{"title": "Dopamine Modulation in a Basal Ganglio-Cortical Network of Working Memory", "book": "Advances in Neural Information Processing Systems", "page_first": 1271, "page_last": 1278, "abstract": "", "full_text": "Dopamine modulation in a basal ganglio-cortical\n\nnetwork implements saliency-based gating of\n\nworking memory\n\nAaron J. Gruber1,2, Peter Dayan3, Boris S. Gutkin3, and Sara A. Solla2,4\n\nBiomedical Engineering1, Physiology2, and Physics and Astronomy4,\n\nNorthwestern University, Chicago, IL, USA.\nGatsby Computational Neuroscience Unit3,\nUniversity College London, London, UK.\n\n{a-gruber1,solla }@northwestern.edu, {dayan,boris}@gatsby.ucl.ac.uk\n\nAbstract\n\nDopamine exerts two classes of effect on the sustained neural activity\nin prefrontal cortex that underlies working memory. Direct release in\nthe cortex increases the contrast of prefrontal neurons, enhancing the ro-\nbustness of storage. Release of dopamine in the striatum is associated\nwith salient stimuli and makes medium spiny neurons bistable; this mod-\nulation of the output of spiny neurons affects prefrontal cortex so as to\nindirectly gate access to working memory and additionally damp sensi-\ntivity to noise. Existing models have treated dopamine in one or other\nstructure, or have addressed basal ganglia gating of working memory ex-\nclusive of dopamine effects. In this paper we combine these mechanisms\nand explore their joint effect. We model a memory-guided saccade task\nto illustrate how dopamine\u2019s actions lead to working memory that is se-\nlective for salient input and has increased robustness to distraction.\n\n1 Introduction\n\nAmple evidence indicates that the maintenance of information in working memory (WM)\nis mediated by persistent neural activity in the prefrontal cortex (PFC) [9, 10]. Critical for\nsuch memories is to control how salient external information is gated into storage, and to\nlimit the effects of noise in the neural substrate of the memory itself. Experimental [15, 18]\nand theoretical [2, 13, 4, 17] studies implicate dopaminergic neuromodulation of PFC in\ninformation gating and noise control. In addition, there is credible speculation [7] that input\nto the PFC from the basal ganglia (BG) should also exert gating effects. Since the striatum\nis also a major target of dopamine innervation, the nature of the interaction between these\nvarious control structures and mechanisms in manipulating WM is important.\n\nA wealth of mathematical and computational models bear on these questions. A recent\ncellular-level model, which includes many known effects of dopamine (DA) on ionic con-\nductances, indicates that modulation of pyramidal neurons causes the pattern of network\nactivity at a \ufb01xed point attractor to become more robust both to noise and to input-driven\n\n\fswitching of attractor states [6]. This result is consistent with reported effects of DA in\nmore abstract, spiking-based models [2] of WM, and provides a cellular substrate for net-\nwork models that account for gating effects of DA in cognitive WM tasks [1]. Other net-\nwork models [7] of cognitive tasks have concentrated on the input from the BG, arguing\nthat it has a disinhibitory effect (as in models of motor output) that controls bistability\nin cortical neurons and thereby gates external input to WM. This approach emphasizes\nthe role of dopamine in providing a training signal to the BG, in contrast to the modu-\nlatory effects of DA discussed here, which are important for on-line neural processing.\nFinally, dopaminergic neuromodulation in the striatum has itself been recently captured in\na biophysically-grounded model [11], which describes how medium spiny neurons (MSNs)\nbecome bistable in elevated dopamine. As the output of a major subset of MSNs ultimately\nreaches PFC after further processing through other nuclei, this bistability can have poten-\ntially strong effects on WM.\n\nIn this paper, we combine these various in\ufb02uences on working memory activity in the PFC.\nWe model a memory-guided saccade task [8] in which subjects must \ufb01xate on a centrally\nlocated \ufb01xation spot while a visual target is \ufb02ashed at a peripheral location. After a delay\nperiod of up to a few seconds, subjects must saccade to the remembered target location.\nNumerous experimental studies of the task show that memory is maintained through striatal\nand sustained prefrontal neuronal activity; this persistent activity is consistent with attractor\ndynamics. Robustness to noise is of particular importance in the WM storage of continuous\nscalar quantities such as the angular location of a saccade target, since internal noise in the\nattractor network can easily lead to drift in the activity encoding the memory. In successive\nsections of this paper, we consider the effect of DA on resistance to attractor switching in\nthe isolated cortical network; the effect of MSN activity on gating and noise; and the effect\nof dopamine induced bistability in MSNs on WM activity associated with salient stimuli.\nWe demonstrate that DA exerts complementary direct and indirect effects, which result in\nsuperior performance in memory-guided tasks.\n\n2 Model description\n\nThe components of\nthe network model\nused to simulate the WM activity during a\nmemory-guided saccade task are shown in\nFig 1. The input module consists of a ring\nof 120 units that project both to the PFC and\nthe BG modules. Input units are assigned \ufb01r-\ning rates rT\nj to represent the sensory cortical\nresponse to visual targets. Bumps of activ-\nity centered at different locations along the\nring encode for the position of different tar-\ngets around the circle, as characterized by an\nangle in the [0, 2\u03c0) interval.\n\nThe BG module consists of 24 medium spiny\nneurons (MSNs). Connections from the in-\nput units consist of Gaussian receptive \ufb01elds\nthat assign to each MSN a preferred direc-\ntion; these preferred directions are monoton-\nically and uniformly distributed. The dy-\nnamics of individual MSNs follow from a\nbiophysically-grounded single compartment\nmodel [11]\n\nFigure 1: The network model consists of\nthree modules: cortical input, basal gan-\nglia (BG), and prefrontal cortex (PFC). In-\nsets show the response functions of spiny\n(BG) and pyramidal (PFC) neurons for\nboth low (dotted curves) and high (solid\ncurves) dopamine.\n\n\u2212C \u02d9V S = \u03b3 (IIRK + ILCa) + IORK + IL + IT ,\n\n(1)\n\nIPF CortexInputBGDApyramidalactivitymedium spinyup stateinputactivationinputEST\fwhich incorporates three crucial ionic currents: an inward rectifying K + current (IIRK),\nan outward rectifying K + current (IORK), and an L-type Ca2+ current (ILCa). The charac-\nterization of these currents is based on available biophysical data on MSNs. The factor \u03b3\nrepresents an increase in the magnitude of the IIRK and ILCa currents due to the activation\nof D1 dopamine receptors. This DA induced current enhancement renders the response\nfunction of MSNs bistable for \u03b3 & 1.2 (see Fig 1 for \u03b3 = 1.4). The synaptic input IT is an\nohmic term with conductance given by the weighted summed activity of the corresponding\nj , where W ST\nji\nis the strength of the connection from the i-th input neuron to the j-th spiny neuron. The\nj ). The\n\ufb01ring rate of MSNs is a logistic function of their membrane potential: rS\nMSNs provide excitatory inputs to the PFC; in the model, this monosynaptic projection\nrepresents the direct pathway through the globus pallidus/substantia nigra and thalamus.\n\ninput unit; input to the j-th MSN is thus given by IT j = P\n\nj = L(V S\n\ni W ST\n\nji rT\n\ni V S\n\nThe PFC module implements a line attractor capable of sustaining a bump of activity that\nencodes for the value of an angular variable in [0, 2\u03c0). \u2018Bump\u2019 networks like this have\nbeen used [3, 5] to model head direction and visual stimulus location characterized by a\nsingle angular variable. The module consists of 120 excitatory units; each unit is assigned\na preferred direction, uniformly covering the [0, 2\u03c0) interval. Lateral connections between\nexcitatory units are a Gaussian function of the angular difference between the correspond-\ning preferred directions. A single inhibitory unit provides uniform global inhibition; the\nactivity of the inhibitory unit is controlled by the total activity of the excitatory population.\nThis type of connectivity guarantees that a localized bump of activity, once established,\nwill persist beyond the disappearance of the external input that originated it (see Fig 2).\nOne of the purposes of this paper is to investigate whether this persistent activity bump is\nrobust to noise in the line attractor network.\n\nThe excitatory units follow the stochastic differential equation\n\nj +P\n\n\u03c4 E \u02d9V E\n\nj = \u2212V E\n\ni +P\n\ni W ES\n\nji rS\n\ni6=j W EE\n\nji rE\n\ni \u2212 rI + rT\n\nj + \u03c3e\u03b7.\n\n(2)\n\nji\n\nThe \ufb01rst sum in Eq 2 represents inputs from the BG; the connections W ES\nconsist of\nGaussian receptive \ufb01elds centered to align with the preferred direction of the corresponding\nexcitatory unit. The second sum represents inputs from other excitatory PFC units; note that\nself-connections are excluded. The following two terms represent input from the inhibitory\nPFC unit (rI) and information about the visual target provided by the input module (rT\nj ).\nCrucially, the last term provides a stochastic input that models \ufb02uctuations in the activities\nthat contribute to the total input to the excitatory units. The random variable \u03b7 is drawn from\na Gaussian distribution with zero mean and unit variance. The noise amplitude \u03c3e scales\nlike (dt)\u22121/2, where dt is the integration time step. The \ufb01ring rate of the PFC excitatory\nj ); as shown in Fig 1, the steepness of this response\nunits is a logistic function rE\nfunction is controlled by DA. The dynamics of the inhibitory unit follows from \u03c4 I \u02d9V I =\ni , where the sum represents the total activity of the excitatory population. The \ufb01ring\nrate rI of the inhibitory unit is a linear threshold function of V I. Dopaminergic modulation\nof the PFC network is implemented through an increase in the steepness of the response\nfunction of the excitatory cortical units. Gain control of this form has been adopted in\na previous, more abstract, network theory of WM [17], and is generally consistent with\nbiophysically-grounded models [6, 2].\n\nj = L(V E\n\nP\n\ni rE\n\nTo investigate the properties of the network model represented in Fig 1, the system of equa-\ntions summarized above is integrated numerically using a 5th order Runge-Kutta method\nwith variable time step that ensures an error tolerance below 5 \u00b5V/ms.\n3 Results\n\n3.1 Dopamine effects on the cortex: increased memory robustness\n\n\fFigure 2:\n(A) Activity pro\ufb01le of\nthe bump state in low DA (open\ndots) and high DA (full dots). (B)\nRobustness characteristics of bump\nactivity in low DA (dashed curve)\nand high DA (solid curve). For\nreference, the thin dotted line indi-\ncates the identity \u2206b\u03b8 = \u2206d\u03b8. The\nactivity pro\ufb01le shown as a func-\ntion of time in the inset (grey scale,\nwhite as most active) illustrates the\ndisplacement of the bump from its\ninitial location at \u03b80 to a \ufb01nal loca-\ntion at \u03b8b due to a distractor input\nat \u03b8d. This case corresponds to the\nasterisk on the curves in B.\n\nWe \ufb01rst investigate the properties of the cortical network isolated from the input and basal\nganglia components. The connectivity among cortical units is set so there are two stable\nstates of activity for the PFC network: either all excitatory units have very low activity\nlevel, or a subset of them participates in a localized bump of elevated activity (Fig 2A,\nopen dots). The bump can be translated to any position along the ring of cortical units, thus\nproviding a way to encode a continuous variable, such as the angular position of a stimulus\nwithin a circle. The encoded angle corresponds to the location of the bump peak, and it\ncan be read out by computing the population vector. The effect of DA on the PFC module,\nmodeled here as an increase in the gain of the response function of the excitatory units,\nresults in a narrower bump with a higher peak (Fig 2A, full dots).\n\nWe measure the robustness of the location of the bump state against perturbative distractor\ninputs by applying a brief distractor at an angular distance \u2206d\u03b8 from the current location\nof the bump and assessing the resulting angular displacement \u2206b\u03b8 in the location of the\nbump 40 ms after the offset of the distractor. The procedure is illustrated in the inset of\nFig 2B, which shows that a distractor current injection centered at a location \u03b8d causes a\ndrift in bump location from its initial position \u03b80 to a \ufb01nal position \u03b8b, closer to the angular\nlocation of the distractor. If \u03b8d is close to \u03b80, the distractor is capable of moving the bump\ncompletely to the injection location, and \u2206b\u03b8 is almost equal to \u2206d\u03b8. As shown in Fig 2B,\nthe plot of \u2206b\u03b8 versus \u2206d\u03b8 remains close to the identity line for small \u2206d\u03b8. However, as\n\u2206d\u03b8 increases the distractor becomes less and less effective, until the displacement \u2206b\u03b8 of\nthe bump decreases abruptly and becomes negligible.\n\nThe generic features of bump stability shown in Fig 2B apply to both low DA (dashed\ncurve) and high DA (solid curve) conditions. The difference between these two curves re-\nveals that the dopamine induced increase in the gain of PFC units decreases the sensitivity\nof the bump to distractors, resulting in a consistently smaller bump displacement. The ac-\ntual location of these two curves can be altered by varying the intensity and/or the duration\nof the distractor input, but their features and relative order remain invariant. This numer-\nical experiment demonstrates that DA increases the robustness of the encoded memory,\nconsistent with other PFC models of DA effects on WM [2, 6].\n\n3.2 Basal ganglia effects on the cortex: increased memory robustness and input gating\n\nNext, we investigate the effects of BG input (both tonic and phasic) on the stability of PFC\nbump activity in the absence of DA modulation. Tonic input from a single MSN, whose\npreferred direction coincides with the angular location of the bump, anchors the bump at\nthat location and increases memory robustness against both noise induced diffusion (Figs\n\nPFC Neuron angular labelactivity0100200300400p3/2ptime (ms)PFC Nuron label*0p/2pp/40p2p000.8DbqDdqp/42p/3p/23p/2qbq00ABqd\fFigure 3: Diffusion of the bump location due to noise in low DA (grey traces in A; dashed\ncurve in B) is greatly reduced by input from a single BG unit with the same preferred\nangular location (dark traces in A; solid curve in B). The robustness to distractor driven\ndrift is also increased by BG input (C).\n\n3A and 3B) and distractors (Fig 3C). Such localized tonic input to the PFC effectively\nbreaks the symmetry of the line attractor, yielding a single \ufb01xed point for the cortical active\nstate: a bump centered at the location of maximal BG input. This transition from a contin-\nuous line attractor to a \ufb01xed point attractor reduces the maximal deviation of the bump by\na distractor.\n\nActive MSNs provide control over the encoded memory not only by enhancing robustness,\nas shown above for the case of tonic input to the PFC, but also by providing phasic input\nthat can assist a relevant visual stimulus in switching the location of the PFC activity bump.\nWe show in Fig 4 (top plots) the location of the activity bump \u03b8b as a function of time\nin response to two stimuli at different locations \u03b8s. The nature of the PFC response to\nthe second stimulus depends dramatically on whether it elicits activity in the MSNs. The\ninitial stimulus activates a tight group of MSNs which encode for its angular position. It\nalso causes activation of a group of PFC neurons whose population vector encodes for the\nsame angular position. When the input disappears, the MSNs become inactive and the\ncortical layer relaxes to a characteristic bump state centered at the angular position of the\nstimulus. A second stimulus (distractor) that fails to activate BG units (Fig 4A) has only a\nminimal effect on the bump location. However, if the stimulus does activate the BG units\n(Fig 4B), then it causes a switch in bump location. In this case, the PFC memory is updated\nto encode for the location of the most recent stimulus. Thus a direct stimulus input to the\nPFC that by itself is not suf\ufb01cient to switch attractor states can trigger a switch, provided it\nactivates the BG, whose activity yields additional input to the PFC. Transient activation of\nMSNs thus effectively gates access to working memory.\n\n3.3 Dopamine effects on the basal ganglia: saliency-based gating\n\nAmple evidence indicates that DA, the release of which is associated with the presentation\nof conditioned stimuli [16], modulates the activity of MSNs. Our previous computational\nmodel of MSNs [11] studied the apparently paradoxical effects of DA modulation, mani-\nfested in both suppression and enhancement of MSN activity in a complex reward-based\nsaccade task [12]. We showed that DA can induce bistability in the response functions of\nMSNs, with important consequences. In high DA, the effective threshold for reaching the\nactive \u2019up\u2019 state is increased; the activity of units that do not exceed threshold is suppressed\ninto a quiescent \u2019down\u2019 state, while units that reach the up state exhibit a higher \ufb01ring rate\nwhich is extended in duration due to effects of hysteresis.\n\nWe now demonstrate that the dual enhancing/suppressing nature of DA modulation of\nMSNs activity signi\ufb01cantly affects the network\u2019s response to stimuli. We show in Fig 5\n(top plot) the location of the activity bump \u03b8b as a function of time in response to four\nstimuli at two different locations: \u03b8A, \u03b8B, \u03b8\u2217\nA is a\nconditioned stimulus that triggers DA release.\n\nA, \u03b8B. Crucially, in this sequence, only \u03b8\u2217\n\n024 0 p/6q 02400.06 0    < q2>-p/6p/60p/2pwith BGwithout BGtime (s)Ddqtime (s)ABCDbq\fFigure 4: Top plot shows the location \u03b8b of the encoded memory as determined from the\npopulation vector of the excitatory cortical units (thin black curve) and the location \u03b8s of\nstimuli as encoded by a Gaussian bump of activity in the input units (grey bars) as a function\nof time. The middle and bottom panels show the activity of the BG and the PFC modules,\nrespectively. Dopamine level remains low.\n\nThe \ufb01rst two stimuli activate appropriate MSNs, and are therefore gated into WM. The\npresentation of \u03b8\u2217\nA activates the same set of MSNs as \u03b8A, but the DA-modulated MSNs\nnow become bistable: high activity is enhanced while intermediate activity is suppressed.\nOnly the central MSN remains active with an enhanced amplitude; the two lateral MSNs\nthat were transiently activated by \u03b8A in low DA are now suppressed. The activity of the\ncentral MSN suf\ufb01ces to gate the location of the new stimulus into WM; the location of\nthe PFC activity bump switches accordingly. Interestingly, this switch from B to A occurs\nmore slowly than the preceding switch from A to B. This effect is also attributable to DA:\nits release affects the response function of excitatory PFC units, making them less likely\nto react to a subsequent stimulus and thus enhancing the stability of the bump at the \u03b8B\nangular position. Once the bump has switched to the angular location \u03b8\u2217\nA to encode for\nthe conditioned stimulus, the subsequent presentation of \u03b8B does not activate MSNs since\nthey are hysteretically locked in the inactive down state. The pattern of activity in the\nBG continues to encode for \u03b8A for as long as the DA level remains elevated, and the PFC\nactivity bump continues to encode for \u03b8\u2217\nA.\nIn sum, DA induced bistability of MSNs, associated with an expectation of reward, imparts\nsalience selectivity to the gating function of the BG. By locking the activation of MSNs\nassociated with salient input, the BG input prevents a switch in PFC bump activity and\npreserves the conditioned stimulus in WM. The robustness of the WM activity is enhanced\nby a combined effect of DA through both increasing the gain of PFC neurons and sustaining\nMSN input during the delay period (see Fig 5, bottom plot).\n\n4 Discussion\n\nWe have built a working memory model which links dopaminergic neuromodulation in\nthe prefrontal cortex, bistability-inducing dopaminergic neuromodulation of striatal spiny\n\n00.511.5200.511.52time (s)time (s)02p 0MSN label0pDA (g)2pp2ppPFC labelABqs, qb\fFigure 5: Top plot shows the location \u03b8b of the encoded memory as determined from the\npopulation vector of the excitatory cortical units (thin black curve) and the location \u03b8s\nof stimuli as encoded by a Gaussian bump of activity in the input units (grey bars) as a\nfunction of time. The second and third panels bottom plots show the activity of the BG and\nthe PFC modules, respectively. Dopamine level increases in response to the conditioned\nstimulus. The bottom plot displays increased robustness of WM for conditioned (solid\ncurve) as compared to unconditioned (dashed curve) stimuli.\n\nneurons, and the effects of basal ganglia output on cortical persistence. The resulting in-\nteractions provide a sophisticated control mechanism over the read-in to working memory\nand the elimination of noise. We demonstrated the quality of the system in a model of a\nstandard memory-guided saccade task.\n\nThere are two central issues for models of working memory: robustness to external noise,\nsuch as explicit lures presented during the memory delay period, and robustness to internal\nnoise, coming from unwarranted corruption of the neural substrate of persistent activity.\nOur model, along with various others, addresses these issues at a cortical level via two basic\nmechanisms: DA modulation, which changes the excitability of neurons in a particular way\n(units that are inactive are less excitable by input, while units that are active can become\nmore active), and targeted input from the BG. However, models differ as to the nature and\nprovenance of the BG input, and also its effects on the PFC. Ours is the \ufb01rst to consider the\ncombined, complementary, effects of DA in the PFC and the BG.\n\nThe requirements for a gating signal are that it be activated at the same time as the stimuli\nthat are to be stored, and that it is a (possibly exclusive) means by which a WM state is\nestablished. Following the experimental evidence that perturbing DA leads to disruption\nof WM [18], a set of theories suggested that a phasic DA signal (as associated, for in-\nstance, with reward predicting conditioned stimuli [16]) acts as the gate in the cortex [4].\nIn various models [17, 2, 6], and also in ours, phasic DA is able to act as a gate through\nits contrast-enhancing effect on cortical activity. However, as discussed at length in Frank\n\n02p 0MSN label0.511.522.530 0   p/2  p 0  p/4 pDA (g)2pp2pptime (s)qs, qbDbqDdqPFC label\fet al [7] (whose model does not incorporate the effect at all), this is unlikely to be the sole\ngating mechanism, since various stimuli that would not lead to the release of phasic DA\nstill require storage in WM. In our model, even in low DA, the BG gates information by\ncontrolling the switching of the attractor state in response to inputs. Frank et al [7] point out\nthe various advantages of this type of gating, largely associated with the opportunities for\nprecise temporal and spatial gating speci\ufb01city, based on information about the task context.\n\nOur BG gating mechanism simply involves additional targeted excitatory input to the cor-\ntex from the (currently over-simpli\ufb01ed) output of striatal spiny neurons, coupled with a\ndetailed account [11] of DA induced bistability in MSNs. This allows us to couple gating\nto motivationally salient stimuli that induce the release of DA. Since DA controls plasticity\nin cortico-striatal synapses [14], there is an available mechanism for learning the appropri-\nate gating of salient stimuli, as well as motivationally neutral contextual stimuli that do not\ntrigger DA release but are important to store.\n\nRobustness against noise that is internal to the WM is of particular importance for line or\nsurface attractor memories, since they have one or more global directions of null stability\nand therefore exhibit propensity to diffuse. Rather than rely on bistability in cortical neu-\nrons [3], our model relies on input from the striatum to reduce drift. This mechanism is\navailable in both high and low DA conditions. This additional input turns the line attractor\ninto a point attractor at the given location, and thereby adds stability while it persists. The\nDA induced bistability of MSNs, for which there is now experimental evidence, enhances\nthis stabilization effect.\n\nWe have focused on the mechanisms by which DA and the BG can in\ufb02uence WM. An\nimportant direction for future work is to relate this material to our growing understanding\nof the provenance of the DA signal in terms of reward prediction errors and motivationally\nsalient cues.\n\nReferences\n\n[1] Braver TS, Cohen JD (1999) Prog. Brain Res. 121:327-349.\n[2] Brunel N, Wang XJ (2001) J. Comp. Neurosci. 11:63-85.\n[3] Camperi M, Wang XJ (1998) J. Comp. Neurosci. 5:383-405.\n[4] Cohen JD, Braver TS, Brown JW (2002) Curr. Opin. Neurobiol. 12:223-229.\n[5] Compte A, Brunel N, Goldman-Rakic P, Wang XJ (2000) Cereb. Cortex 10:910-923.\n[6] Durstewitz D, Seamans J, Sejnowski T (2000) J. Neurophys. 83:1733-1750.\n[7] Frank M, Loughry B, O\u2019Reilly RC (2001) Cog., Affective, & Behav. Neurosci. 1(2):137-160.\n[8] Funahashi S, Bruce CJ, Goldman-Rakic PS (1989) J. Neurophys. 255:556-559.\n[9] Fuster J (1995) Memory in the Cerebral Cortex MIT Press.\n[10] Goldman-Rakic PS (1995) Neuron 14:477-85.\n[11] Gruber AJ, Solla SA, Houk JC (2003). NIPS 15.\n[12] Kawagoe R, Takikawa Y, Hikosaka O (1998) Nat. Neurosci. 1:411-416.\n[13] O\u2019Reilly RC, Noelle DC, Braver TS, Cohen JD (2002) Cerebral Cortex 12:246-257.\n[14] Reynolds JN, Wickens JR (2000) Neurosci. 99:199-203.\n[15] Sawaguchi T, Goldman-Rakic PS (1991) Science 251:947-950.\n[16] Schultz W, Apicella P, Ljungberg T (1993) J. Neurosci. 13:900-913.\n[17] Servan-Schreiber D, Printz H, Cohen J (1990) Science 249:892-895.\n[18] Williams GV, Goldman-Rakic PS (1995) Nature 376:572-575.\n\n\f", "award": [], "sourceid": 2356, "authors": [{"given_name": "Aaron", "family_name": "Gruber", "institution": null}, {"given_name": "Peter", "family_name": "Dayan", "institution": null}, {"given_name": "Boris", "family_name": "Gutkin", "institution": null}, {"given_name": "Sara", "family_name": "Solla", "institution": null}]}