Center for Neuropharmacology and Neuroscience, Albany Medical College, Albany, NY 1
3. ENSEMBLE CODING AND SYNAPTIC PLASTICITY IN PFC
3.1 Neural Assemblies Defined by Membrane Potential
States
Early electrophysiological studies have suggested that distributed networks (neural ensembles) of neurons may mediate information processing in the brain (Hebb, 1949; Kristan and Gerstein, 1970; Eccles, 1971). Recent simultaneous recordings from populations of neurons support this concept (Wilson and McNaughton, 1993; Deadwyler et al., 1996; Nicolelis et al., 1997). Since actual synchronization of action potential firing is either elusive or, at best, weak (Chang et al., 2000), it is possible that ensembles of active neurons are not defined by instantaneous synchronization of spike firing, but by whether a population of neurons is firing or not during a physiologically relevant period. If this is the case, subthreshold membrane potential activity may be a better strategy to define neural ensembles than action potential firing (O'Donnell, 1999, 2003). Thus, information in the PFC may be encoded with ensembles of neurons in their UP or DOWN membrane potential states (Fig. 3A). Since UP state transitions are dependent on excitatory synaptic inputs from other brain structures or cortical regions projecting to the PFC, ensembles of active neurons could be defined as integrating information from the thalamus, limbic structures (hippocampus and amygdala), and other cortical areas including the parietal cortex. The output of PFC neurons as action potential firing is further determined by the arrival of additional inputs during this period. In this sense, PFC neurons are both temporal integrators and detectors of coincident information. This combination renders the PFC suitable for temporal and cross-modal integration of information (Fuster, 1997; Fuster et al., 2000).
UP and DOWN membrane potential transitions have been studied in anesthetized animals. It is unclear whether cortical neurons in awake animals still exhibit such membrane potential fluctuations. The correlation between UP states and slow wave oscillation in the electroencephalogram (EEG) suggests that synchronous alterations between UP and DOWN states in cortical neurons are typical of slow-wave sleep (Steriade et al., 1993; Steriade and Amzica, 1998). Awake animals exhibit higher frequency components in their EEG. However, recent studies also provide indication that sustained depolarization and hyperpolarization can control information processing. For example, cortical activity measured with voltage-sensitive dyes reveals membrane hyperpolarization associated with oculomotor saccades (Seidemann et al., 2002). In addition, in vivo recordings from
striatal neurons in awake monkeys (Kitano et al., 2002) and unanesthetized rats (Wilson and Groves, 1981) indicate the existence of bistable membrane potentials. UP-DOWN membrane potential alternations in anesthetizedanimals resemble slow-wave sleep conditions. Even in those conditions, information processing during UP states may be important for learning and plasticity mechanisms (Steriade, 2001a,b; Lee and Wilson, 2002). It is possible that in awake animals, neuronal populations loose synchrony of membrane potential fluctuations, resulting in disappearance of slow components in the electroencephalogram. In the presence of behaviorally relevant stimuli that activate the mesocortical pathway, a large number of neural ensembles could be set into a persistent UP state (O'Donnell, 2003).
3.2 DA Modulation of Neural Ensembles and Synaptic
Plasticity
The facilitation of UP states may contribute to working memory. A membrane depolarization prolonged by receptor activation can explain the sustained action potential firing typically observed in PFC neurons during working memory tasks in primates. Indeed, but not receptor blockade disrupts sustained spike firing in PFC neurons and working memory performance (Goldman-Rakic, 1995, 1999).
DA may also affect plasticity in the PFC by sustaining UP states. Long- term potentiation (LTP) (Gurden et al., 1999, 2000) and long-term depression (LTD) (Otani et al., 1998; Takita et al., 1999) have been reported in the PFC. DA is known to modulate synaptic plasticity via receptor activation, since both inactivation of the mesocortical projection and receptor blockade disrupt LTP induction in the hippocampal–PFC pathway (Gurden et al., 1999, 2000). A facilitation of synaptic plasticity by DA may be due to receptors sustaining UP states and thereby facilitating NMDA responses by bringing these receptors out of their inactive voltage range. By reinforcing LTP, receptors may ensure the reproducibility of a given ensemble of PFC neurons in the UP state. It is possible that a DA reinforcement of LTD is also voltage-dependent. LTD is more commonly induced in the PFC using the slice preparation (Law-Tho et al., 1995; Otani et al., 1998), in which PFC neuron membrane potential is within the range of the in vivo DOWN state. Although speculative, in the presence of DA and its resulting state-stabilization, LTP may be enhanced only on cells in the UP state, whereas LTD would be the plasticity mechanism enhanced in neurons in the DOWN state. This may be related to pre- and postsynaptic spike timing determining LTP or LTD induction (Markram et al., 1997; Bi and Poo, 1999, 2001). The possibility of DA supporting either LTP or LTD in a
given system is supported by recent evidence that a first DA application may enhance LTD, whereas a second DA application results in LTP induction (Blond et al., 2002). Such a dual effect of DA would certainly contribute to strengthening the pattern of network activity associated with salient stimuli, resulting in the learning reinforcement function that has been proposed for DA (Schultz, 1998, 2002). A combination of synaptic response enhancement during UP states and input attenuation during DOWN states can result in a filtering mechanism by which only strong stimuli (perhaps those effectively reinforced by plasticity) can overcome the “inhibition”; in other words, an increase in the signal-to-noise ratio. The outcome would be that the network of neurons in the UP state during a salient event is both strengthened and filtered of irrelevant information by the multiple facets of DA actions. Memories could be retrieved by the relative ease of reproducing a similar ensemble in conditions resembling the initial context (Fig. 3B, C).