Culbreth Adam J, Westbrook Andrew, Xu Ziye, Barch Deanna M, Waltz James A
Department of Psychological and Brain Sciences, Washington University in Saint Louis.
University of Maryland School of Medicine, Department of Psychiatry and Maryland Psychiatric Research Center.
Biol Psychiatry Cogn Neurosci Neuroimaging. 2016 Sep;1(5):474-483. doi: 10.1016/j.bpsc.2016.07.007.
Midbrain dopaminergic neurons code a computational quantity, reward prediction error (RPE), which has been causally related to learning. Recently, this insight has been leveraged to link phenomenological and biological levels of understanding in psychiatric disorders, such as schizophrenia. However, results have been mixed, possibly due to small sample sizes. Here we present results from two studies with relatively large Ns to assess VS RPE in schizophrenia.
In the current study we analyzed data from two independent studies, involving a total of 87 chronic medicated schizophrenia patients and 61 controls. Subjects completed a probabilistic reinforcement-learning task in conjunction with fMRI scanning. We fit each participant's choice behavior to a Q-learning model and derived trial-wise RPEs. We then modeled BOLD signal data with parametric regressor functions using these values to determine whether patient and control groups differed in prediction-error-related BOLD signal modulations.
Both groups demonstrated robust VS RPE BOLD activations. Interestingly, these BOLD activation patterns did not differ between groups in either study. This was true when we included all participants in the analysis, as well as when we excluded participants whose data was not sufficiently fit by the models.
These data demonstrate the utility of computational methods in isolating/testing underlying mechanisms of interest in psychiatric disorders. Importantly, similar VS RPE signal encoding across groups suggests that this mechanism does not drive task deficits in these patients. Deficits may instead stem from aberrant prefrontal/parietal circuits associated with maintenance and selection of goal-relevant information.
中脑多巴胺能神经元编码一种计算量，即奖励预测误差（RPE），它与学习存在因果关系。最近，这一见解已被用于将精神疾病（如精神分裂症）中现象学和生物学层面的理解联系起来。然而，结果参差不齐，可能是由于样本量较小。在此，我们展示两项样本量相对较大的研究结果，以评估精神分裂症患者的腹侧纹状体奖励预测误差。
在本研究中，我们分析了来自两项独立研究的数据，共纳入87名长期接受药物治疗的精神分裂症患者和61名对照者。受试者在进行功能磁共振成像扫描的同时完成一项概率强化学习任务。我们将每个参与者的选择行为拟合到一个Q学习模型中，并得出逐次试验的奖励预测误差。然后，我们使用这些值，通过参数回归函数对血氧水平依赖（BOLD）信号数据进行建模，以确定患者组和对照组在与预测误差相关的BOLD信号调制方面是否存在差异。
两组均表现出强大的腹侧纹状体奖励预测误差BOLD激活。有趣的是，在两项研究中，两组之间的BOLD激活模式均无差异。当我们将所有参与者纳入分析时是如此，当我们排除那些数据未被模型充分拟合的参与者时亦是如此。
这些数据证明了计算方法在分离/测试精神疾病潜在相关机制方面的效用。重要的是，各组之间相似的腹侧纹状体奖励预测误差信号编码表明，这一机制并非导致这些患者任务缺陷的原因。相反，缺陷可能源于与目标相关信息的维持和选择相关的前额叶/顶叶回路异常。
