Associative Learning-Induced Synaptic Potentiation at the Two Major Hippocampal CA1 Inputs for Cued Memory Acquisition

Abstract

Learning-associated functional plasticity at hippocampal synapses remains largely unexplored. Here, in a single session of reward-based trace conditioning, we examine learning-induced synaptic plasticity in the dorsal CA1 hippocampus (dCA1). Local field-potential recording combined with selective optogenetic inhibition first revealed an increase of dCA1 synaptic responses to the conditioned stimulus (CS) induced during conditioning at both Schaffer collaterals to the stratum radiatum (Rad) and temporoammonic input to the lacunosum moleculare (LMol). At these dCA1 inputs, synaptic potentiation of CS-responding excitatory synapses was further demonstrated by locally blocking NMDA receptors during conditioning and whole-cell recording sensory-evoked synaptic responses in dCA1 neurons from naive animals. An overall similar time course of the induction of synaptic potentiation was found in the Rad and LMol by multiple-site recording; this emerged later and saturated earlier than conditioned behavioral responses. Our experiments demonstrate a cued memory-associated dCA1 synaptic plasticity induced at both Schaffer collaterals and temporoammonic pathways.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12264-024-01327-7.

Introduction

Plasticity at hippocampal synapses is believed to be essential for the acquisition and initial storage of hippocampus-dependent memory, i.e., declarative memory [1, 2]. To understand hippocampal synaptic mechanisms of memory formation, a number of studies have shown experience-dependent molecular or structural changes at hippocampal synapses [3–8], including the learning-induced synaptic trafficking of AMPA (amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptors [9–12] and the modulation of synaptic plasticity-related proteins [11, 13–15], as well as an increase of dendritic spine number [16–20].

On the other hand, experience-induced functional changes in synaptic strength, i.e., long-term potentiation (LTP) and long-term depression (LTD) widely thought to be cellular models for learning and memory [21–23], have also been demonstrated in the hippocampus with electrophysiological recording made in animals exposed to task learning, but in very few reports (as reviewed in refs [1, 2, 8]). As reported by local field-potential (LFP) recording with multielectrode arrays for monitoring field excitatory postsynaptic potentials (fEPSPs) elicited by electrical stimulation of Schaffer collateral axons at various laminar locations in the hippocampal CA1 regions, a spatially restricted LTP induction is detectable in behaving rats over CA1 apical dendritic layers after a single trial of inhibitory avoidance training [9]. In a contemporaneous report of single-electrode recording for measuring CA1 fEPSP responses also evoked by Schaffer collateral stimulation, but in animals receiving trace conditioning of eyelid responses (another classical form of hippocampus-dependent associative learning [1, 2, 24]), an LTP-like potentiation has been found at hippocampal CA3-CA1 synaptic connections when animals receive several sessions of conditioning per day, and an LTD-like depression has been found over the session of memory extinction [25]. With whole-cell recording made from anesthetized and awake rodents, our recent studies reported an LTP-like synapse unsilencing that is induced by trace fear conditioning in a large population of CA1 pyramidal cells, which results in widespread excitatory synaptic or spike responsiveness of CA1 neurons to the conditioned stimulus (CS), as recorded 0–3 days after conditioning [26, 27].

However, despite some research findings, learning-associated hippocampal synaptic modifications analogous to LTP/LTD have been demonstrated by only a few studies and remain largely unknown. Such experimental analysis could be hampered by the technical challenge, as speculated in some literature [1, 8, 9], of accessing the induction of hippocampal LTP/LTD during or after training, due to the sparse nature of the spatial distribution of a limited number of neurons or synapses.

In another line of research, although not directly examining hippocampal synaptic plasticity, many more reports using real-time monitoring of sensory-evoked hippocampal activity over the course of animal training, which may allow selective focus on the potential activity change at a specific subset (sensory-responding) of hippocampal synapses, have yielded a wealth of findings of plastic changes of hippocampal spike or LFP responses to learned stimuli. In a series of studies from animals exposed to trace eyeblink conditioning [24, 28–30], multiple single-unit recordings have shown that hippocampal CA1 pyramidal cells exhibit several stages of learning-related activity, with large increases in spike activity following both the CS and US (unconditioned stimulus) on the day when the task is initially learned, smaller moderate increases in activity on the following days of training, and decreases in activity following the US when the task is well learned. In more complex associative learning, studies in monkeys have reported an alteration of stimulus-selective response properties of hippocampal neurons that occurs at the same time as or after location-scene associative learning [31], as well as an increase of spikes in bursts of hippocampal principal cells, suggested to play important roles in memory formation, that emerges during the spatial navigation of associative memory task [32]. In addition to recording spike activity, LFP measurements from mice trained in a detection task have revealed the development of CS-evoked fEPSP responses in hippocampal CA1 areas, which were initially very small and increased in parallel to task learning over days [33].

The recordings of sensory-evoked hippocampal responses during animal training have advanced the understanding of hippocampal memory representations by showing the plastic changes of hippocampal spikes or synaptic responses to learned stimuli. In addition to these findings, however, the potentially recruited synaptic mechanism has not been further identified, with the issue of whether the learned hippocampal responses could be accounted for by synaptic plasticity induced in the hippocampus, or some upstream brain areas, not addressed. The lack of such identification might be due to certain technical limitations for determining LTP/LTD induction in a given brain region, as would exist in the multiple sessions (days) of task learning commonly applied in the previous reports, for example, with local pharmacological or physiological manipulation conducted over the learning sessions. Resolving this problem could help to further elucidate the hippocampal synaptic mechanisms of memory representation.

In this study, we developed a learning paradigm of water reward-based trace conditioning at a moderate difficulty level, in which animals are able to learn successfully the CS-US association via a single conditioning session consisting of a total of 120–220 conditioning trials. By recording single- or multiple-site LFPs in the dorsal CA1 hippocampus (dCA1) in combination with local pharmacological and optogenetic manipulation over the conditioning session, we investigated the change of dCA1 synaptic responses to the CS and the pathway at which dCA1 response changes take place, as well as the induction of dCA1 synaptic plasticity for the response change by further combination with whole-cell recording for sensory-evoked dCA1 synaptic responses.

Materials and Methods

Ethics Approval

All animal procedures were performed in accordance with the Animal Care and Use Committee of East China Normal University (M20190210). C57BL/6 mice (male, aged 8–12 weeks; commercially purchased, Slaccas Corp., China) were used. Animals were fed ad libitum and housed under a 12/12 h light/dark cycle. After experiments, animals were euthanized with an extra dose of isoflurane.

Animal Preparation for Conditioning and Electrophysiological Recording

Animals were prepared as previously reported [34]. Seven or eight days before trace conditioning or electrophysiological experiments, after anesthesia (pentobarbital sodium; 85 mg/kg, i.p.) the head was restrained in a stereotaxic apparatus (David Kopf Instruments, Tujunga, CA, USA), a metal plate was mounted on the skull. The body temperature is maintained at 37.3–37.8 °C by a heating blanket. The plate, allowing the exposure of Bregma and craniotomy, was attached with Vetbond tissue adhesive and dental cement. After the surgery, analgesia with carprofen (5 mg/kg, s.c.) was administered, and each animal was returned to its individual home cage.

A craniotomy was made 1 day before LFP recording or local drug infusion over the conditioning session, for which animals were anesthetized with isoflurane (1.5%), and the head was fixed in a customized stereotaxic apparatus. A small craniotomy (1–1.5 mm in diameter; the center from Bregma: 2.2–2.3 mm posterior and 1.5–1.6 mm lateral) was made above the cortex of the right hemisphere or both hemispheres (for bilateral drug infusions), after which it was covered with bone wax. For whole-cell recording (from naive animals), the same craniotomy was made but on the day of recording, as previously reported [26, 34].

Water Reward-based Trace Conditioning

Before trace conditioning, animals were habituated to head fixation in the behavioral set-up (in which mice were placed on a plastic plate only rotatable within an arc length of 5−6 cm at the body position to prevent running, with the metal plate attached to the head screwed to a metal bar installed on one side of the head) as well as the behavioral performance of receiving the water reward by continuously licking the spout placed ~5 mm directly in front of incisors (Fig. 1A, B). Animals were typically habituated for 4 sessions each per day and in some cases 5 sessions, and were water-restricted between daily experimental sessions, with the water restriction adjusted daily to maintain the body weight >80% of the initial value. Each habituation session lasted 25−40 min, during which a 4−5 μL drop of water was delivered to the spout at random inter-trial-intervals (ranging from 8 to 14 s) by gravity, controlled by a solenoid valve (Fig. 1A). Licks were monitored by breaking an infrared beam between an 860-nm LED (Panasonic) and a phototransistor (Fairchild Semiconductor) that were placed across the spout. Habituation (other than on the last day) was carried out in a simple cage containing only the behavioral set-up, and on the last day in the cage used on the following day for trace conditioning (containing both behavioral and electrophysiological set-ups), with a similarly low level of background noise (25−30 dB) and luminance (<0.1 lux) in both cages.

Fig. 1.

The procedures of behavioral experiments and the development of licking CRs during trace conditioning. A Schematic of the behavioral set-up, in which water-restricted mice are placed on a plastic plate and head-restrained to a metal bar, with a spout placed directly ahead of their incisors for water reward. B Procedures of the behavioral experiments, all of which were carried out in the behavioral set-up as displayed in A. C Two representative experiments (Mice 1 and 2) showing the development of licking CRs during conditioning. Upper, raster plots for licking behaviors in all individual conditioning trials (with CS-US pairing illustrated by the inset at the top). Each dot represents a lick; time 0 indicates CS onset, with licks >4 s after CS onset not displayed. Lower, time course plots for licking behaviors (10 trials grouped in 1 block for each data point), with CRs (black solid dots and line; initial emergence of significant CRs indicated by arrows) calculated by subtracting lick rates 0–4 s before CS onset (gray dashed line) from those 0–0.6 s after CS onset (just before US delivery; gray solid line). Error bars represent the mean ± SEM.

One day after habituation, a single session of trace conditioning was conducted on water-restricted mice in the cage for both behavioral and electrophysiological experiments. Each conditioning trial consisted of the CS of an auditory stimulus (8-kHz pure tone; intensity, 75 dB; duration, 0.1 s; delivered binaurally through a PC speaker located ~40 cm in front of the head), a 0.5 s stimulus-free period, and then the reward US of a 4−5 μL drop of water delivered to the spout. Conditioning trials were presented at random inter-trial intervals ranging from 8 to 14 s. In nearly all (> 95%) trials, animals could receive the water reward (i.e., CS-US associations) due to the behavioral performance of continuously licking at the spout, as they did in prior habituation; the water reward could be consumed by animals usually within 1.5 s after delivery (as reflected by an apparent increase and then decrease of lick rates during this period, see Fig. 1C, Upper, for two example mice; any unconsumed water was removed from the spout 1.5 s after delivery). The conditioning session was started 15−30 min after animals were head-fixed and terminated when they stopped licking the spout for a water reward, usually consisting of 120–220 trials of CS-US association.

For pseudo-conditioning, the same experimental procedures including the habituation were conducted as for conditioning, and the only difference was exposure to the CS alone or the pairing of CS and US in reversed order (i.e., US-CS pairing). For the presentation of flash visual stimuli, which were randomly paired with the US during conditioning, LED screens (7 × 7 cm²) were placed bilaterally ~2 cm from eyeballs, and whole-screen flashes (100-ms duration, 200 lux) on a dark background (<0.1 lux) were generated by a pulse generator (Master-8; A.M.P.I., Jerusalem, Israel).

Cued memory was tested on water-restricted mice, with the presentation of test trials (consisting of only the CS but not the US; at random inter-trial intervals ranging from 8 to 14 s) started when animals exhibited rare baseline licks after being placed in the behavioral set-up.

Electrophysiological Experiments

For acute single- or two-site LFP recording, glass pipettes filled with saline (tip opening, 2.8–3.2 μm; pulled from borosilicate glass tubing, Kimble Glass Inc.) were placed 5–20 min before conditioning or pseudo-conditioning into the dCA1 with a motor-driven manipulator (MMX7630, Siskiyou Corp.) (recording sites from Bregma (in mm): 2.2–2.3 posterior and 1.5–1.6 lateral; 1.3–1.5 below the cortical surface). LFP signals were acquired using a patch-clamp amplifier (Axopatch 200B, Molecular Devices) and sampled at 20 kHz by a data acquisition card (Digidata 1440, Molecular Devices), with bandpass filtering at 1 Hz and 10 kHz.

For acute multiple-site LFP recording, linear silicon probe arrays with 64 electrodes (equally spaced at 20 μm) were used. With the tip inserted perpendicular to the cortical surface, the probe was lowered at 5–10 μm/s to target the dorsal hippocampus. Electrode signals were amplified and sampled at 25 kHz using the RHD Recording system (Intan Technologies), with bandpass filtering at 1 Hz and 7.5 kHz. The position of silicon probes was guided by LFP and multiunit activity, and the laminar locations of recording sites were reconstructed with the aid of histologically identified electrode tracks stained with the green fluorescent dye DiO (3, 3-Dioctadecyloxacarbocyanine perchlorate).

Perforated whole-cell recordings in hippocampal dCA1 neurons were made in pyramidal cell layers; they were made in awake head-fixed mice that had been habituated to head restraint for conditioning but not water-restricted, as previously reported [26, 34]. After craniotomy, the dura mater was carefully dissected, and the exposed brain surface was kept moist with saline. Patch pipettes had a tip opening of 2.5–3.0 μm (pulled from borosilicate glass tubing, Kimble Glass Inc.) and resistance of 1.6–2.0 MΩ. The internal solution contained (in mmol/L) 136.5 K-Gluconate, 17.5 KCl, 9.0 NaCl, 1.0 MgCl₂, 10.0 HEPES, 0.2 EGTA, and amphotericin B (375 μg/mL), in which a small amount of glass beads (0.5–0.8 mg/mL; 5–15 μm in diameter; Polysciences, Inc.) was included for the availability of precipitate-free solution in the pipette tips. With a positive pressure of 250–350 mbar applied to the pipette interior, patch pipettes were advanced into the brain by a motor-driven manipulator (MMX7630, Siskiyou Corp.), at an initial speed of 15–30 μm/s and then 10–15 μm/s when reaching dCA1. Signals were acquired in current-clamp mode (I = 0) by a patch-clamp amplifier (Axopatch 200B, Molecular Devices) and sampled at 20 kHz by a data acquisition card (Digidata 1440, Molecular Devices), with 10 kHz low-pass filtering. Tone stimuli (8 kHz; 70–75 dB; duration, 100 ms) were delivered as described above for trace conditioning, and tone-elicited responses were measured by whole-cell recording usually for ~100 repeats, at 6 or 12 s inter-stimulus intervals.

Virus Injection

Animals were anesthetized with 5% isoflurane, and the head was restrained in a stereotaxic apparatus (David Kopf Instr.) for virus injection. A total volume of 0.3 μL (in CA3) or 0.8 μL (in entorhinal cortices) of either AAV1/2–hSyn–eNpHR3.0–eYFP or AAV1/2–hSyn–eYFP was injected at 0.03 μL/min in ipsilateral CA3 regions (from Bregma, in mm: 2.2 posterior and 2.6 lateral; 1.8 below the cortical surface) or entorhinal cortices (3.6 posterior and 4.5 lateral; 2.7 beneath the cortical surface) using a microsyringe connected to a glass pipette (tip opening, ~50 μm). After injection, the pipette was left in place for 10 min and then slowly withdrawn. With the head skin sutured, each animal was returned to its home cage.

Local Drug Infusion

For drug infusion, glass pipettes (tip opening, ~60 μm) connected to a microsyringe were placed unilaterally in dCA1 (together with ipsilateral LFP recording; with the tip located within 50 μm of LFP recording sites) or bilaterally in dCA1 (only for behavioral but not electrophysiological experiments; with the tip located at the site corresponding to unilateral infusion). Drugs were injected twice in a given dCA1 hemisphere (for bilateral injections, both hemispheres underwent the same injection over a similar period), with the first one starting 3–5 min before and the second one starting 5–10 min after conditioning onset. For each infusion, 0.1 μL AP-5 (1 μg/μL) or saline was infused at a constant rate for 5 min.

Histological Procedures

Animals were euthanized with an extra dose of isoflurane, and fixed by perfusion with 4% paraformaldehyde in 0.1 M phosphate-buffered saline (PBS). The brain was rapidly removed, post-fixed for 48 h (in the same fixative), immersed in 20% (24 h) and 30% (24 h) sucrose in PBS, and cut at 40 μm on a freezing microtome (CM3050 S, Leica Microsystems). The sections were incubated with 10 ng/ml 4,6-diamidino-2-phenylindole (Beyotime) in PBS to stain nuclei. Images were captured under a fluorescence microscope (DM4000 B LED, Leica Microsystems).

Data Analysis and Statistics

Nearly all (95%) LFP recording data were considered based on animals’ performance during conditioning, among which a small fraction (~105%) with an apparent fluctuation in LFP response amplitudes (>20% as compared to the maximal amplitude) were excluded. A liquid junction potential of − 13 mV was corrected for whole-cell recording data, and the data included had a resting membrane potential lower than − 60 mV.

For time-course analysis of individual data, 10 and 20 trials of measurements were grouped into one block for licking behaviors and LFP responses. The conditioning block (e.g., the first one) with significant changes in licking behaviors or LFP responses was determined according to P <0.05 (in some cases, <0. 08; paired or unpaired Student’s t-test).

One-dimensional current source density (CSD) analysis (second-derivative approximation) of LFP profiles was calculated using a three-point formula [35]: CSD_x = [(V_x–V_x–d) – (V_x+d–V_x)]/d², where V is the voltage, x is the point at which CSD is calculated, and d is the distance between two neighboring points.

Response onset latencies (from stimulus onset) and durations were inspected according to 2-fold-SD amplitudes in baseline noise. Unless otherwise noted, statistical analyses were made using paired Student’s t-test; average values are presented as the mean ± SEM for individual data and the mean ± SD for population data.

Results

Cued Memory Acquisition and Development of Licking CRs Over the Single Session of Trace Conditioning

In the single session of water reward-based trace conditioning applied in this study, water-restricted mice that had been habituated to the training conditions were head-restrained in the behavioral set-up (Fig. 1A, B), and exposed to the paired presentation of an auditory CS (duration, 0.1 s) and reward US (a drop of 4-5 μL water), with the US delivered 0.5 s after CS termination and any unconsumed water withdrawn from the spout 1.5 s after delivery. The conditioning session was continued until animals stopped licking the spout for water reward and usually consisted of a total of 120–220 trials of CS-US pairings.

At the early stage of conditioning, animals exhibited no licking responses (i.e., CRs) to the CS, and received the water reward by continuously licking the spout at a relatively high frequency (3–5 Hz; see Fig. 1C for an example), as in prior habituation sessions (see Materials and Methods). Licking CRs usually began to emerge after the presentation of 20–60 conditioning trials (in the examples in Fig. 1C: seen over 40–50 and 50–60 trials in Mouse 1 and 2, respectively), which was followed by a progressive increase in licking CRs as training continued. Before learning termination (animals stopped licking the spout for water reward; in Mouse 1 and 2 in Fig. 1C, before ~130 and ~180 trials, respectively), the increase of licking CRs was saturated in a small part (~30%) of experiments (as in Mouse 1) but not in the others (~70%; as in Mouse 2) (more detailed characteristics are shown in Fig. 8, together with the following experiments).

Fig. 8.

Similarities and differences in the development of LFP responses between Rad and LMol and further comparison with licking CRs. A, B From silicon probe data summarized in Fig. 7E, conditioning trials at which initial emergence (A; with different dots indicating different numbers of animals) and initial saturation (B) of the LFP increase in Rad or LMol. In B, for the 3 cases (arrows) with saturation only in LMol but not Rad, cut-off values (trials with learning termination) are plotted for Rad. C (upper) For glass electrode recordings either in Rad or Py, conditioning trials at which initial emergence of an LFP increase or licking CRs occur (n = 32). (Middle and lower) Time courses of the changes of LFP responses (middle) and licking behaviors (lower) averaged across population data with the time (trial) aligned to the initial emergence of LFP increase (indicated by vertical dashed lines), using the data from 29 to 32 animals as displayed above with the initial emergence of the former found to be no earlier than the latter. D As in C, for the data from 7 animals with the increases of LFP responses and licking CRs both saturated: upper, conditioning trials at which initial saturation occurs; middle and lower, time courses averaged across population data with the time aligned to initial saturation of the LFP increase.

In the cued memory test, as conducted 1 day (Fig. S1A) or 7 days (Fig. S1B) after task learning, we found that almost all animals (>90%) were able to memorize successfully the CS-US association via the single session of trace conditioning.

Development of CS-Evoked dCA1 LFP Responses Over Trace Conditioning

In mice exposed to the reward-based trace conditioning as shown in Fig. 1, we made acute LFP recordings in dCA1 to examine possible plastic changes of synaptic responses to the tone CS (Fig. 2A). LFP recordings were first made with single glass electrodes in the dCA1 stratum radiatum (Rad; 80–100 μm below the pyramidal cell layer, Py), in which CS-evoked LFP responses with an overall negative polarity were found (e.g., see Fig. 2C1_, right), similar to spontaneous LFP signals recorded in the absence of sensory stimuli (Fig. 2B).

Fig. 2.

Learning-induced increase of CS-evoked LFP responses in dCA1 Rad. A Schematic of dCA1 LFP recording during training. B An example recording from dCA1 Rad shows that spontaneous LFP signals consist primarily of the events of negative polarity. C Representative data from dCA1 Rad showing a learning-induced increase in CS-elicited LFP responses. C₁, left, raster plots of licking behaviors (1 s before and after CS onset; displayed as in Fig. 1C, upper). C₁, right, LFP activity (0.1 s before and 0.2 s after CS onset) in all conditioning trials; data are normalized to peak amplitudes (in negative polarity) and indicated by the color scale (0 represents the baseline and 1 represents the peak amplitudes); superimposed black traces are LFP activity averaged over 45 consecutive trials. C₂, for the example in C₁, time course plots of licking behaviors (upper; displayed as in Fig. 1C, lower) and LFP responses (lower; 20 trials of data grouped in 1 block for each data point; arrow, initial response increase). Absolute values are shown for LFP response amplitudes in this and other figures unless otherwise noted. Bars, SEM. D Summary of all experiments as illustrated in C (n = 12). Left, summary of time course plots: behavioral data are normalized to lick rates 0–4 s before CS onset as measured before the emergence of licking CRs; LFP data are normalized to baseline amplitudes (with 10 trials of data grouped in 1 block); for both behavioral and LFP data, n values corresponding to different trials are indicated by the insets at the top. Right, for all individual LFP recordings, summary of response amplitudes at the early stage (i.e., baseline amplitudes) and late stage of conditioning (for the recordings in which the response increase is saturated, values are measured at saturated levels, and for those not, measured over the last 20–40 conditioning trials), which are connected by lines for the same recordings. Bars, SD; ***P <0.001.

By real-time monitoring of LFPs in dCA1 Rad over the conditioning session, we found in all experiments a development of LFP responses to the CS. As shown in an example (Fig. 2C) and in a summary (Fig. 2D; n = 12), CS-evoked LFP responses remained unchanged during the early stage of training (in the example data, within the first 100 trials), and were of a relatively small amplitude ranging from 0.04 to 0.20 mV (mean ± SD, 0.11 ± 0.06 mV; Fig. 2D, Right; for LFP response amplitudes, absolute values are reported, unless otherwise noted). As training continued, usually when 40–100 conditioning trials had been presented, the amplitudes of LFP responses began to increase (Fig. 2C, C). In a part of the data (4/12), the response increase was found to be rapidly saturated within the first 20-trial block of conditioning over which a response increase was initially detected (as in the example data in Fig. 2C; for time-course analysis of LFP response changes in individual data, 20 trials of data were grouped in 1 block). In the other part (8/12), responses further increased in the following 20-trial blocks of conditioning (see example data in Figs 3C and 4B for other groups of experiments), with the increase saturating before learning termination (2/12) or not (6/12) (more detailed characteristics including the comparison with licking behaviors are shown in Fig. 8 together with the following experiments). The final amplitudes of CS-evoked LFP responses were 0.16–0.57 mV (mean ± SD, 0.36 ± 0.11 mV), as 2- to 10-fold (mean, 4.2 ± 2.4) of baseline responses (measured from data shown in Fig. 2D, Right).

Fig. 3.

Learning-specific increase of dCA1 LFP responses to sensory stimuli. A, B In animals receiving CS alone (A; n = 7) or US-CS pairings (B; n = 8), there is no increase in Rad LFP responses to the CS (left, example data; right, summary). C During conditioning, as in Fig. 1 but with an additional presentation of a flash stimulus randomly paired with the US, no increase occurs in Rad LFP responses to the flash, with a significant increase in CS-evoked responses (n = 12).

Fig. 4.

Reduction of learning-increased dCA1 responses by optogenetic inhibition at Schaffer collateral or temporoammonic axons. A–C Optogenetic photostimulation at eNpHR3.0-expressing Schaffer collateral axons (an example section displayed in A), as delivered during the late stage of conditioning, significantly reduces the CS-evoked LFP responses recorded in Rad, as shown by representative data (B; yellow and black traces in B₁ are responses recorded with light ON and OFF, respectively; yellow bars in B₂ indicate the periods with light ON) and summarized data (C; n = 8). In these experiments, because the time course of LFP response changes is not analyzed on-line during recording, the measurement with optogenetic inhibition starts coherently in all animals from 130 conditioning trials, and data from some animals exhibiting licks for water reward, <250 conditioning trials were discarded in off-line analysis. D–F As in A–C, for the experiments with optogenetic photostimulation of eNpHR3.0-expressing temporoammonic axons and LFP recording in LMol (n = 9). The amplitudes in E₂ and F are measured at the second peak. G Changes of response amplitudes by optogenetic inhibition of Schaffer collaterals (from Rad data in C; left) and temporoammonic axons (from LMol data in F; middle), as well as the results from eYFP controls (with control virus injected in CA3 or entorhinal cortex; n = 4 for each; right). H Comparison of the reduced response amplitudes (as shown in G, left and middle) with baseline response amplitudes measured in the same recordings during the early stages of conditioning. I For data in G, no change in licking CRs (calculated as in Fig. 1C, lower) by optogenetic photostimulation. *P <0.05; ***P <0.001; n.s., no significant difference. Scale bars, 100 μm.

In a further comparison of the CS-elicited responses at late stages with early stages of conditioning, we found a slight decrease in onset latencies (from CS onset, mean ± SD: 18.1 ± 2.3 ms and 16.0 ± 2.2 ms for early and late stages, respectively; P = 0.005; Fig. S2A), a much more marked increase in durations (53.3 ± 8.0 ms and 65.6 ± 15.6 ms for early and late stages; P = 0.004; Fig. S2C), and no significant change in peak times (from CS onset: 34.4 ± 5.6 ms and 35.3 ± 4.7 ms for early and late stages; P = 0.23; Fig. S2B). The slight decrease in onset latencies, as inconsistent with the unchanged peak times, is likely due to the difference in signal-to-noise ratios we encountered in measuring response onsets (according to Z-score ≥2 of baseline noise) on the data from early and late stages of conditioning.

In control experiments from pseudo-conditioned animals, which received the CS alone (Fig. 3A; n = 7) or the pairing of CS and US in reversed order (Fig. 3B; n = 8), no increase in CS-elicited responses was found by the same LFP recording in dCA1 Rad. In another experiment (Fig. 3C; n = 12), when a flash visual stimulus (100-ms duration, 200 lux) was additionally presented during conditioning but randomly paired with the US, we also found no increase in flash-evoked responses in dCA1 Rad, with a similar increase in tone CS-evoked responses (as in Fig. 2C, D). These findings reveal that the development of sensory-evoked dCA1 LFP responses is specifically associated with task learning as well as the stimulus learned.

Development of CS-Evoked dCA1 Synaptic Responses in Both Schaffer Collateral and Temporoammonic Pathways

Hippocampal CA1 receives direct inputs mainly from CA3 and the entorhinal cortex, through Schaffer collaterals to Rad and stratum oriens (Or) and temporoammonic pathways to LMol. With the findings of the development of dCA1 LFP responses to the CS, we next asked at which dCA1 input(s) CS-evoked synaptic responses could undergo learning-induced alterations. For this purpose, we applied an optogenetic approach for the ability to selectively inhibit action potential firing at Schaffer collateral or temporoammonic axons in CA1, by transfecting CA3 or entorhinal cortical neurons 4–6 weeks before training with an adeno-associated virus (AAV, serotype 1/2) vector carrying the yellow light-gated third-generation chloride pump halorhodopsin (eNpHR3.0). In such experiments, we found interestingly that CS-evoked dCA1 responses relayed from Schaffer collaterals and temporoammonic pathways are both enhanced during conditioning.

Figure 4A-C displays the results obtained during the conditioning session from animals that had received ipsilateral CA3 eNpHR3.0 injection (n = 8), in which LFPs were recorded in dCA1 Rad as in Figure 2. At the late stage of conditioning, when CS-elicited LFP responses were readily increased, 589-nm yellow light was delivered to dCA1 via an optic fiber (with the tip located ~50 μm lateral to and ~50 μm above LFP recording sites) for optogenetic photostimulation at eNpHR3.0-expressing Schaffer collateral axons (see Fig. 4A for an example section), with the duration (light on) and inter-stimulus-interval (light off) both being 10 conditioning trials. As shown by the example (Fig. 4B) and summary data (Fig. 4C, G, Left), light delivery resulted in a significant reduction in CS-evoked LFP responses (mean ± SD, from 0.26 ± 0.13 mV to 0.16 ± 0.10 mV; P = 0.001), a change not seen in the measurement from animals injected with control virus (Fig. 4G, Right). Moreover, the reduction of response amplitudes (0.10 ± 0.05 mV) was not smaller than the baseline response amplitude measured at the early stage of conditioning (0.08 ± 0.04 mV; P = 0.10; Fig. 4H, Left), indicating that the reduced responses comprise learned synaptic responses at Schaffer collaterals and not exclusively baseline responses.

Figure 4D, F displays the results from animals that had received ipsilateral eNpHR3.0 injection in entorhinal cortices (n = 9), in which LFPs were recorded in dCA1 LMol (200–300 μm below Pys). Compared to Rad responses, the most notable difference in LMol LFP responses to the CS was the two peaks in negative polarity (Fig. 4B1, E1), which emerged earlier (mean ± SD, 12.3 ± 1.28 ms; unpaired Student’s t-test, p <0.0001) and later (10.8 ± 2.63 ms; p <0.0001) than the peak of Rad responses, respectively (with peak times from another group of data summarized above in Fig. S2B). The first but not the second peak of LMol responses might be due to dendritic spikes rather than synaptic responses per se, as indicated by the following findings when NMDA (N-methyl-D-aspartate) receptors were locally blocked (Fig. 5C1; as discussed below). Despite these differences, LFP responses in LMol were also increased during task learning, and when optogenetic photostimulation was applied to eNpHR3.0-expressing temporoammonic axons (see Fig. 4D for an example section) at the late stage of conditioning, a significant reduction in LMol LFP responses was found, but only at the second peak (mean ± SD, from 0.34 ± 0.07 mV to 0.19 ± 0.08 mV; P <0.0001; Fig. 4E, F, G, middle), not the first peak (P = 0.37; Fig. 4E1). The reduced response amplitudes (0.15 ± 0.04 mV) significantly exceeded the baseline response amplitudes (0.10 ± 0.02 mV; P = 0.012; Fig. 4H, right), indicating that the reduced responses comprise learned synaptic responses at temporoammonic pathways.

Fig. 5.

Increase of CS-evoked dCA1 responses depends on dCA1 NMDARs. A With LFP recording in Rad and local drug infusion close to recording sites, example data from AP-5 and saline groups show that AP-5 infusion almost completely blocks the increase of CS-elicited responses (traces in A₁ are responses averaged for every 45 trials after conditioning onset). B Summary of the experiments in A (presented as in Fig. 2D, lower; n = 12 for the AP-5 group, with n values corresponding to different trials indicated by the insets at the top; n = 9 for the saline group, with all data included for the displayed maximal trial number). C, D As in A, B, for LFP recordings in LMol with local drug infusion close to recording sites (n = 9 for both AP-5 and saline groups). Note that the first peak of LMol responses is absent with AP-5 treatment (C₁; responses averaged for every 40 or 45 trials after conditioning onset); for the saline group, response amplitudes (C₂ and D) are measured at the second peak.

LTP-like Potentiation of dCA1 CS-responding Excitatory Synapses at Both Schaffer Collateral and Temporoammonic Pathways

We next sought to determine the neural mechanism underlying the learning-induced increase of dCA1 synaptic responses to the CS. We had shown that animal training specifically increased the LFP responses to the tone CS but not to a visual stimulus that was also presented during conditioning but randomly paired with the US (Fig. 3C). Such findings provided evidence that the increase of CS-evoked dCA1 synaptic responses may result from synaptic modifications (LTP/LTD) at CS-responding synapses (in CA1 or upstream brain regions), rather than certain nonspecific modulation such as changes of global brain states or local hippocampal states reported to be able to modulate hippocampal CA1 LFPs or spike responses to sensory stimuli [36–38].

Then, the possible induction of dCA1 synaptic plasticity as an underlying mechanism was tested. In one group of experiments, we recorded LFPs in Rad in combination with local infusion of the NMDA receptor (NMDAR) antagonist AP-5 in the ipsilateral dCA1. With this treatment, we found little change in CS-elicited responses over the conditioning session, while a significant response increase was seen in saline controls (see Fig. 5A for an example and Fig. 5B for a summary; n = 12 for AP-5 and 9 for the saline group). Unlike the blocking effects on LFP response changes, the AP-5 treatment did not change the behavioral expression of licking CRs (Fig. S3A), and such a lack of change was consistently found in the above experiments with unilateral optogenetic inhibition (Fig. 4I).

Similar results were obtained in another group of experiments when LFPs were recorded in LMol in combination with local AP-5 infusion (Fig. 5C, D; n = 9 for both AP-5 and saline groups). We also found that, in the presence of AP-5, the first peak of LMol LFP responses to the CS (Fig. 4E1) was absent (Fig. 5C1), implying that such events are likely the result of slow complex dendritic spikes generated postsynaptically with a possible dependence on NMDAR activation, as reported in vitro at the distal apical dendrites of CA1 or neocortical pyramidal cells [39–42]. This possibility is also supported by the resistance of these events to optogenetic inhibition in part of the temporoammonic axons (Fig. 4E1), reminiscent of the all-or-none manner of spike generation. By contrast, the amplitudes at the second peak of LMol responses were not altered by AP-5 (Fig. 5D, Right; unpaired Student’s t-test, P = 0.45), and are displayed in Fig. 5C2, D showing LMol response changes over the conditioning session.

The findings with local dCA1 AP-5 infusion indicated an induction of synaptic plasticity in dCA1 neurons that accounts for the increase of dCA1 synaptic responses to the CS, and such plasticity may be potentiation at excitatory synapses rather than potentiation or depression at inhibitory synapses. This indication was further supported by two lines of evidence from conditioned and naive animals for identifying membrane potential (V_m) changes of dCA1 neurons in response to sensory stimuli. First, in a published study, our whole-cell recordings from anesthetized rats made 0–3 days after trace fear conditioning show an LTP-like synapse unsilencing as induced by task learning in a large population (~70%) of dCA1 neurons at their silent glutamatergic synapses that are potentially responsive to the CS, while significant changes of inhibitory V_m responses to the CS are not detected [26, 27]. Second, in unpublished data, we have found with whole-cell recording from awake naive mice that a tone stimulus identical to the CS elicits both excitatory and inhibitory V_m responses in dCA1 neurons (Fig. 6A), with no significant difference between the amplitudes of excitatory (mean ± SD, 2.54 ± 0.99 mV) and inhibitory responses (2.13 ± 0.57 mV; unpaired Student’s t-test, P = 0.06; Fig. 6C), but only excitatory and very few inhibitory responses are seen within the period of LFP responses to the CS (Fig. 6B, n = 59 responding cells). These findings further suggest that the learning-induced LFP responses to the CS, with a large magnitude of increase (~300% above baseline responses across population data; Fig. 2D, right) and unchanged rise/peak time (Figs S2B for Rad responses and 6D for LMol responses), would not be expected to result from plasticity at CS-responding inhibitory synapses, either potentiation (e.g., resulting in larger volume-conducted current among a sink-source coupling) or depression (e.g., resulting in a disinhibition of synaptic excitation). Thus, based on these findings together with those by optogenetic inhibition (Fig. 4), we conclude that potentiation of CS-responding excitatory synapses is induced during conditioning at both dCA1 Schaffer collateral and temporoammonic inputs.

Fig. 6.

Onset latencies of tone-evoked excitatory and inhibitory V_m responses in dCA1 neurons compared with the period of LFP responses. A Average tone-evoked V_m responses of 3 example neurons recorded from awake naive mice, with excitatory responses in Cell 1, inhibitory responses in Cell 2, and both excitatory and inhibitory responses in Cell 3. Arrows, stimulus onset; shadows on traces, ± SEM; gray dashed line, baseline V_m in average traces as indicated. B Cumulative distribution plots for onset latencies of excitatory and inhibitory V_m responses. In these experiments, a total of 74 cells were recorded, and 59 cells were responsive to the tone by showing either excitatory (n = 18) or inhibitory responses (n = 27) and both excitatory and inhibitory responses (n = 14, from which onset latencies of both excitatory and inhibitory responses were included in this analysis). For comparison with the period of tone CS-evoked LFP responses, the times of the second peak of LMol responses are plotted (from experiments summarized in Figs 4F and 5D (saline group); measured at early stages of conditioning). Note the two different time scales used for the x-axis. C No significant difference between the amplitudes of excitatory and inhibitory V_m responses as measured from data in B. D No significant difference in the times of the second peak of LMol LFP responses between the early and late stages of conditioning (data from Fig. 5D, saline group).

As for the behavioral expression of licking CRs, significant impairment was not found with unilateral AP-5 infusion in dCA1 (Fig. S3A) but was found in the next experiment with bilateral AP-5 infusion (in Rad; Fig. S3B; n = 6). These results confirm that the trace conditioning used in this study is dependent on the hippocampus, as commonly reported for various learning paradigms of trace conditioning [1, 24].

CS-Associated Current Sinks in dCA1 Rad and LMol as well as Similarities and Differences in LFP Response Increases between Rad and LMol

We next examined the laminar profile of LFP responses and net transmembrane currents (current sinks and sources as the local origin of field potentials) elicited by the CS in dCA1 during conditioning, using 64-site recording with silicon linear probes that were placed acutely in the dorsal hippocampus along the dorsoventral axis from the Or of CA1 to the dentate gyrus (DG) (with the tip inserted at around the boundary between the DG and the thalamus; Fig. 7A).

Fig. 7.

Laminar CSD profiles of CS-evoked dCA1 responses and the development of LFP responses in Rad and LMol. A Histologically identified electrode tracks of a 64-channel silicon probe recording in the dorsal hippocampus, with recording sites 1 and 64 indicated by arrowheads. GC, granule cell layer; ML, molecular layer. B₁ For the measurements at the late stage of conditioning from representative data, color plots for laminar profiles of CS-elicited LFP responses (black arrow above, the peak time of Rad responses; gray arrows, the first and second peak times of LMol responses). B₂ For the measurements at early (open symbols) and late stages (filled symbols) of conditioning from the data in B₁, LFP amplitudes (raw values but not absolute values, as displayed in B₁) of all recording sites that were measured at the peak of Rad responses (left) as well as the first (middle) and second peak (right) of LMol responses. C Color plots for the laminar CSD profiles calculated from the LFP data in B₁. D, E Development of CS-elicited LFP responses simultaneously recorded from Rad and LMol, as shown by an example (D; traces in D₁ are responses averaged for every 55 trials after conditioning onset) and the summary (E, presented as in Fig. 2D, lower; n = 12), in which LMol responses are measured at the second peak, and the two sites exhibiting maximal LFP amplitudes in Rad and LMol are selected from the 64-channel recordings. Scale bar, 100 μm.

Silicon probe recordings were made from a total of 12 mice undergoing trace conditioning; representative recordings of the laminar profiles of CS-evoked LFPs at late stages of conditioning are shown in Figure 7B. The CSD was then calculated from the LFP data to determine the laminar profile of net transmembrane currents. We found two current sinks (net inward current) in dCA1—one (Sink 1) in Rad and the other (Sink 2) in LMol—and another sink (Sink 3) in the DG (Fig. 7C). These sinks were seen in all the 12 silicon probe data at the late stage of conditioning, except that Sink 1 was undetectable at early stages of conditioning in two cases. In parallel, two sources (net outward current) were found, one (Source 1) in dCA1 (between Sinks 1 and 2) and another (Source 2) in the DG (between Sinks 2 and 3).

As for the CSD components in dCA1, Sinks 1 and 2 can both be attributed to active inward currents (EPSPs) that are generated at Schaffer collaterals and temporoammonic input, respectively, given the above results of optogenetic inhibition (Fig. 4), local AP-5 infusion (Fig. 5), and whole-cell recording of dCA1 V_m responses to tone stimuli (Fig. 6A–C). In addition, given the apparent source (Source 1) existing between Sinks 1 and 2, the LFP responses (fEPSPs) measured from Rad or LMol (in this and other groups of experiments), particularly those that increased during conditioning, are expected to result mainly from active inward currents at Sink 1 or Sink 2, but not both.

Then, we compared the temporal profiles of the increase in LFPs recorded by silicon probes from different sites in dCA1 during conditioning. We first found that, at various laminar depths within Rad or LMol, the time courses of the LFP increase were very similar (Fig. S4A), as can be expected due to the same local origins of these potentials. Further, between the measurements from Rad and LMol, a simultaneous LFP increase was recorded in the initial time, and this was detected in nearly all (11/12) data over the same 20-trial block of conditioning (see Figs 7D2 and Fig. S4A for examples and Figs 7E, Left, and 8A for summary). On the other hand, the saturation of the increase, that occurred before learning termination in 5 cases in Rad and 8 cases in LMol (which included the 5 with saturation in Rad), did not appear simultaneously in Rad and LMol. In almost all (7/8) cases, the initial saturation of the increase was found in Rad and LMol in notably different conditioning blocks (see Figs 7D2 and Fig. S4A for two examples and Fig. 7E, Left, and 8B for summary). However, the order of initial saturation of the Rad and LMol response increase varied across probe recordings (Fig. 8B), and the mean times were not significantly different over population data (conditioning trials of initial saturation, mean ± SD: 132 ± 69 and 105 ± 45 for Rad and LMol, respectively; P = 0.42). In addition, the final ratios of response increase in Rad and LMol were also found to be similar over population data (response amplitudes at late stage versus early stage, mean ± SD: 4.1 ± 3.6 and 4.0 ± 2.4 for Rad and LMol, respectively; P = 0.68; Fig. 7E, Right).

Increase of CS-Evoked dCA1 Responses Emerges Later and Saturates Earlier than Licking CRs

Finally, we compared the time courses of LFP increase with the behavioral expression of licking CRs induced during conditioning. For this analysis, experimental data from a total of 32 mice were included, in which LFPs were measured from Rad or Py with glass electrodes, by single-site recording in either Rad (n = 12 mice; summarized above in Fig. 2D) or Py (n = 10) and two-site recording in Rad and Py (n = 10; with LFP data in Rad selected for analysis), and the two-site recording data also revealed a similar time course of LFP changes in Rad and Py (Fig. S4B), as in Fig. S4A.

We first compared the initial emergence of the LFP increase and licking CRs. We found in the vast majority of cases (25/32 mice) that the former significantly fell behind the latter, and in the other cases, the former was in the same conditioning block as (4/32) or earlier than (3/32) the latter, with the mean number (conditioning trials as presented) of the former being 68 ± 31 (SD) and the latter 43 ± 21 (P <0.0001; Fig. 8C (upper)). The time lag of the initial emergence of the LFP increase compared to licking CRs was also indicated by the time course analysis of LFP changes and licking CRs that were averaged across population data with the time aligned in each case to the initial LFP increase (Fig. 8C, middle and lower), which was applied to the data of 29/32 mice with the former found to emerge no earlier than the latter (Fig. 8C, upper).

In the same group of experimental data, we further compared the saturation of the increase in LFPs and licking CRs. We found in about half of the mice (14/32) that the increases in neither the LFPs nor the CRs were saturated before conditioning termination. In others (n = 18), an LFP increase was found to be saturated in 16 (89%) and licking CRs saturated in 9 (50%) mice (both saturated in 7 mice), with the percentage of the former significantly higher than the latter (P = 0.027, Fisher’s exact test), suggesting that the increase of LFPs tends to saturate earlier than CRs. Such a difference was also revealed by analysis of the data from the 7 mice with saturation of both increases of LFPs and CRs (Fig. 8D. Upper; LFP, mean ± SD, 91 ± 36; CR, 121 ± 39; P = 0.02) as well as the time courses of LFP changes and CRs that are averaged over population data with the time aligned to the initial saturation of the LFP increase (middle and lower).

These results reveal that the induction of Rad responses to the CS, i.e., potentiation at Rad CS-responding synapses, tends to emerge later and saturate earlier than the development of CRs, and given the overall similar times in both initial emergence (Fig. 8A) and initial saturation (Fig. 8B) of response increases between Rad and LMol, such a conclusion may also apply to LMol.

Discussion

In the single session of water reward-based trace conditioning, we examined hippocampal mechanisms of dCA1 synaptic plasticity that may account for cued memory acquisition. By combining acute LFP recording with selective optogenetic inhibition, we first found an increase of dCA1 synaptic responses to the CS that is induced during conditioning at both Schaffer collateral and temporoammonic inputs. Further experiments with local AP-5 infusion in dCA1 for determining the dependence of the learned dCA1 responses on postsynaptic NMDARs and whole-cell recording in dCA1 neurons for determining the excitatory/inhibitory nature of V_m responses to unlearned stimuli suggest an induction of synaptic potentiation of CS-responding excitatory synapses at these dCA1 inputs. Silicon probe recording revealed similarities between the time courses of the induction of Rad and LMol synaptic potentiation, both of which emerge later and saturate earlier than the development of CRs as indicated by further analysis. With these findings, we conclude an induction of synaptic potentiation of dCA1 CS-responding excitatory synapses that takes place at both Schaffer collateral and temporoammonic pathways but within a relatively short time window over a learning session.

Synaptic Potentiation of CS-Responding dCA1 Excitatory Synapses at Both Schaffer Collaterals and Temporoammonic Pathways

By real-time monitoring of hippocampal activity during task learning, previous reports have provided a wealth of findings on learning-associated changes in hippocampal spike or LFP responses to sensory stimuli (e.g., the CS of trace conditioning) that may account for hippocampal memory representation [24, 28–33]. However, with respect to the learned hippocampal responses as recorded during the task, further identification of the underlying synaptic mechanism is still lacking. This is possibly due to certain technical limitations for determining the induction of synaptic modifications in a given brain region, as would exist in the multiple sessions (days) of animal training being commonly used in the previous reports, for example, with the use of local pharmacological or physiological manipulation over the learning sessions. Resolving this problem could offer potential approaches for identifying hippocampal mechanisms of synaptic plasticity as recruited for memory acquisition and storage.

In this study, we developed a learning paradigm of reward-based trace conditioning with a moderate level of difficulty, in which animals are able to successfully learn the CS-US association in a single conditioning session usually consisting of 120–220 trials. During conditioning, single-site extracellular recording in the dCA1 hippocampus first revealed a learning-induced increase in LFP responses to the CS. The increased dCA1 LFPs were suggested by selective optogenetic inhibition originating from both Schaffer collateral and temporoammonic inputs (Fig. 4). At both of these inputs, we further conclude an induction of LTP-like synaptic potentiation at CS-responding excitatory synapses, as drawn from the following findings: (1) the dependence of the LFP increase on dCA1 NMDARs (Fig. 5); (2) the overall excitatory nature of V_m responses of hippocampal dCA1 neurons elicited by the unlearned tone within the period of dCA1 LFP responses (Fig. 6B); (3) the learning-induced excitatory but not inhibitory V_m responses to the CS as we previously reported in hippocampal dCA1 neurons with whole-cell recordings from anesthetized rats made 0–3 days after trace fear conditioning [26, 27].

Consistent with this, the laminar CSD profiles measured by 64-channel silicon probes showed two current sinks in dCA1, Sink 1 and 2 in Rad and LMol, respectively (Fig. 7C). The existence of an apparent source (Source 1) between Sink 1 and 2 further suggests that the LFP increase as measured from Rad or LMol primarily reflects the increase of dCA1 synaptic responses (i.e., synaptic potentiation) in Rad (Sink 1) or LMol (Sink 2), but not a mixture of both.

Interpretation of the Physiological Significance of Synaptic Potentiation at Schaffer Collaterals and Temporoammonic Pathways

The hippocampal formation contains two distinct excitatory pathways that are topographically organized and referred to as the trisynaptic pathway (which carries information as follows: entorhinal cortex → DG → CA3 → CA1 → entorhinal cortex) and the monosynaptic pathway (entorhinal cortex → CA1 → entorhinal cortex) [43]. The trisynaptic and monosynaptic pathways both receive information from the entorhinal cortex, but from layer II and layer III, respectively [43].

In this study, we provide evidence for the induction of synaptic potentiation of dCA1 CS-responding synapses at both trisynaptic (Schaffer collateral) and monosynaptic (temporoammonic) inputs. With respect to the possible physiological function of synaptic plasticity at these parallel CA1 inputs, we suggest an interpretation that is formulated based on a computational model proposed by György Buzsáki [44]. In this model, it is postulated that neocortical (experienced) representations are funneled to the hippocampus primarily by way of layer III of the entorhinal cortex, and the main function of layer II entorhinal cortical neurons is to compare experienced neuronal representations (from the neocortex) with expected neuronal representations (feedback from the hippocampus), that is, to compute and input the novelty (the “error” in computational jargon). In light of Buzsáki’s model, we thus propose that synaptic modifications at the temporoammonic (monosynaptic) pathway, as created during conditioning at CS-responding excitatory synapses, could underlie the storage of experienced neuronal representations, and those at the Schaffer collateral (trisynaptic) pathway could underlie the storage of more deeply processed information, i.e., more specific knowledge about the novelty (error). Future examination of this issue, theoretically or empirically, may provide new insight into hippocampal synaptic mechanisms of memory processing.

Distinct Temporal Profiles of LFP Responses in Rad and LMol

In the LFP responses elicited by the CS in Rad and LMol, measured with glass electrodes (Figs 4E1 and 5C1) or silicon probes (Figs 7B1, D1), we find two peaks in LMol responses but just one peak in Rad responses. In addition, the LMol responses containing the first peak emerge earlier than the Rad responses (including response onset and peak times; Figs 4E1 and 7B1, D1), and the LMol responses containing the second peak emerge later (including response peak and offset times). These findings indicate a sequential order of synaptic activation in dCA1 temporoammonic pathways and Schaffer collaterals engaged in relaying sensory information to dCA1 circuits.

In the presence of AP-5, LMol responses containing the first peak were completely blocked (Fig. 5C1), while those at the second peak were not significantly altered (Fig. 5D, right), indicating that the former responses might result from the postsynaptic generation of slow complex dendritic spikes (as reported in vitro at the distal apical dendrite of CA1 or neocortical pyramidal cells with a possible dependence on NMDARs [39–42]), but not synaptic responses, as evoked by the CS. Such a possibility is also supported by the findings that the amplitudes at the first peak of LMol responses remained unchanged when temporoammonic axons were partially inhibited by optogenetic photostimulation (Fig. 4E1), reminiscent of all-or-none spike generation, while the amplitudes at the second peak were significantly reduced. However, it is surprising to note that with AP-5 treatment, LFP responses are completely undetectable for the first peak (Fig. 5C1), including those subthreshold and possibly resistant to NMDAR blockade (e.g., AMPAR-mediated EPSPs). Perhaps, the putative subthreshold synaptic responses for generating putative postsynaptic dendritic spikes were too small or very sparsely distributed and thus not detectable in the LFP measurement.

Time Courses of the Induction of dCA1 Synaptic Potentiation Compared with Behavioral CRs

A significant time lag was found in the initial increase of CS-elicited dCA1 responses in Rad and LMol compared to animals’ performance of licking CRs, i.e., initial acquisition of the CS-US association (Fig. 8A, C). This result implies that the development of dCA1 responses to the CS, i.e., the induction of synaptic potentiation at CS-responding synapses, might be accounted for by the acquisition of the CS-US association. On the other hand, because bilateral AP-5 injection in dCA1 greatly impaired animals’ performance of CRs (Fig. S3B), learning the CS-US association may also rely on dCA1 synaptic plasticity, which could occur initially at a very small number of CS-responding synapses or beyond CS-responding synapses.

In a series of reports on animals (rabbits) receiving trace eyeblink conditioning [24, 28–30], it was found across population data that the learning-related increase in hippocampal CA1 spike responses to the CS precedes the behavioral expression of CRs by several conditioning trials, differing from our findings of an overall apparent time lag of the LFP increase compared to CRs (Fig. 8A, C). This discrepancy might be explained by two differences between our experiments and the previous reports. First, in the previous reports, the development of CS-evoked spike responses in CA1 neurons as well as behavioral CRs was recorded when animals had received days of trace conditioning, which possibly involved daily post-conditioning memory consolidation (e.g., during slow-wave sleep and rapid-eye-movement sleep [45–48]) with potential effects of generating synaptic plasticity (functional and/or structural). Second, unlike the result of LFP measurements, alterations in spike responses could be caused not only by synaptic modifications but also by postsynaptic changes of intrinsic neuronal excitability (which was indeed reported by the same research group in hippocampal slices prepared from eyeblink-conditioned animals [29, 49, 50]).

Taken together, in the single session of reward-based trace conditioning, we provide evidence for an LTP-like potentiation of CS-responding dCA1 excitatory synapses that are induced at both Schaffer collateral and temporoammonic inputs, and this occurs overall within a shorter time window than the development of behavioral CRs. Our findings suggest a hippocampal mechanism of dCA1 synaptic plasticity for cued memory acquisition. Future studies, theoretically or empirically, will be needed to determine the specific role of synaptic potentiation induced at Schaffer collaterals and the temporoammonic pathways in mediating hippocampal memory acquisition and storage.

Supplementary Information

Below is the link to the electronic supplementary material.

Competing interests

The authors claim that there are no conflicts of interest.