Effects of reward on spatial working memory in schizophrenia

Abstract

Reward processing and cognition are disrupted in schizophrenia (SCZ), yet how these processes interface is unknown. In SCZ, deficits in reward representation may affect motivated, goal-directed behaviors. To test this, we examined the effects of monetary reward on spatial working memory (WM) performance in patients with SCZ. To capture complimentary effects, we tested biophysically grounded computational models of neuropharmacologic manipulations onto a canonical fronto-parietal association cortical microcircuit capable of WM computations. Patients with SCZ (n = 33) and healthy control subjects (HCS; n = 32) performed a spatial WM task with 2 reward manipulations: reward cues presented prior to each trial, or contextually prior to a block of trials. WM performance was compared with cortical circuit models of WM subjected to feed-forward glutamatergic excitation, feed-forward GABAergic inhibition, or recurrent modulation strengthening local connections. Results demonstrated that both groups improved WM performance to reward cues presented prior to each trial (HCS d = -0.62; SCZ d = -1.0), with percent improvement correlating with baseline WM performance (r = .472, p < .001). However, rewards presented contextually before a block of trials did not improve WM performance in patients with SCZ (d = 0.01). Modeling simulations achieved improved WM precision through strengthened local connections via neuromodulation, or feed-forward inhibition. Taken together, this work demonstrates that patients with SCZ can improve WM performance to short-term, but not longer-term rewards-thus, motivated behaviors may be limited by strength of reward representation. A potential mechanism for transiently improved WM performance may be strengthening of local fronto-parietal microcircuit connections via neuromodulation or feed-forward inhibitory drive. (PsycINFO Database Record (c) 2018 APA, all rights reserved).

Figure 1.. Spatial WM-Reward Task Design.

Each trial required subjects to remember the location of a circle after it disappeared. The task began with a period of central fixation (Fixation), followed by target presentation in one of 20 pseudo-randomized positions arranged in a 415-pixel radius ring from the center (Target). Target position was randomized throughout the task, and unknown to subjects. Once the target disappeared, the subject maintained the memory of the initial target location while continuing central fixation (Delay). A gray circle with positioning attached to a joystick (Probe) appeared and subjects moved the gray circle to where they remembered the target to be. The target was green in the Trial Reward types, and red elsewhere, with descriptions above. Rewarded trials were designed similarly, except for a green colored target to indicate the possibility for reward, and a feedback phase to indicate how well the subject performed. For the block of contextually rewarded trials, the cue was colored red, and subjects were instructed at the start of the block that there was the possibility to win money on each trial and they would not be reminded of this once the block started.

Figure 2.. Effect of Reward on WM Performance.

A. Overall, patients with SCZ demonstrated significantly worse baseline motor and WM performance (as measured by Euclidean distance between target presentation and probe placement), compared to HCS (*** denotes p<0.001). There was a significant WM x Group interaction, p<0.005. Both groups demonstrated relative improvement in WM performance under trial reward conditions (adjusted means shown). Error bars represent SEM (adjusted). B. Scatterplot of raw responses following rotation of all target presentations to (415, 0) under motor, WM (neutral) and trial reward conditions. Compared to HCS, patients with SCZ demonstrated less accuracy across all three conditions. Ellipses show 95% confidence intervals.

Figure 3.. Effects of Contextually Presented Rewards on WM Performance.

A. Adjusted mean spatial WM performance under neutral, contextually rewarded, and trial rewarded conditions for each group. There was a significant interaction of Context x Group (p<0.05). Examination of this interaction demonstrated differential performance between contextually presented and trial rewards in patients with SCZ (p<0.001, d=−1.01), with less differentiation for HCS (p>0.05, d=−0.38). Patients with SCZ demonstrated no performance improvement in response to contextually presented rewards (p>0.1, d=0.01), while HCS showed modest improvement (p>0.05, d = −0.23). Each group demonstrated improved WM performance when rewards were presented in a trial-by-trial manner compared to neutral trials (*** denotes p<0.001; ** denotes p<0.005). Error bars are adjusted SEM. B. Plot demonstrating the adjusted change in WM performance between groups. [Neut. — Context] refers to the change in WM performance between the neutral condition and the context reward condition; [Neut. — Trial] refers to the change in WM performance between the neutral condition and the trial reward condition; [Context – Trial] refers to the change in WM performance between the context reward and trial reward condition. Error bars are adjusted SEM. C. Scatter plot showing raw distribution of responses following rotation of all target presentations to (415, 0). Ellipses represent 95% CI.

Figure 4.. Partial Correlation of WM Performance and Percent Improvement in Trial Reward When Controlling for Motor Performance.

Worse baseline WM performance in neutral trials was significantly correlated with greater percent improvement in trial reward performance when controlling for motor performance (r = 0.472, p<0.001). This suggests that baseline cognitive performance did not preclude improvement in response to short-term rewards. Shaded areas indicate 95% CI.

Figure 5.. Modulation of WM Precision in a Fronto-Parietal Cortical Circuit Model.

(A) The model architecture consisted of recurrently connected excitatory pyramidal cells (E) and inhibitory interneurons (I). Pyramidal cells are labeled by the angular location they were tuned to (0–360°). E-to-E connections were structured and stronger for neurons with similar preferred angles, while all recurrent connections involving I-cells were unstructured and mediated feedback inhibition. Cortical disinhibition was implemented through NMDAR hypo-function on interneurons (decreased NMDA receptor conductance on interneurons, G_EI, by 3.25%), weakening recruitment of feedback inhibition. (B) Three candidate mechanisms for modulation of the WM circuit by reward-related signals: (i) neuromodulatory positive scaling of recurrent NMDA and GABA receptor conductances (default value of 6%) (purple), (ii) increased feedforward glutamatergic background drive to I-cells (default value of 2%) (gray), and (iii) increased background drive to E-cells (default value of 2%) (red). (C) Variance of the encoded angle grew with time during the delay period, indicating decreased WM precision. Disinhibition (orange) decreased precision (i.e., higher variance) compared to the control condition (blue). The three modulations altered WM precision (dashed for no modulation, solid for modulation): precision improved (lower variance, in line with empirical reward effects) with either recurrent scaling (top) or feedforward drive to I-cells (middle), whereas precision worsened (higher variance) with feedforward drive to E-cells (bottom). (D) WM precision changed parametrically over a range of modulation strengths. The variance plotted was from the end of the 10-s delay. Note that recurrent scaling was the only manipulation that was stable under strong modulations. For an increase of feedforward drive to I-cells beyond a few percent, dominance of inhibition prevented WM-related persistent activity, for both control and disinhibition networks. The hashed areas of either color denote corresponding regions in which the WM state was not stable. Circles mark the default modulation strengths used in C. (E) The variance at the end of the 10-second delay period is plotted for the model with no modulation at the far left (green; value indicated by dashed line). Improved WM precision (decreased variance) was seen at the end of the delay period under manipulations of recurrent scaling and feedforward to I-cells, though not under feedforward to E-cells. (F) Mean firing rate of E-cells during the WM state (same notations as (E)), and (G) example firing-rate profiles of the WM state with the modulations. Although recurrent scaling and feedforward drive to I-cells both improved WM precision at the behavioral level, they made dissociable predictions at the neural level: recurrent scaling increased firing-rate activity, whereas feedforward drive to I-cells decreased activity. For panels (E-G), the top and bottom rows correspond to the control and disinhibition regimes, respectively.