Double dissociation between P300 components and task switch error type in healthy but not psychosis participants

Abstract

Event-related potentials (ERPs) during oddball tasks and the behavioral performance on the Penn Conditional Exclusion Task (PCET) measure context-appropriate responding: P300 ERPs to oddball targets reflect detection of input changes and context updating in working memory, and PCET performance indexes detection, adherence, and maintenance of mental set changes. More specifically, PCET variables quantify cognitive functions including inductive reasoning (set 1 completion), mental flexibility (perseverative errors), and working memory maintenance (regressive errors). Past research showed that both P300 ERPs and PCET performance are disrupted in psychosis. This study probed the possible neural correlates of 3 PCET abnormalities that occur in participants with psychosis via the overlapping cognitive demands of the two study paradigms. In a two-tiered analysis, psychosis (n = 492) and healthy participants (n = 244) were first divided based on completion of set 1 – which measures subjects’ ability to use inductive reasoning to arrive at the correct set. Results showed that participants who failed set 1 produced lower parietal P300, independent of clinical status. In the second tier of analysis, a double dissociation was found among healthy set 1 completers: frontal P300 amplitudes were negatively associated with perseverative errors, and parietal P300 was negatively associated with regressive errors. In contrast, psychosis participants showed global P300 reductions regardless of PCET performance. From this we conclude that in psychosis, overall activations evoked by the oddball task are reduced while the cognitive functions required by PCET are still somewhat supported, showing some level of independence or compensatory physiology in psychosis between neural activities underlying the two tasks.

1. Introduction

Syndromes with psychosis show deviations in cognition (Ermel et al., 2017; Frydecka et al., 2014; Hill et al., 2013) and neurophysiology (Ethridge et al., 2012; Johannesen et al., 2013; Kaur et al., 2011). Each edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM) defines criteria for psychosis subgroups differently; on many physiology-based measures, separations between DSM-defined subgroups, for example between schizophrenia and schizoaffective disorder, are less pronounced than the differentiation between psychosis participants and healthy participants (Aykut et al., 2017; Martin et al., 2015; Shepherd et al., 2015; Velthorst et al., 2017). The lack of biology-based boundaries motivates the search for physiological markers sensitive to psychosis manifestations, rather than to psychosis subgroups.

Among the neurophysiological phenotypes associated with psychosis, reduction in P300 event-related potential (ERP) is one of the most replicated and consistent in the literature. In the laboratory, it is often evoked by the auditory oddball paradigm, which consists of a series of identical, standard tones occasionally interrupted by infrequent, higher pitched, target tones. The P300 ERP peaks along the midline of the scalp roughly 260-600 ms after the presentation of a target tone, and it is specific to target tones; the time range is relatively wide because P300 is a composite of multiple waveforms occurring throughout the brain. This ERP is considered a marker for psychosis because diminished P300 amplitude is observed in schizophrenia participants (SZ) (Bestelmeyer et al., 2009; Dichter et al., 2006; Merrin et al., 2006) and their first-degree relatives (Lebedeva et al., 2008; Sevik et al., 2011; Sponheim et al., 2006); similar deficits are reported in bipolar disorder with psychosis (BDP) participants (Bestelmeyer et al., 2009; Hamm et al., 2013; Schulze et al., 2008) and their relatives (Schulze et al., 2008). Though fewer in number, publications of schizoaffective disorder (SAD) also show reduced P300 amplitude (Kaur et al., 2011; Mathalon et al., 2010; Parker et al., 2021). These findings show a common biological deficiency in psychosis and relatives regarding the brain’s response to non-habituated stimuli.

Source analyses of healthy human participants show that auditory P300 originates from a combination of brain regions: dorsolateral prefrontal (DLPFC), orbital frontal (OFC), anterior cingulate cortices (ACC) (Halgren et al., 1998; Huang et al., 2005; Wang et al., 2003), lateral and inferior parietal cortices (He et al., 2001; Huang et al., 2005; Turetsky et al., 1998), medial temporal cortex, hippocampus, and amygdala (Baudena et al., 1995; Hegerl and Frodl-Bauch, 1997; Wang et al., 2003). That is, P300 is a result of multiple brain circuitries responding to the target stimulus via different electrical and spatial patterns.

P300 reductions are not unique to psychosis syndromes; lower P300 amplitudes are also linked to cognition in various age groups, including childhood and aging (Downes et al., 2017; Groom and Cragg, 2015; Tanriverdi et al., 2013), and particularly in those with impaired working memory (Baillargeon et al., 2012; Gironell et al., 2005; Lazzaro et al., 2013; Yang et al., 2013). Impaired working memory and executive function are often considered features of SZ, SAD, and BDP (Dickinson et al., 2007; Farreny et al., 2016; Ferrarelli et al., 2015; Reichenberg and Harvey, 2007). It is unclear if P300 abnormalities track with psychosis status, cognitive dysfunction, or both.

The Penn Conditional Exclusion Task (PCET) (Kurtz et al., 2004) is a computerized derivative of the Wisconsin Card Sort Test (WCST) that reveals impairments in psychosis groups (Fleck et al., 2008; Liu et al., 2011; Robinson et al., 2016; Silver and Bilker, 2013). The WCST is a complex paradigm originally intended to quantify abstract reasoning during changing contexts (Eling et al., 2008). Participants discover a card sorting rule through trial and error; the rule (also referred to as the “set”) is then maintained until examiner feedback indicates it is no longer relevant, at which point the participant must discover a new set. Perseverative errors capture deficits in cognitive flexibility that enable set switching, and are primarily linked to prefrontal circuitry dysfunction in the literature (Boulougouris et al., 2007; Dias et al., 1997; Johnston et al., 2007; Kim and Ragozzino, 2005; Ragozzino et al., 2003). Compared to WCST, PCET captures one additional error type. Regressive errors, unique to PCET, occur when an individual’s choices revert to a previous set after successful switching. This error type is thought to capture a failure in maintenance of behavioral plans in working memory and has been linked primarily to striatal dysfunction (Ragozzino and Choi, 2004; Ragozzino et al., 2002). In PCET, the two types of errors capture different cognitive deficits, each with corresponding correlates in neural circuitry (Hill et al., 2015; Kurtz et al., 2004).

One more outcome variable from WCST and PCET paradigms is the success in set 1 acquisition. Some individuals are unable to learn and utilize any response set preference, failing to complete even the first set (Stuss et al., 2000; Yamada et al., 1987). This deficit in set acquisition has been shown to correlate with severity of positive and negative psychosis symptoms (Merrin et al., 2006).

The present study examined amplitude of P300, extracted from a 2-stimulus oddball task, in psychosis and healthy participants with different levels of PCET performance. We hypothesized that a frontal component and a central-parietal component will emerge, as discovered in previous publications using a similar cohort (Ethridge et al., 2015). We also hypothesized that (1) frontal response to our oddball targets will be reduced among individuals with high levels of perseverative errors. This might indicate a neurophysiological signature of sensory registration difficulty to environmental changes; (2) parietal response to oddball targets will be reduced in participants with high levels of regressive errors. This might indicate a loss of neural ability to maintain higher level directives, which is critical for goal maintenance and cognitive control.

2. Methods

2.1. Participants

As part of the Bipolar-Schizophrenia Network for Intermediate Phenotypes (B-SNIP) consortium, recruitment, interview, exclusion criteria and testing procedures for the current sample were identical across five sites as described in previous publications (Ethridge et al., 2012; Hill et al., 2015; Tamminga et al., 2014) and fully described in Supplementary Methods. Briefly, clinical evaluation of included subjects is as follows: after providing written informed consent, psychosis and healthy participants were assessed using Structured Clinical Interview for DSM-IV-TR (SCID), including the Global Assessment of Functioning portion (GAF) (American Psychiatric Association and American Psychiatric Association. Task Force on DSM-IV, 2000). Reading Subtest of the Wide Range Achievement Test-IV (WRAT; (Berg et al., 2016)), General Positive and Negative Symptom Scale (PANSS, (Kay et al., 1987)), Young Mania Rating Scale (YMRS, (Young et al., 1978)), and Montgomery-Asberg Depression Rating Scale (MADRS, (Montgomery and Asberg, 1979)) were also administered. Detailed information on assessments performed can be found in Tamminga et al. (2013) (Tamminga et al., 2013).

Participants were included in the psychosis group if they received a diagnosis of schizophrenia (N = 194), schizoaffective disorder (N = 119), or bipolar I with psychosis (N = 179) by the interviewing clinical personnel. Medical history and medication profiles were recorded. Psychosis participants were considered one group in our primary analyses to study the dimensional nature of the association between specific neural responses and specific cognitive deviations. Table 1 summarizes demographic, medication and clinical profiles of the sample. 492 total psychosis participants and 244 healthy comparison subjects were included in this study.

Table 1.

Demographics and clinical information for included subjects.

	Psychosis		Healthy		Statistics
N	492		244
% Female	48 %		53 %		X2 = 1.72
% PCET Set 1 Completers	87 %		96 %		X2 = 14.3***
	Mean	SD	Mean	SD
Age	35.8	12.5	37.3	12.5	t = 1.5
Education (years)	13.5	2.4	15.2	2.5	t = 9.1***
WRATa	97.1	15.6	102.1	13.8	t = 4.2***
GAFb	53.4	13.3	86.1	6.7	t = 35.9***
PCET categories completed	2.2	1.0	2.4	0.8	t = 2.1*
PCET perseverative errors	2.7	3.3	2.2	3.4	t = 2.0*
PCET regressive errors	3.8	6.1	1.9	3.8	t = 4.3***
Clinical scales
MADRSc	10.4	9.2	–	–
YMRSd	6	6.1	–	–
PANSSe Positive	15.6	5.4	–	–
PANSS Negative	14.9	5.5	–	–
PANSS General	31.9	8.8	–	–
PANSS Total	62.4	17.0	–	–
Medications	N	%	N	%
Antipsychoticsf	412	83.7	0	0	X2 = 471.2***
Antidepressants	226	45.9	3	1.2	X2 = 153.8***
Mood Stabilizers	249	50.6	0	0	X2 = 188.7***
Stimulants	33	6.7	1	0.4	X2 = 15.2***
Anxiolytics	148	30.1	4	1.6	X2 = 81.5***
Anticholinergics	69	14	0	0	X2 = 38.4***

^aWide-Range Achievement Test (IV): Reading Test.

^bGlobal Assessment of Functioning portion of Structured Clinical Interview for DSM Disorders (SCID).

^cMontgomery-Asberg Depression Rating Scale.

^dYoung Mania Rating Scale.

^ePositive and Negative Symptoms Scale.

^fFirst- or second-generation.

^*p < 0.05.

^***p < 0.001.

2.2. Cognitive measure

Detailed procedures for PCET collection are found in supplementary methods and Hill et al. (Hill et al., 2015). During any set of this WCST-like paradigm, criterion for a participant to advance to a new set is generating correct responses based on the current set 10 trials in a row. If a participant fails to do so within forty-eight trials, the program terminates. After a set switch, three types of errors are possible: perseverative errors, regressive errors, and general errors. Perseverative errors occur prior to acquisition of the new set - erroneous selections according to the previous set rule despite negative feedback. Regressive errors occur when selections are made based on a previous set after one or more correct responses under the new set have been made. General errors were incorrect responses that were neither perseverative nor regressive; they are not common and are not a focus of this study.

To compare ERPs between PCET error groups, psychosis and healthy groups were each divided into “low error” and “high error” groups using a median split for perseverative and regressive errors independently. For example, participants who committed more regressive errors than the total sample median were placed in the high regressive error group. The same participant may be a member of the low perseverative error group if they committed few perseverative errors during PCET. Group performance is summarized in Table 2. Participants who failed to advance to set 2 are listed as “non-completers”. Perseverative and regressive errors were not applicable for these individuals since they did not experience a set switch, so they were considered as a separate group in the first tier of our ERP analysis.

Table 2.

PCET error performance by group, error type, and performance.

	Non-completers	Set 1 completers									Total N
	N	Perseverative error				Regressive error				General error
		Low		High		Low		High
		N	Avg. # errors	N	Avg. # errors	N	Avg. # errors	N	Avg. # errors
Psychosis	64	183	0.6 (0.5)	245	4.3 (3.6)	175	0 (0)	253	6.3 (6.8)	0.8 (2.0)	492
Healthy	10	130	0.6 (0.5)	104	4.2 (4.3)	139	0 (0)	95	4.6 (4.7)	0.7 (1.7)	244
Statistical tests	X2 = 14.3***		t = 1.1		t = 0.2		t = 0.0		t = 2.3*	t = 0.5

Please note that only set 1 completers advanced to the next set and allowed for perseverative and regressive errors to emerge. Numbers of perseverative and regressive errors in each participant were independent, so an individual could be in the “high” group for perseverative error and the “low” group for regressive errors, or any of the 4 combinations. In “average number of errors” columns, means of the subgroup are stated first, followed by standard deviations in parentheses.

^*p < 0.05.

^***p < 0.001.

2.3. Auditory oddball EEG

Recording conditions, equipment, and protocol of the 2-stimulus auditory oddball task were identical across B-SNIP sites. 64-channel EEG were collected with participants seated inside a sound- and voltage-shielded booth, presented with a stream of 1000 Hz tones (standards) randomly interrupted by 1500 Hz tones (targets) by two speakers at 75 dB, with 1-s inter-trial intervals. The task lasted approximately 13 min, with 567 standards and 100 targets, and impedance was kept under 5kΩ. B-SNIP participants typically completed phenotyping procedures within two weeks; that is, there is a possible range of one to fourteen days between PCET and EEG appointments.

Additional information on collection protocol and details about EEG data preprocessing can be found in supplementary methods, Ethridge et al. (Ethridge et al., 2015), Ethridge et al. (Ethridge et al., 2012), and Clementz et al. (Clementz et al., 2016).

The following processing and statistical analysis steps were performed in Matlab software (MathWorks. Inc., Natick, MA, USA). After baseline correction, to extract the most efficient spatial representation of ERPs, the 1000-ms averaged epochs (250 ms before to 750 ms after) containing data from all 64 sensors underwent spatial principal component analysis (PCA). Two distinct components emerged, one with a medial frontal maximum and the other with a central-parietal maximum, each carried variances for sensory ERPs such as N100, N200, P200 in addition to P300 (see Fig. 1).

Fig. 1. Amplitudes of frontal and parietal ERP components between set-1 completers and non-completers.

Spatial principal component analysis (PCA) revealed two major components in ERP responses to auditory oddball targets. One has a frontal focus which captured 39% of the overall variance, and the other has a parietal focus which captured 51% of overall variance from the epochs. 3D topographies of each component are shown next to their corresponding waveforms. Red lines indicate psychosis cases, and black lines are healthy persons; solid lines indicate set1 completers, and dotted lines are non-completers. Shaded areas show plus and minus standard error from the group mean. Yellow segments on the horizontal bars indicate time periods during which the 2 × 2 ANOVAs for set 1 completion (completers vs. non-completers, top bar), subject group (healthy vs. psychosis, middle bar) and the interaction (bottom bar) were statistically significant (p < 0.05) after measures to control for multiple comparisons. Left: In the frontal component, healthy participants showed stronger N100 (negative-going voltage at 100 ms post-stimulus) and P300 (300-400 ms post-stimulus). No group-by-set 1 completion interaction was found in the frontal component throughout the trial period. Right: In the parietal components, ERPs observed in set 1 completers were significantly higher than non-completers between 300 and 530 ms post-stimulus; healthy participants showed higher amplitudes than psychosis during the N200 time range (200 ms post-stimulus) and the late P300 time range (500 ms post-stimulus). No interaction was found between subject group and set 1 completion at any time point throughout the trail period.

2.4. Statistical analysis

Bartlett component scores for the frontal and parietal components were calculated for each participant using coefficients for each sensor weight produced by the spatial PCA (Ethridge et al., 2015), and 2-by-2 ANOVAs between subject group and the types of PCET error were performed on each 10-ms time bin of the 800-ms averaged epoch (100 ms before to 700 ms after stimulus presentation, extracted from the 1-s epochs described in Section 2.3 above). To control for Type-1 error from multiple comparisons, a cluster method which utilized Monte Carlo simulations was applied to determine that three consecutive time bins of p < 0.05 are required to maintain a family-wise error threshold of p < 0.01 (Ethridge et al., 2012). This criterion was maintained throughout the analyses in this study.

3. Results

3.1. PCET outcomes

Overall, psychosis participants were less likely to complete set 1 than healthy participants (X² = 14.3, p < 0.001; Table 2), and they on average completed fewer sets (psychosis: M = 2.2, SD = 1.0; healthy: M = 2.4, SD = 0.8; t(734) = 2.1, p = 0.036). Among participants who completed set 1, psychosis participants committed more perseverative errors than healthy (psychosis: M = 2.7, SD = 3.3; healthy: M = 2.2, SD = 3.4; t(660) = 2.0, p = 0.047), and they also committed more regressive errors than healthy (psychosis: M = 3.8, SD = 6.1; healthy: M = 1.9, SD = 3.8; t(660) = 4.3, p < 0.001). PCET performance, including numbers of set 1 completers, by psychosis and healthy groups are summarized in Table 2.

Non-parametric tests showed that for both perseverative and regressive errors, the median number of errors was identical in psychosis and healthy groups (Supplementary Figs. 1 and 2); therefore, the study sample was divided into high-error and low-error groups for each error type at the global median (perseverative error groups split at the median of 1, and regressive errors groups split at the median of 0. Please also see distributions in Supplementary Tables and Figures). Mean number of errors committed by each group is summarized in Table 2. DSM diagnosis labels are not a focus of this study, but breakdowns of each error type by DSM groups are summarized in Supplementary Tables 1 and 2 for readers’ information.

3.2. ERP signatures of set 1 completion

To examine ERP characteristics of set 1 completion, two (completers vs. non-completers) by two (psychosis vs. healthy) ANOVAs were performed on 10-ms time bins of the frontal and parietal components. For the frontal component, healthy participants showed larger responses than the psychosis group during N100 as well as early P300 time ranges (80-110 ms and 280-380 ms post-stimulus, respectively). Set 1 completers showed a more pronounced resolution following the frontal response than non-completers (400-460 ms post-stimulus, Fig. 1, left). For the parietal component, the P300 peak amplitude (300-500 ms post-stimulus) was significantly higher in set 1 completers, in both psychosis and healthy groups; healthy participants produced stronger amplitudes around the N200 as well as the later P300 time ranges (480-520 ms post-stimulus; Fig. 1, right).

3.3. ERP signatures of perseverative errors

To examine ERP characteristics between perseverative error levels in set 1 completers, two (high perseverative error vs. low perseverative error) by two (psychosis vs. healthy) ANOVAs were performed on 10-ms time bins for the frontal and parietal ERP components. For the frontal component, healthy participants produced higher amplitudes at the N100 and early P300 time ranges (Fig. 2a). For the parietal component, healthy participants also produced stronger amplitudes than psychosis participants around the N200 and the late P300 time ranges (Fig. 2b). The frontal component showed a significant interaction between error rate and group: among healthy participants, those with low perseverative errors showed higher frontal amplitudes at 300 ms than those with high perseverative errors, but among psychosis participants it did not differ (Fig. 2a and c). No interaction was found for the parietal component throughout the time course (Fig. 2b and d).

Fig. 2. ERP components in response to auditory oddball targets between high and low perseverative error groups.

Healthy (blue) and psychosis (pink) subjects were separated into high (dotted lines) and low (solid lines) levels of perseverative errors in PCET task. 3D topographies of each component are shown next to their corresponding waveforms. Yellow segments on the horizontal bars indicate time periods during which the 2 × 2 ANOVAs for perseverative error levels (high vs. low, top bar), subject group (healthy vs. psychosis, middle bar) and the interaction (bottom bar) were statistically significant (p < 0.05) after measures to control for multiple comparisons. a. Frontal component observed in subjects who committed few perseverative errors showed higher amplitudes during the N200 and late (about 500 ms) time rages than those with high errors; healthy subjects produced higher amplitudes during N100, N200, P300, and late time ranges. A group-by-perseverative error interaction was found in the frontal component during the P3a time range (about 270-320 ms). b. Parietal component observed in subjects that committed few perseverative errors showed higher amplitudes during the P300 time rage than those with high errors; healthy subjects produced higher amplitudes during N200 and late P300 time ranges. No group-by-perseverative error interaction was found in the parietal component throughout the epoch. c. and d. Amplitudes of frontal and parietal components were averaged across the 5-time bin cluster which showed significant interaction effect and plotted to better show the interaction effect between error level and P300 component. As observed in a and b, A group-by-perseverative error interaction was found in the frontal component where healthy participants with low error produced higher ERP than healthy participants with high error, but this contrast was diminished in psychosis subjects, and no interaction was observed in the parietal component in the same time range.

3.4. ERP signatures of regressive errors

To examine ERP characteristics between regressive error levels in set 1 completers, two (low regressive errors vs. high regressive errors) by two (psychosis vs. healthy) ANOVAs were performed on 10-ms time bins of the spatial component scores. The main effect of participant group from this analysis was identical to results with perseverative errors, since group membership did not change. Main effect of high versus low regressive error was pronounced in the onset and resolution of frontal P300, with low-error groups showing an earlier onset and a more pronounced resolution. Late parietal P300 showed an earlier offset from the peak in the high regressive error groups. A significant error rate-by-group interaction was found for the parietal component around the late P300 time range (about 480 ms post-stimulus): healthy set-1 completers with low regressive errors produced a higher amplitude than those with high regressive errors, but psychosis participants did not differ on parietal P300 ERP amplitude as a function of regressive error rate (Fig. 3b and d). No interaction was found in the frontal component (Fig. 3a and c).

Fig. 3. ERP components in response to auditory oddball targets between high and low regressive error groups.

Healthy (blue) and psychosis (pink) subjects were separated into high (dotted lines) and low (solid lines) levels of regressive errors in PCET task. 3D topographies of each component are shown next to their corresponding waveforms. Yellow segments on the horizontal bars indicate time periods during which the 2 × 2 ANOVAs for regressive error levels (high vs. low, top bar), subject group (healthy vs. psychosis, middle bar) and the interaction (bottom bar) were statistically significant (p < 0.05) after measures to control for multiple comparisons. a. Frontal component in subjects that committed few regressive errors showed higher amplitudes during the P300 and late (450-500 ms) time ranges than those with high errors; healthy subjects produced higher amplitudes during N100, N200, and P300 time ranges. No group-by-perseverative error interaction was found in the frontal component throughout the epoch. b. Parietal component observed in subjects that committed few regressive errors showed higher amplitudes during the late N200, P300 (350 ms) and late (500 ms) time ranges than those with high errors; healthy subjects produced higher amplitudes during N200 and P300 time ranges. A significant group-by-perseverative error interaction was observed in the late P300 (early LPP) time range. c. and d. Amplitudes of frontal and parietal components were averaged across the 3-time bin cluster which showed significant interaction effect and plotted to better show the interaction effect between error level and P300 component. As observed in a and b, A group-by-regressive error interaction was found in the parietal component where healthy participants with low error produced higher ERP than healthy participants with high error, but this contrast was diminished in psychosis subjects, and no interaction was observed in the frontal component in the same time range.

4. Discussion

This study examined covariations between PCET (a set shifting task) performance and P300 component amplitudes in psychosis and healthy participants. Participants were first divided based on whether they completed set 1 and advanced to set 2 in PCET, then by perseverative errors and regressive errors. For both healthy and psychosis groups, comparison between set 1 completers versus non-completers found a significant reduction in the parietal P300 amplitude in non-completers. In addition, healthy set 1 completers showed a perseverative error-specific frontal P300 reduction, as well as a regressive error-specific parietal reduction. This dissociation was not observed in psychosis participants, who had diminished P300 amplitudes regardless of PCET performance. Repeating previously published findings using the same sample (Ethridge et al., 2015), the N100 amplitude showed a psychosis-healthy difference in the frontal regions, and the N200 amplitude showed a psychosis-healthy difference in the parietal regions. We focus our discussion on each PCET and P300 outcome in the following sections.

4.1. Set 1 non-completers

Unlike studies that assess categories completed (Everett et al., 2001; Hazlett et al., 2015; Moreno-Iniguez et al., 2005; Siegel Jr. et al., 1996), findings on set 1 acquisition failures are scarce (but see (Witt et al., 2006) and (Moore et al., 2003)). In PCET, failure to produce correct responses consecutively causes termination of testing, and the early termination prevents the participants from experiencing set switches. During PCET acquisition, participants first need to remember the stimuli and their response in the previous trial, evaluate that response based on the feedback, then formulate a plan for the next trial. This process resembles reinforcement learning that involves maintenance and flexible manipulation of items in working memory as well as rule induction, and past studies of feedback-based induction reasoning in different modalities converged in cognitive control supported by the ventral frontal and fronto-parietal networks (Babcock and Vallesi, 2015; Furlan et al., 2018; Liang et al., 2016; Marek and Dosenbach, 2018).

As PCET begins, participants are faced with three possible response set options. Under optimal and efficient reinforcement learning, with the negative feedback effectively punishing “wrong” selections, the probability of making a set-appropriate response by chance during the first, second, and third trials is 33 %, 50 %, and 100 %, respectively. In comparable paradigms providing positive and negative feedback, rodents generally acquire a set within 12 attempts (Broschard et al., 2019; Jewell et al., 2018). PCET allows forty-eight attempts before task termination; participants who fail to acquire set 1 with 48 tries may have reduced activity in (1) brain regions involved in the control networks sensitive to reinforcement, like the striatal areas or (2) networks involved in stable maintenance of a learned set that enable 10 or more consecutive correct responses.

Both healthy and psychosis participants who failed set 1 elicited lower parietal P300 amplitudes, but it is unclear from behavioral data whether they did so due to inability to acquire a set or inability to maintain a set. A higher proportion of psychosis participants in our sample did not complete set 1 (26 %), compared to healthy group (4 %). This difference was replicated in a recent study which also identified a substantial subgroup within the psychosis population with severe impairment in goal-directed and flexibility in behaviors that was not present in healthy controls (Suetani et al., 2022). There is also some evidence of alpha band activity in the frontal-parietal network intensifying during tasks demanding high levels of working memory and mental image manipulation (Sadaghiani et al., 2012), and data in the frequency domain from an overlapping sample of the current study showed that theta/alpha power (between 1 and 15 Hz) in the parietal component during the P300 time range was strongly diminished in psychosis subjects, published previously (Ethridge et al., 2015). It would be of interest to examine this frequency band in the parietal cortex during PCET set acquisition. The characteristics of healthy and psychosis participants who failed to acquire set 1 may be useful in differentiating psychosis-specific features from general cognitive deficiencies.

4.2. ERP responses in set 1 completers

Within healthy participants only, diminished frontal P300 amplitude was perseverative error-specific, and diminished parietal late-P300 amplitude was regressive error-specific. Psychosis participants, regardless of PCET performance, did not show these effects. Augmented frontal amplitude observed with low perseverative errors may indicate that effective frontal involvement is required for successful set switching, which is consistent with previous findings in both animals and humans (Boulougouris et al., 2007; Stuss et al., 2000). Perhaps healthy participants with higher levels of perseveration have weaker functional (synaptic strengths of the circuit) or anatomical (grey matter volume or density; number of cells available as potential generators, etc.) properties in anterior cingulate cortex, bilateral vlPFC, or bilateral dlPFC.

Also within healthy participants only, parietal reduction in late P300 time range was associated with high levels of regressive error. Literature on anatomical correlates of regressive errors in humans is less extensive than for perseverative errors. Similar to observations in animals with striatal damage, a disrupted cingulo-opercular network in primates leads to regressive errors (Rossi et al., 2007). Designs that test regressive behavior during brain functional recording (EEG or functional magnetic resonance imaging) in human psychosis participants would be important additions to the literature.

The dissociation in which parietal scalp amplitude was only associated with regressive errors and frontal amplitude only with perseverative errors, was not observed in psychosis. Rather, psychosis participants produced comparable frontal and parietal responses regardless of PCET performance. A universal P300 ERP reduction in psychosis regardless of error performance suggests that burden posed by psychosis manifestation on P300 related networks may only partially overlap with alterations attributable to cognitive flexibility and maintenance, since half of the patients were still able to perform the PCET task at a high level (see, e.g., (Hamm et al., 2015)).

4.3. Limitations

Medication history was variable in this sample. To address the variability, previous publications investigating oscillatory and ERP responses to the oddball paradigm in the same sample of participants have shown no significant correlation between the subcomponents of P300 and medications of any type in probands (Ethridge et al., 2015). Also, although collected close in time (one to fourteen days apart), minimizing longitudinal effect on the two tasks, ERP responses and PCET performance were not collected synchronously. The observed dissociation only provides evidence of covariance, instead of a causal link, between ERP generators under presence or absence of psychosis and cognitive error types. Animal lesion studies show similar dissociation patterns, indicating frontal involvement in perseveration and striatal involvement in set maintenance, but it is unclear whether direct parallels can be drawn to humans. To further isolate relevant networks involved in induction, set switching flexibility, and working memory maintenance, future PCET studies using higher-density EEG will benefit from implementing source analysis techniques like low-resolution brain electromagnetic tomography (LORETA) or comparison with functional MRI during PCET performance. Alternatively, concurrent imaging of neural signaling features using MRS (magnetic resonance spectroscopy) or PET (positron emission tomography) during PCET could confirm the pathway-level mechanism of P300 reductions in psychosis, as well as among healthy participants with high error rates.

4.4. Conclusion

The current study revealed a double dissociation in the healthy population between PCET cognitive error types and related electro-physiology: frontal P300 amplitudes were lower in those with poor cognitive flexibility, and parietal P300 was lower in those with poor working memory maintenance. The double dissociation was not present in psychosis participants. Real-world significance of these results is two-fold: in the absence of psychosis, the specific links between brain network response strength and higher functioning may direct future treatment discovery in neural stimulation. On the other hand, the attenuated frontal and parietal activity in psychosis, even when cognitive functions were intact in some of those individuals, opens doors for further investigation into neural changes, perhaps compensatory, need-based re-routing, upon psychosis manifestation.