Striatal intrinsic reinforcement signals during recognition memory: relationship to response bias and dysregulation in schizophrenia

Abstract

Ventral striatum (VS) is a critical brain region for reinforcement learning and motivation, and VS hypofunction is implicated in psychiatric disorders including schizophrenia. Providing rewards or performance feedback has been shown to activate VS. Intrinsically motivated subjects performing challenging cognitive tasks are likely to engage reinforcement circuitry even in the absence of external feedback or incentives. However, such intrinsic reinforcement responses have received little attention, have not been examined in relation to behavioral performance, and have not been evaluated for impairment in neuropsychiatric disorders such as schizophrenia. Here we used fMRI to examine a challenging "old" vs. "new" visual recognition task in healthy subjects and patients with schizophrenia. Targets were unique fractal stimuli previously presented as salient distractors in a visual oddball task, producing incidental memory encoding. Based on the prediction error theory of reinforcement learning, we hypothesized that correct target recognition would activate VS in controls, and that this activation would be greater in subjects with lower expectation of responding correctly as indexed by a more conservative response bias. We also predicted these effects would be reduced in patients with schizophrenia. Consistent with these predictions, controls activated VS and other reinforcement processing regions during correct recognition, with greater VS activation in those with a more conservative response bias. Patients did not show either effect, with significant group differences suggesting hyporesponsivity in patients to internally generated feedback. These findings highlight the importance of accounting for intrinsic motivation and reward when studying cognitive tasks, and add to growing evidence of reward circuit dysfunction in schizophrenia that may impact cognition and function.

Keywords: intrinsic motivation; memory; reward; schizophrenia; ventral striatum.

FIGURE 1

Visual recognition task. In each of 60 trials, a unique fractal image was presented in a forced choice recognition memory paradigm. Half of the fractals had been previously displayed once each as salient distractors during a visual oddball task performed ~10 min earlier, yielding 30 targets and 30 foils. Stimuli were presented for 3 s in a fast event-related design, with a variable 0-18 s interstimulus interval during which a homogeneous visual noise gray background was displayed.

FIGURE 2

Ventral striatum HIT > baseline ROI analysis. (A) Results in healthy controls. (B) Results from control > patient contrast. Images thresholded at Z > 2.3 and cluster corrected for multiple comparisons with spatial extent p < 0.01. Note: A similar cluster to that shown in (B) was the only significant result of the exploratory whole-brain controls > patients analysis. Left hemisphere shown on left side of images

FIGURE 3

HIT > baseline whole-brain exploratory voxelwise analysis. Coronal (A) and transverse (B) images showing activation of the reward and motivation network regions including anterior insula, dACC, and striatum. Images thresholded at Z > 3.1 and cluster corrected for multiple comparisons with spatial extent p < 0.01.

FIGURE 4

Relationship between ventral striatum activation and response bias. (A) Cluster in VS ROI showing significant correlation between response bias and BOLD activation to HIT > baseline. An analogous within-group analysis in patients revealed no significant results. (B) Clusters showing significant differential correlation (controls > patients) to response bias. (C) Descriptive scatterplot showing mean BOLD activation in the cluster from (A) and response bias for controls (blue) and patients (red).The Pearson’s r value for the data shown here is -0.42 for controls, and 0.06 for patients; however, correlation values for data extracted from significant clusters may be biased and should not be used for statistical inference, which has already been conducted in identifying the cluster (Vul et al., 2009). Images thresholded at Z > 2.3 and cluster corrected for multiple comparisons with spatial extent p < 0.01.