Insula reactivity mediates subjective isolation stress in alexithymia

The risk for developing stress-related disorders is elevated in individuals with high alexithymia, a personality trait characterized by impaired emotional awareness and interpersonal relating. However, it is still unclear how alexithymia alters perceived psychosocial stress and which neurobiological substrates are mechanistically involved. To address this question, we examined freshmen during transition to university, given that this period entails psychosocial stress and frequently initiates psychopathology. Specifically, we used a functional magnetic resonance imaging emotional face matching task to probe emotional processing in 54 participants (39 women) at the beginning of the first year at university and 6 months later. Furthermore, we assessed alexithymia and monitored perceived psychosocial stress and loneliness via questionnaires for six consecutive months. Perceived psychosocial stress significantly increased over time and initial alexithymia predicted subjective stress experiences via enhanced loneliness. On the neural level, alexithymia was associated with lowered amygdala responses to emotional faces, while loneliness correlated with diminished reactivity in the anterior insular and anterior cingulate cortex. Furthermore, insula activity mediated the association between alexithymia and loneliness that predicted perceived psychosocial stress. Our findings are consistent with the notion that alexithymia exacerbates subjective stress via blunted insula reactivity and increased perception of social isolation.


Scientific Reports
| (2021) 11:15326 | https://doi.org/10.1038/s41598-021-94799-w www.nature.com/scientificreports/ Importantly, alexithymia and loneliness seem to affect similar neural pathways: a meta-analysis of neuroimaging studies 18 revealed that high levels of alexithymia are associated with blunted responses to emotional stimuli (e.g. happy and fearful faces) in a limbic neurocircuitry including the amygdala and insular cortex and elevated responses in the anterior cingulate cortex (ACC) that may reflect difficulties in identifying and regulating emotions. Likewise, in highly lonely individuals, pleasant social stimuli elicited less activity in the striatum, amygdala, insula and ACC 19 . Of note, these brain regions have been identified as important neural hubs of stress resilience such that robust amygdala responses to emotional stimuli and functional coupling of ACC-insula circuitry might promote adaptive stress responses 20,21 . Moreover, a recent study demonstrated that targeted amygdala neurofeedback improves stress coping and reduces alexithymia 22 , further strengthening the assumption that alexithymia and loneliness prevent favorable stress response by shared neural response patterns.
The current study thus aims to probe whether alexithymia might affect perceived stress by enhancing feelings of loneliness and to examine which neural substrates are involved. Therefore, we measured alexithymic traits and neural activation patterns in response to social stimuli (emotional faces) during functional magnetic resonance imaging (fMRI) in a sample consisting of 54 healthy freshmen. Participants were monitored during their first 6 months of the transition to university. Each month, participants completed questionnaires measuring their loneliness and their subjective stress experiences during this major life event. Specifically, we hypothesized a positive correlation between alexithymia and subjective stress response across time and that this relationship would be mediated by feelings of loneliness. Given the intertwined phenotype of alexithymia and loneliness as well as the overlapping neural correlates of both constructs, we predicted that both higher alexithymic traits and higher loneliness levels would be associated with altered responses to emotional face stimuli in the ACC, insula and amygdala. To this end, we used alexithymia and loneliness scores as continuous covariates in the analyses. Finally, we expected that the link between alexithymia and perceived stress would be mechanistically mediated by altered activity in these brain regions.

Results
Behavioral results. Stress levels changed significantly over time (F (6,294) = 4.56, p < 0.01, η p 2 = 0.09) and peaked in month four (t (52) = 4.48, Bonferroni-corrected p (p cor ) < 0.01, d = 0.47) and five (t (53) = 3.92, p cor = 0.02, d = 0.49) of the observation period in comparison to the stress levels at study entry (see Fig. 1A), reflecting the first examination phase. In contrast, social network size (F (6,282) = 0.96, p = 0.43, η p 2 = 0.02) and loneliness scores (F (6,288) = 1.69, p = 0.14, η p 2 = 0.03) did not significantly change during the time course (see Table S1). As predicted, both the average loneliness (r (52) = 0.52, p < 0.01; see Fig. 1B) and alexithymia in the first month (T1) positively correlated with the average perceived stress in the 6 months (r (52) = 0.40, p < 0.01; see Fig. 1C), showing that individuals with greater dysfunctional emotional awareness and higher subjective lack of social con-  www.nature.com/scientificreports/ nection experienced more stress during the transition phase. In addition, T1 alexithymia positively correlated with psychosocial stress (r (52) = 0.49, p < 0.01) already at study entry, but was not significantly associated with the increase in stress levels (i.e. maximum stress minus baseline) during the first examination phase (p > 0.05), indicating that alexithymia is associated with consistently increased perceived stress levels rather than increased acute stress responsiveness. Furthermore, depressive symptoms (t (53)  , k E = 537, p FWE < 0.01) in response to happy faces compared to neutral ones. There were no significant whole-brain differences between the first (T1) and the seventh month (T7). In a second step, we added the parameter estimates of the right amygdala, right ACC and right anterior insula as further mediator variables to the model to elucidate the underlying neural mechanisms. For each brain region, two models were calculated to test potential mediation effects on all pathways (i.e., both serial and parallel mediation effects were tested). The analyses revealed that the link between alexithymia and loneliness was driven by reduced insula reactivity, leading to a significant indirect effect of alexithymia on stress via insula reactivity and loneliness (serial mediation: β = 0.06, SE = 0.04, 95% CI 0.01-0.15, see Fig. 3). Specifically, alexithymia predicted the reduced anterior insula reactivity which was linked to enhanced feelings of loneliness which in turn, predicted subjective stress. This mediation was mainly driven by the Toronto Alexithymia Scale (TAS) factors "difficulties describing feelings" (DDF) and "difficulties identifying feelings" (DIF) (see SI Results). No further mediation effects were observed for the insula, amygdala or ACC (all 95% CIs of further indirect effects via brain activation included zero).

Discussion
The present study aimed at elucidating the neural mechanisms moderating the link between alexithymic traits, loneliness and stress reactivity during the transition to university. Our results confirmed that loneliness mediated the noxious association between alexithymia and subjective stress during the first 6 months of university. Moreover, we found that the anterior insula plays a crucial role in this process, by mediating the link between alexithymia and loneliness. Our results provide further support for and extend the stress-alexithymia hypothesis 4 . We were able to replicate previous models suggesting a close link between alexithymia and loneliness 11 and found that individuals with high alexithymia, especially with difficulties in describing and identifying emotions, experience more stress during transition phases partly because they perceive more subjective social isolation. This finding is consistent with previous studies reporting significant associations between the TAS DIF and DDF subscales and loneliness 23 as well as a relationship between the TAS DIF subscale and poor adjustment during transition to university and perceived stress 24 . Intriguingly, our results indicate that this mechanism may be driven by diminished insula responses to emotional signals which directly link alexithymia with loneliness. The insular cortex is a hub for interoceptive processing and conscious affect 25  www.nature.com/scientificreports/ the right insula positively correlate with changes in social interest 26 . Likewise, individuals with high loneliness have been found to exhibit reduced insula responses during interpersonal trust decisions 27 . Moreover, multiple lines of evidence indicate that insula pathology leads to alexithymia. For instance, dopamine D2-type receptor availability in the insula has been linked to higher alexithymia 28 , the gray matter volume of the insular cortex inversely correlated with alexithymia 29 and the extent of damage to the anterior insula predicted alexithymia in lesion patients 30 . It has been theorized that insula dysfunction in alexithymia may reflect a transdiagnostic marker of empathic deficits 31 and our findings in healthy participants point to an additional mechanism such that the dampened insula responses to external emotional cues underlie the association of alexithymia with enhanced perceived social isolation. Along these lines, the observed pattern of results is consistent with the notion that social connectedness requires the ability to flexibly shift between interoceptive and exteroceptive attention 32 which may be based on recruitment of the anterior insula. Furthermore, consistent with previous fMRI studies 18,19 , we found decreased amygdala and ACC responses in individuals with high alexithymia and loneliness, respectively. The amygdala has often been linked to alexithymia 18,33 and a recent study showed that neurofeedback targeting the amygdala during military training not only enhanced stress coping but also decreased alexithymia 22 . Moreover, the amygdala has also been linked to loneliness and social support. For example, a decrease in perceived stress and loneliness was moderated by amygdala volumes 34 and the experience of social support was regulated by amygdala activity 35 . Likewise, the ACC www.nature.com/scientificreports/ has been previously linked not only to loneliness but also to alexithymia: ACC size correlates with alexithymia ratings especially in men 36 and high levels of alexithymia are associated with elevated responses to emotional stimuli in the ACC 18 . Furthermore, the ACC plays a role in social pain processing during social support 37 and overall seems to be a hub for the integration of social information and empathy 38 . Bearing in mind that neither ACC nor amygdala reactivity mediated effects of alexithymia, the insular cortex seems to be a crucial neural processing hub for the interplay between loneliness and alexithymia. Therefore, neurofeedback training targeting insula activation could lead not only to reduced feelings of loneliness but also to reduced psychosocial stress 22 .
In contrast to loneliness, objective social network indices were not significantly associated with alexithymia or perceived stress. Given that in a previous study with college freshmen 39 psychological stress selectively mediated the association between antibody response to the influenza immunization and loneliness, but not social network size and immunization response, our data provide further support for the notion that the subjective perception of social connectedness may be a more important predictor for stress reactivity during transition phases than the objectively available social contacts. Interestingly, the trait-dependent reactivity was no longer evident in the second fMRI session 6 months later, indicating either repetition effects and reduced retest-reliability or that a disrupted plasticity as observed in the prefrontal cortex with an attention-shifting task following long-term psychosocial stress 40 is more pronounced for limbic reactivity to emotional stimuli. Furthermore, we observed an increase in alexithymia scores, potentially elicited by the prolonged subjective stress, which might reflect an acquired secondary alexithymia 41 . As such, these experience-based changes may have masked genuine trait associations in the second fMRI session. Of note, the allostatic load of the transition to university caused a significant increase in depressive symptoms, social interaction anxiety and autistic-like traits after 6 months, thus illustrating that individuals with high alexithymia and loneliness might be at risk not only for poor academic performance but also stress-related psychological disorders due to chronically increased stress levels.
Collectively, our results provide evidence for a close interplay between emotional awareness and perceived social isolation, with dampened insula reactivity serving as a potential underlying mechanism linking alexithymia with loneliness and thus exacerbating the susceptibility to perceived stress. Based on these findings, neurobiologically-informed interventions with cognitive bias modification procedures should target the feeling of social disconnectedness to help students with alexithymic traits to better cope with psychosocial stress during transition phases. Furthermore, neurofeedback training targeting the insula might reduce the feeling of social isolation and therefore potentially enhance stress coping during stressful life events.

Methods
Subjects. Sixty healthy freshman students participated in the study after giving written informed consent.
The study was approved by the institutional review board of the medical faculty of the University of Bonn and carried out in compliance with the latest revision of the Declaration of Helsinki. Subjects were screened prior to the first test session and received monetary compensation for study participation. Subjects had no past or present physical or psychiatric illness, as assessed by a medical history questionnaire and the Mini-International Neuropsychiatric Interview 42 . All subjects started their first semester without ever attending university courses before. Six subjects had to be excluded because they missed the second fMRI appointment (n = 3), showed excessive head motion in the MRI (> 3 mm/º; n = 2) or because of technical failures (n = 1). Therefore, final analyses . Mean loneliness mediated the relationship between alexithymia at study entry and mean perceived stress ratings. Furthermore, activation of the right insula in response to emotional stimuli at study entry mediated the link between alexithymia and loneliness. Numbers show standardized β coefficients. The β coefficient in brackets shows the total effect without mediators. Insula coordinates are shown in Montreal Neurological Institute space. For illustration purpose, the cluster is shown with a significance level of p = 0.05. *p < 0.05, **p < 0.01, T1 study entry, R right. Experimental design. Subjects were monitored during their first semester at university for a total duration of 6 months, starting with a screening session in their first university month. Shortly (average: 14 days, min/max = 0/32 days) after the screening session, a first fMRI session was conducted (T1 = first month). The fMRI measurements were repeated after 6 months (T7 = seventh month; time between the two fMRI measurements = 164 days, min/max = 153/197 days). Participants completed several questionnaires every month between the two fMRI sessions measuring perceived stress, loneliness and social network size (see Fig. S1).
Questionnaires. Subjects completed different sets of questionnaires to continuously monitor social behavior during their first semester. In the screening session and before the second fMRI scan, we assessed alexithymia (TAS [Toronto Alexithymia Scale] 43  fMRI data acquisition. At the start of the experiment, subjects were instructed to lay as calm as possible.
Functional data were acquired with a 3 T Siemens fMRI task. During the fMRI, subjects completed a well-established emotional face-matching paradigm 54,55 .
To ensure the subjects' attention, subjects had to match the identity of two simultaneously presented pictures at the bottom of the screen with a target picture presented at the top. Stimuli consisted of face pictures (neutral, fearful and happy) and houses as non-social control stimuli. Stimuli were presented with Presentation 14 software (Neurobehavioral Systems, Albany, CA, USA) in three blocks for every condition (Happy, Fearful, Neutral, House) with each block consisting of five trials. Stimuli did not vary in emotional expression or in sociality during a block. Trial duration was 5 s with a 10 s pause after each block. In this pause, a fixation-cross was depicted. The identity of the face stimuli varied between T1 and T7 to reduce habituation effects. Participants could choose their responses using an MRI-compatible response grip system (NordicNeuroLab AS, Bergen, Norway). Responses and reaction times (RTs) were measured to evaluate possible attention effects. High-resolution anatomical images were acquired after the functional images. fMRI analysis. The fMRI data were pre-processed and analyzed using standard procedures in SPM12 (Wellcome Trust Center for Neuroimaging, London, UK; http:// www. fil. ion. ucl. ac. uk/ spm) implemented in Matlab (The MathWorks Inc., Natick, MA). Participants with excessive head movements (> 3 mm/° in any direction, n = 2) or missing data due to technical failures (n = 1) were excluded from fMRI analyses. The first five volumes of each functional time series were discarded to allow for T1 equilibration. Functional images were corrected for head movements between scans by an affine registration. Images were initially realigned to the first image of the time series before being re-realigned to the mean of all images. To correct for signal distortion based on B0-field inhomogeneity, the images were unwarped by applying the voxel displacement map (VDM file) to the EPI time series (Realign & Unwarp). Normalization parameters were determined by segmentation and non-linear warping of the structural scan to reference tissue probability maps in MNI space. Normalization parameters were then applied to all functional images, which were resampled at 2 × 2 × 2 mm 3 voxel size. For spatial smoothing, a 6-mm full width at half maximum Gaussian kernel was used. Raw time series were detrended using a high-pass filter (cut-off period 128 s). A two-stage approach based on the general linear model implemented in SPM12 implemented in Matlab (The MathWorks Inc., Natick, MA, USA) was used for statistical analyses. On the first level, participants' individual data were modelled using a fixed-effect model. Onsets and durations of the four experimental condition blocks ('Happy, 'Fearful' , 'Neutral' , 'House') were modelled by a boxcar function convolved with a hemodynamic response function (HRF). Movement parameters were included in the design matrix as confounds. were computed using one sample t-tests. Loneliness and alexithymia ratings were used as covariates for the second level analysis. The following whole-brain analysis was done with a height threshold of p < 0.001. The main analyses of fMRI data focused on regions of interests (ROIs) associated with emotion processing in alexithymia and loneliness consisting of the amygdala, ACC and insular cortex 18 . These ROIs were anatomically defined according to the Wake Forest University PickAtlas (wfu PickAtlas) for both hemispheres. Parameter estimates of significant ROI clusters were extracted using MarsBaR (http:// marsb ar. sourc eforge. net) and further analyzed in SPSS 25 (IBM Corp., Armonk, NY, USA).

Statistical analyses.
Repeated measures analyses of variance (ANOVAs) and Bonferroni corrected posthoc t-tests were calculated using SPSS 25 (IBM Corp., Armonk, NY, USA) to examine changes in stress, loneliness and social network size over time. If the assumption of sphericity was significantly violated as assessed by Mauchly's tests, Greenhouse Geisser corrections were applied. Pearson correlations between parameter estimates of significant ROI clusters, loneliness, perceived stress and alexithymia were calculated. Furthermore, mediation analyses were carried out using the PROCESS macro v3.4 for SPSS 56 . Focusing on mean stress as outcome variable, we used T1 alexithymia as predictor variable and mean loneliness ratings as mediator. As we were interested in the neurobiological mechanisms underlying the link between alexithymia, loneliness and perceived stress, we also tested the hypothesized mediation effects of the neural correlates of alexithymia and loneliness. Parameter estimates of significant clusters associated with alexithymia or loneliness at the first fMRI session were thus included as additional mediator variables and mediation effects were tested for each pathway between the former mentioned behavioral results. For all mediation analyses, 10,000 bootstraps samples were used.