INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The direct perforant path (PP) projection from the entorhinal cortex is a major route of cortical input to hippocampal area CA3, in addition to the disynaptic projection via the mossy fibers (MF). Although sequential processing via the latter projection has been emphasized in the well-known trisynaptic model of hippocampal circuitry (Andersen et al. 1971), parallel processing through the direct projection may be at least as influential in driving CA3 neurons to fire (Amaral 1993). Anatomical evidence indicates that the entorhinal cortex sends a major projection to all fields of the hippocampus proper as well as to the dentate gyrus (Desmond et al. 1994; Steward 1976). The direct PP projection terminates on the distalmost portions of the apical dendrites of pyramidal cells in CA3. Physiological evidence indicates that in spite of its distal termination site, PP input is capable of driving CA3 pyramidal cells to fire (Urban et al. 1998; Yeckel and Berger 1990). Recent computational models have reflected this revised appreciation of the efficacy of the direct cortical projection to the hippocampus proper (O'Reilly and McClelland 1994; Treves and Rolls 1992; Wallenstein et al. 1998). Most theoretical accounts follow the suggestion that, by virtue of its dense network of recurrent collateral fibers (RC), area CA3 is anatomically and physiologically well suited to perform autoassociative memory functions (McClelland et al. 1995; Rolls 1996; Wallenstein et al. 1998).

Formal analyses of the storage capacity of autoassociative memory networks have demonstrated that a fundamental tradeoff exists between the requirements for optimal efficiency in pattern storage and retrieval (O'Reilly and McClelland 1994; Treves and Rolls 1992). Storage of a new memory is accomplished optimally by a strong but sparsely connected input pathway. The reason for this constraint is that such an input can ensure that a memory trace is encoded with a high degree of fidelity by forcing the network into an activation state that is highly correlated with the activity pattern to be stored. Conversely, a weak, diffusely connected input is optimal for pattern retrieval, because during recall, the input pathway carries a retrieval cue that is itself only weakly correlated with the pattern to be recalled. A retrieval cue that is weakly imposed allows the internally stored pattern to dominate the activity state of the network once recall has been initiated. Thus the capacity of an autoassociative network is fundamentally limited if both storage and recall are mediated by a single input pathway.

Hasselmo and colleagues have suggested that this problem might be solved by the differential effects of neuromodulators on extrinsic and intrinsic hippocampal pathways (Wallenstein et al. 1998). An alternative solution was proposed independently by two groups (O'Reilly and McClelland 1994; Treves and Rolls 1992), both of whom suggested that MF and PP are ideal for mediating memory storage and retrieval, respectively. According to this model, the existence of a dual pathway to area CA3 is seen as avoiding the tradeoff in optimal parameters. Treves and Rolls (1992) further demonstrated that the efficiency of a retrieval pathway is greatly increased if it possesses associatively modifiable synapses and predicted on that basis that the direct PP input to CA3 can undergo associative long-term potentiation (LTP). Here we demonstrate that both MF and RC activity can contribute associatively to LTP at the PP synapse in CA3. We further describe and characterize a form of transient depression of the PP synapse induced by stimulation of heterosynaptic pathways in CA3.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Slice preparation

Transverse hippocampal slices were prepared from 24- to 35-day-old male Sprague-Dawley rats as described previously (Urban and Barrionuevo 1996). Briefly, animals were deeply anesthetized with equitheisin and perfused transcardially with an ice-cold modified artificial cerebral spinal fluid [ACSF; concentrations as follows (in mM): 229.0 sucrose, 1.9 KCl, 1.2 Na₂PO₄ · 7 H₂O, 33.3 NaHCO₃, 6.0 MgCl₂, and 10.0 dextrose bubbled with a 95% O₂-5% CO₂ gas mixture, pH 7.4]. The hippocampus was dissected out, and 500-µm transverse slices were cut on a sliding-stage vibratome (Leica VT1000S, Leica Instruments, Nussloch, Germany). The slices were transferred to an incubation chamber with standard ACSF [concentrations as follows (in mM): 125.0 NaCl, 2.5 KCl, 1.2 Na₂PO₄ · 7 H₂O, 33.3 NaHCO₃, 1.0 CaCl₂, 6.0 MgCl₂, and 10.0 dextrose bubbled with a 95% O₂-5% CO₂ gas mixture, pH 7.4] and maintained at room temperature until used 2-8 h after slicing.

Stimulation and recording techniques

Slices were transferred to a submersion chamber and maintained at 31-33°C. The recording ACSF was identical to the incubation media, except that 10 µM bicuculline methiodide was included in the recording medium to isolate a purely excitatory response; additionally, 3.0 mM CaCl₂ and MgCl₂ were used to reduce excitability in the disinhibited slice. In some experiments (as noted), 50 µM D,L-2-amino-5-phosphonovaleric acid (APV), an N-methyl-D-aspartate receptor (NMDAR) antagonist, or 10 µM naloxone, an opioid receptor antagonist, was temporarily introduced into the recording ACSF. Synaptic responses were evoked by bulk stimulation through a bipolar NiChrome stimulating electrode (18-µm diameter). Extracellular field potentials were recorded through glass electrodes (1-3 M) filled with 150 mM NaCl, amplified 100× or 1,000× by a Dagan BVC-700 amplifier (Dagan, Minneapolis, MN), and stored on a Pentium II PC with custom-made data acquisition software written with LabView (National Instruments, Austin, TX).

PP field excitatory postsynaptic potentials (fEPSPs) evoked by a stimulating electrode placed near the hippocampal fissure were recorded in the stratum lacunosum-moleculare (Berzhanskaya et al. 1998). MF responses were evoked by a stimulating electrode in the s. granulosum of the dentate gyrus and recorded in the s. lucidum of CA3 (Fig. 1A). MF fEPSPs were identified by the following criteria: latency of 2 ms, rise time of 4 ms, posttetanic potentiation (PTP) on the order of at least a fourfold increase in fEPSP amplitude above baseline, and localization of sink in s. lucidum (Salin et al. 1996). RC fEPSPs were evoked by backfiring Schaffer collaterals through a stimulating electrode in the stratum radiatum of CA1, close to the CA1-CA3 border. Unless otherwise noted, both LTP and transient heterosynaptic depression (THD) were induced by a high-frequency stimulation (HFS) pattern consisting of 15 trains of 20 pulses, with an interpulse interval of 10 ms and an intertrain interval of 10 s (hereafter referred to as the "standard HFS"). The HFS was delivered either to PP (homosynaptic) or to MF or RC (heterosynaptic).

View larger version (26K): [in this window] [in a new window]   Fig. 1. A: stimulation and recording sites in schematic representation of the hippocampal slice preparation. Note that data in all figures are derived from perforant path (PP) field excitatory postsynaptic potential (fEPSP) amplitudes, in some cases after conditioning stimulation was delivered to mossy fibers (MF) or recurrent collateral fibers (RC). B: time-course from a representative experiment demonstrating associative PP long-term potentiation (LTP) induced by a strong MF high-frequency stimulation (HFS) and a weak PP HFS. Top waveform: test MF fEPSPs evoked to localize MF-stimulating electrode. Bottom waveform: PP fEPSPs evoked immediately before and 5 min after HFS. C: averaged time-course from 6 experiments. As a positive control, LTP was induced in all 6 slices 1 h after drug wash-out (note axis break). Top waveform: PP fEPSPs before and after HFS in APV. Bottom waveform: PP fEPSPs before and after HFS after APV wash-out. D: summary of magnitude of potentiation after 4 stimulation conditions: homosynaptic, weak PP stimulation (PPw); associative conditioning (PPw + MF); homosynaptic, strong PP conditioning (PPs); and PPs in the presence of APV (PPs + APV). Scale bar, 0.2 mV, 10 ms (PP waveforms) or 20 ms (MF waveform). *Significant difference from pre-HFS baseline, P < 0.005 by t-test.

Data analysis

For all experiments, the baseline fEPSP amplitude was defined as the mean amplitude of the last 10 responses evoked before delivery of the HFS. For LTP experiments, potentiation was defined as the mean amplitude of the first 10 responses evoked 5 min after the HFS, divided by the mean amplitude of the baseline response. The interval of 5 min after HFS was chosen to allow for stabilization of the response after PTP. For THD experiments, transient depression was defined as the mean amplitude of the first 10 responses evoked immediately after the HFS, divided by the mean amplitude of the baseline response. All values reported are the mean %change from baseline ± SE. Except as noted, statistical significance of potentiation or depression was assessed by comparing the means of the post-HFS intervals to the pre-HFS baselines with the two-tailed, paired Student's t-tests.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

NMDAR dependence of homosynaptic PP LTP

To evaluate the NMDAR dependence of homosynaptic PP LTP in CA3, we applied the standard HFS at suprathreshold intensity (30% peak fEPSP amplitude) in the presence of 50 µM D,L-APV (Fig. 1C). The subsequent fEPSP amplitude was not significantly different from pre-HFS baseline (+10.9 ± 8.1%, P > 0.05; n = 6). In the same slices, we applied the same standard HFS 1 h after APV wash-out. In APV-free ACSF, the PP fEPSP increased significantly above baseline (+54.4 ± 13.6%, P < 0.05; n = 6). We conclude that LTP at the PP to the CA3 synapse is NMDAR dependent.

Associative LTP between MF and PP synapses

To determine whether MF can contribute to associative LTP at the PP synapse in CA3, we stimulated PP at a low stimulation intensity, which evoked a fEPSP of <20% of the maximum attainable amplitude. Delivery of the standard HFS to PP with this low stimulation intensity (15 brief trains, see METHODS) by itself was not sufficient to induce homosynaptic LTP (Fig. 1B). This same stimulation, however, when combined simultaneously with an identical standard HFS delivered to MF, resulted in PP LTP (+56.7 ± 9.7%, P < 0.05; n = 6). MF stimulation by itself did not result in PP LTP, indicating that the potentiation we observed in this experiment was indeed associative (Barrionuevo and Brown 1983) rather than heterosynaptic (Bradler and Barrionuevo 1990). At least 20 min passed between the subthreshold HFS delivered to PP alone and the combined MF + PP HFS. The failure of MF HFS alone to induce PP LTP, shown in Fig. 1B, was not due to saturation of PP capacity to potentiate because in separate experiments, the same standard MF HFS also failed to induce LTP in unpotentiated PP synapses (Fig. 2). Treating the slices with 50 µM D,L-APV before HFS blocked LTP induction (+3.6 ± 22%, P < 0.05; n = 3; data not shown), indicating that associative LTP between MF and PP is NMDAR dependent, as is homosynaptic PP LTP.

View larger version (19K): [in this window] [in a new window]   Fig. 2. Magnitude of PP transient heterosynaptic depression (THD) varies as a function of the number of MF trains. Each arrow represents 5 bursts of MF HFS (20 pulses at 100 Hz delivered every 10 s). Magnitude of depression increases with the number of MF trains. Average depression values (±SE): 5 trains (13 ± 5%; n = 5; A), 10 trains (17 ± 3%; n = 4; B), and 15 trains (40 ± 7%; n = 6; C). Under all conditions, the fEPSP amplitude was indistinguishable from the pre-HFS baseline after 5 min (P > 0.05 by 2-tailed unpaired t-test). Scale bar, 0.2 mV, 10 ms.

Bulk stimulation of MF might be expected to recruit activity of RC axons within CA3. To evaluate the extent to which RC contribute to PP LTP independently of MF activation, we applied the standard HFS to both RC and PP simultaneously. This treatment also resulted in statistically significant enhancement of the PP fEPSP (+32.1 ± 7.9%, P < 0.05; n = 8; data not shown). Although the magnitude of LTP was ostensibly lower in RC + PP cases than in MF + PP cases, the difference did not reach statistical significance (P > 0.1).

Input-output relation of transient heterosynaptic depression induction

As the example in Fig. 1B illustrates, delivering a standard HFS to MF results in a THD of PP fEPSP amplitude. This depression was substantial (>50% of baseline), recovered after ~5 min, and occurred whether HFS of MF was delivered alone or simultaneously with HFS of PP. In the latter case, the depression was superimposed on LTP and occluded the initial PTP (Fig. 1B). We evaluated the input-output relationship of THD by measuring the amplitude and duration of depression after 5-, 10-, and 15-train stimuli. As Fig. 2 indicates, HFS with fewer trains resulted in less depression, ranging from 13 ± 5% for 5 trains (n = 5; Fig. 2A) to 40 ± 7% for 15 trains (n = 6; Fig. 2C). Under all conditions, the fEPSP amplitude returned to pre-HFS baseline level after 5 min, based on a comparison of 10 data points immediately before and 5 min after HFS (P > 0.05 by two-tailed, unpaired t-test).

THD is not blocked by naloxone

The magnitude and duration of PP THD in the preceding experiments were similar to a form of opioid-mediated heterosynaptic depression reported previously in both MF in CA3 and PP in the dentate gyrus (Wagner et al. 1993; Weiskopf et al. 1993). Accordingly, we tested the hypothesis that the THD we observed in the preceding experiments is mediated by opioid peptides. To ensure that the slices used in this experiment were capable of undergoing THD multiple times, we applied the standard HFS in regular ACSF two times [40.3 ± 4.8%, P < 0.05; n = 6; Fig. 3A (only first HFS shown)]. Each HFS was separated by 20 min. After THD was successfully induced twice, we delivered a third HFS in the presence of 10 µM naloxone, 10 min after wash-in of the drug. Naloxone did not prevent the MF HFS from depressing the PP fEPSP, indicating that THD is not mediated by opioid peptides (33.6 ± 3.3%, P < 0.05; n = 6; Fig. 3B).

View larger version (21K): [in this window] [in a new window]   Fig. 3. A: magnitude of THD induced by standard MF HFS in standard recording artificial cerebrospinal fluid (only the first of 2 control HFS is shown.) B: results of a third conditioning pulse in the same slices as in A after wash-in of 10 µM naloxone (nalox). C: PP THD induced by standard HFS delivered to RC, driven by backfiring Schaffer collatorals in stratum radiatum of CA1. D: magnitude of depression induced in each of above conditions. Scale bar: 0.2 mV, 10 ms. *Significant difference from pre-HFS baseline, P < 0.005 by t-test.

THD is not an exclusive effect of MF HFS

MF HFS, even in cases where a relatively uncontaminated monosynaptic fEPSP response to single or paired pulses could be identified, could nonetheless be expected to recruit a substantial disynaptic component of RC synaptic activation during HFS. This possibility is likely, considering the strong impact of MF activation on CA3 pyramidal cells (Claiborne et al. 1993) and the magnitude of MF frequency facilitation (Salin et al. 1996; Urban et al. 2001). This consideration raises the question of whether the THD reported in the previous experiments was necessarily a result of MF activation or whether RC activation could also be sufficient. Although it is not feasible to eliminate RC activation after HFS of MF, we did evoke RC responses in CA3 by backfiring Schaffer collaterals via a stimulating electrode in CA1. Applying a standard HFS to RC in this configuration also resulted in depression of the PP fEPSP, indicating that THD is not an exclusive effect of MF HFS on PP (48.8 ± 6.0%, P > 0.05; n = 4; Fig. 3C).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Our results demonstrate that LTP of the PP to CA3 synapse is both cooperative (in that recruitment of a suprathreshold number of afferents is necessary to induce potentiation) and associative (in that afferent fibers of different orders are capable of contributing to potentiation). PP LTP under both conditions was found to be NMDAR dependent. This finding is consistent with a previous study of LTP in CA1 (Colbert and Levy 1993) and suggests that backpropagating spikes are necessary in both associative and cooperative conditions (Magee and Johnston 1997). Although it is likely that strong PP HFS during the cooperative induction of LTP also recruits RC axons to some extent, it would only do so at a stimulation intensity high enough to induce spiking in the postsynaptic cell. We did not observe a long-term heterosynaptic effect of MF HFS on PP, in contrast to the previously demonstrated effect of MF HFS leading to heterosynaptic LTP of RC and commissural fibers (Bradler and Barrionuevo 1990). This result indicates that associative activation is a necessary condition for PP LTP induction, as required for an optimal pattern retrieval input (Treves and Rolls 1992).

GABA_A receptors were blocked in all experiments, in accordance with observations that PP LTP is not readily inducible in vitro without blockade of inhibition (unpublished observations; Colbert and Levy 1993). One probable cause of the susceptibility of PP LTP to GABAergic suppression is the fact that bulk stimulation of s. lacunosum-moleculare in slices recruits both PP fibers and axons of inhibitory interneurons (Freund and Buzsaki 1996). It is likely that transmission through PP in the intact brain is ordinarily under strong inhibitory control, in light of a recent report that strong perisomatic shunting typically accompanies depolarization of the distal dendrites during theta rhythm (Kamondi et al. 1998). Given that PP LTP is readily inducible in area CA3 in vivo (Berger et al. 1996; Breindl et al. 1994), inhibitory suppression of PP LTP can be disengaged in the intact brain.

Formal analyses of the memory capacity of associative networks indicate that separate input pathways mediating pattern storage and retrieval are necessary for maximum efficiency. Applying this result to area CA3 has led to the hypothesis that MF and PP are dedicated inputs for memory storage and retrieval, respectively (O'Reilly and McClelland 1994; Treves and Rolls 1992). Treves and Rolls (1992) further demonstrated that the capacity of a memory network is greatly enhanced if the input pathway that mediates retrieval is associatively modifiable. We directly tested the prediction that PP plasticity is optimized for memory retrieval. Our results support the conceptual framework that regards CA3 as an autoassociative memory network, the capacity of which is maximized through dual input pathways.

The THD that we report here is similar in both time-course and magnitude to opioid-mediated forms of heterosynaptic depression at MF and PP in the guinea pig dentate gyrus (Wagner et al. 1993; Weiskopf et al. 1993). However, we found that PP THD induced by MF or RC HFS is not blocked by naloxone. The fact that RC and MF stimulation are equally effective at inducing THD indicates a general effect of heterosynaptic input into CA3 on PP synapses. Heterosynaptic depression on a comparable time scale has been observed in other brain areas, including CA1 and the neocortex (Alger et al. 1978; Zilberter et al. 1999). The fact that several distinct forms of THD appear in many brain areas suggests that transient depression might be an important counterpart to short-term potentiation.

Although long-lasting synaptic changes are necessary to form memories persisting for hours or days, short-term plasticity (such as PTP and frequency facilitation) also contributes to neural information processing by augmenting transmission in recently activated pathways (Urban et al. 2001). We suggest that THD complements this augmenting effect during memory retrieval by temporarily lowering the level of background noise from PP synapses that have not been recently activated. According to this proposal, the role of THD in short-term plasticity is analogous to the role that heterosynaptic long-term depression plays in preventing networks from becoming saturated by repeated LTP due to new memory formation (reviewed in Turrigiano and Nelson 2000). Thus both short- and long-term potentiation are complemented by reduction of background noise by heterosynaptic depression on equivalent time scales.