Increasing evidence suggests that glial cells are endowed with the ability to externalize their activity to the extracellular space and to neurons. Since the same activity is influenced by the extracellular ionic concentrations and the neurotransmitters released by neurons, it is suggested that neurons and glia entertain a continuous exchange of information. This behavior might have a particular significance during cortical oscillations. In this study we analyzed the time and voltage relationships within simultaneously recorded neuron–glia pairs during normal states characterized by a slow (<1 Hz) sleep oscillation and during paroxysmal epileptic discharges. Our data show that cortical neurons and glia display coherent activities during the tested spontaneous oscillations. The onset of the depolarizing phase of the slow oscillation started in neurons and followed with a lag of 88 ms in nearby (1–2 mm) recorded glial cells. In contrast, the beginning of the hyperpolarizing phase was initiated in glial cells, and neurons followed after 79 ms, suggesting that glial activities are not exclusively the reflection of neuronal ones. Moreover, we tested neuronal excitability that resulted in phase opposition with the glial membrane potential, establishing that only the first 30% of the neuronal depolarization is efficient for synaptic volleys within cortical neuronal networks. Seizures were associated with shorter time lags at onset of depolarization (1.8 ms) and with delayed glial offset (102 ms). The voltage slope and amplitude at the onset of the paroxysmal depolarizations were higher than in the case of the slow oscillation. Together with the variation of neuronal excitability, these results suggest that the glial uptake of K+ contributes to the abridged duration of the paroxysmal depolarization.
Sleep and paroxysmal oscillations involve a larger domain than a network of interconnected neurons in the cerebral cortex. Synchronous fluctuations of the membrane potential of large populations of neurons are associated with phasic shifts in the ionic composition of the extraneuronal compartment. The latter is a complex environment, which includes the extracellular space and networks of glial cells. The extraneuronal compartment potentially performs a spatial and temporal integration of ionic gradients, determining the setting of ionic concentrations on the outer face of the neuronal membranes. This is particularly relevant in the case of K+: on one side, K+ ions, once extruded by neurons, undergo a complex managing by the glial syncytium; on the other side, the final extracellular K+ concentration ([K+]out) is a basic factor influencing the resting potential and the excitability of neurons. Moreover, Ca2+-dependent release of glutamate by glial cells has been proposed as a form of direct communication between glia and neurons (Parpura et al., 1994; Araque et al., 2000). Thus, a close loop between neurons, glial networks and extracellular space could be considered as the complex frame within which synchronous cortical oscillations arise. A first step to validate this model is to study the phasic events in the extraneuronal space associated with neuronal oscillations on a cycle-by-cycle basis. The glial membrane potential varies as a function of [K+]out, which in turn is a function of spatial buffering (Amzica et al., 2002) and neuronal activity. A first prediction consistent with the hypothesis of a neuron–glia dialogue is that neuronal and glial potentials display coherent fluctuations. Moreover, systematic study of the time relationships between neuronal and glial potentials during different oscillatory patterns would provide information with a two-fold perspective: first, considering the glial potential as an index of extracellular K+, we can infer the direction of the net current of this ion across the neuronal membrane during the different phases of the oscillatory cycle. Secondly, looking at the glial syncytium as regulatory system of the extracellular ionic balance, we can make some hypotheses on the possible active role glial cells play in the modulation of cortical oscillations.
In this study we performed in vivo double intracellular recordings of neocortical neurons and glial cells during slow sleep oscillatory patterns and their spontaneous transition to interictal paroxysmal depolarizing shifts (PDSs) and spike–wave (SW) seizures (Steriade et al., 1993a; Steriade and Amzica, 1994; Amzica and Neckelmann, 1999; Amzica and Steriade, 2000). If the oscillatory cortical activity results from the interplay between neuronal and glial activities, we expect that interchangeably, glia and neurons would lead within the operational sequence of an oscillatory cycle. We therefore focused on the calculation of correlation indexes and time lags between phasic oscillatory events. Furthermore, we linked the neuronal excitability with the glial activity in order to test whether the latter modulates the former. Extending our view outside the neuronal membrane potentially brings new insight on the nature of intraneuronal processes and represents a first step to disclose a possible extraneuronal contribution to the genesis of sleep and paroxysmal oscillations in the cerebral cortex.
Materials and Methods
The procedures for animal preparations were described in a previous paper (Amzica et al., 2002). Briefly, 80 cats of both sexes were anesthetized with ketamine–xylazine (15 mg/kg and 3 mg/kg, respectively), tracheotomized, and paralyzed with gallamine triethiodide. Artificial ventilation (20–30 cycles/min), end-tidal CO2 concentration (∼3.7%), heart rate (<110 beats/min), and slow-wave EEG pattern were maintained throughout the experiment. The ketamine–xylazine anesthesia has proven to induce a slow oscillatory activity (Steriade et al., 1993a) similar to the one recorded during natural sleep (Steriade et al., 2001). Craniotomy exposed the suprasylvian gyrus, where DC pipettes and field potential coaxial electrodes were lowered into the cortex. To enhance the stability of the recordings, cisternal drainage and bilateral pneumothorax were performed, and 4% agar solution was applied in the hole of the calvarium. At the end of the experiments, the animals received a lethal dose of sodium pentobarbital.
The intracellular recordings were obtained from areas 5 and 7 of the suprasylvian gyrus with glass micropipettes (tip diameter <0.5 μm) filled with a 3 M solution of potassium acetate (impedance in situ 30–40 MΩ). Similar electrodes, with larger tips (1–2 μm), were used for the recording of DC extracellular field potentials. Signals were passed through a high-impedance amplifier with active bridge circuitry (Neurodata). The EEG was recorded monopolarly with coaxial tungsten electrodes at a depth of 1 mm and at the surface of the cortex (reference in the paralyzed neck muscles). These potentials were bandpass filtered between 0.3 Hz and 1 kHz. Pairs of neurons and glia were recorded within close distance (1–2 mm between the penetration point of the pipettes into the cortex) or from distant sites (>10 mm).
In this study, the quantification of the various relationships between neuronal and glial potentials relied on the precise detection of some stereotyped markers during each cycle of the oscillations. Figure 1A illustrates such markers and the principle underlying their detection. The onset of the depolarizing phase of a slow oscillation or SW complex (‘neuron onset’ in Fig. 1A) was considered at the moment where the first derivative of the neuronal potential crossed the zero line in the upward direction. Due to rich synaptic-like activity during the depolarizing phases of the above-mentioned oscillations, a marker had to be preceded by a negative derivative and followed by an average positive derivative for at least 20% of the average oscillatory cycle of the time series to be analyzed. The following maximum of the neuronal derivative determined the marker for the maximum slope of the depolarization (‘neuron maximum slope’ in Fig. 1A). The marker for the onset of the oscillating cycle in the glial trace was set similarly to the neuronal marker. The end of the neuronal depolarization (‘neuron offset’ in Fig. 1A) was considered at the moment where the derivative of the signal crossed the zero line in the negative direction and remained so for at least 20% of the oscillatory cycle. The corresponding moment in the glial potential (‘glia offset’) coincided with the maximum of the oscillatory cycle. Time lags between the neuronal and glial onset (t1 in Fig. 1A), between the neuronal maximum slope marker and the glial onset (t1a), and between neuronal and glial offset (t2) were calculated as differences between the respective time markers in order to evaluate, on a cycle-by-cycle basis, their dynamic evolution. For this kind of analysis, the action potentials in the neuronal traces were clipped in order to avoid the detection of their very steep slopes.
The slopes at the onset and offset of each cycle were calculated as measures for the steepness of the depolarization, and indirectly, as indicators of the amount of the network synchronization (i.e. steeper onsets of the depolarization, higher the recruitment of cells). The neuronal onset slope was taken directly from the first derivative of the membrane potential (Vm) at the moment of the maximum slope (Fig. 1A). Due to the slower course of the glial potentials, their slopes were calculated after each cycle was fitted with a parabolic function (‘fit’ in Fig. 1B), which proved to be the best fit for this potential shape (lowest Chi-square function, in average 25% below the second best fit, the Lorentzian function). Then, the onset and offset slopes were calculated at the beginning and end of the fitting curve (gray angles below the tangents to the respective points in Fig. 1B).
The estimation of the amount of cellular activity during an oscillatory cycle also relied on the surface area below the depolarization (Fig. 1C). This measure was considered because, in the case of neurons, only the depolarizing phase is made of synaptic and intrinsic currents (Steriade et al., 1993a), while the hyperpolarizing phase is due to network disfacilitation (Contreras et al., 1996) caused by decreased synaptic efficacy in the neuronal network (Massimini and Amzica, 2001; Amzica et al., 2002). At the same time, the glial almost exclusive depolarization reflects K+ currents throughout the cycle (Amzica et al., 2002). In both neurons and glia, the surface area was taken as the integral of the Vm curve calculated between the onset of the depolarization (as defined above) and the moment where the same Vm is reached after the offset point. Thus, the parameter discards eventual extremely slow variations of the Vm. Each depolarization surface was further divided into two parts (A and B). The first surface (SNA or SGA, respectively) started at the onset of the depolarization and ended at the offset point (darker filling in Fig. 1C). This calculation was necessary to establish a possible relationship between the amount of neuronal activity during the period where action potentials could be generated and the corresponding increase in glial Vm.
Since one of the aims of this study was to assess the relationship between neuronal excitability and phasic glial activity, we challenged the cells we recorded intracellularly with electrical cortical stimuli. Cortical stimulation was preferred because both spontaneous oscillations under scrutiny (slow and SW) are cortically generated [reviewed in Amzica and Steriade (Amzica and Steriade, 2001)]. For a given period of time containing at least 30 cycles of spontaneous oscillations, we calculated the generic shape of neuronal and glial intracellular potentials [see details for wave-triggered average (WTA) calculations in Amzica et al. (Amzica et al., 2002)]. Then, we estimated the response of each stimulus with respect to the phase of the oscillation. Cortical stimuli were delivered randomly, such to occur at various delays with respect to the phases of the oscillations. Thus, the shape of the WTAs was not systematically affected by synaptic responses. The assessment of neuronal excitability was designed to include only the effective result of an evoked potential. We built a transfer function in which each response was translated into a voltage scale as follows: from each response we retained the maximum Vm reached by the orthodromic EPSP. If this response also triggered action potentials, an additional 10 mV depolarization was added to the triggering threshold (–57 mV) for each action potential.
By adopting this quantification of neuronal excitability we want to emphasize that, from a neuronal network communication point of view, spikes are the only effective communicators, while EPSPs, regardless of their amplitude, are necessary but not sufficient for communication. This is why the production of an action potential was ‘rewarded’ with a 10 mV depolarization, while the bare EPSP was evaluated just with its top voltage. The ensuing transfer function is, however, informative because the ordinate only bears conventional values. Finally, the transfer function was fitted with a Lorentzian function of the following type:
Correlative analysis was performed according to Bendat and Piersol (Bendat and Piersol, 2000).
This study relied on the simultaneous intracellular recording of 132 neuron–glia pairs. The quality criteria for intraneuronal impalements were: stable Vm more negative than –60 mV without hyperpolarizing current, and overshooting action potentials. Glial recordings were considered only if the Vm fell upon impalement at values more negative than –70 mV (Fig. 2A). Additionally, these recordings had to be stable, not requiring application of steady hyperpolarizing currents, and without generation of action potentials following depolarizing pulses [see also Amzica and Neckelmann (Amzica and Neckelmann 1999)].
The synchronization between glial and neuronal cells has been studied using double impalements during three states: slow oscillation, interictal PDSs and SW seizures (see Fig. 2 for summary). First, during the slow sleep oscillations (Fig. 2A), the membrane potential of neurons alternated between depolarizing and hyperpolarizing levels with a frequency slightly below 1 Hz. The depolarizing segment of a cycle was close to the firing threshold and thus triggered somatic action potentials. In parallel with this type of activity, glial cells displayed in-phase depolarizing–repolarizing sequences.
Occasionally, the slow oscillating pattern was overwhelmed by recurrent PDSs (Fig. 2B). It was not clear whether the PDSs were triggered by the depolarizing segments of the slow oscillation or by other (idiopathic) events. The amplitude of the neuronal PDS was higher than that of the depolarization during the slow oscillation, reaching values where the action potentials initially produced bursts, and were afterwards rapidly inactivated. Increased amplitudes also characterized glial PDSs. Such patterns of rhythmic activity could last for minutes before evolving into SW seizures (Fig. 2C) where the frequency of PDSs increased up to 2–3 Hz. Occasionally, the ictal PDSs were associated with recurrent fast runs at ∼10 Hz [not shown; see, however, fig. 4 in Amzica and Neckelmann (Amzica and Neckelmann, 1999)]. Phasic glial potentials were also present during seizures. Both in the case of the slow oscillation and seizures, the cyclic maximum of the glial depolarization was reached close to the moment where the neuron started to hyperpolarize.
During both interictal PDSs and the transition from the slowly oscillating to the paroxysmal pattern, the glial repolarization was completed for every cycle. However, as the pace of the oscillation accelerated toward the seizure, a steady depolarization was recorded (Fig. 2C), in accordance with previous reports (Sypert and Ward, 1971; Dichter et al., 1972; Ransom, 1974). The persistent depolarization came to an end as the pace of the seizure slowed down, allowing the repolarization to reach more hyperpolarized values.
Next, we will assess the glia–neuron synchronization by analyzing, cycle-by-cycle, several parameters such as time lags, surface areas and slopes of phasic events building up the oscillations.
Time Relationships for Phasic Events
The global synchrony between glial and neuronal activities was assessed by means of correlative analyses (Fig. 3). During ongoing slow oscillations, the major coherence was found for a frequency <1 Hz (arrow in Fig. 3A). The average coherence height was 0.61 ± 0.2, corresponding to an average frequency of 0.82 ± 0.13 Hz (mean ± SD; n = 47 pairs). Other frequency values in the range of delta (1–4 Hz) and spindle (∼8 Hz) activities were shared between glial and neuronal activities, although with reduced contributions. The average cross-correlation confirmed that the slow (<1 Hz) component is the major oscillatory event (see the periodicity of secondary correlation peaks in Fig. 3A, bottom panel), but additionally drew the attention on the polarity relationships between intraglial and intraneuronal potentials. The two peaks around zero time lag with relative equal amplitudes, one positive and one negative, suggest that during part of a cycle of the slow oscillation glial and neuronal potentials have similar evolutions, while during the other part of the cycle they have opposite evolutions. This interpretation of the cross-correlation is consistent with the raw data (see Fig. 2 and below).
SW seizures were characterized by more synchronous oscillations (Fig. 3B). The average coherence function (n = 75 pairs) displayed high values (>0.8) for most of the slow frequency range, with a maximum peak ∼2 Hz, the typical frequency recorded during this type of seizures. The feature was also reflected by the cross-correlation function (bottom panel in Fig. 3B), which also showed a more consistent tendency of the two types of cortical cells to evolve in-phase.
In order to quantify the synchrony between neurons and glia, we have estimated the time lags between related events (Table 1). First, in the case of closely recorded pairs (1–2 mm), the time lag between the onsets of the neuronal and glial depolarizations (t1 as described in Fig. 1A) was 88 ± 12 ms (n = 47 pairs) during the slow oscillation and 2.4 ± 1.3 ms (n = 58 pairs) during interictal PDSs. When calculated for a given pair of cells during stable periods of slow oscillation (Fig. 4A) or interictal spikes (Fig. 4B), the individual time lags had a normal distribution and did not show any pattern of systematic variations. In the majority of cases (91%), the neuronal onset preceded the glial onset. During SW seizures, the time lag t1 (glia–neuron) displayed a dynamic evolution (Fig. 5A, black circles), similar to the one observed between pairs of neurons (Steriade and Amzica, 1994). The average time lag during the seizures was 1.8 ± 1.1 ms (n = 75 seizures) and both positive and negative values were disclosed during the same seizure, indicating that the precession was alternating between glia and neurons.
In cases where the glia–neuron pairs were recorded at some distance (>10 mm) from each other (n = 20 pairs), time lags were systematically higher than for closely located pairs of cells.
We also investigated the time relationship between the point of maximum slope during the onset of the depolarization of the neuron and the onset of the depolarization in glial cells (t1a as defined in Fig. 1A). The point at neuronal maximum slope was considered because it may reflect the moment where the maximum number of neurons is recruited within the cortical network at the beginning of a new oscillatory cycle. The time lag t1a followed the same evolution as the time lag between the onsets of glia and neurons during various states of the cortical network, although the amplitude of this variation was reduced by 44% with respect to the t1 variation (n = 75 seizures). Such a situation is depicted in Figure 5A (open circles).
The end of the periodic neuronal depolarization (beginning of the hyperpolarization) was analyzed in relationship with the maximum cyclic glial depolarization (t2 as shown in Fig. 1A). During the slow oscillation and in closely located pairs, glial cells started to hyperpolarize before neurons (Fig. 4A) with a time lag of –79 ± 21 ms (n = 47 pairs). This is a first indication that glial activity is not simply a delayed reflection of neuronal activity. If it had been so, one would have expected the time lag at the offset of the depolarization to follow the same dynamics as the time lag corresponding to the onset of the depolarization. This assumption has been further contradicted by the lack of correlation between the distribution of t1 and t2 time lags (cross-correlations in Fig. 4). For interictal PDSs, the neurons started to hyperpolarize before glial cells (Fig. 4B), and the time lags were 102 ± 23 ms (n = 58 pairs). During transitions from slow oscillating patterns towards SW seizures, the time lags between offsets had an inverse evolution with respect to the corresponding onset time lags t1 (Fig. 5A, ‘OFFset’).
The relationships between time lags t1 and t2, and the duration of the depolarizing phase of oscillations were as follows: the onset time lag (t1) was proportional (average correlation coefficient of 0.73, n = 75) to the duration of the immediately succeeding neuronal depolarization (‘DND’, Fig. 5B1). In other terms, the longer the glial delay with respect to the neuronal onset, the longer the neuronal depolarization of that cycle. The offset time lag (t2) was inversely proportional to the duration of the immediately preceding neuronal depolarization (Fig. 5B2) with an average correlation coefficient of –0.62 (n = 75).
Slope Relationships for Phasic Events
The slopes of the phasic depolarizing events are one of the parameters that best reflect the reaction time of the cortical cells to synaptic/ionic impingements. We analyzed both onset and offset slopes during the various oscillatory patterns described in this study (see Fig. 1B and Methods). First, it was noticed that the transition from slow oscillations to SW seizures was accompanied by pronounced changes of the onset slopes of both neurons and glia (Fig. 6). During the first cycles of a seizure, the onset slope of glia increased faster than that of neurons (94% of the tested seizures). This behavior suggests that glia react promptly to variations of the [K+]out, probably in relation with the long-range spatial buffering reported in a recent study (Amzica et al., 2002). It has been equally observed that the slope of the repolarizing segment of glial phasic activities (offset slope in Fig. 6) followed, although with reduced values and variations, a similar time course as the onset slopes: steeper repolarizing slopes during the seizure than during the slow oscillation. All slope values reached a saturating level during seizures (Fig. 6) and returned at control values after the arrest of the seizure (not shown). The statistical values for slopes are indicated in Table 1. There was no significant difference between the values calculated for interictal PDS and ictal SW complexes.
The most important correlation patterns between glial and neuronal slope values are depicted in Figure 7A. The best correlation resulted between the onset slopes of glia and neurons (Fig. 7A1) and between the onset and offset slopes of glial cells (Fig. 7A3). The correlation between the neuronal onset slope and the glial offset (Fig. 7A2) was lower than each of the previous correlations, suggesting that it results rather from a transitive combination of the two, than from a direct relationship.
A particular type of correlation appeared between the glial slopes and the duration of the neuronal depolarization (‘DND’, Fig. 7B). In contrast with the previously calculated correlation functions, which displayed peaks with zero time lags, the correlations with the duration of the neuronal depolarization (insets in Fig. 7B) led to delayed correlation peaks. The delay was of 1–2 cycles, meaning that longer neuronal depolarizations were predicting steeper onset (positive correlation in Fig. 7B1) and offset (negative correlation in Fig. 7B2) slopes for glia in the subsequent cycles. These correlations were obtained during transitions from sleep patterns to SW seizures and could be related with the amount of K+ that is expelled from the neurons during the depolarizing phase. This suggests that the faster glial reaction at the onset of the seizure is due to a progressive prolongation of the depolarizing phase of the slow oscillation.
Voltage and Surface Area Relationships for Phasic Events
In view of the fact that the activity of glial cells is mainly devoted to the uptake of extracellular K+, which induces a depolarization of the glia, we correlated the amount of depolarization of these cells with the amount of depolarization expressed by neurons. Generally, during slow and paroxysmal oscillations, the resting Vm of cells (measured in the trough of the hyperpolarizations during oscillations) underwent very slow spontaneous variations (e.g. the slow depolarization during the seizure in Fig. 2C). Such trends had similar developments in glia–neuron pairs (correlation coefficient of 0.62 ± 0.1; n = 83 pairs). Glial depolarization resulted from the accumulation of a K+ gradient in the extracellular space, which provoked an inward drive of K+ toward the glia. On the other hand, the increase in [K+]out moved the Nernst equilibrium potential towards a more depolarized value, possibly entraining the steady depolarization of neurons and, at the same time, diminishing the expelling gradient of K+ from the neurons.
Besides the very slow depolarizing trends recorded in cells, we also estimated the amount of depolarization expressed during the phasic events building up the cycles of oscillations. The amount of depolarization was evaluated by means of the surface area below the depolarizing segment of an oscillation (SNA+SNB and SGA+SGB, as described in Fig. 1C; see Methods). In a cycle-by-cycle analysis (Fig. 8), neuronal and glial surface areas followed similar time courses. However, the correlation coefficient was higher during SW seizures (0.84 in Fig. 8B) than during the slow oscillation (0.67 in Fig. 8A). At the level of a representative sample, the correlation coefficient between surface areas was 0.62 ± 0.23 (n = 47) for the slow oscillation and 0.86 ± 0.12 (n = 75) for seizures. Since the neuronal phasic depolarizations are made of entry of Na+, Ca2+ and exit of Cl− and K+, while the glial ones mostly reflect entry of K+, the increased correlation between the depolarization areas of the two cell types suggests that the ionic activity of Na+ and Ca2+ is proportional to the one of K+ and/or that the neuronal phasic depolarization depend on the amount of extracellular K+. Alternatively, it could be that SW seizures are accompanied by a more active glial uptake of extracellular K+.
The previous result was based on surface areas calculated, in the case of glia, for the whole cycle, which also includes the repolarizing phase of the glia. This segment of the slow oscillation as well as of the SW complex is somehow controversial, several mechanisms being proposed to underlie it (see Discussion). It appeared therefore necessary to also calculate the surface area of glia for the first part of the depolarization (SGA), that better corresponds to the duration of the depolarization of the neuron. By correlating the surface areas of the initial part of the glial depolarization with the neuronal depolarization, we obtained an 11% and a 14% drop of the correlation coefficient for the slow oscillation and SW seizures, respectively. In other words, the neuronal depolarization better correlates with the whole glial depolarization than with the one contained only during the rising phase, indicating that the amount of K+ crossing the glial membrane during a whole cycle is a better estimate of what is released during the neuronal depolarization. These data alone do not permit to determine whether all the glial depolarizations reflect uptake of K+ released by neurons during the depolarizing phase or if, reciprocally, the amount of spatially buffered K+ by glia (Amzica et al., 2002) influences neuronal excitability. We therefore addressed the issue of neuronal excitability during the two phases of the slow and SW oscillations.
In order to test neuronal excitability, we took a more pragmatic approach that included the Vm variations during the various oscillatory phases. Our definition of neuronal excitability took into account only the effective output of a neuron for the rest of the network, i.e. the action potential, regardless of the Vm at which the stimulus was given. By doing so, we emphasized the fact that a hyperpolarized Vm producing an ampler EPSP is as useless as a more depolarized Vm, if both responses are devoid of action potentials. Then, for the responses containing action potentials, we counted their number and produced a score as a function of their number. This method was applied on pairs of neuron–glia recordings displaying spontaneous slow (Fig. 9) and SW (Fig. 10) oscillations. In order to minimize the interference of time lags between neuronal and glial activities, only pairs recorded at very short distance (<1 mm; n = 25) were considered.
The WTAs of a neuron–glia pair during slow oscillations (thick lines in Fig. 9) and the superimposed scoring of the responses (dots) demonstrate that the neuronal excitability is maximal toward the beginning of the depolarizing cycle, coinciding with the highest probability for action potentials. Moreover, neuronal excitability was almost zero toward the end of the depolarizing phase of the slow oscillation, in spite of the fact that the neuronal Vm did not change significantly from the beginning of the depolarizing phase. The average evoked responses of neuron (N) and glia (G) are depicted in the inset of Figure 9 and generally consisted of an EPSP–IPSP sequence for the former, and a small depolarization for the latter. The action potential scoring for a given WTA was fitted with a Lorentzian function (thin line in Fig. 9).
The quantitative analysis for cortically evoked neuronal responses (n = 25 neurons) showed that the peak of the fitting curve for each of the cases fell in the first 30% of the duration of the depolarizing phase of the slow oscillation. Translating the fitting curve over the WTA of the glial recording (Fig. 9) shows that the neuronal excitability followed an almost inverse relationship with the glial Vm: the neuronal excitability started to increase as the glia repolarized, reached its maximum close to the trough of the glial Vm, and declined as the glia depolarized. Knowing that the glial Vm follows the evolution of the [K+]out concentration, it results that the latter has a contribution in shaping the depolarizing phase of the slow oscillation.
The neuronal excitability during the recurrent SW complexes is illustrated by the WTAs of a neuron–glia pair in Figure 10. Averaged responses of both neuron and glia are depicted in the inset and produced in all cases potentials with higher amplitude than those evoked during normal sleep activity (inset in Fig. 9). Similarly, the amplitude of the ictal PDS was higher than that of the corresponding depolarizing phase of the slow oscillation, often crossing the threshold for firing action potentials, which made the latter inactivate quite rapidly. The best fit for the neuronal excitability during SW seizures was again a Lorentzian function with duration similar to the one of the PDS. Superimposing the fitting of the excitability curve over the WTA of the glial PDS leads to the conclusion that the neuronal excitability is maximal during the trough of the glial oscillatory cycle, and starts to decrease as the glial potential starts to depolarize, thus as the [K+]out increases.
That glial cells are influenced by neuronal activity is a wellestablished fact (Kuffler et al., 1966; Orkand et al., 1966). More recently, the demonstration that glial cells release neurotransmitters, of which glutamate (Parpura et al., 1994) is of special interest in view of its ubiquitous presence in the cortex, has opened the investigation of glial signaling upon neurons. Recently, we have suggested a new communication protocol at work during normal and paroxysmal oscillatory patterns (Amzica et al., 2002). We have shown that the spatial buffering of K+ ions may underlie a glia–neuron transfer as part of a more complex information exchange occurring in cortical networks. The aim of this study was to establish whether the time sequencing of various phasic activities supports a glia-to-neuron feedback, and even more so in living brains, during spontaneous oscillatory patterns. The oscillations used in this study (slow sleep oscillations and SW discharges) have stereotyped patterns in which the Vm of both neurons and glia undergo cyclic depolarizations and hyperpolarizations (Fig. 2). These oscillations are generated in the cortex (Steriade et al., 1993b; Steriade and Contreras, 1998), and SW seizures evolve gradually from the slow oscillation (Steriade and Amzica, 1994; Steriade et al., 1998a).
The advantage of a rich in vivo behavioral repertoire is, unfortunately, paralleled by a limited ability for pharmacological manipulations. Therefore we cannot claim definite answers to our questions. However, this study has established that: (i) Both duration and amplitude of glial and neuronal depolarizing activities are consistently correlated. (ii) Neurons precede glia at the onset of the cyclic depolarizations, while glial cells announce the end of the neuronal depolarizing phase during the slow oscillation, while still lagging the offset of neuronal epileptic depolarizations. (iii) Neuronal excitability is maximum at the beginning of the depolarizing phase during the slow oscillation and, in spite of a steady Vm, gradually diminishes with the progression of the phasic depolarization, in agreement with the progressive network disfacilitation accompanying the onset of the neuronal hyperpolarization (Contreras et al., 1996).
Onset of the Cyclic Depolarization
The depolarizing phase of the slow oscillation is associated with action potential discharges in the cortical network (Steriade et al., 1993a). The exclusive confinement of the neuronal firing during this phase of the slow oscillation has led to the conclusion that all neuronal types (excitatory as well as inhibitory) are active only during the persistent depolarization. This was further sustained by the finding that spontaneous IPSPs were recorded only during this phase (Steriade et al., 1993a). Thus, the Vm displayed during the depolarizing phase of the slow oscillation in the neuronal soma mainly reflects the balance between excitatory and inhibitory currents flowing from the dendrites. It is hence reasonable to think that both synaptic currents in the dendrites as well as eventual somatic action potentials would induce an overall outward K+ current. The uptake of these ions by glial cells together with their spatial buffering in the extracellular space (Benninger et al., 1980; Dietzel et al., 1980; Lux et al., 1986; Janigro et al., 1997; Amzica et al., 2002) results in a gradual depolarization of the glial membrane.
We found that the onset of the glial depolarization lags the onset of the neuronal depolarization, in agreement with the above scenario. The glial depolarization reflects with virtually no delay the K+ uptake [fig. 3 in Amzica et al. (Amzica et al., 2002)]. The time lags reported here (88 ms for neuron–glia pairs located at 1–2 mm) are longer than the ones obtained from pairs of neurons (12 ms in Amzica and Steriade (Amzica and Steriade, 1995). This apparent discrepancy might be due to the fact that K+ is also continuously recaptured by neurons through Na+/ K+-ATP pumps. The actual onset of the glial depolarization may thus mark the moment where the increase of [K+]out exceeds the capacity of the neuronal pumps. Alternatively, it could be possible that during the initial period, equivalent to the neuron– glia time lag reported here, the spatial buffering works in an efficient manner such to preclude a perceptible extracellular K+ raise. Then, the subsequent [K+]out increase would betray a less efficient spatial buffering. These behaviors suggest that glial cells are essential for maintaining the [K+]out within low values.
Interictal and ictal EEG spikes correspond to intraneuronal PDSs, which are giant EPSPs (Matsumoto and Ajmone-Marsan, 1964; Ayala et al., 1973; Johnston and Brown, 1981, 1984) on top of which inhibitory potentials are superimposed (Steriade et al., 1998b). The onset time lags within neuron–glia pairs during phasic paroxysmal discharges are lower than in the case of the slow oscillation, consistent with the shorter time lags recorded between pairs of neurons during SW discharges (Steriade and Amzica, 1994), and both the amplitude and slope of the depolarization was higher than the one recorded during the slow oscillation, suggesting that glial cells promptly take up the K+ expelled by neurons. The faster rise in [K+]out may also be triggered by a faster saturation of the neuronal ATP pumps, which are known to decrease their activity in response to increased intracellular Ca2+ concentration (Skou, 1957; Fukuda and Prince, 1992). Periodic decreases of the [Ca2+]out, in phase with the SW complexes, have been recently described (Amzica et al., 2002), but the proportion of Ca2+ entering cells at the postsynaptic versus presynaptic sites is not known. A series of evidence indicates the prevalence of Ca2+ uptake at the postsynaptic level (Heinemann and Pumain, 1981; Bollmann et al., 1998; Borst and Sakmann, 1999; Rusakov et al., 1999; King et al., 2000).
Offset of the Cyclic Depolarization
The mechanism underlying the neuronal hyperpolarization at the end of the steady depolarization during slow oscillations or SW complexes is still under debate. Several superimposable mechanisms could shape this component. In the case of the slow oscillation, we have recently shown that variations of the [Ca2+]out contribute to a progressive impairment of synaptic transmission in cortical networks (Massimini and Amzica, 2001), in agreement with the disfacilitation noted toward the end of the depolarizing phase (Contreras et al., 1996) and with the reduced neuronal excitability toward the end of the depolarizing phase of the oscillations (Fig. 9). In spite of this, the neuronal Vm remains at a quasi-stable level throughout the depolarizing phase, and more precisely, beyond the point where the excitability has begun to decline. This phenomenon could be explained by the compensatory effect played by the shift in the Nernst equilibrium potential due to the increased [K+]out. At the same time, after the onset of the depolarizing phase, Ca2+ could also enter glial cells (MacVicar, 1984), and produce Ca2+-dependent release of glutamate (Parpura et al., 1994; Jeftinija et al., 1996; Rutledge and Kimelberg, 1996; Araque et al., 2000), which could further prolong the neuronal depolarization beyond the peak of the glial depolarization.
Another factor contributing to the modulation of neuronal excitability and to the onset of the periodic hyperpolarizations could be the [K+]out itself. It has been demonstrated that moderate increases of the [K+]out (3–5 mM) contribute to enhanced synaptic transmission (Balestrino et al., 1986; Rausche et al., 1990). Such increases are attained toward the top of the glial depolarization during the slow oscillation [(Amzica et al., 2002); and Fig. 9 here]. Further increases of the [K+]out produce decreased synaptic responses (Kocsis et al., 1983; Rausche et al., 1990). Relying on the similar time course of intraglial and [K+]out potentials, and considering the superimposition of the excitability curves (Figs 9 and 10), we suggest that the termination of periodic depolarizations is favored by the crossing of a K+-dependent excitability threshold. The exact value of this threshold could not be established in vivo because of the fact that, especially during seizures, the intraglial potentials might also contain reversed neuronal field potentials (Amzica and Steriade, 2000). However, the crossing points were in the range of the 4–5 mM mentioned in previous studies (Kocsis et al., 1983; Rausche et al., 1990) and it was attained faster during the SW complexes than during the depolarization of the slow oscillation. This could explain why the peak of the glial depolarization is reached before the onset of the neuronal hyperpolarization during the slow oscillation, and why the duration of the depolarization during SW discharges is shorter.
The latter observation correlates with the shorter duration of SW complexes and indicates that the pace of such oscillations may be determined by the amount of K+ expelled in the extracellular space during preceding depolarizing activities. This mechanism could account for the transition from normal slow sleep oscillations toward nocturnal seizures, in agreement with the model of increased K+ seizures (Zuckermann and Glaser, 1968).
These results are at odds with a computational study (Destexhe, 1998), which started with the assumption that GABAB antagonists suppress seizures (Hosford et al., 1992; Snead, 1992), and proposed that the hyperpolarization between ictal PDSs is generated by GABAB inhibitions. First, cortical interneurons discharge during the same depolarizing phase of the slow oscillation (Contreras and Steriade, 1995) and of the SW complex (Steriade et al., 1998b) as excitatory pyramids. Secondly, interneurons should undergo the same modulatory influences of [K+]out and [Ca2+]out and their excitability would therefore drop toward the end of the persistent depolarization. Thirdly, the input resistance of pyramidal cells is increased during the hyperpolarizing phase of these oscillations, suggesting diminished synaptic impingements (Neckelmann et al., 2000). It is therefore unlikely that strong inhibitory activity would occur at the end of the depolarizing phase to trigger the following hyperpolarization.
Another alternative factor contributing to the hyperpolarizing phase of these oscillations could be a Ca2+-dependent K+ current (Steriade et al., 1993a). Although our data do not exclude this possibility, its role might be less important than previously thought, at least in the case of the slow oscillation. The time course of the [K+]out and of the intraglial Vm dropping before the onset of the hyperpolarization and not showing any subsequent increase during the hyperpolarization (Fig. 4A) indicate that no additional K+ current is turned on at the transition between the depolarizing and hyperpolarizing phases. A more pronounced Ca2+ entrance into the neurons during PDSs might increase the contribution of such a current during the ‘wave’ component of the SW complex and explain the delayed peak of the glial depolarization during such states (Fig. 4B).
Cortical network oscillations of the slow sleep or paroxysmal type result from the complex interactions in which neuronal synaptic transmission and glial regulation of the extracellular ionic concentration are involved. Furthermore, the variations of the extracellular ionic environment (mostly of K+ and Ca2+ ions) critically modulate the excitability of the neuronal membrane producing and setting the pace of its oscillatory behavior. The understanding of this approach might yield to therapeutic strategies to prevent nocturnal seizures.
This study was supported by the Canadian Institute for Health Research (grant MT-15 681). F.A. is a Scholar of Fonds de la Recherche en Santé de Québec. M.M. is a doctoral student. We thank P. Giguère and D. Drolet for technical assistance.
|Slow oscillation||PDSs||SW seizures|
|All values are given as means ± SD.|
|Time lags (ms)|
|t1||88 ± 12||2.4 ± 1.3||1.8 ± 1.1|
|t2||−79 ± 21||102 ± 23||97 ± 28|
|t1a||3 ± 1.3||0.5 ± 0.6||0.2 ± 0.4|
|Onset||342 ± 29||17 ± 4.5||1294 ± 289||83 ± 11||1375 ± 243||82 ± 13|
|Offset||–||−13 ± 3.4||–||−22 ± 7|
|Slow oscillation||PDSs||SW seizures|
|All values are given as means ± SD.|
|Time lags (ms)|
|t1||88 ± 12||2.4 ± 1.3||1.8 ± 1.1|
|t2||−79 ± 21||102 ± 23||97 ± 28|
|t1a||3 ± 1.3||0.5 ± 0.6||0.2 ± 0.4|
|Onset||342 ± 29||17 ± 4.5||1294 ± 289||83 ± 11||1375 ± 243||82 ± 13|
|Offset||–||−13 ± 3.4||–||−22 ± 7|