Neural Responses to Reward and Psychosis Dimensions

Abstract

Background and Hypothesis

Deficits in motivational drive, reinforcement learning, and reward prediction have been implicated in psychotic disorders. Electroencephalography (EEG) measures of these processes have demonstrated inconsistent relationships with psychotic disorders. The present study addressed key limitations in the literature and examined the association between multiple EEG neural responses to reward and different psychosis dimensions.

Study Design

The sample included 362 adults (age M = 49.87, 42.3% female): 169 with a history of hospitalization for a psychotic disorder and 193 demographically matched with no history of a psychotic disorder. Participants completed a monetary guessing task, and we measured the time-domain reward positivity (RewP) as well as time-frequency delta activity to win and theta activity to loss. Participants also completed clinical interviews, questionnaires, and neuropsychological measures.

Study Results

Categorical analyses indicated no group differences in the RewP, but participants with schizophrenia or schizoaffective disorder had more blunted delta activity to win and theta activity to loss. Dimensional analyses indicated more blunted RewP, delta activity to win, and theta activity to loss were associated with greater psychosis dimension scores. More blunted delta activity to win was also associated with greater lower-order detachment and higher-order general psychopathology dimension scores. Executive functions mediated the relationship between both delta and theta activity and the psychosis dimension.

Conclusions

Blunted EEG time-domain and time-frequency neural responses to reward are associated with dimensional representations of psychosis. The study also highlights the importance of cognitive deficits when examining the neural response to reward and psychosis.

Introduction

Psychotic disorders are heterogenous and characterized by both positive symptoms (eg, delusions, hallucinations) and negative symptoms (eg, avolition, diminished emotional expression).¹ Antipsychotic medications have limited efficacy for negative symptoms,^2-4 which relate to functional impairments and lead to a wide range of deficits,^5-7 including difficulties initiating and engaging in goal-directed activities. There is still a need to better understand the neurobiological mechanisms of psychotic disorders, particularly for negative symptoms, to improve early identification, prevention, and treatment (eg, drug discovery).

Deficits in motivational drive, reinforcement learning, and reward prediction have been implicated in psychotic disorders.^8-10 Multiple meta-analyses have indicated blunted functional magnetic resonance imaging striatal activation during reward anticipation and feedback in psychotic disorders.^11-13 In electroencephalography (EEG) research, the reward positivity (RewP), an event-related potential (ERP) component thought to index reinforcement learning and reward sensitivity,^14,15 is blunted in depression¹⁶ and more specific features, such as anhedonia.^17,18 In contrast, a meta-analysis indicated the RewP did not demonstrate a consistent relationship with psychotic disorders,¹⁹ which was again found in a more recent investigation.²⁰ The RewP was recently found to be positively associated with a behavioral measure of decision-making,²¹ and cognitive deficits are another hallmark characteristic of psychotic disorders.²² It is possible that the RewP is also associated with cognitive deficits in psychosis. Overall, the literature on the RewP and psychosis has involved important limitations, including small patient sample sizes (n’s = 18-54), categorical assessment of psychosis, and limited assessment of cognition and comorbid psychopathology. It is important to consider alternative approaches toward the assessment of the neural response to rewards in psychotic disorders.

The Hierarchical Taxonomy of Psychopathology (HiTOP) is an empirically derived, dimensional, and hierarchical model of psychopathology that addresses many limitations of categorical diagnoses.²³ The HiTOP model allows for the measurement of psychopathology at narrow levels of specificity (eg, symptoms or traits) as well as higher-order dimensions, including subfactors, spectra, and superspectra, that represent commonalities among the lower-order features. Psychotic disorders are linked to multiple higher-order spectra, such as thought disorder (ie, positive symptoms) and detachment (ie, negative symptoms), which have been validated using a variety of measures.^24,25 A meta-analysis of the structural evidence for the HiTOP model also indicated that thought disorder and detachment spectra converge for a higher-order psychosis superspectrum.²⁶ Finally, all psychopathology dimensions contribute to an overarching general factor of psychopathology. Most studies on the neural response to rewards and psychotic disorders have investigated categorical disorders or took a dimensional approach but examined a single hierarchical level of psychopathology (eg, thought disorder and detachment spectra). It is possible the RewP might demonstrate more consistent relationships with dimensional representations of psychosis than categorical diagnoses have done. In addition, a blunted RewP might be associated with multiple hierarchical dimensions, such as the detachment spectra and the higher-order psychosis superspectrum. If so, it would be informative to know which hierarchical dimension demonstrates the strongest relationship with the RewP.

The RewP is a time-domain component that reflects the combination of overlapping neurophysiological processes. Time-frequency analysis can disentangle frequency contributions, including reward-related delta and loss-related theta activity that is only moderately correlated with the RewP.^27,28 Although the exact functions are unknown, some research suggests that reward-related delta indicates motivational relevance or salience of feedback²⁹ whereas loss-related theta represents the processing of error feedback.³⁰ Time-frequency analyses in non-psychotic samples have indicated that blunted reward-related delta activity is associated with depressive disorders^27,31 and greater depressive symptoms,³² while one investigation found that greater loss-related theta activity was associated with depression.²⁸ In psychotic disorders, both blunted reward-related delta activity³³ and loss-related theta activity³⁴ have been found in schizophrenia (SZ), but at least one investigation found no differences between SZ cases and controls in either reward-related delta or theta activity.³⁵ However, time-frequency investigations have also involved small patient sample sizes (n’s = 17-54) and limited consideration of comorbid psychopathology.

The present study examined the association between multiple EEG responses to reward and psychosis dimensions. We examined the association between time-domain and time-frequency neural responses to reward and multiple lower- (thought disorder, detachment) and higher-order (psychosis, general psychopathology) psychosis-relevant dimensions. We also examined a negative emotionality dimension, given the comorbidity between psychotic and internalizing disorders³⁶ and the association between the RewP and depression.¹⁶ We hypothesized that more blunted neural responses to reward would be associated with greater psychosis dimension scores, including both lower-order detachment and higher-order psychosis. We hypothesized that the neural responses to reward would be positively associated with executive functions. Finally, we tested whether the association between the neural responses to reward and psychosis dimensions was mediated by executive functions.

Methods

Participants

The Suffolk County Mental Health Project^37,38 is an epidemiological longitudinal study of first-admission psychosis. Between 1989 and 1995, participants were recruited from the 12 psychiatric inpatient units of Suffolk County, NY. The eligibility criteria were first admission within 6 months, clinical evidence of psychosis, ages 15-60, IQ > 70, proficiency with English, resident of Suffolk County, and no apparent general medical etiology. Follow-up assessments were conducted 6 months after first admission and again in years 2, 4, 10, and 20. The study was approved annually by the institutional review boards of Stony Brook University and the participating hospitals. Participants provided written informed consent.

The overall sample of participants who were hospitalized for a psychotic disorder at the baseline assessment was 628. The present study focused on the sample of participants who completed the 20-year follow-up assessment. A total of 82 cases died before the 20-year follow-up assessment, and 318 cases did not complete the 20-year follow-up assessment. The final sample at the 20-year follow-up assessment (years since first admission M = 19.58, SD = 1.98) included 228 individuals with a lifetime history of a psychotic disorder. Of the 228 cases, 90 had SZ, 81 had a mood disorder with psychosis, 29 had schizoaffective disorder (SZA), 16 had other psychotic disorders, and 12 had substance-induced psychosis. Of the 228 cases that completed the 20-year follow-up assessment, a total of 56 cases did not complete the doors task and 3 cases did not complete the psychiatric measures, resulting in a final sample of 169 cases. See Supplemental Materials for a comparison of baseline demographics and clinical characteristics for cases who were and were not included in analyses.

A sample of 260 never psychotic comparison group was also recruited at the 20-year follow-up assessment using random digit dialing to zip codes where the psychotic disorder group resided and was matched in age, sex, and race. A total of 65 comparisons did not complete the doors task, and 2 comparisons did not complete the psychiatric measures, resulting in a final sample of 193 comparisons.

Measures

Participants completed interviews, self-report questionnaires, and neuropsychological tests that were conducted by master’s level mental health professionals. Medical records and interviews with significant others were also obtained at this assessment, and this information was compiled to complete several rating scales.

Interviews

Brief Psychiatric Rating Scale (BPRS)

The BPRS³⁹ is a 19-item clinician-rated interview on a 7-point rating scale ranging from 1 (absent) to 7 (very severe). The present study used four domain scores that were calculated by averaging together select items: anxious-depression (anxiety, depressed mood, guilt feelings), detachment (blunted affect, emotional withdrawal), hostile-suspiciousness (hostility, suspiciousness, uncooperativeness), and thinking disturbance (conceptual disorganization, hallucinatory behavior, unusual thought content).

Hamilton Rating Scale for Depression (HRSD)

The HRSD⁴⁰ is a 21-item clinician-rated interview on a 5-point rating scale ranging from 0 and 4, with higher scores indicating greater symptom severity. The present study used the HRSD total score based on the first 17 items, which measure severity of depressive symptoms.

Scale for the Assessment of Negative Symptoms (SANS) and Scale for the Assessment of Positive Symptoms (SAPS)

The SANS^41,42 is a 19-item clinician-rated interview organized into five domains and a global rating for each domain. The SAPS⁴³ is a 31-item clinician-rated interview covering four symptom domains as well as global ratings. The SANS and SAPS items are rated on a 6-point scale ranging from 0 (none) to 5 (severe) to reflect the past month. The item “inattention during mental test” was excluded from analyses due to a high level of missing data, and we only used individual items, not global ratings. The SANS was subdivided into inexpressivity and avolition subscales, and the SAPS was subdivided into reality distortion and disorganization subscales based on prior factor analysis.⁴⁴

Self-report

Big Five Inventory (BFI)

The BFI⁴⁵ is a 44-item self-report measure of the Big Five personality traits. On each item, participants indicate how much they agree with statements on a scale ranging from 1 (Disagree Strongly) to 5 (Agree Strongly). The present study used the traits neuroticism and introversion (which was calculated by taking the inverse of extraversion).

Schedule for Nonadaptive and Adaptive Personality (SNAP)

The SNAP⁴⁶ is an omnibus self-report true/false questionnaire that assesses both adaptive and nonadaptive aspects of personality. Participants completed the items that contributed to three maladaptive personality traits: detachment (ie, interpersonal and emotional distance), eccentric perceptions (ie, odd or unusual beliefs, cognitions, and perceptual experiences), and mistrust (ie, suspiciousness, alienation, and feelings of injustice).

Cognition

Cognition was assessed using multiple neuropsychological tests.^47,48 Trail Making A was administered as a measure of simple attention and processing speed. Trail Making B, Stroop Color-Word, and Letter-Number Sequencing were administered as measures of executive functions. On the Stroop Color-Word and Letter-Number Sequencing tests, higher scores indicate better performance. Trail Making A and B performance is based on time (max 180 s), such that lower scores indicate better performance. Therefore, we calculated the inverse of Trail Making A and B (ie, |180-Trails A|, |180-Trails B|). This computation allowed for higher values to indicate better performance for all measures of simple attention and executive functions. An executive functions composite score was calculated by z-scoring each of the executive function measures (Trails B, Stroop Color-Word, Letter-Number Sequencing) across all participants and averaging together the z-scores. The Wechsler Adult Intelligence Scale–Revised (WAIS-R vocabulary) subtest was administered as an index of crystallized intelligence.^49,50

Procedure

Experimental Task

The doors task was administered using Presentation software (Neurobehavioral Systems, Inc., Albany, CA, USA). On each trial, participants viewed two identical doors and were told that one door contained a monetary prize ($0.50), and the other door contained a monetary loss ($0.25). The difference between monetary wins and losses was based on the general subjective value of losses being twice that of wins.^51,52 Next, participants selected a door using the computer mouse. After selecting, the doors disappeared and a fixation cross appeared for 1000 ms, and then feedback was presented for 2000 ms—an upward green arrow indicating a monetary win or a downward red arrow indicating a monetary loss. The feedback stimulus was followed by a fixation cross presented for 1500 ms, immediately followed by the message “Click for next round”. This prompt remained on the screen until the participant responded with a button press to initiate the next trial. There were an equal number of win and loss outcomes (20 each, 40 total trials) that were presented in a pseudorandom order, such that trials were presented in a random order and no trial type (win or loss) could occur more than twice in a row.

E‌EG Recording and Data Reduction

Continuous EEG was recorded using an elastic cap with 34 (standard 32 electrodes plus FCz and Iz) sintered Ag/AgCl electrodes placed according to the 10/20 system. Electrooculogram was recorded using four additional facial electrodes: two placed approximately 1 cm outside of the right and left eyes and two placed approximately 1 cm above and below the right eye. Data was recorded using the ActiveTwo system. The EEG was digitized with a sampling rate of 1024 Hz using a low-pass fifth-order sinc filter with a half-power cutoff of 204.8 Hz. A common mode sense active electrode producing a monopolar (nondifferential) channel was used as a recording reference.

The EEG data were analyzed using BrainVision Analyzer 2.2 (Brain Products, Gilching, Germany). Data were referenced offline to the average of left and right mastoids; band-pass filtered (0.1-30 Hz), and corrected for eye movement artifacts.⁵³ For the time-domain analyses, a semiautomatic procedure was employed to detect and reject artifacts. The criteria applied were a voltage step of more than 50 μV between sample points, a voltage difference of 300 μV within a trial, and a maximum voltage difference of less than 0.5 μV within 100-ms intervals. These intervals were rejected from individual channels in each trial. Visual inspection of the data was then conducted to detect and reject the remaining artifacts. Feedback-locked epochs were extracted with a duration of 1200 ms, including a 200 ms pre-stimulus and 1000 ms post-stimulus interval. The 200 ms pre-stimulus interval was used as the baseline. Figure 1 displays the grand average ERP waveforms and scalp distribution for the neural response to reward feedback. The ERP signal was quantified as the mean activity from 250-350 ms after the onset of feedback at a pooling of electrodes Fz, FCz, FC1, and FC2, separately for win and loss feedback. The split-half reliability of odd and even trials, corrected using the Spearman-Brown Prophecy Formula, was .87 for the ERP response to win feedback and .83 for the ERP response to loss feedback. The RewP was isolated by regressing the ERP response to win feedback on the ERP response to loss feedback and saving the unstandardized residual, with higher residual scores indicated larger RewP. The present study utilized a residual (instead of a difference score) to better isolate the time-domain neural response to win.

Figure 1.

Event-related potential waveform (left) shaded region is where the response to the feedback was scored at a pooling of electrodes Fz, FCz, FC1, and FC2. The scalp distribution (middle) shows the average activity for the win-loss difference (ie, the RewP). Warmer colors on the scalp distribution indicate a larger neural response to win feedback. Time-frequency plot (right) displays dB activity for the win-loss difference at a pooling of electrodes Fz, FCz, FC1, and FC2. Warmer colors on the time-frequency plot indicate a larger neural response to win feedback and cooler colors indicate a larger neural response to loss feedback. Ms = milliseconds; RewP = reward positivity.

For the time-frequency analyses, an automatic procedure was employed to detect and reject artifacts. The criteria applied were a voltage step of more than 50 μV between sample points, a voltage difference of 175 μV within a trial, and a maximum voltage difference of less than 0.5 μV within 100-ms intervals. These intervals were rejected from individual channels in each trial. Next, the data was segmented from -1500 to 1500 ms for each trial, and a set of complex Morlet wavelets was applied that increased by 20 logarithmic steps from 0.5 to 20 Hz. An estimate of frequency band-specific power at each time point was computed by squaring the absolute value of the complex signal. Power was normalized using a decibel (dB) transformation where the baseline activity was the average power at each frequency condition from -200 to 0 ms prior to feedback onset. Figure 1 also displays the time-frequency plot for the win-loss difference. Delta activity was extracted via wavelet layer 10 (central frequency = 2.870 Hz) from 150 to 250 ms at a pooling of electrodes Fz, FCz, FC1, and FC2, separately for win and loss feedback. Theta activity was extracted via wavelet layer 13 (central frequency = 5.138 Hz) from 300 to 400 ms at a pooling of electrodes Fz, FCz, FC1, and FC2, separately for win and loss feedback. The split-half reliability of odd and even trials, corrected using the Spearman-Brown Prophecy Formula, was .68 for delta activity to win feedback and .62 for the delta activity to loss feedback, and .53 for theta activity to win feedback and .76 for theta activity to loss feedback. Similar to the time-domain RewP, the time-frequency responses were measured using residuals. Delta activity to win was isolated by regressing the layer 10 wavelet response to win feedback on the layer 10 wavelet response to loss feedback and saving the unstandardized residual. Higher residual scores indicated greater delta activity to win. Theta activity to loss was isolated by regressing the layer 13 wavelet response to loss feedback on the layer 13 wavelet response to win feedback and saving the unstandardized residual. Higher residual scores indicated greater theta activity to loss.

Data Analysis

We first examined task effects to determine whether the guessing task elicited the expected time-domain RewP and time-frequency delta activity to win and theta activity to loss. We conducted a repeated measures analysis of variance (ANOVA) with outcome (win vs loss) as the within-subject factor and the neural response as the dependent variable, with separate analyses conduced for the RewP, delta activity, and theta activity. Across all participants, we tested whether the RewP and delta activity were greater to win compared to loss feedback, and theta activity was greater to loss compared to win feedback.

Next, we conducted categorical group analyses for each neural response to reward. Group analyses involved participants with SZ or SZA (n = 86), mood disorder with psychosis (n = 62), and comparison controls (n = 193). Group analysis did not include participants with other psychotic disorders or substance-induced psychosis due to small sample sizes. We conducted a one-way ANOVA and included group (SZ or SZA, mood disorder with psychosis, comparison controls) as the between-subjects factor.

We also examined multiple psychosis-relevant dimensions. First, we examined thought disorder, detachment, and negative emotionality dimensions. Next, the detachment and thought disorder composites were averaged together to create a psychosis dimension. Finally, the detachment, thought disorder, and negative emotionality dimensions were averaged together to create a general psychopathology dimension.

Thought disorder, detachment, and negative emotionality composites were primarily based on a previous investigation with the same sample.⁵⁴ See Supplemental Materials for bivariate correlations between the psychiatric measures. The SANS avolition, SANS inexpressivity, BPRS detachment, SNAP detachment, and BFI introversion scores demonstrated positive correlations. To create a detachment composite, each measure was first z-scored, and then those z-scores were averaged together. Similarly, the SAPS reality distortion, SAPS disorganization, BPRS thinking disturbance, BPRS hostile-suspiciousness, SNAP eccentric perceptions, and SNAP mistrust scores demonstrated positive correlations. To create a thought disorder composite, each measure was first z-scored and then those z-scores were averaged together. Finally, the BPRS anxious-depression, HRSD, and BFI neuroticism scores demonstrated positive correlations. To create a negative emotionality composite, each measure was first z-scored, and then those z-scores were averaged together.

We conducted Pearson’s correlations to examine bivariate correlations between demographics, psychiatric measures, cognition test scores (ie, no demographic corrections applied to the test scores), and the neural responses to reward. Simple attention was calculated as the inverse of the Trails A test score (ie, |180-Trails A|), such that higher scores indicate better simple attention. Executive functions were calculated as the z-scored average of the inverse of the Trails B test score (ie, |180-Trails B|), Letter-Number Sequencing test score, and Stroop Color-Word test score, such that higher scores reflect better executive functioning. See Supplemental Materials for correlations between executive function measures. Next, we used linear regression to examine the association between the neural responses to reward residuals (ie, RewP residual, delta activity to win residual, or theta activity to loss residual) and the psychiatric measures. All linear regression models included demographics as covariates, including age, sex (0 = female, 1 = male), socioeconomic status using the Hollingshead Index rating of occupational prestige (1 = large business owner/major professional/executive to 8 = not working),⁵⁵ and race (0 = White, 1 = Non-White), and separate analyses were conducted for each neural response to reward (RewP residual, delta activity to win residual, theta activity to loss residual). For the linear regression analyses, we included the psychiatric measures as the independent variables since there was more than one variable for most analyses, while the neural response to rewards always involved a single variable. This approach also allowed us to test whether the different psychiatric dimensions (eg, detachment, thought disorder, negative emotionality) were independently associated with the neural response to reward. We conducted a linear regression with detachment, thought disorder, and negative emotionality dimensions as independent variables and the neural response to reward as the dependent variable. Next, we conducted analogous linear regressions that included psychosis and negative emotionality dimensions as the independent variables. Finally, we conducted a linear regression that included the general psychopathology dimension as the independent variable. Sensitivity analyses were conducted for the linear regressions involving n = 362, α = .05, three predictors, and a two-tailed test. Results indicated the sample size is adequate for the detection of a Cohen’s d effect size of 0.07.

In instances where there were relationships between a neural response to reward, cognition, and psychiatric dimension, we conducted a mediational model and tested for indirect effects of the neural response to reward through cognition on the psychiatric dimension. Mediation analyses were completed using the SPSS PROCESS Macro model 4.⁵⁶ The output of this model includes an (a) estimate and P values of the direct effects between the neural response to reward and cognition, (b) estimate and P values of the direct effects between cognition and the psychiatric dimension, (c) estimate and P value of the direct effects between the neural response to reward and the psychiatric dimension, and (c′) a bootstrap estimate (using 5000 bootstrap samples) of the indirect effect between the neural response to reward and the psychiatric dimension and 95% confidence intervals (CI) for the population value of the indirect effects. Confidence intervals that do not include zero indicate a significant indirect effect (P < .05). Analyses included age, socioeconomic status, sex, race, and crystalized intelligence (WAIS-R vocabulary) as covariates, and all variables were z-scored prior to inclusion in analyses so that coefficients represented standardized estimates. All analyses were conducted in IBM SPSS Statistics, Version 27.0 (Armonk, NY, USA).

Results

Doors task effects

As expected, the time-domain response to win feedback (M = 13.74, SD = 7.39) was larger than to loss feedback (M = 10.87, SD = 6.36), t = 12.90, P < .001, supporting the presence of the RewP. Also as expected, the time-frequency delta activity to win feedback (M = 2.17, SD = 2.21) was larger than to loss feedback (M = 1.65, SD = 2.11), t = 5.06, P < .001, and the time-frequency theta activity to loss feedback (M = 3.00, SD = 2.74) was larger than to win feedback (M = 1.65, SD = 1.99), t = -10.43, P < .001. As shown in Table 1, there were positive correlations between the time-domain RewP, time-frequency delta activity to win, and time-frequency theta activity to loss.

Table 1.

Descriptive Statistics and Bivariate Correlations

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
1.   Age	−
2.    Sex	.03	−
3.    Socioeconomic status	−.03	−.05	−
4.    Race	−.01	.01	.02	−
5.    Simple attention	−.09	−.11	−.16	−.18	−
6.    Executive function	−.04	−.05	−.26	−.22	.63	−
7.    Crystallized intelligence	.18	.09	−.24	−.18	.35	.57	−
8.    Group	−.26	.06	.06	.07	−.37	−.40	−.24	−
9.    Detachment	−.10	.12	.17	.11	−.41	−.45	−.23	.49	−
10. Thought disorder	−.10	.09	.17	.01	−.28	−.38	−.20	.48	.51	−
11. Negative emotionality	−.06	−.15	.13	−.05	−.17	−.15	−.06	.33	.50	.43	−
12. Psychosis	−.11	.12	.20	.07	−.41	−.48	−.25	.56	.87	.86	.54	−
13. General psychopathology	−.10	.02	.19	.03	−.36	−.40	−.20	.53	.83	.79	.81	.93	−
14. Reward positivity	−.11	.06	.12	−.04	.04	.04	−.03	−.03	−.02	−.02	.04	−.03	.00	−
15. Delta activity to win	−.15	.08	−.04	.03	.15	.18	−.03	−.04	−.16	−.11	−.06	−.15	−.13	.39	−
16. Theta activity to loss	−.08	.13	.00	−.04	.15	.17	.06	−.14	−.05	−.04	.00	−.06	−.04	.28	.13	−
M	49.87	−	4.19	−	147.21	0.05	21.31	−	0.01	0.01	0.01	0.01	0.01	0.00	0.00	0.00
SD	9.02	−	1.78	−	13.42	0.81	4.77	−	0.80	0.77	0.85	0.68	0.65	4.14	1.79	2.38
Min	31.74	−	1	−	88	−2.55	6	−	−1.05	−0.58	−1.10	−0.78	−0.87	−11.65	−4.98	−7.11
Max	71.30	−	8	−	167	1.68	28	−	3.99	3.14	4.22	2.57	2.64	16.82	5.70	9.22

Sex (0 = female, 1 = male), race (0 = White, 1 = Non-White), and group (0 = never psychotic comparison participants, 1 = psychotic disorder cases) were binary variables. Simple attention reflects the inverse of the Trails A raw score (ie, |180-Trails A raw score|), such that higher scores indicate better simple attention. Executive function reflects the z-scored average of the inverse of the Trails B raw score (ie, |180-Trails B raw score|), Letter-Number Sequencing raw score, and Stroop Color-Word raw score, such that higher scores reflect better executive function. Crystallized intelligence reflects the WAIS-R vocabulary raw score, such that higher scores reflect better general verbal ability. Detachment, thought disorder, and internalizing dimensions were calculated by z-scoring and averaging together the relevant interview and self-report subscales. Psychosis was calculated by averaging together the detachment and thought disorder dimensions. General psychopathology was calculated by averaging together the detachment, thought disorder, and internalizing dimensions. All bolded results are significant at P < .05. All italicized correlations are significant after correcting for false discovery rate.⁵⁷

Bivariate Correlations

Table 1 displays bivariate correlations between demographics, psychiatric dimensions, cognitive measures, and the neural responses to reward. Greater detachment, thought disorder, psychosis, and general psychopathology dimensions were associated with worse simple attention, executive functions, and general cognitive ability. Greater negative emotionality was associated with worse simple attention and executive functions but was not associated with general cognitive ability. As shown in Table 1, the RewP was not associated with simple attention or executive functions. In contrast, greater delta activity to win and theta activity to loss were associated with better simple attention and executive functions. See Supplemental Materials for scatterplots of the association between the neural responses to reward and both cognition and psychiatric dimensions. See Supplemental Materials for results testing whether group moderated the association between the neural response to reward and cognition.

Categorical Analyses

Figure 2 displays graphs for the categorical analyses. Results indicated that groups did not differ in the RewP, F(2, 338) = 0.59, P = .55. In contrast, groups differed in delta activity to win, F(2, 226) = 4.04, P = .02, Cohen’s d = .27, such that participants with SZ or SZA had smaller delta activity to win compared to participants with mood disorder with psychosis (P = .02) and comparisons (P = .001), but there were no differences between participants with mood disorder with psychosis and comparisons (P = .76). Groups also differed in theta activity to loss, F(2, 336) = 4.53, P = .01, Cohen’s d = .33, such that participants with SZ or SZA had smaller theta activity to loss compared to comparisons (P = .004), but there were no differences between participants with SZ or SZA and participants with mood disorder with psychosis (P = .37) and participants with mood disorder with psychosis and comparisons (P = .12).

Figure 2.

Bar graphs displaying mean activity (and standard error) for group comparison of the reward positivity (top), delta activity to win (middle), and theta activity to loss (bottom). SZA = schizoaffective disorder, SZ = schizophrenia. P < .05, ^**P < .01, ^***P < .001.

Dimensional Analyses

Table 2 displays the results of the linear regressions. For the time-domain analyses, the RewP was not associated with the thought disorder, detachment, or negative emotionality dimensions. However, a more blunted RewP was associated with greater psychosis dimension scores. There was no association between the RewP and the general psychopathology dimension.

Table 2.

Linear Regressions

	Reward positivity					Delta activity to win					Theta activity to loss
	R	β	F or t	P	95% CI	R	β	F or t	P	95% CI	R	β	F or t	P	95% CI
Model 1	.21		2.36	.023		.27		3.85	<.001		.20		1.98	.057
Detachment		−.08	−1.23	.219	−1.107, 0.255		−.20	−2.98	.003	−0.730, −0.149		−.10	−1.43	.154	−0.678, 0.108
Thought disorder		−.07	−1.05	.296	−1.024, 0.312		−.08	−1.21	.228	−0.460, 0.110		−.06	−0.97	.334	−0.575, 0.196
Negative emotionality		.10	1.53	.128	−0.138, 1.092		.09	1.36	.176	−0.081, 0.444		.09	1.37	.173	−0.109, 0.602
Model 2	.21		2.76	.012		.26		4.29	<.001		.19		2.30	.034
Psychosis		−.13	−1.97	.049	−1.560, −0.002		−.23	−3.61	<.001	−0.943, −0.278		−.14	−2.07	.039	−0.923, −0.024
Negative emotionality		.10	1.52	.129	−0.138, 1.085		.08	1.26	.208	−0.094, 0.429		.09	1.35	.179	−0.112, 0.595
Model 3	.19		2.56	.027		.23		3.89	.002		.17		2.08	.068
General psychopathology		−.04	−0.70	.483	−0.907, 0.430		−.15	−2.78	.006	−0.693, −0.119		−.05	−0.99	.323	−0.580, 0.192

Each model represents a distinct linear regression, and separate linear regressions were conducted for each neural response to reward. All models included age, socioeconomic status, sex, and race as covariates. Psychosis was calculated by averaging together the detachment and thought disorder dimensions. General psychopathology was calculated by averaging together the detachment, thought disorder, and negative emotionality dimensions. All bolded results are significant at P < .05.

For time-frequency analyses, more blunted delta activity to win was associated with greater detachment, but not thought disorder or negative emotionality dimension scores. More blunted delta activity to win was associated with greater psychosis dimension scores. More blunted delta activity to win was associated with greater general psychopathology dimension scores. In comparing all three results, delta activity to win demonstrated the strongest association with the psychosis dimension. Theta activity to loss was not associated with thought disorder, detachment, or negative emotionality dimensions. However, more blunted theta activity to loss was associated with greater psychosis dimension scores. There was no association between theta activity to loss and the general psychopathology dimension. See Supplemental Materials for analyses testing whether group moderated the associations between the neural responses to rewards and psychopathology dimensions.

Mediational Analyses

Mediational analyses were conducted for the neural response to reward, cognition, and both delta activity to win and theta activity to loss. Delta activity to win and theta activity to loss were both associated with the psychosis dimension; therefore, the mediational analyses focused on the psychosis dimension. Separate mediational models were conducted for delta activity to win and theta activity to loss, and to determine the independence of the results, each model also included the other neural response as a covariate. As shown in Figure 3, results indicated an indirect effect of delta activity to win through executive functions on the psychosis dimension. Similarly, results indicated an indirect effect of theta activity to loss through executive functions on the psychosis dimension.

Figure 3.

Simple attention was measured by the inverse of trails a (ie, |180-trails a|), such that higher values indicate better simple attention. Executive function was computed as the average of z-scored inverse of Trails B (ie, |180-trails B|), letter-number sequencing, and Stroop color-word, such that higher values indicated better executive function. Psychosis reflects the psychosis composite residual after controlling for the negative emotionality composite. Analyses included age, socioeconomic status, sex, race, crystallized intelligence (WAIS-R vocabulary), and the other time-frequency response to reward as covariates. All variables were z-scored prior to inclusion in analyses so that coefficients represented standardized estimates. For indirect effects, confidence intervals that do not include zero indicate a significant indirect effect (P < .05). ^*P < .05, ^**  P < .01, ^***P < .001.

Discussion

The present study examined the association between EEG time-domain and time-frequency neural responses to reward and psychosis dimensions. A categorical group comparison indicated no differences in the RewP but participants with SZ or SZA had more blunted delta activity to win and theta activity to loss. In contrast, all neural responses to monetary feedback were associated with psychosis dimensions. Specifically, a more blunted RewP, delta activity to win, and theta activity to loss was associated with greater scores on the psychosis (ie, composite of positive and negative symptoms) dimension. In addition, more blunted delta activity to win was also associated with greater scores on the lower-order detachment (ie, negative symptoms) and higher-order general psychopathology dimensions. The time-frequency neural responses to reward were positively associated with simple attention and executive functions. Finally, executive functions mediated the relationship between the time-frequency neural responses to reward and the psychosis dimension. Overall, the present study demonstrates that blunted EEG neural responses to reward are associated with both lower-order and higher-order dimensional features of psychosis. The study also highlights the importance of cognitive deficits when examining the neural response to reward and psychosis.

A recent meta-analysis indicated the RewP did not demonstrate a consistent relationship with psychotic disorders.¹⁹ Similarly, the present study did not find group differences between participants with SZ or SZA, mood disorder with psychosis, or comparisons in the RewP. However, greater dimensional psychosis scores were associated with a more blunted RewP. There are several factors that have likely contributed to the mixed literature on EEG neural response to reward and psychosis. First, the categorical assessment of psychosis is problematic due to the heterogeneity of psychotic disorders. HiTOP-informed neuroscience research has been shown to be more reliable and robust than results from studies using categorical disorders.^58-60 A continued focus on the neurobiological origins of categorical psychotic disorders will likely continue to limit progress on this important area of research. Second, brain-psychopathology relationships do not share any method variance, and it is normal to expect small effect sizes.⁶¹ Previous investigations have been limited to small patient samples and thus underpowered analyses. In the present study, sensitivity analyses indicated the sample was powered to detect small effect sizes. Third, there was no bivariate relationship between the RewP and the psychosis dimension. However, in the linear regression, which controlled for comorbid negative emotionality, there was an inverse relationship between the RewP and the psychosis dimension. This pattern is consistent with the presence of mutual suppressor effects,⁶² which has been previously found between the RewP and other psychopathology dimensions.⁶³ Psychotic and depressive disorders contain common features of anhedonia and low motivation, and it is possible that the RewP indexes a shared mechanism that contributes to high levels of comorbidity. Finally, it is important to highlight that the RewP was not related to categorical psychotic disorders, but it was related to dimensional psychosis scores. Other EEG measures of brain functioning (eg, P3⁶⁴) have still been shown to be related to categorical representations of psychosis, and even the RewP has been associated with other heterogenous categorical diagnostic definitions, such as depression.¹⁶ Taken together, this pattern of results suggests the RewP might not be as strongly related to psychosis.

Time-frequency neural activity to rewards might more optimally represent the neural dysfunction in psychosis. A small number of previous investigations indicated blunted reward-related delta activity³³ and loss-related theta activity³⁴ in SZ. The present study found that SZ and SZA were characterized by more blunted delta activity to win and theta activity to loss, and similar results were obtained for the psychosis dimension. In addition, delta activity to win was independently associated with lower-detachment as well as higher-order general psychopathology dimensions, although the relationship was the strongest with the psychosis dimension. Time-frequency analysis allows for the measurement of both phase and non-phased-locked brain activity, and it is possible that the non-phased-locked brain activity indexes more relevant aspects of the neural dysfunction in psychosis. Future studies should always consider examining both time-domain and time-frequency representations of the neural response to rewards in relation to psychosis, particularly since both quantifications can be measured from the same neural signal.

The present study provides novel evidence that cognition mediates the relationship between time-frequency neural responses to reward and the psychosis dimension. Time-frequency delta and theta activity have been linked to various executive functions, including cognitive control, response inhibition, and working memory.^65,66 Cognitive deficits have also been implicated in aspects of psychosis presentation; for example, lower cognitive capacity has been linked to inexpressivity⁶⁷ and deficits maintaining contextual information to predict outcomes, assign value, and compute effort have been linked to avolition.^68,69 Time-frequency representations of the neural response to reward, compared to the time-domain RewP, demonstrated stronger associations with psychosis dimensions, including both lower-order detachment and higher-order general psychopathology. It is possible this result was due to the time-frequency measurements better capturing the neural mechanisms of cognitive deficits. It is important to note that these analyses were cross-sectional, and, when possible, future studies should examine whether changes in time-frequency neural responses to reward precede changes in cognition. If so, time-frequency neural responses to reward might be ideal targets for efforts to improve cognitive deficits.⁷⁰

The present study has important clinical implications. Across the same Suffolk County Mental Health project sample, other EEG-based neural measures have been associated with psychosis, including performance monitoring and the error-related negativity,^44,71,72 attention and the P300,⁷³ semantic processing and the N400,⁷⁴ and emotional reactivity and the late positive potential.⁷⁵ Collectively, these results suggest there may be a panel of neural deficits that best characterize psychosis. There has been an increased interest in moving beyond behavioral and symptom-based representations of psychotic illness, and it is possible these neural deficits, coupled with other critical sources of information (eg, demographics, culture, environment), might better represent the core neurobiological and psychological mechanisms that would be better targets of early identification, prevention, and treatment. These studies have primarily involved cross-sectional analyses, which limit the ability to understand causal relationships between the neural response to reward, cognition, and psychosis dimensions. However, measures like the RewP and delta activity to reward have been shown to predict other forms of psychopathology, such as depression.^31,76 It will be critical to determine whether the same type of prospective relationships exist between these neural measures and psychotic illness.

The present study had several strengths, including a large sample of cases that transcended the psychotic spectrum, multi-method (clinician ratings, self-report questionnaires) assessment of psychopathology, and inclusion of both time-domain and time-frequency representations of the neural response to reward. The present study also contained several limitations that warrant consideration. First, the study contained a limited assessment of comorbid psychopathology. Future studies should consider other psychopathology dimensions that have also been associated with the neural response to rewards (eg, externalizing psychopathology). Second, cases were from a longitudinal cohort where psychotic illness had progressed for 20 years, and the results might differ for those experiencing a first psychotic experience or more acute psychotic illness. Third, the sample was limited to middle- to late-adulthood, and it is unclear if the results generalize to younger individuals. Finally, the assessment of simple attention was limited to a single measure (Trails A).

In conclusion, the present study demonstrates that EEG-based neural responses to reward are associated with dimension representations of psychosis—and indicates that cognitive deficits likely play an important role. Additional research is needed to examine other aspects of reward processing, such as reward anticipation and the neural response to other types of reward (eg, social reward). Longitudinal evaluations are also needed to better understand the causal relationships between these factors and determine whether a blunted neural response to reward can predict changes in psychotic illness, comorbid pathology, and functional outcomes.

Funding

This work was supported by the National Institutes of Health (MH044801 to E.J.B. and MH110434 to R.K.).

Conflicts of Interest

None declared.