found to have little impact on E-LTP, but appeared to be sufficient to block L-LTP (Kandel 1993). A number of experiments indicates that the activation of PKA may be responsible for inducing L-LTP (Kandel 1993). Analogs of cAMP have been confirmed to induce L-LTP that occludes with naturally induced L-LTP. Finally, a link between cAMP and the protein synthesis in memory consolidation has been established by the observation that protein synthesis inhibitors block the induction of L-LTP with cAMP analogs (Kandel 1993).
An increased level of activity of proteases has been found in the wake of LTP (Bliss et al. 1993). Calpain belongs to the most investigated proteases that have been found in the hippocampus (Lynch et al. 1994). Calpain is activated by calcium; hence the possible increase of activity in potentiated neurons (Lynch et al. 1984). Blockade of the enzyme suppresses LTP (Lynch et al. 1984), there is therefore sufficient reason to believe in its involvement in the memory consolidation cascade. It has been observed that the number of glutamate receptors increases as a result of calcium transients, and proteinase inhibitors suppress the effect of calcium-induced increase in the number of receptors (Lynch at al. 1994). Interestingly, fodrin, a protein that is located mostly in the postsynaptic membrane has been proposed as the target of the proteolytic action of calpain. According to the hypothesis put forward by Lynch and Baudry, calpain might expose glutamate receptors by degrading fodrin without the need for the de novo synthesis of the receptor proteins (Lynch at al. 1984). As it has been found that the blockade of calpain does not impair shock-avoidance learning (Lynch et al. 1984), the role of the enzyme in memory consolidation remains unclear.
In the study of molecular basis of memory, a great deal of attention has recently been given to metabotropic glutamate receptor (mGluR). It is known that the following enzymes can be activated by mGluR via protein G (Bliss et al. 1993):
- phospholipase C
- phospholipase A2
- adenylate cyclase.
All the above enzyme have already been mentioned in previous sections as potentially important for establishing memories. Phospholipase C is responsible for increased levels of inositol 3-phosphate that enrolls more calcium from intracellular stores, and diacylgycerol that stimulates the migration of protein kinase C to the membrane. Not surprisingly, activation of mGluR increases the level of inositol 3-phosphate in the cell (Bliss et al. 1993). Phospholipase A2 can enhance the release of arachidonic acid that has been confirmed to be an LTP enhancer and possible retrograde messenger. Finally, activation of adenylate cyclase results in increased levels of cAMP that triggers some elements of the memory consolidation cascade via cAMP-dependent protein kinase. As a result of intense study over the last four years, it has been shown that mGluR agonists increase the response of neurons to NMDA, and can augment LTP (Bliss et al. 1993). Even more so, mGluR agonists have been found to induce memories(Science Vol 268, p. 262); possibly through suppressing the effect of GABA on potentiated cells. ACPD, an agonist of metabotropic glutamate receptor can even induce LTP (Bliss et al. 1993). In the study of mGluR antagonists, it has been shown that they can reduce the duration of LTP (Bliss et al. 1993). The activation of mGluR can induce LTP even if NMDA receptors are blocked, and it is probably needed to induce LTP in the first place (Bliss et al. 1993). In the context of activation of phospholipase C, it is not surprising that mGluR-induced LTP is sensitive to drugs that block calcium (Bliss et al. 1993), though activation of mGluR itself may depend on NMDA-induced calcium transient (Bliss et al. 1993). One of recent hypothesis is that the increase in extracellular potassium upon depolarization of the target cell might act as a retrograde messenger, as potassium enhances coupling of mGluR with protein G (Bliss et al. 1993). However, this would not make use of the coincidence-detection rule, and does not seem, therefore, to have much appeal in elucidating the molecular basis of memory.
The involvement of gene expression in memory consolidation has for long been quite well documented; however, little is known about the genes that become more active, for example, in later stages of LTP. Expression of c-fos transcription factor in the hippocampus has been found to increase in conditioned animals (Kaczmarek 1993). C-fos makes part of the activator protein 1 (AP1) that takes part in regulating the expression of a number of other genes. Increase in c-fos has been observed both pre- and post-synaptically, and the level of c-fos m-RNA has been confirmed to increase severalfold as a result of conditioning (Kaczmarek 1993). The synthesis of c-fos requires a strong stimulus, and it is dependent on the activation of NMDA receptors (Kaczmarek 1993). Interestingly, c-fos cannot be induced at the plateau of performance (Kaczmarek 1993), which might serve as an excellent molecular mechanism accounting for the spacing effect. Additionally, the importance of findings on c-fos genes has been enhanced that it cannot be induced in anesthetized animals (Bliss et al. 1993).
Protein synthesis and memory
The quest for proteins that might serve as memory molecules has been as long as the molecular research on memory. Initially, it was even thought that the memory itself might be coded in DNA or even protein molecules. Naturally, this quite unlikely presumption was replaced with the understanding of neural networks and the way information is stored in the human brain. Nevertheless, protein remain in the focus of interest as the class of biopolymers involved in potentiating synaptic transmission. It has been known for long that upon applying inhibitors of protein synthesis, amnesia usually occurs as early as in several hours (Kaczmarek 1993). For example, a translational inhibitor anisomycin reduces the duration of LTP to 3-6 hours (Bliss et al. 1993). It has also been noted that during the REM phase of sleep, that has long been supposed (and recently documented) as an important memory consolidation participant, the uptake of amino acids in the brain increases markedly. Additionally, in a direct measurement experiments, it has been shown that conditioning increases the synthesis of mRNA and proteins in conditioned cells. Protein synthesis is definitely proved to be a necessary condition for establishing long-term memories (Kandel 1982). It is known that LTP is associated with an NMDA receptor-dependent increase in the protein content of hippocampal perfusates. This increase is quite slow to develop, and indicates that the protein synthesis may start as late as three hours after conditioning and last for much longer. Alkon has found that migration of protein kinase C to the membrane increases the synthesis of proteins (Alkon 1989). As I earlier mentioned, the level of c-fos protein increases markedly as a result of conditioning. Similarly, the level of mRNA for immediate early gene zif/268 increases after tetanic stimulation (Bliss et al. 1993). In recent important report, mRNA for GR33 presynaptic glutamate receptor has been found to increase during LTP in dentate granule cells of the hippocampus (Smirnova et al. 1993). The increase was maintained for about five hours. Protein kinase C that is the subject of transfer from the cytosol to the membrane is not synthesized de novo, as no increase in mRNA for PKC has been found in conditioning (Craig 1993). However, in a recent study, it has been shown that tetanization increases the level of mRNA for CAM kinase II within 30 minutes to 3 hours since the onset of the tetanic stimulus (Bliss et al. 1993). All in all, there is a great body of evidence indicating the importance of protein synthesis in memory consolidation; however, very little is known which proteins play a key role, and the establishment of memory protein does not seem to be an immediate prospect. Nevertheless, a number of authors have put their hypothetical candidates that include among others: glutamate receptors, protein G, synaptic release proteins, active zone proteins, membrane glycoproteins, and many more.
As indicated in the previous section, protein G is one of attractive candidates for a regulatory protein involved in consolidation of memories. Long-term potentiation can be induced by protein G activator NaF/AlCl3 (Bliss et al. 1993). Importantly, protein G may regulate opening of potassium channels in learning and be itself phosphorylated by protein kinase C (Alkon 1989). One of its subspecies, protein Go, has been found to collocate with protein kinase C. CAM kinase II can also phosphorylate protein G (Alkon 1989) and influence the status of potassium channels. David Alkon has found that mRNA for a protein G analog appears in conditioned neurons in Hermissenda (Alkon 1989). Coupling of mGluR with protein G as a result of extracellular potassium has also been proposed as one of potentiation mechanisms (Bliss et al. 1993). The attractiveness of protein G as a candidate for a memory factor is so much greater that it can stabilize the membrane, and thus prolong the changes effected by other factors such as migration of protein kinase C.
It has been shown long ago that the number and properties of post-synaptic channels are changed in LTP (Bliss et al. 1993). Potassium channels are the most likely effector of potentiation. Their phosphorylation reduces the flow of potassium ions during repolarization and results in increased inflow of calcium. This reduced flow as a result of phosphorylation can be observed in conditioned neurons that show increased excitability (Alkon 1989). Potassium channel blockers enhance potentiation(LTP.TXT), and have been shown to even induce LTP on their own (Bliss et al. 1993). Interestingly, this effect is suppressed by alcohol (LTP.TXT). The decrease in the flow of potassium may last for days after conditioning (Alkon 1989). Importantly, the flow of potassium is not reduced by random stimuli that reduce it in association (Alkon 1989). In other words, permeability of the membrane for potassium may be subject to coincidence detection rule. The opening of potassium channels is regulated by protein G which adds importance to the previously presented findings on the possible role of protein G in memory consolidation (Alkon 1989). Both protein kinase C and CAM kinase II can also phosphorylate protein G and affect the status of potassium channels (Alkon 1989). In accordance with the role of the transfer of PKC to the membrane, PKC from the cytosol shows the opposite effect on potassium channels and detectably increases potassium ion flow (Alkon 1989).
In this section I will try to show that experimental data collected in the study of synaptic transmission in memory and learning presented in the preceding subchapter can be combined with the model of two-components of long-term memory discussed at the beginning of the chapter (Wozniak et al. 1995). My first attempt to find molecular correlates of retrievability and strength dates back to the late 1980s (Wozniak 1990); however, as I tried to show in the preceding review of recent research, the beginning of the 1990s has marked a dynamic inflow of new findings resulting, among other things, from application of new research procedures. Consequently, the development of molecular models can be much more accurately substantiated with the existing publication references. For the sake of clarity of the presented reasoning and easiness of reference, I will quote again here the guidelines on establishing the molecular correlates of retrievability and stability derived earlier:
- R should reflect the synaptic potentiation; forgetting should be understood as the decrease in R
- R should reach a high value already after the first repetition, and decline rapidly in the matter of days (the average optimum inter-repetition interval for retention 95% is several days)
- S determines the rate of decline of R (the higher the stability of memories, the slower the decrease of retrievability)
- with each repetitions, as S gets higher, R declines at a slower rate (stability of memory increases with successive repetitions)
- S should assume a high value only after a larger number of repetitions (stability of memories is positively correlated with the amount of training)
- S should not change (significantly) during the inter-repetition interval
- if the value of R is high, repetitions do not affect S significantly (spacing effect)
- R and S increase only as a result of an effective repetition (i.e. repetition that takes place after a sufficiently long interval)
- As S increases, its further increase becomes easier and easier
From Points 1 and 2 we may conclude directly that retrievability should be associated with long-term potentiation (LTP corresponds directly to the time frame of the decline of memory traces exhibited after learning new material). This leads directly to establishing several molecular correlates of retrievability such as: increase in protein kinase C in the membrane, increase in the activity of protein kinase A, phosphorylation of potassium channels(complete), etc. Because of the previously discussed scarcity of data referring to longer time frames, the establishment of the correlates of stability substantially less obvious. Because of its definitely long-term nature (Points 5 and 6), stability is most likely to be correlated with changes in protein content or properties in the synapse (Nobel winner F. Crick has pointedly noticed that no other membrane-bound molecules can provide better half-life characteristics for sustained renewal of memory traces). Because of its unknown nature, I will use the name for the hypothetical group of proteins that are involved in manifestation of stability. The mechanism of establishing LTP is relatively well studied; however, much less is known about the impact of conditioning on long-term changes in the synapse; for example, on protein synthesis. We can presume, however, that stability develops as a result of impulsation complying with coincidence-detection rule (Point 8), but not as a result of LTP itself (Points 6 and 7). In other words, correlates of retrievability cannot underlie the development of stability; however, both components of long-term memory must originate from the same set of molecular phenomena that directly follow the training session. This would indicate that retrievability and stability increase as a result of transient phenomena such as: activation of NMDA, elevated calcium levels, elevated cAMP,(complete) etc. The obvious implication of the proteinaceous nature of stability and the way it is induced is the impact of the transient phenomena on gene expression. As I showed earlier, the link between conditioning and gene expressions has been quite well established, providing a solid support for the proposed interpretation of the properties of stability implied by Point 8. Another important implication of the presented guidelines is that stability should affect the decline of retrievability. This would strongly favor a mechanism by which newly synthesized synaptic proteins have a stabilizing effect on the changes to retrievability. As it has for long been postulated, both retrievability and stability should be somehow related to the synaptic membrane for the sake of specificity of facilitating only a small subset of the dendritic connections of the neuron. Stabilins should then have a consolidating effect on the structural changes in the membrane, including those corresponding with retrievability (Point 3). There is almost no molecular data on long-term changes in the functionality of the synapse upon a number of training sessions. The increase in the active zone proteins makes a notable exception, though it concerns habituation in Aplysia. Active zone proteins and other membrane stabilizing factors that appear as a result of a long-term training process, should be the primary investigative target for uncovering the molecular nature of stabilins (Points 4 and 5). The only unraveled property of the two-component model of long-term memory is the spacing effect, which implies that high levels of retrievability prevent increase in stability (Point 7). The following two interpretations are the most compelling:
- transfer of retrievability factors from the cytosol to the membrane diminishes their cytoplasmic level that may be necessary for the development of stability (e.g. as it is the case with protein kinase C)
- high retrievability changes the properties of the membrane that reduce or eliminate the impact of transient factors responsible for consolidation of memory (e.g. phosphorylation of potassium channels might change the modus of depolarization in a way that diminishes the role of transient inductive factors).
The change of one of the above factors may reduce the level of gene expression, as it has been found with the earlier mentioned lack of c-fos induction at plateau level of performance.
The general conceptual model that illustrates the general relationship between molecular factors correlated with retrievability and stability is presented in the figure below:
Although there is still a long way toward ultimate answers; the above presented model goes definitely far beyond what might be concluded at the time of formulating the previous one based on phosphorylation of proteins and the synthesis of synaptic receptors (Wozniak 1990).
The increasingly evident correlates between molecular and behavioral levels with reference to two-components of long-term memory strengthen the thesis of universal applicability of repetition spacing algorithms for the purpose of minimizing the cost of constant knowledge retention in the practice of learning