Neurocomputing 44–46 (2002) 1035 – 1042

www.elsevier.com/locate/neucom

Circuit simulation of memory !eld modulation by dopamine Dl receptor activation Koki Yamashita, Shoji Tanaka ∗ Laboratory of Cortical Circuits and Computation, High-Tech Research Center, Department of Electrical and Electronics Engineering, Sophia University, 7-1 Kioicho Chiyoda-Ku, 102-8554, Tokyo, Japan

Abstract We have developed a computational model of the PFC circuit that incorporates dopamine e0ects via Dl receptor activation. The computer simulation with this model shows an inverted-U shaped concentration-dependent e0ect of dopamine on the delay-period activity, which is consistent with the experimental result (Nature 376 (1995) 572). The activities of the pyramidal cells (both the baseline and the signal components) are suppressed in high Dl receptor activation levels. The simulation suggests that this is due to feedforward inhibition when the decrement of the AMPA-type currents is weak enough. When it is stronger, however, recurrent inhibition c 2002 Elsevier Science B.V. All rights reserved. dominates the inhibition.  Keywords: Dl receptor; Dopamine; Intracortical inhibition; Prefrontal cortex; Working memory

1. Introduction Dopamine Dl receptors play critical roles in working memory processes [1,6]. Recently, Williams and Goldman-Rakic [15] have shown that dopamine modulates the memory !elds of PFC neurons via the activation and inactivation of Dl receptors. One of the signi!cant consequences of their !ndings is the inverted-U shape modulation of memory !elds [2]. That is, when the dopamine level in the PFC is lower than the optimum, the application of dopamine enhances the memory !elds, but the opposite e0ects would be observed when the dopamine level is higher than the optimum. Recent evidence shows that Dl receptor activation by dopamine enhances NMDA-type postsynaptic currents (PSCs) [7,16]. According to this and recent !ndings, Goldman-Rakic ∗

Corresponding author. Tel.: +81-3-3238-3331; fax: +81-3-3238-3321. E-mail addresses: [email protected] (K. Yamashita), [email protected] (S. Tanaka).

c 2002 Elsevier Science B.V. All rights reserved. 0925-2312/02/$ - see front matter
PII: S 0 9 2 5 - 2 3 1 2 ( 0 2 ) 0 0 5 0 8 - 8

1036

K. Yamashita, S. Tanaka / Neurocomputing 44–46 (2002) 1035 – 1042

and coworkers proposed a conceptual model for dopamine modulation of mnemonic activities of PFC neurons [4,5]. Although the overall e0ects of dopamine on working memory processes are still controversial, it is interesting to examine, at present, if their model works properly to account for the modulatory e0ects observed by Williams and Goldman-Rakic [15]. To do so, we here propose a computational model of the PFC circuit, which is an extended version of the previous models [4a,4b,8–14]. The neurons in this model have AMPA, NMDA, GABAA , persistent sodium, and calciumdependent potassium channels, all of which but GABAA channels are to be modulated by dopamine. The computer simulation of the model circuit will explore how the dopamine modi!es the circuit dynamics for the representation of spatial working memory. 2. Model 2.1. Neuron model The neurons (pyramidal cells and interneurons) are described here with a single compartment, leaky integrate-and-!re neuron model C

dVi + IAMPA + INMDA + IGABAA + INaP + IK(Ca) + Ileak = 0; dt

where IAMPA =




(1)

gAMPA; ji (t − tji )(Vi − EAMPA );

(2)

gNMDA; ji (t − tji )fMg (Vi )(Vi − ENMDA );

(3)

j

INMDA =


j

IGABAA =




gGABA; ji (t − tji )(Vi − EGABAA );

(4)

j

INaP = gNaP (Vi )(Vi − ENa );

(5)

IK(Ca) = gK(Ca) ([Ca2+ ]i )(Vi − EK );

(6)

Ileak = gleak (Vi − Eleak ):

(7)

Here tji =tsp; j +Jtji , where tsp; j is the time at which the presynaptic neuron j spikes and Jtji is the transmission and synaptic delay. The conductances, gAMPA; ji (t), gNMDA; ji (t), and gGABAA , are described by second-order systems. 1 ; 1 + 0:5 exp(−0:062Vi )    Vi + 56 ; gNaP (Vi ) = gNaP; max = 1 + exp − 7 fMg (Vi ) =

(8) (9)

K. Yamashita, S. Tanaka / Neurocomputing 44–46 (2002) 1035 – 1042

1037

Table 1 The e0ects of Dl receptor activation assumed in this model

Dl receptor activation level

rNMDA (pyramidal cell) rNMDA (interneuron) rAMPA (Case A) rAMPA (Case B) rK(Ca) v (mV)

−3

−2

−1

0

+1

+2

+3

+4

+5

0.300 0.001 1.110 1.110 1.110 2.0

0.400 0.080 1.100 1.100 1.100 1.5

0.600 0.300 1.090 1.090 1.090 1.0

1.000 1.000 1.000 1.000 1.000 0

1.010 1.300 0.950 0.900 0.950 −0:5

1.050 1.500 0.940 0.790 0.940 −1:0

1.100 1.800 0.935 0.750 0.935 −1:5

1.180 2.500 0.930 0.700 0.930 −2:0

1.200 3.000 0.900 0.650 0.900 −2:5

gK(Ca) ([Ca2+ ]i ) = K [Ca2+ ]i ;

(10)


[Ca2+ ]i d[Ca2+ ]i = Ca (t − tspike; i ) − ; dt Ca

(11)

spike

The equilibrium potentials are EAMPA = 0 mV; ENMDA = 0 mV; EGABAA = −90 mV; ENa = 50 mV; EK = −80 mV; Eleak = −70 mV. 2.2. E5ects of D1 receptor activation As dopamine e0ects on the ion conductances, our model includes the following four e0ects: (i) (ii) (iii) (iv)

Suppression of IAMPA (replacing gAMPA by rAMPA gAMPA ). Enhancement of INMDA (replacing gNMDA by rNMDA gNMDA ). Enhancement of INaP (replacing VNaP and ENa by VNaP + v and ENa + v). Reduction of IK(Ca) (replacing gK(Ca) by rK(Ca) gK(Ca) ).

The values of the modulated parameters are given in Table 1. 2.3. Circuit architecture The network contains 1320 neurons, of which 1080 (81.8%) are pyramidal cells and 240 (18.2%) are inhibitory interneurons. The network had three layers, which are the super!cial, the intermediate, and the deep layers. Our models describes all of the connectivity pro!les with Gaussian functions with the standard deviations of 5◦ (for the pyramidal-to-pyramidal connections), 8◦ (for the pyramidal-to-interneuron connections), and 30◦ (for the interneuron-to-pyramidal and interneuron-to-interneuron connections).

1038

K. Yamashita, S. Tanaka / Neurocomputing 44–46 (2002) 1035 – 1042

Fig. 1. Time courses of the membrane potentials of the pyramidal cell in the deep layer for the low, intermediate, and high levels of Dl receptor activation. The preferred directions of these neurons are 0◦ (A and C) and 180◦ (B and D). The cue direction is 0◦ . The horizontal bars in the !gure indicate the duration in which the cue-related input is given (200 –300 ms).

3. Results 3.1. Transient and sustained activities The pyramidal cells receive external inputs cueing a target location. Then the activity is transmitted to the super!cial layer and then to the deep layer. Fig. 1 shows the activities of the pyramidal cell in the deep layer. Due to the recurrent connections of the pyramidal cells, the activities of the pyramidal cells in both the super!cial layer and the deep layer sustain during the delay period when the dopamine level is optimally chosen. When the dopamine level is deviated either to a high level or to a low level, the !ring no longer sustains (Figs. 1A and C).

K. Yamashita, S. Tanaka / Neurocomputing 44–46 (2002) 1035 – 1042

1039

Fig. 2. Activity pro!les of the pyramidal cells in the super!cial, intermediate, and deep layers for the low (A), intermediate (B), and high (C) levels of Dl receptor activation (Case A). (D) shows the result when the dependence of the AMPA conductance on the Dl receptor activation level is stronger than (C) (Case B).

3.2. Inverted-U shape modulation of memory 7elds At the optimum dopamine level, the pyramidal cells in the intermediate layer exhibit transient, cue response, while the pyramidal cells in the super!cial and deep layers exhibit well-tuned delay-period activity. When the dopamine level is decreased from the optimum level, the excitability of both the pyramidal cells and interneurons decreases. The maximum !ring rates of the memory !elds decrease, and !nally the memory !elds disappear at a certain level of dopamine. As the dopamine level is increased from the optimum level, on the other hand, the excitatory transmission to both the pyramidal cells and the interneurons increases. However, the maximum !ring rates of the memory !elds decrease because of sharper increase in the excitability of the interneurons. Above a critical level of dopamine, the inhibitory e0ect (feedforward inhibition) completely suppresses the delay-period activity, resulting in no memory !eld formation. The activity pro!les of the pyramidal cells and the interneurons for these three di0erent levels of Dl receptor activation are depicted in Figs. 2A–C.

1040

K. Yamashita, S. Tanaka / Neurocomputing 44–46 (2002) 1035 – 1042

Fig. 3. Signal and baseline component of the delay-period activities of the pyramidal cell (A) and the interneuron (B) whose preferred directions coincide with the target direction (Case A). (C) (pyramidal cell) and (D) (interneuron) show the result when the dependence of the AMPA conductance on the Dl receptor activation level is stronger (Case B). The activities are averaged over the delay period of 3:0 s.

In the above simulation, the decrement of the AMPA conductance due to Dl receptor activation was assumed to be weak (Case A in Table 1). We did another simulation, in which the dependence of the AMPA conductance on the Dl receptor activation level is stronger than the above (Fig. 2D, Case B in Table 1). The result shows that the background activity level of the interneurons becomes lower than that at the optimum level. As a result, the memory !elds of the pyramidal cells and the interneurons do not vanish completely. This shows a marked di0erence from the earlier case (Fig. 2C). The signal and baseline components of the delay-period activities of a pyramidal cell and an interneuron are depicted in Fig. 3. In the shaded regions in the !gure, the activities persist during the delay period and the !ring rates change with the Dl receptor activation level. Outside the shaded region, however, the activities suddenly stop in the

K. Yamashita, S. Tanaka / Neurocomputing 44–46 (2002) 1035 – 1042

1041

delay period. The signal components (or the memory !elds) of both the pyramidal cells and the interneurons show inverted-U shape dependences on the Dl receptor activation level. The background activity of the pyramidal cells decreases with the Dl receptor activation level. The background activity of the interneurons is higher than that of the pyramidal cells in our simulation. Moreover, the background activity of the interneurons increases with the Dl receptor activation level in Case A (Fig. 3B). Because of this, the feedforward inhibition works e0ectively, especially in the high DA regime, to suppress the activity of the pyramidal cells. In Case B, on the contrary, the background activity of the interneurons decreases with the Dl receptor activation level (Fig. 3D). In this case, the recurrent inhibition dominates the inhibition. 4. Discussion This model has taken into account the dopamine e0ects on the AMPA and NMDA channels, the persistent sodium channels, and the calcium-dependent potassium channels. Because the quantitative dependencies of these neurons on dopamine are unknown yet, we have tentatively chosen the several di0erent sets of the values of the parameters that are modulated by dopamine. In this simulation, the sensitivity of the AMPA-channel conductance was assumed to be weaker than that of the NMDA-channel conductance, in accordance with experimental results [6]. The inputs to the interneurons then increased in the high Dl receptor activation regime. This results in higher background activity of the interneurons, which suppress the !ring of the pyramidal cells (feedforward inhibition). Gao et al. [3] reported the reduction of AMPA-channel transmission was greater than the previously reported. Consequences of this e0ect is unknown yet. In our simulation with a greater reduction of AMPA-channel transmission, the background activity of the interneurons decreased with the dopamine level. This means that the feedforward inhibition did not work e0ectively in the high dopamine regime. Instead, recurrent inhibition suppressed the activity of the pyramidal cells in the high dopamine regime. That is, the recurrent inhibition takes the place of the feedforward inhibition when the reduction of AMPA-channel transmission is greater. In the recurrent inhibition, the inputs to the interneurons mainly come from the nearby pyramidal cells. Both scenarios yield the inverted-U shape characteristic of the memory !eld modulation by dopamine. However, the circuit mechanisms in these scenarios are di0erent. More quantitative research is necessary to elucidate the circuit mechanisms of the biphasic modulatory e0ects. Acknowledgements The author (S.T.) acknowledges valuable discussions with Prof. P.S. Goldman-Rakic and Dr. G.V. Williams at the Yale University School of Medicine. This work was supported by Grants-in-Aid for Scienti!c Research on Priority Areas to S.T. (#13210123) from the Ministry of Education, Science, and Technology, Japan.

1042

K. Yamashita, S. Tanaka / Neurocomputing 44–46 (2002) 1035 – 1042

References [1] A.F.T. Arnsten, J.X. Cai, B.L. Murphy, P.S. Goldman-Rakic, Dopamine Dl receptor mechanisms in the cognitive performance of young adult and aged monkeys, Psychopharmacology 116 (1994) 143–151. [2] R. Desimone, Is dopamine a missing link? Nature 376 (1995) 549–550. [3] W.-J. Gao, L.S. Krimer, P.S. Goldman-Rakic, Presynaptic regulation of recurrent excitation by Dl receptors in prefrontal circuits, Proc. Natl. Acad. Sci. U.S.A. 98 (2001) 295–300. [4] P.S. Goldman-Rakic, E.G. Muly III, F.V. Williams, Dl receptors in prefrontal cells and circuits, Brain Res. Rev. 31 (2000) 295–301. [4a] M. Iida, S. Tanaka, Postsynaptic current analysis of a model prefrontal cortical circuit for multi-target spatial working memory, Neurocomputing 44–46 (2002) 855–861, this issue. [4b] K. Morooka, S. Tanaka, Correlation analysis of signal Row in a model prefrontal cortical circuit representing multiple target locations, Neurocomputing 44–46 (2002) 541–548, this issue. [5] E.C. Muly III, K. Szigeti, P.S. Goldman-Rakic, Dl receptor in interneurons of macaque prefrontal cortex: distribution and subcellular localization J. Neurosci. 15 (1998) 10553–10565. [6] T. Sawaguchi, P.S. Goldman-Rakic, The role of Dl-dopamine receptor in working memory: local injections of dopamine antagonists into the prefrontal cortex of rhesus monkeys performing an oculomotor delayed-response task J. Neurophysiol. 71 (1994) 515–528. [7] J.K. Seamans, D. Durstewitz, B.R. Christie, C.F. Stevens, T.J. Sejnowski, Dopamine D1=D5 receptor modulation of excitatory synaptic inputs to layer V prefrontal cortical neurons, Proc. Natl. Acad. Sci. U.S.A. 98 (2001) 301–306. [8] S. Tanaka, Architecture and dynamics of the primate prefrontal cortical circuit for spatial working memory, Neural Networks 12 (1999) 1007–1020. [9] S. Tanaka, Roles of intracortical inhibition in the formation of spatially tuned delay-period activity of prefrontal neurons: computational study Prog. Neuro-Psychopharm. & Biol. Psychiat. 24 (2000) 483–504. [10] S. Tanaka, Post-cue activity of prefrontal cortical neurons controlled by local inhibition, Neurocomputing 32=33 (2000) 563–572. [11] S. Tanaka, Computational approaches to the architecture and operations of the prefrontal cortical circuit for working memory, Prog. Neuro-Psychopharm. & Biol. Psychiat. 25 (2001) 259–281 (Review). [12] S. Tanaka, Multi-directional representation of spatial working memory in a model prefrontal cortical circuit, Neurocomputing 44–46 (2002) 1001–1008, this issue. [13] S. Tanaka, S. Okada, Functional prefrontal cortical circuitry for visuospatial working memory formation: a computational model. Neurocomputing 26=27 (1999) 891–899. [14] S. Tanaka, A. Yoshida, Signal Row in a prefrontal cortical circuit model for working memory loading, Neurocomputing 38– 40 (2001) 957–964. [15] G.V. Williams, P.S. Goldman-Rakic, Modulation of memory !elds by dopamine Dl receptors in prefrontal cortex, Nature 376 (1995) 572–575. [16] P. Zheng, X.-X. Zhang, B.S. Bunney, W.-X. Shi, Opposite modulation of cortical N-methyl-D-aspartate receptor-mediated responses by low and high concentrations of dopamine, Neurosci. 91 (1999) 527–535. Koki Yamashita received B.E. from Sophia University, Tokyo, in 2001. He is a graduate student at the Program of Electrical and Electronics Engineering, Sophia University. He is currently studying computational neuroscience and computer science.

Shoji Tanaka received B.E., M.E., and Ph.D. degrees from Nagoya University, Japan, in 1980, 1982, and 1985, respectively. In 1985, he was a postdoctoral fellow at Japan Atomic Energy Research Institute, Tokai-mura, Japan. He joined the Department of Electrical and Electronics Engineering, Sophia University, Tokyo, in 1986. He is Professor of Electrical and Electronics Engineering at Sophia University. During 1998–1999, he was a Visiting Scientist at the Section of Neurobiology, Yale University School of Medicine, USA.