Trait Rumination and Social Anxiety Separately Predict Stress Induced Rumination and Hemodynamic Responses

We aimed to investigate the stress-reactive rumination response to social stress and its association with social anxiety and trait rumination. From previous investigations we know that people with a certain vulnerability to rumination show increased stress-reactive rumination. However, up to date the possible inuence of social anxiety to this relationship is still unclear. Therefore, we reanalyzed the data of two of our previous studies assessing healthy low and high trait ruminators and depressed patients performing the Trier Social Stress Test (TSST). We measured cortical oxygenation using functional Near-Infrared Spectroscopy (fNIRS) as well as different behavioral outcome measures (subjective stress levels, negative affect, state rumination). On a behavioral level, we found an inuence of both, social anxiety and trait rumination, on state rumination, even when correcting for the other factor, respectively, implying two potentially independent factors of inuence. On a neural level, we observed reduced activation in socially anxious subjects in the right inferior frontal gyrus (IFG), as well as time-dependent changes in the left IFG and in the right dorsolateral prefrontal cortex. Results indicate a specic role of social anxiety, at least on a behavioral level, and therefore implicate a crucial factor to be considered in the treatment of depression. 82 in order to obtain p-values using the Satterthwaite approximation. Graphics were plotted using the R package ggplot2 83 and MATLAB 74 . Using the R-package MuMIn 84 , marginal R² was computed as a measure of variance explained by the xed effects in the mixed models. By reporting changes in R 2 we are comparing the more complex with the corresponding less complex model.


Introduction
Comorbid diagnoses are common in mental disorders. For instance, in nationally representative surveys in the United States, Kessler and colleagues found 60% of participants with lifetime Major Depressive Disorder (MDD) to meet criteria of a comorbid anxiety disorder 1 , and, more speci cally, about one third suffered from comorbid social phobia 2 . Brown and colleagues 3 replicated these ndings, which gave rise to future research investigating the potentially similar underlying mechanisms. Repetitive negative thought -such as rumination -seems to be such a transdiagnostic factor, which has been shown to be related to the development of both depression and anxiety disorders [4][5][6] .
Rumination is de ned as a perseverative, highly self-referential, pessimistic and abstract thinking style with little or no goal and change-orientation 7 . Furthermore, rumination is conceptualized as a form of repetitive negative thinking, such as worry 8 . Originally investigated by Nolen-Hoeksema under the term "depressive rumination" 9 , it was described as "behavior and thoughts that focus one's attention on one's depressive symptoms and on the implications of these symptoms" (p. 569, 10 . Since then, different models have been established and broadened the conceptualization of rumination. In the investigation of depressive rumination, experimental designs often use instructions to think a certain way, for instance about negative emotions/negative personality attributes [11][12][13] or negative life events [14][15][16] to induce ruminative processes. However, these approaches, even though they assess self-referential thoughts, which are a typical content of depressive rumination, do not capture other aspects of rumination like uncontrollability or repetitiveness. Indirect induction methods using social stress tests like the Trier Social Stress Test (TSST) 17 have been able to successfully provoke increased state rumination [18][19][20][21][22][23][24] . In line with longitudinal studies, stress-reactive rumination promises to be a better predictor of depression than trait rumination 25 . Interestingly, this is not in line with the original de nition of rumination by Nolen-Hoeksema 9 , which assumed ruminative processes to pose more of a trait rather than a reactive state condition 4 .
Social anxiety research has to face similar problems in the investigation of post-event processing, which can be de ned "as an individual's repeated consideration and potential reconstruction of his performance following a social situation" (p. 891, 26 , and which is postulated to be a maintaining factor in social anxiety disorders 27 . However, studies in which participants are instructed to imagine hypothetical or recent personal situations 28-30 as well as experimental studies in which impromptu-speech tasks are used 31,32 show highly socially anxious individuals to be more prone to engage in post-event processing.
These effects were maintained even one week after experimentally induced confrontations 33 , as well as in naturalistic reports of diary-based studies 34 . When stressor-induced ruminative processes are compared in depressed patients and socially anxious individuals, studies found the latter to be more strongly 35 and both uniquely associated with negative post-event rumination 32 . However, it remains unclear in how far momentary ruminative responses represent a construct associated with depressive rumination as de ned by Nolen-Hoeksema 9 and/or social anxiety. Potentially, rumination in depression and post-event-processing in social anxiety are to some extent equivalent constructs, which poses the existence of a common underlying mechanism and transdiagnostic phenomenon.
The Trier Social Stress Test (TSST) is a commonly used, highly potent stressor and the gold standard for examining neurocognitive mechanisms of acute stress 36 . Using the TSST in adults with social phobia, Condren, et al. 37 observed signi cantly higher stress-induced cortisol increases in patients compared to a control group, suggesting that patients with social phobia are hyper-responsive when confronted with psychological stress 38,39 . Concerning rumination in general, it could be demonstrated that in subjects with higher trait rumination the stress response after a TSST was either intensi ed or prolonged 23,40,41 . However, there are also studies which could not nd any differences of socially phobic individuals compared to healthy controls 42 , or only subjective but not physiological stress responses to be elevated in the former 43 . Investigating the neural correlates of social anxiety in a public speech task, Tuscan, et al. 44 found elevated frontal cortical oxygenation using fNIRS, but no signi cant differences between high and low socially anxious individuals. Studies using the TSST in a sample of depressed patients revealed hypoactivity in the prefrontal cortex as well as increased state rumination and negative affect when compared to healthy controls 22,45,46 . Generally, increased activity in regions of the Cognitive Control Network (CCN), comprising the dorsolateral prefrontal cortex (DLPFC), anterior cingulate cortex (ACC) and posterior parietal cortex 47 , was observed in fNIRS and fMRI studies as a correlate of social stress induction 48 . Furthermore, in one of our previous studies 22, 45 we found decreased activity in the bilateral inferior frontal Gyrus (IFG) as a mediator of post-stress rumination. As the IFG is relevant in inhibition and attention control tasks 48-55 , this nding let us conclude that inhibition of the IFG and therefore of these tasks might lead to rumination. However, up to date, neurocognitive studies investigating the association of stress-reactive rumination with social anxiety and trait rumination are scarce. Following the assumption of a possible common, transdiagnostic factor of rumination in depression and social anxiety, a neurocognitive investigation of stress-reactive rumination in depressed as well as anxious subjects seems to be useful in order unravel a potentially common ground on a neuronal level.
In two previous studies of our lab 22,45 , we investigated stress-induced rumination but did not test whether social anxiety has an in uence or explains unique variance. More speci cally, we investigated the e cacy of a socially stressful situation (the TSST) under laboratory conditions as well as its effects on postevent processing on a neuronal level to nd possible neuronal activation patterns for rumination in different sub-samples (high vs. low trait-ruminators, depressed vs. healthy subjects). Furthermore, we measured the physiological and the affective adaptation to this high-stress situation. Therefore, we assessed behavioral data such as subjective stress, state rumination and mood changes as well as heart rate, salivary cortisol and cortical oxygenation in both studies. In the present analysis, we expected higher subjective stress and increased state rumination and negative mood in socially anxious subjects. On a neural level, we expected that social anxiety would show a negative effect on CCN functioning, as already shown for trait rumination. Furthermore, we hypothesized elevated salivary cortisol levels as well as cortical oxygenation in both, socially anxious subjects as well as high trait ruminators. Generally, in line with research suggesting common underlying mechanisms of social anxiety and ruminative processes, we expected high correlations of the Liebowitz Social Anxiety Scale (LSAS) and the Rumination Response Scale (RRS) but unique variance to be explained by both factors in our corresponding outcome measures.

Participants
Participants were recruited at the University Hospital of Tübingen and via ambulant psychotherapists. All procedures were approved by the ethics committee at the University Hospital and University of Tübingen and in line with the Declaration of Helsinki in its latest version. All participants gave their written informed consent prior to data collection. Exclusion criteria comprised: diabetes mellitus, kidney insu ciency, hypertension, dysrhythmia, cushing syndrome, substance abuse, adrenal insu ciency, cortisone medication, pacemaker, craniocerebral trauma as well as any medication except for oral contraceptives.
In the rst study, only healthy controls without a history of mental or neurological disorder were assessed. In the second study, healthy controls meeting the same criteria were compared to a group of patients with major depressive disorder (MDD), from which any other primary mental disorder except ICD-10 diagnosis F32.x, F34.1 and F33.x, as well as suicidality, extraordinarily severe depressive symptoms (BDI-II > 50), de cient emotional stability according to the currently treating psychologist and decompensation under social stress in the past led to exclusion.
The combined sample consisted of 45 healthy controls, 23 subclinical high trait ruminators and 22 depressed patients (see table 1). The diagnosis in the patient sample included recurrent Major Depressive Disorder (MDD) (n = 15) and rst episode MDD (n = 7). The comorbid diagnoses included somatic symptom disorder (n = 2), anxiety disorders (n = 2) and personality disorders (n = 2). 60% of the patients were currently receiving psychotherapy and 58% antidepressant medication. The mean age of the total sample was 24.11 years (SD = 5.24 years) and 80% of all participants were female.
Differences between the study samples (e.g. the BDI-II-score) only concerned outcome measures irrelevant for our analysis. We did not differentiate between healthy and depressed participants, as we could observe a high correlation of social anxiety and rumination in both subsamples. Therefore, we prioritized a larger rather than a homogenous sample. In addition, as social anxiety and rumination are dimensional rather than categorical constructs and seem to play a role not just in depressed patients but also in healthy controls 56,57 , we combined the different subsamples from our previous studies for a dimensional approach. Table 1 gives an overview of the main outcome measures of our analysis, subdivided into the different study samples.

Procedures
In the following section we will brie y describe the paradigm and the used questionnaires as well as the physiological measures we did in the experiment. Figure 1 gives an overview of the procedure and when which measurement and which questionnaire was done.
TSST. The Trier Social Stress Test (TSST) 17 is a highly reliable paradigm to induce psychological stress consisting of a free speech and a mental arithmetic task. We adapted the TSST according to the fNIRSsetting by performing a 7-min resting-state previous to two non-stressful control tasks and another 7-min resting-state after the TSST. Through this adaptation, the TSST delivered an ecologically valid and fNIRSconform stress situation, that leads to post-event processing in rumination-vulnerable and/or anxious subjects 22,45 . Due to saliva sampling, participants rested for another 45 min, however fNIRS was not recorded during this part of the post-stress phase. After arriving at the laboratory, participants completed several questionnaires assessing depression symptom severity (BDI-II), trait rumination (RRS), as well as social anxiety symptoms (LSAS). Previous to as well as after the stress induction, we assessed momentary affect (PANAS) and state rumination. Further, at several time points, subjective stress was measured on a Visual Analogue Scale (VAS) ranging from 0-100% as well as samples of salivary cortisol were taken using Salivettes. As this investigation focused on behavioral data assessed by the aforementioned questionnaires and fNIRS data during the post-stress resting-state, in the following we will brie y describe these measures, however more detailed information regarding the stress induction and general procedure is to nd in 22,45 .
Visual Analogue Scale (VAS). Throughout the experiment, participants rated their momentary stress levels on a scale ranging from 0-100%, where steps of 10% were marked at steps of one centimeter. The questionnaire comprised all ratings of the current day of measurement on one page, so participants were able to allow for their last ratings.
Beck Depression Inventory (BDI-II; Hautzinger et al., 2009). In order to screen depression symptom severity, we used the German version of the self-report questionnaire Beck Depression Inventory II. Regarding the previous two weeks, the occurrence of 21 symptoms is rated and symptom severity is assessed as a total score ranging from 0-63. Investigating psychometric properties across different populations and languages, respectively, Wang and Gorenstein 60 could observe overall high internal consistencies (Cronbach's α around 0.9) as well as high retest reliability (mean interval of 2 weeks; r around 0.7-0.9).
Liebowitz Social Anxiety Scale (LSAS; Liebowitz, 1987). The LSAS is a screening measure for social anxiety disorder, comprising 24 social situations that are rated on a 4-point Likert-scale for level of fear (none, mild, moderate, severe) and avoidance (never, occasionally, often, usually) regarding the previous week. The resulting total score (range from 0 to 144) has been shown to have excellent psychometric properties (Cronbach's α = .95) 61 . Also, the LSAS was found to be capable of differentiating clinical as well as non-clinical samples 62,63 which is the reason why in the following we will consider LSAS-scores as a measure of levels of social anxiety. Note that we will report LSAS mean scores in order to account for potential missing values.
Ruminative Response Scale (RRS; Noelen-Hoeksema, 1991). Trait rumination was assessed using 22 items which are rated on a 4-point Likert scale ranging from "hardly ever" to "almost always". The total score (range from 22 to 88) has been shown to have high internal consistencies (Cronbach's α = .88-92) 64-66 . Note that we will report RRS mean scores in order to account for potential missing values.
Positive and Negative Affect Schedule (PANAS; 67 . Using the PANAS we assessed momentary positive and negative affect. 20 items are rated using 5-point Likert scales ranging from 1 ("very slightly") to 5 ("extremely"). Both subscales, positive (PA) and negative affecrt (NA) have acceptable internal consistencies in clinical and non-clinical samples 68,69 of = .85-.86 for NA and = .84-.89 for PA.

State rumination
We assessed stress-reactive rumination pre and post stress using adapted items from the RRS 9 , ARSQ 70 , PTQ 71,72 . The 18 items were answered using a 5-point Likert scale ranging from 1 ("not at all") to 5 ("very often"), totaling to a score between 18 and 90. Note that we will report mean scores in order to account for potential missing values. Subjects were instructed to rate whether the items were in line with their mental state during the last 10 minutes (see supplemental table 1).
fNIRS. During the stress induction as well as both resting-states, an fNIRS measurement was performed to assess cortical activation using a 46-channel continuous wave multichannel fNIRS system (ETG-4000 Optical Topography System; Hitachi Medical Co., Japan) with a sampling rate of 10 Hz. According to our regions of interest, we placed two frontal probesets and one parietal probeset with reference position to Fpz and Cz according to the 10-20 system 73 using an Easycap with sponge rings for additional xation of the optodes (see table 2 and gure S3). Like this, a xed inter-optode distance of 3 cm was set for all optodes. Changes in levels of oxygenated (O2Hb) and deoxygenated (HHb) hemoglobin were computed by means of the modi ed Beer-Lambert Law. Data preprocessing was done in MATLAB 74 , using customized scripts, including interpolation of single noisy channels, correction of motion artifacts using Temporal Derivative Distribution Repair (TDDR) in order to remove spikes primarily caused by head movements in our case 75 and Correlation-based signal improvement (CBSI) 76 to combine both signals (O2Hb and HHb) in their merits: High sensibility (O2Hb) and high resilience to arousal artifacts (HHb). We further used bandpass-ltering to remove low-frequency baseline-drifts (below 0.01 Hz) and highfrequency noise (above 0.1 Hz). Then, a second step of channel interpolation followed in case of artifacts due to data correction. Afterwards, a global signal reduction was performed with a spatial gaussian kernel lter with a standard deviation of σ = 40 77 . For data analysis, we calculated event-related averages for each trial including a 5 seconds baseline-correction for every region of interest (ROI). Heart rate. We assessed heart rate during both resting-states as well as the stress induction using a 1channel electrocardiogram in order to monitor heart rate variability as an index of the stress response. After disinfection of the corresponding skin areas, three standard Ag/AgCl ring electrodes were attached using Ten20 conductive paste. Electrodes were placed above the right clavicle, below the left costal arch and on the neck (reference). The signal was recorded using the BrainAmp ExG ampli er with a sampling rate of 1000 Hz and BrainVision Recorder Software (Brain Products, Munich, Germany).
Salivary cortisol. Saliva was collected in salivettes (Sarstedt AG & Co., REF 51.1534.500) and was stored at − 20°C and later thawed and centrifuged for 2 min at 1000g to collect saliva. Further analysis was performed with enzyme immunoassay (IBL International, Cortisol ELISA, REF RE52611) according to the manufacturer's instructions. Average cortisol levels were taken from duplicate runs if intra-assay variation was below 10%. Finally, daytime was regressed out of cortisol coe cients to account for circadian rhythm uctuations that are not related to the TSST and values were log-transformed. Participants were instructed not to drink alcohol the day before the measurement, to sleep as long as they usually do and to perform no physical activities on the day of the measurement. Also, subjects were told not to drink or eat 30 min before the measurement started.

Data analysis
After data preprocessing, data analysis was done using R 79 and SPSS 80 . Mixed models were tted using the R packages lme4 81 and lmerTest 82 in order to obtain p-values using the Satterthwaite approximation. Graphics were plotted using the R package ggplot2 83 and MATLAB 74 . Using the Rpackage MuMIn 84 , marginal R² was computed as a measure of variance explained by the xed effects in the mixed models. By reporting changes in R 2 we are comparing the more complex with the corresponding less complex model.
In the following, we will analyze subjective stress, negative affect and state rumination. Further, we will analyze heart rate and salivary cortisol, but these analyses are to be found in the supplemental material as they were not related to our primary hypotheses. We investigated in how far social anxiety (LSAS) and trait rumination (RRS) play a role in each measure and set up mixed models with an increasing number of parameters. Note that we did not include BDI-II as a predictor due to its high correlations with social anxiety and trait rumination (see table 4) and issues of multicollinearity. Fitting mixed models offers the chance to include random effects accounting for non-independence 85 , as well as to handle unbalanced data and include continuous predictors 86 . First, we conducted a basic model consisting of the parameters set by the experimental design, such as time (model 1). For the two more complex models, we added z-standardized LSAS (model 2) and z-standardized RRS total scores (model 3) as main effects and interaction effects with time, accounting for social anxiety and trait rumination, respectively. As most complex models, we added the z-standardized RRS (model 4) and LSAS scores (model 5) only as main effects to the previous models. Like that, we controlled for trait rumination and social anxiety and investigated whether there is unique variance explained by the corresponding measure. Note that all predictors were included as xed effects whereas due to repeated measurements on each subject, intercepts for every participant were modeled as random effects. Table 3 gives an overview over the models conducted for each measure.
For the analysis of cortical oxygenation, we tted a MANOVA in order to increase statistical power and assess multiple dependent variables simultaneously. For this, we added the same parameters as predictors as in our previous mixed models.
Prior to our analyses, we checked assumptions of the used methods and corrected for potential violations of assumptions. Note that corrections were not applied unless stated otherwise. In case of subjective stress ratings and state rumination, normality assumption was not met but due to insu cient success of common transformations and the robustness of the mixed models 87 , we decided to perform the analyses as planned. Furthermore, in case of a detection of outliers, we will report the corrected analyses in detail and summarize results of the uncorrected analyses.
Note that physiological data of heart rate was missing in ve subjects, leaving 85 participants, and one participant had to be excluded in the analyses of negative affect for study 1, due to missing data. Note. In case of subjective stress and heart rate, time was modelled as linear and quadratic term. Colons symbolize interaction effects. All included parameters except for time were z-standardized.

Subjective stress (VAS ratings)
First, we wanted to investigate whether the stress induction was re ected by changes in subjective stress. Calculating mahalanobis distances, we identi ed seven subjects´ stress ratings as multivariate outliers (p < .001) and therefore excluded them from the following analyses. From previous investigations of stress paradigms 45,46 we assumed a quadratic relationship of subjective stress ratings and time and therefore tested a linear model against a quadratic model using a Likelihood-ratio-test. Results revealed a signi cantly better t of the quadratic model, X²(1) = 244.58, p < .001, which was subsequently used in the analysis. Fitting the basic model, time yielded a signi cant predictor as a linear t(664) = 6.607, β = 13.395, p < .001) as well as a quadratic term (t(665) = -2.512, β = -1.538, p < .05) (see table 5) and remained signi cant even in the more complex models. In model 2, we observed signi cant interactions of LSAS with both time factors (t(664) = 2.771, β = 2.868, p < .01 and t(664) = -2.719, β = -0.339, p < .01) and more variance to be explained compared to the basic model (change of R² = .069). Higher LSAS scores were associated with higher linear increases in the stress response and a stronger quadratic decrease at post stress (see Fig. 2). In model 3, RRS did not explain further variance (change of R² compared to the basic model = .021). In model 4, we observed the same pattern as before whereby both interactions of LSAS and time got signi cant (t(664) = 2.771, β = 2.868, p < .01 and t(664) = -2.719, β = -0.338, p < .01) and no change in R² was observed compared to model 2. Fitting model 5, only the control variable LSAS yielded a signi cant predictor (t(83) = 4.054, β = 6.416, p < .001). We found the xed effects of the latter model to explain more variance compared to model 3 (change of R² = .044). Higher social anxiety was associated with increased subjective stress.

Negative affect (PANAS)
To investigate changes in negative affect due to the stress induction, we performed a repeated measurements ANOVA with the factor time (pre stress, post stress, post rest2). Note that the third PANAS assessment (post rest2) was different in study 1 (15 min after the stress induction) and study 2 (60 min after the stress induction), which is why we only analyzed the rst two time points. One subject from study 1 was excluded from further analyses due to missing values for negative affect. After identifying another three subjects in study 2 as multivariate outliers, we observed a  5). Rerunning our analyses with inclusion of multivariate outliers, we found the same as the aforementioned results.

State rumination
After exclusion of one subject which yielded a multivariate outlier (p < .001), we tted a basic model including the factor time, which predicted the outcome signi cantly (t(89.00) = 6.63, β = .470, p < .001) and remained signi cant even in the more complex models (see table 5 were uniquely associated with increased state-rumination following the TSST. Note that our analyses with inclusion of multivariate outliers yielded the exact same results.

Discussion
This study investigated the association of social anxiety and trait rumination with the stress response as induced via the Trier Social Stress Test (TSST). This is crucial as the TSST has recently been used to elicit state ruminative responses as life stress has been shown to be one trigger of rumination and the latter to be a mediating factor of stress and psychopathology 88,89 . In addition, previous ndings suggest that social anxiety and depression appear to be uniquely associated with stress induced ruminative processes, although the underlying mechanisms seem to be of a similar manner 32,35 . We analyzed behavioral, neurobiological and autonomous correlates of trait rumination and social anxiety as predictors of the stress response.
Our results showed the same effects of the general stress response as in previously published studies. Concerning our hypotheses, we found a signi cant moderating effect of social anxiety on subjective stress levels. Intuitively, subjective stress was elevated in socially anxious individuals as they were confronted with a social situation. After similar baseline stress, we observed higher ratings in socially anxious individuals prior to and immediately after the stress induction. This is in line with assumed states of social anxiety including anticipatory processing and post-event processing 26,30,33,34,90−92  On a neurophysiological level, we found different impacts of cortical oxygenation for the RRS and LSAS. First of all, we found time-dependent changes in the left dorsolateral prefrontal cortex and the somatosensory association cortex in the whole sample. Concerning different levels of social anxiety, we found time-dependent changes in the left inferior frontal gyrus and in the right dorsolateral prefrontal cortex. Furthermore, we found attenuated cortical oxygenation for high vs. low socially anxious individuals in the SAC. Note that as this latter statistical effect only reached marginal signi cance, it needs to be interpreted with caution. Nevertheless, these results are in line with a body of research that found the SAC to be relevant in anxiety disorders in general 95,96 . Furthermore, the functional connectivity between the SAC and other brain areas, that seems to be crucial in social anxiety, was found to be increased 97 in subjects with SAD. Regarding trait rumination we only found a marginal signi cant change in cortical oxygenation in the left IFG in high trait ruminators. This result doesn't replicate the results in our previous studies, where we found signi cantly lower brain activity in the right IFG in high trait ruminators over the course of the measurement. However, these differential results are due to different statistical analysis. Calculating the same statistical models as in our previous studies, we could replicate the same pattern of frontal hypoactivation and signi cant changes in cortical oxygenation dependent on time. Nevertheless, over all we constantly found a changed neurophysiological pattern in the bilateral IFG in high trait ruminators in the course of the TSST. This difference in contrast to low trait ruminators could be interpreted as a de cit in cognitive control and inhibition 4 , which blends in well with the known functions of the IFG as a central instance in inhibition during cognitive tasks 48-55 .
Importantly, the study at hand shows that these neuronal changes are also associated with social anxiety. However, the latter seems to explain additional variance, which is not covered by ruminative processes alone.
Furthermore, the TSST did successfully induce stress-reactive rumination as well as negative affect. We found both measures to be uniquely associated with social anxiety and trait rumination, even when we controlled for different levels of trait rumination and social anxiety, respectively. This is in line with our hypotheses and recent research investigating ruminative responses to stress 20,21,23,24,88,93 as well as associations of perceived stress in socially anxious individuals 42,98 . As already mentioned, experimental investigations of social stress in socially anxious subjects are scarce. However, it seems to be ostensible that with higher social anxiety there might also be a higher reactivity to socially stressful situations, as indicated by several correlation studies [99][100][101] .
Intuitively, trait rumination and social anxiety did correlate signi cantly, which is well in line with the high comorbidity of mood disorders, like MDD, with anxiety disorders. However, our results also show that the previously reported effects of trait rumination on stress-induced processing seem to be independent of the potential confounder of social anxiety, at least to a certain degree. This in turn allows us to draw some clinical as well as scienti c implications: Once more, we could demonstrate that via the TSST state rumination can be reliably induced. Moreover, this effect seems to be independent from social anxiety and therefore the TSST is a highly e cient tool to induce post-event processing in subjects with and without social anxiety. In a clinical or psychotherapeutic context our results might demonstrate that in the treatment of SAD a training of social skills to handle daily and/or stressful situations will not be su cient, but it is also important to learn strategies to overcome rumination, for example via emotion regulation strategies 95 .
Additionally, social anxiety and state rumination are not interchangeable as they only explain about 25% of common variance. Supporting this, we found trait rumination and social anxiety to independently in uence outcome measures of the TSST on a behavioral as well as on a neurophysiological level. Taken together, these results strengthen the hypothesis of an underlying unifying mechanism which is potentially also transdiagnostic regarding other psychopathologies.
However, some limitations of the study have to be considered when interpreting the results. As not primarily socially anxious participants with diagnosed SAD were recruited, the analyzed data does not cover the full range of variance of social anxiety, especially if it is regarded as a dimensional construct.
Even though we did nd high variations in LSAS scores, investigations with a clinical sample of patients with SAD would potentially extend evidence for the unique effects of social anxiety on responses to the TSST. Further, we did not collect new data, as our hypotheses have not been analyzed with the used data yet, neither separately for each study nor for the merged sample. However, by merging the two datasets of our lab, we were able to increase the power of our statistical investigations and gather more evidence for the presence of our reported effects. Concerning the investigation of the neural correlates, we used fNIRS, which is highly useful in naturalistic investigations due to decreased sensitivity to motor artifacts 102-104 , penetration depth is limited to about 1,5 − 2 cm of the cortex 105 , which limits our ndings to cortical areas. Nevertheless, fNIRS offers the chance to reliably investigate neurobiological mechanisms in an ecologically valid setting 36 .

Conclusion
Taken together and considering the current state of research 106,107 , ruminative responses could be interpreted as a transdiagnostic construct in social anxiety and major depression disorder. However, both factors showed to be uniquely associated with stress-reactive rumination. Going further, previous studies could show that rumination also seems to play a role in other psychopathological syndromes associated with problems in emotion regulation (e.g. eating disorders 108 , personality disorders 109 as well as substance abuse 110 ). Future research should therefore extend ndings of experimentally stress-induced rumination using samples of various mental disorders, as this promises to give support for a transdiagnostic and emotion-reactive understanding of rumination. This, in turn, could help to improve existing and to develop new psychotherapeutic interventions, of which one possible focus should be the treatment of ruminative thinking. For this purpose, fNIRS is a useful tool as it can be assessed easily in an ecologically valid manner (like we did it in the TSST) and it is very helpful to investigate the neurobiological basis of ruminative processes. In the future it could be used in a neurofeedback-design to help patients initially recognizing their ruminative processes and to overcome those. In this context, emotion regulation seems to be a promising strategy, what we are currently examining in a depressive sample.

Declarations
Data availability The data of this study is available upon request from the rst or the last author.