If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Queensland Brain Institute, The University of Queensland, Brisbane, Queensland, AustraliaQIMR Berghofer Medical Research Institute, Brisbane, Queensland, Australia
Institute for Molecular Bioscience, The University of Queensland, Brisbane, Queensland, AustraliaDepartment of Psychiatry, University of Cambridge, Cambridge, United KingdomCambridgeshire and Peterborough NHS Foundation Trust, Cambridge, United Kingdom
Department of Psychiatry, University of Cambridge, Cambridge, United KingdomDepartment of Diagnostic and Interventional Neuroradiology, School of Medicine, Technical University of Munich, Munich, Germany
The processing of salient and rewarding stimuli is integral to engaging our attention, stimulating anticipation for future events, and driving goal-directed behaviors. Widespread impairments in these processes are observed in psychosis, which may be associated with worse functional outcomes or mechanistically linked to the development of symptoms. Here, we summarize the current knowledge of behavioral and functional neuroimaging in salience, prediction error, and reward. Although each is a specific process, they are situated in multiple feedback and feedforward systems integral to decision making and cognition more generally. We argue that the origin of salience and reward processing dysfunctions may be centered in the subcortex during the earliest stages of psychosis, with cortical abnormalities being initially more spared but becoming more prominent in established psychotic illness/schizophrenia. The neural circuits underpinning salience and reward processing may provide targets for delaying or preventing progressive behavioral and neurobiological decline.
Psychoses of the schizophrenia spectrum are characterized by positive symptoms, such as hallucinations and delusions, which are dominant during the psychotic stages of the disease; negative symptoms, such as lack of motivation and emotional blunting; and cognitive symptoms, such as memory dysfunctions (
). Although the underlying mechanisms leading to this complex set of symptoms are not fully understood, one of the most robust findings in psychosis and schizophrenia and even in the prodromal stages of the disease [e.g., (
). Unfortunately, however, approximately 30% of patients with schizophrenia classify as treatment resistant to dopamine D2 receptor antagonists and have worse long-term functional disability, with more severe positive, negative, and cognitive symptoms [e.g., (
). Given the strong associations between dopamine, salience/reward, and negative symptoms, this review is focused largely on subcortical dopamine systems and their associated corticostriatal networks.
In this article, we examine evidence for dysfunction of the neural circuitry subserving reward and salience processing and their links to dopaminergic dysregulation in the psychosis spectrum (Table 1). These dysfunctions are potentially highly relevant for understanding the core aspects of schizophrenia and related symptoms, including positive symptoms and negative symptoms. A brief introduction to the psychological processes included in salience, prediction error, and reward processing is provided in Figure 1. We endeavored to examine the extent to which neural dysfunctions have been linked to clinical symptoms, cognitive impairments, or functional disability because not all neural abnormalities may have deleterious consequences. Much of the work in this field is cross-sectional; however, we have endeavored to discuss jointly all stages of illness to highlight the relevant similarities and differences as much as possible.
Table 1Psychosis Spectrum
Disease Stage
Description
At Risk
By at risk, we mainly refer to people who are at increased risk of psychotic illness due to being help-seeking patients presenting with mild (subthreshold) clinical symptoms, especially subthreshold positive psychotic symptoms, such as suspicions or hallucinations without delusional interpretations. Such individuals are sometimes termed ultra-high-risk, clinical high-risk, at-risk mental state, or prodromal psychosis, with several (slightly differing) operational criteria available to categorize people in such states (
). We note that although such groups are especially at (relatively) high risk of psychosis, they also are at risk for other adverse psychiatric outcomes (
By early psychosis, we refer to the early stages after the onset of established psychotic illness, such as first-episode psychosis and first-episode schizophrenia. While some studies only include patients who meet the diagnostic criteria for schizophrenia, a number of research studies include a broader mixture of patients with first-episode psychosis. These are people presenting with psychotic illness for the first time, many with nonaffective schizophrenia spectrum psychosis, and others with affective psychosis such as bipolar disorder or depressive psychosis.
Chronic Psychosis/Schizophrenia
In general, we use the term chronic psychosis/schizophrenia for referring to patients who have been unwell beyond the early stages of illness/first 5 years of psychotic illness. Where studies have specified a minimum duration of illness, we use the term chronic schizophrenia, but we note that some studies include a mixture of patients with schizophrenia at different stages of illness.
Figure 1Conceptual interactions between external events (cyan), salience and prediction error (red), and reward processing (purple) in decision making. An external event (e.g., action outcome) will be perceived and attended to when it overcomes a certain level of salience (referred to as a salience threshold). The input is then integrated into associative learning networks to inform decision-making processes. Computation of potential rewards available (valuation) and the effort required for each (effort-cost trade-off) are then used to identify the optimal choice and whether or not it is acted upon. In this example, the outcome is rewarding (external feedback as action outcome), which is then compared with our prior expectations. Our prior expectation is associated with how accurate our valuation of the outcome is. In cases where the outcome matches the expectation (green tick), the associations are reinforced. In cases where there is a mismatch between the outcome and our expectation, a prediction error signal is generated. The prediction error is used to update our understanding of input-output relationships. Prediction error information (magnitude, precision, and so on) is then used to update our beliefs and goals associated with the initial stimuli (or action required). This updates our salience threshold so that when we encounter this same event in the future (n + 1), it is more likely to be considered salient. Reward processing (purple) affects multiple stages in this process to subsequently increase attention (reward anticipation), drive associative learning (reward learning), and govern our motivation to work toward a future goal. Here, we focus on a rewarding outcome; however, perceptual and attentional processes of sensory stimuli work similarly, causing belief updating and salience threshold updating via prediction errors.
Salience is a property that characterizes the importance of a stimulus and ultimately attracts attention to drive cognition and behavior. Salience is a multifaceted concept (
) including different dimensions, such as reward or novelty (Box 1). Abnormal salience processing following dysregulation of the dopaminergic system has been linked to the formation and maintenance of positive and negative symptoms (
)] create neurobiological noise, which is misinterpreted as meaningfulness and may lead to the attribution of salience to the otherwise unimportant, ordinary experiences that incidentally co-occur with this experience. The interpretation of these falsely judged–important stimuli may lead to the formation of hallucinations and delusions.
The world around us is highly complex and produces constant noisy and ambiguous sensory input to our brain. The biggest challenge for our brain is to rapidly and efficiently identify important stimuli and to process them effectively. One efficient way that the brain applies is to evaluate the saliency of incoming sensations and prioritize them accordingly considering the context. Imagine, for example, the change of the traffic lights from green to red; the change in the physical qualities of the visual input is highly informative and important to adapt our behavior and decisions. This example shows that there is a strong interaction between stimulus-driven processing and goals or belief of the individual to determine the saliency of incoming information.
We can differentiate between different forms of saliency (
Altered activation and connectivity in a hippocampal-basal ganglia-midbrain circuit during salience processing in subjects at ultra high risk for psychosis.
), the first distinction being between incentive or motivational salience and nonmotivational salience. Incentive salience describes the desire to obtain a reward by increasing attention and motivational drive (see the sections on Prediction Error Signaling in Psychosis and Reward Processing in Psychosis for more information). Tasks used to investigate motivational salience use mainly monetary rewards. Nonmotivational salience, which is usually studied in visual/auditory oddball paradigms or when investigating the processing of irrelevant stimuli in monetary reward paradigms, can be distinguished in novelty salience and surprisal (
), which are essential drivers of intrinsic motivation (not reward related) and attention. Surprise is characterized by a change in condition through the comparison of the expected to the actually perceived (e.g., the physical change of a stimulus or emotional change) and is strongly linked to an unsigned (i.e., without valence indication, worse or better than expected) prediction error (
). For novelty, however, the concept is less concrete. Generally, novelty refers to a sensory input never encountered before (complete novelty) or not encountered for some time (short-term/long-term novelty). See Barto et al. (
When experimentally investigating salience processing, two different aspects might be explored: the processing of salient, such as emotional, novel, or motivational stimuli, or the processing of irrelevant, uninformative, neutral stimuli. In schizophrenia, deficits in the first aspect conceptually relate to the development and preservation of negative symptoms (
); the latter suggests the overweighting of irrelevant stimuli, which is linked to the emergence of positive symptoms. The current literature provides evidence for aberrant salience processing across both aspects in early psychosis and chronic schizophrenia; however, links to symptomatology are less concise.
At the same time, relevant stimuli, such as a reward prediction error or an emotional stimulus, fail to be processed appropriately, leading to a blunted response, potentially explaining negative symptoms (
). Reward prediction errors are intrinsically salient (see Prediction Error Signaling in Psychosis and Reward Processing in Psychosis). The firing of dopamine neurons, however, is not exclusive to reward prediction error but has been reported in response to nonrewarding unexpected events, such as aversive or alerting (
Precision weighting of cortical unsigned prediction error signals benefits learning, is mediated by dopamine, and is impaired in psychosis [published correction appears in Mol Psychiatry 2021; 26:5334].
Aberrant Salience as Altered Processing of Irrelevant Information
Several studies investigated the processing of neutral or uninformative stimuli using different methods and exploring different stages of the disease and the link to symptoms; while theoretically clear, the experimental results show inconsistencies. Roiser et al. (
) adapted a monetary reinforcement learning task (salience attribution task) to investigate the salience assigned to irrelevant stimuli. Their study revealed that schizophrenia patients with delusions showed higher levels of aberrant salience to irrelevant stimuli than patients without delusions (
); however, aberrant salience was correlated with negative symptoms. Applying the same task to at-risk individuals, aberrant salience attribution to irrelevant stimuli was associated with the severity of positive symptoms (
Neural and behavioral correlates of aberrant salience in individuals at risk for psychosis [published correction appears in Schizphr Bull 2016; 42:1303].
). Similarly, using novel computational approaches to investigate the differences between patients with schizophrenia and control subjects in the implicit salience task, Katthagen et al. (
) found that patients had a stronger bias toward irrelevant information, which was associated with stronger negative symptomatology and not with positive symptoms as conceptually expected. Partially, these inconsistencies might be explained by different medication or treatment statuses. Abboud et al. (
), for example, showed that patients with treatment-resistant schizophrenia did not show heightened aberrant salience. This result may be explained by the nonelevated levels of dopamine synthesis capacity (
). A further argument explaining the inconsistencies might be the investigation of task-irrelevant information in a rewarding setting, which might not allow a clear differentiation between the underlying processes.
In a reward-independent learning setting, Ceaser and Barch (
) found that during a cognitive control task, patients with schizophrenia were more likely to inappropriately encode irrelevant distracter stimuli, showing increased prefrontal and striatal activity. In emotion recognition studies, patients with chronic schizophrenia consistently showed increased brain signaling to neutral emotional stimuli, the irrelevant stimuli in the tasks, with effects being especially strong for face stimuli (
). Similar results were also reported for emotional and neutral word processing; patients with schizophrenia and their unaffected first-degree relatives demonstrated increased attention to neutral words compared with control subjects (
Changes in EEG spectral power on perception of neutral and emotional words in patients with schizophrenia, their relatives, and healthy subjects from the general population.
). These studies provide evidence for altered processing of neutral events in subcortical regions associated with dopaminergic dysregulation. When studying the processing of neutral stimuli in a reward-independent setting, there is more consistent evidence of a link between neural dysregulation and performance and symptom strength; increased striatal activity during incorrect distracter trials correlated positively with aberrant salience symptoms (
) showed that people with schizophrenia with severe positive symptoms failed to discriminate between predictive and nonpredictive cues compared with healthy adults. Furthermore, overweighting nonpredictive cues was correlated with more severe positive symptom scores in schizophrenia (
) exploring neutral stimuli in a reward learning setting showed that individuals with psychotic-like experiences overattribute salience to neutral stimuli and underattribute salience to rewards, indicating that abnormal salience attribution is a trait-like feature. Together, these studies show that neutral or irrelevant stimuli are consistently overweighted in patients at different disease stages, with the exception of patients with treatment-resistant schizophrenia, showing a clear indication of aberrant salience, although associations with symptoms are inconsistent.
Aberrant Salience as Altered Processing of Relevant Information
Aberrant brain processing of informative and relevant but nonmotivational salient events in psychosis has been reported in several studies. In a recent study, Knolle et al. (
) to investigate novelty, negative emotional salience, and targetness, which required a button press, in patients with antipsychotic naïve first-episode psychosis. The patients exhibited reduced substantia nigra, ventral tegmental area, and striatal and cingulate signaling to novelty; reduced substantia nigra, ventral tegmental area, amygdala, and striatal and cingulate signaling to negative emotional salience; and reduced substantia nigra, ventral tegmental area, and cingulate signaling to targetness. Modinos et al. (
), using the same paradigm, showed similar results for novelty processing in at-risk individuals using the same task. Similar results of altered salience processing have been reported in patients with Parkinson’s disease exhibiting psychotic, mainly hallucinatory, symptoms (
). Moreover, patients with schizophrenia and early psychosis show deficits when processing emotions and intrinsic salient events, especially in the context of facial recognition (
) showed impaired neuronal signaling in the ventral striatum in response to emotional salient events in people with chronic and acute psychosis. In general, experimental findings are less consistent. A study reporting overall increased arousal in patients with schizophrenia during processing of emotionally neutral and salient stimuli (
) showed that the increase resulted solely from falsely attributing salience to neutral stimuli. This view has been confirmed by a meta-analysis showing similar processing of emotionally relevant information but attribution of aberrant salience to emotionally neutral information (
) using a P300 auditory oddball paradigm reported that reduced P300, which reflects impaired salience processing, indicated both transition to psychosis in at-risk individuals and transition to remission from psychotic symptoms. These studies show consistent findings for altered, mainly reduced, processing of salient stimuli compared with control subjects. Bringing both accounts together, a study by Boehme et al. (
) investigating healthy subjects showed that individual variability in aberrant salience measures related negatively to ventral striatal and prefrontal reward prediction error signals and, in an exploratory analysis, was found to be positively associated with nucleus accumbens presynaptic dopamine levels.
Aberrant Salience and Symptomatology
The literature provides evidence for aberrant processing of nonmotivational (and motivational) salience with regard to relevant and irrelevant events. Dysregulations in salience processing are associated with a range of brain areas in early psychosis and chronic schizophrenia (Figure 2). Here, we wish to argue that aberrant salience may explain positive and negative symptoms via different, although interrelated, mechanisms. As Maia and Frank (
) discussed, the failure to distinguish between salient and nonsalient events may be reflected in the dysregulated activity of phasic firing of dopamine neurons. While increased phasic dopaminergic firing to a nonsalient event, such as a radio in the background, may lead to the attribution of attention to this otherwise irrelevant stimulus, this suddenly important information (content of the radio show) might be reinterpreted causing, e.g., delusional thinking. However, decreased dopaminergic phasic firing to a salient event, such as an unexpected positive emotional expression on a person’s face, may blunt the importance of this information. The inadequate interpretation or evaluation of the situation may be linked to, e.g., anhedonia. The aberrant processing of these informative or relevant (i.e., salient) events, as seen in blunted prediction errors to rewards (see Prediction Error Signaling in Psychosis) (
). In contrast, aberrant processing of irrelevant (i.e., nonsalient) events, such as neutral events in reward learning or oddball paradigms, seems to provide an explanation for positive symptoms (
Figure 2Salience areas and psychosis. Simplified diagram of key regions involved in salience processing and underlying problems observed in psychosis. See text for citations and details. dl, dorsolateral; vm, ventromedial; VTA, ventral tegmental area.
). Several studies reported reduced midbrain, striatal, and/or cortical processing of reward prediction errors in psychosis, which may underpin aspects of the clinical manifestations of psychotic illness (
). Here, we will mainly focus on functional magnetic resonance imaging (fMRI) studies of reward prediction error; other closely related topics such as mismatch negativity, Kamin blocking, latent inhibition, and causal learning prediction error have been previously studied in schizophrenia and discussed elsewhere [e.g., (
) demonstrated abnormalities in patient brain responses correlating with reward prediction error in the dopaminergic midbrain, in striatal and limbic regions, and in cortical regions such as the dorsolateral prefrontal cortex (PFC). Subsequent studies using different psychological paradigms have found similar abnormalities [e.g., (
) documented impaired subcortical (midbrain) reward prediction error signals in an antipsychotic-free early psychosis sample and showed that an at-risk group with mild psychotic symptoms had a degree of midbrain signaling abnormalities. Notably, there was dorsolateral PFC prediction error dysfunction in the early psychosis sample, with comparatively intact cortical function in the at-risk group with mild psychotic symptoms.
Relevant Pharmacological and Molecular Imaging Results
Although PET studies have demonstrated robustly that there is excessive dopaminergic striatal release in schizophrenia (
), neuroimaging studies have generally shown impaired striatal signaling in patients during learning, which may appear perplexing if striatal fMRI signals are considered a pure assay of dopamine release. This apparent paradox was addressed by Bernacer et al. (
), who showed that administration of methamphetamine (which floods the striatum with dopamine) to healthy volunteers leads to a disruption of striatal prediction error–associated activity. The degree to which methamphetamine induced mild psychotic experiences was related to the degree to which it disrupted the expected value signal in the ventromedial prefrontal and posterior cingulate cortices. The study showed that a drug that reliably increases dopamine release is not beneficial to neural or behavioral indices of learning, indicating that caution should be exercised when using fMRI signals as a clean readout of neurochemical processes. A related, nuanced finding is that antipsychotic dopamine D2 receptor antagonist medications may enhance brain activations during reward processing in patients, in contrast to their effects in studies in healthy individuals (
Recent interest has focused not simply on prediction error per se, but on the precision of prediction error, which theory posits should play an important role in belief updating under uncertainty. Simple models update value or beliefs in proportion to the prediction error, but it is thought that a prediction error of a given magnitude should influence belief updating depending on the degree of uncertainty with which it is estimated. Substantially updating beliefs because of an imprecisely estimated prediction error could be maladaptive (
Precision weighting of cortical unsigned prediction error signals benefits learning, is mediated by dopamine, and is impaired in psychosis [published correction appears in Mol Psychiatry 2021; 26:5334].
Precision weighting of cortical unsigned prediction error signals benefits learning, is mediated by dopamine, and is impaired in psychosis [published correction appears in Mol Psychiatry 2021; 26:5334].
) found behavioral and brain imaging evidence that patients with early psychosis have learning abnormalities related to the degree of precision weighting of prediction error in the superior frontal cortex. They focused on unsigned prediction error (i.e., prediction error that indicates surprise without valence evaluation, being worse or better than expected), which is often associated with cortical brain activity in human fMRI studies (
). The degree of cortical abnormality was most pronounced in early psychosis and linked to the severity of positive psychotic symptoms, with relatively less impaired cortical function in at-risk patients with fewer symptoms. This pattern fits with the relatively distinct roles of the cortical areas in modulating the level of certainty of an unsigned prediction error estimation, compared with the role of the subcortical areas in signaling signed prediction errors, and hints toward a key role for the cortical function in the progression from the at-risk state to the frank psychotic illness state (Figure 3).
Figure 3Prediction error areas and psychosis. Simplified diagram of key regions involved in prediction error signaling and underlying problems observed in psychosis. See text for citations and details. VTA, ventral tegmental area.
Prediction Error Signals in Psychosis May Differ Across Illness Stages
We do note that there have been inconsistencies in the literature of prediction error signaling in psychosis. This is reflected in the conflicting accounts of two meta-analyses, one of which found relatively little evidence of abnormal fMRI reward prediction error signals in patients with schizophrenia compared with control subjects, although there were differences between schizophrenia and depression (
). Some studies that do not show group differences in activation have shown relationships at the interindividual level between striatal activation and the severity of anhedonia (
). One theme that may be emerging is that predominantly medicated samples of patients with schizophrenia have relatively intact brain prediction error signals, especially in chronic illness [e.g., (
)], whereas wholly or partly unmedicated samples, especially of early psychosis or schizophrenia, often show brain reward prediction error abnormalities (
Effort, avolition and motivational experience in schizophrenia: Analysis of behavioral and neuroimaging data with relationships to daily motivational experience.
). However, a variety of brain areas are involved in reward, and understanding the discrete contributing processes is critical for an appropriate neurobiological interpretation. Growing evidence supports a specific set of problems in patients with psychosis, related primarily to motivation and reinforcement learning.
Table 2Types of Reward Processes
Process
Role
Anticipation
Increased attention and responsiveness to upcoming rewards. In general, reward anticipation is associated with activation of the striatum, amygdala, and thalamus, and if multiple choices are available, the orbitofrontal and ventromedial prefrontal cortices can also be recruited (
). Within the striatum, the nucleus accumbens is thought to code the expected value, with the dorsal striatum (caudate) more involved in selection, action, and choice (
Comparison of various reward outcomes. Reward valuation is important in multiple stages of decision-making processes. The areas involved in computing and comparing value include the orbitofrontal cortex, nucleus accumbens, amygdala, and ventromedial prefrontal cortex (
). Initially, a comparison of the various rewards that may be available is required before computing the effort-reward trade-off. Once an outcome is acquired, its value is again assessed and used to update our understanding of the relationships between actions, effort, and reward. This is critical for effective reinforcement learning and generating anticipatory or incentive motivation for future rewards.
Effort and Motivation
Calculation of the effort-reward trade-off. The dorsal striatum is potentially involved in the selection of low-effort choices (
). In healthy people, greater endogenous striatal dopamine function or dopamine-stimulating pharmacological manipulations increase the willingness to expend effort when pursuing rewards (
Encoding the presence or absence of a reward and the actual reward value. The areas involved in monitoring and encoding reward outcomes include the nucleus accumbens, orbitofrontal cortex, ventromedial prefrontal cortex, and amygdala (
). Outcome-specific devaluation is a test of goal-directed action, requiring a participant to adjust, or bias, their actions away from an outcome after it has been devalued. This requires effective reward valuation (one outcome is now less rewarding), reward comparison (between the two possible outcomes), and then using this information to guide action selection.
Learning With a Focus on Reversal Learning
Incorporating reward outcomes and experience to navigate future choices. Many approaches have been used to probe reward learning, but one of the most widely used in psychosis research is probabilistic reversal learning, which requires a participant to navigate trial-by-trial feedback to determine which of two stimuli is rewarded more often. The presence of misleading negative feedback on the better choice (often 80% reward probability), as well as positive feedback on the worse choice (often 20% reward probability), means that the participant cannot rely solely on a “follow the win” strategy. To further probe how adaptable reward learning is, after the participant has successfully demonstrated their knowledge of the correct choice (6–10 consecutive correct trials), the contingencies are reversed. The areas commonly involved in probabilistic reversal learning include the striatum, orbitofrontal cortex, and ventral prefrontal cortex (
). Preclinical studies have also shown that dopamine is important for probabilistic reversal learning. For example, systemic amphetamine (a dopamine stimulant) administration in rats alters punishment learning without affecting win-stay use (
). Yet, increasing phasic dopamine signaling in the nucleus accumbens during a choice can increase the tendency to win-stay, even if the choice itself was not rewarded (
). A range of cortical areas have also been implicated in reward and punishment learning. For example, inactivating or lesioning the prelimbic, infralimbic, and orbitofrontal cortex subregions have all been shown to alter reward or punishment learning during reversal learning in rodents and marmosets (
Lesions of the medial striatum in monkeys produce perseverative impairments during reversal learning similar to those produced by lesions of the orbitofrontal cortex.
Lesions of the medial striatum in monkeys produce perseverative impairments during reversal learning similar to those produced by lesions of the orbitofrontal cortex.
), and in schizophrenia it has been thought to contribute to broad motivational deficits. For example, if a reward is perceived to be less valuable, then the effort-reward trade-off is biased toward inaction. However, a growing amount of evidence suggests that people with schizophrenia experience the same pleasure from positive emotional and hedonic outcomes (
Deficits in reward anticipation have consistently been observed before psychosis onset. For example, a meta-analysis (including six studies in those at risk) observed impairments in those at risk (
). A combined fMRI and PET study in healthy individuals found that nucleus accumbens dopamine release during reward anticipation was associated with activation of the dorsal striatum, amygdala, hippocampus, and thalamus (
), suggesting that dopamine release in the nucleus accumbens may be causative in recruiting the necessary networks. Multiple meta-analyses have observed reduced activation of the striatum and anterior cingulate cortex in those with schizophrenia [e.g., (
), whereas decreased activation of the nucleus accumbens in those at risk for and with chronic schizophrenia was associated with increased negative symptoms (and not with positive symptoms) (
). There is some evidence that antipsychotic treatment can improve striatal signaling during reward anticipation in people with schizophrenia but only in those who show significantly decreased positive symptoms (
) found that nucleus accumbens activation during reward anticipation decreased across adolescence in the children of people with schizophrenia (i.e., carrying a higher familial risk of developing psychosis) but not in the children of healthy control subjects. Conversely, it has been suggested that reduced anticipatory motivation may be a byproduct of a decreased ability to accurately maintain value representation (
Another primary negative symptom thought to reflect reward impairments is avolition, or a lack of willingness to do tasks required for achieving a goal. Decreased motivation is evident in self-reports in those with early psychosis (
), and there is strong evidence that people with schizophrenia are less motivated and less willing to expend the same effort to attain rewards as healthy individuals (
Effort, avolition and motivational experience in schizophrenia: Analysis of behavioral and neuroimaging data with relationships to daily motivational experience.
). In the progressive ratio task, the effort required to get a reward increases with each reward delivery, and the breakpoint refers to the point at which participants decide that the effort-reward trade-off is no longer worthwhile. This performance deficit was associated with greater amotivation scores and decreased ventral striatal function (
). Decreases in motivation and willingness to work toward goals may be specific to certain types of rewards and the required efforts. For example, those with early psychosis were less likely to select the high-effort, high-reward options than healthy control subjects (
Effort-based decision-making impairment in patients with clinically stabilized first-episode psychosis and its relationship with amotivation and psychosocial functioning.
). Furthermore, in a task in which increasing the effort improved the chance of receiving a high or low reward, people with schizophrenia were willing to expend the same effort regardless of reward size, whereas healthy individuals heavily biased their effort to increase the chance of higher rewards (
). This was associated with reduced functional changes in the caudate and anterior cingulate cortex during reward presentation and in the caudate during effort selection.
Goal-Directed Actions and Reward
In parallel to motivational and effort-based impairments, people with schizophrenia are less able to use reward information correctly when guiding their actions (
) demonstrated that people with schizophrenia were able to understand that one outcome was worth less after devaluation but failed to alter their actions in response to this information. Our work suggests that this may occur in a specific subgroup of those with chronic psychosis and is less likely to be observed in early psychosis (
). These goal-directed action impairments were due to the inability to correctly relate outcomes causally to actions rather than problems in reward valuation (
). The observed decreases in caudate function during responses in those with schizophrenia were associated with increased negative symptom severity (including avolition) (
). Preclinical studies have highlighted that striatal dopamine is important in establishing causal action-outcome associations and in action selection (
). Whether these impairments are evident at earlier disease stages is not known, but they likely reflect impairments in reward learning rather than reward valuation.
Cognitive Control and Risk
Imaging studies focused on cognitive control have indicated that across adolescence there is a marked improvement in our ability to increase cognitive performance directed at higher reward or risk. This may contribute to impairments in effort allocation in psychosis (
). Moreover, the nucleus accumbens functional coupling was greatest with the dorsal striatum in young adolescents but shifted to the ventrolateral PFC with increased age (
). Model-based reward learning is positively associated with nucleus accumbens dopamine synthesis and activation of the nucleus accumbens and lateral PFC (
). In unmedicated patients with schizophrenia who follow similar reinforcement learning strategies as control subjects, decreased nucleus accumbens activation has been observed (
), but in those who do not follow the same strategies (i.e., poorer performers), decreased activation of both the nucleus accumbens and ventrolateral PFC has been observed (
), suggesting that model-based circuitry may be dysfunctional. In another study looking at high-reward/risk comparisons, caudate and dorsolateral prefrontal coupling increased with age (
). This may indicate that when comparing reward values in choice situations, the maturation of corticostriatal systems critical for focusing cognitive effort are impaired or delayed in people with psychosis. Evidence of functional connectivity alterations in those with early psychosis and chronic schizophrenia demonstrates progressive deviation from healthy control subjects (
). However, other studies have demonstrated more widespread connectivity changes in unmedicated people with early psychosis, with changes in limbic circuits still pronounced (
Functional connectivity in antipsychotic-treated and antipsychotic-naive patients with first-episode psychosis and low risk of self-harm or aggression: A secondary analysis of a randomized clinical trial.
). Improvements in thalamocortical connectivity were associated with antipsychotic treatment, suggesting that medication status is important when interpreting changes in functional connectivity.
Reward Learning
There is ample evidence that reward learning, or reinforcement learning, is altered in psychosis. Deficits in reward learning have been associated with higher general symptoms (
). Furthermore, psychosis is associated with jumping to conclusions, whereby those with psychosis are more likely to update their beliefs using less information [e.g., (
). Decreased activation of the nucleus accumbens, anterior cingulate cortex, and dorsolateral PFC has been associated with impaired reward learning in those with schizophrenia (
). Win-stay refers to a participant selecting the same stimulus after winning a reward on the prior trial. Poorer reversal learning performance in unmedicated people with schizophrenia has been associated with decreased activation of the ventrolateral PFC and nucleus accumbens (
Neural, behavioural and real-life correlates of social context sensitivity and social reward learning during interpersonal interactions in the schizophrenia spectrum.
) provide evidence of how these deficits can increase the functional burden of those with schizophrenia. Overall, it appears that people with schizophrenia are less able to use rewarding feedback to guide learning. Working memory deficits may also present as impaired reward learning processes, which has been demonstrated using task-based and computational modeling–based approaches in those with schizophrenia (
). These conflicting studies suggest that reinforcement learning deficits may not be a universal core trait in psychosis but rather a feature in a large subgroup of those with psychosis. Deficits in reinforcement learning are observed in those at risk for psychosis (
Reinforcement learning as an intermediate phenotype in psychosis? Deficits sensitive to illness stage but not associated with polygenic risk of schizophrenia in the general population.
). In contrast, reinforcement learning studies conducted in those with early psychosis suggest that deficits may include punishment learning, specifically a decreased sensitivity to punishment (
Reinforcement learning as an intermediate phenotype in psychosis? Deficits sensitive to illness stage but not associated with polygenic risk of schizophrenia in the general population.
). However, reversal learning impairments in early psychosis are less robust than in those with persistent schizophrenia, with some studies observing relatively intact performance (
). Clearly more work, including longitudinal studies, is required to determine what reversal learning indices in those at risk for psychosis or with early psychosis mean for subsequent outcomes (both diagnostically and in terms of treatment efficacy).
Reward Systems in Psychosis
Overall, schizophrenia is associated with a broad group of reward deficits spread across multiple brain areas (Figure 4). These include anticipatory and effort-related motivation, reward-based decision making, and reward learning. Nevertheless, all of these processes rely heavily on corticostriatal circuitry, which corroborates well-known alterations in striatal dopamine function (
Functional connectivity in antipsychotic-treated and antipsychotic-naive patients with first-episode psychosis and low risk of self-harm or aggression: A secondary analysis of a randomized clinical trial.
Figure 4Reward areas and psychosis. Simplified diagram of key regions involved in reward processing and underlying problems observed in psychosis. Impairments tend to be less obvious in earlier disease stages or, in the case of reward/punishment learning, may actually be in opposition. See text for citations and details.
Our review is based heavily on cross-sectional data using different experimental paradigms. Although not surprising, this highlights the need for focused longitudinal studies to track how these processes change across illness stages. Nevertheless, a picture is emerging over several studies showing that cortical function during reward and salience processing is impaired in psychotic illness (established/chronic schizophrenia), especially in the ventral and dorsolateral PFC and the anterior cingulate cortex, but relatively spared in the earliest stages of psychosis (Figure 5). However, there is evidence of subcortical reward and salience dysfunction in at-risk patients, consistent with PET studies showing dopaminergic abnormalities, especially increased levels of striatal dopamine (
), early in the course of illness. In psychosis, antipsychotic medication may normalize some learning abnormalities and learning-related brain signals, such as prediction error signals (
), potentially explaining the alleviating effects on positive symptoms. We suggest that subcortical reward and salience dysfunction may be an early manifestation of the illness, with cortical abnormalities in these domains becoming more prominent as the illness progresses. This contrasts with the hypotheses of schizophrenia that have proposed the primary abnormalities as being cortical in origin, which proceed and/or induce subcortical dopamine dysfunction (
). However, our proposal is consistent with some animal models that show proof of principle for cortical dysfunction secondary to primary subcortical lesions (
Figure 5Neurochemical and functional imaging associations with salience and reward processing in those at risk for and in early psychosis compared with (A) chronic psychosis and (B) schizophrenia. Colored arrows reflect key neurochemical changes, and black arrows indicate functional magnetic resonance imaging changes in those with psychosis and their associated behavior. Evidence of salience and reward impairments in those at risk for and in early psychosis are commonly found in subcortical structures, such as the associative striatum, nucleus accumbens, ventral tegmental area (VTA), and substantia nigra (SN). However, evidence of cortical glutamatergic abnormalities in the anterior cingulate cortex (ACC) have been observed. In contrast, there is evidence of widespread functional impairments in those with chronic psychosis or schizophrenia. This includes reduced functional connectivity in thalamocortical and corticostriatal projections (dashed lines) and reduced functional activity during salience and reward processing in the ACC, ventral prefrontal cortex (vPFC), and dorsolateral PFC (dlPFC). Therefore, it may be that subcortical alterations precede cortical impairments in driving negative symptoms and deficits in salience and reward processing.
There are several potentially helpful new interventions at various stages of development, highly relevant for reward and salience processing domains in psychosis. For example, the behavioral intervention of cognitive remediation therapy has already been shown to be capable of modulating the aspects of reinforcement learning, such as sensitivity to rewards and punishments (
). Moreover, reward learning can be used to improve attentiveness during conversational skill learning relevant for everyday functioning, improving outcomes (
). Pharmacological interventions at numerous molecular targets are of interest in the treatment of cognitive deficits in schizophrenia, including reinforcement learning domains, and could potentially be combined with cognitive remediation interventions (
Reward learning as a potential target for pharmacological augmentation of cognitive remediation for schizophrenia: A roadmap for preclinical development.
) indicates that administration of prepubertal benzodiazepines mitigates the deleterious effect of perinatal or adolescent insults that otherwise lead to a hyperdopaminergic state directly relevant for salience processing. Recent advances in noninvasive brain stimulation also show potential for modulating brain circuits discussed in this article; for example, transcranial-focused ultrasound appears to be well tolerated in humans and has recently been shown to have the potential to target not only cortical but also subcortical structures and thus potentially to modulate brain activations in networks throughout the brain (
). fMRI-based techniques that draw on real-time fMRI signal decoding and neurofeedback are under investigation regarding their ability to influence various psychological states, including addressing symptom domains in schizophrenia [e.g., reduction of auditory hallucinations (
Real-time functional magnetic resonance imaging neurofeedback for the relief of distressing auditory-verbal hallucinations: Methodological and empirical advances.
)], and merit further investigation. To fully realize the potential benefits of novel brain stimulation technology, we need to accelerate our understanding of the causal relationships between brain circuits and behavior in patients.
Conclusions
Together, salience and reward processes are integral to engaging our attention, stimulating anticipation of future events, and driving goal-directed behaviors. In this review, we have summarized the current knowledge of behavioral and functional neuroimaging in salience, reward, and prediction error. Although they are specific processes, they interact in multiple feedback and feedforward systems essential for decision making and cognition more generally. Further studies focused on subcortical systems during adolescence and the transition to psychosis are warranted.
Acknowledgments and Disclosures
This work was supported by the National Health and Medical Research Council (Grant No. APP1139960 [to JPK]), the Brain & Behavior Research Foundation (2019 Maltz Prize [to JPK]), and the European Union’s Horizon 2020 (Grant No. 754462 [to FK]).
The authors report no biomedical financial interests or potential conflicts of interest.
Precision weighting of cortical unsigned prediction error signals benefits learning, is mediated by dopamine, and is impaired in psychosis [published correction appears in Mol Psychiatry 2021; 26:5334].
Neural and behavioral correlates of aberrant salience in individuals at risk for psychosis [published correction appears in Schizphr Bull 2016; 42:1303].
Changes in EEG spectral power on perception of neutral and emotional words in patients with schizophrenia, their relatives, and healthy subjects from the general population.
Effort, avolition and motivational experience in schizophrenia: Analysis of behavioral and neuroimaging data with relationships to daily motivational experience.
Effort-based decision-making impairment in patients with clinically stabilized first-episode psychosis and its relationship with amotivation and psychosocial functioning.
Functional connectivity in antipsychotic-treated and antipsychotic-naive patients with first-episode psychosis and low risk of self-harm or aggression: A secondary analysis of a randomized clinical trial.
00159-2&id=gr1.jpg){kind=link}
00159-2&id=gr2.jpg){kind=link}
00159-2&id=gr3.jpg){kind=link}
00159-2&id=gr4.jpg){kind=link}
00159-2&id=gr5.jpg){kind=link}