The Relationship Between Emotion Dysregulation and Error Monitoring in Adolescents with ADHD

Abstract

Attention-deficit/hyperactivity disorder (ADHD) is emblematic of the limitations of existing diagnostic categories. One potential solution, consistent with the Research Domain Criteria (RDoC) initiative, is to interrogate psychological mechanisms at the behavioral and physiological level together to try and identify meaningful subgroups within existing categories. Such approaches provide a way to revise diagnostic boundaries and clarify individual variation in mechanisms. Here, we illustrate this approach to help resolve heterogeneity in ADHD using a combination of behaviorally-rated temperament measures from the Early Adolescent Temperament Questionnaire; cognitive performance on three difference conditions of an emotional go/no-go task; and electroencephalogram (EEG)-measured variation in multiple stages of error processing, including the error-related negativity (ERN) and positivity (Pe). In a large (N = 362), well-characterized sample of adolescents with ADHD, latent profile analysis identified two ADHD temperament subgroups: 1) emotionally regulated and 2) emotionally dysregulated (with high negative affect). Cognitive and EEG assessment in a subset of 272 adolescents (nADHD=151) found that the emotionally dysregulated group showed distinct patterns of change in early neural response to errors (ERN) across emotional task conditions as compared to emotionally-regulated ADHD adolescents and typically-developing controls. Both ADHD groups showed blunted later response to errors (Pe) that was stable across emotional task conditions. Overall, neural response patterns identified important differences in how trait and state emotion interact to affect cognitive processing. Results highlight important temperament variation within ADHD that helps clarify its relationship to the ERN, one of the most prominent putative neural biomarkers for psychopathology.

The Relationship Between Emotion Dysregulation and Error Monitoring in Adolescents with ADHD

Attention-deficit/hyperactivity disorder (ADHD) is emblematic of the problems in existing psychiatric classification systems that the Research Domain Criteria (RDoC) initiative approach (Kozak & Cuthbert, 2016) was intended to help solve. First, inattention (and to a lesser extent hyperactivity-impulsivity) occurs not just in ADHD but across a large number of diagnostic groups. In other words, at least at the behavioral level, symptoms of ADHD are transdiagnostic (although whether these ostensibly similar behaviors reflect similar underlying mechanisms remains a major question for the field). An additional problem, and the focus of the work reported here, is the presence of significant within-diagnosis heterogeneity, including differences between individuals with ADHD in symptom presentation (Sibley et al., 2011), comorbidity (Luo et al., 2019), cognitive abilities (Huang-Pollock et al., 2012; Kofler et al., 2013; Willcutt et al., 2005), trait-based emotional and personality-related elements (Nigg et al, 2020), clinical outcomes, and biological correlates (Sibley et al., 2012; Karalunas & Nigg, 2020). This substantial within-group variation complicates efforts to understand individual differences in etiology and prospective risk. Thus, the field has sought alternative approaches.

As one example, the RDoC initiative seeks to identify novel phenotypes based on biologically-informed dimensional measures that correspond to differences in clinical outcome or etiology (Kozak & Cuthbert, 2016) with a strong focus on and recognition of the critical importance of psychological domains (Lake et al., 2017) and the utility of psychophysiological indices of psychological constructs (Miller et al., 2016). One promising approach emphasizes the role of trait-based emotional variation in ADHD using measures of temperament. However, most work in this area has focused on early and middle childhood and how trait-based emotional variation relates to ADHD at other ages remains unclear (Goh et al., 2020; Karalunas et al 2014). Further, while there is relatively stronger evidence for the clinical utility of trait-based subgroups in ADHD (Karalunas et al., 2019), the biological validation of these trait-based emotion types remains understudied. Here, we apply latent clustering techniques to a clinically-deployable set of temperament ratings and then compare groups using electroencephalogram (EEG)-based measures of error-related brain activity that target both cognitive and emotional response systems. The ultimate goal of this type of work is to link trait-based variation within heterogeneous diagnostic categories to differences in functioning of biological systems that will help move the field towards biologically-informed diagnosis and treatment matching for ADHD and other types of psychopathology.

Trait-based emotional subtyping in ADHD

Weaknesses in cognitive control are well documented in ADHD (Huang-Pollock et al., 2012; Kofler et al., 2012; Willcutt et al., 2005). While this is one important conceptualization, there is also renewed interest in the role of emotion dysregulation in ADHD (Beauchaine, 2015; Bunford et al., 2015; Graziano & Garcia, 2016; Shaw et al., 2014). While several early theories of the disorder (e.g., Barkley’s inhibition model, Barkley, 1997) emphasized emotion dysregulation as an important consideration, diagnostically these emotional symptoms are often included as part of comorbid diagnoses (including oppositional defiant disorder, depression, or anxiety). Several studies have challenged this comorbidity approach and have suggested that emotion dysregulation and heterogeneity may be better understood as a core feature of ADHD (Nigg et al., 2020; Shaw et al., 2014).

Recent efforts to characterize emotional heterogeneity in ADHD have emphasized temperament traits (Clark, 2005; Rothbart & Ahadi, 1994). Although there are several influential models, here we focus on Rothbart’s temperament model (Rothbart, 2011) which provides a set of phenotypic features related to emotional reactivity and regulation that are outside of the scope of formal ADHD diagnostic criteria (Jones et al., 2015; Bates et al., 2008; Clark, 2005; Kotov et al., 2010) but have been long theorized as relevant in ADHD (Nigg et al., 2004). Rothbart’s model of temperament encompasses both top-down cognitive processes (captured via Effortful Control) and bottom-up measures of reactivity and emotional responding (captured by Surgency—related to positive affect and extraversion— and Negative Affect—including fear, sadness, anger, and low soothability). Each of these superordinate traits can be further broken down (e.g., Negative Affect is comprised of subscales related to fear, anger, sadness, etc) and temperament domains also have direct correlates in the RDoC matrix (e.g., Acute Threat (“Fear”) in the Negative Valence Systems; Kozak & Cuthbert, 2016).

Prior work has consistently identified dimensional relationships between ADHD symptom domains and temperament traits (Gomez & Corr, 2014). In addition, a smaller number of studies have shown success in using person-centered approaches to identify reproducible subgroups of children with ADHD in early and middle childhood (Martel, 2016; Karalunas et al., 2014; Karalunas et al., 2019) with at least some consistency in findings, including a group with regulated emotional functioning (characteristic differences are primarily in features that make up the core ADHD symptom domains), a group characterized by high surgency/extraversion, and a group with high negative affect. In analyses conducted in the current sample at younger ages, all three groups (emotionally regulated, high negative affect, high positive affect/surgency) were observed (Karalunas et al., 2019). However, it is less clear to what extent these findings carry forward into adolescence. Initial studies of trait-based variation within ADHD have identified similar emotional subgroups at older ages (Martel et al., 2010; Smith & Martel, 2019) but with less consistent presence of a high surgency/extraversion group (Nigg et al., 2010; Smith & Martel, 2019) and considerable variation in the proportion of individuals with ADHD characterized by high negative affect.

Psychophysiological validation of RDoC-based or emotional subtypes in ADHD

If consistent patterns of emotional variation in ADHD exist, an additional question is whether groups provide useful information above and beyond ADHD diagnosis. Nearly all validation of emotion-based ADHD groups to date has focused on clinical validity— that is, whether emotion-based groups are associated with differences in meaningful clinical outcomes, such as developing comorbidity or impairment. Evidence is reasonably strong that the emotionally-dysregulated profiles predict clinical outcomes over the short term (1–2 years) better than ADHD symptom severity, baseline comorbidity, or impairment (Karalunas et al., 2014; Karalunas et al., 2019; Martel, 2016; Smith & Martel, 2019), suggesting this approach can yield important additional insight into intervention efforts specific to the experiences of those with ADHD.

Some models remain primarily interested in this level of clinical description, which is valuable (Kotov et al., 2017). However, another major goal of RDoC is to identify relationships between traits and associated neurobiological and psychophysiological systems (Cuthbert, 2014; Miller et al., 2016). Temperament-based groups are ideal in this context because temperament traits have direct theorized relationships to biological systems (Shiner et al., 2021; Clark & Watson, 2021). Initial studies in the current ADHD longitudinal sample at younger ages suggested differences between emotion-based ADHD groups in resting-state connectivity between the amygdala and top-down control regions (Karalunas et al., 2014). Additional studies in larger samples using approaches that capture functioning of both emotional and cognitive control systems are needed.

Error-Monitoring

Here, we focus our attempts at psychobiological evaluation on error monitoring – the ability to detect and respond to errors – using event-related potentials (ERPs). Successful error monitoring is fundamentally important for both adaptation to the environment and regulation of goal directed behaviors (Falkenstein et al., 2000; Gehring et al., 1993; Manoach & Agam, 2013). It is related to RDoC constructs of cognitive control and negative valence systems and has been highlighted as a critical juncture point for cognitive and emotional processing (Weinberg et al., 2015). As described in detail below, neural markers associated with error monitoring have also been implicated as putative transdiagnostic biomarkers for both internalizing and externalizing disorders (Meyer, 2022), but how ADHD fits into this internalizing/externalizing framework remains unclear.

Error monitoring is typically understood from an information processing perspective as occurring in several main stages: initial error (or conflict) detection followed by conscious error awareness and subsequent regulation of behavioral responses. Critically for the current purposes, there are well-defined neural markers for different stages of the error monitoring process that can be captured using the high temporal resolution of EEG-measured ERPs (Olvet & Hajcak, 2008). Thus, error monitoring is an excellent starting point for psychophysiological validation of temperament groups because it is a clinically-relevant process at the nexus of cognitive and emotional control with well-defined neurobiological indices. While other ERPs associated with attention (e.g., P3) or inhibition (e.g., N2) are also of interest, they have been more extensively studied in ADHD (including in this sample; Karalunas et al., 2020; Alperin et al., 2019) with somewhat more consistent findings so are not the focus of investigation here.

Error-Related Negativity

Initial error detection is measured using the error-related negativity (ERN). During laboratory tasks, an ERN occurs when an error is detected, regardless of a respondent’s conscious awareness of the error (Olvet & Hajcak, 2008; Meyer, 2017). The ERN appears as a negative deflection at fronto-central electrode sites, primarily at the 10–20 location FCz, 0–100ms following an error of commission (Luck & Kappenman, 2012; Meyer, 2017; Weinberg et al., 2015). Converging evidence from fMRI and EEG studies suggests the ERN is partially generated by the anterior cingulate cortex (ACC), an area of the brain involved in the assimilation of information about pain, threat and punishment, and subsequent changes in behavior if necessary (Debener et al., 2005; Gehring et al., 1993). Maturation of the ACC lasts into adulthood, and the activation of this region increases throughout development (Velanova et al., 2008; Adleman et al., 2007; Davies et al., 2004).

Alterations in the early stages of error monitoring as indexed by the ERN have been linked to maladaptive behaviors and multiple psychopathologies (Weinberg et al., 2015; Meyer et al., 2021; Olvet & Hajcak, 2008; Manoach & Agam, 2013), and thus are already a promising psychopathology biomarker overall. Broadly speaking, the ERN amplitude has been robustly observed as enhanced in internalizing disorders characterized by overactive concern regarding performance (Gehring et al., 2000; Ladouceur et al., 2006; Endrass et al., 2014) and blunted in many externalizing disorders, including during middle childhood and adolescence (Meyer, 2022). Similarly, consistent evidence links enhanced ERN to temperamental fear and negative affect, suggesting it varies with trait-based emotional responding, even in the absence of diagnosable disorder (Fox et al., 2023; Meyer, 2016). It has been proposed to be a transdiagnostic marker of psychopathology (Hajcak et al., 2019; Meyer, 2016).

In contrast, and perhaps because of failure to consider individual differences in emotion or temperament, findings in ADHD have been inconsistent. Studies have sometimes shown blunted ERNs in this group (Albrecht et al., 2008; Meyer et al., 2018; Groen et al., 2008), reflecting a reduced initial detection of errors. However, many studies have failed to find group differences between children diagnosed with ADHD and typically-developing controls (Jonkman et al., 2007; Wiersema et al., 2005; Groom et al., 2010) and one study has even found that the ADHD group showed an enhanced ERN (Burgio-Murphy et al., 2007).

Inconsistencies in the ADHD error-monitoring literature may be due in part to small sample sizes, task differences, or differences in ERN measurement (Luck & Kappenman, 2012; Shiels & Hawk, 2010). However, an untested possibility is that studies of performance monitoring in ADHD have not accounted for variation in trait-based emotional tendencies, which could impact the pattern of ERN attenuation or enhancement reported (Meyer & Hajcak, 2019; Nuñez-Estupiñan et al., 2022). In addition, recent systematic review (Nuñez-Estupiñan et al., 2022) finds that manipulating state emotion (e.g., via valence of stimuli) may lead to either enhancement or blunting of the ERN but among the small number of studies available, the direction of effects is mixed with only slightly more than half of studies reporting an enhanced ERN and the remainder reporting no effects or ERN blunting. Such discrepancies are likely due to a combination of small, underpowered studies and unaccounted for interactions between state and trait-level emotion that have been very rarely considered but are likely important.

In addition, state effects have been assessed primarily in the context of anxiety, where the ERN is most well-studied. Almost no studies have examined state-level emotion effects in the context of ADHD. One prior study (Balogh et al., 2017) did find that ERN was most blunted in ADHD when negative stimuli were used, but the study was small and potential interactions with trait-level variables were not considered.

Error-Related Positivity

The ERN component is followed by the error-related positivity (Pe), which indexes later stages of error processing, including determining error salience (Meyer et al., 2012), the conscious awareness of errors (Herrmann et al., 2004; Nieuwenhuis et al., 2001; Falkenstein et al., 2000; Wang et al., 2020), and affective response to errors (Larson et al., 2011; Overbeek et al., 2005). The Pe is positive going and typically peaks during cognitive tasks 200–500ms following an incorrect response. The Pe component is more consistently implicated than the ERN as being blunted in individuals diagnosed with ADHD (Shiels & Hawk, 2010; Groen et al., 2008; Marquardt et al., 2018; van de voorde et al., 2010), with differences present even in the absence of group differences in ERN amplitudes (Wiersema et al., 2009; Jonkman et al., 2007; Zhang et al., 2009; Shen et al., 2011). This greater consistency is aligned with recent findings that the Pe may not vary with trait-level emotional characteristics, such as narcissism (Mück et al., 2023), but the Pe remains understudied compared to the ERN in relation to both state and trait-level emotion.

Current Study

The current study seeks to: 1) describe trait-based emotional heterogeneity in a large, well-characterized adolescent sample with ADHD. This aim directly addresses RDoC goals of resolving within-diagnosis variation that complicates study of etiology and outcome; 2) examine diagnostic group (ADHD, typically-developing) differences in early and later stages of error monitoring for comparability to prior literature; and 3) determine whether any ADHD error monitoring differences are attributable to specific temperament-based subgroups within the ADHD sample, as well as assess theoretically important trait-by-state interactions.

Based on prior literature and prior work in this sample at earlier ages (Karalunas et al., 2014; 2019), we expected to identify 2–3 unique emotional profiles within the ADHD group: an emotionally well-regulated group, a high negative affect/irritable group, and possibly a high surgency/extraversion group. We also hypothesized that adolescents with ADHD would have blunted ERN and Pe amplitudes when compared to typically-developing controls but that these effects would vary based on trait-based emotional profile. Our primary hypothesis was for an enhanced ERN in adolescents with ADHD and high negative affect but with the possibility that a specific subtype with elevated anger or a more purely externalizing profile might emerge with blunted ERN. We expected Pe to be blunted in all ADHD emotion subgroups given the consistency in diagnostic group level findings for this component but with recognition that Pe is not well-studied in relation to trait-level emotional variation. Lastly, we examined whether the emotional context of the task would affect ERN and Pe amplitudes and hypothesized that trait-state interactions may occur but did not have more specific hypotheses for whether emotion would enhance or blunt these neural responses.

Methods

Adolescents with and without ADHD were part of a larger longitudinal study. The overall sample, baseline enrollment criteria, and details of all assessment procedures are described in Nigg et al. (2023). Briefly, recruitment for the larger study utilized a community-based sampling method that included both community outreach and public advertising. A parent or legal guardian gave written informed consent and all adolescents provided written assent for the study. Ethics approval was obtained from and provided by the Institutional Review Board at Oregon Health & Science University.

Baseline enrollment and diagnosis

At baseline, children age 7–12 underwent extensive clinical evaluation including parent and teacher questionnaires (Conners Rating Scales, 3^rd Edition; Conners, 2001; ADHD Rating Scale [ADHD-RS]; DuPaul et al., 1998; Strengths & Difficulties Questionnaire [SDQ]; Goodman, 1997), child self-report (Child Depression Inventory; Kovacs, 2015; Multiaxial Anxiety Scale for Children; March et al., 1997), parent semi-structured clinical interview (Kiddie Schedule for Affective Disorders and Schizophrenia [K-SADS]; Kaufman et al., 1997), and IQ and academic achievement screening (WISC-IV; WIAT-III; Wechsler, 2003). A diagnostic team that included a psychiatrist and clinical psychologist reviewed cases independently and made diagnostic determinations for ADHD and all other disorders. Children were enrolled to be followed longitudinally if they met criteria for ADHD or as a typically-developing control. Children in both diagnostic groups were excluded if they were: prescribed long-acting psychotropic medications; had a history of neurological concerns or impairment including seizures and head injury that included loss of consciousness; had a substance abuse disorder; had a prior diagnosis of intellectual disability, autism spectrum disorder, psychosis, or tourette’s syndrome; were experiencing a full major depressive episode at the time of initial clinical evaluation, or had an estimated IQ below 70.

Longitudinal follow-up & EEG study

Participants.

Enrolled children were followed annually for up to 12 years with comprehensive clinical and cognitive assessment (see Nigg et al., 2023) that included temperament measures and detailed clinical characterization using the same measures and approaches as at baseline. In addition, a substantial subset completed a single additional EEG testing visit at either year 5, 6, or 8 of the longitudinal study during early to mid-adolescence (ages 11–17). The current study included 463 adolescents, including 342 adolescents with ADHD whose parents completed a temperament rating scale at least one time between years 5–8 of the larger study. Of these 342 adolescents with ADHD, 151 also completed the EEG testing visit. In addition, 121 typically-developing controls who completed the EEG testing visit served as a comparison group for EEG analyses. Baseline characteristics of adolescents who completed an EEG visit did not differ from the full sample for sex assigned at birth, race, ethnicity, comorbidities, or family income. Those completing the EEG visit had a slightly higher baseline IQ scores, but the effect was small (d = .18).

Diagnostic assessment at follow-up.

Diagnostic assessment at all years of follow up was identical to baseline. The clinical diagnostic team achieved kappa > 0.87 for ADHD diagnoses and > 0.70 for nearly all common disorders at all years of the study. Final diagnostic groups for this study used the current ADHD diagnosis at the year of EEG assessment.

Stimulant Medication.

Medications at follow-up were not exclusionary for the larger study, however, children taking non-stimulant medications were excluded from the EEG testing. Children who were taking stimulant medications were eligible for EEG testing but required to washout of their stimulant medications for 24 or 48 hours depending on if the prescription was short or long acting.

Measures

Adolescent Temperament.

To measure trait-level emotion, we utilized a modified version of the parent-rated Early Adolescent Temperament Questionnaire-Revised (Ellis & Rothbart, 2001) derived in an overlapping sample that showed strong correspondence with the original theorized scales (Kozlowski et al., 2023). In the EATQ-R, parents were asked to rate items on a five-point Likert scale (1 = almost always untrue; 2 = usually untrue; 3 = sometimes true, sometimes untrue; 4 = usually true; and 5 = almost always true, e.g., “… worries about getting into trouble” or “enjoys exchanging hugs with people s/he likes”). Temperament ratings were selected to match the year of EEG assessment for those who participated in EEG. For other participants, ratings were taken from Year 6 of the study, or from Year 5 if Year 6 was not available. The current modified form includes seven rather than the ten original factors. These are: Effortful Control (a composite scale of the original Activation Control, Attentional Focusing, and Inhibitory control; ω = 0.91), Anger (a composite of the original frustration and aggression subscales; ω = 0.83), Shyness (ω = 0.87), Depression (ω = 0.73), Fear (ω = 0.72), Affiliation (ω = 0.77), and High Intensity Pleasure (ω = 0.76). A lower order correlated factors CFA demonstrated that this alternate form was an acceptable fit to the data when within-subscale or correlated errors consistent with theory were allowed in the model (χ2 (700) = 1146.36, p < 0.01, CFI = 0.91, TLI = 0.90, RMSEA = 0.043, SRMR = 0.06).

Go/No-Go Task.

Participants completed three conditions of a go/no-go tasks while 32- or 64-channel EEG were recorded. The three tasks differed according to the emotional valence of the condition: neutral, fearful, and happy. All conditions were counterbalanced between participants. 344 trials of grey-scaled faces from the NimStim set were presented in a semi-random order with one break in between for each condition (Tottenham et al., 2009). In the neutral condition of this task, adolescents were asked to press a button if they saw either a male or female face and were asked not to press a button if they saw the opposite sex; this was separated into two blocks and blocks were counterbalanced on whether they responded to male of female faces first. In the fearful condition, adolescents pressed a button if they saw a fearful face and withheld a response if they saw a neutral face. A similar design was used in the happy condition in which a response was required if they saw a happy face and no response was required for a neutral face. Prior to testing, all adolescents were given 10 practice trials to ensure understanding. Adolescents were instructed to remain as still and quiet as possible.

EEG recording and analysis

EEG were recorded at 500Hz using either 32 (n = 177) or 64 (n = 95) AG-AgCl active electrodes based on the International 10–20 system centered at the central midline electrode site (Cz). EEG signals were recorded using PyCorder v1.0.9 and then amplified utilizing a BrainVision actiCHamp amplifier. Impedance level for each electrode was kept at or below 50 kΩ. Data were referenced to Cz online and re-referenced offline to the average of all electrodes. EEG data were pre-processed in BrainVision Analyzer and EEGLAB version 2022.0 (Delorme & Makeig, 2004), following a modified version of the Maryland Analysis of Developmental EEG (MADE) Processing Pipeline (Debnath et al., 2019). EEG data were resampled to 250Hz. Following MADE, data were high and low pass filtered at .3–50Hz with an additional notch filter at 60Hz added, eye movement artifacts were removed utilizing independent component analysis, channels were interpolated following artifact rejection, and then re-referenced offline to the average of all electrodes.

Response-locked segments from −500 to 500ms with baseline correction of −500 to −300 prior to response onset were used. ERN amplitude was measured as mean area amplitude between 0 and 100ms following errors of commission and Pe was measured as mean area amplitude between 200 and 400ms following errors of commission at electrode Cz. Trials were excluded from analyses if they contain baseline drift or movement artifacts greater than 100mV. Out of a possible total of 816 condition observations, a participant’s data for a given condition were excluded if more than 50% of the total trials were rejected due to artifacts, if more than 10% of electrode channels were interpolated, and if participants made fewer than six commission errors to derive psychometrically reliable ERPs (Pontifex et al., 2010; Steele et al., 2016). On average, typically-developing controls made a total of 16 errors across conditions (SD = 15.68) and ADHD adolescents made 21 errors (SD = 19.28). 65 total subjects were excluded for making less than six errors of commission in all three go/no-go conditions (nADHD = 33). The final EEG sample following data exclusion (participants who retained at least one condition) included 207 (nADHD= 136) adolescents.

Statistical Analysis

Latent Profile Analysis.

Latent Profile Analyses (LPA) were conducted in Mplus 8.7 (Muthen & Muthen, 1998–2017) to extract temperament trait profiles. LPA were identified using full maximum likelihood robust estimation methods to handle departures from normality. The CLUSTER command in Mplus was also utilized to account for sibling level non-independence in observations (n = 14). We adopted a step-wise approach applied to 1–5 profile solutions to identify profiles based on evaluation of fit indices and theory (Spurk et al., 2020). Fit indices included Bayesian Information Criterion (BIC), the Sample Adjusted BIC (SaBIC), the Consistent Akaike Information Criterion (CAIC), The Approximate Weight of Evidence (AWE), Entropy, and the Vuong-Lo-Mendell Rubin likelihood ratio test (VLMR-LRT). Per methodological guidance, we also plotted the information criterion values (Nylund-Gibson & Choi, 2018) in selecting a final model.

Group Comparisons.

Cognitive performance and ERPs were analyzed utilizing linear mixed models in SPSS (v.26). Missing data were handled with maximum likelihood estimation methods. Age, sex, and number of available error trials were used as covariates in all analyses as decided a priori due to likelihood of sex and developmental effects (Larson et al., 2011; Meyer et al., 2011; Davies et al., 2004). Number of electrodes available on each electrode cap was tested as a covariate but did not impact results so are not included in the reported results.

Results

Participant Description

Table 1 provides full information for participant demographics. As expected, based on current rates of ADHD diagnosis in community-based samples, the ADHD group had a higher proportion of males than females when compared to the typically-developing group (χ²(1) = 8.65, p = .003). Participant groups did not differ in age, ethnic identity, racial identity, or socioeconomic status (SES). ADHD adolescents had a lower IQ score when compared to their non-ADHD peers (t(267) = 5.17, p < .001, d = .71). 45.4% of those in the ADHD group were prescribed stimulant medications.

Table 1.

Demographics by Revised Parent Rated EATQ-R Temperament Subtype in EEG Sample

	Regulated ADHD	Dysregulated ADHD	Typically-Developing Controls
N	117	116	109
M(SD) Age in years	14.29(1.49)	14.19(1.46)	14.09(1.34)
Sex Assigned at Birth (Male: Female)	88:29	48:20	60:49
IQ M	109.26	108.70	114.96
Race N
American Indian	3	1	4
Asian American/Pacific Islander	7	7	9
Black American	13	9	5
White American	94	99	91
Hispanic N	5	6	7
Family Income	$75,000-$100,000	$50,000-$75000	$75,000-$100,000
Comorbid conditions N
Mood Disorder	5.98%	11.21%	1.83%
Anxiety Disorder	4.27%	25.86%	5.50%
Oppositional Defiant Disorder	2.56%	13.79%	1.83%
Learning Disorder	5.98%	5.17%	0.92%
% on Stimulant	43.30%	43.20%	-

Note: Income is measured in the following ranges: <$25,000; $25,000-$35,000; $35,000-$50,000; $50,000-$75,000; $75,000-100,000; $100,000-$130,000; $130,000-$150,000; >$150,000. IQ is reported for the most recent cognitive testing visit prior to EEG.

Results of Latent Profile Analysis

Table 2 contains the fit statistics for each of the five profile solutions. Based on fit indices alone, it appeared the four-profile solution was the best fit to our data, however, this solution extracted a very small sample, which was not theoretically meaningful. The greatest drop in the information criterion indices occurred between the one and two-profile solution, and the neighboring three class solution did not significantly improve fit over the two-profile solution (Nylund-Gibson & Choi, 2018). We opted to retain the two profile-solution (see Figure 1). Visual inspection of ADHD adolescent profiles indicated: 1) an emotionally-dysregulated profile characterized by low effortful control and high negative affect, such as anger, sadness, and fear (n = 117) and 2) an emotionally-regulated profile with low effortful control but scores for negative and positive affect scales all within 0.5 standard deviations of the typically-developing sample mean (n = 225).

Table 2.

Fit Statistics for Latent Profile Analysis of the Modified Parent-Rated Early Adolescent Temperament Questionnaire-Revised

# of profiles	AIC	BIC	SaBIC	CAIC	AWE	LL	Smallest Class Count	Smallest Class p	Entropy	VLMR-ALRT	VLMR-ALRT p
1	3167.48	3215.79	3171.42	3229.79	3236.79	−1569.74	233	1.00		--	--
2	2859.53	2935.45	2865.72	2957.45	2968.45	−1407.76	116	0.49	0.79	316.69	0.00
3	2787.35	2890.88	2795.79	2920.88	2935.88	−1363.67	26	0.11	0.79	86.20	0.65
4	2707.18	2838.32	2717.88	2876.32	2895.32	−1315.59	11	0.05	0.83	94.01	0.01
5	2666.92	2825.67	2679.87	2871.67	2894.67	−1287.46	11	0.05	0.81	55.00	0.54

Note: LL = log-likelihood; BIC = Bayesian information criterion; SABIC = sample-size adjusted BIC; CAIC = consistent Akaike information criterion; AWE= approximate weight of evidence criterion; VLMR-LRT = Vuong-Lo-Mendell-Rubin adjusted likelihood ratio test.

Figure 1.

shows factor score means for each ADHD temperament-based groups from the latent profile analysis. For display purposes, scores are standardized against a sample of 109 typically-developing children whose parents also completed temperament ratings at the same years as parents in the ADHD group. Deviations above zero indicate greater levels of the trait compared to typical development, while deviations below zero indicate less of the trait than the typically-developing group. Standard errors are included but are small and not visible beyond the point markers. The regulated group consistent of 225 children with ADHD whose scores are generally within 0.5 SDs of the typically-developing groups, while the Dysregulated groups consistent of 117 adolescents with ADHD who showed more than 1 SD higher scores on trait negative affect.

ADHD Diagnostic Group Comparisons

Task Performance.

As expected, children with ADHD had lower no-go accuracy (F(1,267) = 7.50, p = .007, d = .41) and more variable go trial reaction times (F(1,267) = 9.92, p = .002, d = .44) than their typically-developing peers. There was no group difference in mean RT (F(1,267) = .004, p = .953, d = .01). There was a main effect of condition for no-go accuracy (F(2,533) = 9.35, p < .001), go reaction time (F(2,532) = 35.02, p < .001), and go reaction time variability (F(2,532) = 21.77, p < .001). Both groups were more accurate in the positive valence condition than the negative (ADHD d = .33; typically-developing d = .18) or neutral (ADHD d = .16; typically-developing d = .12) conditions. Both groups had their fastest reaction times in the positive condition and their slowest RTs in the negative condition. Both groups had more variable RTs in the negative valence condition compared positive (ADHD d = .27; typically-developing d = .22) and neutral (ADHD d = .21; typically-developing d = .21) conditions. There were no significant group x condition interactions for no go accuracy (F(2,533) = .434, p = .65), RT (F(2,532) = 2.65, p = .072), or RT variability (F(2,532) = .20, p = .819).

ERN.

There was no discernable main effect of diagnostic group for the ERN (F(1,171) = .332, p = .565, d = .08). There was also no meaningful main effect of task condition (F(2,291) = .208, p = .812) and no condition x diagnostic status interaction (F(2,291) = .132, p = .876).

Pe.

There was a robust main effect of diagnostic group for the Pe (F(1,201) = 8.26, p = .004). ADHD adolescents showed a blunted Pe when compared to typically-developing controls (d = .40). There was also a main effect of task condition (F(2,276) = 5.36, p = .005) showing that participants had an enhanced Pe in the positive valence condition. Main effects were qualified by a condition x diagnostic status interaction (F(2,276) = 3.29, p = .039) such that typically-developing controls had an enhanced Pe in the positive condition compared to negative (d = .24) and neutral (d = .14) conditions. In contrast, ADHD adolescents showed no change across conditions as the positive condition were similar to both the negative (d = .05) and neutral (d = .02) conditions. See Figure 3 for the Pe condition x diagnostic status interaction plot.

Figure 3.

shows diagnostic group differences in ERPs. Panels A and B and C show grand average ERN and Pe waveforms at Cz in all task conditions for control and ADHD participants. Panel D shows the ADHD diagnostic group effects on Pe amplitude in all emotion conditions.

Temperament Group Comparisons

Task Performance.

Findings for no-go accuracy and RT paralleled results using diagnostic groups. Main effects of task condition were robust for no-go accuracy (F(2,531) = 11.02, p < .001), RT (F(2,530) = 28.25, p < .001), and RT variability (F(2,530) = 19.50, p < .001). The pattern of findings followed that seen with diagnosis. All groups had a higher accuracy in the positive valence condition compared to negative and neutral conditions. All groups had the quickest reactions in the positive condition and the slowest in the negative condition. All groups had more variable reaction times in the negative condition compared positive and neutral conditions. There was no main effect of temperament group for no-go accuracy (F(2,266) = 2.14, p = .119) or go RT (F(2,266) = .855, p = .426) but there was for RT variability (F(2,266) = 5.47, p = .005). Both temperament groups produced more variable RTs when compared to typically-developing controls (Regulated d = .16; Dysregulated d = .22). Group x condition interactions were not significant for no-go accuracy (F(4,531) = .385, p = .819), RT (F(4,530) = .866, p = .484), or RT variability (F(4,530) = .346, p = .847).

ERN.

There was not a significant main effect of temperament group (F(2,170) = .952, p = .388). or task condition (F(2,287) = .817, p = .443) for ERN. However, effects were qualified by a significant condition x temperament group interaction (F(4,287) = 2.55, p = .039). Post-hoc between-group comparisons found that the dysregulated ADHD adolescents showed a blunted ERN in the negative condition (p = .026, typically-developing d = .16, regulated d = .14) and the emotionally regulated group was blunted in the neutral condition (p = .043 , typically-developing d = .17; dysregulated d = .21) when compared to other groups. In addition, we considered within-group patterns of change across conditions. Whereas the emotionally regulated ADHD group (p = .313) and typically-developing controls (p = .981) had a stable ERN across conditions, the dysregulated group (p = .019) produced blunted ERNs in the negative (d = .19) and positive conditions (d = .14) when compared to the neutral condition. See Figure 4 for the ERN condition x temperament group interaction plot.

Figure 4.

shows temperament differences in ERPs. Panels A and B and C show grand average ERN and Pe waveforms at Cz in all task conditions for typically-developing, regulated, and dysregulated participants. Panel D shows differences in ERN amplitudes observed for the temperament-based ADHD groups across emotion conditions.

Pe.

Consistent with findings using diagnostic group, there was a significant main effect of task condition (F(2,275) = 3.05, p = .049). Participants had an enhanced Pe in the positive condition compared to negative or neutral. There was not a significant effect of temperament group for the Pe (F(2,201) = 2.74, p = .067), but given the marginal p-value we note that pairwise effects did show that the dysregulated ADHD adolescents had a blunted Pe compared to typically-developing controls. The regulated ADHD adolescents fell in between the two other groups but did not significantly differ from either. There was no condition x diagnostic status interaction (F(4,275) = .548, p = .700).

Discussion

The current study used a large, well-characterized sample of adolescents with and without ADHD to integrate trait-based measures of emotion/temperament, experimental cognitive design, and neural markers of error monitoring. Latent grouping identified two temperament-based subgroups within the ADHD sample— an emotionally-regulated and emotionally-dysregulated group— with partially distinct neural responding to errors. Differences in neural response were related to an interaction of trait and state variation in emotion. Results clarify how transdiagnostic difficulties in error monitoring apply in the context of a neurodevelopmental disorder that spans the internalizing/externalizing framework and suggest that considering temperament-based variation in ADHD can add clarity to studies of the neural correlates of the disorder and help to achieve RDoC aims of creating a biologically-informed nosology.

Dimensionally-based models of psychopathology, including RDoC, increasingly emphasize the need to consider features outside of traditional DSM domains, including specific efforts to integrate the broad DSM framework with personality/temperament features (Hankin et al., 2017; Kotov et al., 2017; Kotov et al., 2021). Yet, to date there is little agreement about how to best understand ADHD in terms of temperament or personality variation, particularly during adolescence. Here, we identify two separate temperament groups within ADHD during adolescence: emotionally regulated and emotionally dysregulated. The emotionally regulated group had low effortful control (consistent with ADHD symptoms) but otherwise appeared similar to typically-developing children in terms of temperament variation. In contrast, the emotionally dysregulated group included adolescents with low effortful control and significantly higher ratings of irritability, depressed mood, and fear than their typically-developing peers. Results are broadly consistent with prior work in this sample at younger ages (Karalunas et al., 2014) and suggest that a substantial proportion of adolescents with ADHD experience difficulties with regulation of negative affect broadly, and their diagnostic picture would be enhanced by considering dimensional traits outside of traditional symptoms.

While much of the recent literature has emphasized irritability or anger dysregulation as a primary transdiagnostic emotional risk factor (Evans et al., 2017; Finlay-Jones et al., 2023; Shaw et al., 2016; Vidal-Ribas et al., 2016), the patterns here suggest a broader pattern of emotion dysregulation across multiple types of negative emotion. A number of explanations are possible. First, it may be that the emphasis on irritability in the literature is overstated and considering a broader negative affect phenotype more accurately captures clinically-relevant emotional variation. That said, even if dysregulation of multiple types of negative affect are present, it is still possible that irritability is uniquely important for predicting long-term risk (Karalunas et al., 2019; Leibenluft et al., 2006; Stringaris et al., 2009) and so should remain the focus of study. Another possibility is that clinical risk is conveyed not by a specific type of negative emotion but by poor emotion differentiation (i.e., difficulty distinguishing between types of negative affect) that drives the general pattern of elevated negative affect seen here (Schreuder et al., 2022).

While the profiles observed are partially consistent with prior work in this sample at younger ages, there are also important contrasts. Most notably, the current study did not identify a group of children with ADHD and dysregulated positive affect (i.e., high surgency). Results suggest that the between-person patterns of temperament variation in ADHD shift across development from more distinct types of early dysregulation (i.e., positive and negative valence) to a more general regulated/dysregulated distinction. Such findings are consistent with prior work that has tended to identify groups defined by surgency in younger children, as well as with evidence that positive affect dysregulation and irritability or aggression may stem from similar disruptions in neural circuits (Vogel et al., 2022).

Although temperament is at least partially stable over time, findings also suggest that multiple pathways of early temperament disruption may converge by adolescence into a smaller number of emotional profiles. This pattern is also consistent with the clinical literature implicating transactional developmental processes, such as negative peer interactions, academic difficulties, and strained parent-child interactions, as providing a key link between ADHD and later negative affect comorbidities. Regardless of the type of early dysregulation, negative interactions over time may lead to increasing negative affect that coalesces at the trait level, contributing to the convergence of negative and positive dysregulated profiles during adolescence.

Clarifying how ADHD may be reflected in the context of temperament/personality constructs is directly relevant to RDoC efforts to redefine the structure of psychopathology based on these domains. One neglected question in much of this work is whether heterogeneity at the level of temperament is also related to neurobiological variation. Here, we addressed this question in the context of error processing because of its position at the nexus of cognitive control and positive/negative valence systems of the RDoC framework. At the level of DSM diagnostic groups, results identified blunted Pe amplitudes for ADHD-diagnosed adolescents when compared to typically-developing controls, but no difference in ERN amplitude. The pattern was generally consistent with prior literature, particularly studies that have also found Pe blunting in the absence of ERN differences in ADHD (Albrecht et al., 2008; Meyer et al., 2018; Groen et al., 2008). Impairments in later stages of error processing may help explain group-level differences in performance. However, the group-level results did not accurately reflect within-group variation based on temperament.

When temperament groups were considered, both the regulated and dysregulated groups showed blunted Pe compared to typically-developing children. However, the lack of ERN effects at the group level belied meaningful variation within the ADHD group. Between-group comparisons found that the dysregulated ADHD group had blunted ERN in the negative condition compared to other groups, whereas the regulated group had a blunted ERN in the neutral condition. However, consideration of within-group changes across conditions is also informative. Both the emotionally-regulated ADHD and typically-developing groups showed stable ERN across task conditions (they did not differ from themselves across conditions), while the dysregulated group showed a blunted ERN in emotional conditions when compared to their own ERN in neutral conditions. The neutral task condition included faces that did not show specific emotions; however, it is notable that such conditions are perhaps better interpreted as “ambiguous” (rather than neutral). Findings suggest that overt emotion information may reduce error processing in the emotionally-dysregulated ADHD group, perhaps by using resources that might normally be directed to top-down control (Alperin et al., 2017; Balogh et al., 2017; Karalunas et al., 2020). Results provide evidence that considering trait-level variation in ADHD will be important for understanding its neural correlates. Findings are consistent with a small but growing literature pointing to importance of state-based emotional manipulations and their interaction with trait-based individual differences for understanding cognitive and neural variation in ADHD (Nuñez-Estupiñan et al., 2022).

The primary study aim was to investigate variation within the ADHD group; however, the more general findings for effects of emotional valence are also notable. Across groups, response accuracy and reaction time were improved in the positive valence condition when compared to the negative and neutral conditions. Similarly, the Pe was enhanced in positive contexts. Overall, the patterns of findings indicate that later error processing and behavioral adjustment appear to be enhanced in positive contexts as opposed to negative or neutral ones and is consistent with other literature similarly suggesting that positive affect enhances cognition (Nummenmaa et al., 2015; Bayer et al., 2014; Leppänen et al., 2004).

This study provides important insights into the neural markers of error-monitoring in psychopathology and is the first to additionally investigate the role of temperament groupings, yet there are several limitations to consider. First, while our participant sample is heterogeneous in many ways that allow for more generalizability to the population and utilizes a community-based sampling method, it is largely Caucasian with a relatively high socioeconomic status (SES) demographic, which may limit generalizability. Additionally, future studies should replicate this novel approach to characterizing the emotional heterogeneity using temperament latent groupings in new and larger samples of children and adolescents. Future studies should also consider looking at symptoms of interest (i.e., inattention, hyperactivity, anxiety) on a continuum rather than focusing solely on diagnostic grouping criteria. While beyond the scope of the current manuscript, this would further aid in addressing RDoC goals. Finally, with consideration of different tasks, including more difficult tasks with a larger number of errors may also be informative.

Clinical implications

Results from the current study clarify how the ERN, one well-characterized and promising biomarker for psychopathology (Meyer, 2020), and other EEG-based markers of error monitoring can be used in the context of ADHD— a neurodevelopmental disorder that does not clearly fit in the internalizing/externalizing divide. Such studies can help achieve RDoC goals of linking biological systems to specific behaviorally-measured clinical phenomenon, as well as point towards alternative groups that form the basis of a revised, biologically-informed nosology. Although there is considerable work needed to inform translation into standard clinical care, the findings highlight the importance of assessing dimensional temperament features in clinical assessment of ADHD, as well as how considering trait-based emotional variation can eventually contribute to creation of diagnostic categories that better correspond to underlying neural response. In addition, while the stability of symptoms across contexts has typically been the focus of ADHD conceptualizations, the trait-state interactions highlight the critical role of context in ADHD symptom variation. The logical next steps of this type of work includes using revised categories or neural markers to inform personalized treatments that differentially target phases of error-monitoring to motivate behavior change. Such differences may be integrated into existing behavioral interventions and/or targeted via novel neurofeedback approaches. Current findings suggest that clinical assessment in ADHD should consider differences in both trait-based emotional responding and contextual variation.

Figure 2.

Happy Condition Experimental Run.