Abstract

Local cortical networks in the prefrontal cortex and visual cortex are capable of spontaneously generating sustained activity for periods of seconds or longer. This sustained activity is generated through recurrent excitation between pyramidal cells that is controlled by feedback inhibition and can have both a rapid onset and a rapid offset. The period of activity is associated with a marked increase in neuronal responsiveness to the intracellular injection of current pulses, especially those of smaller amplitude. Independently mimicking the depolarization, increase in membrane conductance and increase in noise associated with sustained activity revealed that the depolarization is largely responsible for the increase in neuronal responsiveness, although an increase in membrane noise also facilitates responses to small inputs. These results indicate that the persistent activity associated with the performance of working memory tasks may be generated largely through recurrent networks. They also suggest that feedback pathways, such as those involved in selective attention, may exert a powerful influence on neuronal responsiveness through synaptic bombardment.

Introduction

Neuronal activity in the monkey prefrontal cortex can exhibit a persistent discharge during the ‘memory’ or delay period of a task that requires the animal to remember specific features, such as location, of a target (Funahashi et al., 1989; Fuster, 1995; Goldman-Rakic, 1995; Miller et al., 1996). The occurrence of persistent activity during working memory tasks is not unique to the prefrontal cortex, and indeed is found in multiple cortical, and even subcortical, areas (Fuster and Alexander, 1973; Fuster and Jervey, 1981; Miyashita and Chang, 1988; Hikosaka et al., 1989; Fuster, 1995; Pesaran et al., 2002). Computational models of this persistent activity typically rely upon the operation of recurrent excitatory networks that are controlled by the activation of inhibitory interneurons (Amit and Brunel, 1997; Lisman et al., 1998; Wang, 1999, 2001; Compte et al., 2000, 2003a; Durstewitz et al., 2000b; Timofeev et al., 2000), although it is also possible that intrinsic membrane properties contribute significantly (Marder et al., 1996; Camperi and Wang, 1998; Haj-Dahmane and Andrade, 1998; Egorov et al., 2002). While computational studies have examined the possible role of different components of neuronal networks in the generation of persistent activity, it has been difficult to examine the cellular mechanisms in vivo, owing in part to the problems of performing intracellular recordings in awake, behaving animals.

Recently, we reported that slices of the ferret prefrontal and occipital cortex generate recurring periods of sustained (persistent) activity when maintained in vitro in an ionic solution that more closely mimics normal cerebrospinal fluid than is traditionally used for maintaining slices (Sanchez-Vives and McCormick, 2000; Shu et al., 2003). This recurrent activity is composed of so-called ‘UP’ and ‘DOWN’ states and appears nearly identical to that underlying the generation of the cortical slow oscillation in vivo during anesthesia or sleep (Steriade et al., 1993a, 2001; Contreras and Steriade, 1995; Contreras et al., 1996). We propose that the mechanisms involved in the generation of this form of recurring sustained activity may have important implications regarding those underlying the generation of persistent activity during working memory tasks in the same neuronal circuits.

Changes in background synaptic activity may not only underlie the generation of persistent neuronal discharge, but also may have strong influences on neuronal excitability and responsiveness to other synaptic inputs (Bernander et al., 1991; Timofeev et al., 1996; Destexhe and Paré, 1999; Anderson et al., 2000a,b; Destexhe et al., 2001; Chance et al., 2002; Fellous et al., 2003; Mitchell and Silver, 2003). Increases or decreases in background synaptic activity may mediate the influence of feedback pathways or local networks on the responsiveness of single neurons. Thus, rapid changes in barrages of fast excitatory and inhibitory postsynaptic potentials may provide a mechanism for neuronal ‘context’ such as the position of gaze in visual–motor transformations (Andersen and Mountcastle, 1983; Andersen et al., 1985; Salinas and Abbott, 1996), selective attention (McAdams and Maunsell, 1999a,b; Treue and Martínez-Trujillo, 1999; Reynolds et al., 2000; Chance et al., 2002), long-range receptive field influences (Carandini and Heeger, 1994; Angelucci et al., 2002; Stettler et al., 2002) and feedback from higher cortical areas (Hupé et al., 1998).

Here we use intracellular and extracellular recordings from ferret prefrontal cortical slices to examine the cellular mechanisms underlying the generation of local sustained activity and how this may influence the neuronal responsiveness of cortical neurons to afferent inputs. These results are relevant to the possible mechanisms of persistent activity during short term memory tasks and may shed light on the mechanisms by which network activity influences neuronal processing.

Methods

Slices (0.4 mm thick) were prepared on a DSK microslicer from the medial prefrontal cortex (anterior to the presylvian fissure) of 2- to 4-month-old ferrets of either sex. The ferrets were deeply anesthetized with sodium pentobarbital (40 mg/kg) and killed by decapitation. The entire forebrain was rapidly removed and the hemispheres were separated with a midline incision. A modification of the sucrose-substitution technique developed by Aghajanian and Rasmussen (1989) was used to increase tissue viability. During preparation of slices, the tissue was placed in a solution in which NaCl was replaced with sucrose while maintaining an osmolarity of 307 mOsm. After preparation, slices were placed in an interface style recording chamber (Fine Sciences Tools, Foster City, CA). For the first 10 min cortical slices were superfused with an equal mixture in volume of the normal bathing medium and the sucrose-substituted solution. Following this, normal bathing medium was switched into the chamber and superfused the slices for ∼1 h, then modified slice solution was used through out the remained of the experiment. Bath temperature was maintained at 34.5–36°C.

Simultaneous extracellular multiple unit and intracellular recordings were performed in layer 5 after at least 2 h of recovery. The normal bathing medium contained (in mM): NaCl, 126; KCl, 3.5; MgSO4, 1; NaH2PO4, 1.25; CaCl2, 1–1.2; NaHCO3, 26; dextrose, 10, and was aerated with 95% O2/5% CO2 to a final pH of 7.4. Sharp intracellular recording electrodes were formed on a Sutter Instruments (Novato, CA) P-80 micropipette puller from medium-walled glass (1B100F-4, WPI Sarasota, FL) and beveled on a Sutter Instruments beveller to final resistances of 60–100 MOhms. Micropipettes were filled with 2 M KAc.

Dynamic Clamp and Data Analysis Techniques

After obtaining a stable intracellular recording from a layer 5 neuron, a number of experiments were performed with a dynamic clamp technique using a DAP-5216a board (Microstar Laboratory; Dorval et al., 2001). To test the responsiveness of cortical neurons in different states, different amplitude current pulses (typically 400 ms duration) were injected into the intracellularly recorded neurons. The effects of various manipulations on the resulting f–I plots, as well as in spike timing, were then examined. The dynamic clamp system operated at 10  000 Hz in standard current clamp (bridge) mode or ∼3000 Hz in discontinuous current clamp (DCC) mode using an Axon Instruments Axoclamp-2B amplifier. In current clamp mode, the bridge was carefully balanced with the periodic injection of current pulses. In DCC mode, the headstage output was continuously monitored to ensure adequate settling time during the injection of current. The large majority of experiments were performed in DCC mode, since this was less sensitive to changes in bridge balance. Data collected were carefully monitored for problems with the dynamic clamp technique, which typically appear as breaks or incontinuities in the current injected. Data containing problems with the dynamic clamp technique were discarded.

The extracellular and intracellular data were collected using Spike-2 software (Cambridge Electronic Design; Cambridge, UK). The data was segregated into three periods, UP, DOWN, and in between states, according to activity in the multiple unit recording. The multiple unit recording was rectified and smoothed to yield an outline of the increase and decrease in activity associated with the onset and offset of the UP state. Two thresholds were positioned and the multiple unit activity was required to rise above the highest threshold to be counted as an UP state. The onset and offset of the UP state were then determined by the crossings of the lower threshold, which was set ∼2× baseline. Only UP states that were at least 0.5 s in duration were considered. DOWN states were defined as those periods for which the activity level stayed below the lower limit for a period of >0.5 s. Periods that were neither classified as UP nor DOWN were not analyzed.

In addition to the injection of current pulses, we also manipulated the level of membrane potential, conductance, and variance (noise) occurring within the cell at the recording (injection) site. Membrane potential was manipulated through the intracellular injection of steady current. Changing membrane conductance was achieved with the dynamic clamp system with the reversal potential of the conductance set to the measured resting membrane potential of the cell. Noise was modeled according to Destexhe et al. (2001). The total synaptic current, Isyn, was composed of two independent conductances according to the following:

Isyn = ge(t)(VEe) + gi(t)(VEi)

where ge(t) and gi(t) are the time-dependent conductances calculated as a one-variable stochastic process similar to the Ornstein–Uhlenbeck process (Uhlenbeck and Ornstein, 1930; Destexhe et al., 2001). In our model Ee is 0 mV and Ei is –75 mV and mean ge was typically equal to mean gi. The tau for correlations within Ge was typically 30 ms for both excitation and inhibition.

Results

Intracellular recordings were obtained from layer 5 regular spiking cells in the ferret medial prefrontal (n = 51 cells) cortex maintained in vitro. Maintaining these cortical slices in an ionic medium that more closely mimicked that of in vivo (see methods) resulted in the spontaneous occurrence of so-called UP and DOWN states (Fig. 1; Sanchez-Vives and McCormick, 2000). UP states, which were typically from 0.5 to 3 s in duration, were associated with an increase in multiunit activity (Fig. 1A; 0.5 s was the shortest acceptable duration to be considered an UP state) and a 4–10 mV depolarization mediated by synaptic potentials (see Table 1). Multiple unit recordings revealed that the start and stop of the UP state was synchronized between neighboring neurons (Fig. 1). Examining the onset and offset of the changes in membrane potential associated with the transitions between the UP and DOWN states revealed that they were relatively rapid, often occurring within a time frame of <0.2 s (Fig. 1). Extracellular single unit recordings (n = 32 cells), performed in conjunction with multiple unit recordings, revealed that single layer 5 cortical cells discharge during the UP state at a rate that ranged from 2 to 47 Hz (mean ± SD = 17.1 ± 11.1 Hz). In the DOWN state, 18 of these cells did not spontaneously generate action potentials, while the remaining 14 cells discharged at a low rate that averaged 3.6 ± 3.0 Hz (n = 14 cells). Recordings in supragranular layers revealed relatively little spontaneous activity during the down state (Sanchez-Vives and McCormick, 2000). Thus, the recurring sustained activity generated in prefrontal slices was associated with a low level of discharge in layer 5 during the DOWN period, intermixed with up to several seconds of maintained discharge of ∼10–20 Hz during the UP period.

Intracellular recordings and manipulation of the membrane potential with the intracellular injection of current revealed that the UP state is generated by the arrival of barrages of EPSPs and IPSPs, and did not provide any evidence for the involvement of persistent inward currents. If the UP state was generated through mechanisms intrinsic to the cell recorded, then hyperpolarization of the cell should have a significant effect on the duration or rate of recurrence of the UP state. However, hyperpolarization of cortical neurons by 10–25 mV, sufficient to prevent action potential discharge during all periods of the activity, did not affect the duration (1.7 ± 0.3 s control; 1.7 ± 0.4 s hyperpolarized) or the rate of recurrence (0.2 ± 0.1 Hz control; 0.2 ± 0.1 Hz hyperpolarized; n = 16 cells) of the UP state (see also Sanchez-Vives and McCormick, 2000).

Recording intracellularly with sharp microelectrodes that contained 50 mM QX-314 to block voltage dependent Na+ currents (e.g. INat and INap) and 2 M CsAc to reduce K+ conductances allowed us to directly examine the voltage dependence of PSP barrages underlying the UP state (Sanchez-Vives and McCormick, 2000; Shu et al., 2003). Depolarization or hyperpolarization of pyramidal cells revealed barrages of fast events at all membrane potentials (e.g. Fig. 1B). At resting membrane potentials, these fast events appeared to ride on top of an envelope of depolarization, the amplitude of which depended upon the membrane potential of the cell. Depolarization with the intracellular injection of current reduced the amplitude of the depolarizing envelope associated with the UP state, and if the membrane potential was depolarized with current past –30 to –40 mV, then the PSP barrage associated with the UP state resulted in an overall hyperpolarization rather than a depolarization (Shu et al., 2003). This activity therefore behaved as a roughly balanced mix of EPSPs and IPSPs, with a reversal potential approximately half way between the reversal potential of GABAA mediated inhibition (–75 mV) and glutamatergic excitation (0 mV) (Shu et al., 2003). Indeed, minimizing the contribution from GABAA-mediated inhibition (by keeping the cell near ECl) revealed strong barrages of excitatory postsynaptic potentials during the UP state. Similarly, depolarizing the cell to 0 mV, and thus minimizing the contribution from glutamatergic excitation, revealed the UP state to be associated with strong barrages of IPSPs throughout its entire duration (not shown). The presence of a strong mixture of EPSPs and IPSPs in the membrane depolarization of the UP state is consistent with the fact that the spike rates of both pyramidal neurons and fast spiking interneurons greatly increase during persistent network activity (Sanchez-Vives and McCormick, 2000; Shu et al., 2003).

The critical involvement of recurrent excitation in the generation of sustained activity of the UP state was also demonstrated by examining the effects of bath application of the non-NMDA glutamate receptor antagonist CNQX (10 µM) and the NMDA receptor antagonist dl-APV (25 µM). Bath application of CNQX resulted in a rapid and complete abolition of UP states in both visual and prefrontal cortical slices (n = 6 visual; n = 4 prefrontal). Interestingly, block of NMDA receptors also resulted in a block of the UP state in the prefrontal cortical slices (n = 5), and a block or strong reduction of this activity in the visual cortical slices (Sanchez-Vives and McCormick, 2000). Thus the generation of the UP state depends critically upon the activation of both NMDA and non-NMDA receptors at fast, glutamatergic synapses. In the presence of both CNQX and APV, a subset of layer 5 neurons remained spontaneously active (Compte et al., 2003a), presumably through intrinsic ionic mechanisms. We assume that these spontaneously active layer 5 cells may be involved in the initiation of spontaneously occurring UP states (Fig. 2B).

The intracellular injection of depolarizing current pulses alone never resulted in action potential activity that outlasted the duration of the current pulse, whether the current pulse was injected during the DOWN or UP state (e.g. Fig. 3A; n = 10 cells; see, however, Timofeev et al., 2000). In addition, the depolarization and action potential activity occurring in single neurons never significantly outlasted the activity occurring in local networks, as indicated by nearby multiple unit recordings (e.g. Fig. 1). In other words, we found no evidence for the natural generation of plateau potentials or the ability to normally generate persistent depolarizations through intrinsic cellular mechanisms in our intracellularly recorded cortical neurons. The action potential activity occurring in each cell that we recorded could be explained simply by the barrage of synaptic activity arriving from the local network.

In contrast to the effects of glutamate ionotropic receptor antagonists, local or bath application of GABAA receptor antagonists (e.g. picrotoxin, bicuculline) resulted in a transition of the UP states into epileptiform activities including paroxysmal bursts of spikes and excitatory postsynaptic potentials (Sanchez-Vives and McCormick, 2000). Together these results indicate that the UP state is generated through recurrent excitation between neighboring pyramidal cells within the cerebral cortex. This recurrent excitation is rapidly and precisely controlled by the activation of local inhibitory interneurons such that a balance is achieved allowing for the generation of a relatively stable membrane potential with a discharge rate (Shu et al., 2003) that is within the range observed in vivo in awake, behaving monkeys performing delayed response memory tasks (Funahashi et al., 1989; Miller et al., 1996).

Impact of Sustained Activity on Neuronal Responsiveness

Neurons in the nervous system are constantly bombarded with synaptic activity (e.g. Destexhe and Paré, 1999; Steriade et al., 2001). The level of this ongoing activity is not static: it dynamically changes, resulting in significant changes in the input–output properties of cortical neurons. Presumably, for example, the generation of sustained activity in prefrontal and other cortical areas modulates the response properties of these cells to the arrival of other synaptic inputs (e.g. Chance et al., 2002). Here we examined how barrages of synaptic activity during the UP state modulate neuronal responsiveness to the intracellular injection of current pulses and sine wave currents of varying frequencies.

The intracellular injection of depolarizing current pulses during the UP and DOWN states (n = 10 cells) revealed that the UP state, and the barrages of synaptic activity that underlie it, are associated with a substantial increase in neuronal responsiveness (Fig. 3). Indeed, this effect was so powerful that current pulses that were completely subthreshold for the generation of action potentials in the DOWN state could generate a strong train of action potentials during the UP state (Fig. 3). The facilitation of responsiveness depended in part on the amplitude of the current pulse. Larger amplitude pulses were less facilitated than smaller amplitude pulses. This amplitude-dependent facilitation resulted in a decrease in slope of the input–output relationship during the UP state (Fig. 3C; see also Chance et al., 2002).

The intracellular injection of constant current pulses in an otherwise quiescent neuron results in stereotyped trains of action potentials, the interspike intervals of which are relatively repeatable from pulse to pulse (Fig. 3B; Nowak et al., 1997). Injection of the same pulses during the UP state, however, resulted in a considerable increase in variation in the interspike interval (Fig. 3B, UP state) such that the cumulative peristimulus histogram gave a much smoother appearance than that obtained in the DOWN state. In this way, the UP state was associated with an increase in variation in spike timing during constant current pulses.

Through what mechanism(s) is the barrage of synaptic activity during the UP state causing an increase in neuronal responsiveness to the current pulses along with an increase of spike timing variability? The barrages of synaptic potentials that underlie the UP state are associated with three general changes in the postsynaptic cell: depolarization of the membrane potential by an average of 7.6 ± 2.1 mV; increase in membrane conductance, ranging from 4–32 nS (n = 15 cells); and an increase in membrane potential variance of 2.6 ± 0.4 mV (n = 9 cells; Table 1). These values are similar or a little smaller than those reported during the slow oscillation in vivo, where the UP state is on average between 9.6 (Steriade et al., 2001) and 14.5 mV (Stern et al., 1997) depolarized to the DOWN state, the SD of the membrane potential is increased by 2–3 mV (Stern et al., 1997; Steriade et al., 2001) and the membrane conductance is increased by an average of 27 nS (Paré et al., 1998; Steriade et al., 2001). To examine which of these three general changes (depolarization, increase conductance, increase membrane variance) were responsible for the increases in neuronal response during the UP state in our preparation, we manipulated each independently and examined the effect on the response to the intracellular injection of constant current pulses (Figs 47).

Changing the membrane potential of layer 5 pyramidal cells through the intracellular injection of current resulted in marked changes in neuronal responsiveness to the current pulses (Fig. 4). Depolarization resulted in a leftward shift in the input–output curve, without a change in slope, such that a smaller current pulse was needed in order to activate the same number of action potentials (n = 5 cells). This is as expected, since depolarization results in a decrease in the distance between the membrane potential and spike threshold. On average, we found a relationship of 0.062 ± 0.049 nA/mV change in the input–output relation, meaning that an 8 mV depolarization would result in a decrease of 0.5 nA in the current required to inject to yield the same spike response.

To examine the effect of increases in membrane conductance on neuronal responses to current pulses, we used a dynamic clamp system (Dorval et al., 2001). Increases in the apparent ‘leak’ conductance of the membrane (with the reversal potential of the leak set at the resting membrane potential of the cell) resulted in a rightward shift in the input–output relationship (Fig. 5B; n = 5 cells). The average shift (n = 5 cells) was 0.016 ± 0.0033 nA per nS increase in baseline membrane conductance, meaning that a 10 nS increase in membrane conductance will result in a requirement of 0.16 nA of additional current to achieve the same spike response. Again, these shifts in the input–output relation occurred without marked changes in slope (Fig. 5B).

Finally, to examine the effects of increases in membrane variance on the input–output relationship, we used the dynamic clamp system to inject noise consisting of both excitatory and inhibitory conductances adjusted to give statistical features similar to that of the synaptic activity associated with natural UP states (n = 6 cells; see Methods). Three different levels of injected noise (with SDs of 0, 2 and 4 nS; for both the excitatory and inhibitory components; see Methods) were investigated. For all cases, the average conductance was set to 2–5 nS for the excitatory (reversal potential of 0 mV) and 2–5 nS for the inhibitory (reversal potential of –75 mV) components of the injected background activity. Increasing the average excitatory and inhibitory noise to 2 nS resulted in an increase in the SD of the membrane potential to 2.87 ± 0.34 mV (n = 7 cells), which is similar to that occurring during the natural UP state.

Increases in membrane potential ‘noise’ through this method resulted in an increase in neuronal response to depolarizing current pulses, especially to smaller inputs (Figs 6 and 7). (Note that the average membrane potential does not change in response to the addition of noise since the injected conductance does not change.) The enhancement by noise was dependent upon the amplitude of the current pulse injected, resulting in a change in slope of the input–output relationship (Fig. 6D). Measurements of the action potential response to the smallest current pulse that generated action potentials in the presence of noise revealed an increase in effectiveness equivalent to an increase in the amplitude of the noise-free current pulse by 48 ± 21 pA per nS of increased SD of both Ge and Gi (n = 4 cells). The addition of noise-2, which results in a similar increase in SD of the membrane as the natural UP state, should then result in an increase in the number of action potentials generated by the smallest current pulse equivalent to increasing the constant current pulse by ∼0.1 nA (see also Chance et al., 2002).

Expansion of the responses to current pulses with and without the addition of membrane potential variance revealed the mechanism of enhancement. The addition of depolarizing current transients on top of the relatively steady depolarization of the constant current pulse resulted in occasional crossings of spike threshold (Fig. 7). Thus, depolarizing current pulses that were normally subthreshold became suprathreshold and generated action potentials. Since the action potentials during the current pulse were generated by the combination of the pulse and the randomly varying noise, the timing of these spikes varied widely from trial to trial, but were confined to the timing of the current pulse. The facilitatory effect on larger current pulses was less marked, since the injected noise adds both depolarizing and hyperpolarizing components. When neurons are spiking rapidly, the addition of noise both increases and decreases spike probability, through these depolarizing and hyperpolarizing components.

Synaptic Activity of the UP State Facilitates the Response to Sine Wave Currents

Prior in vivo investigations have suggested that the spontaneous synaptic activity associated with the UP state may contribute to contrast-invariant responses in visual cortical neurons by removing the so-called ‘iceberg’ effect (Anderson et al., 2000b). Here we investigated this phenomenon by examining the effects of the UP state on the response to the intracellular injection of sine wave currents of varying frequencies and amplitudes (n = 7 cells). The UP state increased the response to 4–5 Hz sine waves considerably (Fig. 8). During the DOWN state, the intracellular injection of the sine wave typically yielded action potentials only near the peak of the depolarizing phase of the current injection (Fig. 8B,C). However, during the UP state, action potentials were generated throughout a greater extent of the depolarizing phase, resulting in a better representation by the average action potential discharge of the amplitude-time course of the sine wave input (Fig. 8B,D). To quantify this effect, we examined the peak cross correlation between the injected sinewave current and the spike rate histogram. The UP state resulted in an increase in the peak cross correlation between injected current and the spike rate of the recorded cell (28% increase from r = 0.44 to 0.57; n = 6; P < 0.01). This increase in correlation by the UP state, however, was frequency dependent (see below).

By examining the response of cortical cells to sine waves of different amplitudes, it became evident that the UP state was associated with a decrease in the iceberg effect (Fig. 8E), owing to a decrease in the amount of current needed to elicit an action potential, an increase in the number of action potentials generated with each current level above firing threshold, and an increase in their temporal representation of the sine wave current (e.g. see Fig. 8B). An additional effect of the UP state on the spike response to sinewave currents was a shift in the peak spike rate to earlier times: a phase advance (Fig. 8C,D,G).

Increasing the frequency of the sine wave current from 10 to 100 Hz resulted in variations of this general finding. Increasing the frequency of current injection into neurons is well known to decrease the amplitude of the membrane deviation, owing to the low-pass filtering characteristics of neuronal membranes (e.g. Nowak et al., 1997). Owing to this decrease in response, the cell was depolarized with the intracellular injection of current at higher frequencies (40–100 Hz) so that sinewave input generated action potentials, typically in both the DOWN and UP state. The UP state facilitated spike responses to all inputs up to the higher frequencies tested (50, 100 Hz; Fig. 9). The increase in peak cross correlation caused by the UP state, as well as the phase advance, both decreased with increasing frequencies of the sinewave current (Fig. 8F,G). These results indicate that although the UP state results in a general increase in responsiveness to a broad range of frequencies, the amplitude of this effect is frequency dependent.

Discussion

Our results demonstrate that slices of the prefrontal cortex maintained in vitro can generate prolonged periods of sustained activity when maintained in a slice solution that more closely resembles the ionic composition of cerebrospinal fluid than is typically used. We utilized this spontaneous network activity to address the mechanisms by which sustained activity may be generated within local circuits within the prefrontal cortex and how it may influence the responsiveness of single cortical cells to other inputs.

Mechanisms of Generation of UP and DOWN states

Our results demonstrate that sustained activity in vitro is generated by barrages of EPSPs and IPSPs, owing to activity in pyramidal and non-pyramidal neurons (Sanchez-Vives and McCormick, 2000; Shu et al., 2003; see also Steriade et al., 1993a). In both of these cell types the PSP barrages bring the cells close to firing threshold and in at least pyramidal cells the intermixed PSPs exhibit an average reversal potential of between –20 and –40 mV (Sanchez-Vives and McCormick, 2000; Shu et al., 2003). The influence of inhibitory interneurons is critical in the generation of relatively stable periods of activity, since blocking GABAA-receptor mediated inhibition results in the transformation of recurring persistent activity into paroxysmal discharges (Sanchez-Vives and McCormick, 2000).

The cyclical generation of UP and DOWN states, which occurs through all areas of the cerebral cortex so far recorded, appears to occur naturally as a result of the recurrent connectivity of the cerebral cortex and the membrane and synaptic properties of the constituent elements (Metherate and Ashe, 1993; Steriade et al., 1993a; Cowan and Wilson, 1994; Contreras et al., 1996; Stern et al., 1997; Sanchez-Vives and McCormick, 2000; Shu et al., 2003). A basic feature of the cerebral cortex is strong, local recurrent excitatory connections between pyramidal cells (see White, 1989; Braitenburg and Shuz, 1998). These recurrent excitatory connections also contact local inhibitory interneurons, which, in turn, inhibit neighboring pyramidal cells. Neuronal models indicate that this type of neuronal network architecture (recurrent excitation tempered by inhibition) can readily generate at least two states of activity, in similarity to those shown here (Amit and Brunel, 1997; Wang, 1999; Compte et al., 2000; Durstewitz et al., 2000a; Timofeev et al., 2000; Brunel and Wang, 2001).

The ability of cortical neurons to maintain activity while discharging at a relatively low rate presumably results from the large degree of convergence and divergence of activity within the cerebral cortex. Each cortical pyramidal cell receives ∼10  000 synapses, of which ∼2/3 to 3/4 are excitatory (see White, 1989; Abeles, 1991; Braitenburg and Shuz, 1998). The fact that connections between any two pyramidal cells involve only a small number of synapses (<5) indicates that each neuron receives input from over one thousand other excitatory cells. Even a low firing rate among this multitude of neuronal inputs may yield, depending upon the strength of inhibitory mechanisms, sufficient depolarization of the postsynaptic neuron to result in maintained action potential discharge. The strong regulation of this activity through the activation of local inhibitory interneurons, as well as intrinsic hyperpolarizing currents (e.g. K+ currents) in pyramidal cells, keeps the firing rate low (see Contreras et al., 1997b).

In our cortical slices, and during slow wave sleep or anesthesia in vivo, this neuronal activity can be expressed as a synchronized depolarization and hyperpolarization of nearly all cortical neurons, leading to so-called ‘UP’ and ‘DOWN’ states (Metherate and Ashe, 1993; Steriade et al., 1993a; Cowan and Wilson, 1994; Contreras et al., 1996; Stern et al., 1997; Sanchez-Vives and McCormick, 2000). We propose that the UP state is initiated by the generation of a burst of activity in a critical number of pyramidal cells (either through intrinsic ionic mechanisms or through recurrent excitation in a subgroup of cells) at a time in which the network has recovered from the refractory period that followed the previous UP state (Sanchez-Vives and McCormick, 2000; Compte et al., 2003a). The spontaneous transition to the DOWN state then may occur owing to the build up of intrinsic K+ currents in pyramidal neurons owing to the activity generated by the UP state (Sanchez-Vives and McCormick, 2000; Compte et al., 2003a). Otherwise the transition to the DOWN state may result in response to the strong activation of inhibitory neurons (Brunel and Wang, 2001; Shu et al., 2003) or perhaps in response to synchronized activation of a large fraction of the neurons in the network, thereby leading to a synchronized refractory period (Gutkin et al., 2001; Shu et al., 2003). In slices, the UP state of the slow oscillation propagates across the slice as it sequentially recruits neighboring regions of tissue into the activity (Sanchez-Vives and McCormick, 2000). The propagation of UP states in vivo should occur in considerably more complicated patterns, given the three dimensional structure of the brain and cortical connectivity.

Is the UP state of the slow oscillation equivalent to the maintained membrane potential and low firing rate of the cortex in the waking, but resting state? The answer to this question is not yet known. On the positive side, stimulation of the brainstem that results in activation of the EEG or the natural transition to waking both result in a loss of the DOWN state, with a maintained membrane potential of cortical cells that is similar to that of the UP state (Steriade et al., 1993b, 2001), suggesting that the UP state and the ‘waking’ membrane potential have similar mechanisms of generation. However, the apparent input resistance of cortical pyramidal cells is higher during the maintained depolarized membrane potential of waking in comparison to the UP state (Steriade et al., 2001), perhaps owing to the increased release of neuromodulatory agents (e.g. acetylcholine). Thus, the membrane potential of cortical neurons in the waking brain may be maintained in a depolarized state through both recurrent excitation as well as the actions of neuromodulators, while in the sleeping brain, the persistent activity of the UP state may be largely due to the recurrent excitatory activities alone. Either way, it is clear that nearly all or all regions of the cortex possess the intrinsic ability to generate persistent depolarized states with a relatively low firing rate through local recurrent connections — a mechanism that is often used to model the generation of persistent activity in cortical networks.

Implications for the Cellular Mechanisms of Persistent Activity during Working Memory

There are two general hypotheses concerning the mechanisms of persistent activity. The first is that the activity is generated through self-sustained excitation within a neuronal network, mediated by recurrent activity. The second is that the persistent activity is generated through intrinsic membrane mechanisms, such as afterdepolarizations. These two hypotheses are not mutually exclusive, and both mechanisms may, at least theoretically, contribute.

Our results lend strong support to the hypothesis that cortical networks can generate local sustained activity through recurrent excitatory interactions. Indeed, the firing rate of our pyramidal cells during the UP state is within a similar range to that occurring during the delay period of working response memory tasks in vivo, namely from a few up to ∼40 Hz (Funahashi et al., 1989; Miller et al., 1996). Action potentials during both the UP state and during the delay period in vivo often occur in a highly variable manner, consistent with a Poisson-like process (Compte et al., 2003b; A. Hasenstaub, C. Ghandi and D.A. McCormick, unpublished observations). Prefrontal cortical persistent activity in vivo can last for periods of seconds in delay response tasks (Fuster, 1995). In normal slice solution, our sustained activity had a maximal duration of ∼4 s, although we have recently found that bath application of the h-channel blocker ZD-7288 results in a marked prolongation of UP states so that they may last for up to 10–12 s (C. Ghandi and D.A. McCormick, unpublished observations). Finally computational approaches successfully model the persistent activity during both the UP state and during the delay period in working memory tasks by using recurrent excitation that is controlled with local inhibition (Wang, 1999, 2001; Compte et al., 2000, 2003a).

However, there are also a number of significant differences between the UP state activity studied here and the properties of persistent activity generated in vivo. First, the UP state in vitro (and during anesthesia and sleep in vivo) is characterized by barrages of synaptic activity in nearly all intracellularly recorded neurons (Metherate and Ashe, 1993; Steriade et al., 1993a; Contreras et al., 1996; Sanchez-Vives and McCormick, 2000), leading to strong increases in extracellularly recorded multiple unit activity. Although intracellular recordings in vivo have not yet been performed to examine the synaptic activity underlying delay period activity, extracellular single unit data has demonstrated that only a minority of cells discharge at an increased rate during the delay period, with many neurons not changing their firing rate, or even decreasing their discharge, during this period (Funahashi et al., 1989). Also, in behaving monkeys, although some neighboring cortical neurons may exhibit correlated activities, others may exhibit very different patterns during the behavioral task (Constantinidis et al., 2001). In contrast, during the UP state in vitro and in vivo, neighboring cells are often correlated in their activities. Finally, in some (but not all) recordings in vivo, the discharge rate of the cortical neuron may vary as a function of some feature of the stimulus (e.g. location, frequency) (Funahashi et al., 1989; Romo et al., 1999). Although action potential activity in vitro varies in intensity from UP state to UP state, this variation occurs spontaneously. One possible mechanism by which neuronal discharge rate could be continually varied in vivo is through changes in network architecture, either through the addition or subtraction of neuronal participation in the network (see Koulakov et al., 2002), or simply through a non-uniform change in the level of activity in participating neurons.

Although there are similarities between persistent activity during working memory tasks in vivo and sustained activity generated spontaneously in vitro, there are also considerable differences. These similarities and differences indicate that while the in vitro activity may yield important information concerning the generation of persistent activity through recurrent excitation, there are important limitations to the in vitro system as a model of delay period discharge in vivo.

Intrinsic Afterdischarges do not Contribute to Persistent Activity

Another leading hypothesis as to the origin of cortical mnemonic activity is that it is generated through intrinsic membrane mechanisms. For example, cortical pyramidal cells can generate prolonged afterdischarges in the presence of high doses of neuromodulatory agents (such as non-hydrolyzable muscarinic receptor agonists) through a Ca2+-activated cation current (Andrade, 1991; Haj-Dahmane and Andrade, 1996, 1998; Egorov et al., 2002; Fransen et al., 2002) or through persistent Na+ currents (Timofeev et al., 2000).

We did not find any evidence that our persistent activity in prefrontal cortex is generated through intrinsic cellular mechanisms. Hyperpolarization or depolarization of recorded neurons did not affect the duration of the UP states nor did it affect the periodicity of the rhythm, which is matched by the duration and periodicity of activity in the extracellularly recorded network (see also Sanchez-Vives and McCormick, 2000). In addition, the UP state is associated with a substantial increase in membrane noise and conductance, consistent with barrages of synaptic activity. Extracellular multiple unit recordings reveal strong increases in action potential discharge in neighboring neurons, which is consistent with the observed PSP barrages in the intracellularly recorded cells (see Fig. 1). Finally, block of excitatory synaptic transmission completely blocks the generation of persistent activity of the UP state (Sanchez-Vives and McCormick, 2000), and the intracellular injection of depolarizing current pulses did not result in persistent activity in any of our cells.

The generation of afterdischarges through a Ca2+-activated cation current in single neurons in vitro is a questionable model of persistent activity generated in vivo. First, the afterdischarges generated in vitro only occur during the strong and continual stimulation of muscarinic, or other metabotropic, receptors. It is not clear that such receptor stimulation occurs in vivo. The intracellular injection of depolarizing currents in awake, sleeping, or anesthetized animals in vivo has not yet been found to result in afterdischarges in cortical neurons (e.g. Sanchez-Vives et al., 2000; Steriade, 2001; Steriade et al., 2001), unless K+ currents are artificially blocked with the intracellular injection of Cs+ (Contreras et al., 1997a). Secondly, persistent activity in vivo during the performance of working memory tasks is associated with a highly variable interspike interval, which is consistent with the generation of this activity through synaptic mechanisms (Brunel and Wang, 2001; Compte et al., 2003b). In contrast, the generation of persistent activity through intrinsic membrane mechanisms typically results in a highly regular discharge rate (Egorov et al., 2002). Finally, the generation of persistent activity through network mechanisms has the distinct advantage that it can be turned on and off quickly (Fig. 1; see also Brunel and Wang, 2001), which does not appear to be the case with afterdischarges generated during constant muscarinic receptor stimulation (see Egorov et al., 2002). At present, the most parsimonious explanation of the available data is that cortical neurons generate persistent activity in vivo largely or exclusively through recurrent excitation that is controlled by inhibition — perhaps in a manner similar to that detailed here for UP state activity. We do not rule out, however, the possibility that intrinsic membrane properties also contribute. It is probable that intrinsic ionic currents, such as persistent Na+ currents (Timofeev et al., 2000) or Ca2+-activated cation currents (Egorov et al., 2002), that facilitate the response to excitatory inputs will also facilitate the generation of persistent recurrent network activity. In this way, we envision intrinsic ionic mechanisms to be facilitatory and regulatory, but not primary sources, of persistent activity in vivo.

Implications of Background Synaptic Activity on Gain Modulation and Attention

Cortical neurons are imbedded in a richly interconnected network and receive tens of thousands of synaptic inputs (reviewed in White, 1989; Braitenburg and Shuz, 1998). Although many of these synaptic connections arise from local circuits, they also arise from distal cortical regions, including feedback pathways, and presumably provide a mechanism by which the responsiveness of single cortical cells can be modulated by behavioral and cortical ‘context’. Here we have demonstrated that barrages of excitatory and inhibitory synapses can result in significant changes in neuronal excitability and responsiveness. The barrages of synaptic activity associated with the UP state result in a significant increase in the responsiveness of cortical neurons to the intracellular injection of depolarizing current pulses, particularly those that were subthreshold (Fig. 3), resulting in a decrease in the slope of the input–output relation (Fig. 3; see also Chance et al., 2002). This effect is mediated largely through depolarization of the membrane potential, so that inputs are more effective at driving action potentials, and through the increase in membrane noise, which changes the slope of the input–output relation and facilitates responses to small inputs (Chance et al., 2002).

Changes in membrane variance (noise) will not only affect the number of spikes generated, but also spike timing. In response to constant current pulses, cortical neurons generate a relatively stereotyped pattern of action potentials that is repeatable from trial to trial (see Figs 3B and 6C). The addition of noise in this situation can have strong effects on spike timing, since the cell will spike at the peaks of the depolarizing components of the noisy current (Mainen and Sejnowski, 1995; Nowak et al., 1997). If the noise varies from trial to trial in a random manner, then the action potentials generated during the current pulse will be smoothed, owing to a decrease in the repeatability of spike times. An additional effect of the added noise, however, is that the average histogram of the spike response is more representative of the amplitude–time course of the current pulse or low frequency sine wave input (see Figs 6C and 8). For low frequency inputs, therefore, increases in membrane variance in a population of neurons may not only increase their responses to small inputs, but also may yield a more accurate representation of the amplitude-time course of the signal.

How does background activity affect the responses to synaptic inputs? This question has been addressed with computational models (Hô and Destexhe, 2000; Destexhe et al., 2001), through the injection of artificial conductances in cortical neurons in vitro (Hasenstaub et al., 2002; Mitchell and Silver, 2003), and through the activation of synaptic potentials in vivo (Timofeev et al., 1996). These studies reveal that membrane depolarization and small to moderate increases in membrane variance may facilitate the responsiveness to synaptic inputs, especially those of lower amplitude. Increases in membrane conductance decrease neuronal responsiveness, causing an increase in threshold. For long duration inputs, increases in membrane conductance affect all inputs about equally, leading to a rightward shift in the input–output curve. However, if the inputs are random trains of EPSPs from a small number of cells, then increases in membrane conductance can cause a decrease in slope, or gain, of the frequency of output spike train versus the mean frequency of input EPSP rate (Mitchell and Silver, 2003).

Functional Implications of Background Synaptic Activity

The convergence of a large number of inputs onto each cortical neuron has the functional implication that the cells activity may be strongly influenced by a wide variety of different neural circuits. This broad influence may provide a mechanism to explain diverse effects such as: (i) the ability of sensory context (i.e. far field stimuli) to modulate the response to discrete sensory stimuli (Carandini and Heeger, 1994; Angelucci et al., 2002; Stettler et al., 2002); (ii) the transformation of sensory to motor maps of external space (Salinas and Abbott, 1996); (iii) increases in neuronal responsiveness with attention (McAdams and Maunsell, 1999a,b; Hahnloser et al., 2002); and (iv) the functional operation of feedback pathways, a common and little understood feature of cortical architecture (Hupé et al., 1998; Hahnloser et al., 2002). For example, the spontaneous barrages of synaptic activity associated with the UP state result in an increase in response probability to small inputs, with relatively little change in the response to larger inputs (Hasenstaub et al., 2002). This is similar to the effects of attention on the contrast-response function curves of neurons in V4 (Reynolds et al., 2000). Here attention results in an increase in neuronal response to low contrast (i.e. weak) stimuli, while responses to high contrast (i.e. strong) stimuli are relatively unaffected. This amplitude-dependent enhancement is similar to our findings that synaptic barrages enhance the responsiveness of cortical neurons to current pulses (Fig. 3). This suggests that barrages of synaptic activity may underlie the attentional affects observed by Reynolds et al. (2000).

However, in other conditions, attention can result in an enhancement of the response to all stimuli in a multiplicative manner (McAdams and Maunsell, 1999a,b; Treue and Martínez-Trujillo, 1999). This effect of attention may also result from barrages of synaptic activity, since it has been shown that decreases in membrane variance that do not change the average membrane potential of the cell and which are balanced with decreases in membrane conductance can change neuronal ‘gain’ (Hô and Destexhe, 2000; Chance et al., 2002). Perhaps attentional mechanisms increase neuronal responsiveness both by depolarizing some cells with barrages of synaptic activity, while increasing the ‘gain’ (slope of the input–output relation) of other cells by decreasing membrane conductance and noise that reverses near resting membrane potential.

Conclusions

A common and pervasive theme of cortical networks is local feedback excitation between pyramidal cells that is controlled and tuned by disynaptic feedback inhibition. This basic architecture, left to its own devices, spontaneously generates two patterns of activity: relative quiescence and persistent activity. The generation of persistent activity in the prefrontal cortex (and other cortical areas) may take advantage of this basic architecture to serve as a mnemonic device, much in the manner of a neuronal ‘flip-flop’. These flips in neuronal network activity may not only serve to temporarily store information, but will also have strong influences on neuronal responsiveness and may provide a neuronal mechanism underlying diverse behavioral effects. In all of these behavioral conditions, however, neuronal activities display important features that are yet to be replicated in vitro and remain to be investigated in full.

We thank Michael Rudolph and Alain Destexhe for helpful suggestions. Supported by the NIH (D.A.M.), Howard Hughes Foundation (A.H.), Institute Henri Poincaré, Centre National de la Recherche Scientifique and Bioinformatique (00N25/0003), and the European Commission (IST-2001–34712) (T.B. and M.B.). Movies and more information concerning the UP and DOWN activity can be obtained at www.mccormicklab.org.

Figure 1. The prefrontal cortical slice in vitro spontaneously generates prolonged periods of activity through synaptic bombardments. (A) Simultaneous intracellular and extracellular recordings in layer 5 of the ferret prefrontal cortex. The network enters into the ‘UP’ state for ∼3 s prior to a rapid transition to the DOWN state. (B) Hyperpolarization of the intracellularly recorded pyramidal cell reveals the barrage of PSPs arriving during the UP state and the reduction of PSPs during the DOWN state. Note that hyperpolarization or depolarization does not affect the duration of the UP state, which is the same as the average duration of action potential discharge in the local network.

Figure 1. The prefrontal cortical slice in vitro spontaneously generates prolonged periods of activity through synaptic bombardments. (A) Simultaneous intracellular and extracellular recordings in layer 5 of the ferret prefrontal cortex. The network enters into the ‘UP’ state for ∼3 s prior to a rapid transition to the DOWN state. (B) Hyperpolarization of the intracellularly recorded pyramidal cell reveals the barrage of PSPs arriving during the UP state and the reduction of PSPs during the DOWN state. Note that hyperpolarization or depolarization does not affect the duration of the UP state, which is the same as the average duration of action potential discharge in the local network.

Figure 2. Summary diagram of the proposed mechanisms for the spontaneous generation of the UP and DOWN states in cortical networks. Cortical pyramidal and local GABAergic interneurons are highly interconnected through local axonal connections such that activity in pyramidal cells excites both other pyramidal cells as well as local interneurons. The inhibitory feedback from the local interneurons controls the level of the UP state. Activation of an afferent input can trigger the transition from the DOWN to the UP state. Build up of intracellular levels of Ca2+ and Na+ can activate K+ currents, which eventually tip the balance back to the DOWN state, during which time the levels of Ca2+ and Na+ inside the cell decrease, thus allowing the network to spontaneously generate another UP state. Another mechanism for making the transition from the UP to the DOWN state is the activation of afferent inputs (Shu et al., 2003).

Figure 2. Summary diagram of the proposed mechanisms for the spontaneous generation of the UP and DOWN states in cortical networks. Cortical pyramidal and local GABAergic interneurons are highly interconnected through local axonal connections such that activity in pyramidal cells excites both other pyramidal cells as well as local interneurons. The inhibitory feedback from the local interneurons controls the level of the UP state. Activation of an afferent input can trigger the transition from the DOWN to the UP state. Build up of intracellular levels of Ca2+ and Na+ can activate K+ currents, which eventually tip the balance back to the DOWN state, during which time the levels of Ca2+ and Na+ inside the cell decrease, thus allowing the network to spontaneously generate another UP state. Another mechanism for making the transition from the UP to the DOWN state is the activation of afferent inputs (Shu et al., 2003).

Figure 3. The UP state is associated with a marked increase in neuronal excitability. (A) Intracellular injection of depolarizing current pulses during the UP and DOWN states reveal a marked increase in neuronal responsiveness during the UP in comparison to the DOWN state. Two different amplitudes of current pulses are illustrated. (B) Stimulus histograms of the action potential response of the cell to repeated injections of the same current pulse (0.7 nA) in the DOWN and UP states. The cell responds with more action potentials in the UP state, and the timing of these action potentials varies from pulse to pulse. (C) Response of the neuron to different amplitude current pulses in the UP and DOWN states. Note the marked increase in neuronal responses, especially to smaller amplitude inputs.

Figure 3. The UP state is associated with a marked increase in neuronal excitability. (A) Intracellular injection of depolarizing current pulses during the UP and DOWN states reveal a marked increase in neuronal responsiveness during the UP in comparison to the DOWN state. Two different amplitudes of current pulses are illustrated. (B) Stimulus histograms of the action potential response of the cell to repeated injections of the same current pulse (0.7 nA) in the DOWN and UP states. The cell responds with more action potentials in the UP state, and the timing of these action potentials varies from pulse to pulse. (C) Response of the neuron to different amplitude current pulses in the UP and DOWN states. Note the marked increase in neuronal responses, especially to smaller amplitude inputs.

Figure 4. Depolarization increases neuronal responsiveness. (A) Examples of a pyramidal cell’s response to the intracellular injection of three different amplitude current pulses at –78 and –90 mV. (B) Graphical illustration of results revealing that hyperpolarization of the cortical pyramidal cell with the intracellular injection of current shifts the input–output curve to the right, without marked effects on slope.

Figure 4. Depolarization increases neuronal responsiveness. (A) Examples of a pyramidal cell’s response to the intracellular injection of three different amplitude current pulses at –78 and –90 mV. (B) Graphical illustration of results revealing that hyperpolarization of the cortical pyramidal cell with the intracellular injection of current shifts the input–output curve to the right, without marked effects on slope.

Figure 5. Increases in background membrane conductance with the dynamic clamp system results in a decrease in neuronal responsiveness. (A) Examples of neuronal response to three different current pulse amplitudes at three different levels of added ‘leak conductance’ that reversed at the resting membrane potential. (B) Graphical plot of the data demonstrating that the input–output curve shifts to the right with increases in artificial membrane conductance. The reversal potential of the background leak conductance was set to the resting membrane potential in order to avoid changes in membrane potential with changes in conductance. Points at 1.1 and 1.2 nA for 0 and 20 nS were discarded because the neuron was not in the same adaptive state as for the rest of the data.

Figure 5. Increases in background membrane conductance with the dynamic clamp system results in a decrease in neuronal responsiveness. (A) Examples of neuronal response to three different current pulse amplitudes at three different levels of added ‘leak conductance’ that reversed at the resting membrane potential. (B) Graphical plot of the data demonstrating that the input–output curve shifts to the right with increases in artificial membrane conductance. The reversal potential of the background leak conductance was set to the resting membrane potential in order to avoid changes in membrane potential with changes in conductance. Points at 1.1 and 1.2 nA for 0 and 20 nS were discarded because the neuron was not in the same adaptive state as for the rest of the data.

Figure 6. Increases in membrane potential variance result in increases in neuronal responses, especially to smaller inputs. (A) Examples of the response to the intracellular injection of three different amplitude current pulses with no added ‘noise’ (membrane variance). The membrane conductance was the same with and without noise, so the addition of noise did not result in a change in membrane potential. The difference between (A) and (B) is that the SD of Ge and Gi in (A) was 0 nS, while in (B) it were 4 nS. (B) Examples of the response of the neuron to the same three pulses when the SD of both Ge and Gi was set to 4 nS. (C) Stimulus histograms of the response of the neuron to repeated injections of the same current pulse with and without additional noise injection. There is an increase in spike timing variability to constant current pulses with the addition of noise. (D) Input–output plot of the response of the cell with and without noise. Note that the addition of a noisy conductance results in an increase in action potential response to small inputs.

Figure 6. Increases in membrane potential variance result in increases in neuronal responses, especially to smaller inputs. (A) Examples of the response to the intracellular injection of three different amplitude current pulses with no added ‘noise’ (membrane variance). The membrane conductance was the same with and without noise, so the addition of noise did not result in a change in membrane potential. The difference between (A) and (B) is that the SD of Ge and Gi in (A) was 0 nS, while in (B) it were 4 nS. (B) Examples of the response of the neuron to the same three pulses when the SD of both Ge and Gi was set to 4 nS. (C) Stimulus histograms of the response of the neuron to repeated injections of the same current pulse with and without additional noise injection. There is an increase in spike timing variability to constant current pulses with the addition of noise. (D) Input–output plot of the response of the cell with and without noise. Note that the addition of a noisy conductance results in an increase in action potential response to small inputs.

Figure 7. Expansion of the response of the neuron to the same current pulse before and after the addition of noise. (A) Injected current with and without added noise. (B) Response of the neuron to these two current injections. Note that the addition of noise occasionally brings the membrane potential above threshold for action potential generation. Same cell as in Figure 6.

Figure 7. Expansion of the response of the neuron to the same current pulse before and after the addition of noise. (A) Injected current with and without added noise. (B) Response of the neuron to these two current injections. Note that the addition of noise occasionally brings the membrane potential above threshold for action potential generation. Same cell as in Figure 6.

Figure 8. The UP state enhances the neuronal response to the injection of a low frequency (5 Hz) sine wave. (A) Simultaneous intracellular and extracellular recordings of the UP state in this layer 5 pyramidal cell. (B) Example of the response of the neuron to the continuous injection of a 5 Hz sine wave before, during, and after the UP state. (C, D) Peristimulus histograms and raster plots of the action potential response to the sine wave injection during the DOWN (C) and UP (D) states for the same number of trials. (E) The UP state facilitates the responsiveness to various amplitude sine wave currents (4 Hz), thereby reducing the ‘iceberg’ effect. (F) Absolute increase (from the DOWN to UP state) in peak cross correlation between the injected sinewave current and the spike rate histogram for several cells (n = 6) versus the frequency of the injected sinewave. Note that the increase in collelation by the UP state diminishes as the frequency of the sinewave increases. (G) Advance in the time of peak firing rate by the UP state versus frequency of the injected current. With low frequency inputs, there is a strong advance in neuronal discharge by the UP state and this effect diminishes with increases in frequency. Cells in (A–D) and (E) are different. Data in (F) are from six cells, while in (G) they are from one cell.

Figure 8. The UP state enhances the neuronal response to the injection of a low frequency (5 Hz) sine wave. (A) Simultaneous intracellular and extracellular recordings of the UP state in this layer 5 pyramidal cell. (B) Example of the response of the neuron to the continuous injection of a 5 Hz sine wave before, during, and after the UP state. (C, D) Peristimulus histograms and raster plots of the action potential response to the sine wave injection during the DOWN (C) and UP (D) states for the same number of trials. (E) The UP state facilitates the responsiveness to various amplitude sine wave currents (4 Hz), thereby reducing the ‘iceberg’ effect. (F) Absolute increase (from the DOWN to UP state) in peak cross correlation between the injected sinewave current and the spike rate histogram for several cells (n = 6) versus the frequency of the injected sinewave. Note that the increase in collelation by the UP state diminishes as the frequency of the sinewave increases. (G) Advance in the time of peak firing rate by the UP state versus frequency of the injected current. With low frequency inputs, there is a strong advance in neuronal discharge by the UP state and this effect diminishes with increases in frequency. Cells in (A–D) and (E) are different. Data in (F) are from six cells, while in (G) they are from one cell.

Figure 9. The UP state facilitates the response to the intracellularly injected 50 Hz sinewave current. (A) During the UP state, the cell generates single action potentials that are phase locked to near the peak of the 50 Hz sine wave. (B) Expansion of part of the response in (A) for illustration.

Figure 9. The UP state facilitates the response to the intracellularly injected 50 Hz sinewave current. (A) During the UP state, the cell generates single action potentials that are phase locked to near the peak of the 50 Hz sine wave. (B) Expansion of part of the response in (A) for illustration.

Table 1


 Intracellular measurements

Duration of UP state at rest Vm  1.7 ± 0.3 s (n = 16) 
Duration of UP state after hyperpolarization  1.7 ± 0.4 s (n = 16) 
Rate of recurrence of UP state at rest Vm  0.2 ± 0.1 Hz (n = 16) 
Rate of recurrence of UP state after hyperpolarization  0.2 ± 0.1 Hz (n = 16) 
Single unit firing rate of UP state 17.1 ± 11.1 Hz (n = 32) 
Single unit firing rate of DOWN state  3.6 ± 3.0 Hz (14/32)  0 Hz (silent, 18/32) 
UP–DOWN change in membrane potential  7.6 ± 2.1 mV (n = 8) 
Increase in membrane conductance during UP state  4–32 nS (n = 15) 
Increase in membrane variance during UP state  2.6 ± 0.4 mV; n = 9 
Depolarization induced shift in input–output curve  0.062 ± 0.049 nA/mV 
Conductance induced shift in input–output curve  0.016 ± 0.0033 nA/nS 
Duration of UP state at rest Vm  1.7 ± 0.3 s (n = 16) 
Duration of UP state after hyperpolarization  1.7 ± 0.4 s (n = 16) 
Rate of recurrence of UP state at rest Vm  0.2 ± 0.1 Hz (n = 16) 
Rate of recurrence of UP state after hyperpolarization  0.2 ± 0.1 Hz (n = 16) 
Single unit firing rate of UP state 17.1 ± 11.1 Hz (n = 32) 
Single unit firing rate of DOWN state  3.6 ± 3.0 Hz (14/32)  0 Hz (silent, 18/32) 
UP–DOWN change in membrane potential  7.6 ± 2.1 mV (n = 8) 
Increase in membrane conductance during UP state  4–32 nS (n = 15) 
Increase in membrane variance during UP state  2.6 ± 0.4 mV; n = 9 
Depolarization induced shift in input–output curve  0.062 ± 0.049 nA/mV 
Conductance induced shift in input–output curve  0.016 ± 0.0033 nA/nS 

References

Abeles M (
1991
) Corticonics. Neural circuits of the cerebral cortex. Cambridge: Cambridge University Press.
Aghajanian GK, Rasmussen K (
1989
) Intracellular studies in the facial nucleus illustrating a simple new method for obtaining viable motoneurons in adult rat brain slices.
Synapse
 
3
:
331–
338.
Amit DJ, Brunel N (
1997
) Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex.
Cereb Cortex
 
7
:
237–
252.
Andersen RA, Mountcastle VB (
1983
) The influence of the angle of gaze upon the excitability of light-sensitive neurons of the posterior parietal cortex.
J Neurosci
 
3
:
532–
548.
Andersen RA, Essick GK, Siegel RM (
1985
) Encoding of spatial location by posterior parietal neurons.
Science
 
230
:
450–
458.
Anderson JS, Carandini M, Ferster D (
2000
) Orientation tuning of input conductance, excitation, and inhibition in cat primary visual cortex.
J Neurophysiol
 
84
:
909–
926.
Anderson JS, Lampl I, Gillespie DC, Ferster D (
2000
) The contribution of noise to contrast invariance of orientation tuning in cat visual cortex.
Science
 
290
:
1968–
1972.
Andrade R (
1991
) Cell excitation enhances muscarinic cholinergic receptors in rat association cortex.
Brain Res
 
548
:
81–
93.
Angelucci A, Levitt JB, Walton EJS, Hupé J-M, Bullier J, Lund JS (
2002
) Circuits for local and global signal integration in primary visual cortex.
J Neurosci
 
22
:
8633–
8646.
Bernander Ö, Douglas RJ, Martin KAC, Koch C (
1991
) Synaptic background activity influences spatiotemporal integration in single pyramidal cells.
Proc Natl Acad Sci USA
 
88
:
11569–
11573.
Braitenburg V, Shuz A (
1998
) Cortex: statistics and geometry of neuronal connectivity. Berlin: Springer.
Brunel N, Wang XJ (
2001
) Effects of neuromodulation in a cortical network model of object working memory dominated by recurrent inhibition.
J Comp Neurosci
 
11
:
63–
85.
Camperi M, Wang X-J (
1998
) A model of visuospatial memory in prefrontal cortex: recurrent network and cellular bistability.
J Comp Neurosci
 
5
:
383–
405.
Carandini M, Heeger DJ (
1994
) Summation and division by neurons in primary visual cortex.
Science
 
264
:
1333–
1336.
Chance FS, Abbott LF, Reyes AD (
2002
) Gain modulation from background synaptic input.
Neuron
 
35
:
773–
782.
Compte A, Brunel N, Goldman-Rakic PS, Wang X-J (
2000
) Synaptic mechanisms and network dynamics underlying spatial working memory in a cortical network model.
Cereb Cortex
 
10
:
910–
923.
Compte A, Sanchez-Vives MV, McCormick DA, Wang XJ (
2003
) Cellular and network mechanisms of slow oscillatory activity (<1 Hz) in a cortical network model.
J Neurophysiol
  (in press).
Compte A, Constantinidis C, Tegnér J, Raghavachari S, Chafee M, Goldman-Rakic PS, Wang XJ (
2003
) Temporally irregular mnemonic persistent activity in prefrontal neurons of monkeys during a delayed response task.
J Neurophysiol
  (in press).
Constantinidis C, Franowicz MN, Goldman-Rakic PS (
2001
) Coding specificity in cortical microcircuits: a multiple-electrode analysis of primate prefrontal cortex.
J Neurosci
 
21
:
3646–
3655.
Contreras D, Steriade M (
1995
) Cellular basis of EEG slow rhythms, a study of dynamic corticothalamic relationships.
J Neurosci
 
15
:
604–
622.
Contreras D, Timofeev I, Steriade M (
1996
) Mechanisms of long lasting hyperpolarizations underlying slow sleep oscillations in cat corticothalamic networks.
J Physiol (Lond)
 
494
:
251–
264.
Contreras D, Durmuller N, Steriade M (
1997
) Plateau potentials in cat neocortical association cells in vivo: synaptic control of dendritic excitability.
Eur J Neurosci
 
9
:
2588–
2595.
Contreras D, Destexhe A, Steriade M (
1997
) Intracellular and computational characterization of the intracortical inhibitory control of synchronized thalamic inputs in vivo.
J Neurophysiol
 
78
:
335–
350.
Cowan RL, Wilson CJ (
1994
) Spontaneous firing patterns and axonal projections of single corticostriatal neurons in the rat medial agranular cortex.
J Neurophysiol
 
71
:
17–
32.
Destexhe A, Paré D (
1999
) Impact of network activity on the integrative properties of neocortical pyramidal neurons in vivo.
J Neurophysiol
 
81
:
1531–
1547.
Destexhe A, Rudolph M, Fellous J-M, Sejnowski TJ (
2001
) Fluctuating synaptic conductances recreate in vivo-like activity in neocortical neurons.
Neuroscience
 
107
:
13–
24.
Dorval AD, Christini DJ, White JA (
2001
) Real-time linux dynamic clamp: a fast and flexible way to construct virtual ion channels in living cells.
Ann Biomed Eng
 
29
:
897–
907.
Durstewitz D, Seamans JK, Sejnowski TJ (
2000
) Dopamine-mediated stabilization of delay-period activity in a network model of prefrontal cortex.
J Neurophysiol
 
83
:
1733–
1750.
Durstewitz D, Seamans JK, Sejnowski TJ (
2000
) Neurocomputational models of working memory.
Nat Neurosci
 
3
:
1184–
1191.
Egorov AV, Hamam BN, Fransen E, Hasselmo ME, Alonso AA (
2002
) Graded persistent activity in entorhinal cortex neurons.
Nature
 
420
:
173–
178.
Fellous JM, Rudolph M, Destexhe A, Sejnowski TJ (
2003
) Synaptic background noise controls the input/output characteristics of single cells in an in vitro model of in vivo activity. Neuroscience (in press).
Fransen E, Alonso AA, Hasselmo ME (
2002
) Stimulations of the role of the muscarinic-activated calcium-senstive nonspecific cation current INCM in entorhinal neuronal activity during delayed matching tasks.
J Neurosci
 
22
:
1081–
1097.
Funahashi S, Bruce CJ, Goldman-Rakic PS (
1989
) Mnemonic coding of visual space in the monkey’s dorsolateral prefrontal cortex.
J Neurophysiol
 
61
:
331–
349.
Fuster JM (
1995
) Memory in the cerebral cortex. Cambridge, MA: MIT Press.
Fuster JM, Alexander GE (
1973
) Firing changes in cells of the nucleus medialis dorsalis associated with delayed response behavior.
Brain Res
 
61
:
69–
91.
Fuster JM, Jervey JP (
1981
) Inferotemporal neurons distinguish and retain behaviorally relevant features of visual stimuli.
Science
 
212
:
952–
955.
Goldman-Rakic PS (
1995
) Cellular basis of working memory.
Neuron
 
14
:
477–
485.
Gutkin BS, Laing CR, Colby CL, Show CC, Ermentrout GB (
2001
) Turning on and off with excitation: the role of spike-timing asynchrony and synchrony in sustained neural activity.
J Comp Neurosci
 
11
:
121–
134.
Hahnloser RHR, Douglas RJ, Hepp K (
2002
) Attentional recruitment of inter-areal recurrent networks for selective gain control.
Neural Comput
 
14
:
1669–
1689.
Haj-Dahmane S, Andrade R (
1996
) Muscarinic activation of a voltage-dependent cation nonselective current in rat association cortex.
J Neurosci
 
16
:
3848–
3861.
Haj-Dahmane S, Andrade R (
1998
) Ionic mechanism of the slow afterdepolarization induced by muscarinic receptor activation in rat prefrontal cortex.
J Neurophysiol
 
80
:
1197–
1210.
Hasenstaub A, Shu YS, Badoual M, Bal T, McCormick DA (
2002
) The effects of recurrent activity on the responsiveness of cortical neurons to synaptic inputs.
Soc Neurosci Abstr
 
753A
.
Hikosaka O, Sakamoto M, Usui S (
1989
) Functional properties of monkey caudate neurons. III. Activities related to expectation of target and reward.
J Neurophysiol
 
61
:
814–
832.
Hô N, Destexhe A (
2000
) Synaptic background activity enhances the responsiveness of neocortical pyramidal neurons.
J Neurophysiol
 
84
:
1488–
1496.
Hupé JM, James AC, Payne BR, Lomber SG, Girard P, Bullier J (
1998
) Cortical feedback improves discrimination between figure and background by V1, V2 and V3 neurons.
Nature
 
394
:
784–
787.
Koulakov AA, Raghavachari S, Kepecs A, Lisman JE (
2002
) Model for a robust neural integrator.
Nat Neurosci
 
5
:
775–
782.
Lisman JE, Fellous J-M, Wang X-J (
1998
) A role for NMDA-receptor channels in working memory.
Nat Neurosci
 
4
:
273–
275.
Mainen ZF, Sejnowski TJ (
1995
) Reliability of spike timing in neocortical neurons.
Science
 
268
:
1503–
1506.
Marder E, Abbott LF, Turrigiano GG, Liu Z, Golowasch J (
1996
) Memory from dynamics of intrinsic membrane currents.
Proc Natl Acad Sci USA
 
93
:
13481–
13486.
McAdams CJ, Maunsell JHR (
1999
) Effects of attention on orientation-tuning functions of single neurons in macaque cortical area V4.
J Neurosci
 
19
:
431–
441.
McAdams CJ, Maunsell JHR (
1999
) Effects of attention on the reliability of individual neurons in monkey visual cortex.
Neuron
 
23
:
765–
773.
Metherate R, Ashe JH (
1993
) Ionic flux contributions to neocortical slow waves and nucleus basalis-mediated activation: whole-cell recordings in vivo.
J Neurosci
 
13
:
5312–
5323.
Miller EK, Erickson CA, Desimone R (
1996
) Neural mechanisms of visual working memory in prefrontal cortex of the macaque.
J Neurosci
 
16
:
5154–
5167.
Mitchell SJ, Silver RA (
2003
) Shunting inhibition modulates neuronal gain during synaptic excitation.
Neuron
 
38
:
433–
445.
Miyashita Y, Chang HS (
1988
)
Neuron
 al correlate of pictorial short-term memory in the primate temporal cortex.
Nature
 
331
:
68–
70.
Nowak LG, Sanchez-Vives MV, McCormick DA (
1997
) Influence of low and high frequency inputs on spike timing in visual cortical neurons.
Cereb Cortex
 
7
:
487–
501.
Paré D, Shink E, Gaudreau H, Destexhe A, Lang EJ (
1998
) Impact of spontaneous synaptic activity on the resting properties of cat neocortical neurons in vivo.
J Neurophysiol
 
79
:
1450–
1460.
Pesaran B, Pezaris JS, Sahani M, Mitra PP, Andersen RA (
2002
) Temporal structure in neuronal activity during working memory in macaque parietal cortex.
Nat Neurosci
 
5
:
805–
811.
Reynolds JH, Pasternak T, Desimone R (
2000
) Attention increases sensitivity of V4 neurons.
Neuron
 
26
:
703–
714.
Romo R, Brody CD, Hernandez A, Lemus L (
1999
)
Neuron
 al correlates of parametric working memory in prefrontal cortex.
Nature
 
399
:
470–
473.
Salinas E, Abbott LF (
1996
) A model of multiplicative neural responses in parietal cortex.
Proc Natl Acad Sci USA
 
93
:
11956–
11961.
Sanchez-Vives MV, McCormick DA (
2000
) Cellular and network mechanisms of rhythmic recurrent activity in neocortex.
Nat Neurosci
 
3
:
1027–
1034.
Sanchez-Vives MV, Nowak L, McCormick DA (
2000
) Membrane mechanisms underlying contrast adaptation in cat area 17 in vivo.
J Neurosci
 
20
:
4267–
4285.
Shu Y, Hasenstaub A, McCormick DA (
2003
) Turning on and off recurrent balanced cortical activity.
Nature
 
423
:
288–
293.
Steriade M (
2001
) Impact of network activities on neuronal properties in corticothalamic systems.
J Neurophysiol
 
86
:
1–
39.
Steriade M, Nunez A, Amzica F (
1993
) A novel slow (<1 Hz) oscillation of neocortical neurons in vivo: depolarizing and hyperpolarizing components.
J Neurosci
 
13
:
3252–
3265.
Steriade M, Amzica F, Nunez M (
1993
) Cholinergic and noradrenergic modulation of the slow (approximately 0.3 Hz) oscillation in neocortical cells.
J Neurophysiol
 
70
:
1385–
1400.
Steriade M, Timofeev I, Grenier F (
2001
) Natural waking and sleep states, a view from inside neocortical neurons.
J Neurophysiol
 
85
:
1969–
1985.
Stern EA, Kincaid AE, Wilson CJ (
1997
) Spontaneous subthreshold membrane potential fluctuations and action potential variability of rat corticostriatal and striatal neurons in vivo.
J Neurophysiol
 
77
:
1697–
1715.
Stettler DS, Das A, Bennett J, Gilbert CD (
2002
) Lateral connectivity and contextual interactions in Macaque primary visual cortex.
Neuron
 
36
:
739–
750.
Timofeev I, Contreras D, Steriade M (
1996
) Synaptic responsiveness of cortical and thalamic neurons during different phases of slow sleep oscillation in cat.
J Physiol (Lond)
 
494
:
265–
278.
Timofeev I, Grenier F, Bazhenov M, Sejnowski TJ, Steriade M (
2000
) Origin of slow cortical oscillations in deafferented cortical slabs.
Cereb Cortex
 
10
:
1185–
1199.
Treue S, Martínez-Trujillo JC (
1999
) Feature-based attention influences motion processing gain in macaque visual cortex.
Nature
 
399
:
575–
579.
Uhlenbeck GE, Ornstein LS (
1930
) On the theory of the Brownian motion.
Phys Rev
 
36
:
823–
841.
Wang X-J (
1999
) Synaptic basis of cortical persistent activity: the importance of NMDA receptors to working memory.
J Neurosci
 
19
:
9587–
9603.
Wang XJ (
2001
) Synaptic reverberation underlying mnemonic persistent activity
Trends Neurosci
 
24
:
455–
463.
White EL (
1989
) Cortical circuits. Boston, MA: Birkhauser.