Multiple sclerosis-related white matter microstructural change alters the BOLD hemodynamic response

Abstract

Multiple sclerosis (MS) results in inflammatory damage to white matter microstructure. Prior research using blood-oxygen-level dependent (BOLD) imaging indicates MS-related alterations to brain function. What is currently unknown is the extent to which white matter microstructural damage influences BOLD signal in MS. Here we assessed changes in parameters of the BOLD hemodynamic response function (HRF) in patients with relapsing-remitting MS compared to healthy controls. We also used diffusion tensor imaging to assess whether MS-related changes to the BOLD-HRF were affected by changes in white matter microstructural integrity. Our results showed MS-related reductions in BOLD-HRF peak amplitude. These MS-related amplitude decreases were influenced by individual differences in white matter microstructural integrity. Other MS-related factors including altered reaction time, limited spatial extent of BOLD activity, elevated lesion burden, or lesion proximity to regions of interest were not mediators of group differences in BOLD-HRF amplitude. Results are discussed in terms of functional hyperemic mechanisms and implications for analysis of BOLD signal differences.

Introduction

Multiple sclerosis (MS) is characterized by inflammatory white matter damage within the central nervous system (e.g. demyelination, astrocytic hypertrophy). Although its pathogenesis remains to be discovered, MS is believed to result from an immune reaction in which lymphocytes, autoreactive to myelin basic protein and proteolipids, invade the central nervous system.^1–3 MS pathology is deleterious to myelin, astrocytes, and other white matter microstructure (WMMS) and can contribute to neuronal degeneration.^1,2,4 A central question of MS research has been how the pathology leads to changes in brain function.⁵ In the present study, we assessed how WMMS damage in MS affects blood-oxygen-level dependent (BOLD) signal changes.

Pioneering work assessing BOLD signal changes in MS has documented the heterogeneous nature of MS BOLD pathophysiology.^6–10 Some studies have shown increases in regional BOLD signal when comparing MS patients to healthy controls (HCs).^6,8–10 Other studies have shown decreases in BOLD signal when comparing MS patients to HCs.^7,9 Accounting for such MS-related BOLD signal changes by considering factors that affect neural activity, such as behavioral performance,¹¹ has revealed a similarly complex picture. While some fMRI studies have shown MS-HC BOLD signal differences in the presence of performance differences,^7,9 others have shown BOLD signal differences in the absence of performance differences.⁸ The results of studies comparing BOLD signal between MS patients and HCs have highlighted that such differences are indexing complex phenomena related to MS pathophysiology.^12–14

BOLD signal is contingent upon local changes in the concentration of deoxyhemoglobin.¹⁵ Increases in local neural activity lead to increases in cerebral metabolic rate of oxygen, resulting in subsequent increases in deoxyhemoglobin and decreases in BOLD signal. Functional hyperemia (i.e. increased perfusion of oxyhemoglobin in response to neural activity) results in increased concentrations of local oxyhemoglobin, which displaces deoxyhemoglobin and increases BOLD signal.^16–21 Under normal physiologic conditions, wherein functional hyperemia is sufficiently produced, local perfusion of oxyhemoglobin exceeds the metabolism of oxygen at a ratio of ≳2:1.^16–21 Under such conditions, BOLD signal reflects approximate linear changes in local neural activity.^22–27 This archetypal relationship between BOLD and neural activity is maintained via functional hyperemia and may be disrupted when there are aberrations in the functional hyperemic response.^12–14

MS features damage to specific WMMS (i.e. astrocytes) which, many models of neural-vascular coupling suggest, produce functional hyperemia.^28–33 Further, MS also features brain-wide aberrations in cerebral blood flow.^31,32,34 We wondered if these pathological changes in brain microstructure and physiology might result in alterations to the BOLD signal. To explore this question, we assessed the BOLD hemodynamic response function (BOLD-HRF) and its relation to WMMS integrity changes in patients with relapsing-remitting MS (RRMSPs).

The BOLD-HRF characterizes task-evoked BOLD changes over time in response to stimulation. BOLD-HRFs in healthy brains are well studied and assume a canonical shape.^35–37 Alterations to this canonical shape are thought to signify aberrations in the functional hyperemic response.^12,38 Thus, we assessed whether RRMS features BOLD-HRF changes compared to HCs.

We further used diffusion tensor imaging (DTI) to assess the extent to which RRMS-related alterations in anisotropic diffusion of molecular water, a marker of WMMS integrity, could mediate possible RRMS-HC BOLD-HRF differences. Changes to the integrity of WMMS have been shown to affect the BOLD signal in health and disease.^39–41 For example, van Eimeren and colleagues⁴⁰ assessed relationships between fractional anisotropy (FA; a measure of WMMS integrity) and BOLD signal. These authors showed that greater FA in specific white matter tracts was predictive of BOLD increases in grey matter regions of interest (ROIs) to their task (mathematical problem solving). These results showed that WMMS integrity could influence task-related BOLD signal. Thus, it might be that WMMS changes related to RRMS could also affect the BOLD-HRF.

In the present study, we assessed (1) RRMS-HC differences in the BOLD-HRF, (2) RRMS-HC differences in WMMS integrity, and (3) whether group differences in the BOLD-HRF were accounted for by changes in WMMS integrity. We used a simple button-press paradigm to evoke task-related BOLD-HRFs in motor and visual cortices.^38,42,43 We modeled key parameters of the BOLD-HRF within these ROIs and assessed whether groups differed on these measures. DTI was used to assess whether groups differed in skeleton-wide FA, and whether individual differences in FA could account for group differences in the BOLD-HRF. Further, we assessed whether our results might be accounted for by RRMS-related factors such as changes to spatial extent of activation, lesion burden, and lesion proximity to ROIs. Results are discussed in terms of WMMS components and functional hyperemic mechanisms. Considerations for the analysis of MS-HC BOLD signal differences are also discussed.

Materials and methods

Participants and procedure

Fifty-five participants were recruited for this study. Patients were recruited from the University of Texas Southwestern Medical Center Clinical Center for Multiple Sclerosis and had a confirmed diagnosis of RRMS by a neurologist. RRMSPs were required to be at least one month prior to their last exacerbation and one month prior to their last use of corticosteroids.⁴⁴ RRMSPs and HCs were both required to be free of MR contraindicators, and without significant medical or psychiatric disease. Participants had normal or corrected-to-normal vision.

Recruiting ensured that RRMSPs and HCs were similar in both mean age and sex (see Table 1). One RRMSP was excluded for reporting psychostimulant use before scanning and one HC was excluded for a history of possible seizures. Further, one patient and one HC were excluded for fMRI quality control issues (detailed below). Thus, our analyses were restricted to 51 participants (n_RRMS = 28; n_HC = 23). Patient characteristics are listed in Table 1. Expanded disability status scores (EDSS) were available for 25 patients and immunomodulatory therapy history was available for 27 patients (see Table 1).

Table 1.

Sample characteristics.

	HCs	MS	p-value
Demographics
Age	42.13 (2.20)	47.34 (2.00)	.085a
Sex (% Female)	73.91 %	82.14 %	.477b
Patient characteristics
Disease duration	–	153.19 (20.00)	–
EDSS	–	2.5 (0–6)	–
Immunomodulation therapy	–	78.57 %	–
Interferon Beta	–	39.29 %	–
Glatiramer Acetate	–	10.71 %	–
Natalizumab	–	35.71 %	–
Task performance
Button-press reaction time	374.37 (16.67)	386.26 (9.54)	.520c

Note: Average age in years (SEM); average disease duration in months (SEM); EDSS = Median Expanded Disability Status Scale score (Range); immunomodulation therapy = Percent reporting immunomodulation therapy during disease course. Average button-press reaction time in milliseconds (SEM).

^ap-Value derived from independent samples t-test with 49 DOFs (N = 28 RRMS, 23 HC).

^bp-Value derived from Pearson χ²-test with 1 DOF. ^cp-value derived from independent samples t-test with 43 DOFs (N = 25 RRMS, 20 HC).

Participants were financially compensated for their time. All participants gave their written consent before undergoing any procedure. Procedures were approved by the University of Texas at Dallas and University of Texas Southwestern Medical Center Institutional Review Boards. All procedures were carried out within APA guidelines. Approximate sample sizes were determined a priori using pilot study.

FMRI scanning parameters and task

Imaging data were collected at the University of Texas Southwestern Advanced Imaging Research Center using a Phillips 3Tesla MRI scanner (Philips Medical Systems, Best, The Netherlands) with an 8-channel SENSE head coil. High-resolution anatomical data were acquired using a T1-weighted MPRAGE pulse sequence. MPRAGE scans were acquired with the following parameters: 160 slices/volume, sagittal slice orientation, 12° flip angle, 256 × 204 matrix, 237 s scan duration. Functional data were acquired using gradient echo planar imaging with the following parameters: echo time (TE) = 30 ms, repetition time (TR) = 2000 ms, 39 slices/volume, 0 mm slice gap, transverse slice orientation, 3.43 × 3.43 × 4.00 mm³ voxel, 70° flip angle, 64 × 64 matrix, six dummy scans, 372 s scan duration.

Participants completed a simple button-press task during fMRI scanning. Participants were trained on this task before entering the MR environment. Experimenters first explained the task to the participants. Next, participants completed several practice trials of the button-press task that they would perform during imaging. The researchers ensured that the participants could complete several successful trials of the practice task, before entering the MR environment. Neither RRMSPs nor HCs reported difficulty completing this task.

The button-press task within the MR environment was kept at a fixed pace with long, jittered rest periods lasting 9, 10, or 11 s after stimulus presentation. Rest periods (i.e. interstimulus intervals) consisted of a fixation cross at the center of the participants’ visual field. The fixed-paced design with long rest periods was used to keep performance demands at a minimum to ensure group-equivalent information processing time and performance,³⁸ with the goal of minimizing the influence of possible group differences in reaction time (RT) on the BOLD-HRF.^45,46 In this event-related paradigm, participants were given two button-boxes and were instructed to simultaneously press and release both thumb-buttons as quickly as possible, whenever a checkerboard stimulus flashed on the screen. The checkerboard stimulus occupied the participants’ full visual field and flickered at a frequency of 6 Hz for 500 ms. There were a total of 20 trials.

Functional and MPRAGE image processing

Data were transformed from Philips PAR/REC format into AFNI (Analysis of Functional NeuroImages)⁴⁷ BRIK/HEAD format. Data were preprocessed using the align_epi_anat.py program in AFNI. The functional data were time-shifted to correct for interleaved slice acquisition. Motion was corrected using a rigid-body (6 degrees of freedom), least-squares transformation. This transformation reduced motion effects by co-registering all volumes to the first functional image volume, ensuring that any movements did not result in incongruent voxel matrices between separate volumes. The skull was removed from the MPRAGE image. Functional volumes were aligned to the MPRAGE image using local Pearson correlation cost functions with 45 degrees of freedom, a 45 mm search parameter, and a center of mass adjustment. Functional data were high-pass filtered (0.015625 Hz) which resulted in the removal of a large portion of the noise spectrum (<.008 Hz), increasing the signal-to-noise ratio of these data. Data were also spatially smoothed (full width at half maximum = 6 mm) to increase the signal-to-noise ratio. Noise was cleared from outside the head by eliminating voxels located outside of the anatomical brain region as well as any voxel for which there was significant functional signal loss. Each participant’s MPRAGE was warped to the Colin TTN27 template using the @auto_tlrc program in AFNI. Subsequently, each participant’s functional data were warped to his or her MPRAGE within Colin space using the @auto_tlrc program. Spatial normalization allowed for ROI demarcations using standard stereotaxic coordinates.⁴⁸ All functional and anatomical data were examined for artifacts and alignment issues. Two participants were excluded from all analyses because their functional data failed to align properly to the standardized brain template.

ROIs were drawn automatically in Colin space using AFNI. This method provided an unfilled, cortical map of bilateral Brodmann’s area 4⁴⁹ (precentral gyrus representing primary motor cortex; 403 voxels) and bilateral BA 17 (striate cortex representing primary visual cortex; 287 voxels; see Figure 1). These ROIs were chosen because primary motor cortex^38,42,43 and primary visual cortex^35,37,43 have shown reliable task-related BOLD changes in this and similar button-press paradigms.

Figure 1.

BA 4 (top) and BA 17 (bottom) masks.

Task-related excursions from baseline (measured during inter-trial interval) were modeled for the functional images using general linear modeling. Because we expected BOLD-HRF differences between groups, a “model-free” method⁵⁰ was utilized as opposed to standard probability distribution impulse response functions often used to model the BOLD-HRF (e.g. Poisson,⁵¹ gamma,³⁵ Gaussian⁵²). Fixed-shape impulse response functions assume that the hemodynamic response adheres to a canonical shape. For example, use of standard parameters of the gamma impulse response function makes the assumption that peak BOLD signal occurs at a fixed time point (i.e. approximately 4.70 s⁵³) after stimulus onset. Consequently, if the data do not meet this assumption, peak amplitude results derived from this impulse response function become biased estimators of BOLD signal.⁴³ Thus, if the functional hyperemic response is slowed differentially between groups that are being compared, the assumption of a canonical time-to-peak could lead to attenuation of the modeled BOLD signal for the slower group. Such attenuation would inflate the probability of false-positives for between-group comparisons. We therefore employed a model-free approach to convolve BOLD estimates; specifically, the data were convolved using piecewise linear B-spline functions in AFNI using the tent function.^53–55

The tent method employed here used a finite number of basis functions to allow for free-formed estimation of subject-specific BOLD-HRFs, without assumptions regarding the shape of individuals’ impulse responses. This function was calculated from baseline during a window of time centered on stimulus onset. The current method modeled each participant’s HRF using BOLD parameter estimates at eight time points, spaced equally at intervals of 2 s (1 TR). These parameter estimates represented relative signal amplitude beginning at stimulus onset (t₀) and extending 16 s (t₇) past the initial event.³⁷ These data were then converted to percent signal change from baseline.

BOLD data were taken from the average of all voxels in bilateral BA 4 and bilateral BA 17. Mean percent signal change was calculated for each participant across all voxels in each ROI. Percent of active voxels was also calculated for each ROI (Supplementary Formula 1). Piecewise B-spline functions were fit for each mean percent signal change time point. This procedure maximized the function’s fit to the data (all participants, BA 4 and BA 17 R²= 1) and resulted in smooth curves approximating each individual’s HRF, within the respective ROIs (see Figure 2).

Figure 2.

Group average (top) and individual participant (bottom) Spline-modeled BOLD-HRFs. HC = healthy control; RRMS = patients with relapsing-remitting Multiple Sclerosis.

Peak amplitude was derived to examine the magnitude of BOLD-HRF activity. This resulted in the maximum BOLD percent signal change value of each BOLD-HRF (see Figure 3). In healthy adults, this measure is thought to reflect the degree of local neural activity.³⁵ Time-to-peak (TTP) was calculated as the time from stimulus onset each BOLD-HRF took to reach its peak. TTP examined potential group differences in latencies of the BOLD hemodynamic response. Full-width at half maximum (FWHM) was calculated as the approximate width of each BOLD-HRF. FWHM is thought to reflect the duration of local neural activity in HCs.^45,46 Minimum amplitude was calculated as the lowest value of the BOLD-HRF and was used to examine possible baseline signal differences between groups.

Figure 3.

BOLD-HRF metrics. (a) Peak amplitude (green) is the maximum BOLD percent signal change value. (b) TTP (blue) is the length of time from stimulus onset that passed until peak amplitude was reached. (c) FWHM (gold) is the width of the curve at half the peak amplitude. (d) Minimum amplitude (red) is the lowest BOLD percent signal change value. Dashed line represents baseline.

DTI scanning parameters and processing

DTI data were acquired using a single-shot echo planar sequence with a SENSE parallel imaging scheme (reduction factor = 2.3). The imaging matrix was 112 × 112 with a field of view of 224 × 224 mm² (nominal resolution of 2 mm), which was zero-filled to 256 × 256. Axial slices of 2.2 mm thickness were acquired parallel to the anterior-posterior commissure line. A total of 65 slices covered the whole brain and brainstem without gap. TE and TR were 97 ms and 7.78 s, respectively, without cardiac gating. The diffusion weighting was encoded along 30 independent orientations⁵⁶ and the b value was 1000 s/mm². Imaging time for each sequence was 5 min and 15 s.

To increase the signal-to-noise ratio, two repetitions were performed. Automatic Image Registration⁵⁷ was performed on raw diffusion weighted images to correct distortion caused by eddy currents. Six elements of the 3 × 3 diffusion tensor were determined by multivariate least-square fitting of diffusion weighted images. The tensor was diagonalized to obtain three eigenvalues (${\lambda }_{1-3}$) and eigenvectors (${\nu }_{1-3}$). Tensor fitting was conducted using DTIStudio.⁵⁸

Average FA (Supplementary Formula 2; threshold = FA > .2) was computed within each participant’s white matter skeleton using tract-based spatial statistics⁵⁹ from FMRIB software library (http://www.fmrib.ox.ac.uk/fsl). This method compared the DTI-derived FA of each group at core or skeletons of the white matter to effectively alleviate partial volume effects. Further, this technique allowed for simultaneous assessment of aggregate WMMS integrity across all white matter tracts. The first step involved using the downsampled 2 × 2 × 2 mm³ JHU ICBM-DTI-81 template for nonlinear registration of FA maps. Each subject’s FA map was then linearly registered to the 1 × 1 × 1 mm³ JHU ICBM-DTI-81 template. Each subject’s FA map (instead of a one single subject map) was used as a template for nonlinear registration so that all subject’s FA data were transformed to JHU ICBM-DTI-81 space.⁶⁰ Skeletonization was performed on the data in the JHU ICBM-DTI-81 space to represent the voxels containing core white matter. Skeleton analysis was based on FA values at the skeleton voxels after TBSS registration, projection, and skeletonization steps.

T2-FLAIR scanning parameters, processing, lesion burden, and lesion proximity

To quantify lesion burden, one T2-fluid attenuated inversion recovery (FLAIR) image was acquired for each participant with the following parameters: TE = 125 ms, TR = 11000 ms, 33 slices, 0 mm slice gap, transverse slice orientation, 1.00 × 1.00 × 5.00 mm³ voxel, 120° refocusing angle, 352 × 212 matrix. T2-FLAIR data were available for 27 RRMSPs and 22 HCs.

After scanning, the skull was completely removed from each participant’s FLAIR image using a semi-automated procedure. Hyperintensities were then demarcated and defined as exhibiting FLAIR signal intensity greater than 2 SDs above the mean on each individual slice. These hyperintensities were then manually delineated as lesions by ruling out spurious voxels owing to fat signal, motion, ventricular edge effects, or coil sensitivity inhomogeneities.⁶¹ Hyperintensities with contact and protruding from a ventricle were qualified as periventricular lesions (PVLs). Hyperintensities not confluent with the margins of the ventricles or a PVL, were qualified as deep white matter lesions (DWMLs). Lesion burden was acquired by adding the number of voxels demarcated as PVLs and DWMLs (total white matter lesion burden; TWML). Inter-rater agreement of lesion burden was calculated using the Dice ratio of PVL, DWML, and TWML volume estimates by two independent raters (LO and JS; κ; see Supplementary Formula 3).⁶² κ > .70 indicates excellent inter-rater agreement.⁶³ For PVLs, DWMLs, and TWMLs, κ equaled .91, .86, .89, respectively.

Lesion proximity to motor and visual cortex was also determined. To determine lesion proximity spherical masks with 5, 10, and 15 mm radii were placed at the center of BA 4 and BA 17 on the Colin brain. These masks where then warped to individuals T2-FLAIR images using linear transformations with 6 degrees of freedom. We then qualified whether participants featured at least one lesion within 5, 10, or 15 mm of the ROIs.

Results

Button-press task performance

RT data were analyzed for 45 participants (n_RRMS = 25; n_HC = 20; data were lost for six participants due to equipment failure). RRMSPs and HCs did not significantly differ in RT on the button-press task (see Table 1).

Group BOLD-HRF comparisons

We utilized MANOVA modeling to test whether the components of the BOLD-HRF differed between groups. Homogeneity of variance in group distributions was examined using Levene’s test of equality of variance. Groups did not differ significantly in variance on any metric, in BA 4 or BA 17 (all ps > .05).

MANOVA of BA 4 and BA 17 peak amplitudes showed a significant effect of group, F(2, 47) = 4.39, p = .036, ${\eta }_{\text{p}}^2$ = .129. BA 4 and BA 17 tests of between-subjects effects were significant (see Figure 4; all ps < .05). MANOVA of FWHM in BA 4 and BA 17 did not show a significant effect of group, F(2, 47) = .118, p = .889, ${\eta }_{\text{p}}^2$ = .005 (see Table 2). MANOVA of TTP in BA 4 and BA 17 did not show a significant effect of group, F(2, 47) = 1.09, p = .345, ${\eta }_{\text{p}}^2$ = .043 (see Table 2). MANOVA of minima in BA 4 and BA 17 did not show a significant effect of group, F(2, 47) = .518, p = .599, ${\eta }_{\text{p}}^2$ = .021 (see Table 2).

Figure 4.

Group regional peak BOLD-HRF amplitude. Means and standard errors of group BOLD-HRF peak amplitudes. (a) Motor cortex (BA 4) and (b) visual cortex (BA 17). * = p < .05; ** = p < .01; HC = Healthy Control; RRMS = patients with relapsing-remitting Multiple Sclerosis.

Table 2.

Group descriptives on non-significant BOLD-HRF metrics.

	BA 4		BA 17
	HC	MSP	HC	MSP
FWHM	4.49 (.23)	4.45 (.17)	4.89 (.27)	5.07 (.31)
TTP	4.12 (.16)	4.36 (.11)	4.32 (.13)	4.27 (.15)
Minima	−0.09 (.02)	−0.08 (.02)	−0.06 (.02)	−0.07 (.02)

Note: Mean (SEM). FWHM = full-width at half maximum (seconds). TTP = time-to-peak (seconds). Minima (percent signal change). All independent samples t-test p-values >.05.

DTI group comparisons and association with peak amplitude

We sought to assess whether BOLD-HRF group differences (i.e. decreased peak amplitude of the BOLD-HRF) were influenced by possible changes to RRMSPs’ WMMS. First, we assessed whether RRMSP showed changes in FA. Levene’s test of equality of variance showed that the groups differed significantly in FA variance (p < .05). Thus, Welch’s t, robust to such differences, was used for group comparisons. Examination of the white matter skeletons of patients and HCs indicated that patients had lower FA compared to HCs (M_RRMS = .416 [SEM = .007] vs. M_HC = .432 [.004], d = −.58, t(41.89) = −2.07, p = .045).

We tested whether individual differences in FA could predict peak BOLD-HRF amplitude using ordinary least-squares regression. Shapiro-Wilk goodness-of-fit tests showed that the distribution of residuals from all regressions did not differ from a normal distribution (all ps > .05). Assessing this relationship by group, FA significantly predicted patients’ peak BOLD-HRF amplitude in BA 4 and BA 17 (all ps < .05; see Figure 5). No such relationships were found for HCs (all ps > .05).

Figure 5.

Peak amplitude and degree of white matter integrity in patients with RRMS. (a) Motor Cortex (BA 4), (b) Visual Cortex (BA 17). Regression lines interpolated using ordinary least-squares regression.

We further sought to assess whether group differences in BOLD-HRF peak amplitude remained when controlling for FA. We used ANCOVA modeling, predicting BOLD-HRF peak amplitude from group status controlling for FA. These results showed that RRMS-HC differences in BOLD-HRF peak amplitude were attenuated to non-significance when controlling for FA, BA4: Marginal M_RRMS = .342 [.028] vs. Marginal M_HC = .410 [.031], F(1, 48) = 2.55, p = .117, ${\eta }_{\text{p}}^2$ = .05; BA 17: Marginal M_RRMS = .416 [.026] vs. Marginal M_HC = .491 [.028], F(1, 48) = 3.68, p = .06, ${\eta }_{\text{p}}^2$ = .07.

Extent of spatial activity group comparisons

Because we averaged across voxels in motor and visual cortices, one alternative explanation for the decreased BOLD-HRF amplitude in patients could be grey matter loss and associated functional reorganization related to RRMS.^8,64–66 Thus, if RRMSPs showed restricted extent of activation compared to HCs, averaging across voxels in these ROIs might be expected to result in artificially attenuated BOLD-HRF peak amplitude. We tested for group differences in the percent of active voxels at each of the eight tent-derived time points (t₀ − t₇) in BA 4 and 17. We utilized independent samples t-tests for each ROI, assessing for a between-groups effect of percent of active voxels. A Benjamini-Hochberg (B-H) procedure⁶⁷ was used to correct for possible Type I error inflation associated with multiple comparisons. Using this procedure, RRMSPs and HCs did not show significantly different extent of active voxels in either BA 4 or BA 17 at any time point (all B-H adjusted ps > .05; Supplementary Figure 1).

Lesion burden, lesion proximity, EDSS and BOLD-HRF peak amplitude

Because some studies have suggested that white matter lesions can result in decreases in perfusion,⁶⁸ another alternative hypothesis for decreased peak amplitude of the BOLD-HRFs is white matter lesion burden. We first tested whether there were group differences in DWML, PWML, and TWML. Levene’s test of equality of variance showed that our groups differed significantly in DWML, PWML, and TWML variance (all p < .05). Thus, Welch’s t, robust to such differences, was used for group comparisons. RRMSPs had significantly greater lesion burden in all group comparisons (all ps > .05; Supplementary Figure 2).

We further tested whether individual differences in DWML, PWML, and TWML could predict peak BOLD-HRF amplitude for patients and HCs. A B-H procedure was used to correct for possible Type I error inflation associated with multiple comparisons. DWML, PWML, and TWML did not significantly predict RRMSPs’ or HCs peak BOLD-HRF amplitude in BA 4 or BA 17 (all B-H adjusted ps > .05; see Supplementary Table 1).

We further compared whether having at least one lesion within 5, 10, or 15 mm of an ROI could significantly predict peak amplitude of the BOLD-HRF. Because the probability of having a lesion in a random region of the brain is highly dependent on the amount of total lesions in the brain, we controlled for lesion burden. These analyses showed that lesion status within 5, 10, or 15 mm of BA 4 and 17 did not significantly predict peak amplitude of the BOLD-HRF in these regions for RRMSPs (all B-H adjusted ps > .05) or HCs (all B-H adjusted ps > .05).

Lastly, we assessed whether individual differences in RRMSPs’ level of disability (EDSS) could significantly predict peak amplitude of the BOLD-HRF in BA 4 and 17. EDSS did not significantly predict peak amplitude of the BOLD-HRF in BA 4 or 17 for RRMSPs (all B-H adjusted ps > .05).

Discussion

In the present study, we assessed RRMS-related changes in the BOLD-HRF and whether these changes were related to WMMS integrity. Results of the BOLD-HRF analyses showed that RRMSPs did not differ from HCs in their HRF full-width at half-maxima, minimum amplitudes, and times-to-peak, in both motor and visual cortices. However, RRMSPs showed significantly attenuated motor and visual cortex BOLD-HRF peak amplitudes compared to HCs. Further, we found that WMMS integrity significantly predicted RRMSPs’ peak amplitudes, where individuals with greater integrity (i.e. higher FA) had higher peak amplitudes.

We also observed that, when controlling for WMMS integrity, RRMS-HC group differences in BOLD-HRF peak amplitude were attenuated to non-significance. This result, coupled with significant RRMS-HC differences in WMMS integrity, and significant prediction of BOLD-HRF peak amplitude by WMMS integrity for RRMS patients suggest that WMMS integrity changes might mediate RRMS-HC differences in BOLD-HRF peak amplitude.⁶⁹ Because there were no observed group differences in RT or limited spatial extent of activation, it is unlikely that these variables affected RRMS-HC differences in BOLD-HRF peak amplitude.⁶⁹ Further, we failed to find a significant relationship between lesion burden or lesion proximity to motor/visual cortices and BOLD-HRF peak amplitude for either groups.

Two points should be noted regarding the lesion burden findings. First, we failed to find significant relationships between lesion burden and BOLD-HRF peak amplitudes at the group level using correction for multiple comparisons (i.e. B-H). However, these results were still not significant using uncorrected significance thresholds (see Supplementary Table 1). Second, for the RRMS patients, several relationships between the lesion burden metrics and BOLD-HRF peak amplitudes, although non-significant, did show small effect sizes (see Supplementary Table 1), suggesting that this effect should be investigated further. Lesions are, however, also a marker of WMMS integrity. For example, lesions are closely related to DTI signal changes (see Supplementary Material) and it is known that there is considerable microstructural damage in lesioned tissue.³³ Lesion burden measures are probably not as sensitive as FA in capturing the extent of RRMS-related WMMS damage, which could explain the smaller relationships between these measures and BOLD-HRF peak amplitude.

The present results showed an MS-related reduction in BOLD signal amplitude related to individual differences in RRMSPs’ WMMS integrity. MS-related damage to the biological substrates that facilitate functional hyperemia could provide an explanation for these results. One possible explanation for our results might be MS-related damage to a specific component of WMMS (i.e. astrocytes). Structural and functional aberrations in astrocytes are found in animal models of MS and in MS patients.^28–33 Astrocytes play a central role in many contemporary models of functional hyperemia.^70–74 Such models suggest that astrocytes, in both white (via fibrous astrocytes)⁷⁵ and grey matter (via protoplasmic astrocytes),⁷⁶ facilitate neural-vascular communication. This communication allows neurons to signal vasculature to perfuse oxygenated blood to the region of active neurons, displacing deoxyhemoglobin, producing positive changes in BOLD signal. Further, in vivo astrocytic changes such as those incurred by patients with MS (i.e. astrocytic hypertrophy)³⁰ also result in decreased anisotropic diffusion of molecular water (analogous to decreases in FA), irrespective of damage to myelin.⁷⁷ Thus, it might be that MS-related changes to astrocytes, and subsequent pathophysiological functional hyperemia deficits, produce BOLD changes such as those observed here.^31,32

Our present findings showed RRMS-related decreases in BOLD signal amplitude in motor and visual cortices. MS-related alterations in functional hyperemia might not only relate to BOLD amplitude decreases. Local neural activity results in increases in local deoxyhemoglobin concentrations, increases in the rate of T₂* decay, and decreases in BOLD signal. Functional hyperemia results in disproportionate local increases in oxyhemoglobin which acts to displace deoxyhemoglobin concentrations, resulting in positive changes in BOLD signal amplitude.^16–21 Disruption of communication between neurons and vasculature, hindering functional hyperemia, could result in an inverted relationship between neural activity and BOLD signal. In this scenario, relatively little local neural activity (with low local concentrations of deoxyhemoglobin and flat, baseline levels of cerebral blood flow) could result in positive BOLD signal. Similarly, task-evoked neural activity (with high local concentrations of deoxyhemoglobin and flat, baseline levels of cerebral blood flow) could result in negative BOLD signal. Because modulation of cerebral hemodynamics can vary regionally within MS,⁷⁸ BOLD increases and decreases relative to healthy controls might also show regional heterogeneity. Such regional heterogeneity would be consistent with the heterogeneous findings of MS-related BOLD increases and decreases reported in extant literature.^6–10 Advanced neuroimaging methods, such as calibrated fMRI, coupled with diverse task paradigms to evoke neural activity in different ROIs, could be employed to test this hypothesis.^{16–19,21,79}

MS-related alterations to functional hyperemia could also influence measurement of the temporal associations used to estimate functional connectivity at rest and during tasks.^51,80,81 MS involves changes in task and resting-state functional connectivity relative to healthy controls.^51,82–85 Functional hyperemic deficits and attenuated BOLD signal would decrease the magnitude of the relationship between the affected brain region(s) and all other regions. For instance, if the magnitude of the BOLD hemodynamic response was attenuated in one region for MS patients, this could artificially attenuate the strength of linear dependencies between this region and all other brain regions.⁸⁶ Comparing a group with such artificially attenuated linear dependencies (e.g. RRMSPs) to a group with a greater hemodynamic range (e.g. HCs) could yield spurious results.⁵¹ Investigation of relationships between MS-related functional hyperemic attenuation and functional connectivity certainly warrants future research.¹⁴

One way to minimize the influence of MS-related pathophysiological factors on BOLD signal is statistical control procedures. Prior fMRI work assessing MS-HC BOLD signal differences has relied upon common BOLD signal estimates across groups, without subject- or group-specific scaling factors that could account for non-neural activity factors influencing BOLD signal. As we have demonstrated here, group differences in WMMS integrity can be controlled statistically using multivariate methods (e.g. ANCOVAs, multiple regressions, partial correlations). These methods can assess whether WMMS damage acts as a possible mediator between BOLD signal and group status. Such methods are easily applied and can be utilized to remove the variance accounted for by WMMS changes and derive group estimates of the BOLD signal independent of such damage. Further, possible MS cerebrovascular changes could also be controlled using BOLD-calibration techniques (e.g. breath hold and hypercapnic challenges, resting state fluctuation of amplitude).^80,81

Alterations to functional hyperemia have also been posited to result in cognitive dysfunction in MS.^31,32 Indeed, decreases in functional hyperemia not only affects the BOLD signal, but results in cognitive slowing,^18,19 a core neuropsychological deficit in RRMS.⁸⁷ Thus, future work should assess whether attenuated functional hyperemia can predict cognitive changes in RRMS.

In conclusion, we have shown that individual differences in WMMS accounted for RRMS-HC group differences in the amplitude of task-related BOLD signal. Decreased neural-vascular communication and decreased functional hyperemia provide one explanation for these findings. The present research suggests that WMMS damage should be accounted for in future research assessing BOLD differences between RRMSP and HC populations.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Friends of Brain Health and the Linda and Joel Roebuck Distinguished New Scientist endowments (to NAH), the National Multiple Sclerosis Society (RG4453A1/2 to BR and EF), and the National Institutes of Health (1R01AG047972 and 1R01AG029523 to BR).

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: NAH, MT, JLH, AO, JS, LO, SS, SD, RB, HH, JH, BBB, and BR are not aware of any potential financial conflicts of interest related to the current study. GR, TF, and EF have received honoraria from speaking engagements with pharmaceutical companies related to multiple sclerosis, but unrelated to the current study.

Authors’ contributions

NAH contributed to all aspects of data analysis, and wrote the manuscript. MT and JLH contributed to manuscript writing and data analysis. AO and JS contributed to data analysis and processing. LO, SS, RB contributed to data analysis and data collection. GR contributed to data collection and manuscript writing. SD, HH, JH, TF, EF, BBB, and BR contributed to the study design and manuscript writing. EF, BBB, BR conceived the ideas for the study and study design.

Supplementary material

Supplementary material for this paper can be found at http://jcbfm.sagepub.com/content/by/supplemental-data