Current theories of neuronal information processing performed by Ca2+/Calmodulin-dependent protein kinase II with support and insights from computer modelling and simulation
Introduction
An important question that neuroscientists are trying to answer concerns how the nervous system lays down memories. The answer to this fundamental question will not only explain how memories are encoded and retained by the nervous system but may offer insight into treating diseases and injuries that often impair memory function, such as Alzheimer's disease and stroke. Many experience-dependent changes that occur in the nervous system are recorded at synapses, the locus of communication between neurons (Fig. 1). Identification of the processes behind these changes will profoundly influence progress toward understanding the encoding and storage of memories. Recent studies have shown that certain enzymes found in synaptic structures play an important role in memory formation. These enzymes, termed protein kinases and protein phosphatases, regulate the phosphorylation state of certain substrate proteins and thereby control a wide range of cellular functions (Dunkley et al., 1986; Gispen and Routtenberg, 1986). A protein kinase catalyses the transfer of a phosphate group of ATP to an amino acid residue of the substrate protein. A protein phosphatase catalyses the removal of this phosphate group. The balance between protein kinase and protein phosphatase activity can significantly change cellular function.
The enzyme Ca2+/calmodulin-dependent protein kinase II (CaMKII) is a particularly appealing contender for keeping a record of past synaptic activity. CaMKII is concentrated in the postsynaptic density (PSD), a submembranous structure believed to be involved in neuronal information processing (Carlin et al., 1980), and 20–50% of all proteins in the PSD consists of CaMKII (Colbran and Soderling, 1990). CaMKII mediates the effects of changes in Ca2+ concentration associated with synaptic activity by phosphorylating receptors and ion channels (McGlade-McCulloch et al., 1993), controlling events in synaptic transmission (Benfenati et al., 1992), and regulating the expression of genes and altering neuronal shape (Halpain and Greengard, 1990; Antoine et al., 1996). CaMKII is known to play a key role in an experimental memory model called long-term potentiation (LTP) (Malenka et al., 1989; Reymann et al., 1988) and may be involved in spatial memory (Lisman, 1994; Silva et al., 1992). The enzyme is also involved in many pathologies. Much of the earlier interest in CaMKII arose because it was found that certain anticonvulsant drugs (e.g. Valium) inhibited CaMKII activity (DeLorenzo et al., 1981). Abnormal changes in CaMKII activity have been implicated in Alzheimer's disease (Simonian et al., 1994) and it is known that ischemic episodes (in which the blood flow to the brain is disrupted, such as in a stroke) cause acute inhibition of CaMKII activity (Aronowski and Grotta, 1996).
CaMKII is composed of 8–10 subunits that are tethered together to form a petal-like structure (Kanaseki et al., 1991). A subunit is expressed from one of four distinct genes—α, β, γ and δ (Tobimatsu and Fujisawa, 1989). Each CaMKII isoform is distributed heterogeneously throughout the brain (Miller and Kennedy, 1986), with the α-CaMKII isoform, found predominately in the forebrain (Kanaseki et al., 1991). All isoforms of subunits have a catalytic domain followed by a regulatory domain and an association domain (Fig. 2). The association domain is responsible for tethering subunits together. The regulatory domain has an autoinhibitory segment that overlaps a calmodulin binding site. Calmodulin is a molecular calcium “sensor” and is activated co-operatively by binding three or four calcium ions. Each subunit in the enzyme is regulated by Ca2+/calmodulin (Ca2+/CaM) (Yamauchi et al., 1989). In the absence of Ca2+/CaM, the autoinhibitory segment effectively disables catalytic activity by blocking access to protein- and ATP-binding sites. However, when a subunit binds Ca2+/CaM it becomes fully active and does not only phosphorylate other protein substrates but also itself on multiple amino acid residues in a process called autophosphorylation (Hanson and Schulman, 1992), which is independent of enzyme concentration and therefore an intramolecular reaction (Lai et al., 1986). The initial Ca2+/CaM-dependent phase of autophosphorylation is known to occur on a threonine residue at position 286 (Thr-286) of α-CaMKII subunits (Meyer et al., 1992). When Ca2+/CaM dissociates from a subunit autophosphorylated on Thr-286, two sites within the calmodulin-binding domain become available for secondary autophosphorylation—Thr-305 and Thr-306 in α-CaMKII subunits (Patton et al., 1990). These sites are rapidly autophosphorylated in a Ca2+/CaM-independent reaction that renders the subunit incapable of binding Ca2+/CaM. Other residues, such as Ser-314 in α-CaMKII subunits, may also become phosphorylated.
Autophosphorylation of CaMKII subunits has attracted considerable interest, especially for its possible role in memory storage (Lisman, 1985, Lisman, 1994) and has recently been the focus of several simulation and modelling studies (Coomber, 1998; Matsushita et al., 1995; Dosemeci and Albers, 1996). Early findings suggested that autophosphorylation of only a few subunits, maybe only one (Lickteig et al., 1988), could promote autophosphorylation of all other subunits (Lai et al., 1986) and full Ca2+/CaM-independence. Consequently, it was proposed that CaMKII might act as a bi-stable molecular switch (Lisman, 1985, Lisman, 1994; Lisman and Goldring, 1988; Miller and Kennedy, 1986). As long as a minimum threshold level of phosphorylated sites was maintained the enzyme could keep the molecular switch “on” by autophosphorylating any sites stripped of their phosphate groups by protein phosphatases. The level of autophosphorylated CaMKII sites was thought to act as a memory storage mechanism. A problem with this hypothesis is that there is no direct experimental evidence for CaMKII acting as a bi-stable switch. In fact, a more recent study (Ikeda et al., 1991) has provided evidence that disagrees with earlier results that autophosphorylation of Thr-286 of only one subunit is necessary to convert all other subunits to a Ca2+/CaM-independent state. It was found that to achieve maximum Ca2+/CaM-independent activity each subunit in the enzyme had to become autophosphorylated on Thr-286. In other words, CaMKII was discovered to be more like a “regulator” with a graded response based on the extent of Ca2+/CaM-independent activity than a simple on–off switch. However, in support of the bi-stable switch hypothesis, a recent mathematical modelling study demonstrated that through a chain of phosphorylation–dephosphorylation reactions it is possible to obtain quasi-switching behaviour as a function of the Ca2+ signal intensity (Matsushita et al., 1995). This model does not take into account the conformational arrangement of subunits (petal-like structure) but instead averages out any co-operative or allosteric efforts that may occur between subunits. Therefore, it is not known whether conformational detail has any bearing on the type of “switch” that may be implemented by CaMKII. Furthermore, the rate of dissociation of Ca2+/CaM from the enzyme was assumed to be much faster than is now recognized. Recent experiments have shown that Ca2+/CaM remains bound to a Thr-286 phosphorylated subunit for many seconds (Hanson et al., 1994), making Thr-286 phosphorylated CaMKII an enzyme with one of the strongest affinities for calmodulin. The observed “trapping” of Ca2+/CaM may be a significant factor in whether quasi-switching behaviour is expressed.
In another study, Ca2+/CaM trapping was proposed to provide potentiation for Ca2+ transients and to thereby act as a Ca2+ Frequency detector (Hanson and Schulman, 1992, Hanson et al., 1994). Experiments with CaMKII monomers relieved of their association domains showed that Thr-286 autophosphorylation is primarily an intersubunit reaction between proximate neighbours of the holoenzyme (Fig. 3). In contrast to early ideas, effective autophosphorylation of Thr-286 can only take place if Ca2+/CaM is bound to the subunit acting as “substrate” for an adjacent, activated subunit. This supports the graded switch theory (Ikeda et al., 1991). A computer simulation employing equilibrium equations based on experimental results was able to demonstrate that trapping confers upon CaMKII the ability to decode the frequency of Ca2+ signals. As the frequency increases, so does the probability of Ca2+/CaM becoming bound and then trapped, which in turn promotes further trapping of Ca2+/CaM as the number of activated subunits increases. However, it has been pointed out that because trapped Ca2+/CaM slowly dissociates (in the order of seconds), this scheme cannot discriminate high-frequency Ca2+, signals (Dosemeci and Albers, 1996). Furthermore, the ability to decode the frequency of Ca2+ signals rests on the assumption that the concentration of free calmodulin is limiting. The model does not consider results of secondary autophosphorylation on Ca2+ signal processing. In related work (Bear, 1995; Mayford et al., 1995), calmodulin trapping was suggested to play a part in regulating the threshold for long-term changes in synaptic strength as a function of stimulation frequency. Calmodulin trapped on autophosphorylated CaMKII would reduce the level of free calmodulin, and would therefore impede other Ca2+/CaM-stimulated reactions.
In another recent simulation of CaMKII behaviour (Dosemeci and Albers, 1996), a probabilistic model was explored for the role autophosphorylation may play in synaptic frequency detection. In this model CaMKII is assumed to be composed of 8–12 subunits, with each subunit having two types of autophosphorylation sites, denoted as A sites and B sites, respectively. Each subunit that binds Ca2+/CaM becomes autophosphorylated on its A site and remains active even when Ca2+/CaM diffuses away. An activated subunit may then autophosphorylate the B site of any other subunit (active or inactive), thereby preventing A site autophosphorylation of an inactive subunit. Therefore, B site autophosphorylation before A site autophosphorylation effectively locks the subunit in an inactive state. The model predicts that A sites and B sites become preferentially autophosphorylated under different periods of synaptic activity. Simulations demonstrated that periods of high Ca2+ concentration lead A sites to become autophosphorylated before B sites, and periods of low Ca2+ concentration lead B sites to become autophosphorylated before A sites. Consequently, the relative activity of CaMKII will reflect the pattern of prior synaptic activity. The durations of Ca2+ pulses and the duration between Ca2+ pulses could be recorded as a permutation of autophosphorylation states. Unlike the calmodulin trapping model, this scheme has the advantages of not requiring a rate limiting concentration of free calmodulin, and it can discriminate between pulse intervals of less than 1 s. However, there is one controversial point that should be noted about the mechanism of B site autophosphorylation, which has a significant bearing on the results. Secondary autophosphorylation (B site) is assumed to occur on any subunit providing at least one subunit is autophosphorylated on its A site. As mentioned before, although earlier reports (Lickteig et al., 1988; Lai et al., 1986) suggested this could occur, more recent evidence appears to indicate otherwise (Ikeda et al., 1991). Moreover, Hanson et al. (1994)observed that Ca2+-independent (B site) autophosphorylation was an intrasubunit reaction because Thr-305 and MY-306 do not make good substrates.
Section snippets
A new simulation model
In view of the strengths and weaknesses of the aforementioned models of CaMKII, the remainder of the present paper will discuss a new model, based as closely as possible upon recent experimental findings, and which incorporates all known pertinent aspects of CaMKII function. Our purpose is to use the model to help answer some questions that cannot be answered readily with earlier models. For example, earlier models have concentrated on single issues, such as co-operative autophosphorylation of
Results of simulation experiments
Computer simulations were performed to assess the behaviour of the model against experimental observations, with particular reference to stimulation protocols that produce upgrading or downgrading of synaptic efficacy by activating certain protein kinases and phosphatases. In many regions of the brain, such as the CA1 hippocampus, a long-lasting increase in synaptic efficacy at excitatory synapses is engendered by a short burst (tetanus) of high-frequency synaptic stimulation (HFS). This
Concluding remarks
The diverse number of protein kinases and protein phosphatases that operate in the nervous system play critical roles in cellular regulation and information processing. In the present paper, we have examined but one of these regulatory proteins, the Ca2+/calmodulin-dependent protein kinase II (CaMKII). Its wide, multifunctional substrate specificity and its interesting self-regulatory properties have seen it proposed variously as a molecular switch, a Ca2+ pulse frequency detector, and a
Acknowledgements
The author would like to thank Dr J. Carminatti for helpful comments on numerical simulation.
References (49)
- et al.
Molecular and Cellular Endocrinology
(1996) - et al.
Brain Research
(1996) Neuron
(1995)- et al.
Current Topics in Cellular Regulation
(1990) - et al.
Biophysical Journal
(1996) - et al.
Neuron
(1990) - et al.
Neuron
(1994) - et al.
Journal of Biological Chemistry
(1991) - et al.
Journal of Biological Chemistry
(1967) - et al.
Journal of Biological Chemistry
(1985)
Journal of Biological Chemistry
Trends in Neuroscience
Cell
Cell
Journal of Biological Chemistry
Journal of Biological Chemistry
Brain Research
Brain Research
Journal of Biological Chemistry
Journal of Biological Chemistry
Nature
Nature
Journal of Cellular Biology
Cited by (11)
Modeling CaMKII in cardiac physiology: From molecule to tissue
2014, Frontiers in PharmacologyThe effect of noise on CaMKII activation in a dendritic spine during LTP induction
2010, Journal of NeurophysiologyPostsynaptic signal transduction models for long-term potentiation and depression
2010, Frontiers in Computational NeuroscienceModeling in the neurosciences: From biological systems to neuromimetic robotics: Second edition
2005, Modeling in the Neurosciences: From Biological Systems to Neuromimetic RoboticsStructure-function of the multifunctional Ca<sup>2+</sup>/calmodulin-dependent protein kinase II
2002, Biochemical Journal