Stress-induced brain activity, brain atrophy, and clinical disability in multiple sclerosis

Significance

Psychological stress is linked to multiple sclerosis (MS) severity (e.g., to a heightened risk of brain lesion development). The exact mechanisms underlying this association are unknown. To investigate the link between brain activity induced by mild psychological stress and MS disease parameters, we conducted a mental arithmetic neuroimaging task involving performance feedback in MS patients and healthy controls and related the brain activity signals to clinical disability and brain volume. In patients, motor and cognitive impairment were related to activity in the insular cortex. Brain volume was related to activity in overlapping cerebellar areas in patients and controls. This overlap suggests that the link between activity and volume cannot reflect a passive response to clinical disability alone.

Abstract

Prospective clinical studies support a link between psychological stress and multiple sclerosis (MS) disease severity, and peripheral stress systems are frequently dysregulated in MS patients. However, the exact link between neurobiological stress systems and MS symptoms is unknown. To evaluate the link between neural stress responses and disease parameters, we used an arterial-spin–labeling functional MRI stress paradigm in 36 MS patients and 21 healthy controls. Specifically, we measured brain activity during a mental arithmetic paradigm with performance-adaptive task frequency and performance feedback and related this activity to disease parameters. Across all participants, stress increased heart rate, perceived stress, and neural activity in the visual, cerebellar and insular cortex areas compared with a resting condition. None of these responses was related to cognitive load (task frequency). Consistently, although performance and cognitive load were lower in patients than in controls, stress responses did not differ between groups. Insula activity elevated during stress compared with rest was negatively linked to impairment of pyramidal and cerebral functions in patients. Cerebellar activation was related negatively to gray matter (GM) atrophy (i.e., positively to GM volume) in patients. Interestingly, this link was also observed in overlapping areas in controls. Cognitive load did not contribute to these associations. The results show that our task induced psychological stress independent of cognitive load. Moreover, stress-induced brain activity reflects clinical disability in MS. Finally, the link between stress-induced activity and GM volume in patients and controls in overlapping areas suggests that this link cannot be caused by the disease alone.

Multiple sclerosis (MS) is an autoimmune disease of the central nervous system leading to demyelination, axonal damage. and neuronal degeneration (1). In addition to sensorimotor symptoms, stress-related syndromes such as depression and anxiety disorders are among the most frequent comorbidities in MS (2).

A role for psychological stress in the pathobiology of MS was hypothesized as early as the 19th century when Charcot first described the disease, and a link between stress and the risk of MS relapse is now supported by numerous prospective clinical studies (e.g., 3). Moreover, MS patients frequently exhibit dysregulated psychobiological stress systems, and these systems interact with the key neurologic characteristics. Neuroendocrine studies revealed a link between MS and altered regulation of both stress systems, the hypothalamic–pituitary–adrenal (HPA) axis and the sympathetic nervous system (SNS) (4). Specifically, pharmacological challenge tests have shown that glucocorticoid responsivity is elevated in MS patients (5) and that impaired HPA axis feedback control is linked to brain atrophy (6) and subsequent deterioration of clinical disability in MS (7). Furthermore, the application of corticotropin-releasing factor has been shown to reduce the severity of experimental autoimmune encephalomyelitis (EAE), the animal model of MS (8). For hormones of the SNS it has been shown that the density of β-adrenoreceptors on peripheral blood mononuclear cells correlates positively with lesion load in MS (9) and that norepinephrine (NE)-related antidepressants reduce the severity of EAE (10). Finally, a stress-reduction intervention based on cognitive behavioral therapy reduced the number of new contrast-enhancing lesions in a randomized controlled trial (11), providing the best evidence to date that stress and MS disease severity indeed might be linked directly.

Given the relatively small number of functional MRI (fMRI) stress studies that experimentally manipulated the degree of psychological stress and measured neural responses in healthy subjects (e.g., refs. 12–14), it is not surprising that stress-related brain activity has not yet been investigated in MS. This investigation is important, however, because psychosomatic studies suggest that the impact of stressors on health depends on the cognitive processing of stressors or affective stimuli (e.g., refs. 15, 16), which is closely reflected by immediate brain responses (17).

Consequently, we investigated in voxelwise fMRI analyses the neural responses to psychological stress in 36 patients with MS and 21 healthy control subjects with an arterial-spin–labeling (ASL) fMRI stress paradigm and the relation of these responses to disease parameters [i.e., clinical disability (18), clinicoradiographic disease severity measures, gray (GM) and white matter (WM) volume]. GM and WM volume were assessed across the whole brain. The fMRI paradigm comprised seven experimental stages (Fig. 1). Brain activity and heart rate were measured during three of the seven stages: stage II, baseline 1; stage IV, stress; and stage VI, baseline 2. Salivary cortisol and perceived stress were measured at four stages: stage I, prebaseline 1; stage III, prestress; stage V, poststress; and stage VII, postbaseline 2. In stage IV, psychological stress was induced by a mental arithmetic task. This task was subdivided in an adaptation stage (IVa) during which the participant’s performance level was determined, and a subsequent performance stage (IVb) comprising performance-dependent adjustments in task frequency and performance feedback. Finally, measures of fast and slow (14) neural stress responses were derived from the task and were related separately to MS disease parameters.

Fig. 1.

The stress task. In the first experimental stage (stage I, pre-baseline 1), an initial salivary cortisol sample was taken, and participants were asked questions related to their currently perceived stress level (referred to as ‘rating’ in the following). These questions were presented on a projection screen and answered with MRI-compatible button boxes. In the next stage (stage II, baseline 1), the first ASL fMRI measurement was conducted; patients were asked to fixate on a crosshair. In stage III, pre-stress, a second rating and cortisol measurement were performed. In stage IV (stress), the fMRI stress measurement was conducted. The participant was asked to perform repeated subtraction tasks having the form operand X minus operand Y. The participant had to select the correct answer from a set of four answers shown on the screen below the operands. The task started with a constant value of X, 43,521, for all subjects. Operand Y ranged from 1 to 99 and was determined randomly in each trial. If a patient gave the correct answer, the difference between the two operands was used as operand X in the next trial. The course of the stress task was divided into two substages, an adaptation stage (stage IVa, duration ≤4 min) and a performance stage (stage IVb, lasting for the remainder of the total 12-min duration of stage IV). In each trial in stage IVa, participants had 8 s to select the response, and response times were recorded. Stage IVa ended either when the participant made 10 correct answers or when a 4-min period had elapsed. Stage IVb began without a break. Stage IVb differed from stage IVa in three aspects. First, the calculation times for correct trials determined during the adaptation stage were used to provide performance feedback in terms of school grades ranging from 1 [Sehr gut (very good)] to 5 [Ungenügend (insufficient)]. Second, if the answers were incorrect or too slow, the subject had to begin again with the initial starting value for X. Third, the time provided for subtraction was adjusted (only) in IVb based on the subject’s performance (i.e., starting with 8 s after the transition from IVa to IVb, this time was decreased or increased by ten percent in a given trial based on the correctness or incorrectness in the preceding trial). Finally, a third rating and cortisol measurement (stage V, post-stress), a second resting fMRI measurement (stage VI, baseline 2), and a fourth rating and cortisol measurement (stage VII, post-baseline 2) were performed.

Results

Demographic and Clinical Participant Characteristics.

Twenty-two of thirty-six patients and 13 of 21 controls were female (χ² = 0.00, P > 0.999). Twenty-one of thirty-six patients and 16 of 21 controls had at least a high school diploma (χ² = 1.86, P = 0.250). The mean age (±SD) of patients was 47.4 (±9.1) y and of controls was 49.1 (±11.7) y (t = −0.59; P = 0.547). Eight of 36 patients were treated with fumarate, 7 with β-interferons, 7 with glatiramer acetate, 6 with fingolimod, and finally 2 with teriflunomide. For a subgroup of 22 patients, a T2-weighted (T2w) brain MRI scan acquired within a time period of roughly 1 y before participation in our study was available [median = 293 d before participation; range, 132–435 d]. Comparing these images with T2w images acquired during study participation revealed that only 7 of 22 patients had developed new lesions in this period (median = 0 new lesions; range, 0–4 new lesions). Consistently, the median number of days since the end of the last relapse across all 36 patients was 654 d (range, 22–3,550 d). Together, these findings suggest that disease activity at or around the time of our study was fairly small. See Table 1 and Fig. S1 for further patient characteristics.

Table 1.

Patient characteristics

Parameter	Median (range)
Disease duration since diagnosis, y	6.4 (0.3–21.2)
T2w lesion volume, 102 mm3	44.6 (0.9–720.9)
Relapses since diagnosis	5 (1–21)
Days since last relapse	654 (22–3550)
EDSS	3.5 (1.5–6)
FSS, BB	1 (0–4)
FSS, BS	1 (0–3)
FSS, CB	2 (0–3)
FSS, CE	1 (0–2)
FSS, PY	2 (0–4)
FSS, SE	2 (0–4)
FSS, VI	1 (0–3)

BB, bowel and bladder; BS, brainstem; CB, cerebellar; CE, cerebral; PY, pyramidal; SE, sensory; VI, visual.

Fig. S1.

GM and WM fraction and lesion load. (A) The fraction of all voxels classified as GM, WM, lesioned tissue, or CSF relative to the sum of these voxels for patients and controls separately. (B) The spatial distribution of lesions across patients superimposed on the WM tissue template included in SPM12, specifically, the percentage of patients having focal T2w lesions for voxel coordinates in MNI space. (C, Left) Group differences between patients and controls with regard to GMF. (Right) Group differences between patients and controls with regard to WMF.

Psychophysiological Stress Responses, Mental Arithmetic Performance, and Cognitive Load.

As expected because of the performance-dependent adjustments in task frequency in the stress paradigm, the link between performance (the number of correct trials during the last 8 min of stage IVb) and cognitive load (the mean duration of intertrial intervals during that period) across participants was strong (t = −18.89; P < 10⁻⁴). Please note that only trials in the last 8 min of IVb were evaluated to control for measurement duration across conditions and equal feedback settings (Materials and Methods, Experimental Design). Patients performed worse than controls (t = −2.16; P = 0.019) and had a lower cognitive task load (t = 2.10; P = 0.021). The paradigm induced a fast psychological stress response (i.e., a significant positive difference in stress ratings for stage V vs. stage III: t = 6.20; P < 10⁻⁶) and a fast response of SNS-related measures (i.e., the difference between the average heart rate in the last 8 min of stage IVb and the average rate across stage II; t = 7.20; P < 10⁻⁷). For salivary cortisol, no fast stress response was observed. The paradigm did not induce a slow or lasting stress response in any of the three response measures (heart rate: differences between stages VI and II; perceived stress and cortisol: differences between stages VII and III). None of the stress responses (i.e., fast increases in perceived stress and heart rate) was (linearly) related to cognitive task load. Please also see Table 2, the supplementary analysis of psychophysiological stress response measures investigating nonlinear associations in SI Materials and Methods, Psychophysiological Stress Responses, Mental Arithmetic Performance, and Cognitive Load, and Figs. S2 and S3 for further details.

Table 2.

Psychophysiological stress response variables and cognitive task load

	Fast response		Slow response
Response variable	t	P	t	P
Heart rate	Stage IVb–II		Stage VI–II
Main effect	7.2	<10−7	−0.6	0.442
Group effect	−1.5	0.133	−0.1	0.900
Task load	0.4	0.724	0.0	0.983
Cortisol	Stage V–III		Stage VII–III
Main effect	−3.0	>0.99	−2.7	>0.99
Group effect	−0.6	0.557	−1.1	0.306
Task load	0.8	0.408	0.7	0.528
Perceived stress	Stage V–III		Stage VII–III
Main effect	6.2	<10−6	1.3	0.145
Group effect	−1.1	0.271	−1.5	0.150
Task load	0.1	0.937	1.1	0.262

Main effects of stress on response variables (response differences between pairs of experimental stages) were tested using sign tests across variables of all participants (corrected for mean-centered gender, age, and cognitive task load but not for the overall mean). Group effects on response variables were tested with regression and permutation testing (covariates of no interest: gender, age, task load, constant) as well as the association of task load and response variables (covariates of no interest: gender, age, constant).

Fig. S2.

Stress-related heart rate, salivary cortisol, and perceived stress data. (A) The course of the heart rate is shown separately for each of the three ASL measurement blocks and is aggregated across each participant group. Black dots in the boxes correspond to the median; the lower edge of the box is the first quartile, and the upper edge is the third quartile for a time window covering 20 s. The whiskers indicate the first (third) quartile minus (plus) 1.5 times the interquartile range. More extreme values are considered outliers and are plotted as empty circles. The gray shading in the graph for stage IV depicts time points during stage IVa and potentially early time points during stage IVb that were discarded from all heart-rate analyses to control for measurement duration across conditions and equal feedback settings across participants. (B) The course of salivary cortisol concentrations. (C) The course of self-reported data on perceived stress across the paradigm.

Fig. S3.

Associations between heart-rate accelerations (i.e., average heart rate during the last 8 min in stage IVb minus the average heart rate during stage II, corrected for linear effects of gender and age) and cognitive task load (i.e., the mean intertrial interval during the last 8 min in stage IVb). R², coefficient of determination.

Stress-Induced Brain Activity and MS Disease Parameters.

Four fMRI analyses were conducted to investigate mechanisms of fast and slow neural stress responses and their link to MS disease parameters. Specifically, subject-specific voxel contrast maps denoting the difference in local cerebral blood flow (CBF, measured in milliliters per 100 g per minute) averaged across the last 8 min of stage IVb (the mental arithmetic task with feedback) minus the CBF averaged across stage II (baseline 1) were used as parameters of fast neural stress effects. Maps assessing this difference between stages VI and II were used as indicators of slow stress effects. We report coordinates significant on a familywise error (FWE)-corrected level (α_FWE = 0.05).

fMRI analysis 1: Neural stress response.

In analysis 1a, one-sample t tests were conducted on the voxel level across all participants to identify brain areas showing fast or slow neural stress responses (i.e., stress-induced increases in activity). In analysis 1b, we tested for differences between MS patients and controls in these responses using two-sample t tests. Both analyses were restricted to coordinates located in a GM group mask (see Materials and Methods, MRI Preprocessing and SI Materials and Methods, MRI Preprocessing), and in both analyses gender and age were modeled as covariates of no interest. Analysis 1a identified a variety of brain areas showing a fast generic stress response across MS patients and controls, primarily in visual, insular, and cerebellar cortex areas. See Table S1 for further details. No slow neural stress responses were found. Analysis 1b showed that fast and slow neural stress responses do not differ in patients and controls.

Table S1.

Brain regions with higher brain activity during stage IVb (the mental arithmetic task with feedback) than at stage II (baseline 1) across all participants

Region	k	x	y	z	t	pFWE
Calcarine gyrus	4.8	18	−94	−1	11.19	≤0.001
Middle occipital gyrus	4.3	−27	−91	2	9.74	≤0.001
Supplementary motor area	2.8	−6	14	47	8.35	≤0.001
Insula	2.0	33	20	−4	7.70	≤0.001
Inferior parietal gyrus	4.1	−27	−55	44	7.32	≤0.001
Cerebellum	0.4	−9	−73	−22	6.67	≤0.001
Middle frontal gyrus	0.8	36	2	59	6.32	≤0.001
Precentral gyrus	0.6	−42	5	32	6.31	≤0.001
Cerebellum	2.1	−36	−70	−22	5.94	0.002
Cerebellum	0.2	9	−73	−19	5.81	0.002
Precentral gyrus	0.5	−30	−1	50	5.67	0.002
Insula	0.6	−30	20	−1	5.55	0.004
Cerebellum	0.1	−24	−52	−28	5.25	0.006
Vermis	0.8	0	−58	−25	5.09	0.01
Angular gyrus	0.6	30	−52	41	4.97	0.017

k, cluster size in 10³ mm³.

fMRI analysis 2: Clinical disability.

In fMRI analysis 2 (and in fMRI analyses 3, and 4) we searched for brain activity that is linked to MS disease parameters using voxelwise regression analyses modeling gender, age, and a constant term as covariates of no interest. Because we were particularly interested in the link between neural stress responses and disease parameters, we restricted these analyses to regions identified in analysis 1a. Because no slow neural stress effects were found in analysis 1a, we analyzed the link between alterations in the slow neural signal and disease parameters across all areas located in GM (SI Materials and Methods, fMRI analysis S4). We report effect size measures (r) for significant associations (weak effect: 0.1 ≤ r < 0.3; moderate effect: 0.3 ≤ r < 0.5; strong effect: r ≥ 0.5) (19).

To investigate the link between fast neural stress responses and clinical disability, we tested the association between the respective activity changes and Expanded Disability Status Scale (EDSS) scores in analysis 2a and the Functional System Scores (FSS) subscales in analyses 2b–2h (2b, bowel and bladder; 2c, brainstem; 2d, cerebellar; 2e, cerebral; 2f, pyramidal; 2g, sensory; and 2h, visual). These analyses consistently revealed negative associations for coordinates in the left anterior insula. In particular, activity in a cluster of voxels surrounding coordinates −30, 23, 2 [t = −3.97; p_FWE = 0.026; cluster size (CS) = 54 mm³; r = 0.57] in the anatomical standard space defined by the Montreal Neurological Institute (MNI) (20) was negatively linked to the EDSS score (analysis 2a). Activity in this peak voxel coordinate also showed a significant negative association (t = −2.38; P = 0.021; r = 0.39) with fatigue [i.e., measured with the Modified Fatigue Impact Scale (MFIS)] (21) (see Discussion). When controlling for fatigue in addition to gender and age in a voxelwise analysis, the link between the EDSS score and activity in this coordinate was no longer significant.

For the FSS cerebral scale (analysis 2e), activity in a cluster of left insula coordinates surrounding MNI −30, 20, 5 (t = −4.39; p_FWE = 0.007; CS = 459 mm³; r = 0.61) was found [in addition to activity in a cerebellar area surrounding MNI coordinates −33, −64, −25 (t = −4.01; p_FWE = 0.019; CS = 432 mm³; r = 0.58)]. Activity in this single insula peak coordinate was not linked significantly with fatigue (t = −1.88; P = 0.065; r = 0.32). The link between activity in the peak insula coordinate and the FSS cerebral scale remained significant in a voxelwise analysis when fatigue was additionally controlled for (t = −3.96; p_FWE = 0.023; CS = 27 mm³; r = 0.58).

Finally, activity in a cluster of voxels surrounding MNI coordinates −30, 23, −1 in the insula (t = −5.24; p_FWE = 0.003; CS = 270 mm³; r = 0.68) was negatively associated with the score on the FSS pyramidal scale (analysis 2f). The association remained significant when additionally controlling for fatigue (t = −4.59; p_FWE = 0.009; CS = 108 mm³; r = 0.64), and activity was significantly linked to the MFIS score on a single-voxel level (t = −2.50; P = 0.017; r = 0.4) (see Fig. S4).

Fig. S4.

The association between stress-induced brain activity and clinical disability. (A) The overlap among coordinates with a significant negative association with stress-induced changes in brain activity (corrected for gender and age) and the EDSS, FSS cerebral, or FSS pyramidal subscale. Axial and coronal slices are shown in neurological orientation. (B) The association between the EDSS score and stress-induced brain activity (corrected for gender and age) in the peak coordinate identified in fMRI analysis 2a (MNI coordinates: −30, 23, 2). (C and D) The corresponding associations for the FSS cerebral (fMRI analysis 2e) (C) and FSS pyramidal (fMRI analysis 2f) (D) subscales. ΔCBF, difference in cerebral blood flow (stress minus baseline I); CE, cerebral; PY, pyramidal; r, effect size (see Materials and Methods).

fMRI analysis 3: Clinico-radiographic measures of disease severity.

In fMRI analyses 3a (total lesion volume), 3b (total number of relapses since diagnosis), and 3c (disease duration), no significant relations were found between stress-induced increases in fast brain activity and clinico-radiographic measures of disease severity.

fMRI analysis 4: Brain volume.

In analysis 4a, a significant positive association between fast neural stress responses and the GM fraction (GMF) in patients was found in the left cerebellum (MNI coordinates −42, −61, −22; t = 4.71; p_FWE = 0.004; CS = 108 mm³; r = 0.64) and in the supplementary motor area (SMA; MNI coordinates 0, 20, 47; t = 4.10; p_FWE = 0.030; CS = 81 mm³; r = 0.59). Interestingly, when we repeated this voxelwise analysis for controls in the peak voxel cluster identified in patients, a significant positive association between stress-induced brain activity and GMF was found for cerebellar MNI coordinates −45, −61, −22 (t = 2.10; p_FWE = 0.047; CS = 27 mm³; r = 0.45). Analysis 4b revealed a negative link between fast neural stress responses and WM fraction (WMF) in the left middle occipital gyrus (MNI coordinates −30, −64, 38; t = −4.64; p_FWE = 0.011; CS = 189 mm³; r = 0.63) in patients. No significant associations were found between stress-induced brain activity and WMF for the corresponding analysis based on data for controls (Fig. 2).

Fig. 2.

The association between fast neural stress responses and brain volume. (Left) The brain slices depict coordinates with a positive link of stress-induced brain activity and GMF in patients surrounding the peak coordinate identified in fMRI analysis 4a (MNI coordinates −42, −61, −22). The scatterplot illustrates the association of GMF and stress-induced brain activity (corrected for gender and age) at MNI coordinates −42, −61, −22 for patients and the association between activity and GMF at MNI coordinates −45, −61, −22 in controls. (Right) The corresponding results for WMF identified in fMRI analysis 4b.

SI Materials and Methods

Multiple Regression and Permutation Testing.

In the present study, several effects and associations were analyzed using linear multiple regression and permutation testing. Specifically, we used this technique to analyze group differences in cognitive load, task performance, changes in psychophysiological stress response variables, and brain volume (i.e., GMF and WMF). In these analyses, a dichotomous group membership vector (1 = MS patients, 0 = controls) served as covariate of interest. Moreover, the technique was used to analyze the associations between the following covariates of interest and criteria: cognitive load and task performance; cognitive load and stress response variables; and fatigue and disability-related activity (identified in peak voxel coordinates in fMRI analyses 2a, 2e, and 2f).

In all analyses, a constant term and the mean-centered factors gender and age were used as covariates of no interest. In analyses of group differences in psychophysiological stress-response variables, the average intertrial interval for each participant during the last 8 min of stage IVb was used as an additional covariate of no interest.

To determine probabilities for observing an effect of a given covariate of interest by chance, we permuted that factor’s vector (but not the vectors of the covariates of no interest) 10,000 times and computed a regression analysis (i.e., one t-statistic for this factor) for each of the permutations with the glm.fit algorithm included in Matlab 2014a (MathWorks). In the next step, we determined the proportion of t-statistics in the permutation distribution with a (undirected test: absolute) magnitude at least as extreme as the (undirected test: absolute) t-statistic observed for the nonpermuted covariate of interest to obtain a permutation-based P value (41). Please note that we chose nonparametric permutation testing for inference in all analyses of MRI and non-MRI data because that method depends less on the fulfillment of distributional assumptions in regression than does the parametric procedure (41). This consideration is important, because EDSS and FSS scores have only an ordinal scale level.

MRI Sequences.

We measured brain images using a 3-T whole-body tomograph (Magnetom Trio; Siemens) and a standard 12-channel head coil. Perfusion brain images were acquired using a pseudocontinuous ASL EPI sequence (40) comprising 22 ascending transversal slices and covering the whole brain [slice thickness, 5.75 mm including a 5% interslice gap; in-plane voxel resolution 3 × 3 mm; repetition time (TR) = 4,000 ms; echo time (TE) = 19 ms; flip angle (FA) = 90°; field of view (FOV) = 192 × 192 mm; matrix size = 64 × 64; label duration 1.5 s, postlabel delay 1.2 s; phase-encoding direction anterior to posterior]. With this sequence, 120 images (60 control, 60 labelled) were acquired in stage II (baseline 1) and stage VI (baseline 2) (duration 8 min each), and 180 images (90 control, 90 labelled) were acquired during stage IV (stress) (duration 12 min). ASL fMRI instead of blood oxygenation level-dependent (BOLD) fMRI was used to measure brain activity because ASL fMRI has better baseline stability and is much less sensitive to signal artifacts, having a slow temporal profile similar to that of the stress response (12, 14).

To enable distortion correction of ASL images (see ASL fMRI images below), two spin-echo EPI reference volumes with opposite phase-encoding directions (anterior-to-posterior and posterior-to-anterior) were acquired in advance of each of the three functional ASL measurements (i.e., in stages II, IV, and VI) with the same parameters as these sequences.

In addition, an anatomical T1w sequence was measured for spatial normalization of ASL images and to determine measures of GM and WM volume or atrophy, respectively (176 slices; slice thickness 1.3 mm; in-plane voxel resolution 1.5 × 1.5 mm; TR = 1,720 ms; TE = 2.34 ms; FA = 9°; FOV = 192 × 192 mm; matrix size = 128 × 128; duration 1 min 43 s).

Finally, we measured a sagittal T2w sequence to facilitate manual lesion mapping (176 slices; 1-mm isotropic voxels; TR = 5,000 ms; TE = 502 ms; FA = 120°; FOV = 256 × 256 mm²; matrix size = 256 × 256; duration 5 min 52 s).

MRI Preprocessing.

Anatomical brain images.

In short, preprocessing of anatomical images comprised (i) manual lesion mapping and lesion volume computation, (ii) determination of a group mask used to constrain fMRI analyses to areas of GM, and (iii) determination of brain volume/atrophy measures.

Lesion mapping and lesion volume computation.

In particular, J.R.B. started the preprocessing of anatomical images by performing a manual lesion mapping based on high-resolution T2w images of individual participants using the OsiriX software toolbox (OsiriX Foundation). The resulting voxel maps or masks, respectively, were used to determine total T2w lesion volume and in further preprocessing steps of anatomical and functional MRI data.

Determination of a GM group mask.

Although it is possible to detect functionally relevant perfusion changes in WM with ASL, doing so requires the use of specific MR sequences (e.g., 42) that were not used in the present study. Consequently, we determined a group mask that comprised all GM areas across all participants to constrain fMRI analysis 1 and several supplementary fMRI analyses (see below) to coordinates with valid and interpretable perfusion signals.

In particular, we first used SPM12 to coregister the T2w images and the lesion masks derived thereof with the T1w anatomical images. Subsequently, we spatially registered/normalized the T1w anatomical image and the coregistered lesion mask of each patient to the anatomical standard space defined by the MNI (20) brain template. For this step, we used the combined spatial normalization and segmentation algorithm included in SPM12. Coordinates located in lesioned tissue as denoted by a subject’s coregistered lesion masks were excluded from this registration procedure. The voxel size of the normalized images was 3 × 3 × 3 mm.

To create the GM group mask, we computed the voxelwise average tissue probability for GM, WM, and cerebrospinal fluid (CSF) across tissue probability maps of all participants determined in the MNI space that did not have a lesion at a given voxel coordinate. The presence of a lesion at a given coordinate was determined based on the spatially normalized lesion masks. In the next step, a voxel coordinate was assigned to that tissue class for which the maximal averaged probability score among the three tissues was computed. To account for spatial smoothing of the functional ASL data, we finally smoothed the GM mask with an isotropic 3D Gaussian filter with an FWHM of 8 mm. After this procedure, the GM mask comprised roughly 60% of the total brain volume.

Determination of brain volume/atrophy measures.

As mentioned above, spatial normalization of an image to a standard space performed during determination of the GM group mask was done by a combined spatial normalization and segmentation algorithm in SPM12. This algorithm simultaneously computes the spatial registration/transformation and voxelwise tissue probability maps for GM, WM, and CSF in the original subject-specific and the anatomical standard space. We used tissue probability maps computed in original space and T2w lesion masks coregistered to the respective T1w images in original space to compute measures of brain volume (i.e., GMF and WMF) for each participant. In particular, we first assigned each voxel in the brain to one of these tissue classes: GM, WM, CSF, or lesioned tissue. Voxels not located in lesioned tissue were assigned to the tissue with the maximal probability parameter for GM, WM, or CSF. Voxels located in lesioned tissue were always classified as lesioned voxels irrespective of their GM, WM, or CSF probability. In the next step, we divided the number of GM voxels by the sum of all voxels located in the brain to assess the GMF. To assess the WMF, we divided the sum of WM and lesion voxels by the sum of all voxels in the brain. Please note that we considered lesion voxels as WM because lesions are located predominately in WM (see Fig. S1) because proportion of myelin is much greater in WM than in GM. Therefore excluding lesioned voxels would systematically underestimate WMF in patients (38).

ASL fMRI images.

Preprocessing of fMRI images involved the following steps: realignment, distortion correction, coregistration to T1w anatomical images, smoothing, determining local CBF, spatial normalization, and computing voxelwise contrast maps denoting the difference in local CBF between experimental stages.

In the first step, ASL perfusion images acquired in the different experimental stages of each subject were realigned to the spin-echo EPI reference volume with the anterior-to-posterior phase-encoding direction acquired directly before each of the functional ASL blocks or stages, respectively. We did so using the realignment algorithm included in the ASL software ASLtbx for SPM (43). Importantly, we used only ASL images acquired during the last 8 min of stage IVb for realignment and all subsequent processing steps performed for stage IV to control for measurement duration across conditions and equal feedback settings within the stress condition (see Materials and Methods, Experimental Design). In the next step, we used the FSL software Topup (44) to correct the realigned ASL images for spatial image distortions caused by susceptibility-induced magnetic field inhomogeneity based on both spin-echo EPI reference images with opposing phase-encoding direction. Subsequently, we coregistered the realigned and undistorted ASL images of a subject and condition to the T1w anatomical image of that subject and smoothed these coregistered images with an isotropic 3D Gaussian filter (FWHM 8 mm). Based on the resulting images, we computed the average local CBF from control-label pairs with ASLtbx (43) across the 8-min measurement interval for each of the three conditions or stages, respectively. Then we applied the transformation parameters determined during spatial normalization of the T1w image to the average local CBF maps for spatial normalization of local CBF maps (voxel size 3 × 3 × 3 mm). In the last step, we computed contrast maps assessing the difference between average local CBF for the last 8 min of stage IVb minus the average local CBF in stage II and average local CBF in stage VI minus average local CBF in stage II. These difference maps served as source data for the fMRI group analyses.

Psychophysiological Stress Responses, Mental Arithmetic Performance, and Cognitive Load.

Measurement of psychophysiological stress response markers.

Heart rate.

Heart rate data were acquired during fMRI scans using the standard pulse oximeter of the physiological monitoring unit (PMU) provided with the MRI scanner (Magnetom Trio; Siemens). The pulse oximeter measures changes in infrared light transmission with a sampling frequency of 50 Hz. Because participants had to use their hands during the stress protocol to respond to the mental arithmetic tasks, the photoplethysmograph detector was placed around a participant’s toe. Furthermore, because occasionally spurious pulse oximeter signals may be detected as heartbeats, falsely suggesting a sudden increase in the heart rate, and because weak signals may lead to skipped heartbeats, falsely suggesting an apparent decrease in the heart rate, we filtered the heart rate signal computed by the PMU software to determine a measure of SNS activity. In particular, we first excluded heartbeats that were detected within the baseline pulse oximeter raw signal (values of the pulse oximeter raw signal below 2,000) and secondly those that induced a heart rate acceleration to 133% or more or a deceleration to 75% or less. The remaining heartbeats were used for heart rate computation. Please note however, that all analyses of heart rates measured during stage IVb were restricted to heart rates assessed in the last 8 min of that stage.

Salivary cortisol.

To measure the activity of the HPA system, we assessed salivary cortisol levels. Salivary cortisol release is a delayed peripheral response to stress and varies according to circadian rhythm (23). To control for the diurnal changes in cortisol secretion, fMRI scans and cortisol measurements in all participants were carried out between 3:00 and 7:00 PM. The study visit included a 30-min resting period for all participants before the fMRI stress protocol; during that period participants completed questionnaires. A salivary cortisol collection system (Sarstedt, Nuembrecht, Germany) was used to take saliva samples. Subjects were asked to keep the saliva collection device in their mouth for 2 min before putting it back into the container (see Materials and Methods, Experimental Design for the timing and frequency of cortisol sampling). Samples were stored at −80 °C until assayed by ELISA (IBL). Salivary cortisol concentrations were measured in nanomoles per liter.

Perceived stress/self-report data.

Participants were asked to rate their current feeling of distress, frustration, anxiety, and relaxation on a 9-point scale presented on a projection screen in front of them. The scale ranged from “not at all” at the leftmost position to “very strong” on the rightmost position on the scale. Participants used MRI-compatible button boxes to make the selection. The rating was done four times during the fMRI stress protocol (see Materials and Methods, Experimental Design for details). Please note that we focused only on self-reported levels of stress in this study.

Nonlinear associations between psychophysiological stress responses and cognitive load.

Recently it was suggested (27) that task-induced heart rate acceleration in a mentally effortful paradigm (serial subtraction) might reflect the functioning of a cardiac mechanism serving to adjust brain glucose delivery to increased cognitive and thus metabolic task demands and might not necessarily reflect psychological stress. To evaluate whether this suggestion might apply in our study, we tested a linear association of heart rate accelerations (average heart rate during the last 8 min of stage IVb minus the average rate during stage II) and cognitive task load (average intertrial interval during that period), as described in the main text. In this supplementary analysis, we further evaluated whether these parameters are linked in a nonlinear fashion. In particular, we first corrected the above-mentioned heart rate measure for gender and age using linear regression and then computed nonlinear associations between the corrected measure and cognitive load with polynomial regression (using polynomials of first to fifth degree). For each of the five analyses, we computed the coefficient of determination and a probability measure assessing the probability that this coefficient of determination occurs by chance with permutation testing (10,000 permutations of the criterion vector).

Group Volume Differences in GM and WM.

Permutation-based regression was used to test potential group differences (MS patients < controls) in GM and WM volume (i.e., GMF and WMF). Gender, age, and a constant term were used as covariates of no interest.

Stress-Induced Brain Activity and MS Disease Parameters.

fMRI analysis S1: Neural stress response and cognitive load.

In this supplementary analysis, we investigated the link between stress-induced increases in neural activity and the average duration of intertrial intervals computed for each participant in the areas identified in fMRI analysis 1a. Gender, age, and a constant term entered the regression analysis as covariates of no interest. We searched for positive as well as negative associations.

fMRI analysis S2: Cognitive task load and the link between fast neural stress responses and MS disease parameters.

In this analysis, we tested whether the results found in fMRI analyses 1–4 described in the main text depend on the (linear) covariation between fast neural stress responses and cognitive task load. To do so, we repeated the corresponding analyses but modeled the mean intertrial interval during the last 8 min of stage IVb as a covariate of no interest in addition to gender, age, and a constant term. All other parameters were identical to the parameters used in the analyses described in the main text.

fMRI analysis S3: Fast neural stress response and MS disease parameters across all GM areas.

We repeated the fMRI analyses included in analyses 2–4 in the main text for MS patients but now without restricting them to areas showing a fast neural stress response as identified by analysis 1a. Instead, the analyses were conducted across all coordinates located in the group GM mask (see above). All other parameters were identical to the parameters used in the analyses in the main text. The aim of these supplementary analyses was to evaluate the specificity of the link between areas showing a fast neural stress response and MS severity parameters.

fMRI analysis S4: Slow alterations in brain activity and MS disease parameters across all GM areas.

In this analysis, we tested the link between alterations in the slow brain activity signal using voxel contrast maps denoting the difference in the average local CBF for stage VI (baseline 2) minus the average local CBF for stage II (baseline 1) across all voxel coordinates located in the GM group mask and MS disease parameters in patients using multiple regression. Gender, age, and a constant term were used as covariates of no interest. We searched for positive as well as negative associations.

Discussion

In this study, we investigated fast and slow neural and psychophysiological stress responses and related these signals to MS disease parameters in a cohort of clinically stable MS patients. We demonstrate that fast neural stress responses are associated with clinical disability and brain atrophy in MS.

First we investigated a basic stress response across patients and controls. On the level of neural activity, this response was investigated in fMRI analysis 1a by evaluating fast increases of activity (from stage II to stage IVb) or slow increases (from stage II to stage VI) across all participants. We searched only for increases in stress-related activity because the great majority of stress-response parameters described in the literature are parameters of increased (not decreased) activity. For example, psychological stress induces increased cerebral perfusion (22), heart rate, growth hormone, prolactin, and cortisol secretion (23), NE secretion (24), and cognitive coping (25). fMRI analysis 1a revealed a distributed set of regions showing a fast stress response that largely overlapped with those areas found in healthy subjects (14), especially in dorsomedial prefrontal cortex (the SMA), inferior parietal areas, and insular cortex. Analysis 1a also identified visual regions and coordinates in the cerebellum, a brain area receiving strong norepinephrinergic inputs (26; also see below). On the contrary, no brain areas showing slow stress responses were found.

The analysis of psychophysiological stress parameters showed that the mental arithmetic task triggered fast increases in perceived stress and heart rate. Importantly, none of these task effects was (linearly) related to individual variations in cognitive task load (see SI Materials and Methods, fMRI Analysis S1 for neural data). Furthermore, a supplementary analysis of psychophysiological stress response measures (SI Materials and Methods, Psychophysiological Stress Responses, Mental Arithmetic Performance, and Cognitive Load, Nonlinear Associations Between Psychophysiological Stress Responses and Cognitive Load) showed that heart rate accelerations and cognitive load are not related in a nonlinear fashion. Together, these findings strongly suggest that we are not falsely interpreting as psychological stress (27) the functioning of a cardiac mechanism that simply serves to adjust brain glucose delivery to increased cognitive and thus increased metabolic demands during the mental arithmetic task (stage IVb) relative to baseline (stage II). Finally, when additionally considering that cognitively demanding and socially evaluative tasks such as our task trigger NE release (24), one can conclude that our task successfully induced psychological stress.

Group differences in fast or slow neuronal stress responses were investigated in analysis 1b, which failed to identify such differences. This result might appear counterintuitive, given that pharmacological challenge studies have found differences in glucocorticoid responsivity (e.g., ref. 5). However, our result is consistent with the absence of group differences in markers of fast or slow psychological (perceived) stress or SNS-related stress (heart rate accelerations) in our study.

fMRI analyses 2–4 investigated the link between brain activity and MS disease parameters. Because we were particularly interested in the link between stress-related brain activity and MS disease parameters, in analyses 2–4 we searched only in those brain areas that showed a significant generic stress response in analysis 1a. Because no slow neural stress responses were found, we first discuss findings made for fast responses in analyses 2–4 and then briefly discuss the findings made in the corresponding supplementary fMRI analysis S4 for slow signal variations.

Specifically, the association between stress-responsive areas and clinical disability was evaluated in fMRI analysis 2. This analysis revealed a consistent negative link between fast neural stress responses and three disability markers (i.e., the EDSS and the cerebral and the pyramidal FSS subscales) in overlapping left anterior insular areas and thus might suggest an important association between insula functioning and MS disability, in general. At this point, it must be noted that the interrelatedness among EDSS scores and FSS subscales might have contributed to these consistent associations (cf. ref. 18). However, because the functions underlying these scales are realized by distributed networks, and MS neuropathology typically evolves in distributed regions across the disease course (28, 29), activity in a single region very well might be coupled to different but interrelated disability markers via a single functional mechanism. Thus, the interrelatedness among clinical scales does not severely affect the validity of this finding.

Findings made in an fMRI study investigating neural foundations of so-called “sickness behavior” in healthy subjects (30) may help explain the functional link between insular activity and disability in MS. Specifically, sickness behavior denotes a group of symptoms observed during the course of systemic infections including fatigue, depression, reduced exploratory behaviors, fever, and impaired cognitive performance (e.g., ref. 31). Sickness behavior is understood as an adaptive mechanism aiming to suppress nonimmunologically relevant behaviors during infections to reserve metabolic resources for immunological processes (30; see ref. 32 for further details on the complex interplay between immune processes and biologically relevant behaviors such as eating). In ref. 30, either a typhoid vaccination or placebo was applied on one of two measurement days. Subsequently, inflammation, self-report variables (including fatigue), and cognitive performance were measured in an fMRI Stroop task. As expected, vaccination induced a marked inflammatory response and simultaneously induced fatigue. Neural activity in the insular cortex was of outstanding importance because it increased in response to vaccination, and inflammation-induced activity in the insula correlated positively with fatigue. Consequently, the insular cortex may be understood as a region monitoring inflammation and (directly or as part of a larger network) inducing sickness behavior (e.g., fatigue) in response to inflammation.

To investigate whether similar mechanisms might contribute to findings made in analysis 2, we tested whether the insular activity related to clinical disability in the peak voxel coordinates identified in analyses 2a, 2e, and 2f was also related to patients’ fatigue (measured with the MFIS). These analyses revealed a significant link between fatigue and EDSS-related activity (t = −2.38; P = 0.021) and activity related to the FSS pyramidal scale (t = −2.50; P = 0.017) and thus supported the relevance of the findings in ref. 30. However, contrary to the findings in ref. 30, the link was negative. Interestingly, such a negative association is consistent with inverse associations of other stress-related parameters and inflammatory processes in MS [e.g., elevated glucocorticoid activity (5) and reduced glucocorticoid suppression of inflammatory cytokine production (33)] and thus might indicate the existence of a depleted neuroimmunological process. Speculatively, this depletion might result from the immune system being confronted with a type of inflammation that it cannot cope with adequately (i.e., inflammation driven by autoimmune processes rather than by external pathogens) and that leads to continuous but ineffective neuroimmunological activity. As a consequence, the immune system triggers sustained fatigue in a misguided attempt to save metabolic resources for a process falsely assumed to be of a short, temporary nature. Please note, however, that other MS studies found a link between fatigue and HPA-axis hyperactivity (34), regional GM and WM atrophy, and lower fractional anisotropy in several WM areas (35). Consequently, dysregulated and inflammation-triggered sickness behavior might be considered as only one among several complex factors contributing to fatigue in MS.

To determine the relevance of the covariation between activity and fatigue for the covariation of activity and disability, we conducted voxelwise analyses modeling the MFIS as an additional covariate of no interest in analysis 2. Importantly, the link between both the FSS subscales and fast insular stress responses remained significant under these circumstances. These findings clearly argue that the link between disability and activity must be driven by further processes in addition to fatigue, an assumption that is consistent with other immunity-related processes regulated by the insula such as autonomic nervous system activity (36).

In analysis 3, we investigated the link between stress-induced activity and clinico-radiographic measures of disease severity, i.e., T2w lesion volume, the number of relapses, and disease duration. However these analyses failed to identify such associations. Speculatively, this lack of association can explained in part be by the rather weak link between brain lesions and clinical disability in MS (e.g., ref. 37).

In analysis 4 we investigated the associations between fast neural stress responses and GM (analysis 4a) and WM volume (analysis 4b). In 4a, a positive association between stress-induced activity and GM volume (i.e., a negative association between activity and atrophy) was found in the SMA and the left cerebellum in MS patients. The latter finding is compatible with the observations that the cerebellum receives major NE projections from the locus coeruleus (26) and that the stress hormone NE is closely linked to inflammatory processes (9, 10). Thus, given that inflammation is a source of GM atrophy (38), which is pronounced in the cerebellum (29, 39), one might conclude that the negative link of stress-induced brain activity and GM atrophy might be mediated by the NE system. Importantly, analysis 4a also showed that stress-induced activity is positively related to GM volume in overlapping cerebellar areas of controls. Thus, contrary to the conclusions that might be derived using only patient data (and supported by the absence of a link between stress-induced activity and disease duration), the association between stress-induced activity and GM volume cannot reflect only heightened stress sensitivity in response to MS. Instead, it might be at least partly indicative of a generic mechanism of neurodegeneration.

We conducted several supplementary analyses in addition to those described above. In these analyses, we found GMF and WMF were lower in patients than in controls, as is compatible with neurodegeneration being a key feature of MS (1). Furthermore, in fMRI analysis S2 we evaluated whether the associations between brain activity and MS disease parameters tested in analyses 2–4 depend on task load and found that they did not. Thus, together with the findings on fast generic stress responses, analysis S2 strongly suggests that the associations between neural signals and MS severity are driven by psychological stress. fMRI analysis S3 evaluated the specificity of the link between areas showing a fast neural stress response and MS disease parameters by repeating fMRI analyses 2–4 across all GM areas and revealed results very similar to those of analyses 2–4. Thus, fMRI analysis S3 nicely confirmed the functional relatedness of neural stress signals and MS disease parameters. Finally, in analysis S4 we tested the association between slow neural signal alterations and MS disease parameters across all GM areas in the brain and found links to the FSS cerebellar scale, the number of relapses, and WMF. However, given the absence of a slow generic neural stress response, the nature of these slow signal alterations and their link to MS disease parameters must be clarified in future studies.

Further explanations are required for the lack of stress-induced increases in cortisol in our study even though we used an experimental protocol almost identical to that described in ref. 14. A possible explanation is that, unlike the Trier Social Stress Test (TSST) (23), our protocol did not include a free speech task and that the induced stress level is lower under these conditions. Alternatively, the delay between the stress task and the last cortisol measurement might have been too short to account for the slow temporal characteristics of the salivary cortisol response (22), and thus a longer interval should be chosen in future studies. However, given the findings we made and that our task is less burdensome than the TSST (an important consideration in a study involving patients), we believe our task is well suited for addressing the research objectives we investigated.

Materials and Methods

Participants.

Forty-three MS patients and 27 controls were recruited for participation in this study. Suitable participants were referred by the Charité MS outpatient clinic. Controls were recruited through advertisements. The inclusion criteria for patients were (i) a diagnosis of relapsing-remitting MS (RRMS) or secondary progressive MS (SPMS) according to McDonald Criteria 2010 (28); (ii) stable disease-modifying treatment for at least 6 mo or no disease-modifying treatment; (iii) age ≥18 y; and (iv) the physical and mental capability to use the test devices without restrictions. Patients were examined and diagnosed by experienced neurologists. Clinical disability was assessed using the EDSS and FSS (18). Potential participants presenting with mental or addictive disorders, neurologic diseases other than MS, acute MS relapses, acute infections, MRI contraindications, or pregnancy were excluded. Except for MS diagnosis, relapses, and treatment, the inclusion and exclusion criteria were the same for patients and controls. From the 64 participants remaining after the application of these criteria, seven subjects were additionally excluded from the analysis because of incomplete imaging datasets. Consequently, the fMRI data from 57 participants [35 females and 22 males; 36 MS patients (27 RRMS; 9 SPMS) and 21 controls] were available for analyses of fast neural stress effects, and data from 52 participants (34 patients) were available for analyses of alterations in slow neural signals. Data from 57 participants were available for analyses of fast and slow variations in perceived stress. Heart rate data for analyses of fast and slow signal variations were available from 46 participants (30 MS patients), and salivary cortisol data were available from 20 participants (13 patients). The two groups (36 MS patients and 21 controls) were comparable in terms of gender and age. Written consent was obtained from participants according to the Declaration of Helsinki. The study was approved by the research ethics committee of the Charité–Universitätsmedizin Berlin (EA1/182/10, amendment V).

Experimental Design.

We used a version of an established ASL fMRI stress protocol (14) derived from the TSST (23) to induce and measure the impact of mild psychological stress (a mental arithmetic task with social performance feedback) on regional brain activity, SNS (heart rate), HPA axis (salivary cortisol), and perceived stress (Fig. 1). Before the experiment started, participants were told that they would participate in a mental arithmetic task and would receive feedback relating their performance to performance parameters in the overall population. After the protocol, participants were informed that the performance evaluation (expressed as a school grade) was computed based on their arithmetic performance in the adaptation stage.

MRI Sequences.

Brain images were measured with a 3-T tomograph (Magnetom Trio; Siemens) and a 12-channel head coil. Stress-related brain activity was measured with a pseudocontinuous ASL echo-planar imaging (EPI) sequence (40) covering the whole brain. Using this sequence, 120 images were acquired during stages II (baseline 1) and VI (baseline 2) (8 min each), and 180 images were acquired during stage IV (stress) (12 min). To determine brain morphological parameters (see MRI Preprocessing below), anatomical T1-weighted (T1w; 1 min 43 s) and T2w (5 min 52 s) sequences were measured. See SI Materials and Methods, MRI Sequences for further details.

MRI Preprocessing.

Preprocessing of anatomical T1w and T2w images comprised manual lesion mapping and determination of lesion volume, generation of a group mask used to constrain several fMRI analyses to areas of GM, and determination of GM and WM volume (i.e., GMF and WMF). Preprocessing of fMRI images included realignment, distortion correction, coregistration to T1w anatomical images, spatial smoothing, determination of local CBF, spatial normalization, and computation of voxelwise contrast maps denoting the difference in average local CBF for fast stress responses (stage IVb minus stage II) or slow responses (stage VI minus stage II). These maps were computed in the anatomical standard space defined by the MNI (20) and served as source data for fMRI group analyses. Please note that only ASL images acquired during the last 8 min of stage IVb were preprocessed and analyzed to control for measurement duration across conditions and equal feedback settings (see Experimental Design, above).

Psychophysiological Stress Responses, Mental Arithmetic Performance, and Cognitive Load.

For details on psychophysiological stress responses, mental arithmetic performance, and cognitive load please see Results and Table 2.

Stress-Induced Brain Activity and MS Disease Parameters.

fMRI analyses were conducted with the SnPM13 (www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/nichols/snpm), and the SPM12 toolboxes (www.fil.ion.ucl.ac.uk/spm) (Wellcome Trust Centre for Neuroimaging, Institute of Neurology, University College London, London) using permutation testing for inference. Effect size measures (i.e., correlations, r) (19) for significant associations between brain activity and MS disease parameters were computed based on the t-statistic of a corresponding regression coefficient and the given degrees of freedom (df) following the equation r = (t²/[t² + df])^0.5. In addition to the fMRI analyses described in the main text, several supplementary fMRI analyses were conducted. In particular, in SI Materials and Methods, fMRI Analysis S1 we investigated the link between neural stress responses and cognitive load. In SI Materials and Methods, fMRI Analysis S2 we tested whether the findings made in voxelwise fMRI analyses 1–4 for patients and controls depend on the (linear) covariation between fast neural stress responses and cognitive task load by repeating the corresponding analyses but modeling cognitive load as additional covariate of no interest (Table S2). Furthermore, we investigated the link between alterations in fast (SI Materials and Methods, fMRI Analysis S3; Table S3) and slow (SI Materials and Methods, fMRI Analysis S4; Table S4) brain activity variations and MS disease parameters in patients across all coordinates contained in the GM group mask (not constrained to the areas showing a stress response as identified in analysis 1a). These analyses were performed to evaluate the functional specificity of areas showing a fast neural stress response for MS severity parameters and to evaluate whether alterations in slow brain activity are at all linked to these parameters.

Table S2.

Fast neuronal stress responses and their link with disease parameters independent of their linear covariation with cognitive task load

Effect/group/region	k	x	y	z	t	pFWE	r
Neural stress response across all participants
Calcarine gyrus	48.6	18	−94	−1	11.79	≤0.001	N/A
Middle occipital gyrus	45.9	−27	−91	2	9.83	≤0.001	N/A
Supplemental motor area	29.2	−6	14	47	8.32	≤0.001	N/A
Insula	21.1	33	20	−4	7.72	≤0.001	N/A
Inferior parietal gyrus	42.9	−27	−55	44	7.41	≤0.001	N/A
Cerebellum	4.3	−9	−73	−22	6.98	≤0.001	N/A
Cerebellum	24.6	−33	−70	−22	6.41	≤0.001	N/A
Middle frontal gyrus	8.4	36	2	59	6.35	≤0.001	N/A
Precentral gyrus	6.5	−42	5	32	6.33	≤0.001	N/A
Cerebellum	1.9	9	−73	−19	6.05	≤0.001	N/A
Precentral gyrus	5.4	−30	−1	50	5.67	0.002	N/A
Insula	7.3	−30	20	−1	5.62	0.002	N/A
Cerebellum	1.1	−24	−52	−28	5.40	0.004	N/A
Vermis	8.6	0	−58	−25	5.06	0.011	N/A
Angular gyrus	7.6	30	−52	41	5.04	0.011	N/A
FSS cerebral in MS patients
Insula	1.1	−30	20	5	−4.17	0.027	0.60
Cerebellum	0.3	−33	−70	−22	−3.96	0.045	0.58
FSS pyramidal in MS patients
Insula	1.9	−30	23	−1	−4.88	0.006	0.66
GMF in MS patients
Cerebellum	0.5	−45	−61	−22	4.51	0.013	0.65
WMF in MS patients
Middle occipital gyrus	1.4	−30	−64	38	−4.61	0.013	0.64
GMF in healthy controls
Cerebellum	0.3	−45	−61	−22	1.96	0.042	0.44

k, cluster size in 10² mm³; N/A, not applicable; r, effect size.

Table S3.

Fast neuronal stress responses and disease parameters across all GM areas

Parameter/region	k	x	y	z	t	pFWE	r
FSS pyramidal
Insula	0.8	−30	23	−1	−5.24	0.035	0.68
GMF
Cerebellum	0.3	−45	−58	−22	5.3	0.024	0.68
WMF
Middle occipital gyrus	0.5	−33	−67	38	−5.3	0.039	0.68

k, cluster size in 10² mm³; r, effect size.

Table S4.

Alterations in slow brain activity and disease parameters across all GM areas

Parameter/region	k	x	y	z	t	pFWE	r
FSS CB
Inferior temporal gyrus	0.3	−39	5	−37	−5.22	0.042	0.68
Number of relapses
Angular gyrus	17.6	54	−58	26	−6.87	0.002	0.77
Supramarginal gyrus	5.9	60	−46	35	6.09	0.011	0.73
WMF
Gyrus rectus	0.3	0	26	−19	−5.22	0.043	0.68

k, cluster size in 10² mm³; r, effect size.

SI Results

Psychophysiological Stress Responses, Mental Arithmetic Performance, and Cognitive Load.

Time course of psychophysiological stress responses separated by group.

Fig. S2 depicts groupwise raw data time courses for the three response measures.

Nonlinear associations between psychophysiological stress responses and cognitive load.

Fig. S3 summarizes the results of the polynomial regression analyses.

Consistent with the findings reported in the main text, the nonlinear relationships between increases in heart rate and cognitive task load (as assessed by the depicted polynomials) are not significant.

Group Volume Differences in GM and WM.

Patients had a significantly smaller GMF (t = −3.57; P < 10⁻³) and WMF (t = −1.96; P = 0.023) and thus more pronounced brain atrophy than controls (Fig. S1).

Stress-Induced Brain Activity and MS Disease Parameters.

fMRI analysis 1: Neural stress response.

Table S1 gives a detailed overview on findings made in fMRI analysis 1a in the main text.

fMRI analysis 2: Clinical disability.

Fig. S4 shows additional details of findings regarding the link between fast neural stress responses and clinical disability measures identified in fMRI analysis 2 in the main text.

fMRI analysis S1: Neural stress response and cognitive load.

Neither positive nor negative associations of stress-induced brain activity and cognitive load were identified in fMRI analysis S1.

fMRI analysis S2: Cognitive task load and the link between fast neural stress responses and MS disease parameters.

Table S2 summarizes the results of fMRI analysis S2.

fMRI analysis S3: Fast neural stress response and MS disease parameters across all GM areas.

Table S3 shows the results of fMRI analysis S3.

fMRI analysis S4: Slow alterations in brain activity and MS disease parameters across all GM areas.

Table S4 depicts the results of fMRI analysis S4.