Our results confirm earlier reports that VS and anterior caudate respond preferentially to successful recognition (von Zerssen et al., 2001
; Han et al., 2010
). We extend this finding beyond prior literature by identifying a relationship of VS activation to response bias, and by demonstrating a deficit in these responses in patients with schizophrenia.
VS RESPONDS TO CORRECT RECOGNITION WITHOUT EXPLICIT FEEDBACK OR REWARD
Extensive literature documents activation of VS to external rewards. However, research examining VS responses reflecting intrinsic motivation is quite sparse. Prior studies have demonstrated that performance feedback activates VS (Rodriguez et al., 2006
; Tricomi and Fiez, 2008
; Daniel and Pollmann, 2010
). Here we tested the hypothesis that striatal reward signals would be present even in the absence of explicit feedback, reflecting monitoring and evaluation of responses by intrinsically motivated subjects. The VS response we observe to correct targets is consistent with this hypothesis as well as the findings of Han et al. (2010)
. In that study, the anterior caudate response to correct targets occurred in the absence of external rewards or feedback, and was further enhanced when a monetary incentive was provided for correct target recognition. While Han et al. (2010)
reported a peak of activation in anterior caudate dorsal to VS, the activated striatal regions overlapped with those observed in this study. Furthermore, Han et al. (2010)
found that activation in VS itself correlated most strongly with personality measures of reward sensitivity. While our study is not a direct replication of Han et al. (2010)
, the differences in methods make the similarities in our findings all the more striking.
VS RESPONSE IS CORRELATED WITH RESPONSE BIAS
By conceptualizing VS responses within the framework of reward prediction errors, we successfully predicted the novel finding that subjects with a more conservative response bias would activate VS more during correct retrieval. Importantly, this result was not driven by discrimination accuracy, which was accounted for in our model. Prediction errors reflect the difference between actual and expected outcome values. Expected outcome value depends on the subjective value of the outcome and its probability of occurring. In contrast, actual outcome value depends on subjective value and the probability that the outcome has actually been obtained. In typical reward tasks feedback is given, making the actual outcome certain. However, in a task like the current one with no feedback, the probability that the desired outcome has occurred (here, a HIT) corresponds to subjective confidence in one’s response. Therefore, VS prediction error signals should be enhanced by a higher subjective confidence in the correctness of a given response. A conservative response bias could therefore increase prediction errors in two ways. First, subjects with a conservative response bias will have lower expectations that any given stimulus will be a correctly identified target, thereby lowering the expected outcome value. Second, subjects with a conservative bias will therefore tend to report recognition only when they have high confidence they are correct (higher actual outcome value). Thus, subjects with a conservative response bias will have low expectations of correctly identifying a target on any given trial, but higher response confidence will generate a greater outcome value when a target is indeed correctly identified. Together, these two factors could produce greater positive prediction errors. Consistent with this account, conservative bias in the current study strongly correlated both with reduced target hit rate (fewer correct identifications) and with an increased proportion of target responses that were in fact correct. Thus, a subject with a very conservative response bias was unlikely to respond “old” to a previously displayed target (reducing hit rate), but had a high likelihood of being correct when she did respond “old.”
One reason that such an effect has not been previously described may stem from the difficulty of the task. Greater task difficulty lowers pre-response confidence, and also may increase the subjective value of correct responses, producing greater prediction errors responses in VS. The task used here was quite challenging, as encoding occurred incidentally during brief presentations of abstract visual fractals as oddball distractors, without any expectation of a memory test. Old vs. new recognition tasks used in prior fMRI studies are typically designed to be easier, in order to reduce the number of errors (which are typically unanalyzed), and to ensure all subjects perform at greater than chance levels. We are currently testing the prediction that greater task difficulty enhances VS response to correct responses in a follow-up study.
VS IS PART OF AN INTRINSIC REINFORCEMENT NETWORK
Our ROI analysis focused on the VS based on its central importance in reward and motivation. An exploratory whole-brain analysis revealed a network of other regions previously associated with reinforcement learning that were also strongly activated to successful retrieval, including midbrain, insula/orbital frontal cortex, and dorsal anterior cingulate cortex (Kirsch et al., 2003
; Sescousse et al., 2010
; D’Cruz et al., 2011
). Importantly, activation of these regions was not due to non-specific elements of the task, as these regions were also recruited in the contrast of HIT > MISS. The activation of this network, which is not typically associated with memory retrieval per se
, further supports our contention that VS responses seen here reflect reward signaling. These findings underline the importance of reward-related circuits in understanding performance and brain responses in cognitive tasks.
PATIENTS WITH SCHIZOPHRENIA HAVE DIMINISHED VS INTRINSIC REINFORCEMENT RESPONSE
The hypothesized deficit in VS activation to correct targets in the schizophrenia group was confirmed. Notably, the result from the VS ROI analysis was also the single statistically significant group difference in the exploratory whole-brain analysis. Multiple fMRI studies have examined recognition memory tasks in schizophrenia patients, identifying abnormalities in various cortical and subcortical regions (Weiss and Heckers, 2001
; Achim and Lepage, 2005
; Leavitt and Goldberg, 2009
; Ragland et al., 2009
). To our knowledge prior recognition memory studies have not identified VS deficits. This may be a consequence of using easier tasks (to reduce group performance differences) that do not activate VS robustly even in controls. In addition, in studies where group differences in VS activation are less robust than ours, a VS ROI approach might have been necessary but has generally not been utilized. However, VS hypofunction in schizophrenia is commonly observed in tasks involving reward (Juckel et al., 2006
; Wolf et al., 2008
; Waltz et al., 2009
). Two studies have reported striatal hypofunction in schizophrenia using different cognitive tasks without an explicit reward component. Koch et al. (2008)
reported reduced dorsal striatum activation in schizophrenia during successful working memory retrieval, whereas Vink et al. (2006)
reported diminished anterior caudate response in both schizophrenia patients and unaffected family members during a response inhibition task. Although these findings were not framed as deficits in intrinsic reinforcement, both these prior studies and the current results are consistent with such an interpretation. Taken together with the existing literature, the current study provides additional evidence of striatal dysfunction in schizophrenia, demonstrates its occurrence during a recognition memory task, and suggests that such dysfunction may be related to deficits in reward-related processing.
In addition to demonstrating reduced VS activation to correct recognition, schizophrenia patients also failed to demonstrate the increase in VS responses with more conservative bias that was found in controls. This occurred despite a similar average and range of response bias in the patient and control groups. This suggests a failure to generate prediction errors in response to changes in expectation and confidence, and is consistent with substantial evidence of blunted prediction error responses in schizophrenia (Corlett et al., 2007
; Murray et al., 2008
; Waltz et al., 2009
; Romaniuk et al., 2010
; Gradin et al., 2011
; Morris et al., 2011
Evidence of blunted VS reward responses in schizophrenia, particularly in those with prominent negative symptoms (Juckel et al., 2006
; Wolf et al., 2008
; Waltz et al., 2009
), leads us to speculate that the abnormal VS responses seen here reflect reductions in intrinsic motivation and self-generated reward signals. Deficits in intrinsic motivation have been found in schizophrenia and related to clinically relevant outcomes. In particular, Nakagami et al. (2008)
demonstrated that intrinsic motivation, neurocognitive performance, and psychosocial functioning are all inter-related in schizophrenia, and that intrinsic motivation mediated the relationship between neurocognition and psychosocial functioning. Similarly, Choi and Medalia (2009)
have shown that an intervention aimed at increasing intrinsic motivation in schizophrenia patients improved performance in a difficult cognitive task. Reward and motivation impairments may therefore contribute to cognitive abnormalities in schizophrenia. Our results extend this literature, and suggest VS dysfunction may be a key neural mechanism.
ALTERNATIVE INTERPRETATIONS AND LIMITATIONS
While our findings are consistent with a prediction error framework, certain alternative interpretations and limitations should be acknowledged. First, anterior caudate responses have been observed during successful response inhibition and response switching (Vink et al., 2005
; Li et al., 2008
; Cameron et al., 2009
). These responses have been interpreted as evidence that anterior caudate is important for overcoming prepotent responses. One can conceptualize “old” responses in the setting of a difficult memory task and a conservative response bias as requiring inhibition of the prepotent “new” response. This response inhibition theory would also predict stronger VS responses in those with greater conservative bias. However, it is unlikely that the response inhibition account fully explains VS activation to correct recognition responses, as we have observed stronger VS activation to correct than incorrect responses in the absence of feedback in a facial recognition memory paradigm (Wolf et al., 2011
) that utilizes simultaneous target and foil presentations and therefore does not involve response inhibition (unpublished VS ROI analysis).
Second, while the presence of apparent reward prediction signals in the absence of external rewards or explicit feedback is one of the most interesting aspects of the study, this design is also a limitation as it does not directly vary reward outcomes. Future studies can bolster such interpretations by including additional task conditions that directly manipulate rewards, feedback, and task difficulty. In addition, future research should also assess relevant subjective states and traits not measured in the current study, including pre-response confidence, post-response confidence, subjective value of correct and incorrect responses, trait reward sensitivity, and intrinsic motivation.
Third, the number of correct targets available for fMRI analysis was relatively small (~15 on average in each group), which may have decreased the reliability of within-subject activation estimates. However, this within-subject variance is accounted for in the imaging analysis, and while it may reduce the statistical significance of our results, it is unlikely to explain them. Regardless, future studies would benefit from the increased statistical power afforded by a greater number of trials, as well as a larger number of subjects.
Fourth, group differences in task performance present a potential confound. As in this study, schizophrenia patients’ performance in cognitive tasks is typically impaired. If VS prediction error responses are more robust in control subjects when task difficulty increases, there is a tension between achieving adequate patient performance and capturing the phenomenon of interest. However, groups did not differ in accuracy or reaction time for the target recognition condition that was the focus of this study. Furthermore, discrimination accuracy did not correlate with VS responses within either group and inclusion of Pr as a confound covariate across groups did not attenuate the group difference. Therefore, group differences in recognition performance are an unlikely explanation for observed VS differences.
Finally, our patient population was treated with antipsychotic medications, which are known to affect dopamine signaling in brain regions including VS, and which may affect reward processing, motivation, and cognition in complex ways. The observation of striatal hypofunction in unaffected family members as well as patients (Vink et al., 2006
), suggests a vulnerability phenotype rather than an effect of frank illness or medication. Furthermore, given that most schizophrenia patients require long-term treatment with antipsychotics, a potential role of medication effects on VS hypofunction does not negate its importance. Nonetheless, clarifying the role of medications will be important and will require further studies including in drug-naïve populations.