Temporal dynamics of learning-promoted synaptic diversity in CA1 pyramidal neurons

Although contextual learning requires plasticity at both excitatory and inhibitory (E/I) synapses in cornu ammonis 1 (CA1) neurons, the temporal dynamics across the neuronal population are poorly understood. Using an inhibitory avoidance task, we analyzed the dynamic changes in learning-induced E/I synaptic plasticity. The training strengthened GABAA receptor–mediated synapses within 1 min, peaked at 10 min, and lasted for over 60 min. The intracellular loop (Ser408−409) of GABAA receptor β3 subunit was also phosphorylated within 1 min of training. As the results of strengthening of α-amino-3-hydroxyl-5-methyl-4-isoxazole-propionate receptor–mediated synapses, CA1 pyramidal neurons exhibited broad diversity of E/I synaptic currents within 5 min. Moreover, presynaptic glutamate release probability at basal dendrites also increased within 5 min. To further quantify the diversified E/I synaptic currents, we calculated self-entropy (bit) for individual neurons. The neurons showed individual levels of the parameter, which rapidly increased within 1 min of training and maintained for over 60 min. These results suggest that learning-induced synaptic plasticity is critical immediately following encoding rather than during the retrieval phase of the learning. Understanding the temporal dynamics along with the quantification of synaptic diversity would be necessary to identify a failure point for learning-promoted plasticity in cognitive disorders.—Sakimoto, Y., Kida, H., Mitsushima, D. Temporal dynamics of learning-promoted synaptic diversity in CA1 pyramidal neurons.

The hippocampus is a primary area for contextual memory (1), known to process spatio-temporal information (2, 3) within a specific episode (4). Long-term strengthening of glutamatergic transmission has been identified as a mechanism of contextual learning in the dorsal cornu ammonis 1 (CA1) area of the hippocampus (5), and CA1specific immobilization or blockade of a-amino-3hydroxyl-5-methyl-4-isoxazole-propionate (AMPA) receptor delivery can impair the performance (6,7), indicating a causal relationship between learning and the receptor delivery into the synapses.
Genetic deficiency of GABA A receptor b 3 subunit severely impairs the contextual freezing response without affecting pain perception (22), and the phosphorylation in the cytoplasmic loop of b 3 subunit (Ser 4082409 ) is known to play an essential role for PKA, PKB, PKC, or Ca 2+ and calmodulin-dependent protein kinase II-dependent plasticity (23). Becausee the phosphorylation is known to increase surface levels of GABA A receptors containing b 3 subunits in cultured neurons (24)(25)(26)(27), we also examined the effect of learning as well as the temporal dynamics.
Pharmacological manipulation of the AMPA or GABA A receptors in the CA1 suggested different roles of the receptors after training (12,(28)(29)(30)(31)(32)(33). Microinjections of the AMPA receptor blocker [7-nitro-2,3-dioxo-1,4dihydroquinoxaline-6-carbonitrile (CNQX)] into CA1 impairs inhibitory avoidance (IA) task training immediately (0-5 min) but the effects are lost 30-60 min after training (29,30,32), whereas GABA A receptor blocker microinjection improves performance if performed immediately following training (28,(31)(32)(33). Although these studies suggest a critical period for plasticity immediately following training, the dynamic changes in learning-induced synaptic diversity are poorly understood. Here, we analyze the dynamic changes seen in learning-induced excitatory and inhibitory (E/I) synaptic function, pre-and postsynaptically. Learning rapidly strengthened both E/I synapses in various ways in individual CA1 neurons, producing a broad diversity of synaptic input across the CA1 neuronal population within 5 min after the training. Moreover, we quantified the diversity levels by calculating the self-entropy per single CA1 neuron.

Animals
Male Sprague-Dawley rats (postnatal 4 wk of age) were obtained from Chiyoda Kaihatsu (Tokyo, Japan). Prior to the experiment, the rats were individually housed in plastic cages for a couple of days (40 3 25 3 25 cm) at a constant temperature (23 6 1°C) under a 12-h light/dark cycle (lights on from 8 AM to 8 PM) with ad libitum access to water and food (MF; Oriental Yeast, Tokyo, Japan). All animal housing and surgical procedures were approved by the Institutional Animal Care and Use Committee of Yamaguchi University Graduate School of Medicine and comply with the Guide for the Care and Use of Laboratory Animals [National Institutes of Health (NIH), Bethesda, MD, USA].

IA task
Hippocampus-dependent IA training procedures were previously described in refs. 6 and 12. The IA training apparatus (length, 33 cm; width, 58 cm; height, 33 cm) was a 2-chambered box consisting of a lighted safe side and a dark shock side separated by a trap door (Fig. 1A). For training, rats were placed in the light side of the box facing a corner opposite the door. After the trap door was opened, the rats could enter the dark box at will. The latency before entering the novel dark box was measured as a behavioral parameter (latency before IA learning, Fig. 1B). Four seconds after the animals entered the dark side, we closed the door and applied a scrambled electrical foot-shock (2 s, 1.6 mA) via electrified steel rods in the floor of the box. The rats were kept in the dark compartment for 10 s before being returned to their home cage. The rats in the 0-min group were quickly euthanized with an overdose of pentobarbital within 1 min. Untrained control rats were not moved from their home cages and were injected with the same dose of anesthesia. The results of unpaired and walk-through controls were previously reported (12).
Thirty minutes after the procedure described above, the rats were placed in the light side. The latency before entering the dark box was measured as an indicator of learning performance (latency after IA learning).

Slice patch-clamp
Acute brain slices were prepared as previously described in refs. 12 and 13. Detailed protocol of slice patch-clamp technique for Figure 1. Diagram of experimental design and IA task. A) Rats were housed in a home cage but moved into the light box used for the task on the training day. A brief electrical foot-shock (2 s) was applied in the dark box in the shock cage. Brain slices were prepared at various time points of the training. B) Thirty minutes after the training, the rats consistently showed a longer latency before entering the dark side of the box. **P , 0.01 vs. training. Error bars indicate 6 SEM. The number of rats is shown at the bottom of each bar.
analyzing learning-induced synaptic plasticity was also published with a short demonstration movie (34).

Miniature recordings
For miniature recordings, we used a modified intracellular solution to adjust the reversal potential of the GABA A receptor response [127.5 mM cesium methanesulfonate, 7.5 mM CsCl, 10 mM HEPES, 2.5 mM MgCl 2 , 4 mM Na 2 ATP, 0.4 mM Na 3 GTP, 10 mM sodium phosphocreatine, 0.6 mM EGTA (pH 7.25)]. Moreover, we added 0.5 mM tetrodotoxin (Wako Pure Chemicals, Osaka, Japan) to perfusate to block action potentials. The voltage was clamped at 260 mV for mEPSC recording and at 0 mV for mIPSC recording (Figs. 2A and 3A). We analyzed the frequency and amplitude of mEPSCs and mIPSCs above 10 pA.
We obtained 4 miniature parameters (mean mEPSC amplitude, mean mIPSC amplitude, mean mEPSC frequency, and mean mIPSC frequency) in individual CA1 pyramidal neurons. For graphic expression, the distribution was visualized 2-dimensionally in the R software environment (R Foundation for Statistical Computing, Vienna, Austria) (amplitude in Fig. 2B; frequency in Fig. 3B). To calculate E/I balance, the value of mEPSC frequency or amplitude was divided by corresponding value of mIPSC frequency or amplitude in each neuron. After recording, we confirmed that mEPSCs and mIPSCs were completely abolished by 10 mM CNQX (MilliporeSigma, Burlington, MA, USA) and 10 mM bicuculline methiodide (MilliporeSigma), respectively.

Paired-pulse stimulation
To analyze presynaptic plasticity at excitatory synapses, we added 0.1 mM picrotoxin and 4 mM 2-chloroadenosine to the perfusate and performed paired-pulse stimulation at 260 mV. To analyze presynaptic plasticity at inhibitory synapses, we added 10 mM CNQX to the perfusate and performed paired-pulse stimulation at 0 mV. To evaluate the pairedpulse ratio from the EPSC or IPSC average, 50-100 sweeps were recorded with paired stimuli at 100-ms intervals. The ratio of the second amplitude to the first amplitude was calculated as the paired-pulse ratio.

Western blotting
Western blotting was performed according to a previous study (13). Rats were deeply anesthetized with pentobarbital at 0, 5, or 30 min after the training. The brain was removed and incubated for 3 min in ice-cold buffer containing 0.32 M sucrose and 20 mM Tris-HCl (pH 7.5). Dissected hippocampal CA1 tissues were homogenized in 200 ml of buffer containing 50 mM Tris-HCl (pH 7.4), 0.5% Triton X-100, 0.5 M NaCl, 10 mM EDTA, 4 mM EGTA, 1 mM Na 3 VO 4 , 50 mM NaF, 40 mM sodium pyrophosphate, 1 mM protease inhibitor, and 1 mM DTT. Insoluble material was removed by a 10-min centrifugation at 15,000 rpm.
Samples containing equivalent amounts of protein based on the bicinchoninic acid analysis (Thermo Fisher Scientific, Waltham, MA, USA) were heated at 100°C for 3 min in Laemmli sample buffer and subjected to SDS-PAGE for 30 min at 200 V. Proteins were transferred to an immobilon PVDF membrane for 1 h at 100 V. Membranes were blocked for 1 h at room temperature in 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 0.1% Tween 20, and 5% skim milk. Then, the membranes were incubated overnight at 4°C with anti-GABA A receptor b 3 subunit (1:1000; Abcam, Cambridge, MA, USA), anti-phosphorylated GABA A receptor b 3 subunit (Ser 4082409 ) (1:1000; Abgent, San Diego, CA, USA), or anti-b-tubulin (1:1000; BioLegend, San Diego, CA, USA). This step was followed by incubation with horseradish peroxidaseconjugated goat anti-rabbit IgG (1:5000; MilliporeSigma) for GABA A receptor b 3 subunit, phosphorylated GABA A receptor b 3 subunit, and b-tubulin. Bound antibodies were visualized using an ECL detection system (GE Healthcare, Chicago, IL, USA) and semiquantitatively analyzed using the ImageJ program (NIH).

Self-entropy analysis
As we previously reported, standard spreadsheet software (Excel 2010; Microsoft, Redmond, WA, USA) was used to calculate the self-entropy per neuron (36). Although we obtained 4 miniature parameters (mean mEPSC amplitude, mean mIPSC amplitude, mean mEPSC frequency, and mean mIPSC frequency) in individual CA1 pyramidal neurons, we determined the distribution of appearance probability of each miniature parameter using 1dimensional kernel density analysis. X 1 , X 2 , . . ., X n denotes a sample of size n from real observations. The kernel density estimate of P at the point x is given by where K is a smooth function called the gaussian kernel function and h . 0 is the smoothing bandwidth that controls the amount of smoothing. We chose Silverman's reference bandwidth or Silverman's rule of thumb (37,38). It is given by: where A = min (SD, interquartile range 1.34). By normalizing integral value in untrained controls, we found the distribution of appearance probability at any point. Then, we calculated the appearance probability at selected points. All data points for probability in untrained and trained rats were converted to self-entropy (bit) using the Shannon entropy concept defined from the information theory (39).
Based on the probability distribution, we calculated individual appearance probability of all recorded neurons. Then, the appearance probability of the neuron was converted to the self-entropy using Shannon's formula [=2LOG (appearance probability of the neuron, 2)]. For graphic expression, we visualized the self-entropy distribution by 2-dimensional kernel analysis in the R software environment (Figs. 2C and 3C).

Statistical analysis
We used the paired Student's t test to analyze the latency. The data of mEPSC, mIPSC, self-entropy, and protein levels were analyzed using 1-way factorial ANOVA in which the between-group factors were the individual time points. The Shapiro-Wilk test and F test were used for normality and equality of variance, respectively. Because the self-entropy data had large variations within a group, we performed log (1 + x) transformation prior to the analysis (40). A value of P , 0.05 was considered significant.

The performance of IA task
To investigate learning-induced synaptic modification in the hippocampus, we used the IA task (Fig. 1A). In this paradigm, rats were allowed to cross from a light box to a dark box, where an electric foot-shock (1.6 mA, 2 s) was delivered. Half an hour after the task, we measured the latency in the illuminated box as contextual learning performance. Figure 1B shows the latency in the training session and the retrieval test. The rats consistently showed longer latency in the retrieval test than in the training session ( Fig. 1B; t 11 = 5.746; P , 0.0001).

Miniature postsynaptic currents
To analyze the learning-dependent synaptic plasticity, we recorded mEPSC or mIPSC in the presence of 0.5 mM tetrodotoxin on the dorsal hippocampus ( Fig. 2A). By changing the membrane potential, we sequentially recorded mEPSCs (at 260 mV) and mIPSCs (at 0 mV) from the same neuron, as previously reported in refs. 12 and 13. We confirmed that the mEPSC and mIPSC events were clearly blocked by the bath treatment of an AMPA receptor blocker (CNQX) or GABA A receptor blocker (bicuculline). The postsynaptic currents are thought to correspond to the response elicited by a single vesicle of glutamate or GABA (41). In contrast, the number of synapses affects the frequency of events.

E/I balance
To calculate the balance ofE/I inputs, mean mEPSC amplitude was divided by mean mIPSC amplitude and mean mEPSC frequency was divided by the mean mIPSC frequency in each neuron. In the E/I balance of the amplitudes, the main effect of training [F (6, 318) = 1.570; P = 0.16] was not significant after the training (Fig. 2F). Conversely, in the E/I balance of miniature frequency, the main effect of training was significant after the training [ Fig. 3F, F (6, 318) = 2.371; P = 0.0296]. Post hoc analysis further showed a significant decrease from 0 to 5 min after the training, suggesting faster postsynaptic plasticity at the inhibitory synapses than that at excitatory synapses (Fig. 3F).

Self-entropy of mEPSC and mIPSC amplitude
Based on the information theory of Shannon (39), we calculated the appearance probability of the mean amplitudes of mEPSCs and mIPSCs. First, we found the distribution of appearance probability in untrained controls (Fig. 2B, left) and then we analyzed cell-specific appearance probability of all recorded neurons one by one (Fig. 2C, upper panels). Each probability of a single neuron was calculated as the self-entropy and plotted 2-dimensionally. For example, a point with a high appearance probability (around the mean level of mEPSC and mIPSC amplitude) indicated low self-entropy, whereas a point with very rare probability (a deviated point of mEPSC and mIPSC amplitude) indicated high self-entropy. Two-dimensional kernel analysis visualized the density (Fig. 2C, lower panels). IA training clearly diversified the amount of information per neuron and sustained. For the mEPSC amplitude, the results are statistically summarized in Fig. 2G. Self-entropy in the mEPSC amplitude exhibited a significant temporal change [F (6, 318) = 13.456; P , 0.0001]. Post hoc analysis further showed a significant effect from 5 to 60 min after the training compared with untrained control. Similarly, self-entropy in the mIPSC amplitude exhibited a significant temporal change [F (6, 318) = 15.057; P , 0.0001; Fig. 2H], and post hoc showed a significant effect from 5 to 60 min after the training. Combined self-entropy (bit) at both synapses also increased within 5 min after the training [F (6, 318) = 8.565, P , 0.0001; Fig. 2I].

Self-entropy of mEPSC and mIPSC frequency
For mEPSC and mIPSC frequency, we found the distribution of appearance probability in untrained controls (Fig. 3B, left), and then we analyzed the appearance probability of all recorded neurons one by one. We found cell-specific self-entropy in all recorded neurons, showing different self-entropy from each other (Fig. 3C, upper panels).
Two-dimensional kernel analysis visualized the density (Fig. 3C, lower panels). IA training diversified the amount of information per neuron and sustained. For the mEPSC frequency, the results are statistically summarized in Fig.  3G. Self-entropy in the mEPSC frequency exhibited a significant temporal change [F (6, 318) = 3.694; P = 0.0015]. Post hoc analysis further showed a significant effect from 5, 10, 20, and 60 min after the training compared with untrained control. In contrast, self-entropy in the mIPSC frequency increased quite rapidly [F (6, 318) = 20.938; P , 0.0001; Fig.  3H], showing a long horizontal kernel distribution at 0 and 5 min after the training (Fig. 3C, lower). Significant effect was observed at 0, 5, and 20 min after the training.

Presynaptic glutamate and GABA release
To analyze presynaptic plasticity, we examined the paired-pulse ratio after the training. At the excitatory synapses in the apical dendrite, the paired-pulse ratio for evoked EPSCs was significantly increased at 60 min after training, suggesting a delayed decrease in presynaptic glutamate release probability [Fig. 4A, B; F

Phosphorylation of GABA A receptor subunits
To analyze phosphorylation of the receptors using Western blot, we trimmed dorsal hippocampal CA1 tissue and extracted the whole-cell fractions (Fig. 5A).

Rapid plasticity at excitatory synapses
Although rapid plasticity of excitatory CA1 synapses is considered as an initial step of memory encoding rather than retrieval (42), conclusive evidence for the dynamic change of synaptic current is still lacking. Here we found a rapid increase in mEPSC amplitude within 5 min after IA training, showing that memory encoding rather than retrieval strengthens AMPA receptor-mediated excitatory synapses. Using fluctuation analysis of CA1 pyramidal neurons, we recently confirmed that the training increased postsynaptic number of AMPA receptor channels without changing cation current per single channel (36). As to the causal relationship between learning and the plasticity, we previously reported the bilateral gene expression of GluA1-containing AMPA receptor delivery blockers in the CA1 neurons impairs IA learning (6). Moreover, a chromophore-assisted light-inactivation technique demonstrated that optical inactivation of synaptic AMPA receptors can erase acquired memory (43). These results showed that newly delivered GluA1-containing AMPA receptors contribute to form contextual memory.
Paired-pulse analysis further revealed the presynaptic glutamate plasticity after the training. The decrease in the paired-pulse ratio at 5 min after the training suggests transient increase in presynaptic glutamate release within 5 min after the training. Because mEPSC frequency is used as an indicator for evoked release (44)(45)(46) or the number of functional synapses (47,48), both pre-and postplasticity may contribute to the increase in mEPSC frequency at 5 min after the training. Although the immobilization of postsynaptic AMPA receptors can also decrease the paired-pulse ratio (49), bilateral CA1 microinjections of the AMPA receptor blocker (CNQX) seem to impair the IA learning immediately (0-5 min) but not at 30-60 min after the training (29,30,32). Optogenetic approach is a powerful technique to investigate spine-specific presynaptic plasticity after contextual learning; glutamate-release probability at the synapses between CA3-engram and CA1-engram cells was significantly greater than that of other pair types of synapses (50). These findings together with the present results support the notion that contextual learning requires the presynaptic acute glutamate release soon after the training.

Rapid plasticity at inhibitory synapses
Conversely, the plasticity at inhibitory synapses seems to be task dependent and region specific (12,13,16). As to hippocampal-dependent contextual learning, IA training clearly increased the mIPSC amplitudes, suggesting a postsynaptic strengthening of GABA A receptor-mediated plasticity (12). Also, the mIPSC frequency was rapidly increased without increase in GABA release probability, suggesting rapid activation of inhibitory silent or  subthreshold synapses to increase the number of over-threshold synapses. It is possible that many mIPSC events are small and below the level of detection threshold (,10 pA) and increased postsynaptic responses may increase the amplitude of these small events above the level of detection (.10 pA), resulting in an apparent increase in mIPSC frequency. Although the mechanism at GABAergic synapses is still unclear, it was well-described regarding the postsynaptic function of GluA3 containing AMPA receptors in the hippocampal CA1 (51). Moreover, we further found a rapid increase in mIPSC amplitude immediately after the training, indicating that the memory encoding rather than the retrieval strengthens the GABA A receptor-mediated inhibitory synapses. This is the first report showing a rapid phosphorylation of the Ser 4082409 GABA A receptor b 3 subunit within 1 min after the training, the sites of which are necessary to attenuate clathrin-dependent endocytosis of the synaptic receptors increasing both amplitude and frequency of mIPSCs in cultured neurons (52).
A possible causal relationship between the GABAergic plasticity and learning has been previously reported. Not only the genetic deficiency of GABA A receptor b 3 subunit but also the prevention of GABA A receptor-mediated plasticity in CA1 impairs the contextual learning (12,22). Optogenetic manipulation of CA1 neurons further proved the timing-specific causal relationship between the GABAergic inputs and the learning; optic inactivation of dendrite-targeting CA1 interneurons during aversive stimuli was sufficient to prevent fear learning (17). In a preliminary study, we found that microinjections of an interference peptide in Ser 4082409 phosphorylation into the CA1 successfully blocked the training-induced strengthening of mIPSCs. Moreover, bilateral microinjections of the peptide resulted in a drastic decrease in IA task-learning performance, suggesting further causal relationship between the learning and the Ser 4082409 phosphorylation of the GABA A b 3 subunit.
Questions arise as to how the training can increase GABA A receptor-mediated currents so rapidly. Mobility of GABA A receptors may be closely associated with the issue, because the removal from the postsynaptic membrane or lateral diffusion decreases the synaptic GABAergic current (53-55). Recent single-particle tracking analysis further demonstrated quick diffusion of a single GABA A receptor (0.07 mm 2 /s) in cultured hippocampal neurons; it can move rapidly between the 2 different synapses within a few hundred milliseconds to a few seconds. Surprisingly, abundant GABA A receptors heterosynaptically locate at glutamatergic synapses, playing a key role in the stimulusdependent rapid changes in the postsynaptic number of receptors (56). Because single CA1 pyramidal neuron possesses around 30,000 excitatory and 1700 inhibitory synapses (57), learning may rapidly recruit the heterosynaptic GABA A receptors to strengthen the inhibitory synapses.
Because the phosphorylation of Ser 4082409 GABA A receptor b 3 subunit is known to prevent clathrin adaptor protein 2-mediated GABA A receptor internalization, the training-induced Ser 4082409 phosphorylation may help to stabilize the surface receptors (61)(62)(63). Although the training-induced Ser 4082409 phosphorylation is rapid and transient, gephyrin may contribute to sustaining large mIPSC amplitude. Finally, using fluctuation analysis of CA1 pyramidal neurons, we recently confirmed that the training increased the postsynaptic number of GABA A receptor channels without changing Cl 2 current per single channel (36).

Overall significance of temporal dynamics
An early-phase long-term potentiation (,1 h) is thought to arise from rapid changes in the functional status of preexisting synapses, which may include conversion of synapses from silent state to an active one (47) and increase in the release probability of presynaptic vesicles (45,46). Because the training did not increase the release probability except some time points ( Fig. 4; glutamate at 5 min and GABA at 30 min), it may increase the number of functional synapses drastically. Considering the peak levels of mEPSC and mIPSC frequency, the number of over-threshold synapses (.10 pA) may increase up to 2-4 times greater than the pretraining levels and then decreased gradually (excitatory synapses, Fig. 3D; inhibitory synapse, Fig. 3E).
Temporal dynamics of mEPSCs and mIPSCs further revealed a time lag of synaptic plasticity. IA training rapidly increased both mIPSC amplitude and frequency within 1 min, whereas the training enhanced both mEPSC amplitude and frequency within 5 min. Moreover, the time lag transiently reduced E/I balance of the mEPSC and mIPSC frequency but returned to the pretraining level within 10 min (Fig. 3F). The quicker plasticity at inhibitory synapses may contribute to reduce seizure vulnerability on the remodeling of hippocampal networks (64).
The E/I imbalance may be caused by training-induced spike bursts of CA1 neurons, because optogeneticallyinduced spike burst reduces E/I balance in CA1 pyramidal neurons (65). Then, concomitant calcium influx may induce the specific phosphorylation of GluA1 and GABA A receptor b 3 subunits. Abundant extrasynaptic membrane GABA A receptors of CA1 pyramidal neurons (66) and the rapid motility (53) may enable rapid activation of GABA A receptor-mediated synapses. In contrast, it is known to take at least 5 min for long-term potentiation induction, because GluA1-containing AMPA receptors are inserted at the plasma membrane and move laterally to the excitatory synapses (67,68).

Quantification of diversity
Shannon's information theory was applied to the recent studies on learning and memory (69,70). Although the induction of long-term potentiation expands synaptic information storage capacity in hippocampal dentate gyrus neurons in vivo (70), it is still unknown whether the learning affects the information content of CA1 neurons. Here, we calculated the self-entropy of each CA1 neuron to quantify the learning-induced synaptic diversity. Each CA1 neuron had a different self-entropy that was clearly increased after the training (Figs. 2C and 3C). Because bilateral blockade of the synaptic diversity clearly impaired the learning performance (12,71), we hypothesized that the increased self-entropy may code a piece of experienced information after training. Considering the total number of pyramidal neurons in rat dorsal CA1 (72), the self-entropy in untrained condition (5.2 3 10 6 bits) would increase up to 42.1 3 10 6 bits at 10 min after the training. In any case, the analysis may be a useful approach to quantify the learning-induced synaptic diversity at the entropy level.

Temporal dynamics and cognitive disorders
Synaptic dysfunction is well-correlated with cognitive decline in Alzheimer's disease (73). Amyloid b peptide 1-42 (Ab 42 ) is well-known as a major causative agent (74)(75)(76)(77), and long-term exposure to Ab 42 (1-3 d) impairs the AMPA receptor trafficking by reducing synaptic distribution of Ca 2+ and calmodulin-dependent protein kinase II in cultured pyramidal neurons (78). In contrast, the effect of soluble oligomeric assemblies of Ab 42 is more rapid, decreasing surface level of AMPA receptors within 30 min (79). Although less is known about the toxic effect at inhibitory synapses, Ab 42 specifically binds to nicotinic a 7 receptors (80), impairing the learning-induced plasticity at the GABA A receptor-mediated inhibitory synapses (12,81). Bath application of Ab 42 weakens GABA A receptormediated synaptic currents within 10 min (82), whereas it directly blocks the nicotinic a 7 receptor-mediated cholinergic response within 3 min (83). Understanding the dynamic changes occurring during learning-promoted plasticity would be necessary to identify a failure point in cognitive disorders.