The uniqueness of each human being is dependent upon his life history, as stored in the individual's memory. Thus, self-perception is dependent upon storage and retrieval from memory. For hundreds of years, philosophers and scientists endeavored to unravel the enigma of memory. Today, as in the remote past, understanding of the essence of memory remains the Holy Grail of modern neuroscience. Despite extensive research into the molecular mechanisms of memory over the last 30 years, we still lack the "big picture" of the intricate processes entailed in memory engram formation.

 

Memory - definitions

​

In contemporary neuroscience memory is discriminated into immediate, short-term and long term forms. Immediate and short term memories have a limited capacity and last only for a period of several seconds to a minute. On the contrary, long-term memory has nearly unlimited capability to store information for unlimited duration. Long-term memory is divided into declarative (explicit) and non-declarative (implicit) types. Declarative memory includes facts whereas non-declarative memory refers to acquisition of skills and habits. These two types of memory are formed by different brain structures, hippocampus along with other medial temporal lobe structures with further consolidation in the neocortex are related to declarative whereas neostriatum and cerebellum mediate non-declarative memory mechanisms. In addition, amygdale has been shown to mediate emotional memory and being involved in memory consolidation. Declarative memory is divided into episodic memory (the personally experienced event specific to a particular context such as time and place) and semantic memory referred to facts taken independent of the context in which they were learned.

​

 

A bit more about the Molecular Mechanisms

​

The current thinking about memory mechanisms focus on two approaches: one approach emphasizes the role of hallmark molecules linked to the functional and morphological changes upon memory formation; the second approach focus on enhancement of interaction between specific neurons within the neural networks (NNs) storing the memory traces. The first approach relates to the well-studied classical molecular and cellular neuroscience, the main dogma of which states that long-term synaptic plasticity is the cellular correlate of memory formation and retention. Alteration in synaptic activity employs a wide range of signal transduction pathways, which regulate changes in synaptic strength and connectivity by governing memory related gene expression and protein translation. The variation in synaptic strength affects summation properties of the neuron and its output propagated by the axons, in turn, entails alteration of specific functional communications between neurons creating memory-related functional NNs. The second approach states the logic of NN as a base of memory formation. In Hopfield model, recurrent NN (RNN) of excitable neurons may store traces of discrete memory as activity patterns, to which partial patterns of activity coverage, based on synaptic weight, mathematical model of synaptic plasticity. Hence, synaptic plasticity bridges between the two approaches aiming to unravel the memory formation mechanisms.

​

​

Synaptic Plasticity: Long-term Synaptic Plasticity & Memory

​

Long-term synaptic plasticity as cellular correlate of memory persists hours, days, months, and perhaps, years, in contrast to short-term versions, which end up within seconds and minutes. On the long run, long-term synaptic plasticity requires protein de novo synthesis and gene expression, which in turn, will maintain and support the chain of events expressing synaptic plasticity, accompanied by time-bound changes in the cellular ultrastructure.  Hence, long-term synaptic plasticity is an extremely intricate phenomenon in which proper timing of consecutive signal transduction events determines the success or failure of memory engram formation. Despite this appealing concept and the volume of research revealing specific sequences of memory-related molecular processes, little is known about gene expression pattern changes incurred during memory formation, or regarding the signal transduction networks leading to these alterations in gene expression. In parallel to genetic mechanisms of regulation, recent studies in different fields of neuroscience have shown a strong epigenetic level of regulation, where noncoding RNAs play one of the most important roles in controlling of synaptic plasticity mechanisms.

​

Phosphorylation and synaptic plasticity

Among all known post-translational modifications, phosphorylation is the most crucial signaling regulator of induction, establishing and maintenance of long-term synaptic plasticity. At the initial steps of long-term synaptic plasticity, in the absence of newly synthesized proteins, long-term potentiation (LTP, a subtype of long-term synaptic plasticity) is maintained by phosphorylation of pre-existing proteins. Serine/threonine protein kinases, PKA, PKC and CaMKII were shown to be crucial for the early events of LTP. More recently, the role of PI3K in LTP expression was also demonstrated. AKT has been shown to be important in LTP dependent LTD (long-term depression, another type of long-term plasticity) regulation via GSK3β, another serine/threonine kinase. Although the role of tyrosine kinases is much less studied, they have been shown to be important for both early and late phase events of long-term synaptic plasticity; for instance, ERK was shown to be involved in regulation of protein translation necessary for maintenance of LTP, whereas p38 were shown to be involved in mGluR dependent LTD regulation. Although a long list of protein kinases, there is still obscure dynamic interplay of these signal transduction components, as well as their spatial and temporal organization and role in different forms of long-term synaptic plasticity.  Our recent findings showed a crucial role of AKT in expression of early phase LTP and the role of membrane tethering in this regulation.  

​

Transcription-independent protein de novo synthesis and synaptic plasticity

Although phosphorylation of pre-existing proteins is sufficient to induce and maintain long-term synaptic plasticity at its early phase, the later events requires de novo protein synthesis. Both forms of long-term plasticity, LTP and LTD have been shown to require protein translation. In case of LTP, two major signaling pathways have been shown to be involved in protein translation regulation: 1) ERK-MAPK pathway with two independent upstream regulators: PKA and PKC; PI3K-mTOR pathway, which regulates translation at its initial steps, similar to EKR-MAPK pathway. Despite the vital role of protein synthesis, we know almost nothing which plasticity relate proteins (PRPs) are de novo synthesized to maintain long-term synaptic plasticity. Proteins, such PKMz, PDE4B3, CaMKII are considered to be candidates for maintenance of LTP, whereas immediate early gene Arc/Arg3.1, STEP and MAP1B are associated with sustaining of LTD.

​

Small RNA dependent regulation of translation related to synaptic plasticity

Although noncoding RNAs may be fundamental in gene regulation during development, very little is known regarding their role in the regulation of synaptic plasticity, and almost nothing is solidly established regarding their role in memory engram formation. Emerging evidence indicates that miRNAs are actively involved in regulating gene expression patterns in the adult brain. Recent evidence has shown that proteins associated with the miRNA pathway, e.g. Dicer, Argounaute, and fragile X mental retardation (FMRP), are localized in dendrites and may be involved in regulation of dendritic mRNA levels and local protein translation. Further studies have shown that at least 20 miRNAs are enriched at the synapse and at least part of them may be involved in the regulation of synaptic plasticity. Moreover, the induction of Long-term Potentiation (LTP) in vivo was found to induce an over 50-fold increase in miR-132 and miR-212 levels. Deficiency of key components of miRNA expression regulation was shown to enhance long-term memory; e.g. Armitage knockdown enhanced long-term olfactory memory, Dicer knockdown enhanced both synaptic plasticity and memory in a variety of learning tasks including spatial learning in the Morris water maze, as well as contextual and trace fear conditioning. These knockdown procedures also reduced miR-124 and miR-132 expression, which were previously shown to be involved in the regulation of dendrite spine formation. Elsewhere, miR-134 was shown to be important in regulation of spinogenesis via targeting Limk1 kinase, while an inverse effect of miR-132 and miR-125b expression has been shown on dendritic spine size and density, and a significant increase in spine volume was observed upon inhibition of miR-138 function. Moreover, a most recent study described involvement of miR-291/b in regulation of dendritic spine morphology through targeting Arpc3 actin nucleation factor. Accumulated evidence unequivocally indicates a crucial role of distinct miRNAs in the regulation of synaptic plasticity and formation of memory engram. However, the evidence still consists of isolated cases of individual miRNAs detected “accidentally.”

​

Memory related gene transcription regulation

Persistence of long-term synaptic plasticity requires transcription dependent translation regulation. Significant experimental evidences enforce a notion that the role of gene expression in memory formation is to be the replenishing of mRNA pool of proteins necessary for maintenance of synaptic plasticity whereas translation is crucial for establishing of long-term plasticity. Consequently, transcription is supposed to be permanently sustained during the whole period of long-term synaptic plasticity maintenance, which may persist years in living animal. Accumulated evidence shows that transcriptional control of long-term synaptic plasticity and memory is strictly regulated via multiple signal transduction pathways controlling TFs responsible for gene expression. Several transcription factors (TFs), e.g.  CREB, C/EBP, AP1, Egr and Rel/NF-kB have been shown to be critical to synaptic plasticity and memory formation. Serine/threonine kinase (S/T-K) and phosphatase pathways are the most intensively studied in the context of synaptic plasticity, including LTP. However, studies conducted during the last decade showed that not only PKA, CaMKII,  CaMKIV and PKC pathways, but also MAPK signaling is involved in the regulation of TFs, e.g. CREB, C/EBP, mediating synaptic plasticity. Moreover, tyrosine kinases: ERK1/2, p38, and MAPK were also shown to be responsible for histone phosphorylation leading to chromatin remodeling, a process necessary for gene expression regulation and crucial to long-term memory.

​

​

Beyond Plasticity: Understanding of Memory & Neural Information Processing beyond Synaptic Plasticity

​

Synaptic plasticity is not the lone contributor to this process: multiple evidences show a significant role of non-synaptic plasticity and modification of intrinsic neuronal excitability in memory traces formation. Moreover, recent findings indicate a dramatic role of glia in memory encoding/consolidation, as well as contribution to generation of gamma oscillations associated with memory formation. Hence, despite the enormous amount of knowledge regarding how memory could be “built up”, the neurobiology requires to be extended by an entirely new approach to reveal how the information is coded, and what is the essence of memory “bit” in the brain. The conceptual progression from cellular structures to the NN level requires drawing of a broad picture describing the physical basis of memory, which can be attributed to an alteration in electrical activity around specific morphological structures incorporated into the functional NN. Local field potentials (LFPs), reflecting spatially averaged weighted variations of synaptic currents exerted during activity changes in synapses, evidences alterations in electrical activities underlying memory encoding and retrieval. However, LFPs do not provide a reproducible pattern, which could be associated with coding of a memory chunk. On the other hand, mathematical models of artificial NNs (ANNs) presented tremendous achievements in mimicking numerous brain features, by successful implementation of advanced algorithms in different fields of mathematics, data analysis, business and economic modelling and predictions. Regardless, artificial imitations of the NN are minimalistic in contrast to native NN. These imitations depict neurons as simplified processing units, lacking geometry, irrespective of the various types of dendrites and axons with permanently altering electrical properties, serving as specialized data processing and transmission components. As a result, ANNs are deterministic, behaving as digital processing units and are incapable of incorporating the complexity of the native NNs.