A dimensional investigation of error-related negativity (ERN) and self-reported psychiatric symptoms

Alterations in error processing are implicated in a range of DSM-defined psychiatric disorders. For instance, obsessive-compulsive disorder (OCD) and generalised anxiety disorder show enhanced electrophysiological responses to errors—i.e. error-related negativity (ERN)—while others like schizophrenia have an attenuated ERN. However, as diagnostic categories in psychiatry are heterogeneous and also highly intercorrelated, the precise mapping of ERN enhancements/impairments is unclear. To address this, we recorded electroencephalograms (EEG) from 196 participants who performed the Flanker task and collected scores on 9 questionnaires assessing psychiatric symptoms to test if a dimensional framework could reveal specific transdiagnostic clinical manifestations of error processing dysfunctions. Contrary to our hypothesis, we found non-significant associations between ERN amplitude and symptom severity of OCD, trait anxiety, depression, social anxiety, impulsivity, eating disorders, alcohol addiction, schizotypy and apathy. A transdiagnostic approach did nothing to improve signal; there were non-significant associations between all three transdiagnostic dimensions (anxious-depression, compulsive behaviour and intrusive thought, and social withdrawal) and ERN magnitude. In these same individuals, we replicated a previously published transdiagnostic association between goal-directed learning and compulsive behaviour and intrusive thought. Possible explanations discussed are (i) that associations between the ERN and psychopathology might be smaller than previously assumed, (ii) that these associations might depend on a greater level of symptom severity than other transdiagnostic cognitive biomarkers, or (iii) that task parameters, such as the ratio of compatible to incompatible trials, might be crucial for ensuring the sensitivity of the ERN to clinical phenomena.


Supplemental Information for A dimensional investigation of error-related negativity (ERN) and self-reported psychiatric symptoms
Supplemental Methods ERN, demographics and error rate. In existing work, age, gender and IQ (Falkenstein et al., 2001;Fischer et al., 2016;Larson et al., 2016;Zijlmans et al., 2019) yield various relationships with ERN amplitude. We explored their effects on ERN in our data. IQ (β = -0.38, SE = 0.20, p = 0.06) and age (β = 0.39, SE = 0.20, p = 0.05) showed a trending effect with ERN amplitude shifts, while gender not associated (both p < 1). We also observed that error rate was related to the ERN (β = 0.40, SE = 0.20, p = 0.04). However, inclusion of age and IQ nor error rate did not change the effect of questionnaires scores on ERN amplitude (all p > 0.13, uncorrected).
ERN and medication status. The ERN has also been previously influenced by various psychotropic medication (Bates et al., 2002;de Bruijn et al., 2006;Endrass et al., 2008;Henderson et al., 2006;Riba et al., 2005 (Clayson et al., 2013). Here we report the supplementary analyses showing that the main results were not due to our chosen analysis approachwhether it was from electrode site (Supplemental Figure S6) or ERN quantification method (Supplemental Figure S7). For non-adaptive mean, ERN amplitude calculated as the mean of ±40ms at 37.61ms post-response, which was the mean latency of the most negative peak across participants. For peak, the most negative peak was identified, and amplitude was extracted, for each participant by searching for the largest preceding negativity within -20ms to 120ms post-response. For trough-peak, the trough was identified for each participant by searching for the largest preceding positivity within -100ms before the peak. The amplitude of this positive peak was then subtracted from the negative peak amplitude.

ERN controlled for CRN variation.
A common method thought to isolate activity specific to error monitoring is calculated by the subtraction of the CRN from the ERN i.e. ERN-CRN (ΔERN) (Gehring et al., 1993). However, using the subtraction method is conceptually problematic as the ERN and CRN are highly correlated across individuals (here, ERN and CRN correlate: r = 0.30, p < 0.001). This is because difference scores are not independent from the constituent measures (i.e. not an error processing measure independent of the CRN) and may conflate effects relating to either signal (Meyer et al., 2017). An alternative approach to control for variation of the CRN is to use the variation left over from a regression of CRN predicting ERN (ERNresid) as the ERN amplitude measure. ERNresid was correlated to ERN (r = 0.95, p < 0.001) but not to the CRN (r = ~0, p = 1), suggesting that it specifically indexes error-related activity and is a more interpretable measure. We report the associations of these two different ERN measures, ΔERN and ERNresid with questionnaire scores ( Figure S8 and Table S3), and note both findings do not reveal any significant effects (all p > 0.12, uncorrected).

ERN, depression and anxiety. Previous studies have suggested that depression
can reduce the increased ERN amplitudes effect associated with anxiety (Weinberg et al., 2016(Weinberg et al., , 2015(Weinberg et al., , 2012. We tested if this was true in our data by regressing depression and anxiety total scores against ERN estimates in the same model. Both effects remained non-significant; but the direction of effects was perhaps more representative of the literature with anxiety leaning towards a larger ERN (β = -0.33, SE = 0.30, p = 0.27; standardised β = -0.12 (a metric that is comparable to r (Peterson and Brown, 2005)) and depression towards a smaller ERN (β = 0.43, SE = 0.30, p = 0.16).
Goal-directed learning. The same sample of participants (N = 234) completed the two-step reinforcement learning task (Daw et al., 2011). Several exclusion criteria were applied to ensure data quality, on a rolling basis. i) Participants who responded with the same key in stage one >90% (n = 135) of the time (N = 10). ii) Participants whose probability of staying after common, rewarded trials was less than 5% likely to be at chance, based on a binomial distribution with 50% (chance) probability and the total number of common-rewarded trials experienced by each participant (N = 11). iii) Participants who missed >20% (n = 30) of the trials were excluded (N = 3). (iv) Participants who incorrectly responded to a "catch" question within the questionnaires: "If you are paying attention to these questions, please select 'A little' as your answer" were excluded (N = 7). (v) As we intend to analyse the EEG data collected for this task, we additionally excluded participants whose EEG data were incomplete (N = 5) or corrupt (N = 2) from the analysis. 38 participants (16.24%) were excluded in total, leaving 196 participants for analysis. To clean the task data, we excluded individual trials with very fast reaction times (<150ms) (Gillan et al., 2016). Inclusion of these demographics did not change the pattern of effect to compulsivity (β = -0.08, SE = 0.04, p = 0.04).
Seow et al. Figure S1. Across participants, the distribution of:

Supplemental Figures and Tables
(A) Mean error rate.

(D) Mean RT by trial congruency.
(E) Mean RT by trial accuracy.

(F) Mean RT by post-trial accuracy.
Vertical lines denote mean error rate/RT for respective trial type.

Figure S2. Mean error rate and response times (RT) for various trial types. cC: congruent trials preceded by a congruent trial, cI: incongruent trial preceded by a congruent trial, iC: congruent trial proceeded by an incongruent trial), iL: incongruent trial preceded by an incongruent trial). White dots represent individual participants, red marker indicates mean and SD.
Conflict adaptation. Conflict adaptation effects refer to the phenomenon wherein previous-trial congruency affects current-trial performance, which have consistently been shown as behavioural adjustment in error rates and RTs in Flanker tasks (Clayson and Larson, 2011;Larson et al., 2016). We replicate these effects, where mean error rates were smaller for iI than for cI trials ( Figure S4. Associations between questionnaire scores with mean error rate (%).
Error bars denote standard errors. The Y-axes indicate the change in error rate as a function of 1 standard deviation (SD) increase of questionnaire scores. No questionnaire score was significantly associated to changes in error rate. Figure S5. Scatterplots of ERN amplitude and total questionnaire scores. Coloured markers represent an individual's total score for the corresponding questionnaire.
See Figure 2 and