Phase Sensitivity of Synaptic Modifications in Oscillating Cells of Rat Visual Cortex

Introduction

In numerous structures of the nervous system and in particular in the cerebral cortex, neuronal activity is often characterized by an oscillatory patterning in the β- and γ-frequency range (20-60 Hz) that is associated with precise synchronization of discharges (Silva et al., 1991; Singer, 1993, 1999; Buzsaki and Chrobak, 1995; Steriade, 1999; Tallon-Baudry and Bertrand, 1999). Evidence suggests that this synchronization serves response selection in the context of various cognitive functions such as binding, perceptual rivalry, attention, and short- as well as long-term memory (Fernandez et al., 1999; Gray, 1999; Miltner et al., 1999; Singer, 1999; Tallon-Baudry and Bertrand, 1999; Engel et al., 2001; Fries et al., 2001; Parkhurst et al., 2002). The likely mechanism is that coincident EPSPs summate more effectively in target structures than temporally dispersed EPSPs (Usrey and Reid, 1999). This enhanced saliency predisposes synchronized responses for further joint processing.

Temporal relations among the discharges of interconnected neurons are also an important variable in synaptic plasticity. Synapses undergo long-term potentiation (LTP) when the EPSP precedes the postsynaptic spike within a few tens of milliseconds, and they undergo long-term depression (LTD) when the EPSP follows the spike [spike time-dependent plasticity (STDP)]. Thus, synapses strengthen activity that is causally related to postsynaptic firing (Markram et al., 1997; Egger et al., 1999; Feldman, 2000; Sjöström et al., 2001; Froemke and Dan, 2002).

Interactions between presynaptic and postsynaptic activity become hard to predict when neuronal networks engage in high-frequency oscillations, as is often the case in vivo. Previous studies on the phase sensitivity of synaptic modifications have investigated θ oscillations (7 Hz) in the hippocampus and relied on field potential recordings for both the assessment of oscillation phase and synaptic modifications. These studies demonstrated LTP of field EPSPs when pairing afferent volleys with the depolarizing phases of the oscillations and a resetting of LTP when applying the stimuli during the hyperpolarizing phases (Huerta and Lisman, 1993, 1995; Hölscher et al., 1997). At these low frequencies, one expects little interference between successive activation cycles because STDP effects are weak when intervals between EPSPs and postsynaptic spikes exceed 100 msec (Markram et al., 1997; Egger et al., 1999; Feldman, 2000; Sjöström et al., 2001; Froemke and Dan, 2002), as is the case with θ oscillations. Thus, synaptic modifications are determined essentially by the level of membrane potential (V_m) depolarization at the time of EPSP arrival. Conditions become much less predictable at oscillation frequencies in the β and γ range that are common in the cerebral cortex. Here, even at phase offsets of 180°, EPSPs precede and follow postsynaptic spikes at intervals ≤25 msec, and thus STDP effects can be expected to cause strong interference between successive cycles. Furthermore, because discharges are periodic in oscillating networks, phase shifts can lead to ambiguous timing relations between presynaptic and postsynaptic firing. Such phase shifts are frequent and commonly observed in multisite recordings from the cerebral cortex (Varela et al., 2001). With 180° phase shifts, EPSPs are equidistant to preceding and following spikes, and in this case no clear predictions follow from STDP rules, provided that oscillation frequencies are high enough to bring EPSPs sufficiently close to the respective preceding and following spikes. With β and γ oscillations, these intervals are <25 msec and hence well within the critical range of STDP.

These considerations call for an experimental answer to the question of whether and how synapses change in networks that engage in high-frequency oscillations and whether the modifications remain sensitive to the precise timing relations between presynaptic and postsynaptic discharges. Clarification of this question is crucial in the context of the hypothesis that precise synchronization of discharges serves response selection and the definition of relations in signal processing.

Materials and Methods

All experiments were performed in accordance with European laws and National Institutes of Health guidelines for the treatment of animals in research.

Slice preparation. Cortical tissue was obtained from 3- to 5-week-old [postnatal day 18-33 (P18-33), P22 ± 4; mean ± SD] Wistar rats under deep halothane or isoflurane anesthesia. In a subset of experiments (13 cells), slices were taken from 2-week-old rats (P12-14, P13 ± 1). Coronal slices (300-400 μm) of primary visual cortex were prepared in ice-cold low-calcium artificial CSF (ACSF) containing (in mm): either 125 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 25 NaHCO₃, 1 CaCl₂, 2 MgCl₂, 25 glucose or, alternatively, 87 NaCl, 4 KCl, 1.25 NaH₂PO₄, 25 NaHCO₃, 0.5 CaCl₂, 7 MgCl₂, 10 glucose, 75 saccharose, saturated with 5% CO_2-95% O₂. Slices were then transferred for 30 min to warm (37°C) ACSF and then stored at room temperature. For recording, slices were transferred to a submersion chamber and perfused at 5 ml/min at 36°C with ACSF of slightly different composition (in mm): 125 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 25 NaHCO₃, 2 CaCl₂, 1 MgCl₂, 25 or 10 glucose.

Recordings. Whole-cell patch-clamp recordings were performed on 201 layer II/III pyramidal neurons in the current-clamp mode (Axoc-lamp 2B, Axon Instruments, Foster City, CA). Patch electrodes (5-10 MΩ in ACSF) were filled with intracellular solution. Solution 1 contained (in mm): 100 K-gluconate, 20 KCl, 4 ATP-Mg²⁺, 10 phosphocreatine, 0.3 Na⁺-GTP, 10 HEPES. Solution 2 contained (in mm): 130 K-gluconate, 10 KCl, 3 ATP-Mg²⁺, 10 phosphocreatine, 0.3 Na⁺-GTP, 10 HEPES, and creatine phosphokinase (50,000 U/l). Neurobiotin (0.5%) was added to the intracellular solution for subsequent anatomical reconstruction. Pyramidal neurons were identified by means of differential interference contrast-infrared microscopy and characterized electro-physiologically as regular firing neurons. They had overshooting action potentials, a resting membrane potential of -74 ± 7 mV, a firing threshold of -40 ± 5 mV, and a membrane resistance of 140 ± 61 MΩ (mean ± SD). Input resistance was monitored throughout the experiment by measuring voltage responses to hyperpolarizing current pulses (-40 pA, 300 msec). Only cells with stable input resistance (±20%) were analyzed. Series resistance (10-20 MΩ) was compensated for by adjusting the bridge, and unstable recordings were discarded.

Stimulation. Horizontal afferents were stimulated extracellularly with a concentric bipolar microelectrode positioned in layer II/III, 0.5-1 mm apart from the recorded neuron. Test stimuli consisted of trains of four pulses (pulse duration 0.1 msec; interpulse interval 50 msec; intensity 2-25 V), that were repeated at intervals of 10 sec. Once they had become stable, test responses were recorded for at least 10 min before the onset of conditioning and for up to 60 min after conditioning. Because of the relatively large distance between the recorded neuron and the stimulation site, monosynaptic inhibitory GABAergic PSPs were not evoked, because PSPs were completely blocked by coapplication of CNQX (20 μm) and APV (50 μm) (n = 14 of 14; data not shown); however, we cannot exclude activation of polysynaptic GABAergic PSPs during high-frequency stimulation.

Oscillation pairing. Suprathreshold membrane potential oscillations (±10 mV, 20 Hz and 40 Hz) were evoked by sinusoidal current injection from a depolarized holding potential (-55 to -40 mV), as is common in vivo (Gray and McCormick, 1996; Steriade et al., 1998; Azouz and Gray, 2000), and paired with 20 Hz EPSP trains. Phase angles were adjusted by imposing subthreshold oscillations at resting potential to avoid induction of synaptic plasticity during the adjustment. Stimulus sequences were timed individually as a function of the EPSP rise time (onset to peak, 7 ± 2 msec) so that EPSP peaks coincided with either the peaks or the troughs of the V_m oscillations (see Fig. 1). With peak pairing, EPSP onsets preceded oscillation peaks and the concomitant spike by 10 ± 3 msec (20 Hz oscillations) or 9 ± 2 msec (40 Hz oscillations) and followed the discharges of the preceding cycles by 40 ± 3 msec (20 Hz) or 16 ± 2 msec (40 Hz). Because of the latency jitter of oscillation-evoked spikes and variable EPSP rise times, the EPSP peaks coincided only on average with the spikes but in individual cases could either precede or follow spike onset with an uncontrollable scatter of ±2 msec. With trough pairing, EPSP onsets lagged oscillation peaks and concomitant spikes by 13 ± 3 msec (20 Hz) and 5 ± 2 msec (40 Hz) and preceded the discharges of the subsequent peak by 37 ± 3 msec (20 Hz) or 20 ± 2 msec (40 Hz). For conditioning, sequences of 10 oscillation cycles were applied at 1 sec intervals and repeated until 100 EPSPs had been delivered, i.e., 10 times for 20 Hz oscillations and 20 times for 40 Hz oscillations. Occasionally a depolarizing cycle failed to evoke a spike discharge, but there was no significant difference in the number of spikes evoked during peak and trough pairing, respectively (p > 0.05; ANOVA).

Evaluation of synaptic plasticity. The four test EPSPs were recorded continuously before and after pairing. We used repetitive stimulation (four test EPSPs) to be able to assess a potential redistribution of synaptic efficacy (Markram and Tsodyks, 1996). To this end, the amplitude changes of the first, second, third, and fourth EPSP of the test response were determined. In the majority of cases (193 of 201 cells), the polarity of modifications was similar for all four EPSPs, and a redistribution of synaptic amplitudes occurred in only eight cells. Therefore, we only used amplitude changes of the first EPSP for the quantification of synaptic modifications.

We routinely measured and plotted several EPSP parameters: EPSP amplitude, EPSP time to peak, EPSP area under the curve, and EPSP slope. Because EPSP slope, amplitude, and area changed in exactly the same way, we opted for the EPSP amplitude for the assessment of synaptic changes (supplemental Fig. 1; supplemental material, available at www.jneurosci.org).

EPSP amplitudes were compared during epochs of equal length, one immediately before and the other after the pairing by 15-20 min, with each epoch including 20 normalized and averaged data points. Changes were classified as LTP or LTD if they were significant at the p < 0.05 level (paired Student's t test). In experiments during which kainate was washed in and out, baseline EPSPs preceding kainate application were compared with EPSPs after the pairing by 35-40 min. Pairwise differences between protocols were tested with the Mann-Whitney rank sum test unless stated otherwise. We opted for the nonparametric Mann-Whitney rank sum test to avoid false assumptions about the distribution of the data. In addition we examined the significance of changes for each experimental protocol by computing grand averages. We considered all cells in this calculation as well, including those in which changes had not reached significance, to avoid selection of only those cells showing strong effects.

Pharmacology. Almost all drugs were bath applied and present throughout the experiments. The cholinergic agonist carbachol (CCH) (5 μm) was always coapplied with the antagonist gallamine (50 μm) to avoid depression of release (Bröcher et al., 1992). This allowed us to start experiments with middle-sized EPSPs (∼3 mV) potentially able to undergo both LTP and LTD without having to recruit a larger synaptic circuit by simply increasing the stimulation intensity. Effects of CCH wash-in were monitored in a subset of cells. CCH application led to a slight V_m depolarization (+3 to +8 mV; n = 16 cells) and occasionally shortened interspike intervals in response to depolarizing current steps, increasing the number of spikes generated by a constant current pulse by a factor of 1.58 (n = 7). CCH had no measurable effect on input resistance (106 ± 6%; n = 9 cells). After repetitive spiking, we sporadically observed an afterdepolarization of +1.0 to +2.5 mV (n = 3) that lasted ∼1000 msec. Application of CCH had no effect on the incidence of spikes evoked during pairing (p > 0.5; Mann-Whitney rank sum test, compared with pairing in ACSF), nor did CCH induce population oscillations in the slice.

The NMDA-receptor blocker APV (50 μm) was added to the ACSF alone or in addition to carbachol-gallamine. The muscarinic receptor antagonist scopolamine (10 μm) and the nicotinic receptor antagonist hexamethonium (100 μm) either were separately applied in addition to carbachol-gallamine or were coapplied in ACSF. The muscarinic agonist muscarine (20 μm) was added to the ACSF.

Although carbachol was present throughout the experiments, kainate (200 nm) was washed in just before and washed out immediately after pairing. The application of this low dose of kainate did not remarkably alter the V_m of the cells (slight depolarization of 2 mV on average) but resulted in a drop of input resistance (-16 ± 4%; n = 12 cells) associated with decreased EPSP amplitudes (-16 ± 7%; n = 12 cells) and increased synaptic background activity (data not shown). Furthermore, kainate accelerated EPSP decay and reduced EPSP summation (see Fig. 6, inset).

In six experiments, postsynaptic Ca²⁺ was buffered with BAPTA (10 mm) in the intracellular solution.

Results

We examined synaptic plasticity after synchronized β and γ oscillations in rat visual cortex slices. Slices were taken mainly from juvenile to adult rats (P18-33) and analyzed at close to physiological temperature (36°C). Inhibition was not antagonized, extracellular [Ca²⁺] was set to 2 mm, holding V_m was maintained at levels characteristic of “up-states” (Cossart et al., 2003; Shu et al., 2003), the amplitude of the imposed V_m oscillations was in the same range as observed during γ oscillations in vivo (Azouz and Gray, 2000), and compound multifiber EPSPs, as are likely to occur when afferent cells synchronize their discharges, were evoked.

Layer II/III pyramidal cells recorded in whole-cell configuration were forced to discharge at 20 or 40 Hz by injection of sinusoidally modulated currents. For the induction of synaptic modifications, we paired these oscillatory, high-frequency responses with trains of synchronized multifiber EPSPs that oscillated at 20 Hz and were evoked by electrodes placed laterally to the recorded cells. Stimulus trains were timed so that the peaks of the EPSPs were synchronized with either the depolarizing (peak pairing) or hyperpolarizing maxima (trough pairing) of the membrane potential oscillations (Fig. 1a,b). Synaptic modifications were assessed by comparing the amplitudes of the first of four EPSPs (responses to four pulses separated by 50 msec, repeated every 10 sec) before and after pairing (Student's t test; see Materials and Methods) (Fig. 1c,d).

Discussion

A phase dependency of synaptic modifications has been reported previously in the hippocampus for slow network oscillations in the θ range [5-8 Hz, in vitro (Huerta and Lisman, 1993, 1995), in vivo (Hölscher et al., 1997)]. In these studies, synaptic strength was assessed from field potentials. These were found to undergo potentiation after pairing of electrically evoked afferent volleys with oscillation peaks, and this potentiation could be reset by subsequent trough pairing. No evidence was obtained that LTD could be induced de novo by trough pairing. The present results extend these experiments to the neocortex and demonstrate with intracellular recordings that synaptic modifications remain highly sensitive to the phase relation between periodic presynaptic and postsynaptic activity even if presynaptic and postsynaptic elements engage in sustained activity in the high-frequency range of β and γ oscillations. This phase sensitivity depended on the oscillatory V_m modulation of the postsynaptic cell on NMDA receptors and in slices of older rats on the presence of cholinergic agonists. The phase sensitivity of synaptic modifications (contrast between effects of peak and trough pairing) still persisted during increased network activity driven by kainate application.

In the present experiments, peak pairing led to NMDA receptor-dependent LTP and trough pairing led to NMDA receptor-independent LTD. In contrast to slices of young rats, LTP induction in slices of older rats required activation of muscarinic receptors, suggesting that carbachol effects had compensated for the age-dependent elevation of the LTP threshold. Evidence suggests that this elevation is caused in part by the maturation of inhibitory mechanisms (Kato et al., 1991) and changes in the subunit composition of NMDA receptors (Nase et al., 2003). Carbachol had caused a slight depolarization of resting V_m and a reduction of spike-frequency adaptation (data not shown), probably because of a blockade of K⁺ currents (I_AHP, I_M) (Nicoll et al., 1990). This could have enhanced LTP by increasing dendritic depolarization and facilitating backpropagation of action potentials. Other nonexclusive possibilities are that carbachol enhanced NMDA receptor efficiency directly (Kirkwood et al., 1999) and/or increased Ca²⁺ release from endogenous stores, both effects being mediated via the IP₃ pathway (Markram and Segal, 1992; Nakamura et al., 2000). With peak pairing, activation of muscarinic receptors with carbachol or muscarine had increased the incidence of LTP. This is best explained by assuming a lowering of the threshold that separates the activation levels required for the induction of LTD and LTP (Artola and Singer, 1987; Artola et al., 1990; Hansel et al., 1997).

LTP facilitation was found to depend on the activation of muscarinic receptors; however, a contribution of nicotinic receptors might be considered, because the LTD observed in carbachol after blockade of muscarinic receptors was much weaker than the LTD obtained in control solution and after blockade of both receptor subtypes. The balanced activation of both muscarinic and nicotinic receptor subtypes might act synergistically to support the phase sensitivity of synaptic plasticity during fast oscillations.

Trough pairing never induced LTP in either P13 or P22 slices treated with carbachol and kainate, suggesting that this protocol provides robust conditions for LTD induction. This is remarkable in view of the sustained high-frequency discharges (20 or 40 Hz) imposed on both presynaptic and postsynaptic elements and the high affinity of NMDA receptors for glutamate (see below).

The role of the V_m oscillations appeared to be more than a clocking device to time action potentials, because coincident step-pulse pairing induced LTD and failed to induce LTP despite the presence of carbachol (supplemental material, available at www.jneurosci.org). Because we did obtain LTP in the experiments with oscillatory V_m modulation, all other conditions being identical, we conclude that the susceptibility to undergo synaptic modifications differs in regimens in which cells undergo an oscillatory modulation of their V_m and in which they are just driven above threshold by brief current injections. We propose the following possibilities. First, sinusoidal currents are likely to result in larger dendritic depolarization than brief current pulses because of dendritic low-pass filtering. Sinusoidal currents therefore should be more efficient in facilitating activation of NMDA receptors and voltage-dependent calcium channels and backpropagation of spikes (Stuart and Häusser, 2001). Second, with sinusoidal V_m oscillations, the V_m of the cells occupies depolarized levels over longer time than with step pulses. With peak pairing, EPSPs occurred in the middle of the depolarizing stretches, which should provide particularly favorable conditions for prolonged activation of NMDA receptor-mediated Ca²⁺ entry. Third, the hyperpolarizing phases of V_m oscillations could have favored deinactivation of previously activated, voltage-gated Na⁺ and Ca²⁺ channels, which could have contributed to a stronger depolarization with peak pairing.

The conditions of our experiments (recordings at physiological temperature with intact inhibition, synchronized EPSP multifiber volleys exhibiting an oscillatory modulation at 20 Hz; oscillatory modulation of the V_m of postsynaptic cells at depolarized V_m levels associated with periodic firing at 20 or 40 Hz; stimulation of cholinergic and kainate receptors) closely resemble those in vivo when neuronal networks engage in synchronized gamma oscillations (Gray and McCormick, 1996; Azouz and Gray, 2000). The major difference is that in the present experiments V_m oscillations were induced with current injection, whereas in vivo they are attributable to the alternation of synchronized volleys of EPSPs and IPSPs (Bringuier et al., 1997). Hence, in vivo, oscillations are associated with periodic increases of membrane conductance and shortened membrane time constants (Traub et al., 1996; Lampl et al., 1999). We predict from this that the phase sensitivity of synaptic modifications should be even higher in vivo than under the present conditions, when neuronal networks engage in high-frequency oscillations. This prediction is supported by the finding that the differences between the effects of peak and trough pairing persisted in a kainate-driven increased conductance state.

Unlike in previous studies on STDP, we evoked periodic discharge patterns in both presynaptic and postsynaptic elements, imposed an oscillatory modulation of V_m on the postsynaptic cells, and varied the phase angles between the periodic discharge sequences between in-phase and of-phase conditions. In the first case, EPSPs were coincident with postsynaptic discharges and the peaks of V_m depolarizations; in the second, they were equidistant from the preceding and following spikes and fell in the troughs of the V_m oscillations. Under these conditions, causal relationships between presynaptic and postsynaptic responses are no longer clearly defined. Still, the changes in phase angle led to a reversal of the polarity of synaptic changes, suggesting that the Hebbian LTP mechanism evaluates not only causal relations between presynaptic and postsynaptic discharges, as demonstrated in STDP experiments, but also the synchronicity between oscillatory presynaptic and postsynaptic activity. We propose that the mechanisms are the same as those supporting STDP. Peak pairing leads to maximal cooperativity between presynaptic and postsynaptic activity even if EPSPs jitter around the interval of postsynaptic spiking, because EPSPs coincide with phases of maximal V_m depolarization and postsynaptic spikes. Both factors increase Ca²⁺ entry through NMDA receptors, which was a necessary prerequisite for LTP induction in the present experiments, and in addition, both enhance the probability of Ca²⁺ entry through voltage-gated Ca²⁺ channels. In contrast, with trough pairing, EPSPs are less likely to activate NMDA receptor-gated Ca²⁺ channels, so that synaptic modifications, in this case LTD, should depend mainly on voltage-gated Ca²⁺ channels and/or metabotropic glutamate receptors. This conjecture is supported by the finding that blockade of NMDA receptors had no effect on the incidence of trough pairing-induced LTD and that LTD was also inducible with presynaptic stimulation alone.

The question remains, however, of why NMDA receptor activation contributed so little to the effects of trough pairing, especially when cells oscillated at 40 Hz. In this case, EPSPs peaked ∼12 msec before the respective next spikes. In STDP experiments, such a condition leads to NMDA receptor activation and to LTP, especially when discharge frequencies are high (Sjöström et al., 2001). The most likely interpretation is that in the present experiments the contribution of NMDA receptors was reduced by inhibition and by the oscillatory V_m modulation that caused rapid succession of hyperpolarizing phases. We could exclude the possibility that the test stimuli used for the assessment of synaptic gain changes evoked monosynaptic IPSPs that could have contaminated the test EPSPs (see Materials and Methods), but it is very likely that the high-frequency trains used for the induction of modifications recruited polysynaptic inhibition. Thus, inhibition evoked by the trough stimuli was probably still effective around the peak of the afterdepolarization cycle. We propose that this reduced the probability of lifting the Mg²⁺ block of NMDA receptors that are still occupied by glutamate released by the trough stimuli. In addition, because of the high-frequency oscillations, the time window for the removal of the Mg²⁺ block was short because of the intervening hyperpolarization phases. Both mechanisms are likely to sharpen the phase sensitivity of the synaptic modifications by focusing the window for LTP induction on the peaks of the V_m oscillations.

The present results indicate that connections between oscillating cell groups strengthen or weaken depending on whether the cells oscillate in phase or in antiphase. Such a mechanism establishes associations between cells exhibiting synchronous firing patterns, regardless of the origins of correlated firing. Precisely synchronized firing can have several causes. It can result from stimulus locking of responses to temporally contingent stimuli (Rager and Singer, 1998), it can be caused by common input from diverging afferents, or it can be caused by dynamic, self-organizing interactions among reciprocally coupled neurons. In the latter case, the synchronizing effects are mediated by both excitatory and inhibitory interactions (Buhl et al., 1998), which are mediated by reciprocal corticocortical connections in the neocortex. A mechanism that enhances coupling between synchronously active cell groups this is ideally suited to support associative learning and the formation of cell assemblies. It can establish associations as a function of temporally contingent external events, but it can also associate neurons that have synchronized their responses because of internal grouping and binding operations.

Of particular relevance for functional interpretations is the finding that synaptic plasticity remained sensitive to small shifts in synchrony even when cells engaged in sustained, periodic, high-frequency discharges such as occur during γ oscillations in vivo. This suggests that precise synchrony of discharges serves as a signature of relatedness in Hebbian LTP. An important implication is that precise synchrony is then likely to also serve as a tag of relatedness in signal processing. The reason is that learning of false or spurious conjunctions can be avoided only if the code that defines neuronal responses as related operates on exactly the same time scale in signal processing and synaptic plasticity.