The most effortless supposition concerning the mechanism of memory is the formation of new pathways in brain during learning. When I.P. Pavlov suggested development of a new neuronal connection between cortical representations of conditioned and unconditioned stimuli, he used an analogy of a telephone exchange. In the early twentieth century, when one user wanted to establish communication with another user, he called to a telephone operator and she connected the two users by a cable. I.P. Pavlov suggested that something similar is also happened in the brain (only, of course, without participation of an operator or homunculus).
The nature of plasticity has been linked with changes in the reserves of neurotransmitter in the presynapse, in the configuration of the synaptic space, the rate of transmitter destruction, in the density of receptors in the subsy-naptic membrane and in the sensitivity of the postsynaptic receptor to the neurotransmitter. These hypotheses can be grouped together on the grounds that they all relate plasticity to changes in the efficiency of synaptic transmission. If memory traces would be localized in any location, in the pre- or in the postsynaptic structure, memory will be determined by the place, and a one-to-one correspondence will have to exist between the memory elements and the points of the brain. Another group of hypotheses relates plasticity to changes in the efficiency of excitation of neurons as a result of a change in the level of their membrane polarization or the threshold of AP generation. Clearly, the various hypotheses differ only in the localization of the plastic component while a firm fixation of memory trace in brain is supposed in any case. The problem of plasticity itself is reduced to a change in the efficiency of spread of the signal along the nerve net. Memory traces have been identified with stable changes in the conduction of excitation through a neuron or synapse. This hypothesis became extremely popular during tens years and almost forced out an alternative chemical hypothesis. Really, change in synaptic response during learning was established reliably. All regularities of behavioral learning were reflected in changes in synaptic efficacy. Particularly, when conditioned response increased, corresponding EPSPs in many neurons of brain also increased. The utmost form of synaptic hypothesis is postulation of the presynaptic nature of synaptic plasticity. No attention has been paid under these circumstances to the role of intracellular biochemical reactions. However, location of a memory trace in the postsynapse admits perception of all current synaptic influences as a whole (by chemical means). Certainly, this is not a trivial task, but it at least has a decision.
New neuronal connections might also be developed. Death of neurons or the birth of new neurons may presumably serve as the basis for long-term memory. The formation of long-term declarative memory has been tried to explain by means of the involvement of constantly appearing new neurons in specific areas of the brain, which differentiate from stem cells. They develop stimulus-selective (grandmother) and behavioral selective (which encode particular behavioral acts) long-term neurons. Final differentiation of long-term memory neurons may be triggered by the ‘novelty signal" from newly formed neurons of hippocampus. Such newly formed neurons included in a new environment form new synaptic contacts that then convert these neurons in "gnostic units" [1158]. For example, hypothalamic neurons in canary responsible for new song are anew born neurons [943]. Naturally occurring synapse elimination and permanent loss of axonal input has also been supposed as the attractive mechanism for information storage [738]. Instead of a new pathway corresponding to a new memory, it has also been supposed the formation of new systems of mutually activating neurons [43]. Evidently, whether a new pathway arises after learning or a preexisting path changes its efficacy, or a new combination of neurons arises, we are within the scope of the same structural hypothesis.
Intracellular electrical activity is almost impossible to record from a presy-naptic area, but it is accessible from the body of postsynaptic neuron. Indirect proofs of change in the presynapse were found 30-40 years ago. It was established that during learning the membrane potential, input resistance, threshold level, membrane capacity, membrane time constant and other characteristics of cellular membrane do not change [217, 1442, 376]. Sensitivity of postsynaptic neurons to certain neurotransmitters, in many cases, also does not change. If we proceed from the principle of "on the contrary", absence of alterations in the postsynaptic site denotes the reason for the excitatory postsynaptic potential (EPSP) change in the presynaptic area. However, this principle is not appropriate in the given case. In reality, change in the response during learning concerns only a specific stimulus related to the learning procedure, such as the CS+, while the response to the CS~ does not change. In order to examine membrane properties we need to affect neurons, for instance by a current pulse for examination of excitability or to administrate a corresponding neurotransmitter close to the cell soma in order to evaluate chemosensitivity. Such a measurement gives us insight into characteristics concerning the impact of the matter used for the measurement. For example, it shows excitability in response to current pulse and not in response to the CS+. Whether excitability, membrane potential, etc. are unspecific parameters? We will demonstrate that this is not always correct, but now we are only pointing out: if one considers a parameter to be unspecific, this parameter cannot explain specific changes of the response and it was not necessary to examine the role of that parameter at all. Curiously, similar difficulties arise if plasticity would be located within synapses accounting for a CS+. Specificity between changes in responses to the CS+ and CS~ is here easily explained, but correspondence between the alterations in CS+ and CS~ effects is modified during learning in a nontrivial way (see Sec. 1.3). Therefore, during learning, it is insufficient to change synaptic efficacy. Synaptic patterns accounting for CS+ and CS~ signals have to be reorganized and in fact become more distinctive in the course of training. Synaptic plasticity had also been investigated by means of quantal analysis of instability of EPSP, modified during learning. Neurons, it is known, secrete neurotransmitters from synaptic terminals into portions contained in synaptic vesicles. Therefore, EPSP consists of a few portions, quanta. Alterations in synaptic strength can be encoded presynaptically as alterations in the machinery releasing the neurotransmitter, say, glutamate, or postsynaptically by changing the number or function of receptors sensing the glutamate signal. If instability of EPSP is connected with quantal fluctuations, then either the value or the number of quanta will alter after learning. A change in value of quantal size means postsynaptic localization of plasticity, while a change in the number of quanta (quantal content) means an alteration in the probability of secretion of elemental portions from the presynaptic terminal. In order to perform quanta analysis, an amplitude histogram of EPSP was approximated by the Puasson’s (or in some studies by binomial) distribution [77] and experimental distribution did not reveal significant differences from the theoretical distribution. Thus, a change in EPSP amplitude was explained by the change in quantal content. However, this method is impossible to implement to a given task [1255, 890]. The same experimental distributions did not have significant differences also from Gauss’s distribution (we evaluated the above mentioned data by Kolmogorov’s criterion A = 0.4 — 1.1). Quantal analysis allows evaluating the difference between the experimental and theoretical distributions, but did not give any possibility to evaluate similarity between the two distributions. Paradoxically, the smaller is the set of experimental data, the easier it is to determine an absence of significant difference between the distributions. Therefore, implementation of quantal analysis is also proof of "on the contrary". By the way, the quantal size may differ across similar synapses (with the difference more than 10 fold) [153].
The strongest evidences of existence of material traces in synapses after aversive learning come from morphological data. The number, size and vesicle complement of presynaptic sensory neuron active zones are larger in sensitized animals than controls [77]. Neural terminals are markedly expanded (Fig. 1.11). Sensory neurons from long-term sensitized Aplysia exhibit a twofold increase in the total number of synaptic varicosities, as well as an enlargement in the size of each neuron’s axonal arbor. Similarly, after one-trial aver-sive learning in chickens, morphological traces of memory were found in their forebrain: when the chicks learned to discriminate edible grains (each chick only once pecked a bitter grain) the number of dendritic spines in the fore-brain of the chick increased up to 60% [1039]. Eyeblink conditioning increases the number (1.6 fold) of excitatory synapses within the interpositus nucleus that in the brain region is essential for long-term retention of the conditioned response [640]. We are considering that if some morphological or biochemical alterations after learning are too extensive, they, probably, do not concern memory. It is enough to remember how large an amount of experience any living being collects during its life. Such massive traces cannot serve normal memory and they are more similar to the traces of injury, evoked by excessive aversive stimulation. Really, it is well known that aversive events and excessive excitation lead to cell damage and tissue growth [797, 832, 337, 1389].
Nevertheless arguments like "on the contrary" sometimes are exploited even at the present time [1017, 45, 704, 86]. The fact that the response of a neuron to frequent stimulation is attenuated, despite the fact that the same unit did not decrease in responsiveness to rare stimuli (ordinary specificity), ones earlier [217, 1442, 376, 1438] and sometimes at the present time [371, 1282] is interpreted as evidence that habituation is mostly a result of synaptic depression rather than cellular-level excitability reduction.
There are many various forms of synaptic plasticity and it is impossible to consider these phenomena here in detail. We will point out only some important features in the well-known models of synaptic plasticity. Mechanisms of plasticity for Aplysia, rat hippocampus, chick and honeybee can be slightly different, but their important features are similar [515, 1071, 86]. A logical rule for the setting up of a new pathway during learning is suggested by D.O. Hebb [530]. He postulated that changes in synaptic efficacy take place when a presynaptic cell participates in the firing of a postsynaptic cell, or, in other words, both cells generate AP, but the postsynaptic cell generates a later AP. The Hebbian scheme implies participation of both the presynaptic and postsy-naptic neurons. The cellular mechanisms underlying Hebbian plasticity have been well studied, mostly using the example of long-term potentiation (LTP) in regions of hippocampus and neocortex: enhancement of neuronal responses after frequent or intensive stimulation. Low-frequency stimulation was found to induce long-term depression (LTD). Yet, firing produces LTP at low frequency Correspondingly Besides, LTP is induced by stimulation that results in strong postsynaptic activation, while LTD is induced by stimulation that results in weaker postsynaptic activation [746]. Similarly, a single tail shock in Aplysia produces short-term sensitization that lasts for minutes, whereas repeated tail shocks given at spaced intervals produce long-term sensitiza-tion that lasts for up to several weeks. Short-term behavioral sensitization lasts minutes to hours and correlates with an increase in synaptic strength between the sensory to motor neuron connection referred to as short-term facilitation.
![Growth of synaptic terminals in the sensory neuron of Aplysia after sensitization. The Fig. 1.11 was redrawn in accordance with the data [77].](http://what-when-how.com/wp-content/uploads/2011/06/tmp2311_thumb.jpg "Growth of synaptic terminals in the sensory neuron of Aplysia after sensitization. The Fig. 1.11 was redrawn in accordance with the data [77].")
Fig. 1.11. Growth of synaptic terminals in the sensory neuron of Aplysia after sensitization. The Fig. 1.11 was redrawn in accordance with the data [77].
Thousands of investigations have been undertaken in the field of synaptic plasticity. At present, this is the most broadly studied phenomenon in neurobiology. Typically, the problems discussed were: this circuit is critically involved or does not involved, presynaptic mechanism or postsynaptic, excitability changed or did not change, these protein kinases are essential and those are not essential, Ca?+ participates or does not participate, protein synthesis is important or ribonucleic acid (RNA) one is important, etc. [21]. Nevertheless, many fascinating details3 of the cellular and molecular mechanisms that underlie synaptic plasticity have been established and they, mostly, relate to postsynaptic phenomena [389, 1033, 890, 731]. The contribution of postsynaptic mechanisms to learning was earlier underestimated. Really, the induction of LTP/LTD by correlated pre/post spiking is accompanied by an immediate and persistent enhancement/reduction of the intrinsic excitability of the presynaptic neuron. This was established in the first report of LTP [135]. In addition to changes in global presynaptic excitability, correlated pre-and postsynaptic activity also results in modification of the local postsynaptic excitability [290].
Cooperation of many presynaptic fibers is often needed to produce LTP, since individual unitary synapses are too weak to cause strong postsynap-tic activation, whereas LTP depends on general integral power of excitation [1148]. Long-term potentiation or depression of the synapses may be dependent not only on strength and frequency of stimulation, but also on the temporal order of pre- and postsynaptic activity (spike-timing-dependent plasticity) [1279]. The timing of the postsynaptic AP relative to the EPSP determines the sign and magnitude of synaptic modification. Small differences in the timing of pre- and postsynaptic activity can determine whether synapses are strengthened or weakened. When a presynaptic AP precedes a postsynaptic AP (usually 10-25 ms), this leads to LTP. Reversed order of excitations lead to LTD. Spike-timing-dependent plasticity can be evoked in almost natural conditions. Repeatedly pairing visual stimulation and neuronal spiking induces rapid changes in the spatiotemporal receptive field of the neuron. The sign and magnitude of the receptive field modification depends on the relative timing of the pairing, in a manner consistent with spike-timing-dependent plasticity of the excitatory synapses on the recorded cortical neuron (±25 ms) [833]. The order of signals (the US after the CS+) is critical for behavioral plasticity, too, if one does not bear in mind that the time delay for real conditioning is in the order of seconds [740]. Additional peculiarities of the LTP phenomenon have been found. The NMDA receptors are largely blocked by Mg2+ at hyperpolarized membrane potentials, but the block can be relieved by depolarization, leading to the idea that this receptor can serve as the coincidence detector for pre/post activity [555, 290] and for change in synaptic efficacy. Synaptic transmission is enhanced if the NMDA receptor detects the co-activity of the presynaptic (release and binding of glutamate) and postsynaptic neuron (enough depolarization to expel Mg2+ from the channel pore) [1316]. So, strong dependence on the temporal order of pre- and postsynaptic excitations (mechanism of Hebbs’s rule) is excellently explained by the link of the voltage-dependent blockade of the NMDA receptor with Mg2+.
In mammalians, excitatory synapses, the object of plastic reorganization, are often located at the dendritic spines, at a long distance from the cell soma and separated from dendrites by thin necks. The regulation of diffusional coupling across this thin neck provides a possible mechanism for determining the amplitude of postsynaptic potentials [138]. It was supposed also that back-propagating of action potentials into dendrites of hippocampal and cortical pyramidal cells provide synapses by depolarization and are responsible for Hebbian plasticity. However, this logical scheme may not be absolutely correct. A back-propagating AP invades the dendrites, but may be insufficient to relieve the Mg2+ block of NMDA receptors. Really, back-propagating of action potentials is not always required for LTP and a strong synaptic stimulation is able to induce LTP even when back-propagating of action potentials is blocked [751].
Our memory is constantly renewed and there is a problem to explain how modified synapses keep old information when new information comes in. Retention is an active process and memory-storing synapses must somehow retain the capacity for ongoing plasticity if old information would be preserved in the face of new learning. Electrically induced LTP in the hippocampus is rapidly reversed when animals receive novel information through a new or enriched environment [290]. New learning incorporates a partial trial of old patterns and this retards the recording of new information [5]. At least the early phase of LTP and LTD can be reversed by neuronal activity following an induction protocol: de-potentiation and de-depression. The seizure activity following tetanus and low-frequency stimuli was found to reverse LTP. Similarly, reversal of LTD can be achieved by subsequent activity, typically high-frequency stimulation. De-potentiation stimuli are milder in intensity than those used in induction of LTP and LTD [1432]. The phenomenon of de-potentiation is in poor agreement with properties of memory, since memory continuously changes, but never disappears.
LTP can develop either gradually, or by the all-or-none principle, depending on the intensity of stimulation, but gradual development, if it is observed, takes tens of minutes after powerful stimulation [586, 15, 414, 1432, 124, 374, 1367]. The learning process may be prolonged and learning traces are usually exceptionally protracted, but this time is not the phase of time-consuming ripening. A necessity for time delay in order to develop an effect is not typical for learning, whose consequences are the most noticeable immediately after training. At the same time, traces of damage are usually expanded in time and appear in space gradually. An LTP is comparable with cell excitotoxic damage [1255, 785] and an LTP-related cascade of biochemical events coincides with the damage-related cascade: participation of glutamate receptors, Ca?+ elevation, dependence on protein synthesis, neural growth, dependence on strong excitation, blockage by GABA-inhibition or hyperpolarization, etc. We will discuss this problem further.
Thus, localization of short-term plasticity within synapses has never been demonstrated. Some indirect but important data such as spike-timing-dependent plasticity, the role of NMDA receptor as a coincidence detector, properties of dentritic spines, etc. are in accordance with supposed mechanisms of synaptic plasticity during LTP, but they are poorly timed with behavioral plasticity: Hebbian plasticity requires too short a delay in a millisecond scale between pre- and postsynaptic excitations and this contradicts the properties of behavioral plasticity. Long-term plasticity within synapses was demonstrated only on the level of morphology and only for anxious or inadequate learning, such as LTP. If current electrical activity could influence the structural organization of brain tissue, this is a very long lasting process and not compatible with the rate of behavioral plasticity. It would be possible to suppose that morphological reorganization may play a role in long-term memory storage, which is formed after memory consolidation over the course of several hours and keeps for days and years. Really, compared with non-stable biochemical processes and with fast- alternating electrical processes, morphological structure of brain, when it is represented at the histological level, looks steady, fine ordered and well suitable as a memory substratum. Nevertheless, synaptic proteins and structures are not stationary, but rather are highly dynamical. Synapses, even in adults, continuously arise and degenerate [267]. New-born neurons usually also are short lived [492, 332]. Video recordings from hippocampal neurons allow the visualization of actin dynamics in living neurons. These recordings revealed large actin-dependent changes in dendritic spine shape. Spines contain high concentrations of actin, suggesting that they might be motile. Visible changes occurred within tens of seconds, suggesting that anatomical plasticity of synapses can be sufficiently rapid for long-term memory, but are too slow for current memory [389] and it is not clear how synaptic efficacy can be precisely maintained without suffering from accumulative drift in the face of molecular turnovers of synaptic machinery [1316]. On the other hand, a large body of data has accumulated that it is in disagreement with the morphological nature of memory traces. In behaving animals, single neurons can intermittently participate in different computations by rapidly changing their coupling without associated changes in firing rate [352].
Is LTP an authentic model of learning? There are many parallels between LTP development and behavioral plasticity between synaptic plasticity and aversive or stressed (hippocampus- and amygdala-dependent) learning relative to pharmacological, morphological and physiological characteristics [1218, 86]. Rodents can readily be tested about their memories for places, objects, and odors and these studies have revealed that lesions of the hippocampus and related structures interfere with long-term storage of these kinds of memory [845]. However, nobody has demonstrated that modification of synap-tic pathways during normal, non-stressed learning hold permanent features.
Synaptic reactions do change during learning, but any signal usually activates many synapses at the given neuron and we record the reaction of an entire synaptic pattern. If the neuron recognizes a synaptic pattern, all together, the synapse may have a different efficacy, when it is included in different patterns [1264, 1272, 1292, 1064]. This means that when single neuron recognizes a synaptic pattern, as a whole, synaptic memory does not exist at all. Learning may depend on rapid modulation of effective connectivity with time constants of modulation in the range of from ten to one hundredth of a millisecond. Structural or anatomical connectivity should be distinguished from functional connectivity. At present, proofs relating to a connection between a change in synaptic efficacy during LTP and memory are absent [750].
Currently, it is evident that the Hebbian process is an oversimplification [281, 377]. A coincidence of pre- and postsynaptic excitations cannot be a decisive mechanism of plasticity. If we perceive stimuli, they must induce postsynaptic AP in some neurons. This automatically satisfies the Hebbian rule. The environment sometimes may demand that this reaction fail (for example, as a result of habituation), but Hebb’s rule requires an increase in this pathway. A change in synaptic efficacy often depends on its inherent properties and is not dependent upon behavioral needs. For instance, synapses with a low release probability display facilitation and augmentation, whereas the synapses with a high release probability supplied by the same axon may exhibit paired-pulse and frequency-dependent depression [1225]. On the other hand, transgenic mice (lacking a subunit of the AMPA glutamate receptor) may develop LTP and spatial learning in the absence of synaptic strengthening [1112]. Moreover, in most studies there is a specific range of repetition rates that must be used for successful induction: at frequencies below 10 Hz, LTP cannot be induced [751]. We would like also to remind that synaptic plasticity localized in determined points of the brain meets difficulties in furthering a plausible explanation of the numerous facts of memory recovery after injury After all, individual cell participates in the storing of many patterns of activity.
