Multicomponent Multimethod Assessment of Emotional Change in Psychotherapy Research: Initial Validation of a Neurobehavioral Paradigm

Both in imaging research, for example when relating the activity of the amygdala to negative stimuli in borderline personality disorder (Schulze et al., 2016a, b) and in psychotherapy research, for example when relating emotional change to outcome (Peluso & Freund, 2018), the great variety of experimental methods may limit generalizability and comparability across samples. It was argued that broad categories and processes assessed in a variety of experimental tasks overlook the individual anchoring of the functional emotional response (Pascual-Leone et al., 2016). Speed and colleagues (2020, p. 2) noted that for studies on emotion regulation that “one broad limitation of prior emotion regulation studies is that they have primarily relied on standardized emotional words”: it has been demonstrated that the goal-relevance - or relevance to the Self – of a stimulus impacts the emotional response (Ellsworth & Scherer, 2003). One study has shown that idiographic stimuli are more effective in eliciting emotional reactivity compared to standardized stimuli, for specific emotions such as anger and sadness (Kuo et al., 2014). Therefore, individualizing the assessment of emotional change is needed to capture the functional change, which should be done under certain methodological (under the conditions highlighted by Pascual-Leone et al., 2016).

A clinically relevant context of emotion-elicitation was described by Whelton and Greenberg (2005). These researchers elicited self-critical expressions anchored in the client’s episodic memory within an experiential two-chair dialogue, as used in Gestalt and emotion-focused psychotherapy approaches. A series of studies have shown the relevance for the activation of specific self-critical processes pertaining to the individual’s goals, needs and concerns (Kramer et al., 2018; Kramer & Pascual-Leone, 2016; Nardone et al., 2022). In particular, Whelton and Greenberg (2005) showed that the emotional expression of self-contempt while criticising the Self was related with the intensity of depressive symptoms, rather than the expressed cognitive content. Consistently, Kramer et al. (2016) showed in the same experiential paradigm that expressed self-contempt interfered with the individual’s access to adaptive and healthy needs (see also Nardone et al., 2022). In this context, self-contempt – or cold self-hatred – may be understood as a maladaptive secondary emotion that is central in diverse psychopathologies and which can be elicited reliably in such experiential two-chair dialogues (Beuchat et al., 2023). In a process-outcome analysis on brief treatment for BPD, Sallin and colleagues (2021) showed that in-session expressed self-contempt, as rated on video and audio material, was associated with the intensity of borderline symptoms and negatively predicted the course of the patient’s therapeutic alliance over time, while it positively predicted the course of the therapist’s therapeutic alliance over time. These observations are consistent with earlier research which related the intensity of personality pathology with low self-esteem and self-disgust (Winter et al., 2017) and echo other studies that point to the importance of self-contempt in clinical presentations of BPD (Krawitz, 2012a, b). Finally, research shows that these emotional processes are linked with the in-session activation of specific cognitions – so-called hot cognitions – related with low self-esteem (Hochschild Tolpin et al., 2004; Kramer et al., 2022; Santangelo et al., 2017; Shafir et al., 2017; Zeigler–Hill & Abraham, 2006).

From a neurofunctional perspective, emotional responses to specific stimuli evoking such self-critical content may be manifold. In tasks evoking self-contempt and presumably associated cognitive and emotional processes, researchers demonstrated a heightened activity in the prefrontal regions, for instance the medial prefrontal cortex (mPFC), because it appears to be commonly activated during generic emotional tasks related to awareness. In addition, the dorsolateral PFC (dlPFC) my be involved due to its role in emotion regulation (Phan et al., 2002, 2004a; Schmahl et al., 2018a, b). The orbitofrontal PFC (OFC/ventromedial PFC) may play a role in social and emotional behaviour, through its projections to the amygdala and anterior cingulate cortex (ACC) while this latter and other parts of the cingulate cortex may be important in general types of emotional processing (Schmahl et al., 2018a, b). The insula whose activation seems to be associated with the experience and processing of negative emotions such seeing facial expressions of disgust (Calder et al., 2001; Phillips et al., 1998; Schmahl et al., 2018a, b) may be relevant to study responses to self-contemptuous stimuli. Self-contemptuous responses may fundamentally be linked with processes of social exclusion and ostracism, interpersonal processes known to affect clients with BPD. In a study using the trust game, King-Casas et al. (2008) showed that the insula activity was associated with felt emotions of relationship distrust and lacking access to social norms in these patients when compared with healthy controls. Concerning subcortical regions, the amygdala is highly relevant as specific salience detector, whose activation is frequent amongst several types of emotional stimuli, especially disgust (Costafreda et al., 2008; Fitzgerald et al., 2006; Schmahl et al., 2018a, b; Sergerie et al., 2008). Another subcortical region of interest to the study of self-contempt may be the ventral striatum, whose activation seems to be associated with emotional intensity and stimuli self-relatedness (Phan et al., 2004b; Schmahl et al., 2018a, b). We may also assume that the hippocampus, also subcortical, seems to play a crucial role in emotional memory (Bouton et al., 2006; Ji & Maren, 2007; Milad et al., 2007; Schmahl et al., 2018a, b). Negative stimuli pertaining to the Self of the individual may have numerous neurobiological underpinnings, in severe personality pathology, but also in healthy controls. A challenge is the diversity of tasks used to assess the neurofunctional responses.

In a pilot study, Kramer et al. (2018) demonstrated feasibility of an integrated multimethod assessment of emotional change using idiographic models, with a focus on change in emotional arousal. In a pre-post assessment paradigm across brief treatment, these researchers observed in clients with BPD that the decrease in self-reported emotional arousal was associated with the brain responses to the idiographic self-critical words emitted by the client previously in a two-chair dialogue, in particular in brain regions associated with the processing of information relevant to the Self (i.e., the putamen and the bilateral precuneus). These changes predicted symptom reduction across the brief treatment. These changes were observed to be associated with the activation of a specific emotion category, observed verbally and non-verbally in the psychological assessment, when the client criticizes oneself: self-contempt (Kramer, Renevey et al., 2020a, b). As suggested by Kandel (1999) when reformulating the aims of psychiatry, the client’s words impact not only processes in the brain, but may represent the individual’s core aims or needs in the symbolization process across their life, and in psychotherapy.

The present study has two goals. Firstly, we aim to provide first validation data for micro-fluctuations and macro-stability for healthy controls over four months of pre-post observation. Our design combines self-report assessments, observer-rated assessments and fMRI to evaluate emotional changes in healthy controls. Secondly, we aim to illustrate the multicomponent assessment approach of emotional change with a client presenting with BPD undergoing brief treatment.

We assume for the first aim above significant within-subject intra-task micro-fluctuations (within 30 min) and between-task stability (over four months). In addition, we assume for the neurofunctional aspects that being exposed to individualized stimuli with high emotional meaning results in a greater neural activation in the amygdala, insula, anterior cingulate cortex (ACC) or prefrontal cortex (PFC) than being exposed to standard stimuli. We also assume that a higher level of expressed self-contempt in a two-chair dialogue is related with a higher neural activity when exposed to the same – the individual’s own - words in an fMRI task one week later. We expect this to take place in brain regions associated with either emotional responses (e.g., as the insula or amygdala) or to the self (e.g., as the medial prefrontal cortex, the insular cortex or the insula).

Female undergraduate students (N = 28) (to avoid gender bias (see Wrase et al., 2003), aged between 19 and 31 years (M = 22.29; SD = 2.65), were recruited in University classrooms. All of them agreed to participate and have their data used for research. Inclusion criteria were being aged between 18 and 35 years old and speaking French while exclusion criteria were either having a psychiatric diagnosis or taking any psychopharmacology. Participants underwent two assessment sessions including a self-critical task (two-chair dialogue described below). Out of the 28 participants, a sub-sample of n = 24 participated in two fMRI assessments in addition to the psychological assessments (roughly one week after the latter each time), in which the self-critical words from the previous task (the individualized stimuli) were presented, along with non-individualized negative stimuli (Kherif et al., 2011). This research was approved by the competent institutional ethics board (2017–02167) and took place as part of a larger psychotherapy research trial (design see Kramer et al., 2020a, b).

Emotional changes were assessed using three methodological perspectives: (a) self-report measures, (b) observer-rated assessment of video data from the two-chair dialogue and (c) neurofunctional imaging assessments. We assessed two fundamental emotion components: (a) emotion arousal, and (b) self-contempt (and self-esteem), using a combination of the three methods.

All participants underwent two sessions of a two-chair dialogue focusing on the elaboration of self-criticism – four months apart. In the present study, this therapeutic intervention from Gestalt and emotion-focused therapy is used as an emotion-eliciting research context designed to increase emotional arousal, to process self-criticism that is most relevant to the individual (Greenberg, 2002; Kramer & Pascual-Leone, 2016; Stiegler et al., 2018; Whelton & Greenberg, 2005). Self-reported assessments were given at three time-point during task: (a) at the beginning, (b) right after the imagination task and (c) in the end. The sessions were video-taped for the observer-rated assessment.

In this task, after the assessment (a), the participant is first (1) invited to imagine a personal situation of failure of their life, as vividly as possible (without reporting verbally; followed by assessment b). Then, for step (2), the interviewer asks them to change chair onto a “self-critical” chair. On this chair, the participant is asked to adopt the stance of the inner self-critical voice and express self-criticism to the self, as imagined on the initial – experiential – chair. For step (3), the participant is invited to move back on the initial chair and to explore their current emotional reaction to the self-criticism (for a complete description of the two-chair dialogue used in research, see Kramer & Pascual-Leone, 2016).

An assessment in the fMRI environment was organized for the participants roughly one week after the psychological (i.e., two-chair dialogue) assessment. The task consisted of two series of 20 words which served as the contrast tested in the current study: the individual’s own self-critical words vs. generic and validated negative words. The former were extracted from the step 2 of the two-chair dialogue when the participant was invited to criticize oneself. For example, these could be words like “failure”, “dumb”, “ugly” or “loser”. These words were projected during the fMRI in a block design. Each block contained ten images, each image containing one of the twenty words was projected for 2 s. The total of 20 blocks were pseudo-randomized and the design was repeated two times in order for the twenty images in each category to be projected.

The fMRI data was acquired on a 3-T MRI scanner with a 64-channel head coil using a 2D EPI sequence. The acquisition parameters were as follows: voxel size 3 × 3 × 3 mm3, (slice thickness = 2.5 mm with a distance factor of 20% that makes it = 3 mm); 45 slices; slice TR = 80 ms; volume TR = 80ms x 45 = 3.6s; TE = 17.4 ms; phase oversampling = 12%; flip angle = 90°; BW = 260 Hz/Px; GRAPPA with acceleration factor = 2 along phase direction. The structural MRI data consisted of T1-weighted MPRAGE images (TR = 2,000 ms; TI = 920 ms; TE = 2.39ms; α = 9°; BW = 250 Hz/Px; read-out in inferior–superior direction; FoV = 256 × 232 mm; 176 slices) at 1 mm resolution; GRAPPA with acceleration factor = 2 along phase direction. A whole-brain analysis was used. EPI images were realigned to the participant’s average image across runs, corrected for spatial distortions using the SPM field-mapping tools (Hutton et al., 2002). The parameters of registration to standardised Montreal Neurological Institute (MNI) space were calculated on the anatomical image and the default settings of the ‘unified segmentation’ framework followed by the diffeomorphic registration algorithm DARTEL (Ashburner, 2007). The spatial registration parameters were applied to the functional time series co-registered to the corresponding individual’s anatomical scan. Prior to statistical analysis, a spatial smoothing with a Gaussian kernel of 8 mm full-width at half-maximum was applied. Our experiments used the methodology of blood oxygen level-dependent (BOLD) imaging. The pre-processing and statistical analyses of the MRI data were performed with Statistical Parametric Mapping (SPM12) fMRI software (Wellcome Trust Centre for Human Neuroimaging, http://​www.​fil.​ion.​ucl.​ac.​uk/​spm) running on MATLAB R2021a (Mathworks, http://​www.​mathworks.​com).

Within-task Pearson correlations between the variables SAM, SSES and ESC were all non-significant for assessment at pre (-0.13 < r < 0.03) and at post (-0.14 < r < 0.03). Within-task Pearson correlations between the three emotional variables and the intensity of borderline symptomatology were all non-significant for assessment at pre (-0.13 < r < 0.18) and at post (-0.22 < r < 0.32). Pre-post within-measure Pearson correlations were all significant at p = 0.01 (0.48 < r < 0.81). As expected, we found no difference in borderline symptomatology between pre- (M = 2.14, SD = 1.88) and post- (M = 1.46, SD = 1.77), t(27) = 1.78, p = 0.09), and all means were below the clinical threshold.

Self-rated emotional arousal was tested using a Paired Sample t-test, comparing mean SAM, at pre- and post-. We found a medium-sized significant decrease in mean arousal between pre- and post- (M = 5.48; SD = 1.48 at pre-; M = 4.38, SD = 1.43 at post-; t(26) = 3.49; p = 0.001; d = 0.74). Therefore, for the subsequent analyses, we used the SAM self-rating score of emotional arousal at pre- as covariate.

There was no difference in the means of expressed self-contempt between pre- (M = 4.10, SD = 0.67) and post- at four months (M = 3.91, SD = 1.01); F(27) = 0.91, p = 0.37; d = -0.03; controlling for SAM at pre-). There was a small increase between the means of self-reported intensity of problems in self-esteem between pre- (M = 3.47, SD = 0.66) and post- at four months (M = 3.74 SD = 0.059); F(26) = 2.57, p = 0.04; d = 0.43; controlling for SAM at pre).

Within task fluctuations were small to large at pre- (for SAM between assessment (a) and (b): d = 0.80; SAM between assessment (b) and (c): d = 0.31, for SSES between assessment (a) and (b): d = 0.30; between assessment (b) and (c): d = 0.36). Within task fluctuations were moderate to large at post- (for SAM between assessment (a) and (b): d = 0.92 and between assessment (b) and (c): d = 0.53, for SSES between assessment (a) and (b): d = 0.59 and between assessment (b) and (c): d = 0.81).

There is a positive link between the level of expressed self-contempt at pre- and the intensity of borderline symptomatology at post- (r = 0.48, p = 0.01; partial correlation, controlling for the mean level of self-reported arousal at pre). This model predicted 19% of the variance of the borderline symptomatology at post-.

Neural activation in control participants was not higher when exposed to their self-contemptuous individualized stimuli compared to non-individualized negative stimuli: using whole brain analysis, to control for p < 0.05 FEW corrected, the applied threshold was T = 5.65, no suprathreshold clusters were found. For exploration purposes, we did the same analysis with uncorrected p < 0.001. The applied threshold was T = 3.48. Suprathreshold clusters are found as shown in Table 1; Fig. 1.

On the level of the main effect, expressed self-contempt in the psychological assessment was not linked to a higher neural activation during the condition in which participants were exposed to individualized stimuli. Then, the level of expressed self-contempt was introduced as a covariate at 2nd level analysis. Using whole-brain analysis, controlling for p < 0.05 FEW corrected, the applied threshold was T = 5.75 and no clusters survived this threshold. For exploration purposes, we did the same analysis with uncorrected p < 0.001. The applied threshold was T = 3.50. Suprathreshold clusters are found as shown in Table 2; Fig. 2.

To further illustrate the relevance of the above-described multicomponent multimethod assessment of emotional change to psychotherapy research, this section will present the case of Maya to which we have also applied this neurobehavioral paradigm.

The 25 years old client, presented with six out of the 9 DSM-5 criteria of BPD and requested treatment. She underwent 10 sessions of brief psychiatric intervention (Gunderson, 2014). Her symptom level decreased slightly over the course of the 4 months brief treatment (mirroring the time span used for the controls), between pre- (T1, beginning of treatment) and post- (T3, end of brief treatment; from 15 on the first ZAN-BPD assessment to 13 on the one four months later; see Blanco Machinea et al., submitted, for a complete narrative account of this specific case example). Her observer-rated expressed self-contempt was elevated at pre-assessment when she was criticizing herself, and it decreased between pre- and post-treatment, when expressing self-criticism to the Self (from 4.2 to 2.1). Mean self-rated self-esteem remained stable between pre- and post-therapy (between 2.9 and 2.5). Emotional arousal was self-rated on average 7.3 at pre- and 5.0 at post-therapy, and thus decreased over time (see Fig. 3). Analysis of within task fluctuations showed that within task assessment point (b, right after the imagination task) was systematically highest (or lowest for self-esteem), compared with assessment points (a) and (c) for both pre- and post-therapy for all four assessment modalities.

Among the words used by Maya in the two-chair dialogue, there were “wrong”, “not do better”, “no value”, “not enough”, “not good”, “too hopeful”. Neurofunctional analysis revealed that at pre-therapy, when contrasting the BOLD signal in response to her own self-critical words compared to negative words, the amygdala as well as the insular cortex showed an increased activation (see Fig. 4a). At post-therapy, however the same BOLD contrast (her own self-critical words vs. negative words) revealed increased activation of the dorsolateral-prefrontal cortex (see Fig. 4b).