Although it is generally assumed that the hippocampus is involved in associative learning, the specific contribution of the different synapses present in its intrinsic circuit or comprising its afferents and efferents is poorly defined. We studied here activity-dependent changes in synaptic strength of 9 hippocampal synapses (corresponding to the intrinsic hippocampal circuitry and to its main inputs and outputs) during the acquisition of a trace eyeblink conditioning in behaving mice. The timing and intensity of synaptic changes across the acquisition process was determined. The evolution of these timed changes in synaptic strength indicated that their functional organization did not coincide with their sequential distribution according to anatomical criteria and connectivity. Furthermore, we explored the functional relevance of the extrinsic and intrinsic afferents to CA3 and CA1 pyramidal neurons, and evaluated the distinct input patterns to the intrinsic hippocampal circuit. Results confirm that the acquisition of a classical eyeblink conditioning is a multisynaptic process in which the contribution of each synaptic contact is different in strength, and takes place at different moments across learning. Thus, the precise and timed activation of multiple hippocampal synaptic contacts during classical eyeblink conditioning evokes a specific, dynamic map of functional synaptic states in that circuit.
It is commonly accepted that learning and memory processes are able to evoke more-or-less stable changes in synaptic weights in selected cortical and subcortical neural sites (Bliss and Collingridge 1993; Kandel 2001; Neves et al. 2008; Wang and Morris 2010)—that is, that newly acquired motor and/or cognitive abilities are stored in the brain in the form of functional and/or structural modifications of synaptic efficiency (Ramón y Cajal 1909–1911; Konorski 1948; Hebb 1949). For example, it has been shown that the hippocampal-dependent trace eyeblink conditioning (Weiss et al. 2005; Bangasser and Shors 2007), a form of associative learning, evokes a concomitant change in strength at the hippocampal perforant pathway (PP)–dentate gyrus (DG) (Weisz et al. 1984) and CA3–CA1 (Gruart et al. 2006) synapses of behaving mice. It has also been shown that these changes in synaptic strength share many properties with experimentally evoked long-term potentiation (LTP) (Bliss and Collingridge 1993; Kandel 2001; Gruart et al. 2006; Wang and Morris 2010). However, a still-open question is whether all hippocampal synapses included in its intrinsic circuit, and those involved in its main inputs and outputs, present similar increases in strength and with similar changing rates across the learning process.
The hippocampus seems to participate in many different functions, such as Pavlovian associative learning (Berger et al. 1983; Moyer et al. 1990; McEchron and Disterhoft 1997; Gruart et al. 2006), spatial orientation (Moser et al. 2008), object recognition (Clarke et al. 2010), and other forms of memory acquisition, storage, and retrieval (Bliss and Collingridge 1993; Neves et al. 2008; Wang and Morris 2010). In this regard, it is unresolved whether those different functions are carried out by localized sites within the hippocampal circuits (McHugh et al. 2007; Nakashiba et al. 2008), or are dependent upon the specific, timed activation of the multiple synaptic contacts present in those circuits.
We have addressed these important questions here, studying the evolution in strength of 9 different hippocampal synapses during the hippocampal-dependent trace eyeblink conditioning in behaving mice. For this, animals were chronically implanted with stimulating and recording electrodes at selected sites of the intrinsic hippocampal circuit and in the main hippocampal afferent or projecting sites. First, we determined the basic functional properties of the selected synapses—namely, input/output curves, paired-pulse facilitation, and LTP evoked by high-frequency stimulation (HFS) of presynaptic neurons (see Madroñal et al. 2009). For eyeblink conditioning, animals were presented with a tone as a conditioned stimulus (CS) followed 500 ms from its end by an electrical shock of the trigeminal nerve as an unconditioned stimulus (US). CRs were determined from the electromyographic (EMG) activity of the ipsilateral orbicularis oculi muscle. Field excitatory postsynaptic potentials (fEPSPs) were evoked following CS presentations across training sessions in all of the selected synapses. Results indicate that there is a precise and quantifiable activation sequence of the multiple hippocampal synaptic contacts that takes place during the acquisition and extinction of a classical eyeblink conditioning, involving a dynamic map of synaptic-learning states.
Material and Methods
Experiments were carried out in C57Bl/6 male adult mice (3–4 months old; 25–30 g) obtained from an official supplier (University of Granada Animal House, Granada, Spain). Upon arrival, animals were housed in separate cages (n = 8 per cage), but they were switched to individual cages after surgery. Mice were kept on a 12-h light/dark cycle with constant ambient temperature (21.5 ± 1 °C) and humidity (55 ± 8%). Food and water were available ad libitum. Experiments were carried out in accordance with the guidelines of the European Union (2003/65/CE) and Spanish regulations (BOE 252/34367–91, 2005) for the use of laboratory animals in chronic studies. All experimental protocols were also approved by the local Ethics Committee.
Animals were prepared for the chronic recording of the EMG activity of the orbicularis oculi muscle and fEPSPs evoked at 9 different hippocampal synapses during classical conditioning of eyelid responses (Figs. 1 and 2). Only animals that reached all behavioral criteria and with proper electrode placement and expected field potential recordings were analyzed. We considered successful those animals that finished the experimental protocols presenting extracellular recordings (i.e., fEPSPs and/or LFPs) that did not deteriorate over time. A total of 10 animals per group (i.e., per selected synapse) were used for classical conditioning and fEPSP recordings. Some additional animals (n = 16) were used in a preliminary study to determine the most appropriate recording and stimulating sites (Fig. 3). Finally, 10 animals/selected synapse were used for input/output curves, paired-pulse facilitation, and the LTP study (Fig. 4). In accordance, a grand total of 196 mice were used in this study.
Animals were anesthetized with 0.8–1.5% isoflurane delivered via a mouse anesthesia mask (David Kopf Instruments, Tujunga, CA, USA). The anesthetic gas was supplied from a calibrated Fluotec 5 (Fluotec-Ohmeda, Tewksbury, MA, USA) vaporizer, at a flow rate of 1–2 L/min oxygen (AstraZeneca, Madrid, Spain). The 9 synapses selected for this study were the following: PP–DG, DG–CA3, PP–CA3, CA3–CA1, contralateral CA3 (cCA3)–CA1, thalamic reuniens nucleus (REU)–CA1, PP–CA1, CA1–subiculum (SUB), and CA1–medial prefrontal cortex (mPFC). As shown in the diagrams included in Figure 2A, animals were implanted with bipolar stimulating and recording electrodes at the selected sites, following the corresponding stereotaxic coordinates (Paxinos and Franklin 2001). Electrodes were made from 50 µm, Teflon-coated, tungsten wire (Advent Research, Eynsham, UK). A bare silver wire was affixed to the occipital bone as ground. All the implanted wires were soldered to a 6-pin socket (RS Amidata, Madrid, Spain) and were then fixed to the skull with dental cement (Fig. 1A). Further details of this chronic preparation are indicated elsewhere (Gruart et al. 2006).
Recording and Stimulation Procedures
EMG recordings were carried out with the help of Grass P511 differential amplifiers within a bandwidth of 0.1 Hz–10 kHz (Grass-Telefactor, West Warwick, RI, USA). Field EPSP recordings were carried out with a high-impedance probe (2 × 1012Ω, 10 pF). Electrodes were surgically implanted in the selected hippocampal and prefrontal areas using as a guide the field-potential depth profile evoked by paired (40 ms of interval) pulses presented at the corresponding synaptic input (Fig. 3). The recording electrode was fixed at the site where a reliable monosynaptic fEPSP was recorded (Fig. 2B). Synaptic field potentials were evoked in the 9 selected synapses during habituation, conditioning, and extinction sessions by paired pulses (100 µs in duration, 40 ms of interpulse interval) applied to Schaffer collaterals 300 ms after CS presentation. Stimulus intensities ranged from 50 to 350 µA. For each animal, the stimulus intensity was set usually at 30–40% of the intensity necessary to evoke a maximum fEPSP response (Gureviciene et al. 2004, Fig. 4A). An additional criterion for selecting stimulus intensity was that the second stimulus, presented 40 ms after the conditioning pulse, evoked a larger (>20%) synaptic field potential (Bliss and Gardner-Medwin 1973, Fig. 1B,D).
For input/output curves (Fig. 4A), animals were stimulated with single pulses at increasing intensities (0.02–0.4 mA, in steps of 0.02 mA). We also checked the effects of paired pulses at different (10, 20, 40, 100, 200, and 500 ms) interstimulus intervals while using intensities corresponding to 40% of the amount necessary to evoke a saturating response (Fig. 4B).
For evoking LTP, each animal was presented with five 200 Hz, 100 ms trains of pulses at a rate of 1/s. This HFS protocol was presented 6 times in total, at intervals of 1 min. The 100 µs, square, biphasic pulses used to evoke LTP were applied at the same intensity used for the single pulse presented during baseline and post-HFS recordings. Baseline records were collected at a rate of 3/min for the 15 min preceding HFS presentation. After the HFS protocol, recordings at the same stimulus intensity and frequency were collected for 2 h, and then for 15 min on each of the 2 following days (Fig. 4C).
Classical Eyeblink Conditioning
For recordings, 2 animals at a time were placed in separate small (5 × 5 × 10 cm) plastic chambers located inside a Faraday box (30 × 30 × 20 cm). Classical conditioning was achieved using a trace paradigm (Fig. 1B). For this, a tone (20 ms, 2.4 kHz, 85 dB) was presented to the animals as a CS. The US consisted of a 500 µs, 3× threshold, square, cathodal pulse. The US started 500 ms after the end of the CS. A total of 2 habituation, 10 conditioning, and 5 extinction sessions were carried out for each animal (Fig. 1C). A conditioning session consisted of 66 CS–US presentations, and lasted ≈30 min. In 10% of cases, the CS was presented alone. CS–US presentations were separated at random by 30 ± 5 s. For habituation and extinction sessions, only the CS was presented, also for 60 times per session at intervals of 30 ± 5 s. As criterion, we considered a “CR” the presence, during the CS–US period, of EMG activity lasting >10 ms and initiated >50 ms after the CS onset. In addition, the integrated EMG activity recorded during the CS–US interval had to be at least 2.5 times greater than the averaged activity recorded immediately before CS presentation (Gruart et al. 2006).
At the end of the experiments, mice were deeply re-anesthetized (sodium pentobarbital, 50 mg/kg) and perfused transcardially with saline and 4% phosphate-buffered paraformaldehyde. Selected sections (50 µm) including the implanted wires were mounted on gelatinized glass slides and stained using the Nissl technique with 0.1% Toluidine Blue, to determine the proper location of stimulating and recording electrodes (Fig. 2B).
Data Collection and Analysis
EMG and hippocampal activity, and 1V rectangular pulses corresponding to CS and US presentations, were stored digitally on a computer through an analog/digital converter (CED 1401 Plus; Cambridge Electronic Design, Cambridge, UK), at a sampling frequency of 11–22 kHz and an amplitude resolution of 12 bits. Commercial computer programs (Spike 2 and SIGAVG, from Cambridge Electronic Design) were modified to represent EMG and extracellular synaptic field potential recordings. Data were analyzed off-line for quantification of CRs and the fEPSP slopes (Gruart et al. 2006). The slope of evoked fEPSPs was collected as the ratio between the difference of amplitudes (in mV) and the corresponding difference of times (in ms) for the selected points from fEPSP recordings (i.e., mV/ms). The relatively different evolution of fEPSP slopes evoked by the 2 pulses presented during the CS–US intervals was calculated as a second/first × 100 ratio (see Madroñal et al. 2009).
Only data from successful animals (i.e., those that allowed a complete study with a proper functioning of both recording and stimulating systems) were computed and analyzed. All the analyses were developed with the help of homemade quantification and representation programs. The computational tools, the analytical approach of synaptic-learning-state functions, and the timing–strength evolution indices proposed here (equations 1–3) were designed and developed with the help of modified versions of the MATLAB (The MathWorks, Natick, MA, USA) functions, as well as, by means of new applications (see Sánchez-Campusano et al. 2009; 2011; 2012) and customized scripts (Figs 5–8).
Computed results were processed for statistical analysis using the Sigma Plot 11.0 package (Sigma Plot, San Jose, CA, USA) and the MATLAB Toolbox (“Multivariate Statistics” and “Circular Statistics” [Berens 2009] tools) for Windows. For multivariate statistics assessments, both parametric (analysis of variance [ANOVA] F-tests, with or without repeated-measures [RM]) and nonparametric (ANOVA tests on Ranks, with RM [Friedman RM ANOVA] or without RM [Kruskal–Wallis ANOVA]) methods were used to assess the statistical significance of differences among groups, followed by the appropriate test (Holm-Sidak, or Tukey, or Student–Newman–Keuls, in this order of priority) for all the pairwise multiple-comparison analyses.
When the normality (Shapiro–Wilk, or Kolmogorov–Smirnov test) and equal variance of the errors (Levene Median test) assumptions were satisfied, the statistic with its corresponding orders m (number of groups), n (number of animals), and l (number of multivariate observations), was reported (Sánchez-Campusano et al. 2009). For ANOVA F-test with/without RM, the first number (m−1) represents the variability between groups (i.e., the variability due to the differences among the column means) and determines the numerator degree of freedom in the F-distribution table. The second and the third numbers in the bracket following the F statistic represent the variability within groups (i.e., the variability due to the differences between the data in each column and the column mean). For ANOVA F-test with RM the second number (m− 1) × (n− 1) determines the denominator degree of freedom in the F-distribution table, while the third number informs about the number of multivariate observations. For ANOVA F-test without RM, the second number informs about the number of experimental subjects, while the third number (l− m) determines the denominator degree of freedom in the F-distribution table (Hogg and Ledolter 1987). The factors related to the statistical model proposed here were the training phase and days, corresponding to 1-way or 2-way ANOVA F-tests.
When the normality assumption was not verified, the significance (P value) of χ2 statistic was calculated using the ranks of the data rather than their numeric values. Ranks are found by ordering the data from smallest to largest across all groups, and taking the numeric index of this ordering. The rank for a tied observation is equal to the average rank of all observations tied with it. Note that ANOVA test on Ranks is a nonparametric version of the classical ANOVA F-test, and an extension of the Wilcoxon rank sum test to >2 groups. Thus, the usual F statistic is replaced by a χ2 statistic (Hollander and Wolfe 1999). Finally, the parametric and nonparametric methods were also applied to the data taking in to account the sessions/days as repeated measures.
For the circular statistics inferences (Batschelet 1981; Mardia and Jupp 1999; Berens 2009; Sánchez-Campusano et al. 2011), we used both the Rayleigh and the Watson hypothesis tests, in addition to the von Mises distribution (the circular analog of the normal distribution), for the probability density of the circular mean of the time data and the corresponding circular dispersion indices: for the simultaneous timing–strength characterization of each synapse (see eq. 1, Figs 6A and 7D, and Supplementary Table 1 and Supplementary Material, Appendix A) and relative dispersion index (RDI) for the relative dispersion patterns between different synapses (see eq. 2, Figs 6A and 7D, and Supplementary Table 2 and Supplementary Material, Appendix A). Regression analysis was used to study the evolutions of the percentage of CRs (Figs 1C, 5B, and 7C) and the fEPSP slopes (Figs 1D, 5A, and 7A,B) across conditioning, and for the trend of the synaptic evolution index (SEI) (see eq. 3, Figs 6B,C, and 8, and Supplementary Table 3 and Supplementary Material, Appendix B).
In general, for all the statistical tests (multivariate, regression, and circular statistics), the significance level (P) was indicated. It is common to declare a result significant if the P value is <0.05 (*), 0.01 (**), or 0.001 (***). In particular, for each regression test, the correlation coefficient (r), and the parameters and equation of the best polynomial/sigmoidal fit, were reported. Unless otherwise indicated, data are represented by the mean ± standard error of mean (SEM). The different timing–strength evolution indices proposed in this work are defined below:
Relative dispersion index RDIi1,i2|r between the circular timing–strength patterns of 2 different (i1 and i2) synapses or between the i1th-synapse and the rth-circular pattern for the level of expression of the CRs:
SEI (in the same unit of measurement as the fEPSP slope —i.e., mV/ms, or the % of baseline) was designed to estimate the changes between different synaptic-learning states of the same synapse or to measure the dynamic changes between different synaptic-learning states of different synapses:
In equations 1 and 2, is the circular kurtosis and is the radius of the circumference that describes the centroid with respect to the origin. In equation 3, represent the synaptic strengths (fEPSP slopes, as % of baseline) at the synaptic-learning states respectively. Notice that the evolution index SEI measures the simultaneous changes in the 2 physiological variables of each synapse—that is, the timing T(j) and the synaptic strength Si(j), as well as the relative dynamic evolution between different synapses at different learning states. For further details about these analytical approaches, the readers may refer to Supplementary Material (see Supplementary Tables 1–3, and Appendices A and B).
Basic Functional Properties of Hippocampal Synapses Included in This Study
Mice were prepared for the chronic recording of fEPSPs evoked at the 9 selected synapses belonging to the hippocampal intrinsic circuit or involved in its main inputs and outputs (Fig 1A) during classical conditioning of eyelid responses. The final location of recording and stimulating sites was checked with histological procedures at the end of the recording sessions (Fig. 2). Nevertheless, during surgery, the electrode location in the selected recording or stimulating sites was decided with the help of field-potential depth profiles collected in a preliminary set of experiments (Fig. 3).
In a first series of experiments (n = 10 mice/group) we checked the general functional properties of the 9 selected synapses. Input/output curves presented increased fEPSP slopes with increasing intensities (in mA), although some synapses (REU–CA1, CA1–SUB, and CA1–mPFC) reached lower asymptotic values than the others (Fig. 4A), indicating some important functional differences between them. In addition, perforant pathway inputs evoked differential effects on their hippocampal targets. Specifically, input/output relationships were steeper for the PP–DG synapse than those evoked at the PP–CA3 and PP–CA1 synapses, a fact already reported in anesthetized guinea pigs (Bartesaghi et al. 2006).
We also checked the paired-pulse facilitation present in the selected synapses. Paired-pulse stimulation is a form of short-term synaptic modulation used as an indirect measurement of changes in the probability of release of neurotransmitter at the presynaptic terminal (Thomson 2000; Zucker and Regehr 2002). According to the multiple-comparison analyses, most of the selected synapses presented a significant (P < 0.05) paired-pulse facilitation at short intervals (10–40 ms), but some of them extended the facilitation to 100 ms (DG–CA3, PP–CA3, cCA3–CA1, and CA1–mPFC), or even 200 ms (REU–CA1) of interpulse interval (Fig. 4B).
Finally, an LTP was evoked in all of the synapses by HFS of the corresponding presynaptic terminal (see Materials and Methods). When evoked in behaving mice, most LTPs presented profiles similar to those already described for the CA3–CA1 synapse in the same species (Gruart et al. 2006; Madroñal et al. 2007). However, and as reported recently (Jurado-Parras et al. 2012), the CA1–mPFC synapse presented a significant (P < 0.05) delayed building up of the expected synaptic potentiation, a fact also reported recently in behaving mice (Eleore et al. 2011). Finally, the REU–CA1 synapse showed an evident and significant (P < 0.05) long-term depression in response to the presentation of the HFS protocol (Fig. 4C). On the whole, these results suggested the presence of different functional properties of the selected synapses that could make them act in diverse, but complementary, ways during the acquisition of actual associative learning tasks.
Differential Changes in Strength Evoked in the Selected Hippocampal Synapses by Classical Eyeblink Conditioning of Behaving Mice
In a second series of experiments, animals were prepared for the classical conditioning of eyelid responses using a hippocampal-dependent trace conditioning paradigm (Fig. 1A,B; Weiss et al. 2005; Bangasser and Shors 2007). Mice (n = 10 per group) acquired this type of associative learning progressively, reaching 50% of CRs by the fourth conditioning session and reaching top values (75% of CRs) by the 10th (Fig. 1C). In parallel with the increase in the percentage of CRs, and as already reported for this synapse (Gruart et al. 2006), we were able to record a learning-dependent change in fEPSP slopes evoked by a pair of pulses presented to the CA3–CA1 synapse during the CS–US interval (Fig. 1B,D). It has already been shown that this synaptic potentiation is significantly related with the increase in the percentage of CRs (Gruart et al. 2006).
We used the same conditioning paradigm to check the evolution of activity-dependent changes in strength for the 9 synapses included in this study. As illustrated in Figure 5A,B, changes in synaptic strength reached maximum (or minimum) values at different times and rates. Although most synapses presented a progressive, sustained increase in fEPSP slopes, others presented an early potentiation that also disappeared very soon (DG–CA3), were progressively depressed (CA1–mPFC), or were not affected by the training (PP–CA1).
The appropriate statistical analysis (combination of parametric multivariate methods [ANOVA F-test] and multiple-comparison analyses [Tukey's test]) carried out with the collected data indicated that some synapses (PP–DG, DG–CA3, CA3–CA1) belonging to the intrinsic hippocampal circuit (Amaral 1993) presented maximum synaptic changes significantly [F3,27,36 = 16.51; P < 0.001] different (in both strength and session presentation) from those evoked by the CA1–mPFC synapse (Fig. 5D, left histogram).
We also observed important differences in the changes evoked at the 2 hippocampal output pathways (CA1–SUB and CA1–mPFC) included in this study (Fig. 5A,D). Thus, one-way ANOVA F-test with factors (days) revealed that fEPSPs evoked at the subiculum (CA1–SUB) were potentiated [F1,9,18 = 30.37; P < 0.001] during the conditioning sessions relative to baseline, whilst effects on the prefrontal cortex (CA1–mPFC) were depressed [F4,36,45 = 11.70; P < 0.001] in comparison with the maximum synaptic strengths in other synapses (intrinsic hippocampal circuit and CA1–SUB).
A similar statistical approach (Fig. 5D, right histogram) showed that the PP–CA1 synapse did not follow the changes in strength [F4,36,45 = 6.22; P < 0.001] taking place in the other intrinsic synapses (PP–CA3, CA3–CA1, cCA3–CA1, and REU–CA1) of the hippocampal circuit. Finally, 4 synapses (PP–DG, CA3–CA1, cCA3–CA1, and CA1–SUB) presented similar synaptic strengths [F4,36,45 = 1.86; P > 0.05] at the asymptotic level (10th conditioning session) of the acquisition process.
Different Timing–Strength Evolution Patterns Evoked at the Selected Synapses During Classical Eyeblink Conditioning
As illustrated in Figure 5A, the classical conditioning of eyelid responses carried out in the experimental animals not only evoked specific changes in strength in the 9 selected synapses, but also those changes took place at precise moments across the successive training sessions. In accordance, we decided to analyze both processes simultaneously. Thus, we also analyzed the timing–strength dispersion pattern (Batschelet 1981; Mardia and Jupp 1999; Sánchez-Campusano et al 2011) observed between the mean and maximum synaptic strength values across conditioning (Fig. 6A). In this analytical approach, the normalized fEPSP slope determined the strength/magnitude of the vector (red or green arrow in Fig. 6A), while the training session determined the timing/orientation of the vector.
According to the data illustrated in Figure 6A, 3 synapses showed a dispersion pattern similar [F3,27,36 = 3.68; P > 0.01] to that of eyelid activity during conditioning sessions (% CRs, 100%; PP–DG, 96%; CA3–CA1, 121%; and CA1–SUB, 112%), whilst other synapses (cCA3–CA1; REU–CA1; PP–CA1) presented smaller [<32%; F3,27,36 = 341.91; P < 0.001] timing–strength dispersion patterns, suggesting a lower degree of correlation with the acquisition process (see Supplementary Tables 1 and 2 and Supplementary Material, Appendix A). Finally, 3 synapses (DG–CA3, PP–CA3, and CA1–mPFC) presented the smallest values [<13%; F3,27,36 = 415.68; P < 0.001] of the dispersion indices, with angular displacements in opposite directions (see blue bent arrows in Fig. 6A).
The SEI was designed to estimate the changes between different synaptic-learning states of the same synapse or to measure the dynamic changes between different synaptic-learning states of different synapses. In particular, for each synapse, the analysis of the SEI between the mean and maximum values of the synaptic strength and their corresponding times of occurrence (see Supplementary Table 3 and Supplementary Material, Appendix B) enabled us to propose an alternative distribution of the synapses (in this order: 2, DG–CA3; 3, PP–CA3; 4, CA3–CA1; 8, CA1–SUB; 1, PP–DG; 5, cCA3–CA1; 6, REU–CA1) (see Fig. 6B,C), and a hypothetical model of input–output physiological connections at the hippocampal circuit (see the diagram of connections in Fig. 6C) while the animal learns the associative learning task.
For comparative purposes, in Figure 7 are illustrated the different changes in synaptic strength evoked by the second of the pair of pulses presented to each of the selected synapses during the training sessions, in comparison with the effects evoked by the first (fEPSPs 1 illustrated in Fig. 7A; and fEPSPs 2 illustrated in Fig. 7B). Changes evoked by the second pulse evolved across training in a way similar to those evoked by the first one. Thus, some synapses increased in learning-dependent strength across training (PP–DG, CA3–CA1, cCA3–CA1, REU–CA1, and CA1–SUB); other synapses increased their synaptic strengths at the beginning of the conditioning sessions and decreased at the end (DG–CA3 and PP–CA3); others did not present any significant change in this type of associative learning (PP–CA1); and, finally, others decreased in their synaptic strength (CA1–mPFC). Although the general evolution of changes in synaptic strength was similar for the fEPSPs evoked by the 2 pulses, some differences (≥25%; P ≤ 0.05) in intensity (PP–DG, PP–CA3, CA3–CA1, and CA1–SUB) were also evident (Fig. 7A,B). In addition, some minor differences in the timing–strength dispersion pattern were observed in 2 of the synapses (PP–CA3 and PP–CA1) for the evolution of fEPSPs evoked by the first (Fig. 6A) and the second (Fig. 7D) pulses across learning.
In accordance with an early report (Madroñal et al. 2009) paired-pulse facilitation (i.e., the relative changes in strength of the second vs. the first pulse) was significantly modified (P ≤ 0.05) in all of the studied synapses (apart the PP–CA1 one) across learning and extinction (not illustrated). As a whole, and as already proposed for many other synaptic circuits (see Thomson 2000; Zucker and Regehr 2002 for details), the different synaptic effects evoked by the pair of pulses suggest that some presynaptic mechanisms are involved in classical eyeblink conditioning. In this regard, it has been already proposed the involvement of presynaptic components of the CA3–CA1 synapse in the same type of associative learning in mice (Madroñal et al. 2009).
The analysis of the timing–strength synaptic evolutions across conditioning (Figs 5–8), including the calculation of the dispersion patterns ( and RDI) (see equations 1 and 2, and Supplementary Tables 1 and 2) and the SEI (see eq. 3 and Supplementary Table 3), allowed us to propose the putative functional organization of the 9 synapses included in the present study, which did not coincide with their sequential organization according to anatomical criteria and connectivity (Figs 5C and 6B,C; Amaral 1993). The picture emerging (Figs 5D, 6C, and 8) is that following an early, short-lasting activation of the DG–CA3 synapse, there is an increasing process of activation of the intrinsic hippocampal circuit (CA3–CA1), including one of the hippocampal main outputs (CA1–SUB) and its contralateral counterpart (cCA3–CA1). Perforant pathway inputs (PP–CA3 and PP–DG synapses) are activated in a delayed form, a fact that has also been reported during a tone-food association test carried out in behaving rats (Segal and Olds 1972; Segal et al. 1972), whilst the PP–CA1 synapse seems not to participate actively in this type of associative learning (Figs 5A,D, 6C, and 8C). Interestingly, whereas synaptic projections to the subiculum (CA1–SUB) are potentiated, those to the medial prefrontal cortex (CA1–mPFC) are disfacilitated (Figs 5A,D and 6C).
The functional and dynamic approach used here to better understand the plastic changes taking place in 9 selected hippocampal synapses during the actual acquisition and extinction of a classical eyeblink conditioning task allows us to propose the hippocampus as a neuronal structure enabling the specific, timed combinations in synaptic changes in strength characterizing this type of associative learning (Berger et al. 1983; Moyer et al. 1990; McEchron and Disterhoft 1997; Weiss et al. 2005; Gruart et al. 2006; Bangasser and Shors 2007; Madroñal et al. 2007). It can be assumed that other cognitive functions carried out by hippocampal networks, such as spatial orientation, object recognition, and memory storage and recall (Moser et al. 2008; Neves et al. 2008; Clarke et al. 2010; Wang and Morris 2010), are carried out by spatial-temporal patterns of synaptic activities different to the ones described here for classical eyeblink conditioning.
The collected results have enabled us to evaluate the trends of the proposed SEI (see eq. 3, Fig. 8A–C, and Supplementary Table 3) and to define a 3D learning-dependent array (Fig. 8D) characterizing the contribution of the hippocampal synapses (included in its intrinsic circuit and those involved in its main inputs and outputs) to trace eyeblink conditioning. In this respect, it is very important to consider that the evolution index SEI measures the simultaneous changes in the 2 physiological variables of each synapse—that is, the timing and the synaptic strength, as well as the relative dynamic evolution between different synapses at different learning states. In general, the 3 indices defined here (, eq. 1; RDI, eq. 2; and SEI, eq. 3) enable a better determination of the relationships between different functional states—that is, between different synaptic-learning states involved in the acquisition of an associative learning task. The present results strongly suggest that this type of associative learning is a multisynaptic process in which the contribution of each synaptic contact is different in strength and takes place at a different acquisition moment across the conditioning process.
The significance of the dynamic patterns set up by the intrinsic and/or extrinsic afferents to both CA3 and CA1 networks was also established. Synaptic terminals impinging upon CA3 postsynaptic pyramidal neurons (DG–CA3 and PP–CA3) presented similar activation profiles (Fig. 5A,D) across conditioning; but quite different circular means of time distributions (Fig. 6A) and different mean values of the SEI (Fig. 8 and Supplementary Table 3). These differing data suggest the need for 2 distinct input systems to the hippocampal CA3 network while the animal learns this classical conditioning task. The computational relevance of these 2 afferents to CA3 pyramidal neurons has been suggested by other authors (Treves and Rolls 1992; Treves 2004; Rolls and Kesner 2006) and supported by the present results. We found that the synapses from the dentate granular cells onto CA3 neurons (DG–CA3 synapse) via mossy fibers (first input system) presented an early and significant potentiation (Fig. 5A,D) and very high absolute values of the SEI (Fig. 8 and Supplementary Table 3) in order to force, at the beginning of the learning process, the CA3 cells into a pattern of activity relatively independent of any input being received from CA3 recurrent collaterals.
Interestingly, at the precise moment that the synaptic strength of the DG–CA3 synapse began to decrease, the PP–DG synapse presented a progressive, sustained increase in fEPSP slopes (Fig. 5A,D). In fact, the effects of these dynamic changes were reflected as different dispersion patterns (; Figure 6A and Supplementary Tables 1 and 2) and very different values of the proposed evolution index (SEI) (Fig. 8 and Supplementary Table 3). In this respect, we can propose that following an early, short-lasting activation of the DG–CA3 synapse, there is an increasing process of activation of the PP–DG synapse in order to regulate the intensity of the DG–CA3 synapse and to relay enough specific signals to initiate the retrieval process, via perforant path input (second input system) to CA3 pyramidal cells (PP–CA3 synapse, see Fig. 8D). In this regard, it has already been shown that single neuron changes in CA3 are not directly dependent upon those in DG, and that the sequence of changes taking place in those neurons does not directly confirm the classical intrinsic hippocampal circuit (Segal and Olds 1972; Segal et al. 1972). In addition, it was shown years ago that the PP–DG synaptic activity is potentiated during nictitating membrane response conditioning in rabbits (Weisz et al. 1984).
According to this dynamic approach, synapses impinging upon the same set of CA1 postsynaptic pyramidal cells (PP–CA1, CA3–CA1, cCA3–CA1, and REU–CA1) seemed to present different activation profiles (Fig. 5A,D) at the asymptotic level (10th conditioning session) of the acquisition process. Additionally, these synapses have different dispersion patterns () (Fig. 6A and Supplementary Table 1) and different values of the SEI (Fig. 8 and Supplementary Table 3). This evidence allows us to suggest the need for at least 3 different operant input systems (from CA3, cCA3, and REU) to the hippocampal CA1 network (see diagram in Fig. 8D) during the acquisition of the trace eyeblink conditioning. The information relayed by such different afferent connections determines the timing and intensity of the fEPSPs evoked in CA1 neurons. Thus, we can propose that whereas CA3 cells project both to CA1 via the Schaeffer collaterals (first input system, CA3–CA1 synapse) and to other CA3 cells via recurrent collaterals, those from the contralateral counterpart (second input system) project to CA1 pyramidal cells (cCA3–CA1 synapse) with maximum potentiation (Fig. 5A,D), as well as with an extensive dispersion pattern () (Fig. 6A, and Supplementary Tables 1 and 2) and the highest values of the SEI (Fig. 8 and Supplementary Table 3). In contrast, the synapse formed by the extrinsic afferents from the thalamic reuniens nucleus (third input system) to CA1 neurons (REU–CA1) presented a lower activation profile than the cCA3–CA1 synapse (Figs 5A,D and 6C).
Furthermore, and following the dynamic approach developed here, the dispersion patterns of both REU–CA1 and CA3–CA1 synapses with respect to the level of CR expression were significantly different (; Supplementary Tables 1 and 2); and the calculation of the synaptic evolution index between REU–CA1 and cCA3–CA1 synapses indicates that there are important differences (SEI; Figs 5D and 8) in the progressive potentiation of these synapses. The absence of a definite LTP (Eleore et al. 2011), or even the presence of a long-term depression, instead of the expected LTP, in the REU–CA1 synapse when activated by the HFS protocol, is coherent with its late activation during the extinction sessions (see Fig. 5A). In this regard, the existence has been proposed of a closed REU–CA1 circuit that allows the reuniens nucleus to modulate the activity in CA1, depending on hippocampal output (Dolleman-Van der Weel et al. 1997).
Whereas the CA1–SUB synapse is potentiated, the synaptic projections to the medial prefrontal cortex (CA1–mPFC) are disfacilitated, and the differences between them were highly significant (Fig. 5A,D). These differences were reflected directly in the dispersion patterns () (Fig. 6A, and Supplementary Tables 1 and 2) and in the SEI values (Fig. 8 and Supplementary Table 3). The facilitation of learning-related commands sent to the subiculum could be related to the diffusion across sensory-motor circuits of the increased CS–US associative strength in order to generate the corresponding overt motor responses (i.e. the CRs), whilst the disfacilitation of medial prefrontal circuits could contribute to the proper release of the newly formed CR. As recently shown (Alexander and Brown 2011; Leal-Campanario et al. 2013), the mPFC needs to be inhibited in order for the CR to occur. As already reported Jurado-Parras et al. (2012), and confirmed here, the small, delayed LTP observed in the CA1–mPFC pathway is indicative of the peculiar role played by this synapse in the release of newly acquired motor responses (Alexander and Brown 2011; Leal-Campanario et al. 2013).
Finally, the PP–CA1 synapse seems to play a specific role (Izumi and Zorumski 2008) erasing unwanted memories, and is thus not involved in this type of associative learning (see SEI values in Fig. 8 and Supplementary Table 3). In addition, and as already suggested for the hippocampal CA3–CA1 synapse (Madroñal et al. 2009), collected data suggest the presence of some presynaptic mechanisms, at selected synapses of the hippocampal circuitry, in the acquisition of this type of associative learning.
As already shown for the CA3–CA1 synapse during classical eyeblink conditioning using a trace paradigm (Gruart et al. 2006), significant changes in synaptic strength also took place in some of the selected hippocampal synapses during the extinction process (Figs 5 and 7). These results further confirm that extinction is not a mere passive process during which acquired memories are erased. To the contrary, extinction seems to be an active process involving the activation of selective synaptic patterns and underlying molecular events (Inda et al. 2005).
In summary, we can propose that the results obtained with the dynamic approach presented here for determining the functional organization of hippocampal circuits during actual learning are incompatible with the excessive localization of hippocampal functions proposed by others (McHugh et al. 2007; Nakashiba et al. 2008). Thus, depending on the specific, timed activation of its multiple synaptic contacts, the hippocampus can be involved in many different functions, such as spatial orientation (Moser et al. 2008), object recognition (Clarke et al. 2010), and various other forms of memory acquisition and retrieval (Bliss and Collingridge 1993; Neves et al. 2008; Wang and Morris 2010). It can therefore be proposed that for each associative and/or cognitive function carried out by the hippocampal circuit there is a corresponding functional map of different synaptic weights characterizing it. In future studies, it will be important to systematically manipulate each of individual components of the hippocampal circuits, piece-by-piece, using powerful genetic approaches, to reveal their respective roles in learning and memory processes.
This study was supported by Spanish MINECO (BFU2011-29089 and BFU2011-29286) and Junta de Andalucia (BIO122 and CVI7222) grants to A.G. and J.M.D.-G.
The authors thank J.M. González-Martín for his technical assistance and Roger Churchill for his editorial help. Conflict of Interest: None declared.