Glutamatergic synapses are the main excitatory synapses in the brain particularly in the cerebral cortex and hippocampus. More than 80% of synapses in the cortex are glutamatergic (Micheva et al., 2010). Glutamatergic transmission plays a major role in neuronal functions in the brain. Imbalances in glutamatergic signaling can lead to several neurodegenerative and psychiatric conditions (Moretto et al., 2018).

Calcium (Ca²⁺) signaling is an essential component in signal transduction at glutamatergic synapses. Calcium signals are tightly regulated since sustained elevation in Ca²⁺ levels can lead to toxicity. In glutamatergic synapses, the spike patterns of Ca²⁺ signals are thought to encode information. Decoding these signals requires the participation of efficient protein machineries that convert them into long-lasting biochemical and cellular changes representing memories. Calcium (Ca²⁺)/calmodulin (CaM)-dependent protein kinase II (CaMKII) at synapses plays a significant role in decoding Ca²⁺ spike patterns and in converting them to corresponding biochemical states. Thus, CaMKII has gained the status of a “memory molecule” by being the initiator of biochemical memory in the brain. However, the multiple isoforms and splice variants of CaMKII that assemble in varying combinations to give rise to several holoenzyme subtypes, makes it so versatile that it is involved in several other functions both in the brain and in other tissues. The phylogenetic relations of CaMKII with other kinases, its structure, its different isoforms and splice variants, biochemical and physiological functions, especially in long-term potentiation (LTP) and long-term depression (LTD), and its role in various diseases have been reviewed recently (Bayer and Schulman, 2019; Giese, 2021; Sloutsky and Stratton, 2021). Its functions specifically in the glutamatergic postsynaptic compartment have also been previously described (Hell, 2014). This article covers the basics on CaMKII including the recent advances in structure, isoforms, activation mechanisms, role in LTP and LTD, regulation of its translation, role in synapse morphology regulation, role in presynaptic mechanisms and role in various pathological conditions with emphasis on its functioning at glutamatergic synapses. In vivo models of CaMKII mutants with the associated behavioral phenotypes and CaMKII mutations reported in neurodevelopmental disorders and learning disabilities in humans have also been included in the present review.

Even though CaMKII has four distinct isoforms (α, β, γ, and δ) encoded by four different genes with molecular weight ranging from 52 to 83 kDa, α and β are the predominant ones in neurons. CaMKIIα has distinct roles in neuronal plasticity and memory. It is predominant in the hippocampal and in the neocortical areas of the brain. CaMKIIβ is enriched in cerebellum and is involved in neuronal development. While both CaMKIIα and CaMKIIβ are expressed in excitatory pyramidal neurons in the cortex and hippocampus, only CaMKIIβ is found in inhibitory interneurons in these regions (Nicole and Pacary, 2020). CaMKIIδ isoform participates in long-lasting memory storage in the hippocampus (Zalcman et al., 2018, 2019). CaMKIIγ isoform is attributed with the main function of synapse-to-nucleus communication, conveying Ca²⁺ signals to the nucleus and regulating gene expression that is essential for neural plasticity involved in memory (Ma et al., 2014; Cohen et al., 2018).

CaMKII forms large homo or hetero oligomeric assemblies of either single or multiple isoforms (Hoelz et al., 2003; Bayer and Schulman, 2019). The core sequence for the CaMKII isoforms includes an N-terminal catalytic domain, followed by a CaM binding autoregulatory domain containing Thr²⁸⁶/Thr²⁸⁷, a variable domain that is subject to alternative splicing and a C-terminal self-association domain. A linear representation of a CaMKII subunit is shown in Figure 1A.

Under basal state, the enzyme is present in an autoinhibited state with the regulatory segment bound to the substrate-docking groove on each kinase domain. It can be activated by the binding of Ca²⁺/CaM to the autoregulatory domain which releases the regulatory segment off the kinase domain. Activation of adjacent subunits can result in trans-autophosphorylation at Thr²⁸⁶ site (Rich and Schulman, 1998) in the regulatory segment which generates ‘autonomous’ kinase activity even after the initial Ca²⁺-stimulus subsides (Miller and Kennedy, 1986) by preventing the regulatory segment binding on the kinase domain. This inter-subunit autophosphorylation mechanism enables Ca²⁺-spike frequency-detection by CaMKII (De Koninck and Schulman, 1998). The autophosphorylation at Thr²⁸⁶ can also increase the affinity of the enzyme for CaM by about 1000-fold, a process termed as CaM trapping. A single autophosphorylated subunit can also rapidly phosphorylate its neighbors. Thus, a brief Ca²⁺ stimulus in the synapse can lead to the persistence of Thr²⁸⁶-autophosphorylated CaMKII that represents molecular memory (Figure 2). Autophosphorylation at Thr²⁸⁶ is an essential event in the induction of LTP that underlies memory formation.

Once Ca²⁺/CaM dissociates from the kinase, cis-autophosphorylation occurs at the CaM binding domain of CaMKII at the Thr^305/306 position. Phosphorylation at these sites, termed as “inhibitory” or “burst” autophosphorylation, prevents the binding of Ca²⁺/CaM and hence kinase cannot be further stimulated. Autophosphorylation at Thr³⁰⁵ and Thr³⁰⁶ before phosphorylation of Thr²⁸⁶ makes the kinase non-responsive to Ca²⁺/CaM stimulus and such a kinase cannot be activated. On the other hand, if Thr²⁸⁶ gets autophosphorylated first, it leads to a holoenzyme in which Thr³⁰⁵ and Thr³⁰⁶ are protected by Ca²⁺/CaM and cannot be phosphorylated (Bhattacharyya et al., 2020). It is also reported that CaMKII phosphorylation at Thr^305/306 is selectively promoted by LTD inducing stimuli and not by LTP inducing stimuli, and phosphorylation at Thr^305/306 directs CaMKII movement during LTD from excitatory to inhibitory synapses. This phosphorylation can also reduce the activity of phospho-Thr²⁸⁶ CaMKII in the absence of Ca²⁺ (Cook et al., 2021).

The first snapshot of the 3D structure of this enzyme was an electron microscopy (EM) image of CaMKII purified from rabbit skeletal muscle (Woodgett et al., 1983) that revealed a symmetrical hexagonal structure, composed of two stacked 6-membered rings. Since then, several hypotheses have been proposed about its structure in relation to its function. The catalytic/autoregulatory domains of each subunit are attached to the hexameric ring by a stalk-like appendage that presumably allows subunits to behave independently of one another for activity and Ca²⁺/CaM binding, but in concert with one another for the intra-holoenzyme autophosphorylation reaction (Figure 1B). Most of the crystallographic studies provided structures at atomic resolution of truncated forms having single or multiple domains (Hoelz et al., 2003; Rosenberg et al., 2006) giving insights on the mechanism of catalytic activity and atomic level details of the interactions holding the 3D structure and interactions between domains.

The recent studies based on single-particle EM (Myers et al., 2017; Bhattacharyya et al., 2020) in combination with biochemical and live-cell imaging experiments (Buonarati et al., 2021) further substantiated the multimeric structure of CaMKII holoenzyme having a rigid central hub complex formed by the association domains. The kinase domains are linked to the hub by the intrinsically disordered and highly flexible linker regions (residues 301–344) so that they can readily perform inter-subunit autophosphorylation. The holoenzymes range from 15–35 nm in diameter. This model also predicts that CaMKII holoenzymes can exist in three different conformations. Among these three conformations, <3% of the holoenzymes are in the compact conformation, ∼20% appears to form kinase dimers and most of the kinase domains are ordered independently both in vitro and inside the cells. CaMKII holoenzymes which appear as either compact or kinase dimers are inactive, whereas the fraction with fully extended kinase domains is in the activatable state (Figure 1B; Myers et al., 2017; Bayer and Schulman, 2019).

The formation of extended intra-holoenzyme kinase dimers could enable cooperative activation by CaM in both α and β isoforms (Myers et al., 2017; Bhattacharyya et al., 2020; Buonarati et al., 2021) but there could be distinct steric positioning of kinase domains in the CaMKIIα versus β holoenzyme due to differences in the linker length. This explains the differences in the autophosphorylation states of both the isoforms; CaMKIIα acquires Thr²⁸⁶ phosphorylation more readily than Thr^305/306 phosphorylation whereas inhibitory autophosphorylation at Thr^306/307 in CaMKIIβ occurs more readily. Inter-holoenzyme kinase dimer formation is thought to involve a high order clustering among CaMKII holoenzymes and is present in minimal quantities under normal physiological conditions for both the isoforms. But it is enhanced in both excitotoxic and ischemic conditions and the high-order CaMKII clustering formed by inter-holoenzyme kinase domain dimerization is reduced for the β isoform for both basal and excitotoxicity-induced clusters, both in vitro and in neurons (Buonarati et al., 2021). Much of the studies on holoenzyme structure have been carried out using homomers of either α or β isoforms. However, heteromultimeric CaMKII formed by α and β is known to play key functions in the brain. Structural insights into heteromultimeric forms of CaMKII would help in further advancing the understanding of the physiological functioning of this enzyme. It has been also noted that a small percentage (<4%) of holoenzymes of CaMKIIα were found as 14-mers even with full-length kinase domains (Myers et al., 2017) whereas CaMKIIβ can even exist in 16-mers (Buonarati et al., 2021). The existence of a full-length 14-mer is thought to be an intermediate state in which the exchange of subunits is possible (Myers et al., 2017) and it entails the exchange of activated subunits between two activated, or an activated and a non-activated holoenzyme (Bhattacharyya et al., 2020). This hypothesis was supported by the finding that proteolytic cleavage of the kinase domains from a 12-meric holoenzyme preparation results in the subsequent formation of 14-meric hub domain assemblies (Rosenberg et al., 2006). The function of this kind of subunit exchange is currently unknown, but it is speculated that it can be a part of repair mechanisms of individual subunits and/or synaptic plasticity mechanisms (Bayer and Schulman, 2019).

N-Methyl-D-aspartate receptor (NMDAR) is an ionotropic glutamate receptor with high Ca²⁺ permeability that plays an important role in excitatory neurotransmission in the central nervous system (CNS). Glutamate binding to α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid receptors (AMPARs) can induce depolarization in the postsynaptic membrane of glutamatergic synapses. The binding of glutamate and glycine and the depolarization-induced removal of Mg²⁺ block causes NMDAR to open and conduct Ca²⁺ and Na⁺ into the cell. This Ca²⁺ influx activates several important signaling pathways involving different protein kinases including CaMKII and phosphatases. Activated CaMKII can bind to various membrane proteins as listed in Table 1. The enzyme can interact with each of these proteins either in the Ca²⁺/CaM activated form or in the autophosphorylated form. It can bind with high affinity at the GluN2B subunit of NMDAR and phosphorylate GluN2B-Ser¹³⁰³ (Omkumar et al., 1996). GluN2B-binding can also happen at the T-site of CaMKII (site where Thr²⁸⁶ is bound in the inactive state) making the enzyme permanently active even after the Ca²⁺ stimulus subsides (Bayer et al., 2001). In addition, the kinetic parameters of CaMKII activity and its affinity for ATP are altered in an allosteric manner upon binding to GluN2B (Pradeep et al., 2009; Cheriyan et al., 2011; Madhavan et al., 2020) and this regulation is limited only to the subunit of the enzyme that binds GluN2B (Cheriyan et al., 2012). CaMKII activated in the cytosol can translocate to the postsynaptic membrane where the NMDAR complex is embedded in the postsynaptic density (PSD). CaMKII reversibly translocates to synaptic sites in response to brief stimuli, but its resident time at the synapse increases after longer stimulation (Bayer et al., 2006). It is also reported that the phosphorylation status of GluN2B at Ser¹³⁰³ also regulates GluN2B-CaMKII interaction (Raveendran et al., 2009), whereas the phosphorylation status of Ser¹³⁰³, in turn, is regulated by the action of kinases and phosphatases (Ramya et al., 2012). In the GluN2B-bound state, the enzyme becomes resistant to the action of phosphatases (Cheriyan et al., 2011) indicating GluN2B-induced structural changes which can be abolished by specific mutations in CaMKII (Mayadevi et al., 2016). This could be a possible reason for the resistance of phospho-Thr²⁸⁶-CaMKIIα to phosphatases in the PSD (Mullasseril et al., 2007). Autonomy of CaMKII due to GluN2B-binding can be terminated only by dissociation of CaMKII from GluN2B. Repeated Ca²⁺ influx through NMDAR promotes the persistent binding of CaMKII to GluN2B which occurs during LTP (Bayer et al., 2006).

Gene expression needed for long-lasting synaptic plasticity is tightly regulated. In particular, protein synthesis, regulation of mRNA transport and mRNA stability contribute to the control of gene expression. mRNA translation happens in synaptic locations - dendrites and dendritic spines, which are filled with polyribosomes, translation factors, and mRNAs (Steward and Levy, 1982; Crino and Eberwine, 1996; Job and Eberwine, 2001; Steward and Schuman, 2001).

The neurotrophins (NTs) are involved as major players in synaptic development and synaptic plasticity (Poo, 2001). Among the NTs – Neuregulin (NRG), BDNF, NT-3 and NT-4, extensive research has been done on BDNF and its role in synaptic plasticity. Postsynaptic NMDAR gating is regulated by BDNF signaling (Levine et al., 1995, 1998). BDNF is important in LTP, as seen by lack of proper establishment of LTP in BDNF knockout (KO) mouse models (Korte et al., 1995; Patterson et al., 1996). BDNF supports high-frequency transmitter release, which is required for LTP induction (Figurov et al., 1996; Gottschalk et al., 1998; Pozzo-Miller et al., 1999; Abidin et al., 2006).

Moro et al. (2020) reported reduced BDNF secretion in mouse deficient in α and β CaMKII [αβCaMKII double-knockout (DKO)] hippocampal neurons. These neurons had drastically reduced levels of BDNF and fewer BDNF containing dense core vesicles (DCV) targeted to the axon, leading to fewer DCVs per synapse and thus reduced BDNF secretion upon stimulation. CaMKIIβ is crucial for increasing the amount of secreted BDNF by CaMKIV and phospho-cAMP-response element binding protein (CREB) pathway. Interestingly, active CaMKIIβ and not CaMKIIα or inactive CaMKIIβ/CaMKIIα could restore the reduced levels of BDNF expression (Moro et al., 2020). BDNF binds to TrkB and this activates CaMKIIβ further leading to a series of downstream signaling events. Subsequently, Ca²⁺/CaM enters into the nucleus and CaMKIV gets activated, phosphorylating CREB at Ser¹³³ position, along with nuclear-localized neurogranin. Phosphorylated CREB promotes BDNF transcription (Wheeler et al., 2008; Ma et al., 2014; Wang et al., 2017). Thus, BDNF-mediated activation of CaMKIIβ acts as a positive feedback loop to initiate the expression of the neuromodulator (Moro et al., 2020).

Voltage gated calcium channels (VGCCs) are present throughout the neuronal membrane and are a major source of Ca^2+. especially in dendritic spines after a depolarization of the membrane. Different subtypes of VGCCs are known with distinct functions; mainly involved in Ca²⁺ influx into the cell as well as in regulating gene transcription. Activation of dendritic VGCCs can generate LTP, STP (short-term potentiation) or LTD. Perhaps because of the distinct subcellular localization of VGCCs, LTP induced due to their activation may use mechanisms distinct from NMDAR-dependent LTP (Malenka and Nicoll, 1999). With aging, LTP induction through NMDAR becomes lesser compared to VGCC-dependent LTP, as shown by the limited sensitivity of LTP generated in slices from older rats to NMDAR antagonists and increased sensitivity to antagonists of L-type VGCC (Izumi and Zorumski, 1998). Studies have also shown that repetitive activation of VGCCs is involved in LTD (Pöschel and Manahan-Vaughan, 2007) in a Ca²⁺-dependent manner. Among the various categories of VGCCs, L-type VGCCs are mainly involved in synaptic plasticity mechanisms.

In the CA1 area of hippocampus, an LTP component has been found that is dependent only on the activation of VGCCs without NMDAR (Grover and Teyler, 1990; Alkadhi, 2021) which was later termed as VDCC LTP. Ca²⁺ entry through VGCCs mediates LTP at thalamic input synapses to the lateral nucleus of amygdala, which may be mechanistically different from the NMDAR-dependent form of plasticity found in the hippocampus but is still dependent on activated CaMKII (Weisskopf et al., 1999). The conditional hippocampus/neocortex Cav1.2 (L-type VGCCs) KO mouse demonstrates an essential role of Cav1.2 in CREB signaling during LTP and spatial learning (Moosmang et al., 2005). In the cortical neurons, activation of T-type VGCCs enhanced LTP and CaMKII autophosphorylation (Moriguchi et al., 2012a). Even in the NMDAR-dependent mechanisms of LTP and LTD (Di Biase et al., 2008), Cav channels are involved (Zhao et al., 2021) by enhancing Ca²⁺ influx into the synaptic site and through CREB mediated events.

Upon aging, the expression of NMDAR diminishes and its subunit composition also changes (Zhao et al., 2009), whereas VGCCs, especially the L-type channels, increase in expression (Thibault and Landfield, 1996; Wang and Mattson, 2014) and can majorly involve in LTP or LTD mechanisms. Activation of L-type VGCCs, especially Cav1.2 localized in the postsynaptic membrane (Patriarchi et al., 2018) leads to Ca²⁺ influx into the spine, which can activate CaMKII. Even if the expression levels of GluN2B are lower, CaMKII can still tether to the postsynaptic site by binding with the C-terminus of Cav1.2 (Hudmon et al., 2005). This binding, however, does not lead to constitutively active CaMKII and hence, cannot support molecular memory. The enzyme tethered at the membrane can easily get activated with the trains of depolarization stimulus and can facilitate further Ca²⁺ influx through these channels (Ca²⁺-dependent facilitation).

LTD is an activity-dependent reduction in the efficacy of neuronal synapses (Malenka and Nicoll, 1999) and is thought to be involved in learning and memory. It brings about a long-lasting decrease in synaptic strength or a reversal of LTP mechanisms. LTD is triggered by synaptic activation of either NMDARs or metabotropic glutamate receptors (mGluRs). A low frequency stimulation (LFS) of NMDARs (700–900 pulses at 1 Hz) can activate LTD mechanisms (Figure 4). If the Ca²⁺ influx is low in intensity (if the activation is only for a postsynaptic compartment), it will majorly activate phosphatases and result in LTD (Baltaci et al., 2019). Initially it was thought that protein kinases are required for LTP and phosphatases are involved in LTD. But recent findings suggest that kinases are involved in LTD mechanisms also. It has been noted that the bath application of CaMKII inhibitor KN-62 could block LTD during low-frequency SC collateral stimulation (1 Hz/15 min) (Stanton and Gage, 1996). Experiments with CaMKIIα KO mice also pointed to the role of CaMKII in LTD (Stevens et al., 1994). Even though these initial experiments indicated the role of CaMKII in LTD, the exact mechanism by which CaMKII participates in the process is unknown. In contrast to the previously accepted dogma, it has also been shown by using T286A mutant mouse that Thr²⁸⁶ autophosphorylation is a requisite for LTD (Coultrap et al., 2014). The most recent studies on CaMKII autophosphorylation indicates that the autophosphorylation at Thr^305/306 is selectively induced by LTD stimuli and the mutation of these residues impairs LFS-induced LTD but not HFS-induced LTP (Cook et al., 2021). Both the autophosphorylations are necessary for LTD but the exact role of Thr²⁸⁶ with respect to Thr^305/306 in LTD remains controversial. The death-associated protein kinase 1 (DAPK1) can regulate CaMKII-GluN2B interaction to facilitate LTD. DAPK1 is a CaM kinase family member and is enriched in excitatory synapses. They can bind to GluN2B at a site overlapping the CaMKII binding site. The enzyme gets activated by CaN, a Ca²⁺-activated protein phosphatase. LTD-stimuli can activate DAPK1 in hippocampal slices in a CaN-dependent manner. Inhibition of DAPK1 or CaN allowed the accumulation of CaMKII at excitatory synapses after LTD-stimuli (Goodell et al., 2017). This indicates that during LTD, DAPKI activated by phosphatases will compete for GluN2B binding and would reduce the binding of activated CaMKII generated by the low frequency stimuli.

CaMKII can phosphorylate Ser⁵⁶⁷ residue of GluA1 subunit of AMPAR, a unique phosphorylation site for CaMKII in the C-terminal loop of GluA1. The C-terminal tail of GluA1 is involved in AMPAR trafficking from extra-synaptic pool to the synapses. Phosphorylation of GluA1 at Ser⁵⁶⁷ by CaMKII inhibits AMPAR trafficking to the synapses (Lu et al., 2010). It has been noted that LTD-inducing stimulation of hippocampal slices produced a robust phosphorylation of Ser⁵⁶⁷ whereas LTP-inducing stimulus could yield only Ser⁸³¹ phosphorylation. The differential phosphorylation of GluA1 by CaMKII under the two synaptic plasticity conditions underlies the role of CaMKII in LTD (Coultrap et al., 2014).

In contrast to spine enlargement in LTP, LTD is associated with spine shrinkage aided by the removal of the AMPA receptor regulatory scaffold protein, A-kinase anchoring protein (AKAP) 79/150. The synaptic removal of AKAP79/150 is brought about by the phosphorylation of the substrate sites within the AKAP79/150 N-terminal polybasic membrane-cytoskeletal targeting domain (residues 1–153) by CaMKII. Phosphorylation by CaMKII inhibits AKAP79/150 association with F-actin, thus facilitating AKAP79/150 removal from spines (Figure 4). In addition to the direct phosphorylation of AKAP79/150, CaMKII is also responsible for its depalmitoylation on two Cys residues within the N-terminal targeting domain. Depalmitoylation also promotes synaptic elimination of AKAP79/150. Since the protein harbors PKA and protein phosphatase 2B (PP2B) at the PSD, it can regulate both synaptic insertion and elimination of AMPARs. Under LTP stimulation, PKA can phosphorylate Ser⁸⁴⁵ of GluA1 of AMPAR and thereby more AMPAR trafficking to the synapse occurs, whereas in LTD conditions due to the elimination of AKAP79/150 along with activation of phosphatases, AMPAR dephosphorylation at Ser⁸⁴⁵ and its endocytosis is promoted which eventually leads to spine shrinkage (Woolfrey et al., 2018).

The stimulation pattern-dependent activation of NMDAR that yields either LTP or LTD, causes activation of CaMKII in either case. With the differing stimuli the enzyme targets different substrates and thereby activates specific signaling mechanisms to yield either form of synaptic plasticity.

CaMKII plays an important role in several physiological pathways including synaptic plasticity and hence its localization in the cytosol and PSD are crucial determinants of its function. Immunoelectron microscopy studies show that CaMKIIα is significantly higher in dendritic shafts when compared to dendritic spines. When it gets any proper stimulus, it will abundantly translocate to the spines (Shen and Meyer, 1999; Shen et al., 2000; Ding et al., 2013). In the basal condition, more CaMKII will be available in the dendritic shaft than in spines. Whenever activation happens the activated CaMKII can translocate to the spines.

Translocated CaMKII can bind with various protein ligands in the PSD as indicated in Table 1. One such protein is densin-180, which is a core protein in the PSD that does not span the membrane. Though densin-180 is the only documented interaction partner for the association domain of CaMKII, it will not bind with CaMKII holoenzymes which contain β isoform (Penny and Gold, 2018). The PDZ domain of densin-180 contributes to its binding to α-actinin. A distinct domain of α-actinin interacts with the kinase domains of both α and β subunits of CaMKII. Thus, these three proteins can form a ternary complex in the PSD stabilized by multiple interactions (Walikonis et al., 2001). This ternary complex within the PSD is an additional mode of localization of CaMKII to PSD apart from its binding to GluN2B.

SAP97, a member of membrane-associated guanylate kinase protein family, has been implicated in the processes of targeting ionotropic glutamate receptors such as NMDARs and AMPARs at postsynaptic sites and is enriched in PSD. SAP97 shares its interaction with AKAP79/150 in addition to the C-terminal region of GluA1. AKAP79/150 in turn harbors PKC, PKA and PP2B. This molecular arrangement inside the PSD works in accordance with the stimuli received. The most important function of this complex is the regulation of AMPARs in synapses including both potentiation and trafficking. CaMKIIα displays a high degree of co-localization with SAP97. CaMKII phosphorylation of Ser³⁹ in the N-terminus of SAP97 modulates trafficking of SAP97 (Mauceri et al., 2004) and the associated proteins; in contrast, CaMKII phosphorylation of Ser²³² in the first PDZ domain of SAP97 may modulate binding of other proteins, such as NMDAR and AMPAR subunits (Nikandrova et al., 2010), especially GluA1 of AMPAR. SAP97 is in close association with AKAP 79/150, but the phosphorylation of SAP97 at Ser³⁹ by CaMKII disengages AKAP79/150 from regulating GluA1-AMPARs.

Another complex associated with CaMKII in the PSD is the complex formed by SynGAP, MUPP1 and CaMKII. SynGAP and CaMKII are brought together by direct physical interaction with the PDZ domains of MUPP1, a multi-PDZ domain-containing protein (Krapivinsky et al., 2004). In this complex, SynGAP is phosphorylated by CaMKII which enhances its Ras GTPase activity which in turn promotes AMPAR trafficking as shown in Figure 3.

CaMKII has an important role in dendritic spine remodeling upon synaptic stimulation. Electron micrographic studies showed that at physiological molar ratios, single CaMKII holoenzymes cross-linked multiple F-actin filaments at random, whereas at higher CaMKII/F-actin ratios, filaments bundled. From this bundled state CaMKII is released upon Ca²⁺/CaM activation, triggering network disassembly and expansion leading to spine enlargement. Upon subsequent disappearance of Ca²⁺, compaction will occur (Khan et al., 2019).

Excitotoxicity is a pathological condition triggered by excessive stimulation of receptors by excitatory neurotransmitters, primarily glutamate, causing Ca²⁺ overload in the cytosol and thereby resulting in neuronal dysfunction and cell death. Increased Ca²⁺ influx and high intracellular Ca²⁺ ([Ca²⁺]_i) rise trigger gene expression (Ortuño-Sahagún et al., 2012) and long-lasting activation of CaMKIIα in hippocampal neurons (Otmakhov et al., 2015). Autophosphorylation of CaMKII at Thr²⁵³, Thr²⁸⁶ (Vest et al., 2010; Otmakhov et al., 2015; Rostas et al., 2017) and simultaneous S-nitrosylation at Cys²⁸⁰/Cys²⁸⁹ by nitric oxide (NO) (Coultrap and Bayer, 2014) generates autonomous activity of the kinase during excitotoxic cell death. Activated CaMKII redistributes to the spines (Otmakhov et al., 2015), promotes its interaction with synaptic GluN2B (Wang N. et al., 2014; Buonarati et al., 2020) and mediates the NMDA-induced caspase-3-dependent cell death pathway (Goebel, 2009). During a glutamate-induced excitotoxic event, CaMKII can also modulate the activity of neuronal nitric oxide synthase (nNOS) (Araki et al., 2020), can cause axonal degeneration via necroptosis (Hou et al., 2009; Arrazola and Court, 2019) and also contribute to the regulated necrosis (RN) pathway (Wang S. et al., 2019).

Contrastingly both overexpression (Vest et al., 2010) and sustained CaMKII inhibition during excitotoxicity can exacerbate cell death of cultured neurons (Ashpole and Hudmon, 2011; Ashpole et al., 2012). Loss of CaMKII activity in astrocytes results in dysregulated Ca²⁺ homeostasis and reduced glutamate uptake (Ashpole et al., 2013) by excitatory amino acid transporter 1 (EAAT1) (Chawla et al., 2017). On the whole, dysregulated CaMKII function upon excitotoxic insult shifts the tight homeostatic balance maintained between kinases and phosphatases in the cell, resulting in dysfunction of excitatory synaptic transmission (Farinelli et al., 2012). The following section reviews the role of CaMKII at glutamatergic synapses in a few diseases in which excitotoxicity is one of the causes.

CaMKII plays a versatile role in different regulatory processes involved in synaptic plasticity. This section reviews the different CaMKII mutant animal models generated to study the physiological role of the kinase in synaptic plasticity and its associated behavioral phenotype. Targeted disruption of CaMKIIα/β/γ function in vivo dysregulates different types of synaptic plasticity (Silva et al., 1992a; Stevens et al., 1994; Mayford et al., 1995; Giese et al., 1998; Elgersma et al., 2002; Miller et al., 2002; Cho et al., 2007; van Woerden et al., 2009; Yamagata et al., 2009; Yin et al., 2017; Cohen et al., 2018; Kool et al., 2019) and impairs learning (Silva et al., 1992b,1996; Bach et al., 1995; Giese et al., 1998; Elgersma et al., 2002; Irvine et al., 2005; Yamagata et al., 2009; Borgesius et al., 2011; Achterberg et al., 2014; Cohen et al., 2018), memory (Miller et al., 2002; von Hertzen and Giese, 2005; Cho et al., 2007) and the emotional state (Chen et al., 1994; Yamasaki et al., 2008; Hasegawa et al., 2009; Bachstetter et al., 2014). Although the behavior exhibited varies slightly with the genetic background of the mouse strain used (Gordon et al., 1996; Silva et al., 1996; Hinds et al., 1998; Need and Giese, 2003), the molecular and electrophysiological alterations remain largely unchanged.

In humans, de novo mutations in CaMKII have been identified and reported majorly in cases of neurodevelopmental disorders (NDDs) (Study, 2017; Akita et al., 2018) and intellectual disability (ID) (Küry et al., 2017). The role of CaMKII and glutamatergic signaling in neuropsychiatric diseases has been reviewed by Robison (Robison, 2014; see also Nicole and Pacary, 2020). Supplementary Table 2 summarizes the different CaMKII variants reported with their functional implications and clinical manifestations if any. The type of mutation (synonymous, missense, splice region, frameshift, deletion), the specific CaMKII isoform (α, β and γ) that is mutated and the protein domain (catalytic, auto-regulatory or association) affected determine the disease phenotype. The zygosity of inheritance (heterozygous/homozygous) can also influence the pathogenicity of the variant (Chia et al., 2018); however, intrafamilial variations in the expression of disease symptoms by subjects carrying the same heterozygous variant, have also been reported (Heiman et al., 2021).

Clinical manifestations of the identified mutations range from global neurodevelopmental delay, seizures, mild to severe ID, hypotonia, delayed development of motor and speech/language skills, abnormal emotional behavior, cerebellar atrophy, facial dysmorphism, visual impairment and gastrointestinal issues. Dysfunction of CaMKIIα can cause seizure-associated activity in the forebrain (Akita et al., 2018) and pronounced motor delay (Küry et al., 2017), while individuals with CaMKIIβ variants exhibit severe ID accompanied with hypotonia (Küry et al., 2017) and cerebellar atrophy (Akita et al., 2018). Facial dysmorphisms along with severe ID and severe hypotonia has been reported in patients carrying a CAMK2G variant (Proietti Onori et al., 2018). The vast majority of the variants identified, affect amino acids conserved across species (Küry et al., 2017; Stephenson et al., 2017; Akita et al., 2018; Chia et al., 2018; Proietti Onori et al., 2018), which may explain the degree of severity of pathogenicity.

Stephenson et al. (2017) reported the first characterization of a de novo missense mutation in the CAMK2A gene, encoding for CaMKIIα, that was found in a patient with autism spectrum disorder (ASD) (Iossifov et al., 2014). Replacement of Glu with Val at 183^rd position in the catalytic domain of CaMKIIα (CaMKIIα^Glu183Val) disrupts the interaction of CaMKII with ASD-associated proteins, such as Shank3 (SH3 and multiple ankyrin repeat domains 3) (Jiang and Ehlers, 2013), GluN2B (Pan et al., 2015) and the metabotropic glutamate receptor mGlu5 (Chana et al., 2015), which can reduce targeting of CaMKIIα to spines (Stephenson et al., 2017). Neuronal expression of CaMKIIα^Glu183Val disrupts AMPAR-mediated synaptic transmission, interferes with CaMKII autophosphorylation and reduces dendritic spine density. Heterozygous (Camk2a^WT/E183V) and homozygous (Camk2a^E183V/E183V) KI mice displayed enhanced repetitive behaviors and deficits in social interactions, which mimic symptoms of ASD (Stephenson et al., 2017). Decreased autoinhibition and increased Thr²⁸⁶ autophosphorylation of the CaMKIIα^Pro212Gln mutant, identified in an individual with NDD, affects the efficiency of excitatory synaptic transmission by enhancing K⁺ currents in dendrites in vitro (Akita et al., 2018).

A biallelic, germline, loss-of-function CAMK2A missense mutation, CAMK2A^{p.(His477Tyr)} in the association domain of CaMKIIα, was reported in two siblings displaying psychomotor retardation, frequent seizures and severe ID (Chia et al., 2018). Compared to the WT enzyme, the mutant form disrupts CaMKIIα self-oligomerization and holoenzyme assembly which in turn affects its subcellular localization in neurons and ability to support synaptic function in vivo (Chia et al., 2018). Recently, Brown et al. (2021) characterized six heterozygous variants of CAMK2A found in patients with schizophrenia. The p.(Arg396*) mutation in the association domain of CaMKIIα ablates holoenzyme formation, impairs GluN2B binding and consequently fails to accumulate at excitatory synapses in response to a LTP stimulus. While both p.(Arg396*) and p.(Arg8His) variants of CAMK2A exhibited impaired autophosphorylation at Thr²⁸⁶, only the p.(Arg8His) mutation in the kinase domain significantly affected the Ca²⁺/CaM-stimulated kinase activity (Brown et al., 2021). The absence of impaired function or expression for the remaining four mutants studied indicates that the mere occurrence of a mutation in a patient does not imply that the disease is caused by the mutation (Brown et al., 2021).

In addition to NDDs and ID, CAMK2A variants/single nucleotide polymorphisms (SNPs)/single nucleotide variants (SNVs) have been reported to be associated with risk for bipolar disorder (BD) in cohorts of European descent (Ament et al., 2015), in sporadic AD patients belonging to the Han Chinese population (Fang et al., 2019) and mild cognitive impairment (MCI) subjects in a Spanish population (Bufill et al., 2015). Deletion of the chromosome at 5q32, covering CAMK2A, might be responsible for mild ID observed in two patients diagnosed with mandibulofacial dysostosis (Vincent et al., 2014). Interestingly, CAMK2A genetic variants have been reported to be nominally associated with non-verbal communication in ASD cohorts (Chiocchetti et al., 2018) and logical memory performance in the elderly people (Rhein et al., 2020). CAMK2A polymorphisms can also influence spatial working memory in Caucasian adolescents (Easton et al., 2013) and cognitive ability in Taiwanese senior high school students (Lee et al., 2021). Recruitment of higher number of subjects from distinct populations is warranted to further validate the association of CAMK2A SNPs in genotype-phenotype association studies (Chiocchetti et al., 2018; Rhein et al., 2020).

Apart from mutations in CAMK2A, de novo mutations in CAMK2B have also been reported in 10 unrelated individuals exhibiting mild to severe ID (Küry et al., 2017). There are 19 rare variants of CAMK2A and CAMK2B that are heterozygous nonsense, missense or splice-site mutations affecting the catalytic or auto-regulatory domain of CaMKII. The identified variants could affect protein expression and autophosphorylation at Thr²⁸⁶/Thr²⁸⁷ when expressed in vitro, and cause deficits in neuronal migration in vivo (Küry et al., 2017). De novo mutations in CAMK2A and CAMK2B can also result in varying neurodevelopmental phenotypes (Akita et al., 2018). The missense variants disrupted the interaction between the catalytic domain and the regulatory segment, leading to increased Ca²⁺-independent activity (Akita et al., 2018).

A heterozygous c.85C>T, p.(Arg29*) mutation in CAMK2B was found in a patient with mild ID, delayed speech development and seizures (Küry et al., 2017). This mutation was also reported in a 3-year-old European girl with complex focal seizures and global neurodevelopmental delay (Heiman et al., 2021). This maternally inherited pathogenic variant of the CAMK2B gene only mildly affected the patient’s sibling, with the same variant, while the mother was phenotypically healthy and intellectually normal (Heiman et al., 2021). Similarly, a heterozygous c.416C>T, p.(Pro139Leu) variant of CAMK2B found in four Caucasian patients presented with severe ID, global developmental delay, hypotonia and microcephaly (Küry et al., 2017), was also reported in a 22-year-old South Asian woman (Rizzi et al., 2020) as well as in a MECP-2 (methyl-CpG binding protein 2) negative proband of Japanese origin (Iwama et al., 2019). The recurrence of a few pathogenic variants of CAMK2A and CAMK2B (Küry et al., 2017) calls for elaborate functional studies of the mutant proteins both in vitro and in vivo (Onori and van Woerden, 2021).

De novo mutations in CAMK2G have also been identified and reported in cases of NDDs and severe ID (De Ligt et al., 2012; Study, 2017; Proietti Onori et al., 2018). Whole-exome sequencing performed on two patients revealed c.875G>C, p.(Arg292Pro) mutation in the auto-regulatory domain of CaMKIIγ, which is a putative CaM trapping region. One research group showed that CAMK2G^{p.(Arg292Pro)} affects protein stability in vitro and functions as a pathogenic gain-of-function mutation by rendering it constitutively active and by blocking neuronal migration during development in vivo (Proietti Onori et al., 2018). The pathogenicity of the mutant is dependent on its catalytic activity (Proietti Onori et al., 2018). Cohen et al. (2018) reported that the ID observed in these patients might be due to the inability of CaMKIIγ^Arg292Pro to effectively trap CaM and shuttle Ca²⁺/CaM complex to the nucleus, thereby disrupting a major link connecting activation of NMDARs and Cav1 channels to nuclear transcription of BDNF, c-Fos and Arc. This in turn adversely affects synaptic strengthening and LTM in vivo. Similar to CAMK2A SNPs reported, a genetic cluster containing CAMK2G polymorphisms has been identified to be associated with episodic memory performance (Dominique and Papassotiropoulos, 2006).

Although the CAMK2 variants reported so far shed light on the probable role of the kinase in mediating disease symptoms, the number of human subjects identified with the mutation is insufficient, compared to the samples tested, to correlate the variant to the disease with good statistical power. Neither is it mandatory for the identified variant(s) to be a causal factor in the diseased phenotype (Brown et al., 2021), nor can an indirect role by the mutant protein be overlooked. More detailed functional characterization of the identified and reported CaMKII mutations than what is already reported, both in vitro (Küry et al., 2017; Stephenson et al., 2017; Akita et al., 2018; Chia et al., 2018; Cohen et al., 2018; Proietti Onori et al., 2018; Brown et al., 2021) and in vivo (Stephenson et al., 2017; Chia et al., 2018), can further substantiate the critical role of CaMKII mutants in disease conditions. Nonetheless, screening for either sporadic or inherited CAMK2 variants in disorders majorly affecting cognition, can help in unraveling the theragnostic potential of CaMKII, if any.

Antagonists against glutamate receptors, majorly NMDARs, have been designed, synthesized and evaluated for their efficacy in preventing excitotoxicity in CNS diseases (Liu et al., 2020; Chandran et al., 2021). Signaling molecules downstream to NMDARs, like CaMKII, can also be targeted to restore Ca²⁺ and glutamate homeostasis at synapses (Vest et al., 2010). Likewise, CaMKII has been exploited as a potential drug target in neuropsychiatric and neurodegenerative diseases (Sałaciak et al., 2021).

Based on the differential regulation of CaMKII function during neurotoxicity, the modulators either enhance (Yamamoto et al., 2009; Zeng et al., 2010; Wang D.M. et al., 2013; Wei et al., 2013; Wang S.Q. et al., 2014) or inhibit CaMKII activity (Wang D. et al., 2013; Jiang et al., 2019). Their modus operandi includes binding to Ca²⁺/CaM binding site of CaMKII (Brooks and Tavalin, 2011; Wong et al., 2019), targeting autonomous CaMKII activity (Coultrap et al., 2011; Wang D.M. et al., 2013; Wang et al., 2016; Deng et al., 2017), interacting with CaMKII hub domain (Leurs et al., 2021), preventing CaMKII translocation to the synaptic membrane (Matsumoto et al., 2008), inhibiting GluN2B-CaMKII binding (Tullis et al., 2021) or by modulating CaMKII-mediated signaling pathways (Liu et al., 2012; Matsumoto et al., 2013; Wei et al., 2013; Zhang et al., 2017; Islam et al., 2019; Wu et al., 2019; Izumi et al., 2020; Chen et al., 2021). The different CaMKII modulators reported from studies involving glutamatergic synapses in neurons are listed below:

Although a majority of the modulators have been widely employed to combat neuronal glutamatergic dysfunction in vitro and in vivo, their clinical use will require extensive studies on possible side effects and toxicity to cardiac health too (Nassal et al., 2020).

Majority of the biochemical and structural studies on CaMKII have been performed on homomeric holoenzymes of one of the isoforms. However, under physiological conditions the enzyme can form hetero-multimers with different subunit stoichiometries. The relative abundance of different heteromeric subtypes under different developmental stages and in different regions would be an important determinant in the physiological functioning of CaMKII. Detailed studies on the functional variation among heteromers and their participation in specific cellular functions would be essential in making further progress in understanding the physiological functions of CaMKII. Further refinement of molecular genetic techniques for ectopic expression of isoforms and mutants of CaMKII with better control on heteromer formation would be essential for progress toward this goal.

The binding of CaMKII to GluN2B modulates the kinetic parameters of CaMKII enzyme activity and attenuates dephosphorylation of CaMKII (Pradeep et al., 2009; Cheriyan et al., 2011). These regulatory events strongly support the bistable switch model of molecular memory involving CaMKII and PP1. However, the physiological relevance of these regulatory events and the existence of the bistable switch in vivo needs to be demonstrated. Elucidation of the structure of the CaMKII-GluN2B complex could contribute significantly toward understanding the physiological functions of this complex.

CaMKII activation is a prerequisite for both LTP and LTD and recently it has been shown (Woolfrey et al., 2018) that two types of substrates (high autonomy, low autonomy) are preferred under each condition. If so, what are the exact signaling events with respect to the amount of Ca²⁺ entry into the postsynaptic site responsible for each of these events?

The recurrence of a few pathogenic variants of CAMK2A and CAMK2B (Küry et al., 2017) calls for elaborate functional studies of the mutant proteins both in vitro and in vivo (Onori and van Woerden, 2021). It also encourages more extensive screening for genetic variants of CaMKII in human populations.

CaMKII is an enzyme highly enriched in the brain. It has important roles to play in the functioning of glutamatergic synapses. Significant advances have been made in understanding the structure, function and physiological role of CaMKII. Its contribution to learning and memory has been investigated extensively with the help of most modern techniques. This has unequivocally established the integral part played by this enzyme in learning and memory. However, there is still more to be understood about the exact manner in which CaMKII participates in the underlying cellular mechanisms such as synaptic plasticity. Novel features of the structure and biochemical regulation of CaMKII are still being revealed by biophysical and biochemical experiments. Since the molecular properties of CaMKII holoenzyme are among the foundations on which most of the models of synaptic plasticity, learning and memory are built, progress in structural studies would continue to necessitate revisions in these models. Powerful molecular genetic techniques have permitted the controlled expression of isoforms and mutants of CaMKII in specific cell types in the brain of model organisms leading to important insights into its role in the cellular and systemic mechanisms. However, it has been difficult to dissect out all of its synaptic functions from cellular functions due to technical hurdles. The occurrence of heteromeric subtypes of CaMKII and the redundancy in the function among the isozymes also poses challenges to molecular genetic interrogation of its cellular functions. Progress in the understanding of CaMKII has prompted attempts to pursue it as a therapeutic target for pharmacological and genetic interventions since it is part of impaired Ca²⁺ signaling in many disease conditions. A limited number of genetic variants of CaMKII have been found associated with human neurological disease conditions. The central role of CaMKII in brain functions calls for large scale screening for CaMKII variants in human populations.

AM and RO contributed to conception and design of the manuscript. AM, SG, RJ, and RO co-wrote the manuscript. All authors contributed to manuscript revision, read and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.