INTRODUCTION

Dopamine innervation in the prefrontal cortex (PFC) originates from the ventral tegmental area (Lindvall et al, 1974; Sesack et al, 1998; Thierry et al, 1978), a pathway that is critical for the regulation of cognitive functions including working memory and decision-making (for review, see Floresco, 2013). The neurobiology underlying the modulation of these PFC functions is thought to be mediated by dopamine’s ability to enable context relevant inputs to enhance the activity of selective neuronal ensembles for hundreds of milliseconds to seconds (Funahashi et al, 1989; Goldman-Rakic, 1995). In addition to the role of D1 and NMDA receptors in facilitating prefrontal output responses (Baldwin et al, 2002; Flores-Barrera et al, 2013; Floresco and Phillips, 2001; Gurden et al, 1999; Jay, 2003; Tseng and O’Donnell, 2005), dopamine action in the PFC includes activation of fast-spiking interneurons (FSIs) (Gorelova et al, 2002; Tseng and O’Donnell, 2007b). They are a subset of GABAergic interneurons critical for determining the timing and spatial selectivity of pyramidal cell firing (Rao et al, 2000). Accordingly, it has been suggested that FSI could shape the response pattern of prefrontal pyramidal neurons to mesocortical dopamine drive (Tseng et al, 2006; Tseng and O’Donnell, 2004). For example, ventral tegmental area stimulation frequently results in suppression of pyramidal cell activity in the PFC, an inhibitory response that matches the temporal course of local prefrontal FSI excitation (Lewis and O’Donnell, 2000; Tseng et al, 2006). At the cellular level, there is ample evidence that part of the inhibitory action of dopamine in the PFC is due to an enhancement of local GABAergic tone (Gorelova et al, 2002; Gulledge and Jaffe, 1998; Pirot et al, 1992; Tseng et al, 2006; Tseng and O’Donnell, 2004). In fact, GABAergic interneurons in the PFC do express dopamine receptors (Le Moine and Gaspar, 1998; Mrzljak et al, 1996; Muly et al, 1998; Vincent et al, 1995) and FSI excitability becomes positively modulated by D1 and D2 receptors in the adult PFC, a functional maturation that occurs late in adolescence (Tseng and O’Donnell, 2007b). Thus, a fine tuning between local PFC GABAergic transmission and pyramidal cell firing by dopamine has been proposed to have a critical role in the regulation of working memory processes as disruptions of such interactions are implicated in the pathophysiology of cognitive deficits observed in schizophrenia and related psychiatric conditions (Lewis and Gonzalez-Burgos, 2006; O’Donnell, 2011; Tseng et al, 2009). Furthermore, deficits in PFC GABAergic function can result in reduced cognitive flexibility as shown in animal models (see review by Floresco, 2013). We therefore hypothesize that dopamine-dependent facilitation of FSI function in the PFC is needed to improve the signal detection ratio between task-dependent stimuli and distractors by enhancing clusters of neuronal activity that encode such stimuli. Here, we employed a modified version of the well-established computational model of working memory developed by Brunel and Wang (2001) to determine how dopamine modulation of FSI transmission in the PFC enables input selectivity in pyramidal cells to sustain working memory and its reset. The latter will be simulated by a transient phasic elevation of dopamine that is sufficient to cause rapid increases in FSI activity.

MATERIALS AND METHODS

Our computational model is a modified version of that introduced by Brunel and Wang (2001) with the inclusion of a dopamine component on GABAergic interneurons (see equation (10)). Briefly, a PFC network (Figure 1) of 2000 cells composed by 80% pyramidal neurons and 20% FSI was modeled. Two subsets of 240 pyramidal neurons respond selectively to input stimuli S1 or S2 (S-responding neurons). The remaining 1120 pyramidal neurons belong to a cluster of non-selective (NS) responding cells. Per simulation, 100 independent trials were run for each condition using parameters that were adjusted to avoid potential ceiling/floor effects, especially to modifications of dopamine and GABA. Changes in neuronal firing rate were estimated using a non-overlapping window of 25 ms. Differential equations were determined by an integration time step of dt=0.01 ms.

Figure 1
figure 1

(a) The PFC model is comprised by groups of selective (S) and non-selective (NS) pyramidal neurons, and fast-spiking inhibitory interneurons (FSIs). The group of S-responding pyramidal neurons is composed by two clusters of 240 cells each that respond selectively to input stimuli S1 and S2, respectively. Synaptic connections among pyramidal neurons within each of the S-responding group are stronger (w=1.9) than those between S1- and S2-responding cells (w=0.84). A synaptic efficacy of w=1 was used for connections within NS pyramidal neurons and FSI, and between pyramidal neurons (both S and NS) and FSI. The global inhibitory tone is provided by FSI with a synaptic efficacy of w=1. (b) Raster of cell firing showing examples of neuronal activity across the different populations before and after input stimulus S1 (indicated by the arrow and gray line; calibration bar: 250 ms).

PowerPoint slide

Full size image

The membrane potential Vm results from the integration of external and recurrent excitatory and inhibitory currents:

Cm=0.5 nF/0.2 nF is the membrane capacitance whereas gm=25 nF/20 nF and VL=−70 mV account for the leakage conductance and leakage potential for excitatory/inhibitory neurons, respectively. If Vm surpasses the threshold potential VTH=−50 mV, an action potential is elicited, and the membrane potential is held at the reset potential Vr=−55 mV for the duration of the refractory period (2.5 ms for pyramidal neurons and 1.5 ms for FSI).

Synaptic currents for both pyramidal neurons (Isyn-pyr) and FSI (Isyn-FS) were computed as follows:

AMPAext, AMPArec, and NMDArec account for AMPA and NMDA receptor-mediated currents resulting from external and recurrent glutamatergic drive. The inhibitory tone from FSI originates from GABA-A receptor-mediated transmission. Dopamine modulation of neuronal activity in the developmentally mature PFC network is modeled by including both the well-documented D1-positive modulation of NMDA receptor-mediated response (ID1-NMDA) and the facilitatory action of D1 and D2 receptors on FSI excitability (IDA) (Tseng and O’Donnell, 2004, 2005, 2007b).

AMPA and GABA-A currents are computed as follows:

where gAMPAext=2.08 nS/1.62 nS, gAMPArec=0.104 nS/0.135 nS, and gGABA=1.25 nS/0.973 nS are the nominal values for AMPA and GABA conductances for both excitatory and inhibitory neurons, respectively. VE=0 mV and VI=−70 mV are the reversal potentials for excitatory and inhibitory cells, respectively. SAMPAext, SAMPArec, and SGABA are action potential-driven variables that modulate the magnitude of the net conductances. The instantaneous values for recurrent conductances are also affected by the connectivity parameter, which is wj=1 between FSI and the three groups of pyramidal cells (S1, S2, and NS), and among pyramidal neurons of the NS group. The connectivity parameter is wj=1.9 among neurons within the same S-responding group and wj=0.84 between the two S-responding groups (see Figure 1).

The nominal value of the NMDA current is modulated by a voltage-dependent Mg2+ block of the channel.

where gNMDA=0.327 nS/0.258 nS is the NMDA conductance for excitatory/inhibitory neurons, respectively and φD1-NMDA accounts for the D1 facilitation of NMDA currents (see equation (9)).

Action potential-driven variable s accounts for the dynamics of AMPA and NMDA conductances. Fast conductances for AMPA- and GABA-mediated transmission were modeled as exponential functions:

For NMDA conductances, we used the following differential equation:

Here, Kronecker’s delta (δ) is the epoch for the onset of presynaptic spikes whereas τ is the time constant, τAMPA=2 ms, τGABA=5 ms, τNMDArise=2 ms, and τNMDAdecay=100 ms. In the model, the number of spikes arriving from external inputs follows a Poisson function with a basal λ value of 2400 spikes/s. Thus, λ value increases only in selective neurons in response to stimulus presentation, a function that is determined by the stimuli contrast, which is typically 10% in our simulations (Supplementary Figure 1). In this regard, when a stimulus with a given contrast value c is presented, an increase in λ (defined as c.λ) will be applied only to those inputs that were previously defined as selective for the S-responding cells.

To simulate the D1 action on pyramidal neurons and FSI cells, the ΦD1-NMDA was included in equation (9), a variable that depends on basal dopamine levels (Brunel and Wang, 2001).

where DA is the dopamine level and ϕ is a parameter that shifts the sigmoid function depending on the cell type (ϕ=1 for pyramidal cells and ϕ =1.05 for FSI) as in Brunel and Wang (2001).

The dopamine modulation of FSI activity was computed by means of a Na+ conductance-like current with a reversal potential of VDA=55 mV:

Here, changes in dopamine levels were simulated by modifying the basal gDA=0.35 nS conductance in a linear-dependent manner. This basal gDA value was chosen to maintain FSI activity in the adult PFC within physiological ranges as seen in vivo (Tseng et al, 2006).

To test how the model responds to selective stimuli, we simulated the presentation of one of the selective stimuli S for 250 ms followed by a delay of 1 s, and computed the following measures:

Both the mean firing rate (μFRSi) of S-responding neurons and its variance (σ 2FRSi) were taking into account. For d’, μFRSi and σ2FRSi were calculated from the stimulus presentation period whereas the WMI was obtained from a 250-ms window at the end of each trial 500 ms after the stimulus offset.

RESULTS

Our PFC model comprises a network of 2000 neurons with an inhibitory/excitatory ratio of 0.25 (Figure 1), and includes the following physiological features of dopamine action: (i) dopamine facilitation of prefrontal GABAergic transmission via activation of local FSIs (Tseng and O’Donnell, 2007b); (ii) D1 facilitation of NMDAR-mediated response in both pyramidal neurons and FSI. We first determined the differential effects of transient steps of dopamine elevation on the spontaneous activity of pyramidal neurons and FSI. For each simulation, changes in neuronal mean firing rate were estimated using a 25-ms non-overlapping window. Under basal dopamine tone, pyramidal neuron discharge activity was ∼0.8 Hz whereas the mean firing rate for FSI was ∼9 Hz. Consistent with the biological effects found in the PFC in response to phasic ventral tegmental stimulation in vivo (Tseng et al, 2006), the simulations revealed that increasing steps of dopamine (mimicking phasic dopamine) also elicited facilitation of FSI activity concurrent with a coordinated suppression of pyramidal cell firing (Figure 2a). In our model, a complete cessation of pyramidal cell activity was observed when a step of threefold dopamine increase and subsequent elevation of FSI activity by ∼35% were simulated. These results pointed toward a non-linear relationship between pyramidal cell activity and FSI function. Accordingly, a steady downregulation of local GABAergic tone (GABA degradation) exponentially increased the mean firing rate of pyramidal neurons (Figure 2b). Interestingly, the hyperactive state resulting from the reduced GABAergic inhibition can be normalized by augmenting basal dopamine levels (Figure 2c). Together, these results indicate that one critical mechanism to control pyramidal cell activity is the increased responsiveness of FSI to dopamine.

Figure 2
figure 2

(a) FSI and pyramidal cells in the PFC exhibit opposite firing responses to increasing steps of dopamine level (from 0, 0.5, 1, 2, 3, 4, and 5 folds) above baseline. Black arrow indicates the onset of dopamine increase. (b) Impact of downregulation of local prefrontal GABAergic inhibitory tone (GABA degradation) on pyramidal cell activity. Note the exponential increase in pyramidal cell firing following GABA degradation >20%. (c) Relationship between basal dopamine tone and pyramidal cell firing at different degrees of GABA degradation.

PowerPoint slide

Full size image

We next assessed how changes in basal dopamine levels (from −10 to +10%) affect the ability of the PFC network to discriminate between stimuli by measuring d’ (see Materials and Methods for details). In our model, there are two populations of pyramidal neurons that respond exclusively to external input stimuli S1 or S2 (Figure 3a). We first determined how different contrast values between S1 and S2 inputs (ranging from 5 to 80%) affect the d’ discrimination curve (Supplementary Figure 1). Data from these simulations show that a minimal 10% contrast value is required to enable the emergence of a dopaminergic modulation of d’. Thus, input selectivity (computed as d’) is maximal under the basal dopamine state (Figure 3b) because S1-responding neurons typically increase their activity to S1, but remain unresponsive to S2 and vice versa (Figure 3c). Interestingly, such selectivity begins to diminish as dopamine tone moves away from baseline irrespectively of the direction of change (Figure 3b). Further analyses revealed that the deterioration of selectivity in the low dopamine state arises from an increased responsiveness of S1- and S2-responding pyramidal neurons to the non-preferred input stimuli (Figure 3c). This latter effect is likely due to an insufficient level of GABAergic inhibition needed to maintain network stability (Figure 2b and c) as similar degrees of diminished selective response can be obtained following GABA degradation (Figure 3d). Conversely, the reduced selectivity observed upon increasing levels of dopamine is due to an augmented GABAergic tone exerting a strong inhibition on both S1- and S2-responding pyramidal neurons to the preferred input stimuli (Figure 3c). In fact, GABA degradation effectively improves selectivity only in the high dopamine state as indicated by a right-shift of the normalized d’ curve (Figure 3d). Together, these results indicate that a fine homeostatic interplay between dopamine function and local GABAergic transmission is required for maintaining PFC network stability and output selectivity.

Figure 3
figure 3

(a) Diagram of the PFC network highlighting the two populations of S-responding pyramidal neurons and their exclusive responsiveness to external input stimuli S1 or S2. In this example, activation of S1-responding pyramidal neurons results in inhibition of the S2-responding group through a fast-spiking interneuron (FSI) mechanism. (b) Relationship between basal dopamine tone and prefrontal discrimination of selective stimuli. The selectivity of the PFC response was determined by d’. Higher d’ values reflect better selectivity as seen at baseline dopamine while decreasing or increasing dopamine tone deteriorate stimuli selectivity. (c) Examples of S1- and S2-responding pyramidal neurons activity during the presentation of S1 input stimulus at three levels of basal dopamine. Data show the mean firing rate of S1- and S2-responding cells across 40 trials (vertical calibration bars: 4 Hz). (d) Impact of GABA degradation on stimuli selectivity at increasing levels of basal dopamine (from 0 to +40% above baseline). Vertical dashlines indicate the % of GABA degradation associated with the highest degree of selectivity for each dopamine state. A full color version of this figure is available at the Neuropsychopharmacology journal online.

PowerPoint slide

Full size image

Among the different PFC-dependent functions, working memory is of interest because it is associated with the ability of neuronal populations to sustain activity from hundreds of milliseconds to seconds in response to a cue presentation for solving a task several hundred of milliseconds later (Goldman-Rakic, 1995). Typically, such sustained activity is resistant to interferences produced by distractors, that is, task-independent stimuli capable of perturbing the correct response (Brunel and Wang, 2001). Thus, retention of cue-associated information over distractors is critical to correctly execute the task. Here, we simulated this paradigm and asked how working memory becomes affected by changes in dopamine and GABA functions. In each trial, 250 ms duration of S stimulus presentation is followed by a 300-ms delay interval before a distractor is turned on. For S1-responding pyramidal neurons the distractor was the S2 stimulus, and vice versa for S2-responding cells. Working memory retention was considered successful if the average activity of S-responding pyramidal neurons (measured 250 ms after the offset of the distractor) remained higher than the rest of the pyramidal cells. We first examined how changes in the functional connectivity among pyramidal neurons of the same S-responding group alter working memory. By increasing and decreasing the strengths of such connectivity, we found that the relationship between working memory and synaptic weights follows an inverted U-shape function (Figure 4a). Thus, only a range of synaptic weights favors retention of working memory by enabling sufficient input selectivity and recurrent excitation to sustain persistent activity in S-responding pyramidal neurons. This balance between input selectivity and recurrent activity becomes disrupted when too much or too little synaptic strengths were simulated, all of which prevent the formation of working memory (Figure 4b). Next, we investigated how changes in dopamine and GABA impact the retention of working memory (Figure 5). These simulations were conducted using a value of synaptic connectivity among S-responding pyramidal neurons that was sufficient to elicit working memory at baseline (w=1.9, Figure 4a). The data show that high values of working memory performance (ie, working memory index >0.85) can be achieved only within a particular combination of dopamine/GABA function (Figure 5a). Interestingly, we found an inverted U-shape relationship between working memory and dopamine tone, a function that is maintained while increasing the levels of GABA degradation (Figure 5b). In fact, the working memory deficits induced by GABA degradation can be rescued by increasing dopamine tone, resulting in a shift of the working memory index curve to the right (Figure 5b). Further analyses revealed a non-linear interplay between dopamine and GABA function in sustaining working memory stability in the PFC (Figure 5c).

Figure 4
figure 4

(a) Relationship between working memory performance (measured as working memory index, WMI) and changes in synaptic weight (w) among pyramidal neurons of the same S-responding group at baseline dopamine and GABA levels. Arrow indicates the original synaptic weight (w=1.9) used in the simulations shown in Figures 2, 3, 4. Note that a reduction in the WMI occurs at both high and low values of w. (b) Examples of S1 (black)- and S2 (red)-responding pyramidal neurons’ activity during and after the presentation of S1 stimulus (250 ms duration) showing the different mechanisms by which a change in w can diminish the WMI (horizontal calibration bar: 250 ms). Increasing values of w (eg, w=2.4) eliminate the selectively of S-responding neurons whereas decreasing values of w (eg, w=1.6) reduce the level of recurrent activity needed to maintain persistent neuronal firing. A full color version of this figure is available at the Neuropsychopharmacology journal online.

PowerPoint slide

Full size image
Figure 5
figure 5

(a) U-shape relationship between basal dopamine tone and PFC working memory capacity (computed as working memory index) measured at normal GABA function. The synaptic weight among S-responding pyramidal neurons was 1.9 (see Figure 4). (b) In the presence of increasing levels of GABA degradation, a higher basal dopamine tone is needed to achieve working memory. Interestingly, the inverted U relationship between dopamine and working memory is maintained despite the GABA degradation. Thus, a particular combination of dopamine and GABA function is required to achieve optimal working memory performance (above the dashed line; working memory index >0.85). (c) Scatter plot summarizing the non-linear relationship between the degrees of GABA degradation needed to restore working memory equilibrium (arrows in b) in response to increasing levels of basal dopamine.

PowerPoint slide

Full size image

We next determined whether phasic activation of FSI is sufficient to bring S-responding pyramidal cell firing back to baseline and reset PFC working memory. To test this, a step of transient dopamine increase (500 ms duration) above the baseline level was introduced after the selective stimulus S and once the S-responding pyramidal neurons had reached to a steady state of increased activity (Figure 6a). Results from the simulations show that phasic augmentation of dopamine (ie, fold change above baseline) can effectively decrease the elevated firing state of S-responding neurons in a concentration-dependent manner (Figure 6b). In fact, the ability of dopamine transients to reset the activity of S-responding cells is determined by the magnitude of inhibition elicited during the 500-ms phasic dopamine increase (Figure 6b). In our model, this inhibitory action is dictated by FSI (Figure 6c), whose level of responsiveness increases linearly to dopamine differentials (Figure 6d), and by the duration of the dopamine transient (Supplementary Figure 2). Interestingly, the extent of dopamine-induced FSI excitation needed to reset the S-responding neurons follows a non-linear function (Figure 6d). Further analyses revealed that the point of inflexion for the reset occurs when the activity of FSI surpasses 40 Hz (Figure 6e). Together, these results underscore the critical role of dopamine–FSI interaction in the regulation of working memory reset in the PFC.

Figure 6
figure 6

(a) Firing rate time course of S-responding pyramidal neurons and FSI during and after the selective stimulus S followed by a 500-ms step of dopamine (DA) increase. Here, a step of dopamine increase (3-fold above the basal tone) was used to inhibit and reset (R) the activity of S-responding pyramidal cells back to baseline values. (b) Summary of the effects of increasing dopamine levels from 0.5- to 3-fold above baseline on S-responding neuronal activity measured at 500 ms after the offset of the selective stimulus (S), 250 ms after the onset of the dopamine transient (DA), and 250 ms after the offset of the dopamine step (R). Note that while the degree of inhibition in S-responding cells is positively correlated with the dopamine transients, a critical level of inhibition is required to enable reset (ie, +1.5 DA). (c) FSI response to increasing steps of dopamine transients (same as in b). Note the rapid, but transient increase in firing rate after the onset of the dopamine pulse (dashed line). (d) Relationship between dopamine transients and the FSI peak response or the mean firing rate of S-responding pyramidal neurons at 250 ms post-dopamine offset (ie, R from a). (e) Summary of the FSI peak response to dopamine steps and the mean firing rate of S-responding pyramidal neurons at 250 ms post-dopamine offset. A full color version of this figure is available at the Neuropsychopharmacology journal online.

PowerPoint slide

Full size image

DISCUSSION

Our PFC model predicts that behavioral outcomes associated with an initial dopamine elevation will increase FSI activity. As a result, the signal detection ratio in the PFC network increases by virtue of reduced recurrent activity in pyramidal neurons. Further simulations revealed that a fine homeostatic interplay between dopamine and FSI is needed to enable PFC output selectivity and stability. A similar dopamine–FSI interaction is required for the formation and retention of working memory, especially in the presence of distractor stimuli. Finally, our model also predicts that phasic activation of FSI by dopamine is an effective mechanism to reset the PFC working memory state back to baseline. Together, these results show for the first time that a critical gain of prefrontal FSI function by dopamine is necessary for maintaining PFC network stability, which enables working memory retention and reset.

A major inhibitory action of dopamine in the PFC results from local activation of GABAergic interneurons (Gorelova et al, 2002; Gulledge and Jaffe, 1998; Tseng and O’Donnell, 2004, 2007a, 2007b). In the adult PFC, pharmacological stimulation of both D1 and D2 receptors converge to enhance FSI excitability (Tseng et al, 2008; Tseng and O’Donnell, 2007b). Hence, the net effect of mesocortical dopamine is to drive FSI and subsequently inhibit pyramidal output neurons. We simulated these interactions and found that the activity of pyramidal neurons in the PFC becomes inhibited following phasic dopamine increase. This is consistent with in vivo studies showing suppression of pyramidal cell firing following stimulation of the mesocortical dopamine pathway (Ferron et al, 1984; Jay et al, 1995; Lewis and O’Donnell, 2000; Pirot et al, 1992; Tseng et al, 2006). Although several mechanisms have been proposed to explain this inhibition, paired recordings of interneurons and pyramidal cells revealed that a coordinated activation of FSI in the PFC could account for the mesocortical-induced prefrontal output inhibition (Tseng et al, 2006). We tested this hypothesis and found a non-linear relationship between pyramidal cell firing and FSI. In fact, the mean firing rate of pyramidal neurons begins to increase exponentially when the FSI-dependent GABAergic tone decreases >20%. Interestingly, a reinstatement of the FSI function by augmenting basal dopamine levels was sufficient to reduce the hyperactive PFC state to baseline levels. Thus, one critical mechanism for shaping the activity of PFC output neurons is the responsiveness of FSI to dopamine.

Parvalbumin-positive FSI is the most abundant class of GABAergic cells in the adult PFC (Gabbott et al, 1997). As a result of their non-adapting and fast-spiking firing pattern, FSI is functionally positioned to exert fast-feedforward inhibition onto pyramidal neurons and controls the signal detection ratio of the cortical output (Bartos and Elgueta, 2012; Rao et al, 2000). Our simulations predict that a proper level of FSI function is required for enabling input-specific processing of afferent information among PFC pyramidal neurons. For example, when the basal dopamine tone is low, the input selectivity begins to deteriorate due to insufficient levels of FSI activity to sustain network stability. Interestingly, too much dopamine tone also reduces PFC selectivity because of excessive inhibition of pyramidal neurons by FSI. Thus, a disruption of FSI function is expected to reduce the PFC capacity for discriminating contextually and emotionally salient signals, in particular those originating from the ventral hippocampus (Abela et al, 2012; Chudasama et al, 2012; Floresco et al, 1997; Seamans et al, 1995; Wang and Cai, 2006) and the basolateral amygdala (Davis and Whalen, 2001; Garcia et al, 1999; Gilmartin and Helmstetter, 2010; Milad and Quirk, 2012; Morgan and LeDoux, 1995), which are key for the development of mature cognitive abilities associated with adult behavior (Best and Miller, 2010; Casey et al, 2000, 2008). In this regard, the well-established role of dopamine on PFC-dependent cognitive functions (see review Floresco, 2013) could be attributable to the fine homeostatic interplay between FSI and dopamine transmission found in our simulations.

Neurocomputational models of PFC function have been successfully used to study neuronal dynamics of dopamine modulation and working memory (Brunel and Wang, 2001; Durstewitz et al, 2000; Tanaka, 2001). Of particular interest is the work by Durstewitz et al (2000) wherein the GABAergic component was modeled as a dependent variable of the dopamine’s effect, and noticed that concurrent augmentation of the GABA-A conductance in pyramidal neurons is required for proper functioning of the network, in particular to dopamine-driven actions. However, the role of GABA interneurons has never been taken into account as an independent variable in these PFC simulations (see Wang et al, 2004) despite the fact that acute blockade of prefrontal GABAergic transmission in animal models has been repeatedly shown to impair cognitive functions within the working memory domain (Enomoto et al, 2011; Paine et al, 2011; Sawaguchi et al, 1988, 1989). Thus, a more pressing question is how dopamine regulation of GABAergic interneurons interplays with the recurrent activity of pyramidal neurons to enable working memory in the PFC, and whether these emerging compensatory mechanisms could be implemented to restore the normal prefrontal output function in pathophysiological conditions of aberrant GABAergic function and/or dopaminergic transmission. By incorporating physiological features of dopamine’s action onto FSI (see Discussion above), we have uncovered a range of non-linear interactions between dopamine and FSI for sustaining an optimal working memory performance. For instance, working memory deficits resulting from downregulation of FSI inhibition can be restored by supra-linearly increasing basal dopamine levels. Results from the simulations also revealed a series of inverted U-shape relationships between working memory and prefrontal dopamine. A similar inverted U-shape function between PFC dopamine and cognition has been often described in animal studies (see review Floresco, 2013), yet the precise neuronal substrate underlying such a relationship remains largely unknown. On the basis of our model, we predict that only an optimal combination of dopamine transmission and FSI function in the PFC will enable appropriate network stability to sustain working memory. In this regard, increasing dopamine tone in the PFC is expected to improve working memory if a functional deficit in local GABAergic transmission is responsible for the sub-optimal prefrontal performance.

In addition to working memory impairments, deficits in prefrontal GABAergic interneurons have been often associated with reduced cognitive flexibility and perseverative behaviors as seen in animal models (Brady, 2009; Gruber et al, 2010) as well as in psychiatric disorders including schizophrenia (Lewis and Gonzalez-Burgos, 2006; Uhlhaas and Singer, 2006). Although several mechanisms have been proposed, a common theme thought to account for the lack of behavioral flexibility resulting from prefrontal disinhibition is the inability of the PFC network to reset (Floresco, 2013; Gruber et al, 2010). Here, we asked whether features of dopamine–FSI interaction found in our PFC model could have a permissive role to reset prefrontal activity. Results from the simulations indicate that once PFC working memory is formed, a phasic activation of FSI by dopamine (ie, 500 ms) can effectively bring the steady state of working memory-dependent persistent activity down to baseline only if the dopamine transient occurs after the offset of the conditioned stimuli (see Figure 6) and last >350 ms in duration (Supplementary Figure 2). Interestingly, our model also predicts that the level of pyramidal cell inhibition needed to achieve reset requires that activation of FSI surpasses 40 Hz. Thus, a failure to engage sufficient FSI activation is expected to reduce the PFC network cognitive capacity to switch from one state to another. Future studies are warranted to validate these neurocomputational observations in animal models.

In summary, we have presented a neurocomputational model of the PFC that takes into account the often-overlooked role of dopamine modulation of prefrontal GABAergic transmission. This model provides new insights on how the dopaminergic system interacts with both pyramidal neurons and GABAergic interneurons to achieve the complex balance necessary to sustain network stability and input selectivity for the formation and reset of working memory in the PFC.

FUNDING AND DISCLOSURE

The authors declare no conflict of interest.