A possible coding for experience: ripple-like events and synaptic diversity

The hippocampal CA1 is necessary to maintain experienced episodic memory in many species, including humans. To monitor the temporal dynamics of processing, we recorded multiple-unit firings of CA1 neurons in male rats experiencing one of four episodes for 10 min: restraint stress, social interaction with a female or male, or observation of a novel object. Before an experience, the neurons mostly exhibited sporadic firings with some synchronized (≈ 50 ms) ripple-like firing events in habituated home cage. After experience onset, restraint or social interaction with other rats induced spontaneous high-frequency firings (super bursts) intermittently, while object observation induced the events inconsistently. Minutes after experience initiation, CA1 neurons frequently exhibited ripple-like firings with less-firing silent periods. The number of ripple-like events depended on the episode experienced and correlated with the total duration of super bursts. Experience clearly diversified multiple features of individual ripple-like events in an episode-specific manner, sustained for more than 30 min in the home cage. Ex vivo patch clamp analysis further revealed experience-promoted synaptic plasticity. Compared with unexposed controls, animals experiencing the female, male, or restraint episodes showed cell-dependently increased AMPA- or GABAA receptor– mediated postsynaptic currents, whereas contact with a novel object increased only GABAergic currents. Multivariate ANOVA in multi-dimensional virtual space revealed experience-specific super bursts with subsequent ripple-like events and synaptic plasticity, leading us to hypothesize that these factors are responsible for creating experience-specific memory. It is possible to decipher encrypted experience through the deep learning of the orchestrated ripple-like firings and synaptic plasticity in multiple CA1 neurons.


Introduction
The hippocampus is a primary site for episode-like memory development (Scoville & Milner, 1957), known to process spatio-temporal information (Mitsushima et al, 2009;Wills et al, 2010) within a specific episode (Gelbard-Sagiv et al, 2008). The dorsal CA1 neurons may encode place and context when animals are exploring novel context (Tanaka et al, 2018), and temporal inactivation of firings before exploration of novel objects can impair test performance in a what-where-when episodic-like memory task (Drieskens et al, 2017). These observations indicate the importance of firing activity during or immediately after episodic experience, but specific firing patterns during the early learning period are not clear.
Emotions such as happiness, fear, and sadness influence the strength of a memory (Christianson et al, 1992;McGaugh et al, 2000;Richter-Levin & Akirav, 2003;LeDoux, 2000). Emotional arousal enhances learning via noradrenergic stimulation of the dorsal CA1 neurons to drive GluA1-containing AMPA receptors into the synapses (Hu et al, 2007). Tyrosine hydroxylase-expressing neurons in the locus coeruleus may mediate post-encoding memory enhancement with co-release of dopamine in the hippocampus (Takeuchi et al, 2016). However, conclusive evidence regarding whether emotional arousal affects CA1 neuron firing and the associated temporal dynamics is lacking in freely behaving animals.
The first aim of this study was to examine the temporal dynamics of CA1 neural activity in the early learning period in rats. The second aim was to examine differences in these firing patterns among different experiences. For this purpose, we used four emotionally distinct episodes and compared the associated early-learning processes.
Temporo-spatial firing patterns by multiple neurons may orchestrate a possible code (Grinvald et al, 2003;van Hemmen & Sejnowski, 2006), so we also monitored changes in multiple-unit firings after various episodic experiences. Finally, by analyzing postsynaptic currents induced by a single vesicle of glutamate or GABA ex vivo , we examined experience-specific plastic changes at excitatory/inhibitory synapses onto the CA1 pyramidal neurons.

Animals
Sprague-Dawley male rats (CLEA Japan Inc., Tokyo, Japan) were housed at 22°C with a 12-h light-dark cycle (lights on from 8:00 A.M to 8:00 P.M.). Rats were allowed at least 2 weeks of ad libitum food (MF, Oriental Yeast Co. Ltd., Tokyo, Japan) before surgery. Rats at age 15 to 25 weeks were used, and all experiments were conducted during the light cycle. These studies were reviewed and approved by the Yamaguchi University Graduate School of Medicine Committee of Ethics on Animal Experiments.
All manipulations and protocols were performed according to the Guidelines for Animal Experiments at Yamaguchi University Graduate School of Medicine and in accordance with Japanese Federal Law (no. 105), Notification (no. 6) of the Japanese Government, and the National Institutes of Health Guide for the Care and Use of Laboratory Animals , revised in 1996.

Surgery
Animals were anesthetized with sodium pentobarbital (50 mg/kg, intraperitoneal) and placed in a stereotaxic apparatus. Movable recording electrodes (Unique Medical Co., LTD, Japan) were chronically implanted above the hippocampal CA1 (posterior, 3.0-3.6 mm; lateral, 1.4-2.6 mm; ventral, 2.0-2.2 mm) and fixed with dental cement. Rats were housed individually and excluded for analysis if cannulas or electrodes did not target the region.

In vivo recording of multiple-unit firing activity
Neural signals were passed through a head amplifier and then into the main amplifier (MEG-2100 or MEG-6116; Nihon Kohden, Tokyo, Japan) through a shielded cable.
Signals were band-pass filtered at 150-5000 or 150-10000 Hz and digitized using a CED 1401 interface (Cambridge Electronics Design, Cambridge, UK) controlled by Spike2 software (Cambridge Electronics Design, Cambridge, UK). Signal data were mostly sampled at 25 kHz, but a few were sampled at 17 kHz.
Isolation of single units was initially performed using the template-matching function in Spike2 software. As we reported previously (Ishikawa et al, 2015), all spikes used in subsequent analysis were clearly identified, with a signal-to-noise ratio of at least 3 to 1. Following the initial separation of spikes, we applied principal component analysis of the detected waveforms. Single units had to show cluster separation after their first three principal components were plotted (Fig. 1E). However, in this experiment, the sorting was not always reliable, especially in super bursts and ripplelike events, because one electrode recorded many units. Therefore, we analyzed all recording data as multiple unit activity.
We identified a spontaneous event of super burst (Figs. 1G and 2) as one showing a higher firing rate for each point greater than 3 standard deviations (SDs) of the mean firing rate before the episodic experience. Figure 2F shows profile plots of individual super bursts over the threshold. A silent period (Figs. 1H and 3A) was one that showed a longer duration of inter-spike interval than 3 SD above the mean of the baseline. A ripple-like event was defined as one that involved long-lasting, high-frequency firing (> 10 ms) with a signal-to-noise ratio of at least 6 to 1 (Figs. 1I and 3A). Behavioral states during the ripple-like events were immobile eye-closed 16.0 ± 7.0 %, immobile eyeopened 59.7 ± 11.0%, or eye-opened moving 24.3 ± 11.3% (1631 events in 9 rats).
For analysis of duration, arc length, and amplitude, the initiation time of a ripplelike event was defined as the time point at which the root mean square (RMS) reached +3 SD of baseline, and the end time of a ripple-like event was defined as the time that RMS came down to +3 SD of baseline (Fig. 4A). The calculated duration appeared shorter than the visual feature. For detection of ripple-like events, the signal was filtered at 150-300 Hz (band-pass), and RMS was calculated. The threshold for the event detection was set to +6 SD above the mean of the baseline.
For amplitude analysis (Fig. 4B), we calculated the difference between the lowest and highest peaks. Arc length (L: Fig. 4F) was calculated using the following equation: where Fs is the sampling rate for signal acquisition and p(t) indicates voltage value at time point t. The total number of sampling points during the duration of a ripple-like event is n (Fig. 4D). For peak analysis (Fig. 4H), we counted the number of negative peaks during the ripple-like firing events. For graphic expression (Figs. 5A and B), three of four parameters were multi-dimensionally plotted using MATLAB ® (The MathWorks, Inc. MA).

Experimental schedule
Rats were handled for at least one week before surgery and allowed to recover from surgery for at least 2 weeks before the recording experiments. On the day of the experiment, a recording electrode was inserted into the pyramidal cell layer of CA1 neurons. We started the recording of multiple-unit CA1 firings, and spontaneous behavior was simultaneously monitored while the animals were in their well-habituated home cages. Details of the recording schedule are described below. To confirm the formation of memory for each experience, behavioral responses were monitored again on the day following the first experience (Fig. 1B).

Neural recording schedule
At least 15 min after the recording of basal condition, the rats were exposed to either the restraint stress or a first encounter with a female, male, or object for 10 min (Fig. 1A).
For restraint, rats were taken from the home cages, their legs were strapped using soft cotton ties, and the animals were fixed onto the wood board for 10 min (Mitsushima et al, 2006(Mitsushima et al, , 2008. The animals then were returned to their home cages, and the multipleunit firings were sequentially recorded for more than 30 min. For the other experiences, a sexually mature female (postnatal age, 8-12 weeks), young male (postnatal age, 6-7 weeks), or novel object [yellow LEGO ® /DUPLO ® brick, 15 cm (h) × 8 cm (w) × 3 cm (d)] were placed in the home cage for 10 min. The bricks have a weak plastic smell and were fixed onto the side wall using double-sided tape. After the removal of the intruder or object, multiple-unit firings were sequentially recorded for more than 30 min.

Histology
At the end of experiments, animals were deeply anesthetized with sodium pentobarbital (400 mg/kg, i.p.) and immediately perfused transcardially with a solution of 0.1 M phosphate buffer containing 4% paraformaldehyde. The brain was removed and then post-fixed with the same paraformaldehyde solution and immersed in 10%-30% sucrose solution. Coronal sections (40 μm thick) were stained with hematoxylin and eosin. The locations of cannulas, recoding electrode tips, and tracks in the brain were identified with the aid of a stereotaxic atlas (Paxinos & Watson, 2013).

Slice-patch clamp analysis
Forty minutes after onset of the episodic experience, we anesthetized the rats with pentobarbital and prepared acute brain slices (Mitsushima et al, 2011(Mitsushima et al, , 2013. The inexperienced control rats were injected with the same dose of anesthesia in their home cage. Whole-cell recordings were performed as described previously (Kida et al, 2016).
Whole-cell recordings were obtained from CA1 pyramidal neurons from the rat hippocampus using an Axopatch 700A amplifier (Axon Instruments).

Self-entropy analysis
We calculated the self-entropy per neuron as reported previously .
First, we determined the distribution of appearance probability of mean mEPSC and mIPSC amplitudes separately using one-dimensional kernel density analysis.
Geometric/topographic features of the appearance probability were calculated using kernel density analysis, as follows: Let X1, X2, . . . , Xn denote 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 (Sheather, 2004, Silverman, 1986 integral values in inexperienced 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 inexperienced and experienced rats were converted to self-entropy (bit) using the Shannon entropy concept, defined from the Information Theory (Shannon, 1948).

Statistical analysis
Data and statistical analyses were performed using SPSS or StatView software. To analyze behavioral parameters, we used paired t-tests. To compare temporal dynamics in the duration of the silent period or number of ripple-like events, we used the Friedman test followed by the post hoc Wilcoxon signed-rank test. Differences in the ripple-like events or super bursts or silent period or features among experiences were compared using the Kruskal-Wallis test followed by the post hoc Mann-Whitney U test. We applied Pearson's correlation coefficient to examine correlation among super bursts, ripple-like events, and silent period.
The number and total duration of super bursts or features of ripple-like events, such as amplitude, duration, arc length, and peak, were analyzed using two-way analysis of variance (ANOVA) with the between-group factors of time and experience, followed by post hoc ANOVAs and Fisher's post hoc least significance difference test. Each miniature parameter and each self-entropy value were analyzed using one-way factorial ANOVA in which the experience was the between-group factor.
Overall differences in two features of super bursts, four features of individual ripple-like events, or four miniature parameters (mEPSC and mIPSC amplitudes and frequencies) were analyzed using multivariate ANOVA (MANOVA) with Wilks' Lambda distribution. Specific differences between two experiences were further analyzed using post hoc MANOVAs. The Shapiro-Wilk test and F-test were used to determine normality and equality of variance, respectively. Because the data had large variations within a group, we performed log (1+x) transformation prior to the analysis (Mitsushima et al, 1994). P < 0.05 was considered statistically significant.

Evaluation of experienced memory
The male rats being recorded vigorously resisted and vocalized frequently during the 10 min of restraint (Mitsushima et al, 2006). The males checked, chased, and sometimes mounted a female intruder, but no intromission or ejaculation was observed (Mitsushima et al, 2009). With a male intruder, the males attacked him minutes after the check and hesitation. With the novel object, the animals checked it and sometimes smelled and baited it.
To evaluate the acquired memory, we compared behavioral parameters during the first experience with those during a second experience (Fig. 1B). The rats that experienced the 10-min restraint stress showed a smaller number of vocalizations in the second exposure (t6 = 3.476, P = 0.0129). Similarly, those that encountered a female, male, or novel object consistently in the second encounter showed a shortened latency to vaginal inspection (t8 = 3.492, P = 0.0082) and attack (t7 = 4.192, P = 0.0041) and a shorter observation time (t9 = 2.901, P = 0.0176), suggesting acquisition of experienced memory.

Multiple-unit firings
To record multiple-unit firings of CA1 neurons, we implanted a depth-adjustable tungsten electrode with a resistance of 50 to 80 kΩ (Fig. 1D). We extracted super bursts ( Fig. 1G), silent periods ( Fig. 1H), or ripple-like events (Fig. 1I) based on basal firing rate in their habituated home cage. Per the criteria, spontaneous super burst events showed a firing rate and silent periods showed a duration of inter-spike interval greater than 3 SD of basal firings. Ripple-like events were short-term high-frequency firings (≈ 50 ms) with more than a 1:6 signal-to-noise ratio. Although spike sorting was not successful in super bursts and ripple-like events, coordinated firings of multiple single neurons formed a single event of super burst or ripple-like firings (Fig. 1E). Single events of a silent period lasted 0.89 ± 0.04 s (N = 278).

Super bursts
Before the test experience, the rats in their habituated home cage mainly exhibited sporadic spontaneous firings. After the onset of an experience, they frequently showed super burst events ( Fig. 2A Both events and their duration were two-dimensionally compared using repeated measures of MANOVA (Fig. 2E), with experience as the between-group factor and time as the within-group factor. The main effect of experience (F6, 970 = 11.408, P < 0.0001) and the post hoc analyses indicated specific differences between two experiences, suggesting episode-specific temporal dynamics of the super bursts (Table 5).
The duration and relative firing rate of individual super bursts were further analyzed and plotted two-dimensionally (triangles in Fig. 2F). X-axis represents duration (s), and Y-axis represents times over the threshold (+3 SD of the mean firing rate before the experience). The two features were compared using MANOVA, with experience as the between-group factor. The results indicated a significant main effect of experience (F6, 554 = 29.201, P < 0.0001). Post hoc MANOVA showed specific differences between two experiences, suggesting episode-specific features of individual super bursts (Table 6).

Silent periods
We found a within-group temporal change in the duration of silent periods with restraint stress ( To examine episode specificity, we compared both incidence and duration twodimensionally using MANOVA, with experience as the between-group factor and time as the within-group factor. MANOVA results showed no significant main effect of experience (F6, 210 = 1.566, P = 0.158), suggesting that the temporal dynamics of silent periods may not be episode-specific (Table 5).

Ripple-like events
We found a significant temporal change in the incidence of ripple-like events within the group for restraint stress (c 2 2 = 9.805, P = 0.007) and female contact ( after the onset of experience. Figure 3G (Table 7) shows detailed differences among experiences.

Correlation of super bursts with ripple-like events or silent periods
Both the incidence and duration of super bursts were significantly correlated with the relative number of ripple-like events (Figs. 3H and 3I) but not with silent periods (Figs. 3D and 3E). The results suggest a role for super bursts in increasing ripple-like events.
Detailed correlations of multiple features with ripple-like events are shown in Figure   5C.

Features of individual ripple-like events
We extracted four features of individual ripple-like events (Fig. 4A) and analyzed them using two-way ANOVA, with experience as the between-group factor and time as the within-group factor (Table 8). Further post hoc ANOVAs and the temporal dynamics are shown for amplitude in Figure 4B (Table 9), for duration in Figure 4D (Table 10), for arc length in Figure 4F (Table 11), and for peaks in Figure 4H (Table 12). Values for all of these features significantly increased with restraint stress and female contact. In contrast, only a partial and different effect was seen with contact with a male or object.
Duration and arc length increased with contact with a male, but amplitude and peak number did not. The number of peaks decreased with contact with a novel object, but other features did not change. Differences between two specific groups are shown for amplitude in Figure 4C (Table 13), for duration in Figure 4E (Table 14), for arc length in Figure 4G (Table 15), and for peaks in Figure 4I ( Table 16).
To examine further the diversified features of ripple-like events, we compared the

Synaptic plasticity
To further examine experience-induced synaptic plasticity, we prepared ex vivo brain slices 30 min after the experiences (Fig. 6A). With sequential recording of mEPSCs (at -60 mV) and mIPSCs (at 0 mV) from the same neuron (Mitsushima et al, 2013), we measured four parameters from individual CA1 neurons: amplitudes and frequencies for both mEPSCs and mIPSCs. Figure 6B shows cell-specific plots of the means of AMPA receptor-mediated excitatory currents vs. GABAA receptor-mediated inhibitory currents. We used kernel analysis to visualize two-dimensionally the distribution of appearance probability in lower panels. Although inexperienced rats exhibited a low and narrow distribution range, experience diversified the input strength. The results of one-way ANOVA in individual parameters are shown in Figure 6C ( Table 18).
To compare experience-induced plasticity, we plotted the four parameters in a four-dimensional virtual space to analyze both amplitude and frequency of mEPSC and mIPSC events in individual CA1 neurons. We used one-way MANOVA with experience as the between-group factor, which showed a significant main effect of experience ( Fig.   6C; F16, 551 = 4.729, P < 0.001). Post hoc MANOVA further showed multiple differences between two specific experiences (Table 19), suggesting experience-specific synaptic plasticity.
Based on Shannon's information theory (1948), we calculated the appearance probability of the cell-specific synaptic strength . Using the appearance probability in inexperienced controls, we analyzed the appearance probability of recorded neurons one by one. Figure 6D shows cell-specific self-entropy and the visualized density distribution. The results of one-way ANOVA in individual self-entropy parameters are shown in Figure 6E (Table 20).
To examine experience specificity, we further analyzed the four self-entropy parameters in four-dimensional virtual space. We used one-way MANOVA with experience again as the between-group factor, and the results showed a significant main effect of experience ( Fig. 6E; F16, 551 = 4.361, P < 0.001). Post hoc MANOVA further showed multiple differences between two specific experiences (Table 21), suggesting experience-specific information content in CA1 pyramidal neurons.

Discussion
Here we recorded multiple-unit firings in CA1 from freely moving male rats in a habituated home cage (Fig. 1C, Movie S1) and subjected the rats to the one of four episodic experiences for 10 min: restraint stress, social interaction with a female or male, or observation of a novel object (Fig. 1B). Based on the basal firings in the habituated home cage (Fig. 1F), we extracted three firing events of multiple-unit recording: super bursts, silent periods, and ripple-like events (Figs. 1G-I). After the onset of episodic experience, we found episode-specific intermittent generation of super bursts (Fig. 2) and frequent induction of ripple-like events with silent periods (Movie S2; Fig. 3). The four features (amplitude, duration, arc length, peaks) of these thousands of ripple-like firing events were also experience-specific (Figs. 4 and 5). Finally, ex vivo patch-clamp analysis showed experience-specific plasticity at excitatory/inhibitory synapses (Fig. 6).

Synaptic plasticity
Long-term potentiation (LTP) has been considered as a synaptic model of learning and memory (Bliss & Lømo, 1972). The LTP not only enhances the presynaptic release of glutamate (Dolphin et al, 1982) but also increases the number of GluA1-containing AMPA receptors in CA1 pyramidal neurons (Shi et al, 1999, Hayashi et al, 2000. Moreover, the high-frequency stimulation at the Schaffer collaterals that induces LTP of evoked field response enhances the number of sharp-wave ripples (Buzsáki 1984(Buzsáki , 2015. Although high-frequency electrical stimulation and concomitant pre-and postsynapse excitation can induce synaptic plasticity (Harris & Teyler, 1976;Nicoll et al, 1988), intrinsic high-frequency stimuli have not been established in animals during learning. Here we extracted spontaneous high-frequency firing events of CA1 neurons (super bursts) during and soon after the episodic stimuli.
Individual super bursts showed different firing rates and durations (Fig. 2F), and the multiple features of the super bursts were episode-specific (Table 5 and S6). The episode-specific super bursts may trigger diversification of the ripple-like events because high-frequency stimulation at CA3-CA1 synapses increases both incidence and amplitude of sharp-wave ripples in CA1 (Behrens et al, 2005). In the present study, both incident and duration of super bursts were positively correlated with the multiple features of ripple-like events (Fig. 5C), and the bilateral inactivation of basolateral amygdala by a muscimol/baclofen mixture before the restraint not only attenuated the super bursts but also blocked the diversification of ripple-like events (unpublished preliminary data). These results suggest a causal relationship between the super bursts and the ripple-like events.

Restraint stress
Because physiological stress rapidly occludes LTP induction (

Contact with a female
In contrast to restraint stress, contact with female may be a "positive" experience and has been used as a reward for conditioned memory (Ramirez et al, 2015;Coria-Avila, 2012). Some neurons in the dentate gyrus and basolateral nucleus of male mouse amygdala respond to a female mouse (Redondo et al, 2014;Ramirez et al, 2015).
Moreover, contact with a female activates specific engram cells in the hippocampal dentate gyrus, and re-activation of the neurons reduces stress-induced depression-related behavior in male mice (Ramirez et al, 2015). Contact with a female in the present study may have been surprising for recorded males, which had never met a breedable female and exhibited long-lasting super bursts (Fig. 2E). Moreover, the diversified ripple-like events were comparable to those after the restraint (Figs. 5A and B). Although it is unclear whether this situation is applicable to humans, humans do tend to develop episodic recall of the time and location of an initial heterosexual encounter of interest (Turgenev, 1860;Carpenter & Carpenter, 1975).

Contact with a male rat
The contact with a male was not quite as novel given that the recorded males had been reared in same sex/age group. In experiences like these, the male normally checks superiority with the unknown intruder (Whishaw & Kolb, 2005). Heterosexual contact prevents the development of conditioned same-sex partner preference in male rats, suggesting a difference between the two social memories (Ramirez-Rodriguez et al, 2017 (Sakurai et al, 2013). Here, we found episode-specific changes in the ripple-like firings ( Fig. 5A and B). Analyzing changes in these firings in hippocampal CA1 may be a first step toward being able to identify which animal is experiencing one of the four episodes.

Synaptic diversity in individual CA1 neurons
The question arises as to why the firing pattern changed after the experience and following episode-specific super bursts. Here we hypothesized that experience-induced plasticity at CA1 synapses may cause episode-specific changes in ripple-like firings. To evaluate plasticity at CA1 synapses, we analyzed ex vivo mEPSCs and mIPSCs in individual CA1 pyramidal neurons from experienced rats (Fig. 6A). Each neuron exhibited different mEPSCs and mIPSCs and frequencies at excitatory and inhibitory synapses, showing a synaptic diversity (Fig. 6B). Moreover, the distribution of the diversity differed among experiences. Although the figures are two-dimensional, experience created a specific distribution of four-dimensional plots suggesting an experience-specific synaptic diversity (Table 19). The mEPSCs and mIPSCs are thought to correspond to the response elicited by a single vesicle of glutamate or GABA (Pinheiro & Mulle, 2008), while the number of synapses affects the frequency of events.
It is possible that the neuron-dependent synaptic current or number of individual neurons may contribute to creating specific firings of CA1 neurons (Fig. 7).