Although electrotonic coupling has been described in a variety of central structures in mammals, at least for neocortex it is common in early stages of circuit formation and decreases during later development (Connors et al., 1983; Peinado et al., 1983). Among the exceptions to this rule are the inferior olive in which the morphological correlate of the electrotonic coupling, gap junctions, is present at birth (Bourrat and Sotelo, 1983) and RE neurons in which spikelets were recorded in our experiments on adult cats. In these two structures, the role of electrotonic coupling may be that of a synchronizing device.

Experimental and modeling studies have shown that electrotonic coupling underlies the rhythmicity of complex spike activity in the olivo-cerebellar pathway (Welsh and Llinás, 1997; Makarenko and Llinás, 1998; Loewenstein et al., 2001). A combination of electrical and chemical synapses among local-circuit basket inhibitory neurons has been proposed to entrain fast rhythms, in the gamma frequency range, in rat neocortex (Tamas et al., 2000), and electrical synapses are also thought to generate gamma oscillations in the hippocampus (Draguhn et al., 1998; Traub et al., 1999a; Traub et al., 1999b).

As to the RE nucleus, besides chemical synapses among these GABAergic neurons, which have been implicated in the generation and synchronization of spindle rhythms in experimental (Steriade et al., 1987a) and modeling (Destexhe et al., 1994a; Bazhenov et al., 1999; Bazhenov et al., 2000) studies, electrotonic coupling may be an additional factor in this synchronizing processes. In fact, our computer simulations showed that activity in the RE nucleus can spread not only between pairs of neighboring electrotonically coupled neurons but also at greater distances. This spreading activity could not be due to single spikelets because they are not able to trigger action potentials. However, LTSs may be able to activate a neighbor cell and thus contribute to the propagation and synchronization of spindle activity. This could be expected due to the low-pass properties of gap junctions (Landisman et al., 2002), which strongly filter fast signals (such as action potentials) but not slower signals (as LTSs). Besides a role in spreading slow activities in the RE nucleus, predicted by modeling studies, we propose that electrical coupling in the RE nucleus may be functionally relevant for the synchronization of thalamic oscillations.

The two different modes of membrane bistability are associated with different degrees of neuronal responsiveness. The active state is around the threshold for action potential generation, while the silent state is subthreshold. A broader range of depolarizing inputs’ amplitudes may be processed during the silent state, without the generation of a short-latency and stereotyped spike, than in the active state. On the other hand, small-amplitude EPSPs, which are ineffective during the silent state, may well trigger action potentials during the active state. When excitatory inputs occurring during the silent state are strong enough, transition to the active state might occur. Such transition amplifies the voltage change produced by transient depolarizing signals.

Actually, corticofugal volleys elicit complex depolarizing responses in RE neurons, composed by several EPSPs followed by all-or-none events resembling dendritic spikes or, in less numerous RE neurons, presumably unitary dendritic spikes (Contreras et al., 1993). The dendritic spikes may contribute to the generation of spindle oscillations by boosting distal inputs and depolarizing the soma as well as by triggering dendritic low-threshold spikes (Huguenard and Prince, 1992; Destexhe et al., 1996; Huguenard, 1996) that are crucial in the generation of spindle oscillations. Thus, although an intrinsic membrane property, bistability may strongly be modulated by synaptic activity.

Membrane bistability in a subgroup of RE neurons may play an important role in different patterns of spindles displayed by thalamocortical neurons. Consequently, any change in the bursting pattern of RE neurons would affect their targets, thalamic relay neurons. Our intracellular recordings of thalamocortical cells showed at least two different patterns during spontaneously occurring spindles. Although simultaneous recordings of RE and thalamocortical neurons have not been performed in the present experiments, the two patterns displayed may be related to the actions exerted by non-bistable and bistable RE neurons, respectively. Indeed, non-bistable neurons fired stronger bursts, with higher intra-bursts frequencies, which are assumed to generate deeper and longer IPSPs in TC neurons, giving rise to the usual frequency range of spindles under barbiturate anesthesia, ~7-10 Hz. By contrast, IPSPs with lower amplitudes and higher frequency, up to 20 Hz, are likely to be mainly generated by single action potentials in RE neurons, as they occur during the depolarizing plateau in bistable cells. In either case, the crucial role of RE neurons in initiating spindles, even in the absence of feed-back excitatory effects from thalamocortical neurons, is shown by the absence of rebound bursts with fast action potentials after the first three or four IPSPs in relay cells (Timofeev et al., 2001). Supporting these results, computational models of thalamic networks, including bistable RE neurons, showed a significant shaping of thalamic oscillations in thalamocortical neurons by bistable RE neurons. While spindles are initiated in the RE nucleus (Steriade et al., 1985; Steriade et al., 1986; Steriade et al., 1987a), this oscillation is maintained by reciprocal actions between RE and thalamocortical neurons (Steriade et al., 1993a; von Krosigk et al., 1993; Bal et al., 1995a, b).

Therefore, one single intrinsic membrane property can have important effects on the network behavior, as is the case of membrane bistability which may sculpt the shape of spindle in thalamic networks. However, the other way around is also true, since ongoing synaptic activity can modulate the dynamics of intrinsic membrane bistability in single cells in the RE nucleus.

The effect of synaptic background activity on ongoing cellular activity has previously been studied, especially for the cortex, with computational models (Ho and Destexhe, 2000; Destexhe et al., 2003; Fellous et al., 2003), through the generation of different artificial states of background noise in vitro (Stacey and Durand, 2001; McCormick et al., 2003; Mitchell and Silver, 2003), and through the activation of synaptic potentials or local suppression of network activity in vivo (Destexhe et al., 1996; Timofeev et al., 1996). Those studies showed that membrane depolarization and small to moderate membrane fluctuations may facilitate the responsiveness to synaptic inputs, especially those of lower amplitude. Our results showed that volleys of synaptic activity can result in significant changes in neuronal excitability and responsiveness in RE neurons.

Even if spindles were not to be considered as cortical high-conductance states, they are clearly active network states. Active network states imply a series of consequences for integrative properties of neurons. One consequence is that responsiveness of neurons is markedly different in the presence of fluctuating background synaptic activity. Due to the presence of membrane potential fluctuations, neurons respond stochastically to a given stimulus, and their behavior is best described by probability functions; this has been demonstrated for cortical cells (Ho and Destexhe, 2000). Consistent with these observations in the cortex, our experiments showed that RE neurons respond to cortical activation in a way that is well fitted by probability functions. Another effect of active network states is on temporal processing. The reduction of the space constant in states of high conductance is accompanied by a marked reduction in the membrane time constant (Destexhe et al., 2003). This result was apparent in our experimental data, since active responses were faster to injected positive current pulses. As it has been proposed, the reduced time constant should favor finer temporal discrimination of distant synaptic inputs. Modeling studies have predicted that cortical neurons can resolve higher frequency inputs in active membrane states than when silent; therefore, cortical cells in high-conductance states can efficiently track synaptic inputs (Destexhe et al., 2003). It is expected that a similar case might be found in RE neurons, though evidence should be provided by future experiments.

Thus, our results support the idea that synaptic and intrinsic activities in the RE network is highly modulated at the cellular level by thalamic oscillations, which are generated by the RE nucleus itself.

The presence and contribution of persistent Na⁺ currents cannot be discarded since its range of activation is close to the threshold for spike generation (5-10 mV below) and it is also blocked by QX-314 (Crill, 1996). Besides, studies in inhibitory neurons from the hippocampus have demonstrated the presence and importance of active dendritic Na⁺ conductances (Martina et al., 2000). It is thus possible that a joint contribution of Ca²⁺ and Na⁺ dendritic currents acts in RE neurons for integration of cortical inputs.

Another effect of QX-314 on evoked responses was the two- to three-fold decrease in the rising slope. The rising phase of any active response depends on both its location in the dendritic tree and the active conductances that regulate its conduction to the soma (Rall, 1995; Magee, 2000). In our experiments, electrical stimulation of corticothalamic fibers frequently elicited fast-rising responses (5-10 mV/ms slope) in RE neurons. Recent experiments have shown that cortically evoked EPSPs are mainly AMPA-receptor-dependent, with a negligible element of NMDA-receptor response, which is virtually absent at hyperpolarized levels of membrane potential (-70 mV, (Gentet and Ulrich, 2004)). Even though the effect of QX-314 on the rising phase was not voltage dependent, a clear reduction was seen at different voltages in all cases, suggesting an additive contribution of active conductances to the AMPA component in the response.

The voltage dependent dendritic channel distribution in RE neurons provides them with important properties for integration of cortical inputs. We suggest that, in comparison with the effects these inputs exert on thalamocortical neurons, at which level dendritic currents generate high-frequency rhythms (Pedroarena and Llinás, 1997) that define brain alertness, the parallel activation of RE-cells’ dendrites by synchronized cortical volleys produces low-frequency oscillations (Contreras et al., 1993; Contreras and Steriade, 1996) that characterize the disconnected state of slow-wave sleep.

PHPs were present spontaneously in one third of the neurons studied (10 of 32), and in those cells, it was detected in ~60% of spindles, in apparently random fashion. Activation of corticothalamic afferents was able to elicit PHPs with similar characteristics as spontaneous ones (amplitude and duration), but in all those cases it was preceded by evoked EPSPs. The spontaneous PHP was itself occasionally headed by EPSPs (~30%) whose origin could therefore be attributed to corticothalamic inputs. These results suggest the possibility that PHPs might have, at least in some cases, a cortical origin.

Both spontaneous and evoked PHPs produced a significant drop in the R_in of RE neurons, suggesting the involvement of active inhibition processes. Therefore, disfacilitation processes, like those occurring during slow-oscillation activities (Steriade et al., 2001) were discarded as possible origin of the long hyperpolarizations heading spindles. The reversal potential of PHPs was quite hyperpolarized (~-100 mV), suggesting the activation of K⁺ conductances (Huguenard and Prince, 1994; Ulrich and Huguenard, 1996b). The reversal potential for GABA_A-mediated IPSPs in the RE nucleus has been estimated in ~-70 mV (Zhang et al., 1997; Shu and McCormick, 2002). However, PHPs displayed significant amplitude at that range of membrane potential, discarding the possibility of GABA_A-mediated IPSPs implicated in the generation of PHPs. Moreover, the I-V relation for PHPs was linear for a large range of membrane potential values (-110 mV to -50 mV), implying a very little contribution, if not any, of GABA_A potentials at membrane potentials more depolarized than -70 mV. Consistent with the activation of K⁺ conductances during PHPs, QX-314 (50 mM) in the pipette decreased both their amplitude and incidence. QX-314 affects a wide range of membrane conductances, however, given the fact that PHPs are inhibitory potentials; it is unlikely that depolarizing currents (Ca²⁺ and Na⁺) are implicated in the genesis of PHPs. On the other hand, it has been described that QX-314 blocks also G-protein-activated, inwardly rectifying K⁺ (GIRK) channels (Andrade, 1991). This fact opens the possibility that PHP is an inhibitory potential activated by membrane metabotropic receptors coupled to second messenger cascades (Hille, 1992, 1994). Candidates for such an effect are GABA_B receptors (Ulrich and Huguenard, 1996a), as well as the peptidergic receptors for somatostatin (SST) (Sun et al., 2002) and neuropeptide Y (NPY) (Sun et al., 2001a; Sun et al., 2001b). All these molecules are expressed by RE neurons, as well as their respective agonists (Graybiel and Elde, 1983; Bendotti et al., 1990; Hoyer et al., 1995; Ulrich and Huguenard, 1996b; Sun et al., 2001a). Additionally, all these molecules initiate second-messenger cascades which end up in the activation of GIRK channels (Hille, 1994; Dolphin, 1998; Yamada et al., 1998). The case of GABA_B receptors might be controversial, since experiments in slices have shown that rodents express very little GABA_B responses in the RE nucleus (Ulrich and Huguenard, 1996b), contrary to the case of thalamocortical neurons, where GABA_B responses are quite large (Ulrich and Huguenard, 1996b). Baclofen application to thalamic cells activated a K⁺ conductance that was four times smaller in RE neurons with relay neurons (Ulrich and Huguenard, 1996b). These data suggested that the number of available GABA_B receptors on RE neurons is smaller, and that the slow IPSPs that occur in only a subpopulation of RE neurons may remain subthreshold for LTS generation. However, these experiments performed in vitro did not asses the critical point, which is to know how much GABA_B conductance is activated during thalamic oscillations, in which large populations of neurons fire in synchrony (Contreras et al., 1996a; Contreras and Steriade, 1996). Also, experiments in ferret slices have shown not so small GABA_B components in response to glutamate applied in the perigeniculate nucleus (Sanchez-Vives and McCormick, 1997). In those experiments early bursting rebounds were blocked by the application of CGP 35348, and antagonist of GABA_B receptors (Sanchez-Vives and McCormick, 1997). Therefore, the possibility that the RE nucleus expresses a significant GABA_B response during spindle oscillations cannot be discarded. From our experiments it is not possible to asses the precise ionic bases of PHP in RE neurons, but we can propose that at least one of the mentioned receptors (GABA, SST and NPY) constitutes the molecular basis of PHPs in the RE nucleus.

Computational models as well as experimental evidence have suggested strong electrical compartmentalization between dendrites and soma in RE neurons (Mulle et al., 1986; Destexhe et al., 1996). It is therefore reasonable to expect that an input to the distal dendrites would not have a dramatic effect on the soma, if its transfer to the soma depends only on passive properties. In the same way, current injection in the soma might well exert a local effect and not reach the dendrites. Early experiments in the RE nucleus showed that neurons have to be strongly hyperpolarized in the soma in order to be able to de-inactivate T-current and generate LTSs (Mulle et al., 1986). This was interpreted as distal location of T-current from the soma, likely dendritic. Indeed, later experiments proved the dendritic location of T-current in RE neurons (Destexhe et al., 1996). A similar reasoning suggested that PHP might well have a dendritic origin, since strong soma depolarization did not always increased PHP amplitude, but abolished it leaving instead a low-frequency discharge period. GIRK channels could be preferentially located in dendrites where they might mediate dendritic hyperpolarizations (Sun et al., 2001b). This is consistent with the idea of dendritic genesis of PHPs.

Our simultaneous intracellular and extracellular recordings proved that some neurons in the RE network were able to discharge during intracellularly recorded PHPs. These results suggest, although they do not directly demonstrate, that PHPs are locally generated in the RE nucleus. The main inputs to the RE nucleus arise from dorsal thalamic nuclei, but especially from the cortical areas (Liu and Jones, 1999). However, all of them are of excitatory nature (Liu and Jones, 1999), and cannot be responsible for the direct generation of PHPs. The possibility of a disynaptic cortico-RE origin was supported by our results.

We propose that some cortical neuronal assembles may fire synchronously, thus activating RE neurons by their collaterals to the thalamus. Those RE neurons receiving projections from the cortex will display EPSPs of variable amplitudes, in some cases eliciting powerful LTSs. Indeed, our experiments show that to electrical stimulation of corticothalamic fibers induces responses which vary widely from cell to cell, even though the same stimulation parameters are conserved. Thus, a fraction of RE neurons will be excited and discharge over their neighbors, generating in some cases PHPs. In fact, there is evidence in RE neurons for the presence of both axo-axonic and dendro-dendritic synapses (Pinault et al., 1997), which mediate inhibitory interactions (Sanchez-Vives and McCormick, 1997; Shu and McCormick, 2002). Thus, the activation of PHPs might be relevant for the generation of spindle oscillations since these long-lasting potentials can hyperpolarize RE membranes to de-inactivate T-current, and thus elicit spike-bursts, essential for spindle oscillations (Steriade et al., 1990).

The model we propose is not closed or exclusive, since it supports the large line of work dedicated to elucidate the cellular basis and functional significance of spindles, as highly coordinated and complex corticothalamic oscillations, initiated in the RE nucleus, maintained by RE-thalamic interactions, and later on transferred to the cortex.

The presentation of the model will start on an arbitrary point. However, given the cyclic nature of these oscillations any point in the circuit could be chosen to proceed.

Thus, in an initial step, the simultaneous activation of a significant number of corticothalamic inputs will generate EPSPs of variable amplitude in RE neurons, depending on the current state of activity and the synaptic contacts made on that particular sector of the RE nucleus. The existing state of activity in conjunction with the integrative properties of the secondary depolarizing component of the EPSP, will then determine the output for every single cell receiving the cortical input. As a result, some RE neurons will display a subthreshold EPSP, while others will present a suprathreshold EPSP, whose spiking output might well be controlled by the properties of the secondary depolarization. If the cortical volley is strong enough, some other RE neurons will display an EPSP which will develop into a LTS, with several consequences on the neighboring cells (1, Fig. 9.1).

Some RE neurons are connected synaptically with some of their neighbors (~10%). Therefore, a presynaptic LTS will be transmitted not only to the relay nuclei from the dorsal thalamus, but also to some RE neurons were it will generate a prominent and prolonged hyperpolarization. Such hyperpolarizing potential generated by inhibitory neurons from the RE nucleus itself will deinactivate in some cases the T-current of the target neurons. In this way, a favorable environment for the generation and propagation of spindle activity will be attained in the RE nucleus (2, Fig. 9.1).

In addition to intrinsic chemical synaptic contacts, RE neurons also present electrical synapses which could display an important role in the propagation of thalamic oscillation. As LTSs are generated in RE neurons, neighboring cells connected by means of gap junctions will suffer a change in activity. Spike-to-spike synchronization is precluded due to the low-pass filter properties of gap junctions; however slow activities as the depolarizing envelope of LTSs is well transmitted through electrotonic contacts. Therefore, spindle waves could not only be propagated by electrical synapses but also synchronized to long distances (3, Fig. 9.1).

Coming back to the cortically generated EPSPs, which can generate multiple responses in target RE neurons, they can also produce the transition to a sustained active state in a subpopulation of cells. This sustained active state was shown to be related with the presence of membrane bistability, due to the generation of two clear-cut membrane modes in RE neurons recorded at resting conditions. Such a behavior is a consequence of intrinsic membrane properties of RE neurons and is relevant for the integration of incoming inputs. Most important however is the relevance for network activities in the thalamus. The activation of membrane bistability in some RE neurons seems to be sufficient to shape the general form of spindles not only in RE neurons, but also in thalamocortical neurons, the major targets of the RE nucleus and the secondary element in the generation of thalamic oscillations. Thus, while thalamocortical neurons receiving inputs from canonical RE neurons would display the typical waxing and waning pattern of sequential IPSPs during spindles; some thalamocortical neurons will receive also inputs from bistable RE neurons, and in those cases their spindling pattern should be different, affecting the general properties of the network oscillations (4, Fig. 9.1).

Finally, all these properties of RE neurons which are related to spindles are modulated by this oscillation. In fact, both synaptic and intrinsic membrane properties of RE neurons are modulated by spindles, both being enhanced during active network states. This suggests that the spindle rhythm is not necessarily a condition of low brain activity as has been classically considered, but it represents an internal, active state with its own dynamics (5, Fig. 9.1).