Investigation of neuro‐vascular reactivity on fMRI study during visual activation in people with multiple sclerosis using EEG and hypercapnia challenge

Abstract

Background

People with multiple sclerosis (MS) exhibit a different pattern of blood oxygenation level‐dependent (BOLD) activation on functional magnetic resonance imaging (fMRI) studies when compared to healthy control (HC).

Purpose

The objective of this study is to determine whether observed differences in BOLD activation between people with MS (pwMS) and HC participants are due to the differences of neurovascular coupling, cerebral blood flow (CBF) or actual neuronal activity.

Methods

We investigated the neuronal activation in pwMS (n = 11) and age‐ and sex‐matched HC participants (n = 15) using simultaneous electroencephalogram (EEG) and fMRI measures during a visual task (VT) and hypercapnia condition.

Results

Significant neurovascular coupling is observed in both HC and pwMS. Neuro‐vascular coupling ratios are not significantly different between groups. However, we observe significantly lower CBF increase during VT and higher quantitative CBF at a rest state in pwMS than in HC (p < 0.05). From the multiple regression model, in HC group, we found that the BOLD contrast change during VT is best predicted by the EEG power change during VT (Student t‐score = 2.64, p = 0.022), and the CBF change during hypercapnia (Student t‐score = 2.59, p = 0.024). In pwMS, the BOLD contrast change during VT is negatively predicted by the CBF change during VT (Student t‐score = −4.02, p = 0.003).

Conclusion

These findings could explain that BOLD activation in pwMS is mainly determined by the blood flow change during activation rather than the direct neuronal activation measures or hemodynamic vascular reactivity during hypercapnia challenge, suggesting that altered vasodilatory effects in response to task activation in pwMS might be linked to impaired cerebral hemodynamics, possibly leading to the widely observed abnormal BOLD activation in fMRI studies of pwMS.

1. INTRODUCTION

Multiple sclerosis (MS) is a chronic inflammatory and neurodegenerative demyelinating disorder that is characterized by neurological deficits disseminated over space and time. ¹ , ² In people with MS (pwMS), white matter (WM) focal demyelination leads to tissue changes, which results in gliosis and axonal degeneration. ³ , ⁴ , ⁵ Depending on the location and severity of these tissue changes, pwMS may experience neurological deficits in several brain functions including vision, working memory, or motor skills. ⁶ , ⁷ , ⁸ , ⁹

The increased energy demand and reduced ATP production in a demyelinated axon induce a virtual hypoxia in demyelinated tissue, ¹⁰ , ¹¹ leading to vascular dysfunction. This hypoxia condition translates to neurological worsening. ¹² For this reason, a better understanding of the mechanism between neurological deficits and the neuro‐activation with vascular reactivity has implications for therapeutic development. ¹³ , ¹⁴

Altered blood oxygenation level‐dependent (BOLD) activation has been consistently reported in functional magnetic resonance imaging (fMRI) studies of pwMS. Previous research has shown that during attention and memory tasks, pwMS demonstrated greater BOLD activation on fMRI than age‐ and sex‐matched healthy controls (HC), and this increased activation has been shown to correlate with T2 lesion load. ¹⁵ Other studies have found that during working memory tasks, pwMS displayed a different brain activation pattern from that one HC presented. ¹⁶ , ¹⁷ , ¹⁸ In a study incorporating a delayed recognition task, pwMS demonstrated greater activation than HC in the left inferior parietal cortex. ¹⁹ The authors also found that performance on the Paced Auditory Serial Addition Test (PASAT) was correlated with deactivation in an area of the brain involved in the default mode network. Similar findings have since been reported in multiple fMRI studies assessing a variety of cognitive tasks in pwMS. ²⁰ , ²¹ , ²² , ²³ , ²⁴

Functional connectivity (FC) in pwMS has also been assessed in various fMRI studies. The reduced FC between right‐ and left‐hemisphere primary motor cortices was observed in pwMS. ²⁵ The reduced default mode network was observed in pwMS and the default mode network abnormalities was correlated with PASAT score. ²⁶ Two different diffusion tensor imaging studies specifically demonstrated a positive correlation between FC and radial diffusivity in WM regions, ²⁷ , ²⁸ suggesting that reduced functional connectivity is correlated to impaired structural connectivity. However, in a more recent resting‐state (rs‐) fMRI study, pwMS demonstrated patterns of both increased and decreased FC across various brain networks when compared with HCs. ²⁹ In another recent study, reduced dynamic FC was observed on rs‐fMRI in pwMS, and this reduction was found to be correlated with motor and cognitive impairment. ³⁰

These various fMRI findings have been widely interpreted to be evidence of plasticity, or functional reorganization in response to the disease. ³¹ , ³² However, it is also known that pwMS have impaired cerebral hemodynamics when compared with HCs. Reduced cerebral oxygen consumption rate (CMRO₂) and cerebral blood flow (CBF) were reported in both WM and peripheral cortical grey matter (GM) in pwMS, compared to a group of HC. ³³ Decreased CBF and cerebral blood volume (CBV) values were also reported in both normal‐appearing WM and deep GM in pwMS compared with HC. ³⁴ , ³⁵ Enhancing lesions demonstrate increased CBF and CBV, compared to contralateral normal‐appearing WM in pwMS. ³⁶ Because BOLD contrast change during the activation depends on the change in CMRO₂ and CBF during the activation, both of which are known to be compromised in pwMS, increased BOLD activation on fMRI may reflect altered cerebral hemodynamics rather than neuronal reorganization.

Several recent studies have used resting‐state electrophysiological recordings such as electroencephalogram (EEG) and magnetoencephalogram (MEG) in an attempt to further explore the changes in neuronal activity seen in pwMS. These studies found that pwMS showed the increased frequency power in alpha‐1 band, ³⁷ and alpha‐1 and 2 bands ³⁸ while another study presented the decreased frequency power in alpha band ³⁹ to be compared to HC. However, the relationship between neuronal activity and neurovascular changes during visual paradigms remains unclear.

In this study, we sought to compare BOLD and CBF change during VT and hypercapnia condition in HC and pwMS. Also, rs‐fMRI, quantitative CBF (qCBF), CMRO₂ and EEG power increase during VT were compared between groups. We investigated the coupling of BOLD contrast change during VT to multimodal EEG and other MRI measures during VT and hypercapnia condition in pwMS and age‐ and sex‐matched HC using multi‐linear regression models. Using these measures, we sought to determine whether there were significant differences between pwMS and HC in the linear relationship of BOLD contrast change during VT to any EEG and/or MRI measures during neuronal activity and/or hypercapnia challenge.

2. METHODS

Thirteen pwMS and 15 age‐ and sex‐matched HC participants were recruited into the study. Enrolled MS participants were required to be on stable disease modifying therapy for at least 6 months and free from relapse or steroid treatment for 8 weeks. A diagnosis of hypertension was exclusionary for all potential subjects. In addition, all participants were instructed to prohibit from taking caffeine‐containing food or beverages 24 h before the study date. Two MS patients were excluded from the study due to issues with the EEG acquisition on the scan date (see EEG acquisition and analysis section). A total of 11 pwMS (mean/standard deviation age = 53 ± 6 years; seven women) and 15 HC participants (53 ± 9 years; nine women) met the inclusion criteria (Age > 18) and were included in the final analysis. Data were collected with approval from the Institutional Review Board after informed consent. The Expanded Disability Status Scale (EDSS) score was measured in 11 pwMS.

2.1. MRI sequences

MRI scans were performed on a 3T scanner (Siemens Healthineers, Erlangen, Germany) using a 20‐channel receive‐only head/neck coil. Anatomical images were collected using a T1‐weighted magnetization‐prepared rapid gradient‐echo (MPRAGE) sequence (voxel size = isotropic 1 mm³, inversion time (TI)/repetition time (TR)/echo time (TE) = 1.1 s/1.86 s/3.4 ms, flip angle = 10°, scan time = 5:02). fMRI data were collected using a custom simultaneous multislice (SMS) dual‐echo pseudo‐continuous arterial spin labeling (pcASL) sequence ⁴⁰ (voxel size = 3.8 × 3.8 × 5 mm³, slice gap = 1 mm, TR/TE1/TE2 = 4 s/13.4 ms/34 ms, pseudo‐continuous tagging period = 1.5 s, post delay after tagging = 2 s, SMS acceleration factor = 3, total 18 slices). Rs‐fMRI data were collected using a single‐shot gradient‐echo echo‐planar imaging (EPI) sequence (TR/TE = 2.8 s/29 ms, flip angle = 80°, voxel size = 2 × 2 × 4 mm³, 31 slices, 132 volumes, scan time = 7 min 36 s). qCBF were measured using pulsed tagging 3d gradient and spin‐echo (GRASE) readout ASL sequence with background suppression at the resting state (TR/TE = 4s/20 ms, FOV = 240 mm² voxel size = 3.8 × 3.8 × 4.5 mm³, TI/TI1 = 2 s/0.8 s, 10 pairs of control and tagged images, total scan time = 5:40). M0 image was acquired with the identical MR parameters of 3D GRASE ASL scan without the background suppression and inversion pulses.

2.2. Visual stimulus

During 2 SMS dual‐echo pcASL scans (See MRI sequences), a flashing checkerboard pattern alternating at 4 Hz was projected to the screen for 48 s (On period), followed by 48 s of black screen (Off period). Four On and three Off periods were repeated in an interleaved mode before and after 1‐min of pre‐and post‐Off periods. Each SMS dual‐echo pcASL scan acquires 57 volumes of control and tagging, resulting in a scan time of 7 min 36 s.

2.3. Hypercapnia challenge

Hypercapnia condition was achieved in participants by delivering medical gas with a 5% CO₂ mixture via a non‐diffusible Douglas bag and a nonrebreathing 3‐way valve. The first gas was delivered for 2 min after 2 min of room‐air breathing, followed by 3 min of room‐air breathing, a second 2 min of gas delivery, and another 3 min of room‐air breathing. An SMS dual‐echo pcASL scan including 90 volumes of control and tagging was acquired (total scan time of 12 min).

2.4. MRI analysis

Head motion was estimated in the time series of second echo images in the pcASL scan using the rigid volume motion assumption. Using the estimated volume motion of the second echo images, the first and second echo images of the pcASL scan were aligned to the first volume. To generate time‐series CBF weighted images, the first echo images with tagging were subtracted from the averaged adjacent first echo control images, ⁴¹ and the time‐series of the second echo images were used as BOLD contrast images. Finally, 3‐dimensional 8‐mm full‐width half‐maximum Gaussian spatial filtering was applied to the time series of CBF and BOLD weighted images.

The activated region of interest (ROI) was defined using uncorrected p < 0.01 on time‐series CBF images with voxel‐wise multiple regression to the VT paradigm, including a constant regressor (e.g., zero‐order detrending). The activated ROI was limited to the visual cortex region through visual checks and manual selection by the first author (fMRI scientist with over 15 years of experience). The hemodynamic response to the block‐designed paradigm was assessed on time‐series BOLD images with third‐order detrending regressors (3dDeconvolve in AFNI). The relative percent changes in CBF and BOLD during VT within the activated ROI were calculated and reported as ΔCBF_VT and ΔBOLD_VT. The relative percent changes in CBF and BOLD during hypercapnia were calculated within the activated ROI and reported as ΔCBF_CO2 and ΔBOLD_CO2 terms. Percent resting‐state fluctuation amplitude (RSFA) was calculated in the aligned activated ROI of the rs‐fMRI dataset. ⁴² qCBF values during the rest sate (qCBF₀) were calculated from 3d GRASE ASL scan in the activated ROI.

The relative percentage change in CMRO2 during VT (ΔCMRO_2VT) were calculated using Davis’ model, ⁴³ assuming that CMRO₂ is not changed during the hypercapnia condition;

$$\begin{array}{cc}& \mathrm{\Delta }CMRO{2}_{VT}={\left(1-\frac{\mathrm{\Delta }BOL{D}_{VT}}{M}\right)}^{1/\beta }\hfill \\ & {\left(1+\mathrm{\Delta }CB{F}_{VT}\right)}^{\left(\beta -\alpha \right)/\beta }-1\hfill \end{array}$$ (1)

where $M=\frac{\mathrm{\Delta }BOL{D}_{CO2}}{(1-{(1+\mathrm{\Delta }CB{F}_{CO2})}^{\alpha -\beta })}$

The constant α between CBF and cerebral blood volume and the constant β that related BOLD contrast change to the field strength were assumed to be 0.38 and 1.5 from literature reports. ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ The neuro‐vascular coupling ratio, n (= ΔCBF_VT /ΔCMRO2_VT) was calculated.

2.5. EEG acquisition and analysis

EEG was acquired using 64‐channel gold‐cup electrodes placed on the scalp via a standard 10‐10 EEG electrode montage system (Ives EEG Solutions, Newburyport, MA, USA). EEG electrodes were attached to the participant's scalp using a collodion adhesive. A vendor‐supplied EEG gel was used to reduce the electrode impedance to less than 20K ohm; the impedance threshold for safety is <50K. EEG data were acquired at a 5000‐Hz sampling frequency.

EEG was acquired with participants both inside and outside the MRI scanner during rest (eyes closed) and during VT described above. With the participant outside the MRI scanner, EEG was acquired in a dimly lit room. We monitored the compliance of the participant to task performance via visual assessment. Acquisition of EEG outside the scanner allowed us to (1) train the participant on the VT before the simultaneous EEG‐fMRI study and (2) compare the quality of EEG acquired inside the scanner with that acquired outside the scanner. With the participant inside the MRI scanner, we bundled the EEG leads and connectors together and placed them away from the participant's scalp to minimize imaging artifacts. The cables were routed to a battery‐powered MRI‐compatible Brainvision EEG amplifier (BrainProducts, Gilching, Germany) located at the rear of the MRI scanner bore. The helium compressor for the MRI scanner was switched off before EEG collection to minimize artifacts. The impedance of EEG leads was also checked and corrected when the participant was inside the scanner to mitigate any additional risk.

2.6. EEG data processing

The acquired EEG data were processed using the pipeline described in Figure 1. Template subtraction was used to remove MRI gradient artifacts (Figure 1B) and to down‐sample the EEG data to 1000 Hz. Cardioballistic artifacts were identified and removed from the EEG data ⁴⁷ (Figure 1C). Additionally, independent component analysis decomposition was performed to identify and remove eyeblink artifacts, vibration‐related artifacts, and residual scanner artifacts ⁴⁸ (Figure 1E).

FIGURE 1.

The processing pipeline for simultaneous EEG‐fMRI involving EEG recordings acquired inside and outside the MRI scanner. The pipeline includes multiple steps for artifact correction and spectral decomposition of EEG data to quantify electrophysiological correlates to blood oxygenation level–dependent (BOLD) signal changes. The final step involves estimating the percentage change in 4‐Hz signal power during the task and relating it to the BOLD signal change. ICA, independent component analysis.

2.7. Spectral analysis of EEG data

We segmented the EEG data collected during VT into 4‐s segments to match the temporal resolution (TR) of the ASL‐BOLD MRI acquisition. Our experimental design necessitated the development of novel signal processing paradigms to reduce signal artifacts and generate meaningful EEG features that can be correlated with BOLD studies. The development of a heuristic function to relate simultaneous EEG measurement and BOLD response has been previously discussed in studies by Rosa et al and Kilner et al. ⁴⁹ , ⁵⁰ Both of these studies used spectral decomposition of EEG to develop features that could be associated with BOLD changes. In this study, we used normalized EEG spectral power at the stimulation frequency of 4‐Hz frequency.

Spectral analysis was performed for EEG contacts sampling the occipital regions (O1, O2, Oz). The percentage change in normalized spectral power at 4 Hz at Oz from baseline to stimulation was determined as ΔEEG_VT. Additionally, ΔEEG_VT inside the scanner were compared with ΔEEG_VT outside the scanner across subjects to test the efficacy of MR artifact removal in EEG data analysis. It should be noted that ΔEEG_VT values inside and outside of the scanner are not expected to be identical or similar even within the same subject due to the additional MR noise components removal process in inside of scanner data (see Supporting Information).

2.8. Statistical analysis

We investigated the group difference of all measures (ΔBOLD_VT, ΔBOLD_CO2, ΔCBF_VT, ΔCBF_CO2, RSFA, qCBF₀, ΔEEG_VT) in each group using two sampled Student t‐test and conducted multicollinearity test on all MR and EEG measures. We tested the correlation between clinical and MRI/EEG measures within each group. Also, Student t‐test of each MRI/EEG measures was conducted between male and female groups. Using a step‐wise regression approach, the linear relationship between ΔBOLD_VT and other MR and EEG measures was tested in both groups, progressively including all combinations of predictors (i.e. regressors) except ΔCMRO2_VT because ΔCMRO2_VT is a calculated measure from other MR measures and is strongly correlated to ΔCBF_VT (see results). We report only statistically significant regressors and regression models (p < 0.05).

3. RESULTS

3.1. Clinical demographics

The EDSS scores in pwMS were measured with 3.75 ± 1.34 (mean/standard deviation) and an inter‐quartile range of 2.63–4.38. The age and sex of the two populations did not differ significantly (p > 0.05).

3.2. Group analyses

A significant correlation of ΔEEG_ACT in Oz was observed between inside and outside the MRI scanner (p = 0.0029 and 0.0267 in HC and pwMS, see Figure 3 and Table S1). The different ratio of ΔEEG_ACT values inside to outside of the scanner comes from the additional MR artifact removal process (see Figure S1).

FIGURE 3.

Plots of multicollinearity of MR and EEG measures. White and black circles indicate healthy controls (HC) and people with multiple sclerosis (pwMS) groups, respectively. The dashed and solid lines indicate the statistically significant linear trend (p < 0.05) in HC and pwMS.

We find the significant positive correlation between ΔEEG_VT and EDSS scores (p = 0.04). The scatter plot is displayed in Figure S2. The larger EDSS score indicates the severe disability. We do not observe any significant correlation between clinical and MRI measures. Any sex difference of EEG or MRI measure was not observed within a group.

PwMS demonstrated higher averaged ΔCBF_VT than HC (57.6 ± 20.8% vs 41.1 ± 10.5%, p < 0.05) and lower qCBF₀ in the activated ROI than HC (57.6 ± 20.8 vs. 41.1 ± 10.5 mL/100 g/min, p < 0.05, see Figure 2).

FIGURE 2.

Percentage change of EEG power and MRI measures during visual activation in healthy control (HC, white) and people with multiple sclerosis (pwMS, gray) groups. * Indicates the significant difference between groups (p < 0.05) between patients with multiple schelosis (pwMS) and healthy controls (HC). ΔBOLD/ΔCBF_VT/CO2: BOLD contrast/CBF change during visual activation/hypercapnia challenge (%); qCBF₀: quantitative CBF at the rest (mL/100/min); RSFA, resting state fluctuation of amplitude (%); ΔCMRO2_VT, cerebral metabolic rate of oxygen consumption change during visual activation (%); ΔEEG_VT, EEG power change during the visual activation (%).

Results from multicollinearity test is plotted in Figure 3, and Pearson coefficient correlation (CC) values are reported in Table S1. A significant positive correlation was seen between (1) ΔBOLD_VT and ΔBOLD_CO2, (2) ΔBOLD_VT and RSFA, and (3) ΔBOLD_CO2 and RSFA in HC group (all p < 0.05). A significant negative correlation between (1) ΔBOLD_VT and ΔCBF_VT (p < 0.01), and between (2) ΔBOLD_VT and ΔCBF_CO2 (p < 0.05) was observed in pwMS. A significant positive correlation between (1) ΔBOLD_CO2 and ΔCBF_CO2, and between (2) ΔCBF_VT and RSFA (both p < 0.05) was observed in pwMS. ΔCBF_VT and ΔCMRO2_VT show a significant linear correlation in both HC (p < 0.01) and pwMS (p < 0.0001). No significant difference was observed for the neuro‐vascular coupling ratio (n) between the groups (2.45 ± 0.96 and 2.41 ± 1.18 in HC and pwMS, respectively). ΔEEG_ACT is not correlated to any MR measure in either HC or pwMS.

3.3. Linear regression models

HC Group: We found that the best fitting (p < 0.01) predictive model of ΔBOLD_VT is ΔCBF_CO2 (Student t‐score, 2.59, p = 0.024) and ΔEEG_VT (Student t‐score = 2.64, p = 0.022).

pwMS Group: The best fitting linear regression model of ΔBOLD_VT was ΔCBF_VT (student t‐score = −4.02, p = 0.003). While the multi linear regression models with A) ΔCBF_VT and ΔCBF_CO2 (p = 0.0048) and B) ΔCBF_VT, ΔCBF_CO2 and qCBF₀ (p = 0.0045) were statistically significant, F‐tests from the nested regression model of ΔCBF_VT to multi‐regression models were not statistically significant (p > 0.05), indicating that ΔCBF_VT is single strong predictor to be correlated to ΔBOLD_VT in pwMS. The result of the tested multiple linear regression models with all significant predictors are listed in Table 1.

TABLE 1.

Step‐wise multi linear regression model tested with ΔBOLD_ACT as measured and other MRI values and EEG as predictors in health control (HC) and MS groups.

(A) HC group
	DOF	SS	MS	F	Significance F
Regression	2	1.9130	0.9565	4.0127	0.0463
Residual	12	2.8604	0.2384
Total	14	4.7735

	Coefficients	SE	t Stat	p‐value
Intercept	0.2320	0.3892	0.5962	0.5621
ΔCBFCO2	0.4042	0.1980	2.0419	0.0638
ΔEEGVT	0.0060	0.0039	1.5431	0.1488

(B) HC group
	DOF	SS	MS	F	Significance F
Regression	1	3.7070	3.7070	16.1989	0.0030
Residual	9	2.0596	0.2288
Total	10	5.7666

	Coefficients	SE	t Stat	p‐value
Intercept	3.8547	0.6086	6.3339	0.0001
ΔCBFACT	−0.0579	0.0144	−4.0248	0.0030

Statistically significant model is only reported (p < 0.05).

Abbreviations: DOF, degree of freedom; MS, mean SS; SE, standard errors; SS, summation of the squares; t Stat, student t‐score.

4. DISCUSSION

In this study, we find the significant neuro‐vascular coupling in both HC and pwMS groups. However, incorporating simultaneous EEG and fMRI, a multiple regression model demonstrated that ΔBOLD_VT is best predicted by ΔEEG_VT and ΔCBF_CO2 in HC, and only with ΔCBF_VT in pwMS. The finding in HC suggests that the BOLD response depends on the amount of underlying neuronal activity (ΔEEG_VT) and the change in blood flow to hypercapnia condition (ΔCBF_CO2) which is related to cerebrovascular reactivity. In contrast, BOLD response in pwMS can be predicted from change in blood flow during the visual activation alone, strongly suggesting that BOLD activation is not reflected by neuronal activity or cerebrovascular reactivity due to the different vasodilatory responses in pwMS.

In this study, a positive correlation was seen between ΔBOLD_VT and ΔBOLD_CO2 in HC participants. Previous studies have similarly shown that fMRI BOLD contrast change during a task is proportional to BOLD contrast change during hypercapnia, suggesting that assessment of these factors can be used in a “normalization” process to reduce intersubject and intrasubject variation. ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ However, pwMS group does not show a significant correlation between ΔBOLD_VT and ΔBOLD_CO2. We also observed a positive correlation between ΔBOLD_VT and RSFA in HC participants, again confirming the results of previous studies. ⁵¹ , ⁵⁶ , ⁵⁷ This result is not found in pwMS. A strong negative correlation was also seen between qCBF₀ and ΔCBF_VT in HC participants. This finding also agrees with previous research, ⁵⁸ which is also not observed in pwMS. Taken together, these results indicate altered CBF baseline and CBF reactivity to hypercapnia and neuronal activation in pwMS.

We observe a significant negative correlation between ΔBOLD_VT and ΔCBF_VT in pwMS (Figure 3). There are no other reports of the relationship between these in prior studies, so this constitutes the first observation of this in pwMS. The negative correlation may imply that the blood flow response to neuronal activation is either unrelated or poorly related to the underlying neuronal activation in pwMS. It is known that a large baseline blood flow (qCBF₀) will result in a low blood flow response to activation (ΔCBF_VT) due to a ceiling effect limiting the CBF increase, resulting in a low BOLD signal change (ΔBOLD_VT), which is observed in pwMS (see Figures 2 and 3). Inspection of Figure 3 and Table S1 support this, in that ΔEEG_VT appears to be unrelated to ΔCBF_VT for pwMS (See a grey colored subplot)_. Indeed, although the current study is not properly powered to measure this, the nature of the relationship between ΔEEG_VT and ΔCBF_VT is different for HC and pwMS.

This study presents the multi‐linear regression model of ΔBOLD_VT to other MR measures including ΔEEG_VT. We do not find a significant correlation between ΔEEG_VT and ΔCMRO2_VT in either HC or pwMS. CMRO₂ measure is derived by the complex non‐linear function forms of BOLD and CBF terms with various assumptions. ⁴³ , ⁵⁹ While α and β are assumed to be 0.34 and 1.5 in most of calibrated fMRI studies, the studies find that α is much lower than 1.5 ⁴⁴ , ⁶⁰ , ⁶¹ and depends on the regions. ⁶² , ⁶³ , ⁶⁴ Since β is an exponent contrast between CBF and CBV assuming with the vascular size and distribution, the different tissue composition could lead to the different value, for example, the deformed tissue due to the disease. For this reason, the multi‐linear regression model used here explores simple linear relationships between ΔBOLD_VT and other measures, which is not probing the details of the BOLD contrast mechanism. For example, ΔBOLD_VT is positively correlated to ΔEEG_VT and ΔCBF_CO2 in HC group (see Table 1), indicating that the indirect neural activation measure, ΔBOLD_VT is linearly predicted when the vascular reactivity (not CVR here) of ΔCBF_CO2 are considered in addition to the direct neuronal activation measure of ΔEEG_VT. These relationships are not observed in pwMS.

While we find a positive correlation between ΔEEG_VT and EDSS scores in pwMS, the correlation is mainly determined by a patient with EDSS score = 6 (see Figure S2). This finding is opposite to the previous studies. The decreased latency and reaction time of P3 event‐related potentials to EDSS scores was observed during the cognitive test. ⁶⁵ The positive correlation between P3 event‐related potentials to PASAT score was presented in pwMS. ⁶⁶ Our finding should be validated with a large sample in a future study.

This study was limited by the small number of subjects included in the analysis. For this reason, anatomical MRI‐derived information, for example, MRI disease burden, atrophy, lesion load etc was not considered within pwMS. Also, the location and burden of the focal lesion in pwMS, which can determine functional disability are not considered in this study. Cerebrovascular reactivity, another factor often assessed in such studies, is defined as the ratio of percent hypercapnic blood flow change to change in end‐tidal CO₂ ⁶⁷ was not accessible in this study due to the lack of ability to measure end‐tidal CO₂. In addition, it must be noted that EEG data collected inside the MRI scanner are prone to contamination from multiple sources, including artifacts related to MRI gradient, cardioballistic artifacts, and vibration‐related artifacts. Even with preventative measures and EEG postprocessing strategies to reduce these artifacts, the quality of EEG recorded inside the scanner is suboptimal when compared to the quality of EEG recorded outside. Additionally, small movements of the participant inside the scanner can induce large EEG potentials, leading to improper estimation of spectral power. These factors could account for the systematically lower EEG spectral measures recorded inside the scanner (see Supporting Information).

5. CONCLUSION

In this study incorporating concurrent fMRI and EEG measurements, we find statistically significant neurovascular coupling in both HC and pwMS. However, the coupling ratio is not significantly different between groups. We observe significant group differences of ΔCBF_VT and qCBF₀. It is observed that ΔBOLD_VT is positively correlated to ΔEEG_VT and ΔCBF_CO2 in HC and negatively to ΔCBF_VT in pwMS. All of these results imply very different vasodilatory behavior in response to task activation between pwMS and HC. These results also provide further evidence of impaired cerebral hemodynamics in pwMS, possibly leading to some of the widely observed abnormal BOLD activation in fMRI studies.

CONFLICT OF INTEREST STATEMENT

Daniel Ontaneda declares a conflict of interest: Research support from the National Institutes of Health, National Multiple Sclerosis Society, Patient Centered Outcomes Research Institute, Race to Erase MS Foundation, Genentech, Genzyme, and Novartis. Consulting fees from Biogen Idec, Bristol Myers Squibb, Genentech/Roche, Genzyme, Janssen, Novartis, and Merck. Other authors have no conflicts to disclose.

Supporting information

Supporting information

Supporting information

Shin W, Krishnan B, Nemani A, Ontaneda D, Lowe MJ. Investigation of neuro‐vascular reactivity on fMRI study during visual activation in people with multiple sclerosis using EEG and hypercapnia challenge. Med Phys. 2025;52:5081–5090. 10.1002/mp.17772