Striatal disorders dissociate mechanisms of enhanced and impaired response selection - Evidence from cognitive neurophysiology and computational modelling

Abstract

Paradoxically enhanced cognitive processes in neurological disorders provide vital clues to understanding neural function. However, what determines whether the neurological damage is impairing or enhancing is unclear. Here we use the performance of patients with two disorders of the striatum to dissociate mechanisms underlying cognitive enhancement and impairment resulting from damage to the same system. In a two-choice decision task, Huntington's disease patients were faster and less error prone than controls, yet a patient with the rare condition of benign hereditary chorea (BHC) was both slower and more error prone. EEG recordings confirmed significant differences in neural processing between the groups. Analysis of a computational model revealed that the common loss of connectivity between striatal neurons in BHC and Huntington's disease impairs response selection, but the increased sensitivity of NMDA receptors in Huntington's disease potentially enhances response selection. Crucially the model shows that there is a critical threshold for increased sensitivity: below that threshold, impaired response selection results. Our data and model thus predict that specific striatal malfunctions can contribute to either impaired or enhanced selection, and provide clues to solving the paradox of how Huntington's disease can lead to both impaired and enhanced cognitive processes.

Keywords: AMPA, a-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid; BHC, benign hereditary chorea; Basal ganglia; Benign hereditary chorea; Computational modelling; EEG; EEG, electroencephalography; ERP, event related potential; Executive control; FSIs, fast spiking interneurons; GABA, ?-aminobutyric acid; Huntington's disease; MMN, mismatch negativity; MMSE, Mini Mental Status Examination; MSN, medium spiny neuron; NMDA, N-methyl-d-aspartate; RON, reorientation of attention.

Fig. 2

Signal selection in healthy-state, Huntington's disease-like and BHC-like models of the striatum. The medium spiny neuron (MSN) populations receive inputs representing competing response signals, with the more salient response having a step increase (top); response selection is read-out from the population outputs after the step, by checking that both upper (?_H) and lower (?_L) thresholds are crossed. All example simulations show responses to the same input. (A) The healthy-state model. Line-widths in the model schematic (right) indicate relative synaptic weights. (B) A BHC-like model, with loss of medium spiny neuron intra-connectivity. (C) A Huntington's disease-like model, with loss of medium spiny neuron intra-connectivity due to cell death and enhanced cortical input weight due to increased NMDA receptor sensitivity.

Fig. 3

Selection by the striatum is differentially modulated by increased NMDA sensitivity and loss of striatal connectivity. In each panel, we show a schematic of the model circuit (right), and the analytically-computed minimum input step required for unambiguous signal selection (left) as a function of the inhibitory and input weights. Each heat map's colour intensity encodes the required minimum input step (firing rate, in Hz) for those weights, and the white line is the analytical boundary that separates the regions of smaller (above) and larger (below) step-sizes than the healthy-state model (bottom left-hand corner). Heat maps are plotted for specific choices of model parameter values to illustrate the general insights from the analytical solutions (Supplementary results), and so these results do not depend on specific choices of parameter values. (A) The baseline model of weakly-competing medium spiny neuron (MSN) populations. (B) The effect of adding inhibitory feedback to competing medium spiny neuron populations. (C) The effect of adding feed-forward inhibition from fast spiking interneurons (FSIs) to competing medium spiny neuron populations. The fast spiking interneuron population is a “broadcast” model that equally samples from the cortical inputs and equally outputs to the medium spiny neuron populations.

Fig. 4

Tradeoff between increased NMDA sensitivity and decreased striatal connectivity is robust to nonlinear neuronal dynamics. (A) The reduced spiking microcircuit model. Two spiking model medium spiny neurons (red, blue) represent the two populations of medium spiny neurons (MSNs). Each neuron model receives glutamatergic input (arrows) via AMPA and NMDA receptors from its afferent cortical populations, and GABAergic input (circles) from striatal fast spiking interneurons (FSIs) and from the other medium spiny neurons. Each box represents a source of spikes based on the number of spikes received from the indicated sources. The numbers of connections between each indicated population are derived from a full three-dimensional model of striatal micro-anatomy (Humphries et al., 2010)). Inset: an example of the output of the two medium spiny neuron models during a single selection test. Cortical input to MSN1 increased at 7 s (grey line), representing a more salient response signal. (B) Landscape of minimum input step required for unambiguous signal selection (in total spikes/s from cortical spike-trains). The value for each parameter pair (missing medium spiny neurons, NMDA conductance) is the mean of 20 numerical searches for the minimum input step. The white line gives the boundary between better (smaller step) and worse (larger step) signal selection performance than the healthy-state model (bottom left).

Fig. 5

Tradeoff between increased NMDA sensitivity and decreased striatal connectivity occurs for selective loss of D2 medium spiny neurons (MSNs). (A) The D1–D2 microcircuit model. Right: we split the two medium spiny neuron populations into separate sub-populations of medium spiny neurons predominantly expressing the D1 and D2 receptors for dopamine, and modelled the resulting connections within and between them. Both the D1 and D2 medium spiny neuron populations received the same input signal from their corresponding cortical population; the selection was read-out from the D1 medium spiny neuron population's output. Left: minimum input needed to achieve unambiguous signal selection following selective loss of the D2 medium spiny neurons only: all D2 medium spiny neuron output weights were changed whilst the D1 medium spiny neuron output weights were held constant. Heat map conventions are the same as in Fig. 3; the results plotted here are from numerical simulations. (B) The D1–D2 microcircuit model with the addition of feedback self-inhibition to each medium spiny neuron population: the heat map gives the minimum input needed to achieve unambiguous signal selection following selective loss of the D2 medium spiny neurons only. (C) The D1–D2 microcircuit model with the further addition of feed-forward fast spiking interneuron (FSI) input to each medium spiny neuron population: the heat map gives the minimum input needed to achieve unambiguous signal selection following selective loss of the D2 medium spiny neurons only.

Fig. 1

Behavioural and electrophysiological differences in the auditory discrimination task. (A) Error rates and reaction times on distractor (black) and standard trials (white) in controls, manifest Huntington's disease and the BHC case (±SEM). The data shows better selection performance (fewer errors, faster reaction times) in Huntington's disease, but reduced selection performance in BHC, compared to controls. (B) Differences (standard minus distractor) at electrode Fz for controls (red), manifest Huntington's diseases (black) and BHC (blue). The different parts of the difference waves are labelled MMN, P3a and RON. Paralleling the behavioural data, the MMN and RON were increased in Huntington's disease and decreased in BHC, relative to controls.