In Vitro Reconstitution of a CaMKII Memory Switch by an NMDA Receptor-Derived Peptide

Abstract

Ca²⁺/Calmodulin-dependent protein kinase II (CaMKII) has been shown to play a major role in establishing memories through complex molecular interactions including phosphorylation of multiple synaptic targets. However, it is still controversial whether CaMKII itself serves as a molecular memory because of a lack of direct evidence. Here, we show that a single holoenzyme of CaMKII per se serves as an erasable molecular memory switch. We reconstituted Ca²⁺/Calmodulin-dependent CaMKII autophosphorylation in the presence of protein phosphatase 1 in vitro, and found that CaMKII phosphorylation shows a switch-like response with history dependence (hysteresis) only in the presence of an N-methyl-D-aspartate receptor-derived peptide. This hysteresis is Ca²⁺ and protein phosphatase 1 concentration-dependent, indicating that the CaMKII memory switch is not simply caused by an N-methyl-D-aspartate receptor-derived peptide lock of CaMKII in an active conformation. Mutation of a phosphorylation site of the peptide shifted the Ca²⁺ range of hysteresis. These functions may be crucial for induction and maintenance of long-term synaptic plasticity at hippocampal synapses.

Introduction

In the brain, Ca²⁺/Calmodulin (Ca²⁺/CaM)-dependent protein kinase II (CaMKII) has been shown to play a critical role in learning and memory, including synaptic long-term potentiation (LTP) (1–3). CaMKII has been theoretically hypothesized to be a memory switch, which transforms a transient signal into persistent LTP. CaMKII forms a large symmetrical holoenzyme in which 12 kinase domains are tightly packed around a central ring-shaped scaffold (4,5), and the cooperative autophosphorylation at threonine 286 (T286) by neighboring subunits form two stable steady states: a dephosphorylated OFF state and a phosphorylated ON state (6,7). Upon LTP induction, memories are established via a transient Ca²⁺ signal that switches CaMKII from the dephosphorylated OFF state to the phosphorylated ON state.

To realize such a memory switch at synapses, the interaction between CaMKII and the N-methyl-D-aspartate receptor subunit GluN2B has been shown to be critically important. The interaction between CaMKII and GluN2B is required for persistent activation and translocation of CaMKII (8,9), persistent T286 phosphorylation (10), and LTP maintenance (11,12). Thus, the interaction between CaMKII and GluN2B is a key factor in memory formation. However, the interaction between CaMKII and GluN2B as a memory switch has not been well described.

Using an in vitro reconstitution system (13), we report that CaMKII operates as a memory switch only in the presence of a GluN2B-derived peptide (residues 1289−1310; N2Bs). Hysteresis, which is the dependence of a state on both its current and past environment (14), shows that the Ca²⁺ range of the memory switch is regulated by phosphorylation of N2Bs at serine 1303 (S1303). Our results show that this CaMKII system can be a minimal component of an erasable memory switch, and no complex molecular interaction with multiple synaptic targets is needed for enabling this function.

Materials and Methods

CaMKII expression and purification

Wild-type (WT) rat alpha-CaMKII cDNA and baculovirus containing alpha-CaMKII cDNA were kindly provided by Dr. Yasunori Hayashi (RIKEN Brain Science Institute) (15) and Haruhiko Bito (The University of Tokyo). The virus containing alpha-CaMKII was added to 175-cm² monolayer cultures of Sf21 cells for 60–72 h, and the cells were then harvested and stored at −80°C until protein purification (13,16–19).

Alpha-CaMKII was purified essentially as described previously (13,16–19). Briefly, cell pellets were resuspended in lysis buffer (50 mM PIPES at pH 7.0, 1 mM EGTA, 1 mM EDTA, 2.5% Betaine, and 1× complete protease inhibitor cocktail (Roche, Basel, Switzerland)) (16), lysed by brief sonication, and clarified by centrifugation at 100,000 × g, 4°C for 0.5 h. The supernatant was loaded onto a 2 mL phosphocellulose column (P11 cation exchange resin; GE Healthcare, Buckinghamshire, UK) equilibrated with wash buffer (50 mM PIPES at pH 7.0, 200 μM EDTA, 100 mM NaCl, and 1× complete protease inhibitor cocktail) (19). The column was washed in wash buffer, and CaMKII was eluted with wash buffer plus 500 mM NaCl. The eluted fractions were then applied to a Sephacryl S300 Sepharose gel filtration column (GE healthcare) that had been equilibrated in gel filtration buffer (30 mM MOPS at pH 7.2, 200 mM KCl, 10 μM EDTA, and 1 mM 2-mercaptoethanol) (19). The CaMKII-containing fractions were collected and concentrated by a centrifugal filter unit (NMWL: 100 kDa, Amicon Ultra-4; Millipore, Bedford, MA) with gel filtration buffer plus 50% glycerol. 30–50 μM CaMKII (subunit concentration) was subdivided into 1.5 mL tubes, and stored at −80°C until use.

Protein phosphatase 1 expression and purification

The His-tagged protein phosphatase 1a (PP1a) catalytic subunit cDNA in the vector pDR540 was kindly provided by Dr. Angus Clark Nairn (Yale University) (13,20,21). Escherichia coli BL21 cells harboring the vector were grown at room temperature in 1 L Lennox broth (LB) medium with 50 mg/mL ampicillin and 100 mM CoCl₂ until OD₆₀₀ = 0.4, and the expression of PP1 was induced overnight with 0.5 mM isopropyl β-D-thiogalactoside. The cells were harvested and resuspended in resuspension buffer (10 mM Tris-HCl at pH 8.0, 30 mM imidazole, 10% glycerol, 300 mM NaCl, 1 mM CoCl₂, and 1 mM phenylmethylsulfonyl fluoride). The cell suspension was lysed by 30 s sonication, and centrifuged at 30,000 × g, 4°C for 30 min. The supernatant was loaded onto a 2 mL Ni-NTA agarose column (Qiagen, Hilden, Germany) that had been equilibrated in resuspension buffer. The column was washed with wash buffer (30 mM MOPS at pH 7.2, 200 mM KCl, and 10% glycerol), and PP1 was eluted with wash buffer plus 400 mM imidazole. The PP1-containing fractions were pooled, applied to a HiTrap desalting column (GE Healthcare), equilibrated with wash buffer to remove imidazole, and concentrated using a centrifugal filter unit (NMWL: 10 kDa, Amicon Ultra-4; Millipore). Approximately 5 μM PP1 was subdivided into 1.5 mL tubes and stored at −80°C until use. Our PP1 had K_m = 10 μM and k_cat = 0.043 s⁻¹ for dephosphorylation of CaMKII at T286.

Preparation of CaMKII phosphorylated at T286

After incubation with 19 μM free Ca²⁺ for 10 min at 4°C, almost all CaMKII showed an upward band shift in Phos-tag Western blotting (22), indicating phosphorylation. The shifted band reacted to the mouse monoclonal antibody 22B1, which specifically recognizes CaMKII phosphorylation at T286. We used this preparation of CaMKII phosphorylated at T286 as a phosphorylated ON state for coming down experiments. We used a twofold concentrated system of Ca²⁺-EGTA buffer for the preparation of indicated free Ca²⁺ solutions (see Figs. 1–3; Fig. S1 and Fig. S2) (13) and calculated the free Ca²⁺ concentrations based on a fluorescence titration with Fluo-4 (the dissociation constant of Ca²⁺ for EGTA, 110 nM; Invitrogen, Carlsbad, CA).

Figure 1.

In vitro reconstitution of a memory switch comprising CaMKII and N2Bs. (A) In vitro reconstitution system with a Ctl peptide (GluN2B 1095−1119), where CaMKII functions as a memoryless switch (left), and with N2Bs (GluN2B 1289−1310), where CaMKII functions as a memory switch (right). (B) Ca²⁺ pulse stimulation (black); 20 μM Ca²⁺ for 10 min and 0.56 μM (basal Ca²⁺ concentration) afterward (top). T286 phosphorylation of CaMKII (T286P) with Ctl (bottom, left) or with N2Bs (bottom, right) in response to the pulse stimulation (solid circles) and to the basal Ca²⁺ concentration (open circles). The CaMKII kinase inhibitor (K252a) (dark blue circles) or control buffer (light blue circles) was added at 3 h. Error bars indicate mean ± SD (n ≥ 3). To see this figure in color, go online.

Figure 2.

Hysteresis of T286 phosphorylation in the presence of N2Bs. (A) Coming down Ca²⁺ dose-response curve of T286 phosphorylation (T286P) from a phosphorylated ON state (solid circles) and that of going up from a dephosphorylated OFF state (open circles) in the absence (left) or the presence of N2Bs (middle and right). The concentrations of PP1 were 0.025 μM (left and middle) and 0.5 μM (right), and the incubation time was 6 h. Solid and dashed lines denote the best fits to the Hill equation. The EC₅₀s and Hill coefficients (n_h) are indicated with 95% confidence intervals. (B) T286 phosphorylated (black) and dephosphorylated (white) CaMKII at 3 h was mixed with ratios of 3:1 (red), 2:2 (green), or 1:3 (blue). To see this figure in color, go online.

Figure 3.

N2Bs phosphorylation state regulates the Ca²⁺ range of hysteresis. Going up (solid circles) and coming down (open circles) Ca²⁺ dose-response curves in the presence of the dephosphomimetic mutant peptide (S1303A) (A), the WT peptide (B), or the phosphomimetic peptide (S1303D) (C). Solid and dashed lines denote the best fits to the Hill equation. The EC₅₀s and Hill coefficients (n_h) are indicated with 95% confidence intervals.

CaMKII autophosphorylation assay

CaMKII autophosphorylation experiments were performed based on a previous study (13) at 4°C in 20 μL of 2 μM CaM, 0.025 μM PP1, 2 μM CaMKII (subunit concentration), 50 μM N2Bs or Ctl, 5 mM Ca²⁺-EGTA, 4 mM MgCl₂, 2 mM ATP, 200 mM KCl, 30 mM MOPS at pH 7.2, and 0.1% w/v bovine serum albumin (BSA) (standard condition). All reactions in this study were performed under this standard condition, unless specified otherwise. N2Bs and Ctl peptides were derived from residues 1289−1310 (KAQKK NRNKL PRQHS YDTFV DL) and 1095−1119 (SAKSR REFDE IELAY RRRPP RSPDH) of GluN2B, respectively (GL Biochem, Shanghai, China) (8), and CaM was purchased from Wako Pure Chemicals (Osaka, Japan). For coming down experiments from a phosphorylated ON state of CaMKII, 8.4 μL of solution A (2.38× solution of CaM, PP1, CaMKII, N2Bs/Ctl, and BSA in 400 mM KCl and 30 mM MOPS at pH 7.2) was first added to 5 μL of solution B (2× solution of Ca²⁺-EGTA in 30 mM MOPS at pH 7.2) (Calcium Calibration Buffer Kit, Invitrogen). Then, 1.6 μL of solution C (12.5× solution of MgCl₂ and ATP in 200 mM KCl and 30 mM MOPS at pH 7.2) was added to initiate a reaction for preparation of a phosphorylated ON state of CaMKII. The mixture of solutions A, B, and C gave 19 μM free Ca²⁺ in 15 μL of 2.7 μM CaM, 0.33 μM PP1, 2.7 μM CaMKII, 67 μM N2Bs or Ctl, 3.3 mM Ca²⁺-EGTA, 5.3 mM MgCl₂, 2.7 mM ATP, 267 mM KCl, 30 mM MOPS at pH 7.2, and 0.13% w/v BSA. CaMKII was phosphorylated at T286 for 10 min incubation (22). Then, 5 μL of solution D (2× solution of K⁺-EGTA and Ca²⁺-EGTA mixture in 30 mM MOPS at pH 7.2) (Calcium Calibration Buffer Kit, Invitrogen) was added to obtain the final reaction condition with the indicated free Ca²⁺ concentration. For going up experiments from a dephosphorylated OFF state of CaMKII, the order of addition of solutions C and D was reversed, and solutions A, B, D, and C were sequentially added.

Autophosphorylation reactions were stopped by applying the reaction buffer to Laemmli’s sodium dodecyl sulfate (SDS) sample buffer, and the samples were subjected to SDS polyacrylamide gel electrophoresis, followed by immunoblotting with the mouse monoclonal antibody 22B1. T286 phosphorylation was quantified by the software Phoretix 1D (TotalLab, Newcastle upon Tyne, UK). The Ca²⁺ dose-responses of CaMKII phosphorylation were fitted by the Hill equation:

$$f_{\text{Hill}}\left({\text{Ca}}^{2+}\right)=\frac{A_{\min }\cdot {\left[{\text{EC}}_{50}\right]}^{{\text{n}}_{\text{h}}}+A_{\max }\cdot {\left[{\text{Ca}}^{2+}\right]}^{{\text{n}}_{\text{h}}}}{{\left[{\text{EC}}_{50}\right]}^{{\text{n}}_{\text{h}}}+{\left[{\text{Ca}}^{2+}\right]}^{{\text{n}}_{\text{h}}}},$$

where A_max and A_min are the maximum and minimum of T286 phosphorylation, respectively, n_h is the Hill coefficient, and EC₅₀ is the half maximal effective concentration.

For the kinase inhibitor (K252a) experiment, 4 μL of 50 μM K252a in 25% v/v DMSO was added to 40 μl of the solution mixture 3 h after initiating the reaction (Fig. 1 B). A control experiment was performed using the same solution without K252a.

Results

Memory switch devices usually have two stable states (ON and OFF) and store information by switching between them. A push button is a typical memoryless switch because it responds just during stimulation (Fig. 1 A, left), whereas a toggle switch is a typical memory switch because it maintains a persistent response even after stimulation (Fig. 1 A, right). To examine whether CaMKII can actually serve as a memory switch, we reconstituted CaMKII phosphorylation in vitro based on an earlier study (13). Purified CaMKII together with CaM, PP1, ATP, and MgCl₂ was incubated in a Ca²⁺-EGTA buffer system at 4°C where free Ca²⁺ concentration could be controlled (Fig. 1, A and B). We examined CaMKII phosphorylation in the presence of a control peptide (residues 1095−1119 of GluN2B, Ctl) or the GluN2B-derived peptide that has been shown to interact with CaMKII (residues 1289−1310, N2Bs) (8). Consistent with the previous finding (13), in the presence of Ctl, the addition of a Ca²⁺ pulse transiently increased phosphorylation of CaMKII at T286, which subsequently decreased to a basal level (Fig. 1 B, left black line). By contrast, in the presence of N2Bs, phosphorylation of CaMKII at T286 increased and was persistently retained (Fig. 1 B, right black line). The persistent phosphorylation was blocked by the addition of K252a, a CaMKII inhibitor, indicating that the persistent phosphorylation requires the kinase activity of CaMKII and is not due to irreversibility or slow dephosphorylation. This persistent phosphorylation is consistent with the previous finding that N2Bs locks CaMKII in an active conformation in vitro (8,9). However, unlike the previous finding, the persistent phosphorylation required basal Ca²⁺ concentration (0.56 μM). If the basal Ca²⁺ concentration was set below 0.2 μM, the phosphorylation of CaMKII at T286 decreased to a basal level within 6 h even in the presence of N2Bs (Fig. S1 B, T286P at 0.13 μM Ca²⁺).

Hysteresis is an essential property of a memory switch. Therefore, we examined whether reconstituted CaMKII phosphorylation exhibits hysteresis, which can be seen as a distinctive splitting of Ca²⁺ dose-response curves for coming down from a phosphorylated ON state and for going up from a dephosphorylated OFF state (13,14). We measured the going up Ca²⁺ dose-response from a dephosphorylated OFF state to a phosphorylated ON state by increasing Ca²⁺, and the coming down Ca²⁺ dose-response from a phosphorylated ON state to a dephosphorylated OFF state by decreasing Ca²⁺ (Fig. 2 A). As previously described (13), the going up and coming down Ca²⁺ dose-response curves in the presence of Ctl exhibited similar switch-like responses with comparable EC₅₀ values (concentration of Ca²⁺ that gives 50% of the maximum response) (Fig. 2 A, left; Fig. S1 A). Although both Ca²⁺ dose-response curves exhibited similar switch-like responses in the presence of N2Bs, the coming down Ca²⁺ dose-response curve shifted to the left of the going up Ca²⁺ dose-response curve (Fig. 2 A, center; Fig. S1 B). This result indicates that CaMKII phosphorylation exhibits hysteresis. Six hours after the initiation of a reaction, CaMKII phosphorylation was maintained at a Ca²⁺ concentration between 0.2 and 0.75 μM in the coming down Ca²⁺ dose-response curve, which is lower than the EC₅₀ of the going up Ca²⁺ dose-response curve. Thus, CaMKII phosphorylation can remember the previous Ca²⁺ concentration (23). In addition to being a Ca²⁺ concentration-dependent process, hysteresis of CaMKII phosphorylation is also dependent on PP1 concentration. Increasing the concentration of PP1 caused the hysteresis to completely disappear, independent of the presence of N2Bs (Fig. 2 A, right; Fig. S1 C). Thus, hysteresis of CaMKII phosphorylation is a Ca²⁺- and PP1-dependent reversible process. We saw persistent hysteresis for up to 17 h, although the difference in Ca²⁺ sensitivities (EC₅₀) gradually decreased (Fig. S1).

CaMKII kinase activity was required for persistent phosphorylation (Fig. 1 B). This suggests that the persistent phosphorylation is due to a regenerative cycle of phosphorylation and dephosphorylation, rather than direct CaMKII activation by N2Bs binding. It has been reported that once CaMKII becomes activated by Ca²⁺/CaM in the absence of PP1, N2Bs keeps CaMKII persistently activated even after Ca²⁺ removal by the addition of EGTA (8). This irreversible activation of CaMKII by N2Bs binding is either a Ca²⁺- or PP1-independent process. Therefore, the hysteresis in this study is likely to be a different process from the previous observation in terms of Ca²⁺- and PP1-dependent reversibility (8). In addition, we did not observe the PP1 dose-dependent shift of the EC₅₀, unlike the findings of a previous study (13). This PP1 dose-dependency of the EC₅₀ may be due to the effect of imidazole in the PP1 solution, because imidazole has a large ionic strength and chelates Ca²⁺, which lowers the free Ca²⁺ concentration. Thus, imidazole may shift the apparent EC₅₀ (13). We removed imidazole with a desalting column and did not see the PP1-dependency of the EC₅₀.

We then analyzed whether persistent phosphorylation occurs in an intermolecular or intramolecular fashion. Phosphorylated ON and dephosphorylated OFF states of CaMKII were mixed in various ratios. The ratio between phosphorylated and nonphosphorylated CaMKII did not change for >6 h after mixing (Fig. 2 B), suggesting that cross-phosphorylation between holoenzymes did not occur and that the persistent phosphorylation occurred only by autophosphorylation within a holoenzyme. This result is consistent with the previous observation that cross-phosphorylation between holoenzymes does not occur ((24), but see (25)).

N2Bs contains the region of GluN2B with serine 1303 (S1303), a major CaMKII phosphorylation site, and S1303 phosphorylation has been reported to interfere with CaMKII binding to GluN2B (26–28). Indeed, the phosphomimetic (S1303D) and dephosphomimetic (S1303A) mutants of N2Bs have been shown to have lower and higher binding affinities to CaMKII, respectively (26–28). We asked whether these N2Bs mutants affected the hysteresis of CaMKII phosphorylation. Compared with WT N2Bs (see Fig. 3 B, dashed line; Fig. S2 B), the going up Ca²⁺ dose-response curve shifted left in the presence of N2Bs dephosphomimetic (S1303A) (Fig. 3 A, dashed line; Fig. S2 A) and shifted right in the presence of N2Bs phosphomimetic (S1303D) (Fig. 3 C, dashed line; Fig. S2 C). This suggests that dephosphorylated N2Bs is more effective than phosphorylated N2Bs at triggering CaMKII phosphorylation, consistent with the previous observation of lower and higher binding affinities of dephosphomimetic and phosphomimetic N2Bs to CaMKII, respectively (26–28). By contrast, the coming down Ca²⁺ dose-response curves appeared the same in the presence of any type of N2Bs (Fig. 3, solid lines; Fig. S2), indicating that the mutation of N2Bs only affects the going up process. Of importance, the hysteresis occurred regardless of the amino acid mutation at S1303 (Fig. 3; Fig. S2). As a result, CaMKII hysteresis had the narrower Ca²⁺ range in the presence of N2Bs dephosphomimetic (S1303A), whereas CaMKII hysteresis had the broader Ca²⁺ range of hysteresis in the presence of N2Bs phosphomimetic (S1303D). This heterogeneity in the Ca²⁺ range of hysteresis may contribute to a modifiable Ca²⁺ threshold for memory induction and persistent memory maintenance such as metaplasticity (29) in concert with other signaling mechanisms (30).

Discussion

Accumulating experimental evidence has clearly shown that the interaction between CaMKII and GluN2B is required in hippocampal LTP. The disruption of the interaction between CaMKII and GluN2B prevents synaptic translocation of CaMKII (9), persistent T286 phosphorylation (10), LTP induction (10,31), LTP maintenance (11,12), and extracellular signal-regulated kinase-dependent synaptic structural plasticity (32). The CaMKII and GluN2B interaction also controls spatial learning in behaving mice (10,33). Together with the function of the postsynaptic density to protect CaMKII from dephosphorylation (34), the CaMKII memory switch can provide a theoretical basis for the maintenance of modifiable and reversible LTP in the hippocampus (11,12). Maintenance of modifiable and reversible LTP cannot be attained by simple intermolecular irreversible binding reactions. Such irreversible binding reactions easily fall in a single stable steady state by noisy molecular interactions, and the coupled molecules do not function as a memory device. By contrast, the CaMKII memory switch accompanies Ca²⁺ and PP1 concentration-dependent hysteresis (Figs. 2 and 3). With this hysteresis, the dephosphorylated OFF state of CaMKII is robust against temporal noise of basal Ca²⁺ concentrations, and the dephosphorylated OFF state switches to a phosphorylated ON state only when the Ca²⁺ signal exceeds a modifiable threshold for LTP. Similarly, the phosphorylated ON state of CaMKII is robust against temporal noise of basal Ca²⁺ concentrations, and the phosphorylated ON state switches to a dephosphorylated OFF state only when PP1 concentration increases. Such phosphatase-dependent reversal of LTP (depotentiation) has been observed (35,36), and modeled (37).

A fluorescence resonance energy transfer experiment has shown that CaMKII is not persistently but transiently activated in spines by LTP inducing stimulation (38), suggesting that CaMKII does not serve as a memory switch. Similarly, a theoretical study has revealed that a CaMKII memory switch without GluN2B does not operate under physiological conditions (39). Our result indicates that the memory switch of CaMKII phosphorylation can be seen only in the presence of a GluN2B-derived peptide (N2Bs). This result supports the idea that GluN2B-associated CaMKII constitutes only a subpopulation of total spine population in LTP, and the signal of GluN2B-associated CaMKII activation is below the detectable level of multiphoton microscopy (3,10,38). Although this idea can account for the discrepancy among experiments (10,11,38), further studies in actual synapses are necessary to clarify the precise roles of CaMKII and GluN2B in LTP.

Hysteresis in Ca²⁺ sensitivity eventually disappeared under our experimental conditions (Fig. S1 and Fig. S2). This may raise the possibility that hysteresis is derived from slow convergence to a monostable state, rather than the bistable dynamics in the dynamical systems theory. However, considering that phosphorylation/dephosphorylation is a stochastic event, stochastic phosphorylation/dephosphorylation may account for the disappearance of hysteresis. Miller et al. (40) computationally analyzed the stochastic transition between the ON and OFF states of CaMKII phosphorylation, and they estimated that the ON/OFF state of T286 phosphorylation is stable over a few years. They obtained this long lifetime where the PP1 activity was saturated. In our experimental conditions, however, the PP1 activity was not saturated (2 μM CaMKII where K_m = 10 μM of PP1 for dephosphorylation). With unsaturated PP1, the lifetime of the ON/OFF state can be much shorter (40). This short lifetime may explain the gradual decrease of hysteresis (Fig. S1 and Fig. S2).

Hysteresis in Ca²⁺ sensitivity occurred at Ca²⁺ concentrations between 0.2 and 0.75 μM, in the presence of 2 μM CaM (Fig. 2 A). This Ca²⁺ range is above the basal Ca²⁺ concentration in neurons (0.05−0.1 μM) (39,41), therefore one may think that the CaMKII memory switch does not operate in neurons. However, when CaM concentration was increased to 10 μM, hysteresis appeared between 0.12 and 0.5 μM Ca²⁺ (data not shown), and the region of hysteresis is expected to cover 0.05−0.1 μM Ca²⁺ with the 100 μM CaM in neurons (42). Thus, CaMKII can operate as a memory switch under physiological conditions.

The CaMKII memory switch was seen only in the presence of a GluN2B-derived peptide (N2Bs). Although the CaMKII memory switch could be explained by the conventional bistable mechanism in the dynamical systems theory (6,7,40,43), such bistable dynamics do not require GluN2B. Therefore, the CaMKII memory switch may actually be based on other mechanisms. Although the mechanisms are unknown, the slow T286 dephosphorylation caused by GluN2B binding to CaMKII (44) and the decreased dissociation rate of Ca²⁺/CaM from CaMKII (8) may not be the direct causes of hysteresis. The CaMKII memory switch required not only the presence of a GluN2B-derived peptide (N2Bs), but also >0.2 μM Ca²⁺ for CaMKII activation (Fig. 1 B, right; Fig. 2 A). If the Ca²⁺ concentration was below 0.2 μM, the dephosphorylation rates of CaMKII in the presence and absence of N2Bs was similar (Fig. S1). This result suggests that hysteresis requires both N2Bs binding and a Ca²⁺ concentration >0.2 μM. T286 phosphorylation leads to >1000-fold decrease in the dissociation rate of Ca²⁺/CaM from CaMKII (45), whereas N2Bs leads to only a 20-fold decrease (8). Despite the much stronger effect of T286 phosphorylation relative to N2Bs, the hysteresis did not appear in the absence of N2Bs, even when T286 phosphorylation occurred (Fig. 2, A and B) (13). This suggests that hysteresis does not occur by simply decreasing the dissociation rate of Ca²⁺/CaM from CaMKII.

Memory switches using artificial gene regulatory networks (46) and engineered molecules (47) have been proposed; however, few examples of memory switches in natural signaling systems have been clearly described (48,49). To our knowledge, this is the first in vitro reconstitution and description of an erasable memory switch based on enzymatic reactions of a natural signaling system. The CaMKII memory switch occurred within a single holoenzyme, and no cross-phosphorylation between holoenzymes occurred. This indicates that the memory density of CaMKII can be extremely high (i.e., each CaMKII holoenzyme can remember different information in a solution). Thus, the CaMKII memory switch provides a design principle for memory devices that are more energy efficient, high-speed, and high-density memory devices than switches using artificial gene regulatory networks.