REINFORCEMENT LEARNING ABNORMALITIES IN THE ATTENUATED PSYCHOSIS SYNDROME AND FIRST EPISODE PSYCHOSIS

Abstract

Prior studies indicate that chronic schizophrenia (SZ) is associated with a specific profile of reinforcement learning abnormalities. These impairments are characterized by: 1) reductions in learning rate, and 2) impaired Go learning and intact NoGo learning. Furthermore, each of these deficits are associated with greater severity of negative symptoms, consistent with theoretical perspectives positing that avolition and anhedonia are associated with impaired value representation. However, it is unclear whether these deficits extend to earlier phases of psychotic illness and when individuals are unmedicated. Two studies were conducted to examine reinforcement learning deficits in earlier phases of psychosis and in high risk patients. In study 1, participants included 35 participants with first episode psychosis (FEP) with limited antipsychotic medication exposure and 25 healthy controls (HC). Study 2 included 17 antipsychotic naïve individuals who were at clinical high-risk for psychosis (CHR) (i.e., attenuated psychosis syndrome) and 18 matched healthy controls (HC). In both studies, participants completed the Temporal Utility Integration Task, a measure of probabilistic reinforcement learning that contained Go and NoGo learning blocks. FEP displayed impaired Go and NoGo learning. In contrast, CHR did not display impairments in Go or NoGo learning. Impaired Go learning was not significantly associated with clinical outcomes in the CHR or FEP samples. Findings provide new evidence for areas of spared and impaired reinforcement learning in early phases of psychosis.

Introduction

Negative symptoms have been considered a core component of schizophrenia (SZ) since the pre-neuroleptic era (Bleuler, 1950; Kraepelin, 1921). Modern empirical studies confirm the importance of negative symptoms, indicating that they are highly prevalent (Strauss and Cohen, 2017) and predictive of a range of poor clinical outcomes (e.g., lower rates of recovery, quality of life, subjective well-being, functional outcome) (Fervaha et al., 2015; Strauss et al., 2012, 2010). Unfortunately, there are currently no phramacological or psychosocial treatments that have proven efficacious for remediating negative symptoms (Fusar-Poli et al., 2015).

Several recent mechanistic accounts of negative symptoms have been proposed, which suggest that they result from disrupted cortico-striatal circuitry and abnormalities in reward processing (Barch and Dowd, 2010; Gold et al., 2008; Kring and Barch, 2014; Strauss et al., 2014). A key reward processing domain that has been implicated in negative symptoms is reinforcement learning (i.e., the ability to dynamically update learned associations between outcomes and the stimuli or actions associated with them). Several studies indicate that SZ patients have difficulties in learning to make stimulus-outcome or action-outcome associations in probabilistic decision-making tasks that require learning from positive feedback (Reinen et al., 2016; Strauss et al., 2011; Waltz et al., 2011, 2007). Such deficits may result from aberrant D1 receptor functionality and disrupted phasic striatal dopaminergic signaling that leads to impaired “Go learning”. Several studies show that greater negative symptom severity is associated with: 1) impairments in rapid, trial-to-trial behavioral adaptation in response to recent changes in reinforcement contingencies during the early phase of learning; 2) impairments in Go learning and learning from positive feedback (Gold et al., 2012; Strauss et al., 2011; Waltz et al., 2007). Neuroimaging studies suggest that these deficits result from aberrant prefrontal activation and reduced striatal response during positive prediction errors (Gradin et al., 2011; Morris et al., 2012; Murray et al., 2008; Radua et al., 2015; Waltz et al., 2009).

Although a wealth of studies support an association between reinforcement learning abnormalities and negative symptoms in the chronic phase of illness among patients who are typically prescribed antipsychotics (Strauss et al., 2014), it is unclear whether this relationship holds in the prodromal and first episode phases of illness among medication naive individuals. Psychotic disorders are typically preceded by a prodromal (i.e., pre-illness) phase characterized by functional decline and subthreshold positive symptoms that progressively worsen over the course of several months to years (Fusar-Poli et al., 2013). Negative symptoms play a critical role in the prodromal phase. For example, they are highly prevalent (e.g., present in 82% of cases of at least moderate severity) and longitudinally stable, one of the earliest indicators of psychosis risk (emerging years before attenuated positive symptoms), a frequent reason why individuals first come into contact with the treatment system, and one of the strongest predictors of conversion to psychosis (Carrión et al., 2016; Häfner et al., 1999; Piskulic et al., 2012). Given that the prodromal period is of interest both as a window for investigating processes involved in disease onset and as a potential point of intervention and prevention, identifying mechanisms associated with negative symptoms during this phase is critical (Addington and Heinssen, 2012). Few studies have examined reinforcement learning in those at clinical high-risk for psychosis, with results generally suggesting subtle reinforcement learning deficits that are associated with reduced striatal activation (Karcher et al., 2019, 2015; Millman et al., 2019; Rausch et al., 2015; Roiser et al., 2013; Schmidt et al., 2017; Waltz et al., 2015). Greater severity of negative symptoms has also been linked to reduced activation of the ventromedial prefrontal cortex in CHR participants (Millman et al., 2019).

Similarly, studies suggest that negative symptoms are also prevalent during the first episode of psychosis; however, they tend to be transient and driven by secondary factors (e.g., depression, anxiety, medication effects) during the first year of treatment, after which they become more persistent (Chang et al., 2011). Although the first episode tends to be more transitory, when negative symptoms are present and persistent in this phase, they tend to predict poor long-term outcomes and are therefore also important early course treatment targets (Chang et al., 2011). Reinforcement learning deficits are also present during the first episode of psychosis (Chang et al., 2016; Huddy et al., 2011; Leeson et al., 2009; Murray et al., 2008; Schlagenhauf et al., 2014), and have been tied to dysfunctional prediction error signaling (Corlett et al., 2007). However, they may be mild and their association with negative symptoms is less clear during this phase (Chang et al., 2016).

The current study examined reinforcement learning in CHR and first episode psychosis (FEP) participants using a well-validated Go and NoGo reinforcement learning paradigm (Moustafa et al., 2008). Consistent with past studies (Chang et al., 2016; Gold et al., 2012; Strauss et al., 2011), we hypothesized that FEP and CHR would display impaired Go learning, which would be selectively associated with increased severity of negative symptoms. In contrast, whereas past studies on medicated patients in the chronic phase of schizophrenia have demonstrated intact NoGo learning potentially due to stabilizing effects of medication on D2 driven reinforcement learning, we hypothesized that unmedicated CHR and FEP participants with only limited medication exposure would display impaired NoGo learning. Furthermore, greater reinforcement learning impairments were expected to be associated with greater probability of conversion to a psychotic disorder measured via a cross-sectional risk calculator in those at CHR.

Study 1

Participants

The sample consisted of 60 participants – 26 healthy controls (CN) and 34 first episode psychosis (FEP) patients. FEP patients were recruited from outpatient clinics, inpatient units and the emergency room at the University of Alabama at Birmingham (UAB). Studies were approved by the University of Alabama at Birmingham Institutional Review Board. All participants were between 14 and 55 years old and were capable of understanding the study procedures and able to provide informed consent (over 18). For minors assent was obtained with written-informed consent provided by their parent or legal guardian. Exclusion criteria were major neurological or medical conditions, history of significant head trauma, substance use disorders (excluding nicotine and cannabis) within 1 month. Healthy controls were recruited by flyers and advertisements in the local Birmingham community. All participants signed informed consent for a protocol approved by the UAB institutional review board. FEP and CN groups did not significantly differ on age, ethnicity, sex, personal education, or parental education (see Table 1).

Table 1.

Group Demographic Characteristics for Study 1 and Study 2

	FEP (n = 34)	CN (n = 25)	Test Statistic	CHR (n = 17)	CN (n = 17)	Test Statistic
Age (M, SD)	22.8 (4.9)	22.4 (4.3)	F = 0.13	20.0 (2.1)	20.82 (2.0)	F = 1.2
Female (n, %)	10, 29.4%	10, 40%	X2 = 0.72	15, 88.2%	13, 76.5%	□2 = 0.8
Personal education	12.7 (1.7)	14.6 (2.1)	F = 15.3***	13.7 (1.5)	14.6 (1.7)	F = 3.1
Parental education	-	-	-	15.7 (2.6)	15.6 (1.8)	F = 0.1
Race (n, %)			X2= 4.2			X 2 =2.7
African American	24, 70.6%	13, 52%		-	-
Asian American	2, 5.9%	1, 4%		2, 11.8%	6, 35.3%
Caucasian	7, 20.6%	11, 44%		14, 82.4%	10, 58.8%
Hispanic/Latino	-	-		1, 5.9%	1, 5.9%
Other	1, 2.9%	0, 0.0%		-	-

Note. FEP = First Episode; CHR = Clinical High-Risk; CN = Healthy Control

^*p <.05

^**p<.01

^***p<.001

Procedures

Data were collected as part of evaluations of two larger neuroimaging studies in FEP. The first recruited initially antipsychotic-medication naïve subjects who were then placed on risperidone in a longitudinal trial (Briend, Nelson, et al. 2020; Briend, Armstrong, et al. 2020; Maximo et al. 2020; Kraguljac et al. 2020). The second was a cross-sectional study in medicated FEP with an average illness duration of approximately one year (Reid et al. 2019; Overbeek et al. 2019; Kraguljac et al. 2018; Lottman et al. 2019). Consensus diagnoses were made according to DSM-5 criteria by two board certified psychiatrists (ACL and NVK) from all historical and direct assessment information available including medical records and information from the Diagnostic Interview for Genetic Studies (DIGS; Nurnberger et al., 1994) or the MINI International Neuropsychiatric Interview (Sheehan et al. 1998). The Scale for the Assessment of Negative Symptoms (SANS; Andreasen, 1989), Scale for the Assessment of Positive Symptoms (SAPS; Andreasen, 1984), and Brief Psychiatric Rating Scale (BPRS; Overall and Gorham, 1962) were completed following a standard clinical interview by raters trained to reliability standards to severity of positive, negative, disorganized, and general symptoms. Neuropsychological impairment was assessed via the Repeatable Battery for the Assessment of Neuropsychological Status (Randolph et al., 1998). Participants received monetary compensation for participation in the studies.

Temporal Utility Integration Task

It is important to note that neuropsychological testing, including the Temporal Utility Integration Task, was obtained when FEP patients were medicated and deemed clinically stable enough by one of the study physicians to proceed with testing, typically within one to two months of initiation of antipsychotic medications for those enrolled in the longitudinal trial.

The reinforcement learning task originally developed by Moustafa et al. (2008) and examined in chronic schizophrenia patients by Strauss et al. (2011) was administered to all participants. In this task, subjects saw a clock face, which had a single revolving arm that made a full cycle over the course of 5 seconds. Participants were asked to press a button before the arm made a full cycle. After their response, a screen appeared to inform participants whether they had won points on that trial and the amount of points won (“You win XX points!”). Each trial ended after participant response or if 5s elapsed without response. The inter-trial-interval was 1s.

The task had 4 conditions, with a total of 200 trials (50 per condition). In each condition, reward probability and magnitude varied as a function of the length of time that elapsed on the clock until participants made a response. There were 3 primary conditions in which the expected value (probability x magnitude) either decreased (DEV), increased (IEV), or remained constant (CEV), across the 5 s trial. As result of these expected value manipulations, the IEV condition was used to measure NoGo learning (i.e., slower responses resulted in a higher number of points), the DEV condition was used to measure Go learning (i.e., faster responses resulted in a higher number of points). When evaluated in conjunction with the CEV condition, the IEV and DEV conditions allow for estimates of NoGo and Go learning while accounting for any general differences in response speeding across individuals.

A secondary condition, CEVR, was also included. In CEVR, expected value remained constant (like CEV), but reward probability increased and magnitude decreased as time elapsed on the clock (i.e., the opposite to CEV). Given that CEV and CEVR have equal expected values across the entire clock face, difference in RT observed in these conditions may signify whether a subject has a bias to learn more from reward probability than magnitude or magnitude than probability. Practically, if a subject has slower RTs in CEVR than in CEV, they are risk averse (i.e., their responses are biased toward obtaining higher frequency of reward than higher magnitude of reward).

The order in which the 4 conditions were administered was counterbalanced across participants, with a brief rest period between conditions. Each condition had a different color clock face, with colors counterbalanced across participants. At the beginning of each condition, subjects were instructed to respond at different times in order to find the interval on the clock that would allow them to win the most points; however, they were not told about the different rules for each condition (e.g., slowing is optimal in IEV and speeding is optimal in DEV). E-Prime was used to present the task.

Data Analysis

A 2 Group x 10 Block Mixed Models ANOVA was calculated separately for DEV-CEV and IEV-CEV contrast conditions. Significant Group X Block interactions were followed up by post hoc one-way ANOVAs and within group paired-samples t-tests. Bivariate correlations were calculated to examine associations between Go (DEV-CEV) and NoGo (IEV-CEV) learning with SAPS and SANS factor scores. The following SANS items not part of the negative symptom construct were removed from summary scores: poverty of content of speech, inappropriate affect, all attention items Two sets of task variables were conducted for both Go and NoGo learning (learning rate: end block – beginning block) and global performance (mean DEV/IEV-CEV for the entire block).

Results

Go Learning

Mixed Models ANOVA indicated a significant Group X Block interaction for Go learning (DEV-CEV), F (9, 522) = 1.93, p = 0.046 (partial eta square = .032). The main effect of Block was also significant, F (9, 522) = 2.24, p = 0.018 (partial eta square = .037); however, the main effect of Group was nonsignificant, F (1, 58) = 0.70, p = 0.41 (partial eta square = .01). Post hoc one-way ANOVAs indicated a significant Group difference for DEV Block 5 (F = 7.03, p = .01) and a trend for Block 4 (F = 3.51, p = 0.066). Within group paired samples t-tests conducted in CN indicated a significantly faster decrease in RT (i.e., Go Learning) from Block 1 to 7 (where the learning effect was maximal) in CN (t = 3.12, p =0.005); however, this effect was nonsignificant in FEP (t = 1.62, p = 0.11). All other sequential pairwise comparisons (i.e., block 1–2, 2–3 etc) were nonsignificant in both groups. These findings suggest poorer Go learning in FEP than CN.

NoGo Learning

Mixed Models ANOVA indicated a nonsignificant group X Block interaction for NoGo learning (IEV-CEV), F (9, 513) = 0.37, p = 0.95 (partial eta square = .01). The main effect of Block was also nonsignificant, F (9, 513) = 1.69, p = 0.088 (partial eta square = .029); however, the main effect of Group was significant, F (1, 57) = 5.01, p = 0.029 (partial eta square = .08), such that FEP (IEV-CEV: M = 149, SE = 90) had poorer NoGo learning than CN (IEV-CEV: M = 466, SE = 108).

Correlations

In FEP, Go and NoGo learning were not significantly associated with SAPS total, SANS total, or SANS domain scores for anhedonia, avolition, asociality, blunted affect, or alogia.

Study 2

Participants

Participants included 17 CHR and 17 healthy controls (CN). CHR participants were recruited from the Georgia Psychiatric Risk Evaluation Program (G-PREP: directed by G.P. Strauss), which received referrals from local clinicians (e.g., Psychiatrists, Psychologists, Social Workers, School Psychiatrists) to perform diagnostic assessment and monitoring evaluations for youth displaying early psychosis. CHR youth were also recruited via online and print advertisements, and in-person presentations to community mental health centers.

CHR participants were included if they met criteria for a prodromal syndrome on the Structured Interview for Prodromal Syndromes (SIPS; McGlashan et al., 2001). SIPS criteria included: Attenuated Positive Symptoms (i.e., SIPS score of at least 3–5 on at least one positive symptom item) (n = 16) and Brief Intermittent Psychosis Syndrome (BIPS) (n = 1). SIPS criteria can include Genetic Risk and Deterioration Syndrome (i.e.,1st degree relative with a psychotic disorder and decline in global functioning over the past year), but no participants in this sample met criteria for this syndrome. Within the syndromes, 11 of the CHR participants met criteria for progression, five for persistence, and one for partial remission status. CHR youth did not meet lifetime criteria for a DSM-5 psychotic disorder as determined via SCID interview (SCID; M B First et al., 2015). No CHR participants had been prescribed an antipsychotic.

CN participants were recruited from the local community using posted flyers and electronic advertisements. CN participants had no current major (former Axis I) DSM-5 diagnoses or schizophrenia-spectrum personality disorder diagnoses (cluster A) as established by the SCID and SCID-PD (M B First et al., 2015; Michael B. First et al., 2015). CN also had no family history of psychosis and were not taking psychotropic medications. All participants were free from lifetime neurological disease. All participants provided written informed consent for a protocol approved by the University Georgia Institutional Review Board and received monetary compensation for their participation. Groups did not significantly differ on age, ethnicity, sex, personal education, or parental education (see Table 1).

Procedures

Prior to completing the Temporal Utility Integration task, examiners who were trained to reliability standards (ICC > 0.80), conducted a structured diagnostic interview with all participants to complete the SCID-I, SCID-PD, SIPS, BNSS, GFS:S, GFS:R. SIPS interviews were either performed directly by the PI or by a clinical psychology doctoral student or lab staff member trained to reliability standards who consulted with the PI on all cases for consensus. Negative symptoms were rated using the Brief Negative Symptom Scale (BNSS; Kirkpatrick et al., 2011), which was previously adapted for use in CHR populations with good reliability and validity (Strauss and Chapman, 2018). Functional outcome was assessed using the Global Functioning Scale Social and Role instruments (GFS:S and GFS:R; Cornblatt et al., 2007). Participants completed the same Temporal Utility Integration task described in Study 1.

Data Analysis

The analytic strategy was the same as study 1. Bivariate correlations were calculated between Go and NoGo learning measures and CHR clinical measures: SIPS positive, SIPS disorganized, SIPS general, BNSS total and domain scores, GFS:S, GFS:R, and NAPLS risk calculator.

Results

Go Learning

Mixed Models ANOVA indicated nonsignificant effects for the Group X Block interaction, F (9, 297) = 0.38, p = 0.95 (partial eta square = .01), main effect of Block, F (9, 297) = 0.80, p = 0.62 (partial eta square = .02), and main effect of Group, F (1, 33) = 0.22, p = 0.64 (partial eta square = .01).

NoGo Learning

Mixed Models ANOVA indicated a nonsignificant Group X Block interaction, F (9, 297) = 0.85, p = 0.57 (partial eta square = .025), trend toward a main effect of Block, F (9, 297) = 1.78, p = 0.07 (partial eta square = .02), and nonsignificant main effect of Group, F (1, 33) = 0.04, p = 0.85 (partial eta square = .001).

Correlations

In CHR, Go and NoGo learning were not significantly associated with positive symptoms, negative symptoms, or functional outcome.

Discussion

Several key findings emerged in the current study. First, FEP participants displayed impairments in both Go and NoGo learning relative to CN. When contrasted to our prior study on medicated outpatients in the chronic phase (Strauss et al., 2011), which found impaired Go learning and spared NoGo learning in schizophrenia, these findings may shed light onto the role of illness chronicity and long-term antipsychotic medication exposure in Go and NoGo learning. It is possible that antipsychotics stabilize NoGo learning due to their effect on the D2 pathway responsible for NoGo learning. In contrast, Go learning impairment may be relatively unmodulated by antipsychotic medications and therefore present in FEP and chronic phases because it is dependent on D1 pathway function. Go learning deficits may emerge early in the FEP period and persist into the chronic phase, in part because antipsychotics fail to act on this pathway.

Second, as a group, CHR participants were unimpaired in both Go and NoGo learning. These findings are contrary to some prior studies indicating that those at CHR display subtle impairments in reinforcement learning (Karcher et al., 2019, 2015; Millman et al., 2019; Rausch et al., 2015; Roiser et al., 2013; Schmidt et al., 2017; Waltz et al., 2015). Discrepancies across studies may be due in part to methodological differences in the types of reinforcement learning paradigms used. Several prior studies examined associations between reward predictive cues and the ability to associate those cues with reward or loss outcomes. Their primary outcome measures were neurophysiological indicators of brain response to the cues and outcomes themselves, rather than pure behavioral responses, such as those measured in the current study. It is possible that CHR subjects obtained intact behavioral performance due to compensatory mechanisms, despite having neural abnormalities. Future neuroimaging studies are needed to evaluate this possibility. Another explanation for inconsistencies across studies might be differences in the proportion of CHR participants who eventually develop a psychotic disorder. The CHR status is one that is markedly heterogeneous, with only approximately 35% of those designated as being at-risk actually transitioning to a psychotic disorder (Cannon et al., 2008). It is unclear what percentage of our subjects might go on to develop a psychotic disorder since study 2 was cross-sectional. Prospective longitudinal studies of CHR participants are needed to determine whether CHR converters do indeed display greater impairments in Go learning than non-converters.

Third, Go and NoGo learning were not significantly associated with positive or negative symptoms in CHR or FEP samples. These findings are contrary to past studies on chronic schizophrenia, which have reported an association between negative symptoms and Go learning impairments specifically. Since reward processing mechanisms are thought to be associated more with primary than secondary (e.g., depression) causes of negative symptoms in the chronic phase of schizophrenia (Strauss et al., 2014), it is possible that the increased rate of secondary negative symptoms in the CHR and FEP samples attenuated the association with reinforcement learning. It is possible that reward processing abnormalities are more core to primary negative symptom patients (that are more prevalent in the chronic phase), which form a more heterogeneous group than those with secondary negative symptoms(that can result from numerous factors, such as depression, anxiety, disorganization, positive symptoms). Additionally, we also failed to find an association with positive symptoms. In the chronic phase and early psychosis, reinforcement learning abnormalities have also been implicated as a mechanism underlying positive symptoms. Specifically, it has been proposed that tonic dysregulation of dopaminergic activity impacts phasic dopaminergic signaling, leading individuals to ascribe aberrant salience toward environmental cues and outcomes that are inherently more neutral, while reducing salience ascribed toward motivational cues that are relevant. Abnormal reward prediction errors (i.e., mismatches between obtained and expected outcomes) that occur during reinforcement learning may drive this form of aberrant salience that has been associated with positive symptoms. Prior studies using the temporal utility integration task have also failed to find an association with positive symptoms, potentially suggesting that this measure is not ideally suited for measuring the types of reinforcement learning abnormalities underlying positive symptoms. Finally, CHR subjects are generally more intact than chronic schizophrenia or FEP samples in cognitive function. It is possible that general cognitive impairments might drive reinforcement learning abnormalities in FEP and schizophrenia moreso than in less impaired CHR subjects. Moreover, the lack of correlation between task performance and symptom scores in the CHR and FEP samples may highlight equifinality in positive and negative symptom mechanisms across phases of illness (i.e., it is possible for two participants to receive similar scores on a rating scale across different phases of illness as result of different underlying neurobiological mechanisms). Large prospective longitudinal studies using a multimethod biomarker approach are needed to identify mechanisms associated with positive symptoms, negative symptoms, and community functioning.

Certain limitations should be considered. First, both studies were cross-sectional. Prospective longitudinal studies are needed in both FEP and CHR samples. Second, the effect of antipsychotics could not be definitively determined since second generation antipsychotics have different receptor profiles and the FEP group was on a range of medications. However, we speculate that antipsychotics may have a greater normalizing effect on NoGo learning, given its purported involvement in the D2 pathway, whereas Go learning is thought to be driven by a D1 pathway. Future studies that initially test CHR/FEP patients unmedicated and then again once stably medicated are needed to evaluate this possibility. In CHR, it would be beneficial to administer reinforcement learning tasks at baseline and then again at multiple follow-ups over the course of several years to determine if task performance predicts conversion. Third, our data were purely behavioral. Conclusions about neurobiological mechanisms are inferred from past studies, and follow-up fMRI and EEG experiments are needed. Fourth, although our results are valuable for hard to ascertain early psychosis populations, the sample sizes included in each study were modest and may have been under-powered to detect some effects. Fifth, we administered a single reinforcement learning task targeting learning rate for both Go and NoGo conditions. Although well-validated, this task does not lend itself to other comparisons of interest, such as learning from gains versus losses. Sixth, different negative symptom scales were used across studies. This was because the studies were started at different times and the SANS would not be appropriate for use in a CHR population (e.g., lack of validation, probes about sexual behavior that are not appropriate for adolescents).

Despite these limitations, the current findings have important implications for understanding reinforcement learning abnormalities in psychosis. Specifically, we found that a group of FEP participants displayed impairment in both Go and NoGo learning. These findings may suggest that prior studies examining participants in the chronic phase of schizophrenia found intact NoGo learning due to normalizing effects of long-term antipsychotic treatment. We failed to find associations between reinforcement learning deficits and measures of negative or positive symptoms, potentially suggesting equifinality in mechanisms underlying these symptoms across phases of illness. If future studies replicate these findings and extend them with fMRI, results may suggest that prefrontal regions are appropriate targets for non-invasive brain stimulation treatments.

Figure 1. Go and NoGo Learning in First Episode Psychosis and Clinical High-Risk for Psychosis Participants Compared to Healthy Controls.

Note. A = First Episode Go learning; B = First Episode NoGo learning; C = Clinical high-risk Go learning; D = Clinical high-risk NoGo learning