Improvement of the thalamocortical white matter network in people with stable treated relapsing – remitting multiple sclerosis over time

Advanced imaging techniques (tractography) enable the mapping of white matter (WM) pathways and the understanding of brain connectivity patterns. We combined tractography with a network-based approach to examine WM microstructure on a network level in people with relapsing – remitting multiple sclerosis (pw-RRMS) and healthy controls (HCs) over 2 years. Seventy-six pw-RRMS matched with 43 HCs underwent clinical assessments and 3T MRI scans at baseline (BL) and 2-year follow-up (2-YFU). Probabilistic tractography was performed, accounting for the effect of lesions, producing connectomes of 25 million streamlines. Network differences in fibre density across pw-RRMS and HCs at BL and 2-YFU


| INTRODUCTION
5][6] Tractography is one such advanced imaging technique that can detect alterations in the organisation and density of white matter (WM). 2,7Tractography enables the detection of WM organisation and density alterations and has been applied to pathological cohorts, including MS, to explore markers of neurodegeneration. 7,8remarkable advantage of tractography is that even in the presence of lesions, the method does not systematically result in erroneous underestimations of WM integrity. 2Deconvolution methods such as Orientation Distribution Functions (ODFs) 9 offer compelling investigation tools and could provide diagnostic and prognostic markers for MS.However, mass univariate statistical testing of tractography leads to sizable false positives and does not account for noise inherent in the data. 102][13] Network-based statistics (NBS) overcomes these issues by identifying connected components of a graph (e.g., a connectome) above a predefined threshold. 13,14NBS offers greater statistical power through cluster enhancement, a method of clustering p values in a certain neighbourhood. 15e advantage of using a network-based method to analyse WM microstructure in MS is that it (1) overcomes the impact that lesions have on local microstructure, which can override any weaker but equally important effects; and (2) allows multiscale (local, regional and global) analysis, reflecting the heterogeneous nature of MS pathophysiology.This study examines a cohort of clinically stable treated people with relapsingremitting MS (pw-RRMS) over 2 years to see a difference in the WM microstructure on a network level between this cohort and matched healthy controls (HCs).By quantifying the network differences at the two timepoints, we can observe any change in WM microstructure throughout treatment and evaluate any association with clinical parameters.We will perform NBS on the baseline (BL) and 2-year follow-up (2-YFU) WM connectomes to compare: (1) differences between HCs and pw-RRMS connectomes at BL; (2) differences in connectomes of HCs and pw-RRMS at 2-YFU; and (3) differences in connectomes of pw-RRMS over time.If a network is identified as changing over time, its association with clinical parameters will be investigated.Applying network neuroscience and tractography in pw-RRMS, this study potentially provides lesion-agnostic and region-specific biomarkers for prognosis and understanding of MS pathophysiology.The selection of HCs was based on (1) no prior/current psychiatric or neurological disorders; and (2) not currently on any medications.All participants completed an MRI safety checklist and all the study components (including imaging and neuropsychological evaluations).Approval to conduct this study was obtained from the human research ethics committee of the local healthcare district (HREC no's: 14/09/10/3.01 and 14/09/10/3.02).

| Clinical assessments
The clinical assessment for both cohorts (pw-RRMS and HCs) has been outlined in a previous study. 16Neurological tests were performed, including the audio-recorded cognitive screen (ARCS), 17,18 which measures executive functioning/attention, memory, language, verbal fluency and visuospatial functioning.A symbol digit modalities test (SDMT) 19 was performed, measuring information processing, speed and attention.Fatigue was assessed with the modified fatigue impact scale (MFIS). 20All clinical assessments were conducted on the same day as the imaging session.

| MRI data acquisition
Data were acquired from a 3-T Siemens Prisma (Siemens, Erlangen, Germany) with a 64-channel radiofrequency head coil.T1-weighted magnetisation-prepared rapid gradient-echo (MP-RAGE) was acquired for anatomical referencing with the following parameters: repetition time (TR), 2000 ms; echo time (TE), 3.5 ms; inversion time (TI), 1100 ms; and a flip angle of 7 degrees.The field of view (FOV) was 256 mm 2 with a voxel size of 1 mm 3 and an acquisition time of 5 min.
For the assessment of lesions, a 3D T2 fluid-attenuated inversion recovery (FLAIR) sequence was acquired with TR/TE/ TI = 5000/386/1800 ms, and a flip angle of 120 degrees.The FOV was 256 mm 2 with a voxel size of 1 mm 3 and an acquisition time of 6 min.
Axial diffusion-weighted imaging (DWI) was obtained using a fat-suppressed single-shot echo-planar imaging (EPI) sequence.Bipolar gradient pulses with a duration of effective δ = 17.9 ms and effective diffusion time Δ = 31.9ms were applied in 64 equally spherically distributed directions; TR/TE: 9400/69 ms; in 70 slices; slice thickness: 2 mm; no slice gap; FOV: 240 mm 2 , with a voxel size of 2 mm 3 ; and b-value = 3000 s/ mm 2 .In addition, three EP images were acquired without gradient (b-value = 0 s/mm 2 ) in anterior-posterior (AP) phase encoding and posterioranterior (PA) phase encoding for distortion correction.The total acquisition time was 10 min.
To account for the effects of lesions on fibre orientation distributions (FODs), we estimated fibre density using a single-shell three-tissue (SS3T) technique as implemented in MRtrix3Tissue v. 3.0.1. 30The SS3T method is more robust to lesion or signal dropout by iterating over all FODs using an unsupervised algorithm of tissue type, losing less 'WM-like' FODs to low contrast-to-noise (such as in lesions), demonstrating validity of approach in WM-hyperintensities in stroke lesions. 31This method has also been shown to be very reliable in test-retest and has longterm stability. 32Twenty-five million streamlines were generated for each individual using the iFOD2 algorithm. 9We used a well-known standard atlas, the Automated Anatomical Labelling (AAL) v. 3 with 166 parcels 33 (nonlinearly registered from MNI152 1-mm space to participant diffusion space using ANTs 29 ) to divide the cortex into separate regions for network-based analysis.We normalised the entire cohort (HCs and pw-RRMS) at both timepoints to the maximum FOD lobe integral of the entire cohort using mtnormalize. 34To remove spurious tracts (false positives), we used the tcksift2 algorithm, which also resulted in quantitative proportionality coefficients for each streamline. 35For consistency with previous literature and current nomenclature in tractography studies, 35,36 hereafter, we refer to the proportionality coefficient of a streamline as 'fibre density'.The assortation of streamlines to regions of the AAL atlas was determined using tck2connectome with a search parameter of 2 mm, resulting in dense connectomes for each participant.To further remove spurious fibres, we collated every connectome (across cohorts and time) to find the variance between each connection (node i to node j) and thresholded at 10% density of these values.As such, we were left with the streamlines that explained the most variance across all streamline lengths within all the connectomes, while leaving the connectomes fully connected. 37This was desirable, as we did not want isolated networks, making comparing individuals difficult (as any two individuals may not have the same isolated networks).8][39] In addition, we applied the main analyses at 15% and 20% thresholds (see Figure 1J,K) to examine whether any results are dependent on the threshold level chosen (Figure S1).

| Network-based statistics
Network-based statistics 13 in MATLAB (2021a, v. 9.11) was used to determine a network of connections that distinguishes the groups (all pw-RRMS against HCs).NBS performs better at Type I and II errors than conventional multiple comparison corrections 13 and works by identifying clusters of connections (i.e., components) within a network.First, mass univariate testing of the network is performed, giving a t-statistic of the contrast that is being tested.Then an a priori threshold (t-statistic) of 2.6 was set, and the connections with t-statistics above were considered connected components.These components were given a score based on their weighted (by fibre density) sum of edges, resulting in a weighted network related to a particular contrast.Finally, family-wise error (FWE)-corrected p values and probabilities were calculated based on shuffling the labels for each datapoint (the matrices in each contrast, see Figure 1) 5000 times to create a null distribution.If the t-statistic of the null distribution was lower than the t-statistic in the empirical network, this was counted as a success for the empirical network.The p value was calculated by 1 minus how many times the real network was successful (over the shuffled network) over the number of permutations (e.g.p = 1 -[4750 successes/5000 permutations] = 0.05).
We hypothesised that pw-RRMS have an altered network that can be determined using NBS, based on lower fibre densities in these connections compared with HCs, and we also hypothesised that there might be a fibre density difference between the two timepoints within pw-RRMS.
F I G U R E 1 Analysis and preprocessing pipeline.(A) Diffusion-weighted imaging preprocessed using an in-house pipeline.Sixty-four directions at b = 3000 s/m 2 , three directions at b = 0 for susceptibility and eddy current distortion correction; (B) T1-weighted images preprocessed using freesurfer 7.0 and FSL 6.0.1;(C) T2-weighted Fluid-Attenuated Inversion Recovery (FLAIR) for lesion segmentation.T2/FLAIR images were acquired in RRMS participants only to calculate lesion burden and segmentation.(D) After performing single-shell multitissue fibre orientation deconvolution (FOD) as implemented in MRtrix3Tissue, we normalised the entire cohort (HCs and pw-RRMS) by the maximum FOD lobe integral to ensure that the data were comparable between groups.We also normalised the cohorts over time to ensure that the data were comparable longitudinally.(E) We selected a well-known and well-utilised anatomical parcellation, Automated Anatomical Labelling Atlas version 3 (AAL v. 3), to define where our streamlines started and ended when we collated the tractograms into a connectome.(F) The lesion segmentation tool (LST) was utilised in MATLAB 2021a (v.9.11) to calculate lesion volume to regress out of the RRMS analyses.(G) An appropriate cross-sectional area multiplier was applied to each individual's tractograms using the tcksift2 algorithm implemented in MRtrix3Tissue.This resulted in a fibre proportionality coefficient of each connection, which was summed in (H) For each individual, resulting in a connectome.(I) For thresholding by variance across the entire cohort, we collated the connectomes (HCs BL + RRMS BL + HCs 2-YFU + RRMS 2-YFU ) into one 3D matrix, consisting of dimensions (N, 166, 166), where N is the total number of participants X 2 (240).(J) For the main NBS analyses, the variance threshold was 10% (see Results), while the NBS analysis was repeated at 15% and 20% variance thresholds (see Figure S1).(K) NBS was run on baseline (HCs BL vs. RRMS BL ), 2-year follow-up (HCs 2-YFU vs. RRMS 2-YFU ) and longitudinal contrasts (RRMS 2-YFU vs. RRMS BL ) with 5000 permutations in each contrast.*Data shown in C and F are from one RRMS participant, while data shown in the other panels are from one HC participant.HC, healthy control; NBS, network-based statistics; pw-RRMS, people with relapsing-remitting multiple sclerosis.
We present (1) BL differences between pw-RRMS and HCs (t-test: fibre density of HCs at BL vs. fibre density of pw-RRMS at BL); (2) 2-YFU differences between pw-RRMS and HCs (t-test: fibre density of HCs at 2-YFU vs. fibre density of pw-RRMS at 2-YFU); and (3) differences between timepoints of pw-RRMS (t-test: fibre density of pw-RRMS at 2-YFU vs. fibre density of pw-RRMS at BL).All analyses were corrected for age and sex effects.See Figure 1 for the complete pipeline.

| Statistical analyses
Baseline changes in clinical parameters were quantified by Student's t-test of unequal sample sizes and corrected for multiple comparisons using Benjamini-Hochberg correction for false detection rates. 12 explore any possible relationship within the NBS analysis to clinical outcomes, we conducted post hoc Pearson correlations and least squares fitting of trendlines of the network fibre density to the clinical measures.We did not report test-retest or familywise error correction on these post hoc correlations, as these were only exploratory.

| Demographic, clinical and cognitive assessments
Overall, age and sex showed no statistically significant difference between HCs and pw-RRMS groups ( p > 0.05).Compared with the other two treatment groups, the DMF cohort was younger on average ( p ≤ 0.05) and had recently started treatment, while the other two groups were stable on treatment.Pw-RRMS showed higher mean fatigue scores ( p ≤ 0.01) compared with HCs; no statistical difference was found between treatment groups.The gap between HCs and pw-RRMS widened at the follow-up, showing more statistically significant differences in most scores.In addition, there were no statistically significant differences in HCs in clinical parameters over time.Detailed demographic characteristics and clinical details of both groups and pw-RRMS subgroups at two timepoints are shown in Table 1.There were no radiological differences over time, that is, pw-RRMS were radiologically stable.

| Baseline network analysis
When testing whether HCs had a network significantly higher in fibre density than pw-RRMS at BL, we detected a widespread network in the frontal, parietal and temporal regions (supplementary motor area, precuneus, insula, temporal gyrus), as well as subcortical regions (caudate, thalamus, basal ganglia).This network was significantly weaker in pw-RRMS than in HCs ( p = 0.03 ± 0.01, Figure 2).The fibre density in this network did not correlate with lesion volume in pw-RRMS.

| Two-year follow-up network analysis
When testing whether HCs had a network significantly higher in fibre density than pw-RRMS at 2-YFU, we detected a widespread network in the frontal, parietal and temporal regions (medial frontal gyrus, lateral frontal gyrus, supplementary motor area, precuneus, insula, temporal gyrus, temporal pole), as well as subcortical regions (caudate, thalamus, basal ganglia) and parts of the cerebellum.This network appears to have more connections around the midbrain when compared with the BL network differences, although the significance of the network during bootstrapping was the same as the BL contrast ( p = 0.03 ± 0.01, Figure 3).The fibre density in this network did not correlate with lesion volume in pw-RRMS.

| Longitudinal network analysis
To examine whether there were any changes over time within pw-RRMS, we used NBS on pw-RRMS (BL and 2-YFU, testing whether network fibre density at 2-YFU is greater than at BL).We found a network with significantly higher fibre density at 2-YFU ( p = 0.04 ± 0.01, Figure 4).middle occipital gyrus (MOG) (see Figure S2).By far, the left and right mPUL had the highest degree in this network (number of connections going to and from the node), serving as hubs for the network.We also extracted the proportionality coefficient of these connections at BL and at 2-YFU for post hoc correlations (Table S1).
To validate our approach, we repeated the longitudinal NBS contrast in HCs (i.e., HCs 2-YFU vs. HCs BL ), showing that while there was an increase in fibre density in several small-scale networks, this effect is likely due to noise ( p = 0.05 ± 0.01; see Figure S3; there was no significant network with higher fibre density at BL in HCs, that is, in the contrast HCs BL vs. HCs 2-YFU ).There was no significant network in pw-RRMS that was greater than HCs over time.

| Correlation of longitudinal network with clinical and behavioural parameters
To determine if there is a relationship between this longitudinal network (the significant network in the contrast RRMS 2-YFU vs. RRMS BL ) and clinical assessments, we explored correlations of fibre density in this network with total MFIS score, tARCS score (as well as the attention, memory and language parts of this score) and EDSS at 2-YFU.We performed Pearson rank correlation at both timepoints to see if there is an effect pretreatment and post-treatment, reporting uncorrected p values for exploratory analysis.There were no correlations of any of these measures, except for a significant (although very low) positive correlation of attention with fibre density at BL (r = 0.237, p = 0.044 uncorrected) (Figure 5).
In other words, higher fibre density at BL weakly correlated with higher attention scores.This correlation disappeared at 2-YFU (p ≥ 0.3).There were also no correlations between the fibre density in this network with lesion volume or disease duration.In addition, there was no correlation between the fibre density in this network with handedness, despite the lateralised appearance of the network.After familywise error correction, all correlations were no longer significant.Using advanced neuroimaging techniques, we identified a network that differentiated pw-RRMS from HCs, showing widespread WM differences even after accounting for the effect of lesions.These differences appear in networks responsible for cognition, 40 motor function, 41 somatosensory function, 42 balance, 43 emotion processing and regulation, [44][45][46] and memory, 47 which overlap with the symptom profile of MS (i.e., that there are disruptions to all of these processes in the brain 48 ).This is consistent with previous cross-sectional studies showing extensive differences in fibre density between HCs and pw-RRMS, 3,49,50 as well as the more general understanding of the pathology of MS, which is that of neuroinflammatory and deleterious auto-immune attacks resulting in extensive and diffuse WM demyelination and lesioning. 51Interestingly, these network differences presented regardless of lesion burden and disease duration in our participants, demonstrating a fundamental reduction in both normal-appearing white matter (NAWM) and lesioned fibres in RRMS.This result supports previous studies that show global alterations in measures of fibre integrity in both NAWM and white matter lesions (WML), [52][53][54][55] and shows a region-level and longitudinal specificity unique to this study.
A subset of this network was detected as improved over time in pw-RRMS, showing strong evidence that it could be a sign of remyelination.
NBS is robust to noise inherent to probabilistic tractography, 13,14 while thresholding by variance accounts for Type I errors. 37Our results were also robust across different thresholds (see Figure S1).The course of the disease is understood to be that remyelination often occurs in the early stages, but is not long-lasting. 51However, our finding shows that in our cohort of stable and treated RRMS, regardless of disease duration, we found a network that is improved over time.This appears although pw-RRMS have more connections with reduced fibre density overall at 2-YFU compared with HCs over the same timeframe (see Section 3.3).The improved fibre density in this network probably occurs in tandem with reductions in fibre density overall in stable RRMS, reflecting regional neuroplasticity within global neurodegeneration.This could be related to the heterogenous topographical evolution of neurodegeneration in MS, 56 although this improved network appeared regardless of disease duration and lesion burden.[59][60] F I G U R E 3 NBS results of HCs versus pw-RRMS at 2-year follow-up.Nodes and connections that contribute to greatest difference between pw-RRMS and HCs at 2-year follow-up (projected in BrainNet viewer) after 5000 permutations (p = 0.03 ± 0.01).The medial pulvinar nucleus of the thalamus in both hemispheres formed the hubs for this network (i.e., they had the greatest number of connections).The mPUL has efferent projections to many parts of the association cortex, including the cingulate and posterior parietal, premotor and prefrontal cortical areas. 61The thalamus is heavily implicated in MS pathology, [62][63][64] while atrophy of the thalamus is common, 65 and has been demonstrated to help predict conversion to MS and future disability accumulation. 65,66The thalamus has high levels of connectivity with the cortex, often characterised as a hub of communication within the brain. 67,68Furthermore, the pulvinar nucleus receives inputs predominantly from the cortex, 69,70 and has been proposed, among other higher order regions of the thalamus, to mediate corticocortical information transfer within the cortex. 70In the larger context of the entire network, the thalamus would seem to serve in a multisensory integrative role, with connections from visual centres, 71 executive regions, 47 language and auditory processing centres, 72 as well as motor and premotor cortices. 73If these connections are enhanced over time, this may transfer to improvements in information processing and fatigue.Motivated by this finding, we evaluated exploratory correlations of this network to clinical and cognitive assessments that covered these domains.Damage to WM tracts innervating and enervating from the thalamus and association cortex have been implicated in symptoms of MS, namely, fatigue, 74 lack of attention 75 and cognitive dysfunction. 75,76Our post hoc correlations showed a small but significant correlation of average fibre density in this network at BL with attention.These findings should be tempered by the caveat that the correlation was weak and did not survive familywise error correction.This correlation disappeared at the follow-up, coinciding with a decline in attention scores (although insignificant), suggesting that the timescale of remyelination could be different from the timescale of cognitive remediation associated with that remyelination.In other words, while a decline in these scores may have appeared at 2-YFU, remediation of these scores may occur at longer timepoints.This is consistent with studies that show no immediate or mediumterm effects of DMTs on cognitive dysfunction. 77We assert that such effects could arise over longer periods (such as 5 years).To fully capture this effect, any longitudinal study should consider a wider scope of timepoints, and include a wide range of functioning and ability in their cohort.
Whether or not this remyelination in our participants results from direct exposure to DMTs is uncertain.DMTs are not believed to directly lead to remyelination; instead, they appear to facilitate neuroprotective processes by reducing neuroinflammation and the relapse rate, 78 although the mechanism regarding how this takes place is less well understood.Glatiramer acetate has been shown in MS to promote neuroprotection and repair of myelin. 79The oligodendroglia pathway of interferon-β has been implicated in remyelination, 80 although this seems to inhibit differentiation into mature oligodendrocytes.Interferon-β also has direct neurotrophic effects in the neural precursor cell (NPC), producing neurotrophin and neurotensin/NTSR1, and may facilitate an indirect effect in the T-cells, contributing to the survival of NPCs. 81These NPCs can differentiate into all types of neural cells, which could contribute to the repair of myelin and lesions.Fingolimod reduces aggressive lymphocyte infiltration into the central nervous system, offering neuroprotection and facilitating the restoration of damaged myelin, 82 while its effects on reducing the relapse rate in RRMS are well known. 83DMF acts along the nuclear factor erythroid 2-related factor 2 transcriptional pathway, reducing oxidative stress and increasing anti-inflammatory action and neuroprotection. 84We assert that the longitudinal effect across all our RRMS participants is attributable to all the aforementioned factors, and this effect is independent of lesion burden in these tracts.Therefore, there seems to be a restorative effect of these DMTs in NAWM and within WML, although this could reflect spontaneous protective effects in this relatively stable cohort independent of DMTs.This probably arises from the neuroprotective factors and anti-inflammatory properties that inhibit auto-immune aggression throughout the brain.
Our large longitudinal cohort and the removal of potential confounders go a long way to showing the robustness and generalisability of this finding.Moreover, the network is functionally distinct, with each region serving diverse and specialised functions: the thalamus acts as a way station for communication and rewards circuits 69 ; the precuneus serves as an executive function region 47 ; the calcarine cortex is dedicated to visual processing 71 ; the superior temporal gyrus is involved in speech processing and multisensory integration 72,[85][86][87] ; among others.The network is also cytoarchitecturally and regionally distinct, involving some of the earliest 61,67 and latest developing, 47,88 and among the longest axonal projections in the brain. 89Further investigation is needed to reinforce these network findings, to understand its interplay with treatment type, and to understand its role in RRMS over the lifespan of the disease.

| Limitations and future work
The effect size of connectome differences between HCs and clinical cohorts is variable.It can be quite small and intersubject variability high, leaving a dependence on sample size and thresholding for statistical power. 13,14As such, the analysis in this paper relies on the relatively large sample size of a homogenous cohort (i.e., pw-RRMS) to show an effect.We also did not randomise our participants or blind them to which DMT they received.Therefore, we could not examine the efficacy of the different types of DMTs (fingolimod, b-IFN/GA and DMF).In addition, while there seems to be a lateralised aspect to the improvement (i.e., that fibre density increases over time more in the right hemisphere than in the left), this was unrelated to lateralised aspects of behaviour, such as handedness.
Future work would examine whether the network that improved over time persists, grows or shrinks within 5 years, and whether this is related to cognitive decline or remediation.Motor control, value processing and reward learning associated with this network could also be examined in conjunction with task-based functional MRI, such as with delayed-reward paradigms, go/no-go tasks or stop-signal tasks.Longitudinal studies of these tasks may uncover the functional correlates of this network and their development over time.

2. 1 |
ParticipantsStructural and diffusion-weighted magnetic resonance imaging (dMRI) scans were acquired from 76 clinically stable treated pw-RRMS (mean age and standard error) at recruitment: 43.4 ± 11.1 years; 29 on beta-interferon/glatiramer acetate (b-INF/GA), 35 on fingolimod, 12 on dimethylfumarate (DMF) and 43 age-and sex-matched HCs (mean age and standard error at recruitment: 40.8 ± 11.5 years) at BL and 2-YFU.Selection for patients was based on (1) fulfilling RRMS diagnosis according to McDonalds' criteria as assessed by their treating neurologist; and (2) Expanded Disability Status Scale (EDSS) of 4.0 or less.All patients were on their treatment for at least 3 months and were excluded upon treatment change.
These make up the right superior frontal gyrus (SFG); the right supplementary motor area (SMA); the right ventral anterior nucleus of the thalamus (VA); bilaterally, the medial part of the pulvinar nucleus of the thalamus (mPUL); the right postcentral gyrus (PCG); the right precuneus; bilateral superior, middle and inferior temporal gyri (STG, MTG and ITG, respectively); bilateral post cingulate cortex (PCC); the left inferior parietal cortex (IPC); left Heschl's gyrus (HG); left angular gyrus (AG); bilateral lingual gyrus (LG); bilateral cuneus; bilateral calcarine cortex (CC); and bilateral T A B L E 1 Mean demographic scores and disease-related variables for pw-RRMS as combined and separated disease-modifying therapies (DMTs) subgroups (b-IFN/GA, fingolimod and DMF) and HCs at BL and 2-YFU.

F I G U R E 2
NBS results of HCs versus pw-RRMS at baseline.Nodes and connections that contribute to greatest difference between pw-RRMS and HCs at baseline (projected in BrainNet viewer) after 5000 permutations (p = 0.03 ± 0.01).(A) Coronal view of significant network.(B) Sagittal view of significant network.(C) Axial view of significant network.(D) 3D rotated view of significant network, looking from a superiorright position down on the projection.The right hemisphere is positioned at the front, and the left hemisphere at the back.Axes labels: S: superior; I: inferior; L: left; R: right; A: anterior; P: posterior.Nodes in orange are located according to their centroid stereotaxic MNI coordinates, scaled in size by the number of connections (nodal degree); binary (unweighted) connections are shown in blue.HC, healthy control; NBS, network-based statistics; pw-RRMS, people with relapsing-remitting multiple sclerosis.
(A) Coronal view of significant network.(B) Sagittal view of significant network.(C) Axial view of significant network.(D) 3D rotated view of significant network, looking from a superior-right position down on the projection.The right hemisphere is positioned at the front, and the left hemisphere at the back.Axes labels: S: superior; I: inferior; L: left; R: right; A: anterior; P: posterior.Nodes in orange are located according to their centroid stereotaxic MNI coordinates, scaled in size by the number of connections (nodal degree); binary (unweighted) connections are shown in blue.HC, healthy control; NBS, network-based statistics; pw-RRMS, people with relapsing-remitting multiple sclerosis.F I G U R E 4 NBS results of longitudinal contrast within pw-RRMS (fibre density network RRMS 2-YFU vs. RRMS BL ).A significant network is made up of nodes and connections that contribute to the greatest difference between pw-RRMS at 2-YFU and pw-RRMS at BL (projected in BrainNet viewer) after 5000 permutations (p = 0.04 ± 0.01).(A) Coronal view of significant network.(B) Sagittal view of significant network.(C) Axial view of significant network.(D) 3D rotated view of significant network, looking from a superior-right position down on the projection.The right hemisphere is positioned at the front, and the left hemisphere at the back.Axes labels: S: superior; I: inferior; L: left; R: right; A: anterior; P: posterior.Nodes in orange are located according to their centroid stereotaxic MNI coordinates, scaled in size by the number of connections (nodal degree); binary (unweighted) connections are shown in blue.NBS, network-based statistics; pw-RRMS, people with relapsing-remitting multiple sclerosis.F I G U R E 5 Correlation of clinical assessment 'Attention' with fibre density (log 10 scaled for visualisation).Attention (as part of the ARCS) was plotted against fibre density (log 10 scale) yielding r = 0.237, p = 0.044 corresponding to a weak positive correlation.ARCS, audio-recorded cognitive screen.*p value ≤ 0.05.