Persistent activity is observed in many cortical and subcortical brain regions, and may subserve a variety of functions. Within the prefrontal cortex (PFC), neurons transiently maintain information in working memory via persistent activity patterns; however, the mechanisms involved are largely unknown. The present study used intracellular recordings from deep layer PFC neurons in vivo and patch-clamp recordings from PFC neurons in organotypic brain slice cultures to examine the ionic mechanisms underlying persistent activity states evoked by various inputs. Persistent activity had consistent features regardless of the initiating stimulus; it was driven by non-NMDA glutamate receptors yet consisted of an initial GABA mediated component, followed by a prolonged synaptically mediated inward current that maintained the sustained depolarization on which rode many asynchronous GABA-mediated events. The stereotyped nature of the multiple-component persistent activity pattern reported here might be a common feature of interconnected cortical networks but within PFC could be related to the persistent activity required for working memory.
Seminal work by Patricia Goldman-Rakic established the idea that working memory, or the ability to transiently maintain information that will be used to guide subsequent action, relies on the prefrontal cortex (PFC) (Goldman-Rakic, 1996). Accordingly, she had argued that dysfunction of the PFC might underlie the cognitive and working memory deficits of schizophrenia (Goldman-Rakic, 1999). This disorder is characterized by alterations in limbic and dopamine inputs to PFC, as well as alterations in synaptic function within PFC (Moghaddam, 1994; Egan and Weinberger, 1997; Lewis et al., 1999). In behaving primates, deep layer PFC neurons exhibit periods of persistent activity that is believed to actively maintain previously encoded information in working memory (Goldman-Rakic, 1996; Fuster, 1998). This type of persistent activity is initiated and modulated by inputs extrinsic to the PFC (Goldman-Rakic, 1996; Fuster, 1998; Seamans et al., 1998). In spite of the ubiquity of persistent activity and its relevance to normal and pathological states involving the PFC, its cellular basis remains elusive.
Persistent activity is not unique to the PFC since spontaneous rhythmic shifts in membrane potential from a hyperpolarized down-state to a depolarized up-state have been observed in a variety of brain regions during various stages of the natural sleep–wake cycle and in reduced cortical slabs or organotypic cultures (Steriade et al., 1993a,b; O’Donnell and Grace, 1995; Plenz and Aertsen, 1996; Plenz and Kitai, 1996; Stern et al., 1998; Lewis and O’Donnell, 2000; Timofeev et al., 2000, 2001; Steriade, 2001; West and Grace, 2002). These up-states share similar characteristics in that they are driven by synaptic input, show a transition in membrane potential to ∼–60 mV for hundreds of milliseconds to tens of seconds, exhibit strong fluctuations in membrane potential and intermittent firing at 2–10 Hz and are synchronous in pairs of cells but evoke asynchronous firing. The similarities in these features across studies suggest that persistent activity may be common property of intact neural circuits. While up-states and the persistent activity underlying working memory could be different, they may share the basic common mechanisms that can be studied in more convenient reduced preparations.
In vivo, cortical up-states of persistent activity may be spontaneous or initiated by stimulation of the hippocampus or neuromodulatory midbrain or brain stem regions (Steriade et al., 1993a; O’Donnell and Grace, 1995; Lewis and O’Donnell, 2000). Moreover, up-states can be modulated by neuromodulatory inputs as well as input from the medial temporal lobe (Fuster, 1998). Stimulation of neuromodulatory inputs produce much more prolonged up-states than those occurring spontaneously in cortex and evoke neural activity that is similar to natural brain state of arousal (Steriade et al., 1993a). In computational models, extrinsic input to realistically simulated deep layer PFC neurons also evokes robust prolonged persistent activity that is critically dependent upon synaptic currents, such as the sustained NMDA current and GABAergic currents (Wang, 1999; Durstewitz et al., 2000). Given the ubiquity of persistent activity and its possible links to working memory processes in PFC, it is of interest to understand the synaptic basis of self-sustained activity evoked by various afferents to PFC such as those arising in the ventral tegmental area (VTA) and hippocampus.
Here we examined the synaptic basis of persistent activity in PFC using in vivo intracellular recordings and whole-cell patch-clamp recordings from PFC neurons in organotypic co-cultures. Persistent activity was evoked by limbic or neuromodulatory afferents to PFC (Steriade et al., 1993a; Lewis and O’Donnell, 2000). Evoked persistent activity in PFC was mediated by a complex interplay of glutamatergic and GABAergic currents, with glutamatergic currents providing sustained depolarization and GABAA currents bombarding the neuron with IPSCs that produced characteristic membrane fluctuations. These unexpected mechanisms allowed groups of PFC neurons to maintain depolarization near threshold for many seconds.
Intracellular In Vivo Recordings
Recordings under current-clamp were obtained from neurons within the prelimbic cortex (coordinates from Bregma: AP 3.2 mm; L –0.6 mm; V 3.2–5.0 mm @ 10° head inclination) of male rats (300–400 g, >p60) anesthetized with an i.p. injection of chloral hydrate (400 mg/kg) and placed in a stereotaxic apparatus. The temperature of the animal was maintained at 36°C with a heating pad. Supplemental anesthesia (as needed) and glutamatergic drugs were delivered via a lateral tail vein catheter. Bipolar concentric electrodes were positioned in the VTA. Sharp intracellular pipettes were pulled from 1 mm Omegadot borosilicate glass tubing using a Flaming–Brown puller, and filled with 1.5% biocytin in 3 M K+ acetate and attached to a head stage connected to a Neurodata amplifier. Custom-made software (Neuroscope; Labview) was used for data collection, storage and analysis.
Organotypic Cultures and Patch-clamp Recordings
P1–P3 rats were anesthetized by placing them on ice for 2–3 min. Sections (350 µm) of the prelimbic region of the PFC and the midbrain (containing the VTA), and/or the hippocampus or the basal forebrain (containing the septum) were obtained in oxygenated, sucrose-substituted solution (Seamans et al., 2001) using a Leica vibratome. Slices were placed next to each other on a Millipore millicell insert in a six-well culture dish. One milliliter of serum-based media fed the slices from below. The plating media contained: 50% basal medium Eagle, 25% Earle’s balanced salt solution, 25% horse serum plus 6.5 mg/ml glucose, 25 mM HEPES–NaOH (pH 7.2), 100 µg/ml streptomycin and Glutamax for first 3 days. Every 3–4 days thereafter inserts were placed in a fresh six-well dish with 850 µl of media as above except 70% basal medium Eagle, 25% Earle’s balanced salt solution and 5% horse serum was substituted. After 15 days, 10 µl of 5-fluoro-2-deoxyuridine (0.08 mM) plus uridine (0.2 mM) in MEM was added to the media to prevent cell division. After an average of 24 ± 1 days in culture a cutout of the tissue with attached Millipore membrane was placed under an upright Zeiss FS2 microscope and viewed using DIC–IR optics. Recordings were made in whole-cell current- or voltage-clamp mode using a HEKA EPC9/3 amplifier running TIDA software. Capacitance and series resistance (5–20 MΩ, 70% compensated) artifacts were compensated automatically and optimized manually. Glass pipettes had tip resistances of 3–7 MΩ when filled with internal solution containing in mM; 110–135 K-gluconate, KMeSO4 or 75–110 K2SO4, 2–10 KCl, 2 MgCl2, 10 HEPES, 1 EGTA, 4 Na-ATP, 0.3 Tris–GTP, 14 phosphocreatine and sometimes 2 QX-314. Liquid junction potentials were corrected when possible. Bathing media, containing in mM 125 NaCl, 3.8 KCl, 25 NaHCO3, 1.2 CaCl2, 1 MgCl2, 10 dextrose and 0.4 ascorbic acid (pH 7.4), was saturated with 95% O2–5% CO2 and maintained at 33–36°C. This modified ‘physiological ACSF’ not only has Ca2+ and Mg2+ concentrations closer to the in vivo conditions (Sanchez-Vives and McCormick, 2000), but also Cl– concentrations that are close native chloride concentrations in the brain. The Cl– reversal potential was calculated based on this external solution and 14 mM [Cl–]i. CNQX (10 µM), bicuculline (10 µM) and d-APV (50 µM) (Sigma, St Louis) were bath applied. N-Methyl-d-aspartate (NMDA,100 µM, 10 ms) was applied via a puffer pipette in the relevant ACSF. A stimulating electrode was placed in the afferent tissue next to the PFC slice and high (>20 mA) or low (<20 mA) intensity square wave (0.2 ms) pulses were delivered. The intensity necessary to produce an up-state varied greatly across preparations. The location of the stimulating electrode relative to the septum was verified by locating acetylcholine-containing neurons in the septal-PFC co-cultures optically using fluorescence for the 192IgG-congugated-Cy3 antibody (1:100) against the P75 receptor (Hartig et al., 1998) that was bath-applied for 30 min at the end of the experiments. The septum was chosen because it is the main source of acetylcholine specifically to the medial PFC/prelimbic cortex as 192IgG-sapporin lesions of the septum most effectively eliminated AchE staining in prelimbic, but not dorsal PFC (Nieto-Escamez et al., 2002; J. Conner, personal communication). For PFC-midbrain co-cultures, the location of the VTA was verified after recordings by processing the tissue for tyrosine hydroxylase (TH) (Gomez-Urquijo et al., 1999). At the end of the recordings, the co-cultures were removed from the Millicell membrane and fixed in 4% paraformaldehyde dissolved in 10 mM phosphate-buffered saline (PBS) for at least 4 h and then placed in PBS. The tissue was pre-incubated in 2% Triton-X in PBS for 24h, and then rinsed again 3 × 5 min in PBS. The tissue was incubated in 1:5000 rabbit TH (New England Biolabs) in 0.1% BSA/PBS for 24 h at 4°C on a shaker. The tissue was rinsed in PBS for 10 min, and incubated in CY3 1:1000 for 1 h followed by 3 × 10 min rinses in PBS, dried and mounted. Alternatively, an avidin–biotin complex was used instead of CY3 and tissue was developed for diaminobenzadine.
In the present study 37% (n = 19/51) of neurons recorded intracellularly in vivo in chloral-hydrate anesthetized animals showed spontaneously occurring depolarizations from –72 ± 2 mV to a mean steady-state potential of –56 ± 2mV. In organotypic cultures 40% (n = 18/45) of PFC neurons showed spontaneously occurring persistent depolarizations from –70 ± 0.4 mV to a mean steady-state potential of –58 ± 1.6mV. However, if a culture was spontaneously active, all recorded cells tended to exhibit persistent depolarizations mainly at irregular intervals, as in cortical slabs (Timofeev et al., 2000). Firing rate during the spontaneous depolarizations varied with the membrane potential, but at a mean Vm of –66 ± 0.4 mV the firing rate was 2.1 ± 0.45 Hz for all cells and 3.7 ± 0.5 Hz with cells that did not fire during the depolarization excluded. These properties are consistent with the properties reported previously for up-states in PFC and other brain regions (Steriade et al., 1993a,b; O’Donnell and Grace, 1995; Plenz and Aertsen, 1996; Plenz and Kitai, 1996; Stern et al., 1998; Lewis and O’Donnell, 2000; Timofeev et al., 2000, 2001; Steriade, 2001; West and Grace, 2002).
Stimulation of neuromodulatory inputs providing cholingergic or dopaminergic input to cortex have been shown to produce persistent activity patterns that were much more prolonged than those occurring spontaneously, sometimes lasting tens of seconds (Steriade et al., 1993a; O’Donnell and Grace, 1995; Lewis and O’Donnell, 2000). In the present study, low intensity single-pulse stimulation of the mibrain region containing the ventral tegmental area (VTA) subthreshold for an up-state, evoked a fast depolarizing response followed by a prolonged hyperpolarization (Fig. 1b). Similar depolarizing-hyperpolarizing responses following VTA, fornix, cortico-cortico or whisker stimulation have been observed in cortical neurons recorded in vivo (O’Donnell and Grace, 1995; Zhu and Connors, 1999; Lewis and O’Donnell, 2000) and appears to be a common effect of activating cortical networks. In the present study, VTA stimulation at higher intensities or with burst patterns (5–10 pulses/20 Hz), effectively produced persistent activity (Fig. 1c) that was 51% longer than that occurring spontaneously in VTA–PFC co-cultures (spontaneous = 3.7 ± 0.5 s, VTA induced = 7.19 ± 0.5 s) and that often outlasted the sampling period of 10s in vivo. However, there was considerable variability across preparations. Persistent activity could also be evoked by other types of stimulation, including the septum in PFC–basal forebrain co-cultures (n = 13; see Methods), stimulation of the CA1 region in hippocampal–VTA–PFC triple cultures (n = 30) or focal application of NMDA in PFC (n = 6, Fig. 2a–d).
Like spontaneous up-states previously recorded from pairs of neurons in vivo (Contreras and Steriade, 1995; Stern et al., 1998; Timofeev et al., 2000), persistent activity in PFC neurons that occurred spontaneously or by synaptic input were highly synchronized between pairs of neurons even though the firing of the neurons was asynchronous (n = 24 pairs). For example, in Figure 2b, following VTA stimulation, one of the two neurons fired vigorously while the nearby neuron did not, in spite of similar membrane potentials. As shown previously in vitro (Sanchez-Vives and McCormick, 2000), in all cases application of a glutamatergic non-NMDA antagonist abolished persistent activity in vivo (12 mg/kg NBQX, n = 4) and in co-cultures (10 µM CNQX, n = 5), indicating that non-NMDA glutamatergic transmission was required for the generation of persistent activity.
Persistent activity that appeared as unitary phenomenon under current-clamp was composed of multiple components revealed under voltage-clamp (Fig. 3). In cells where spontaneous and evoked persistent activity was of similar duration, direct comparison revealed similar components (Fig. 3a,b), and in fact these components were observed regardless of the initiating stimulus. The three main components were: (i) an initial EPSC and large amplitude response that reversed with membrane depolarization (component 1); (ii) a second slower response that was maintained throughout the depolarization (component 2); on which rode (iii) many small asynchronous synaptic events (component 3).
In most cells, stimulation of an afferent input initially evoked an EPSP–IPSP response, similar to that shown in Figure 1b. This EPSP–IPSP sequence made up what we term component 1. When a persistent depolarization was observed the later IPSP of component 1 tended to be much larger than the small initial excitatory response (Figs 1b, 3b and 4). The IPSP of component 1 was also observed when persistent activity was evoked by focal NMDA application directly in the PFC (Fig. 4). In these experiments the puff pipette was located far (>100 µm) from the recorded neuron so as not to evoke a direct excitatory response. This indicated that the IPSP of component 1 could be evoked by a purely excitatory stimulus, so long as it evoked persistent activity. The IPSP of component 1 was inward at hyperpolarized potentials and outward at depolarized potentials (Fig. 4a,b). By fitting a line to the group data (Fig. 4b, n = 23), the reversal potential was –59 mV (Fig. 4b), which was the predicted Cl– reversal according to the Nernst equation (–59.8 mV). Pharmacological investigation in eight co-cultures showed that while the current was not eliminated by bicuculline, it no longer reversed near the Cl– reversal potential but had an extrapolated reversal of near 0 mV and was blocked by CNQX (Fig. 4a). This indicated that at the onset of the persistent activity, glutamatergic currents (revealed after blockade of GABA currents by bicuculline) drove a synchronous GABA-mediated response (i.e. an EPSP–IPSP sequence). In spite of a GABA current mediating the IPSP of component 1, GABA stimulation alone was not sufficient to initiate persistent activity (Fig. 4c). In fact quite the opposite occurred as focal GABA stimulation drove the membrane to near –60 mV and strongly inhibited firing (Fig. 4c). Thus another current is required to bring the neuron from –60 mV to spike threshold.
A prolonged inward current underlied the period of persistent activity that was revealed by smoothing the fluctuations in the membrane current by fitting a double exponential to response in voltage-clamp (Figs 3 and 5b). This current, that underlied the period of persistent activity was blocked or reduced by either non-NMDA antagonists (12 mg/kg NBQX in vivo, n = 4 or 10 µM CNQX in co-cultures, n = 5) or NMDA antagonists (1 mg/kg CPP in vivo, n = 6 or 50 µM APV in co-cultures, n = 12, Fig. 5a,b). The difference was that non-NMDA antagonists blocked all three components (i.e. no persistent activity was evoked) while APV and CPP left component 1 intact while removing component 2, suggesting that component 2 was mediated by NMDA currents. Neurons were then voltage-clamped at different holding potentials and the amplitude of the smoothed current was measured in order to assess whether the voltage dependence of component 2 fit with that of an NMDA current. It should be noted that in order to get an accurate I–V relationship, long inter-stimulus intervals were used (30 s) because at shorter intervals later up-states in the series tended to be of smaller amplitude, possibly due to synaptic depression. In some cells the I–V relationship of the smoothed current resembled that of an NMDA current (not shown), however for most cells and overall the relationship was quite linear (Fig. 5c). This may be because of poor voltage control, or the fact that the NMDA current was counteracted by the large outward IPSCs and long-lasting K+ currents in this voltage range [the importance of intrinsic K+ currents to up-states has been noted previously (Wilson and Kawaguchi, 1996)]. Although the I–V relationship was more linear than expected, it still exhibited an extrapolated reversal between –20 to +20 mV, suggesting it was glutamate mediated.
Component 2 also lasted much longer than a typical NMDA EPSC produced by a single stimulus to layer V (Fig. 5d). In contrast, a 20 ms ‘puff’ application of NMDA resulted in a response (with synaptic transmission and up-states blocked by TTX or CNQX and bicuculline) that was qualitatively similar to the curve fitted to the previously recorded persistent activity state (Fig. 5d, n = 8), suggesting that a prolonged NMDA response could contribute to the period of persistent activity, as suggested previously (Steriade et al., 1993b). Thus the NMDA current has a similar time course and appears necessary to maintain persistent activity; however, the total underlying current is likely the result of the summed inward and outward currents impinging on the cell of both intrinsic and synaptic origin. Moreover, because either NMDA or non-NMDA antagonists completely eliminated persistent activity, the excitatory synaptic drive is likely initiated by a mix of NMDA and non-NMDA currents.
One counter-intuitive aspect of the NMDA contribution to up-states is that robust up-states are routinely observed in animals anesthetized with ketamine (Steriade et al., 1993a,b; Stern et al., 1998) which is an NMDA antagonist. As a test of the hypothesis that NMDA receptors are involved in up-states, Steriade et al. (1993b) showed that a supplemental dose of ketamine (2.5 mg/kg) to animals anesthetized with urethane dramatically reduced up-state duration and spike frequency. But what happens in the case of animals anesthetized with ketamine? We also observed robust up-states in PFC neurons from five rats anesthetized with ketamine (20 mg/kg) and xlyazine (6 mg/kg). In these animals, when a supplemental injection of ketamine (1.5 mg/kg) was delivered a rapid and potent decrease in up-state duration (Fig. 5e) and amplitude was observed. Furthermore, subsequent application of CPP 10 min after the supplemental ketamine injection further reduced up-states, as shown in Figure 5a,e. When viewed in light of the data presented in Figure 5, it suggests that the amount of NMDA blockade by ketamine anesthesia is incomplete, and that the NMDA component of up-states can be reduced further by supplemental application of ketamine or the more selective NMDA antagonist CPP.
Previously it was suggested that during up-states associated with sleep–wake cycles, neurons recorded intracellularly in vivo were bombarded by IPSPs (Timofeev et al., 2001). Accordingly, unitary synaptic events were present during the entire period of persistent activity (Fig. 6a). In the present experiments, most unitary events were eliminated by bicuculline (n = 8) indicating they were mainly mediated by asynchronous GABA-mediated Cl– currents (Fig. 6b). However, it should be noted that in bicuclline the slice exhibited signs of epileptiform activity. Thus the reversal potentials of all events were analyzed in ACSF lacking bicuculline. All events tended to be inward (downward going) at potentials more negative than ∼–60 mV. Figure 6d shows the data from all these responses recorded at potentials below –60 mV plus all outward events (upward going) recorded at potentials above –60 mV. Like component 1, reversal of these events (–61.4 mV) obtained by fitting a line to the group data shown in Figure 6d, occurred near the predicted reversal potential for Cl– (–59.8), given the present Cl– concentrations. The reversal potential measured here using patch-pipettes was very close to the reversal potential of –57 ± 4 mV for fluctuating events in somatosensory cortex recorded during up-states using sharp KAc filled electrodes that do not disrupt the [Cl]i (Plenz and Kitai, 1996). This analysis therefore revealed mainly the inhibitory or Cl– mediated unitary events during the up-state. At holding potentials above the reversal it was possible to dissociate inward (downward going) versus outward (upward going) events in order to determine the relative excitatory versus inhibitory drive. Analysis of up-states from 17 neurons clamped at –40 mV revealed that 69.6% (n = 1684) of all events were outward and 30.4% (n = 512) events were inward at this potential, suggesting that ∼70% of the asynchronous synaptic drive during persistent up-states was inhibitory.
The average frequency of the asynchronous events across all potentials and across all cells was 41 ± 3.9 Hz (Fig. 6c), similar to that reported for somatosensory cortex (Plenz and Kitai, 1996). However, when event intervals of all up-states across all neurons were grouped in 10 ms bins, the first two bins were largest, with the 0–10 ms bin containing 27.9% (6504/23305 events) and the 10–20 ms bin containing 26.7% (6240/23305 events) of all events. This skewed distribution reflects the fact that the neurons with the most events also had the highest event frequency. In all cells the event frequency during the period of persistent activity was higher than the baseline event frequency of 9.4 ± 2.9 Hz during the ‘down state’ (Fig. 6C). However, it is noteworthy that the inward currents measured during up-states from 17 neurons clamped to –40 mV showed an average frequency of 8.9 ± 1 Hz, which was not significantly different from the basal event frequency. Therefore, while large mixed glutamatergic responses seem to underlie the persistent depolarization during up-states (component 2) it is mainly the increase in inhibitory events that are responsible for the fast membrane fluctuations.
The offset of persistent activity was marked by a sharp drop in synaptic activity, supporting the claim that up-states are terminated by a lack of synaptic input (Contreras et al., 1996; Timofeev et al., 2001).
Persistent activity patterns in the present study share many characteristics of up-states reported previously. Both are driven by synaptic input, show a transition in membrane potential to ∼–60 mV, exhibit strong fluctuations in membrane potential and intermittent firing at 2–10 Hz (Steriade et al., 1993a,b; O’Donnell and Grace, 1995; Plenz and Aertsen, 1996; Plenz and Kitai, 1996; Stern et al., 1998; Lewis and O’Donnell, 2000; Timofeev et al., 2000, 2001; Steriade, 2001; West and Grace, 2002) and appear to be a network property since they are synchronous in pairs of cells but evoke asynchronous firing patterns (Contreras and Steriade, 1995; Stern et al., 1998; Timofeev et al., 2000). These similarities raise the possibility that the type of persistent activity reported here for PFC networks may be common property of many intact neural circuits. However, the present study focused mainly on the mechanisms of persistent activity evoked by various inputs that tended to be of longer duration than up-states occurring spontaneously.
Since persistent activity is a network phenomenon requiring sufficient connectivity between neurons it is not observed in acute brain slices from rats or primate PFC [but is present in ferret slices (Sanchez-Vives and McCormick, 2000)] where connections are removed during slicing, but is observed in cortical and subcortical tissue where sufficient connectivity is still present, such as in vivo or in organotypic slice preparations. In organotypic co-cultures of cortex and basal forebrain, thalamus or striatum, afferent fibers entered the cortical region and only formed connections within the layer in which connections normally formed in vivo, regardless of the orientation of the two slices in the culture (Yamamoto et al., 1992; Bolz, 1994; Gomez-Urquijo et al., 1999; Klostermann and Wahle, 1999; Molnar and Blakemore, 1999). Physiological properties of pyramidal, interneurons and striatal neurons in organotypic mono- and co-cultures also closely match those of neurons characterized in vivo or in acute slice preparations (Plenz and Aertsen, 1996; Plenz and Kitai, 1996; Klostermann and Wahle, 1999). These cultures maintain a well-balanced state of excitation and inhibition, suggesting that mechanisms intrinsic to cortex are sufficient for the expression of cell-type-specific electrophysiological properties and persistent activity. Specifically in the case of co-cultures containing the PFC and midbrain, they exhibit similar electrophysiological properties to those observed in vivo (e.g. Fig. 1). In midbrain containing co-cultures tyrosine hydroxylase positive midbrain fibers innervate mainly deep layers of the rat PFC (Gomez-Urquijo et al., 1999; Trantham et al., 2002) and release dopamine (Cragg et al., 1998), similar to the in vivo situation [for a detailed histological description of VTA–PFC co-cultures, see Gomez-Urquijo et al. (1999)]. In both preparations PFC neurons also exhibit up-state behavior that can be initiated or prolonged by VTA stimulation (Lewis and O’Donnell, 2000) (Fig. 1). Nevertheless, we do not presume that the co-culture is identical to the brain of a 24-day-old animal in every respect (see De Simoni et al., 2003). Rather, it is a convenient reduced model system that allows investigation of persistent activity with greater experimental control than the in vivo preparation.
Although midbrain containing cultures contain tyrosine hydroxylase and release dopamine, the contribution of dopamine to up-state initiation was not investigated here but is the subject of ongoing investigations (Trantham et al., 2002). However, as shown previously, VTA stimulation (Lewis and O’Donnell, 2000) or stimulation of other neuromodulatory inputs to cortex such as the locus coeruleus or pedunculopontine nuclei (Steriade et al., 1993b), are capable of initiating the complex mix of synaptic currents that underlie prolonged persistent activity. Thus traditionally ‘neuromodulatory’ inputs to cortex may be capable of evoking persistent activity that is ultimately dependent upon glutamate and GABA currents. The present study focused not on how these neuromodulatory inputs initiate persistent activity, but rather exploited the fact that these subcortical regions are potent activators of such activity.
Persistent activity required non-NMDA currents (i.e. it was eliminated by NBQX or CNQX) to activate a large EPSP/IPSP component (component 1) and a prolonged inward current (component 2) with overlying asynchronous events (component 3). Approximately 70% of all asynchronous events recorded during the period of persistent activity reversed near –60 mV (Figs 3b and 6a) and were eliminated by bicuculline (Fig. 6b). It is not clear why the unitary EPSC rate did not increase significantly during up-states, especially in the presence of bicuculline where almost all events were eliminated. One reason may be that recordings were made from the soma where mostly inhibitory rather than excitatory inputs converge. Recently it was shown that in striatal neurons the Ca2+ signal recorded in dendrites correlated well with up-states recorded from the soma (Kerr and Plenz, 2002). Given the massive drop in input resistance during the up-state, EPSPs on electrotonically remote dendrites may have become invisible to the soma. Nevertheless, from the perspective of the soma (and presumably the axonal spike initiation zone), evoked persistent activity was characterized by a balance of a prolonged inward current and fast IPSPs and not by a balance of fast excitatory and inhibitory unitary events. Very similar results have also been observed in primary cortical cultures (Opitz et al., 2002). Because the characteristic fluctuations in membrane potential were produced by the interplay of glutamate and GABA currents it explains why they were removed by CNQX, APV or bicuculline.
The periodic IPSPs tended to force the membrane potential towards the Cl– reversal potential. In combination with intrinsic currents, these events may bring the mean Vm of up-states to near–60 mV, as is commonly observed (Steriade et al., 1993a,b; O’Donnell and Grace, 1995; Stern et al., 1998; Destexhe and Pare, 1999; Lewis and O’Donnell, 2000; Timofeev et al., 2000, 2001; Steriade, 2001; West and Grace, 2002). These data also highlight the importance of coordinated activity in PFC pyramidal and non-pyramidal neurons to persistent activity, as suggested previously (Goldman-Rakic, 1996).
Both computational modeling (Wang, 1999; Durstewitz et al., 2000) and in vivo intracellular recording studies (Steriade et al., 1993b) have shown that the synaptic drive needed to maintain persistent activity depends strongly on NMDA currents. In support of this claim, the current underlying persistent activity was reduced or eliminated by NMDA antagonists. However, persistent activity was also eliminated by NBQX or CNQX, indicating that activation of NMDA receptors depended on activation of non-NMDA receptors. In many neurons the inward current did not possess the voltage dependence of a pure NMDA current, suggesting that the inward current that drove the up-state is composed of a mix of NMDA and non-NMDA synaptic currents (as well as intrinsic currents). Yet unlike AMPA receptors, NMDA receptors have a long decay time constant and produce a current that can outlast the period of agonist application by as much as 1000-fold (Spruston et al., 1995) and as a result, are capable of producing sustained depolarizations for seconds. This explains how persistent depolarizations could be sustained for seconds despite the asynchronous and low firing rates (<5 Hz) recorded from individual neurons in the network and in spite of the massive inhibitory bombardment (Fig. 6) onto pyramidal cells that might otherwise terminate excitation dependent solely on recurrent firing and EPSPs. However, recurrent AMPA excitation may reinitiate activation of NMDA currents to prolong the period of persistent activity. Indeed the prolonged inward current (component 2) could very likely be the result of summed EPSPs that are smeared out to produce what looks like a large unitary inward current. In many cases, in the presence of bicuculline, massive EPSPs like those shown in Figure 4 were observed throughout the period of the normal up-state. These large EPSPs were of much lower frequency and larger amplitude than the spontaneous events analyzed in Figure 6 and therefore did not contribute to the membrane fluctuations. One possibility is that NMDA currents and possibly other prolonged inward currents bridge the gap between these large infrequent EPSPs to produce what looks like the smoothed EPSP shown in Figure 5. Thus, component 2 may be the result of large AMPA mediated EPSPs that are connected by prolonged NMDA currents.
Although NMDA antagonists CPP, APV and phencyclidine (PCP) (O’Donnell and Grace, 1998) have been shown to block evoked persistent activity, potent spontaneous up-states have been recorded routinely in the presence of ketamine anesthesia (Steriade et al., 1993b; Destexhe and Pare, 1999; Timofeev et al., 2000, 2001). However, supplemental ketamine delivered to anesthetized animals has been shown to significantly reduce up-states (Steriade et al., 1993b). We replicated this observation and found that in animals anesthetized with ketamine/xylazine, up-states were prominent yet were greatly reduced in duration by an additional supplemental dose of ketamine. Since supplemental ketamine to ketamine anesthetized animals reduced up-states, it suggests that the block of NMDA receptors needed for anesthesia was not complete. The differences in the effectiveness of ketamine and phencyclidine versus CPP and APV in eliminating up-states likely reflects the fact the former agents have potent effects on other neurotransmitter systems including dopamine and GABA (Moghaddam et al., 1997; Yonezawa et al., 1998). Furthermore, because PCP and ketamine-like compounds inhibit the NMDA-induced release of GABA in cortex (Drejer and Honore, 1987; Pin et al., 1988), the reduced GABAergic tone might actually facilitate up-state initiation.
During evoked persistent activity, PFC neurons display synchronous depolarizations, not synchronous firing. Therefore the present study actually described a mechanism for producing persistent depolarizations in PFC. Maintaining cortical neurons near threshold in the presence of strong membrane fluctuations is an especially effective means of producing reliable action potential initiation (Mainen and Sejnowski, 1995; Ho and Destexhe, 2000). The prolonged depolarization may produce a ‘ready’ state near threshold, allowing PFC neurons to effectively respond to even small inputs. Sufficiently strong inputs from the VTA or hippocampus that encode information relating to behavioral significance (Schultz, 1997) or context respectively could place working memory buffers into this ‘ready-state’. During this period, PFC neurons may or may not fire (e.g. Fig. 2). However, given a ready state has been evoked, other sources of input perhaps from other cortices, could then effectively cause sustained firing on top of this persistent depolarization. If we extend on these ideas it would imply that during the encoding of behaviorally significant stimuli by VTA neurons, working memory buffers in the PFC should more effectively encode recently acquired information as persistent activity. This is in addition to the prolonged neuromodulatory action of dopamine in the PFC that stabilizes persistent activity, once it is evoked (Durstewitz et al., 2000). Therefore, while persistent activity can be initiated by a variety of extrinsic and intrinsic sources (Fig. 2), the VTA may be unique in that it exerts an additional neuromodulatory control that increases the stability and robustness of already initiated up-states (Durstewitz et al., 2000).
The present findings might also shed light on aspects of PFC dysfunction. As highlighted by Goldman-Rakic (1999) and others (Egan and Weinberger, 1997), schizophrenia is characterized by a dysregulation of PFC function and working memory. The neurobiological deficiencies associated with schizophrenia are exactly those we propose would disrupt sustained activity, namely alterations in midbrain and limbic inputs to PFC, and disruptions in GABAergic and NMDA function in the PFC (Moghaddam, 1994; Egan and Weinberger, 1997; Lewis et al., 1999). Thus, strategies aimed at GABA and NMDA systems that would produce more robust persistent activity patterns in PFC evoked from the VTA or limbic regions may prove beneficial in the treatment of this disease.
The authors thank Dr D. Plenz and Dr I. Timofeev for helpful discussions and Heather Trantham and Chris Lapish and Laurence Neely for help with the preparation and immunostaining of the co-cultures, and Dr Judson Chandler for kindly providing much of the tissue. This work was supported by funds provided by MUSC to J. Seamans and A. Lavin.