Discussion

In this study, we utilized linear regression models based on robust single-trial ERP analysis as implemented in the EEGLAB toolbox LIMO EEG to test for correlations between a range of well-studied ERPs and symptom and trait measures of psychopathology compatible with the transdiagnostic framework the HiTOP. Recruiting a mixed sample of 50 patients with emotional disorders undergoing 14 weeks of transdiagnostic psychotherapy as well as 37 healthy comparison subjects matched in age and sex, we assessed longitudinal correlations across the full psychopathology spectrum. In the following, we pragmatically follow a top-down discourse in that we first treat results covering the HiTOP spectrum and subfactor levels. After this, we look at ERPs which only correlated with a single or a few sub scales and therefore are of relevance to the lowest symptom and maladaptive trait level.
The most consistent result in this study was that a reduced P3b elicited to Target stimuli in the AO paradigm correlated with worse symptomatology. In fact, significant correlations between higher scores and reduced P3b was found in all psychopathology measures with two exceptions: MEDI Positive temperament - where, as expected, we found the opposite pattern - and PID36 Psychoticism. These findings were corroborated by results from the Correct response stimulus-locked Flanker ERPs elicited to Congruent stimuli. Although not significant at the corrected level, at 5% a reduced stimulus-locked Flanker P3b correlated with higher scores in several of the psychopathology measures, including MEDI total score11Note that because the P3b in the Flanker paradigm falls immediately before button press, it cannot be as reliably analysed as the P3b from the AO paradigm due to some trials overlapping.. The P3b is thought to be an index of cognitive processes such as context updating and memory processing (Luck & Kappenman, 2011; Polich, 2007). While P3b is consistently reduced in chronic schizophrenia and less consistently reduced in depression, we are not aware of studies relating P3b to specific symptoms in any of the emotional disorders included in this study (Klawohn et al., 2020; Onitsuka et al., 2013). In schizophrenia, some studies have reported associations between a reduced P3b and increased symptoms of cognitive deficits (Giordano et al., 2021; Kruiper et al., 2019). Our results are in line with these findings and support that cognitive impairment is a mainstay also of the emotional disorders. In the HiTOP, our results suggest that a reduced P3b is a marker not of subfactor or lower symptom and trait levels but of the Internalizing spectrum as a whole. Given the above as well as evidence of reductions also in Externalizing disorders, it can be speculated that a reduced P3b is a marker not only of the Internalizing, but also of the Externalizing and Thought disorder spectra in the HiTOP (Pasion & Barbosa, 2019; Patrick et al., 2006). Indeed, cognitive impairment is a symptom of most, if not all, psychiatric disorders (Etkin et al., 2013). As such, a reduced P3b could be a marker of the general p-factor sometimes included at the top of the HiTOP hierarchy (Levin-Aspenson et al., 2021). However, there is some evidence of an increased P3b in OCD (Gohle et al., 2008; Mavrogiorgou et al., 2002). This highlights the unique features of OCD in a transdiagnostic framework (see below) (Faure & Forbes, 2021). Finally, it should be noted that cognitive impairment is not only a core symptom of psychiatric disorders, but also a known side effect to treatment with psychotropic medication (Cowen & Sherwood, 2013; Paterniti et al., 1999). Even though we accounted for medication status in our analysis, 86% of the patient population was treated with at least one psychiatric medication. Given this high proportion, we can’t be certain that cognitive impairment, as indexed by a reduced P3b, were due to disorder or psychotropic medication. However, a separate analysis with a model excluding medication status did not yield stronger P3b regional effects, suggesting that medication status did not contribute to the results. Indeed, evidence indicate that the P3b is not altered by SSRI treatment, which was the most prevalent psychotropic medication in our sample (d’Ardhuy et al., 1999; Oranje et al., 2008-06-31; however, see Wienberg et al., 2010).
After the P3b, the second most consistent findings involved the response-locked Flanker ERPs, indexing conflict or performance monitoring (Larson et al., 2014). Starting with the ERN, elicited to errors, we had hypothesized than an increased ERN would correlate with higher scores, especially in the psychopathology measures indexing different forms of anxiety and symptoms related to OCD. This was based on findings of correlations between increases in the ERN and transdiagnostic dimensions related to OCD and anxiety, especially Fear-based anxiety, within the Internalizing spectrum (Gorka et al., 2017; Pasion & Barbosa, 2019; Riesel et al., 2022). Contrary to this hypothesis, however, we found consistent and strong correlations between a reduced ERN and the majority of symptom dimensions in MEDI, as well as the Anankastia, Detachment and Negative Affect maladaptive traits in PID36. This apparent discrepancy with the literature invites to several interpretations. First, a few studies have indeed found correlations between a reduced ERN and Internalizing psychopathology. Tanovic et al. (2017) found associations between a reduced ERN and symptoms of ruminations, but only when controlling for effects of anxiety. Olvet et al. (2010) found associations between depression severity and both an increased CRN and a reduced dERN, the difference wave between ERN and CRN. Our results therefore corroborate these results in finding correlations with increased scores in MEDI Intrusive Cognitions and MEDI Depression, the latter for which we also found correlations with an increased CRN as in Olvet et al. (2010) (see below). Second, owing to the nature of linear regression models as utilized in this study, regional effects can either indicate a reduced or an increased ERN, but not both simultaneously. Consequently, if the ‘ground truth’ is that both types of ERN deviations correlate with higher scores, the dominant feature ‘wins’, or, alternatively, the effects cancel out and the correlation is not significant. Perhaps this is why Seow et al. (2020) failed to find associations between the ERN and transdiagnostic measures in a community sample. In Riesel et al. (2022), correlations between increases in the ERN and an anxious-misery dimension was in a mixed sample across the OCD spectrum. In that study, the patient group had a significantly increased ERN compared to the healthy comparison group. For our present sample, we have recently shown that a sub sample based on the HiTOP subfactor Distress - containing the ICD-10 diagnoses of Depression and GAD - had a reduced ERN at baseline compared to healthy comparison subjects (Randau et al., 2023). In that study, the ERN of the Fear subfactor - containing agoraphobia, OCD, PD and SAD - was not significantly different to either the Distress or HC group. As such, our sample characteristics differs from Riesel et al. (2022) in that the group contributing the most to the psychopathology variance is not defined by an increased ERN, but rather by a decreased one. Third, it is established that the ERN is sensitive to not only manipulations of experimental factors, but also to preprocessing and analysis methods (Clayson, 2020; Clayson et al., 2021; Feuerriegel & Bode, 2022). In all paradigm-related aspects, our study closely followed established conventions in the literature. However, we cannot rule out that a desensitizing effect across sessions took place for the patient group, essentially decreasing the perceived threat of errors and thereby the ERN. That being said, patients showed a reduced ERN compared to HC already at baseline and the ERN has been shown to be stable across sessions (Olvet & Hajcak, 2009a; Randau et al., 2023). Neither can we rule out a fatigue effect from the rather long paradigm, even though such an effect has not been demonstrated (Olvet & Hajcak, 2009b). In addition, our paradigm was not longer in duration than the one used in Riesel et al. (2022) and divided into 10 rather than 6 blocks. Taken together, we do not find it likely that our divergent effects were paradigm- or study design-related. While we believe that the robust single-trial analysis method utilized in this study constitutes an improvement, we did not conduct a formal analysis comparing our methods against traditional methods. However, we can report that applying traditional baseline subtraction methods at intervals commonly reported in the literature (-500 to -300 and -200 to 0 ms pre-stimulus, respectively) did not noticeably alter the group-wise grand average ERN waveforms in terms of maximum peak amplitude or latency. In Gorka et al. (2017) on a transdiagnostic sample, OCD was an exclusion criteria and yet higher scores in a derived Fear dimension - but not in a Distress dimension - correlated with increases in a residual ERN-measure, 22Defined as the ERN activity when the CRN is ’regressed out (Gorka et al., 2017).. In our study, even though the ERN beta coefficient represents residual activity when effects of the CRN beta coefficient (and the adjusted mean) are accounted for, we also informally tested an explicit ERN-CRN difference contrast. However, results for this dERN was in the same direction as the ERN. Therefore, we must conclude that when it comes to the ERN, ‘all roads lead to Rome’ in the sense that both decreases and increases can be observed in clinical populations and that both types of deviations are associated with worse symptomatology. In this regard, it must be noted that past studies have utilized rating scales based on a categorical understanding of psychopathology and converted these into transdiagnostic dimensions through factor analysis conducted on the study sample or from weights derived from previous studies. Needless to say, into what latent dimension a given rating scale is allocated will affect the direction of correlations, if any. While neither MEDI nor PID36 are directly derived from the HiTOP, both measures are validated in large populations and index distinct transdiagnostic dimensions consistent with the HiTOP framework. In terms of the HiTOP, then, as we saw for the P3b, a reduced ERN seems to be associated with worse symptomatology across the whole Internalizing spectrum. In addition to ruminations and depressive symptoms, reductions in the ERN have been associated with symptoms and traits belonging to the Externalizing spectrum (Hall et al., 2007; Lutz et al., 2021; Pasion & Barbosa, 2019). However, we found no significant correlations between the ERN and the two Externalizing traits in PID36 (Disinhibition and Antagonism). Then again, our sample did not include Externalizing disorders. Indeed, scores for Antagonism were comparably low and did not differ between the two groups. Finally, in the HiTOP, the placement of OCD within the Internalizing spectrum is not fully established, with results indicating that symptoms cross-load on the Fear subfactor within the Internalizing spectrum but also on the Thought-disorder spectrum (Faure & Forbes, 2021). As such, we can raise the possibility that a reduced ERN is specific to the Internalizing and the Externalizing spectra, as conceptualized in the HiTOP, while an increased ERN is a specific marker of some other construct encompassing both anxiety symptoms contained in the Fear subfactor as well as symptoms related to the Thought disorder spectrum. The uniqueness of OCD in terms of ERP abnormalities is also supported by associations with an increased P3b discussed above (Gohle et al., 2008). We can conclude that more studies are needed to understand the associations between the ERN and psychopathology, especially studies looking to discern divergent effects in different patient populations and using validated transdiagnostic measures.
Results for the CRN, elicited to correct responses, were somewhat more specific and in the expected direction. Firstly, we can corroborate results from Riesel et al. (2022) in finding correlations between an increased CRN and PID36 Anankastia. PID36 Anankastia must be considered to capture much of the same psychopathology as the dimensions Compulsiveness and Personal standards examined in Riesel et al. (2022). In addition, we can corroborate results from Olvet et al. (2010) in finding correlations between an increased CRN and depressive symptoms as indexed by MEDI Depression. Indeed, we find that increased CRN correlates strongly with transdiagnostic dimensions which can be considered to load unto the HiTOP Distress subfactor (MEDI Depression and PID36 Negative Affect, but also PID36 Detachment33We note that Detachment is in itself a spectrum in the HiTOP.), whereas it correlates more weakly or not at all with Fear subfactor dimensions, e.g., MEDI Autonomic Arousal, Neurotic Temperament, Social Anxiety44Here, correlations were present but considerably weaker than for MEDI Depression and PID36 Negative Affect., Somatic Anxiety and PID36 Avoidance. As such, we find some evidence of an increased CRN as a marker of symptoms and traits in the HiTOP Distress subfactor. At first sight, this contradicts the results from our above-mentioned study, where the Distress subfactor - containing patients with a primary ICD-10 diagnosis of depression and GAD - had a significantlyreduced CRN compared to healthy comparison subjects (Randau et al., 2023). However, group-level differences between HiTOP subfactors based on primary ICD-10 diagnosis do not necessarily translate to correlations with transdiagnostic symptomatology. While the above-mentioned Distress dimensions must be considered to be a core part of both ICD-10 Depression and GAD, both HCs as well as patients with disorders which would be allocated to the Fear subfactor contributed to the correlations. Therefore, it is not a contradiction to state that categorical diagnoses associated with Distress show a reduced CRN compared to healthy comparison subjects, but when considering the full spectrum of psychopathology, an increased CRN correlates with dimensions which primarily load unto the Distress subfactor. It is likely that other factors which are common to Distress disorders and not captured by our psychopathology measures, such as cognitive impairment, influence to reduce the CRN. Some evidence of a negative (reduced) effect of cognitive impairment on the response-locked Flanker ERPs exist (Eppinger et al., 2008; Simó et al., 2018; Swainston et al., 2021).
ERN and CRN are followed by the Pe and Pc, which are believed to index the conscious awareness of correct and error responses, respectively (Overbeek et al., 2005; Wessel, 2012). In the main analysis, only correlations between a reduced Pe and MEDI total survived the corrected level of 0.1%. Post-hoc , a reduced Pe correlated strongly with increased scores in MEDI Avoidance and Social Anxiety and more weakly with Autonomic Arousal, Intrusive Cognitions and Somatic Anxiety. Even though PID36 total scores did not survive the main analysis, a look at the sub scales reveals significant but considerably weaker correlations with Anankastia, Disinhibition, Negative Affect and Psychoticism. Given this, and the absence of strong correlations with Distress subfactor dimensions and traits, it is possible that a reduced Pe is a marker of the HiTOP Fear subfactor rather than of the whole Internalizing spectrum. Again, OCD would not fit well into Fear in having an increased Pe compared to healthy comparison subjects (Bellato et al., 2021). As for Pc, correlations with MEDI and PID36 total scores were not significant in the main analysis. Post-hoc , we find quite specifically that a reduced Pe correlates only with increased scores in PID36 Negative Affect. This speaks for the specificity of the Correct response-locked Flanker ERPs in that both CRN and Pc correlate with Distress symptoms, the latter perhaps at the lowest symptom and trait level.
Having discussed the major results spanning HiTOP spectra and subfactor levels, we now turn to more specific sub scale results representing effects at the symptom and maladaptive trait level. It should be noted that the following results were not tested at the main analysis corrected level of 0.1% but at the more conventional 5%. Speaking for their specificity, neither of the MMN measures (with the exception of cMMN with LPFS) nor the Novelty P300 elicited to Distractor stimuli in the AO paradigm or the stimulus-locked Flanker N2 yielded significant, strong correlations with any of the four major psychopathology measures (K10, LPFS, MEDI and PID36) at either level.
The Novelty P300 is rightfully distinguished from the more typical P3a elicited to deviant stimuli in the Unattended oddball paradigm (Polich, 2007). Perhaps due to some confusion in the literature, we are not aware of studies examining correlations between the Novelty P300 and measures of psychopathology. We find, quite specifically, that a reduced Novelty P300 at Fz, FCz and Cz correlates with increased scores in PID36 Negative Affect. For MEDI Depression, a reduced Novelty P300 correlates with higher scores, but weaker and more posterior at CPz and Pz. We note that this effect of central versus posterior effects might to some extent be due to correlations with latency in addition to amplitude. Nevertheless, PID36 Negative Affect and MEDI Depression can be considered to index roughly similar symptom and trait level dimensions. As such, given the absence of other correlations, we can postulate that a reduced Novelty P300 is a marker of a negative affect dimension at the lowest level of the HiTOP hierarchy, or alternatively, of the Distress subfactor.
For LPFS at the main analysis level, increases in the latter part of the cMMN as well as in the following combined deviant dP3a correlated with higher scores. This curiously indicates an association between MMN/dP3a and measures of personality functioning or maladaptive traits. MMN is an index of pre-attentive auditory processing and is reduced in chronic schizophrenia. However, studies examining the role of MMN in personality disorders and associated symptoms are, to our knowledge, rare or inconclusive. Increases in MMN have been associated with schizotypal and antisocial personality disorders as well as with treatment-resistant depression when controlling for comorbid borderline personality disorder (He et al., 2010; Liu et al., 2007). Given this potential connection between MMN and the personality disorders, it is interesting to have a look at the PID36 sub scales. Here, increased MMN correlated with Detachment, Disinhibition and more weakly with Psychoticism and Anankastia. Decreased MMN correlated with Antagonism and Negative Affect, the latter consistent with reduced MMN in depression (Tseng et al., 2021). These results stand in an interesting contrast to results for the MEDI sub scales, where the only significant correlation was between increased MMN and Social Anxiety. Taken together, we find compelling evidence of MMN being a quite specific marker of maladaptive personality traits loading unto the Internalizing (Detachment, Negative Affect), Externalizing (Antagonism, Disinhibition) and Thought-disorder (Psychoticism) spectra in the HiTOP. Such a specificity was not seen in the following dP3a, which, even though an increased dP3a correlated with LPFS, was equally related to both MEDI and PID36 sub scales. Curiously, dP3a was most strongly associated with with MEDI Avoidance and PID36 Antagonism where a decreased dP3a correlated with higher scores. However, we believe it is beyond the scope of this paper to discuss similarities between these two measures.
The last ERPs we consider in this discussion are the stimulus-locked Flanker N2s elicited to correct response to congruent and incongruent stimuli, respectively. Like the CRN and the ERN, the Flanker N2 is an index of conflict monitoring and cognitive control (Larson et al., 2014). First, we note that only the Flanker N2 elicited to congruent stimuli yielded strong correlations. Second, while both the CRN and the ERN correlated with maladaptive traits as indexed by LPFS and PID36, as well as with symptom dimensions as indexed by K10 and MEDI total score, the Flanker N2 correlated only with a few sub scales in MEDI. Specifically, a reduced Flanker N2 correlated with higher scores in Intrusive Cognitions and Traumatic Re-experiencing. Even though the dimensions in MEDI are distinct and validated, these two sub scales must be considered to capture closely related psychopathology. As such, we can postulate that a reduced Flanker N2 to congruent stimuli is a marker of a single or a few specific symptom dimensions at the lowest level of the HiTOP hierarchy. However, it is unclear to us to what extent Intrusive Cognitions and Traumatic Re-experiencing loads unto the Internalizing and Thought-disorder spectra (Kotov et al., 2020). We also saw that an increased Flanker N2 to congruent stimuli correlated with increased scores in MEDI Neurotic Temperament, a core part of the Internalizing spectrum (Watson et al., 2022).
Our study has several limitations. First, our setup did not allow us to infer to what extent treatment and group influenced the correlations. As such, we cannot rule out that our results are driven by correlations which are the strongest in, e.g., patients at baseline. Second, while we believe that we controlled for false positives with the conservative level in the main analysis, we did not define how large a significant region shall be for it to be considered a true correlation. Add to this that we found several significant effects at regions not corresponding to traditional ERP evaluation windows and channels. Rather than considering only the strongest correlations clearly corresponding to traditional ERPs, we opted to describe all significant regions above some arbitrary visual threshold and to quantify these correlations in terms of strong or weak. It is likely that with more data, some of these regions would either become larger or vanish.
To our knowledge, this is the first comprehensive examination of the associations between ERPs and transdiagnostic psychopathology. The ERPs included in the study are easily measured in a clinical setting and index pre-attentive auditory processing, cognition and performance monitoring. Some ERPs, e.g., the MMN, appear to be exclusively related to maladaptive personality traits at the lowest level of the hierarchy, whereas others, e.g., the P3b, cut across and are related to entire spectra or even the general p-factor. The ERN remains elusive in that we found solid evidence of a reduced ERN correlating with higher scores at the spectrum level. Conversely, increases in the CRN correlated with worse symptomatology at the subfactor level, results which are in line with the literature. In showing that abnormalities in such basic brain processes are associated with transdiagnostic symptoms and traits at several levels of the HiTOP hierarchy we have taken yet another small step toward biomarkers in psychiatry. We have also shown the advantages of utilizing a consistent framework such as the HiTOP, which allowed us to pinpoint associations between ERPs and diagnostic measures to specific levels in the hierarchy. While we did not directly compare our results against traditional ERP methods, we can state that robust single-trial ERP analysis as implemented in LIMO EEG is an excellent tool for a pragmatic analysis of ERP features across channels and time frames. In future steps, after replication, machine learning and related advanced method would be obvious candidates in translating ERP features into transdiagnostic symptom profiles at the subject level (Nielsen et al., 2020).