Reviews

Mind the gap: from neurons to

networks to outcomes in multiple
sclerosis
Declan T. Chard   1,2 ✉, Adnan A. S. Alahmadi   3, Bertrand Audoin4,5, Thalis Charalambous1,
Christian Enzinger6,7, Hanneke E. Hulst   8, Maria A. Rocca   9, Àlex Rovira10,
Jaume Sastre-Garriga   11, Menno M. Schoonheim   8, Betty Tijms12, Carmen Tur   1,13,
Claudia A. M. Gandini Wheeler-Kingshott1,14,15, Alle Meije Wink   16, Olga Ciccarelli1,2,
Frederik Barkhof   2,16,17 and the MAGNIMS Study Group*
Abstract | MRI studies have provided valuable insights into the structure and function of
neural networks, particularly in health and in classical neurodegenerative conditions such as
Alzheimer disease. However, such work is also highly relevant in other diseases of the CNS,
including multiple sclerosis (MS). In this Review, we consider the effects of MS pathology on
brain networks, as assessed using MRI, and how these changes to brain networks translate
into clinical impairments. We also discuss how this knowledge can inform the targeting of
MS treatments and the potential future directions for research in this area. Studying MS is
challenging as its pathology involves neurodegenerative and focal inflammatory elements,
both of which could disrupt neural networks. The disruption of white matter tracts in MS is
reflected in changes in network efficiency, an increasingly random grey matter network
topology, relative cortical disconnection, and both increases and decreases in connectivity
centred around hubs such as the thalamus and the default mode network. The results of
initial longitudinal studies suggest that these changes evolve rather than simply increase
over time and are linked with clinical features. Studies have also identified a potential
role for treatments that functionally modify neural networks as opposed to altering
their structure.

Connectomics
The comprehensive study
of neural networks with the
ultimate aim of generating
maps that represent all
connections in the CNS.
✉e-mail: d.chard@ucl.ac.uk
https://doi.org/10.1038/
s41582-020-00439-8
Long before the advent of connectomics, the brain was
known to rely on neural connections to function, but
technology has only recently provided us with the prac-
tical tools to assess the integrity and function of brain
networks in life. Despite ongoing methodological devel-
opment and known limitations, studies investigating
neural networks have already provided fundamental
insights into disease processes and how these translate
into disability. To date, most work has been performed in
classical neurodegenerative conditions. For example,
in Alzheimer disease, evidence suggests that the sequen-
tial involvement of brain regions can be explained by the
spread of pathology through neural networks, in particu-
lar the default mode network (DMN), and that patterns
of network involvement can explain clinical phenotypes1.
Multiple sclerosis (MS) is an inflammatory, demy-
elinating and neurodegenerative disease of the CNS2, in
which sodium channel function3, energy consumption4
and tissue blood perfusion5 are altered. The disease is
the most common neuroinflammatory cause of neu-
rological disability in younger adults in Europe6; how-
ever, the cause of MS and the mechanisms underlying
long-term disability remain uncertain. The existence
of a disparity in MS between neurological (and cogni-
tive) impairment and the brain pathology that is visible
with MRI is well recognized in the field7. This disparity
has been observed in trials of potential treatments for
MS, for example, in a recent phase II study, treatment
with ibudilast was associated with an effect on whole
brain and cortical atrophy, but no effect of treatment on
clinical outcomes was detected8. Resolving this issue is
important, as MRI measures are now the main outcome
in early-phase clinical trials in MS and are increasingly
used in clinical practice to assess the progression of MS
pathology. Therefore, the disparity between conven-
tional MRI measures and clinical outcomes has prac-
tical implications for the development and subsequent
application of disease-modifying treatments.

NaTure RevIeWS | NEuROlOGy volume 17 | March 2021 | 173

Reviews
Key points
• Multiple sclerosis (MS) pathology affects neuroaxonal structure (for example,
via axonal transection) and function (for example, via demyelination), and does
so in both lesions and extra-lesional tissues.
• Multiple pathological processes can combine to affect neural network function.
• In individuals with MS, white matter tracts are disrupted by lesions, the cortex is
relatively disconnected and the grey matter network topology is more random than
in healthy controls.
• Increases and decreases in connectivity, centred around hubs such as the thalamus
and default-mode network, are seen; abnormal connectivity seems to evolve rather
than simply progress over time, for example, early increases can be followed by
later decreases.
• Both structural and functional network changes are associated with clinical outcomes,
and treatments that modify neural network function — not structure — might still have
a beneficial effect.
In this Review, we briefly describe the relevant clin-
ical and pathological features of MS and introduce
brain networks and the methods used to model them
in vivo, before discussing the potential applications of
network-based approaches to bridge the gap between
clinical outcomes and conventional MRI measurements.
We focus on the effect of MS pathology on brain net-
works, the translation of network disruption into neu-
rological and cognitive impairments, and ask how this
knowledge can inform the targeting of MS treatments.
Author addresses
1
NMR Research Unit, Queen Square MS Centre, Department of Neuroinflammation,
UCL Institute of Neurology, Faculty of Brain Sciences, University College London,
London, UK.
2
National Institute for Health Research (NIHR) University College London Hospitals
(UCLH) Biomedical Research Centre, London, UK.
3
Department of Diagnostic Radiology, Faculty of Applied Medical Science,
King Abdulaziz University (KAU), Jeddah, Saudi Arabia.
4
Aix-Marseille University, CNRS, CRMBM, Marseille, France.
5
AP-HM, University Hospital Timone, Department of Neurology, Marseille, France.
6
Department of Neurology, Research Unit for Neuronal Repair and Plasticity,
Medical University of Graz, Graz, Austria.
7
Department of Radiology, Division of Neuroradiology, Vascular and Interventional
Radiology, Medical University of Graz, Graz, Austria.
8
Department of Anatomy and Neurosciences, MS Center Amsterdam, Amsterdam
Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam,
The Netherlands.
9
Neuroimaging Research Unit, Institute of Experimental Neurology, Division of
Neuroscience, San Raffaele Scientific Institute, Vita-Salute San Raffaele University,
Milan, Italy.
10
Section of Neuroradiology, Department of Radiology Hospital Universitari Vall
d’Hebron, Universitat Autònoma de Barcelona, Barcelona, Spain.
11
Servei de Neurologia/Neuroimmunologia, Multiple Sclerosis Centre of Catalonia
(Cemcat), Hospital Universitari Vall d’Hebron, Universitat Autònoma de Barcelona,
Barcelona, Spain.
12
Alzheimer Center Amsterdam, Department of Neurology, Amsterdam
Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam,
The Netherlands.
13
Department of Neurology, Luton and Dunstable University Hospital, Luton, UK.
14
Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy.
15
Brain MRI 3T Research Center, IRCCS Mondino Foundation, Pavia, Italy.
16
Department of Radiology & Nuclear Medicine, Amsterdam Neuroscience,
Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands.
17
Institutes of Neurology and Healthcare Engineering, University College London,
London, UK.
*A list of authors and their affiliations appears at the end of the paper.
174 | March 2021 | volume 17 www.nature.com/nrneurol
We end by discussing the future directions for this area
of research.
Clinical features of MS
Although MS is usually thought of as causing motor and
sensory symptoms, the disease can affect any aspect of
CNS function; for example, a third to two-thirds of indi-
viduals with MS have cognitive impairment9. The symp-
toms of MS can develop acutely (during relapses) or
progressively over months to years. Relapses occur when
lesions form in clinically eloquent parts of the CNS, for
example, the optic nerves, although most lesions occur
without having direct clinically apparent effects10. The
mechanisms underlying progressive MS are not as well
understood as those underlying the relapsing–remitting
form. Correlations between clinical measures of disease
progression and the accrual of white matter lesions are
relatively modest; although clinical measures correlate
more strongly with measures of brain and spinal cord
atrophy, this association is still insufficient to explain or
predict clinical impairments in individuals with MS and
leaves around half of the variability in disability scores
unexplained11.
In this Review, we use motor and cognitive func-
tions as our main examples as both are affected early
in the course of MS, increase during progressive MS,
and are substantial causes of disability and unemploy-
ment. Motor function, as measured using the Expanded
Disability Status Scale (EDSS)12, is the most commonly
assessed outcome in clinical trials and cognitive dys-
function is increasingly also being included. We will also
discuss visual disturbances and fatigue, which frequently
occur in MS. The visual system is commonly one of the
first to be affected by MS and, for 30–40% of individuals,
an episode of optic neuritis is the first symptom of the
disease 13. Detailed clinical and neurophysiological
methods exist for assessing visual pathway function and
we therefore consider the visual system when discuss-
ing the challenges of linking structure and function.
Unfortunately, fatigue is particularly difficult to study, as
a range of different MS symptoms are often reported
as fatigue by patients14 and a consensus definition on
how to classify or measure fatigue is lacking. However, as
previous work has noted that patients with fatigue show
extensive changes in cognitive networks15,16, we briefly
consider studies of fatigue as an example of diverse
symptoms that are, in part, influenced by a common
underlying cause.
MS pathology
MS pathology can affect neuronal network function
in a variety of ways (Box 1). For example, neuro­axonal
loss will stop signal conduction across a network,
and demyelination will slow and disperse transmis-
sion. Both of these changes can have a substantial
effect on neurological function, as has been observed
in individuals with optic neuritis 17. White matter
lesions are the most-studied aspect of MS pathology.
Acute inflammation in white matter lesions is asso-
ciated with demyelination and axonal transection18,
along with trans-synaptic neurodegeneration 19–21.
In extra-lesional (normal-appearing) white matter,

Hub
A network node that has a high
number of direct connections
with other nodes.
NaTure RevIeWS | NEuROlOGy
axonal loss, demyelination and gliosis occur22, although
the degree to which these changes are secondary to, or
independent from, axonal transection in lesions is not
yet clear22. In individuals with MS, grey matter is often as
extensively demyelinated as white matter, if not more so23.
Axonal loss is observed in grey matter lesions24, but
synaptic and neuronal loss is not confined to lesions and
occurs with a similar severity in extra-lesional tissues25.
Deep grey matter structures are not spared; the thala-
mus seems to be affected early in the disease course —
even after a single inflammatory episode26 — and shows
substantial neuronal loss, more so than in other deep or
cortical grey matter27. This finding might be of particular
relevance when considering the effect of MS pathology
on network performance as the thalamus is thought to
have a pivotal role as a hub in many brain networks28.
However, the thalamus is a not a homogenous structure
but instead consists of nuclei that seem to be differen-
tially associated with specific clinical outcomes, for
example, cognition and fatigue29.
Physiological alterations also occur in MS and,
although they are reported less often than tissue struc-
tural alterations, they can have an important influence
on brain function. These physiological changes include
grey matter hypo-perfusion, with delayed arterial bolus
arrival transit times30, and sodium channel polymor-
phisms, for example, in Nav1.8 channels that are ectopi-
cally expressed in cerebellar Purkinje cells in individuals
with the disease and have been linked with effects on
cerebellothalamic functional connectivity31. Energy defi-
cits associated with tissue hypoxia have also been shown
to correlate with processing speed in MS32,33. Although
neurological and cognitive deficits in people with MS
are often thought of as disconnection syndromes result-
ing from white matter pathology34, grey matter changes
have substantial implications for network function and
are also associated with the progression of cognitive
and motor disability35–38.
Assessing brain networks with MRI
The three main MRI methods that have been used
to study brain networks are diffusion tensor imaging
(DTI), 3D T1-weighted scans and functional MRI
(fMRI). DTI assesses white matter tissue microstructure,
3D T1-weighted scans provide anatomical imaging of white
and grey matter, and fMRI measures brain activity, either
during the resting state39 or in response to a task. For con-
text, an ~8 mm3 DTI voxel contains ~200,000 neurons and
a typical fMRI cluster is ~80 mm3 (~2 million neurons)40,41.
In addition, the temporal resolution of fMRI is ~10 seconds,
whereas EEG and magnetoencephalography can detect
changes in neural activity over milliseconds42.
Network architecture is described by connections and
their layout (topology). Connectivity can be determined
by tracing (anatomical) links from region to region, by
looking for (functional) associations in neural activity or
by assessing similarity in structural features, for exam-
ple, cortical thickness. Currently, the main approach for
describing and quantifying brain network topology is
‘graph theory’43, where a ‘graph’ represents a connectiv-
ity map, with ‘nodes’ representing brain areas and the
connection between nodes being termed ‘edges’. Graph
Reviews
Box 1 | Factors that could affect network function
Structural
• Demyelination
• Axonal transection and degeneration
• Synaptic loss
• Neuronal loss
• Gliosis
Functional
• Inflammation
• Hypoxia
• Mitochondrial dysfunction
• Sodium accumulation
• Neurotransmitter deficits
theory can be applied to both structural MRI and fMRI;
however, structural and functional connectivity metrics
are usually different as they represent different properties
of the brain connectome. In both cases, nodes are located
in grey matter regions. Edges are more complicated and
they can represent white matter tracts traced between
grey matter regions, synchronization of fMRI activity
between grey matter regions or co-variation in cortical
structure (for example, thickness). With the latter two
definitions ‘connectivity’ can exist in the absence of
specific white matter tracts connecting the two regions.
White matter connections can be traced on DTI scans
but DTI measures are affected by MS lesions, making
this tracing difficult where tracts cross or pass through
white matter lesions44,45. However, an alternative way of
considering white matter connectivity is from the per-
spective of a ‘disconnectome’46, which maps the extent of
disconnection between grey matter regions that results
from focal lesions. This approach might be particularly
pertinent to the study of MS.
Functional connectivity is usually assessed by iden-
tifying direct or indirect correlations between the
activation of different grey matter regions, for exam-
ple, by measuring variation in the blood oxygenation
level-dependent signal. Direct assessment identifies
correlations between regional measures regardless
of the state of the brain (resting-state fMRI). Indirect
assessment identifies associations with a common fea-
ture, for example, brain regions that show simultaneous
functional activation during a motor task (task-based
fMRI). Falling between direct and indirect assessment is
effective connectivity47, which relies on the construction
of models that incorporate both types of connections
and aims to define the directionality of information
flow between regions. There are several methods used
for the assessment of effective connectivity, including
structural equation modelling48–50, psychophysiological
interactions47 and dynamic causal modelling51. In clinical
trials and clinical practice, robust measures of connec­
tivity in individual participants or patients will be
required. This kind of individual analysis has proven
possible with structural MRI52; however, given the rather
‘noisy’ nature of data on individual connections, most
network-based analyses have averaged findings across
groups of participants.
volume 17 | March 2021 | 175
Reviews
Interpreting results
Connectomics is a rapidly developing field and new
measures of connectivity are still being proposed. The
statistical methods underlying network-based analyses
are complex and the approach used can substantially
influence apparent network structure and function.
As such, links between the cortical network connections
identified with these analyses and classical functional or
anatomical connections remain controversial53.
Of the several topological features that can be
extracted from brain networks (Box 2), it is unclear which
have the greatest relevance to health and disease. In part,
this uncertainty reflects the bespoke nature of brain net-
works; different topologies are optimized to serve differ-
ent functional outcomes, so to generalize across a diverse
range of neurological and cognitive functions is difficult.
In disease states, some network features are preserved,
whereas others are lost or emerge (discussed below).
Of these changes, the appearance of new features has
proven most difficult to explain; specifically, whether
this emergence represents the unmasking of a (poten-
tially deleterious) network feature or is the result of
functionally useful neural plasticity remains unclear54.
Structural connectivity established by tracing white
matter tracts, known as tractography, clearly demon-
strates a link between regions and a decrease in this con-
nectivity can reasonably be interpreted as representing
the disruption of a connection. Some tractography-based
Box 2 | Graph theory measures
Integration
• Characteristic path length
• Global efficiency
Segregation
• Clustering coefficient
• Transitivity
• Local efficiency
• Modularity
Centrality
• Closeness centrality
• Eigenvector centrality
• Betweenness centrality
• Within-module degree z-score
• Participation coefficient
Network motifs
• Anatomical and functional motifs
• Motif z-score
• Motif fingerprint
Resilience
• Degree distribution
• Average neighbour degree
• Assortativity coefficient
Other
• Degree distribution preserving network randomization
• Measure ‘of network small-worldness’
• Rich club coefficient
Data from ref.43.
176 | March 2021 | volume 17 www.nature.com/nrneurol
studies have shown apparent increases in connectivity in
adults55. However, in the adult CNS, entirely new axonal
connections are not thought to occur (although neuro-
genesis with synapse formation is seen56) and a loss of
crossing fibres, which disrupt tractography, provides a
more plausible explanation for the apparent increase
in connectivity. Using tract templates as opposed to
tractography to extract measures such as fractional ani-
sotropy from DTI overcomes this limitation but such
measures then provide an assessment of tract structural
integrity as opposed to a direct measure of connectivity57.
When measures of structural connectivity are made
on the basis of grey matter characteristics (for exam-
ple, cortical thickness), processes that heterogeneously
and randomly affect the cortex will tend to reduce the
detection of correlations between regions, whereas
characteristics that show a regional predilection might
increase the strength of correlations58 even if the axonal
connectivity between regions is unchanged. Similarly,
tract-specific pathology might reduce correlations when
it disrupts connections or increase correlations where
pathology simultaneously affects both the origin and
target of a tract. To add to the complexity, multiple
disease effects — some network-mediated and others
not — might act simultaneously on the same area58,59.
In fMRI studies, although reduced correlations between
regions might be a result of decreases in structural con-
nectivity, they might also represent a de-synchronization
of activity. For example, demyelination might slow and
disperse (as opposed to stop) action potential conduction,
but would still result in seemingly reduced functional
co-activation. An increase in fMRI connectivity could be
the result of increased neural communication but deter-
mining whether it represents an unmasking of previously
hidden features (through the loss of competing functional
activity or disinhibition), structural re-organization, func-
tional adaptation or compensation is challenging. The
interpretation of fMRI results is further complicated by
the dynamic nature of functional connectivity, which can
vary over seconds, for example, during the course of a
single scan60.
Findings in MS
Effect of MS pathology on brain networks
Function. Evidence indicates that MS pathology sub-
stantially alters the structural and functional integrity
of brain networks and does so through a combination of
effects on both grey matter and white matter. However,
studies have yielded seemingly inconsistent results,
sometimes finding different abnormalities in the same
regions. These inconsistencies are likely to be due, in
part, to differences in the location and magnitude of
tissue damage observed in individuals with different
clinical subtypes of MS or with different durations of
disease or degrees of disability61.
To date, most functional connectivity studies in MS
have used (mainly motor and cognitive) task-based fMRI,
although resting-state fMRI has been used more recently,
particularly in studies of fMRI correlates of cognitive defi-
cits. Findings have been variable (Table 1)62; some studies
using task-based fMRI have shown an increase in func-
tional brain activation in individuals with MS compared
 Reviews

Table 1 | fMRI studies that compared individuals with MS with healthy controls
Study Functional changesa MS phenotype Disease Age (years)b Number of EDSS (median,
duration (years)b participants range)
Motor task-associated activation
Wegner et al. ↑ activation; RRMS or SPMS 6.7a MS, 35c; 56 MSd; 60 HC; 2.0, 0–7.5
2008 (ref.128) ↓deactivation HC, 30c (multi-centre)
Manson et al.
2008 (ref.129)
Mancini et al.
2009 (ref.130)
Colorado et al. ↑ activation RRMS 10.2 MS, 41.8; 22 MS; 23 HC 0, 0–1.5
2012 (ref.131) HC, 38.1
Rocca et al. ↑ and ↓activation RRMS, sub- Non-fatigued, MS non- 29 MS non-fatigued; Non-fatigued: 1.5,
2016 (ref.132) related to fatigue grouped based 10.6; fatigued, fatigued, 40.0; 50 MS fatigued; 26 HC 0–4.0; fatigued: 2.0,
on Fatigue Impact 12.9 MS fatigued, 1.0–4.0
Scale score 42.6; HC, 39.2
Resting-state connectivity
Rocca et al. ↑ and ↓ connectivity RRMS 9.0 MS, 41.4; 85 MS; 40 HC 2.0, 0–6.0
2012 (ref.133) HC, 40.6
Schoonheim ↓ cortical centrality RRMS, SPMS 7.7 MS, 40.98; 128 MS (112 RRMS, 9 2.0, 0.0–8.0
et al. 2014 and ↑thalamic or PPMS HC, 40.38 SPMS, 7 PPMS); 50 HC
(ref.74) connectivity
Rocca et al. ↓ functional Relapse-onset MS 10.8 MS, 37.5; 69 MS; 42 HC 1.5, 0.0–6.5
2015 (ref.134) connectivity HC, 36.4
Rocca et al. Loss of hubs and RRMS, SPMS 13.7 MS, 42.3; 256 MS (121 RRMS, 80 3.0, 0.0–9.0
2016 (ref.135) ↓ in nodal degree or BMS HC, 41.7 SPMS, 45 BMS); 55 HC
Eijlers et al. ↑ connectivity RRMS, SPMS 14.6 MS, 48.1; 332 MS (243 RRMS, Cognitively
2017 (ref.136) (cognitively impaired or PPMS HC, 45.9 53 SPMS, 36 PPMS); impaired: 4, 0–8;
Meijer et al. MS only) 87 MS cognitively mildly impaired
2017 (ref.137) impaired, 65 MS mildly and preserved
impaired and 180 MS both: 3, 0–8.
preserved; 96 HC
Rocca et al. ↓connectivity, no global CIS, RRMS, SPMS, 12.1 MS, 41.0; 215 MS (13 CIS, 119 2.0, 0.0–8.5
2018 (ref.66) differences between PPMS or BMS HC, 42.7 RRMS, 41 SPMS, 13
MS phenotypes PPMS, 29 BMS); 98 HC
Hidalgo de la ↑ and ↓connectivity; RRMS or Non-fatigued, MS non- 122 MS (100 RRMS, Non-fatigued: 1.5,
Cruz et al. ↑ and ↓ connectivity progressive MSe; 10.8; fatigued, fatigued, 35.0; 22 progressive); 86 MS 0–8.0; fatigued: 4.0,
2018 (ref.77) in fatigued compared sub-grouped 13.4 MS fatigued, non-fatigued; 36 MS 0–6.5
with non-fatigued MS based on Fatigue 44.3; HC, 41.5 fatigued; 94 HC
Impact Scale score
Tommasin ↓ connectivity only RRMS or SPMS 8.63 MS, 38.3; 119 MS (91 RRMS, 28 2.0, 0–7.5
et al. 2018 MS with EDSS >3 sub-grouped by HC, 35.6 SPMS); 79 MS EDSS
(ref.138
) EDSS ≤3 or >3 ≤3; 40 MS EDSS >3;
42 HC
Meijer et al. ↑ and ↓ functional RRMS, SPMS Information MS, 48.14; 330 MS (243 RRMS, Information
2018 (ref.97) connectivity; or PPMS processing HC, 45.9 51 SPMS, 36 PPMS); processing impaired:
↑ functional sub-grouped impaired, 15.82; 130 MS information 4.0, 3.0–6.0; MS
connectivity in based on impaired information processing impaired; information
information processing or preserved processing 200 MS information processing
impaired compared information preserved, 9.80 processing preserved; preserved: 3.0,
with preserved processing speed 96 HC 2.0–4.0
Cordani et al. ↑ and ↓ functional RRMS or 12.6c MS, 43.0c; 366 MS (251 RRMS, 2.5, 1.5–5.5
2020 (ref.104) connectivity progressive MSe HC, 38.0c 115 progressive MS);
134 HC
For motor task studies, based on previous work highlighting that small sample sizes may yield unreliable results139, only studies with ≥20 participants per group
are shown. For resting state studies, only studies with ≥40 participants per group are shown140. BMS, benign multiple sclerosis; CIS, clinically isolated syndrome;
EDSS, Expanded Disability Status Scale; HC, healthy controls; MS, multiple sclerosis; PPMS, primary progressive multiple sclerosis; RRMS, relapsing-remitting
multiple sclerosis; SPMS, secondary progressive multiple sclerosis. aMS compared with HC unless stated otherwise; bMean value unless stated otherwise; cMedian;
d
RRMS/SPMS numbers not given; ePPMS or SPMS not specified.

with healthy controls, whereas other studies have found
increases in some brain regions and decreases in others
(motor task examples are given in Table 1). Several fac-
tors could explain this inconsistency. First, the distribu-
tion and severity of pathology might differ between the
cohorts of participants in the different studies. Second,
functional connectivity abnormalities can reflect both
pathological abnormalities and compensatory mech-
anisms. Last, task-based fMRI can also be confounded
by variations in the design and performance of the task.
The latter limitation can be circumvented by using
resting-state fMRI and, with this approach, some unifying

NaTure RevIeWS | NEuROlOGy volume 17 | March 2021 | 177

Reviews
themes have started to emerge, principally that functional
connectivity is increased in early MS63,64 and decreased
in individuals with more disabling or longer-duration
disease65–67 (Table 1). Consistent with these observa-
tions, a recent longitudinal resting-state fMRI study in
relapsing-remitting MS identified increases in functional
activity early in MS followed by a tendency of a decrease
in connectivity with disability progression68.
The changes in resting-state connectivity observed
in MS seem to be mostly centred around hub areas such
as those that comprise the DMN. Interestingly, changes
to the DMN in MS seem somewhat non-specific, as
evidence suggests that they are related to several symp-
toms of the disease, including cognitive impairment69,
disability66 and fatigue70. This broad influence might
be explained by the notion that dysfunction within this
important hub network could alter the efficiency of the
entire brain network, which in turn could relate to many
symptoms in MS71. Fatigue has also been hypothesized
to relate to alterations in these cognitive and motor net-
works via a chronic mismatch between expected and
measured output, owing to erroneous signals arising
from these networks15. However, given the aforemen-
tioned difficulties in quantifying fatigue, such hypothe-
ses remain difficult to prove experimentally, warranting
additional studies on the topic. Preliminary work on
treating fatigue through stimulation of the cortical areas
that show altered fatigue-related connectivity (that is, the
DMN, motor network and the insula) seems promising
but still requires validation in larger samples16.
Most studies in this area have focused on cortical grey
matter, but deep grey matter (particularly the thalamus)
and the cerebellum are substantially affected in MS72,73.
Similar to in the cortex, functional connectivity between
the thalamus and other regions often seems to be altered
in MS: both increased and decreased functional connec-
tivity between the thalamus and various cortical regions
has been observed in different studies74–78. As mentioned,
studying individual thalamic nuclei within this context
is of interest. Data suggest that patients with fatigue79
or cognitive impairment 29 show a combination of
hyper-connectivity and hypo-connectivity depending on
the thalamic nucleus being studied; both changes were
correlated with a worsening of symptoms. Therefore,
more work is needed to not only disentangle the effects
of fatigue from those of cognitive impairment but also to
identify why individual thalamic nuclei behave so differ-
ently. As part of the thalamo-striato-cortical loop, altered
basal ganglia connectivity is also frequently observed in
MS, especially in the context of fatigue79.
Interestingly, the evolution of functional connectivity
during MS seems to differ between the deep grey matter
and the cortex. Functional connectivity among the deep
grey matter structures as well as between the deep grey
matter and the cortex tends to increase as the disease
progresses, whereas inter-cortical connectivity tends to
decrease80. The topology of networks can also change
as hubs (for example, the thalamus) usually seem to be
preserved68,78. Few studies have investigated how cor-
tical networks are affected by white matter pathology
but local network efficiency seems to decrease as the
whole-brain white matter lesion load increases68.
178 | March 2021 | volume 17 www.nature.com/nrneurol
Structure. Considering structural features, tissue atro-
phy in MS affects some deep grey matter structures81 and
cortical regions82 more than others. In the cortex, work
using source-based morphometry identified several
overlapping patterns of atrophy in a cohort of individu-
als with MS58, raising the possibility that a combination
of network-mediated and non-network-mediated effects
might be responsible. However, although shared regional
disease effects might increase some associations between
regions, grey matter neural networks also seem to be
less structured in individuals with MS than in healthy
individuals. For example, one study found a more ran-
dom topology in people with long-standing MS than in
healthy controls61. In another study, a more random net-
work topology was found to explain deficits in cognitive
functioning in MS in excess of those deficits attributed to
either localized atrophy or lesion measures83. This ran-
dom topology might be more apparent relatively early on
in the course of MS before network-mediated pathology
results in structural correlations between regions. In one
study in individuals with primary progressive MS, com-
pared with healthy controls, similarities in cortical thick-
ness among regions initially decreased following the first
symptoms and then increased over time (participants
were followed-up for 5 years)84.
In white matter, tractography has identified abnor-
malities in global and local network efficiency that var-
ied among studies. For example, in one study, reductions
in local and global network efficiency were observed in
individuals with MS compared with healthy controls85,
whereas in another study, increases in efficiency and
changes in network topography (increased modularity
and clustering) were observed in participants with MS55.
Cohort and methodological differences could explain
the apparent discord between the results of these two
studies — despite similar median disease durations,
the two populations substantially differed in terms of
maximum disease duration. In the context of fatigue,
reductions in DMN as well as caudate connectivity have
been observed in individuals with MS, in line with the
aforementioned effects on functional connectivity86.
Linking structure and function. Linking the results of
structural and functional studies of connectivity in MS
has proven difficult, both for methodological reasons and
because MS pathology seems to differentially affect struc-
tural and functional networks throughout the course of
the disease. Here, we consider the visual system, a dis-
crete functional entity, the main pathways of which are
well characterized and — crucially — can be assessed
with different complementary modalities, providing
specific information about its structure (for example,
optic nerve MRI, DTI of the optic tracts and radiations)
and its function (for example, visual acuity and visual
evoked potentials). Thus, the visual system provides a
unique opportunity to disentangle the complex relation-
ships between an acute insult to the visual pathways and
subsequent structural and functional changes.
Evidence indicates that the structural consequences
of optic neuritis are not necessarily restricted to the
optic nerve and can extend progressively to the ante-
rior and posterior visual pathways20,87,88. Functionally,

during the acute phase of optic neuritis, reduced acti-
vation is observed in the visual cortical areas and this
is associated with reduced visual acuity89–92. Greater
activation in extrastriate visual regions — particularly
the lateral occipital complex — is associated with bet-
ter visual acuity regardless of the level of structural and
functional damage within the anterior and posterior
visual pathways93 and, importantly, greater activation
in this region has been found to predict eventual visual
recovery94. More recently, a difference between cortical
functional responses to the first and subsequent epi-
sodes of optic neuritis was observed with resting-state
fMRI. In one study, participants who had experienced a
single episode of optic neuritis had increased functional
connectivity within the visual network compared with
participants who had experienced a single inflamma-
tory episode elsewhere in the CNS, even when the optic
radiations were apparently structurally intact95. After
multiple episodes of optic neuritis, increased func-
tional connectivity was still seen in the lateral occipital
complex but coexisted with decreased functional con-
nectivity in other parts of the visual cortical network95.
Intriguingly, stimulation of the unaffected eye in some-
one who has had optic neuritis has also been found to
induce abnormal functional responses91,92. Considered
together, these observations highlight that functional
changes do not simply mirror structural damage but,
instead, reflect potentially adaptive and non-adaptive
functional changes that can extend beyond the cor-
tical regions that are immediately connected to the
affected eye.
When studying cognition in MS, similarly complex
relationships between structural damage and functional
Assessment methods Process and outcome
Structural damage
• Disconnection – axon and dendrite loss, demyelination,
MRI and PET
metabolic dysfunction
• Loss of neurons
Adaptation Maladaptation
• Changes in functional • Changes in functional
connectivity connectivity
fMRI, EEG
• Changes in network • Hub overload
topology
and MEG
Stability
• Brain functional reserve
• Hub preservation
Clinical function
High
Threshold effects
Clinical
examination Non-linear associations
Low
High Low
Network performance
Fig. 1 | From neurons to clinical outcomes in MS. The clinical outcomes observed
in individuals with multiple sclerosis (MS) represent the effect of combinations of
pathological processes on network performance, compensated for by network
adaptation or augmented by network maladaptation and offset by innate network
stability. Each element of this process can be assessed in life using different techniques,
but bridging the gaps between imaging, neurophysiological and clinical measures to
provide an integrated model of MS pathology and its clinical consequences has yet
to be achieved. fMRI, functional MRI; MEG, magnetoencephalography.
NaTure RevIeWS | NEuROlOGy
Reviews
activation emerge. For example, a study on cognitive
outcomes in MS found that thalamic volume and acti-
vation were both — and, in part, independently — asso-
ciated with information processing speed and executive
function96. More recently, structural damage — meas-
ured with a composite score of lesions, atrophy and
fractional anisotropy — was found to correlate better
with impaired information processing speed than with
functional changes in individuals with MS97. However,
in participants with a similar degree of structural dam-
age, those with severe functional changes showed lower
information processing speeds than those with only mild
functional changes. In another study, changes in network
structure (assessed using white matter tractography)
were apparent soon after first symptoms suggestive of
MS occurred, but changes in functional networks only
become detectable in participants with clear ongoing
disease activity98. In addition, correlations between
decreased structural connectivity and decreased func-
tional connectivity were most apparent in subcortical
networks. Lesions and extra-lesional changes within
relevant tracts might have different effects on clini-
cal outcomes. For example, one study showed limited
overlap between tracts and lesions99, suggesting that
extra-lesional damage is relevant; however, another
study found that, when lesions occur in a tract, they
dominate associations with clinical outcomes100.
Although drawing all of this evidence together is
difficult at present, clearly, we must carefully take into
account the clinical and MRI characteristics of study
cohorts to build a coherent model of brain network
changes over the course of MS (Fig. 1). We must also rec-
ognize that white matter and grey matter pathology can
have opposing effects on apparent network performance
at different points in the course of MS. For example, the
authors of one study simulated the potential effects of
grey matter and white matter pathology on functional
connectivity and observed a global increase in func-
tional connectivity associated with cortical and thalamic
pathology101. The authors also observed an increase
followed by a decrease in connectivity associated with
increasing white matter pathology. This observation also
suggests that a proportion of the apparent changes in
functional connectivity observed in MS represents the
direct effects of structural damage and is independent
of adaptive or maladaptive network changes.
Neurological and cognitive impairment
Studies that investigate the associations between brain
connectivity and disability in MS have produced
inconsistent results. For example, changes in network
topology83 as well as increases and decreases in struc-
tural and functional connectivity have been found to
correlate with disability66,67,102. However, importantly,
fMRI measures correlate with clinical outcomes at
least partly independently of measures of structural
connectivity66,75,103,104, suggesting that functionally mean-
ingful effects on network performance can be achieved
through the modulation of neuronal function.
Indeed, this observation has already proven rele-
vant to clinical practice: in the results of a clinical trial
in 301 participants with MS, fampridine, an agent that
volume 17 | March 2021 | 179
Reviews
Table 2 | Mechanisms by which treatments could sustain and promote brain network function in MS
Neuronal Substrate Slows or prevents
treatment aim network dysfunction via
Structural Neuroaxonal loss Neuroprotection
Synaptic loss Neuroprotection
Functional Signal conduction Preventing demyelination
NA, not applicable.
modifies potassium channel excitability as opposed to
promoting remyelination or neuronal repair, was asso-
ciated with improved walking speed105. The results of a
smaller study in 12 participants with MS suggested that
amifampridine can increase functional connectivity106.
Similarly, treatment with the cholinesterase inhibitor
rivastigmine was associated with an increase in func-
tional connectivity in participants with MS (n = 15)
performing cognitive tasks, and this was also associated
with a trend towards improvements in neuropsycholog-
ical performance107. Notably, the clinical effects of struc-
tural damage, measured as grey matter atrophy, could
also be offset by a higher baseline cognitive reserve,
measured as verbal intelligence, and this too is reflected
in the relative preservation of functional connectivity108.
Evidence indicates that functional connectivity is
itself dynamic, waxing and waning over time, and this
observation is also clinically relevant. Better memory
function has been associated with more consistent
hippo­campal functional connectivity109, higher infor-
mation processing speeds have been associated with a
greater increase in dynamic connectivity when switch-
ing from resting to task-based states110, and executive
functioning seems to be more closely linked to dynamic
than to static functional connectivity111. Chronic neu-
ropathic pain in MS also seems to be associated with
increased dynamic connectivity in the DMN and greater
connectivity between the DMN and salience network112.
Considering the vulnerability of brain networks to
MS pathology, in one study, researchers found substan-
tial differences in the number of structural connections
associated with two cognitive tests commonly used in
MS research: 160 structural connections correlated
with scores on the Paced Auditory Serial Addition Test,
whereas only 11 connections correlated with scores
on the Symbol Digit Modalities Test113. They con-
cluded that the Paced Auditory Serial Addition Test was
more cognitively demanding and so more vulnerable
to pathology than the Symbol Digit Modalities Test.
Clinical progression has been hypothesized to represent
a ‘network collapse’71, with unfolding pathological pro-
cesses having ever greater effects on clinical outcomes
as brain networks deteriorate. In one study, simulations
were used to compare the effects of lesions on connec-
tivity in individuals with relapsing-remitting MS and in
individuals who have subsequently gone on to develop
progressive MS (and so who have more abnormal
neural networks). More hubs were lost in individuals
with progressive MS but the majority of differences in
connectivity between the two groups were unchanged
by the simulated disconnection caused by lesions114.
180 | March 2021 | volume 17 www.nature.com/nrneurol
Repairs or improves network function via
NA
Promoting synaptogenesis; slowing synaptic
stripping
Promoting remyelination; improving signal
conduction
Other researchers have set both functional connectiv-
ity and lesions in the context of clinical outcomes, with
the aim of identifying which functional connectivity
changes are due to lesions themselves and which are
network adaptations that either enhance or impair clin-
ical performance73. They found that, in some regions,
functional connectivity was associated with lesions and
in others it was not, indicating that the lesions do not
necessarily have a consistent effect.
Treatment effects
A variety of mechanisms exist through which therapeutic
intervention could affect brain network function (Table 2);
most of these mechanisms are not directly targeted by
current MS treatments. For treatments that alter net-
work structure, it could take many months of treatment
for changes to become apparent, whereas treatments that
affect network function should have a more immediate
impact. Relatively few studies have assessed the effect
of treatment on connectivity in MS but the results do
suggest that improvements in clinical function can be
achieved without necessarily altering the structure of a
network. Cognitive rehabilitation is associated with sub-
stantial increases in DMN activity and variable changes
in task-based fMRI activity115 but functional changes are
not necessarily mirrored by structural alterations116. This
observation is further supported by evidence from stud-
ies of transcranial magnetic stimulation in MS that have
shown rapid improvements in fatigue117 and working
memory118 as well as reduced spasticity with associated
fMRI changes119 that cannot plausibly be mediated by
structural change in networks. Similarly, as noted ear-
lier, treatment with amifampridine is associated with
increased motor-evoked fMRI activation106.
On the basis of these observations, it is tempting
to conclude that functional connectivity measures are
more promising MRI markers of treatment efficacy in
MS than structural connectivity measures. However,
whether or not treatments that suppress inflammation,
slow neuroaxonal loss or promote re-myelination have
substantial effects on MRI-measurable network topol-
ogy and connectivity is yet to be determined — for such
effects to become apparent, longer-term studies will be
required. Importantly, even transient improvements in
clinical function with transcranial magnetic stimulation
or drug treatments imply that the underlying network is
still sufficiently structurally intact for some function to
be preserved or regained, so these improvements could
provide a useful marker of those individuals who have
the most to gain from treatments designed to prevent
neurodegeneration or promote re-myelination.
 Reviews

Nexopathies
The interactions of pathogenic
proteins with neural networks
that are vulnerable to
their effects, resulting in
characteristic patterns
and spread of abnormalities
across the CNS.
Next steps
Much work remains to be done before brain network
measures can be used routinely in clinical trials and in
clinical practice both in MS and in neurological condi-
tions in general. To facilitate meaningful comparisons
between studies, methods need to be standardized and
results reproduced. Some work on comparative connec-
tomics has been performed, with a view to identifying
common themes in networks across species120. Similarly,
meta-connectomics aims to identify consistent obser-
vations across studies121. However, the issue of scale
remains, with current MRI (and EEG and magnetoen-
cephalography) techniques assessing neural networks in
multi-millimetre to centimetre terms, potentially over-
looking small but highly relevant grey matter or white
matter features. This issue is further complicated by a
lack of temporal resolution: even measurements made in
milliseconds using EEG can prove difficult to link with
a neurological or cognitive function, except where EEG
changes can be associated with discrete neurological or
cognitive events, thus enabling event-related potentials
to be identified122. Without methods to infer missing ele-
ments that could explain observed discrepancies, these
limitations of resolution make the task of reconciling
structure and function more difficult. Furthermore,
understanding the link between cellular architecture
and physiology as well as between large-scale network
function and structure will require concerted interdis-
ciplinary work123. A unifying network topology derived
from fMRI and structural MRI might be a more tractable
aim, albeit still a challenging one.
Although the pathological substrates of network
changes in MS are not known, one post mortem study
has shown that graph theory descriptions of network
topology are linked to neuronal size and axonal den-
sity in individuals with the disease124. Nevertheless,
whether a decrease in network node activation in MS
represents the loss of neurons or their axons, arborisa-
tion, or non-structural factors such as inflammation or
mitochondrial dysfunction is not yet clear. This point
is highly relevant when we consider the potential treat-
ment targets and ways to assess treatment efficacy in
early phase clinical trials. The possibility that patho-
logical processes in MS are, in part, mediated through
neural networks and can interact with other factors,
such as regional vulnerabilities to pathology, has already
been raised58,125. In Alzheimer disease, the concept of
nexopathies has recently been proposed to explain how
pathology might spread through neural networks and
interact with intrinsic vulnerabilities, resulting in the
observed patterns of neurodegeneration126. The nexopathy
approach could perhaps also be pursued in MS.
Very few longitudinal studies of functional or struc-
tural brain networks have been performed; therefore, we
still have little insight into the dynamics of brain net-
work degeneration, repair and plasticity. In particular,
determining which elements of structural or functional
network change represent disease effects and which
are adaptive or compensatory is proving very difficult
without knowing how each element relates to changes
in neurological or cognitive function. In order to resolve
these uncertainties, studies will have to characterize the
evolution of structural and functional abnormalities
simultaneously and do so reproducibly.
Attention will also need to be given to clinical out-
come measures. The main measure currently used in
MS clinical studies is the EDSS12, which is essentially
a measure of impaired mobility. However, as impaired
mobility can arise from impairment of motor, cerebellar
and sensory function, the EDSS is intrinsically a poor
measure of any particular neurological function and so
does not enable the identification of a specific underly-
ing network. Thus, network-specific outcome measures
are clearly needed. Cognitive outcomes are more net-
work specific than outcomes based on impaired mobility
but are not perfect as they can rely on visual function,
which is often also affected in MS. In addition, some
cognitive outcomes are designed for diagnostic rather
than monitoring purposes127.
Conclusions
Network-based functional and structural studies have
provided useful insights into the pathogenesis of MS and
the cause of neurological and cognitive symptoms. MS
is associated with the disruption of white matter tracts,
which is reflected in changes in network efficiency, an
increasingly random grey matter network topology,
relative cortical disconnection, and both increases and
decreases in connectivity centred around hubs such as
the thalamus and DMN. With the caveat that longitu-
dinal studies remain rare, these changes seem to evolve
(as opposed to simply increase) over time and are linked
with clinical phenotype and disability. Network-based
studies also highlight the potential role of treatments
that functionally modify neural function rather than
those that structurally alter networks.
Published online 12 January 2021

1. Fornito, A. & Bullmore, E. T. Connectomics –
a new paradigm for understanding brain disease.
Eur. Neuropsychopharmacol. 25, 733–748
(2015).
2. Lassmann, H. Multiple sclerosis pathology.
Cold Spring Harb. Perspect. Med. 8, a028936
(2018).
3. Correale, J., Marrodan, M. & Benarroch, E. E. What is
the role of axonal ion channels in multiple sclerosis?
Neurology 95, 120–123 (2020).
4. Campbell, G., Licht-Mayer, S. & Mahad, D. Targeting
mitochondria to protect axons in progressive MS.
Neurosci. Lett. 710, 134258 (2019).
5. Lapointe, E., Li, D. K. B., Traboulsee, A. L. &
Rauscher, A. What have we learned from perfusion
MRI in multiple sclerosis? AJNR Am. J. Neuroradiol.
39, 994–1000 (2018).
6. Deuschl, G. et al. The burden of neurological
diseases in Europe: an analysis for the Global
Burden of Disease Study 2017. Lancet Public Health
5, e551–e567 (2020).
7. Barkhof, F. The clinico-radiological paradox in multiple
sclerosis revisited. Curr. Opin. Neurol. 15, 239–245
(2002).
8. Fox, R. J. et al. Phase 2 trial of ibudilast in progressive
multiple sclerosis. N. Engl. J. Med. 379, 846–855
(2018).
9. Sumowski, J. F. et al. Cognition in multiple sclerosis.
Neurology 90, 278–288 (2018).
10. McDonald, W. I., Miller, D. H. & Thompson, A. J.
Are magnetic resonance findings predictive of clinical
outcome in therapeutic trials in multiple sclerosis?
The dilemma of interferon-beta. Ann. Neurol. 36,
14–18 (1994).
11. Goodin, D. S. Magnetic resonance imaging as a
surrogate outcome measure of disability in multiple
sclerosis: have we been overly harsh in our assessment?
Ann. Neurol. 59, 597–605 (2006).
12. Kurtzke, J. F. Rating neurologic impairment in multiple
sclerosis: an expanded disability status scale (EDSS).
Neurology 33, 1444–1452 (1983).
13. Tintore, M. et al. Defining high, medium and low
impact prognostic factors for developing multiple
sclerosis. Brain 138, 1863–1874 (2015).
14. Ayache, S. S. & Chalah, M. A. Fatigue in multiple
sclerosis – insights into evaluation and management.
Neurophysiol. Clin. 47, 139–171 (2017).
15. Manjaly, Z. M. et al. Pathophysiological and
cognitive mechanisms of fatigue in multiplesclerosis.
J. Neurol. Neurosurg. Psychiatry 90, 642–651
(2019).

NaTure RevIeWS | NEuROlOGy volume 17 | March 2021 | 181

Reviews
16. Bertoli, M. & Tecchio, F. Fatigue in multiple
sclerosis: does the functional or structural
damage prevail? Mult. Scler. https://doi.org/
10.1177/1352458520912175 (2020).
17. Martínez-Lapiscina, E. H. et al. The visual pathway
as a model to understand brain damage in multiple
sclerosis. Mult. Scler. 20, 1678–1685 (2014).
18. Trapp, B. D. et al. Axonal transection in the lesions
of multiple sclerosis. N. Engl. J. Med. 338, 278–285
(1998).
19. Bodini, B. et al. White and gray matter damage in
primary progressive MS: the chicken or the egg?
Neurology 86, 170–176 (2015).
20. Audoin, B. et al. Selective magnetization transfer ratio
decrease in the visual cortex following optic neuritis.
Brain 129, 1031–1039 (2006).
21. Singh, S. et al. Relationship of acute axonal damage,
Wallerian degeneration, and clinical disability in
multiple sclerosis. J. Neuroinflammation 4, 57
(2017).
22. Allen, I. V., McQuaid, S., Mirakhur, M. & Nevin, G.
Pathological abnormalities in the normal-appearing
white matter in multiple sclerosis. Neurol. Sci. 22,
141–144 (2001).
23. Bø, L., Vedeler, C. A., Nyland, H. I., Trapp, B. D. &
Mørk, S. J. Subpial demyelination in the cerebral
cortex of multiple sclerosis patients. J. Neuropathol.
Exp. Neurol. 62, 723–732 (2003).
24. Jürgens, T. et al. Reconstruction of single cortical
projection neurons reveals primary spine loss in
multiple sclerosis. Brain 139, 39–46 (2016).
25. Magliozzi, R. et al. A Gradient of neuronal loss
and meningeal inflammation in multiple sclerosis.
Ann. Neurol. 68, 477–493 (2010).
26. Henry, R. G. et al. Regional grey matter atrophy
in clinically isolated syndromes at presentation.
J. Neurol. Neurosurg. Psychiatry 79, 1236–1244
(2008).
27. Cifelli, A. et al. Thalamic neurodegeneration in
multiple sclerosis. Ann. Neurol. 52, 650–653
(2002).
28. Bell, P. T. & Shine, J. M. Subcortical contributions
to large-scale network communication. Neurosci.
Biobehav. Rev. 71, 313–322 (2016).
29. Lin, F. et al. Altered nuclei-specific thalamic functional
connectivity patterns in multiple sclerosis and their
associations with fatigue and cognition. Mult. Scler.
25, 1243–1254 (2019).
30. Paling, D. et al. Cerebral arterial bolus arrival time is
prolonged in multiple sclerosis and associated with
disability. J. Cereb. Blood Flow. Metab. 34, 34–42
(2013).
31. Roostaei, T. et al. Channelopathy-related SCN10A
gene variants predict cerebellar dysfunction in
multiple sclerosis. Neurology 86, 410–417 (2016).
32. Desai, R. A. et al. Cause and prevention of
demyelination in a model multiple sclerosis lesion.
Ann. Neurol. 79, 591–604 (2016).
33. Fan, A. P. et al. Quantitative oxygen extraction
fraction from 7-Tesla MRI phase: reproducibility
and application in multiple sclerosis. J. Cereb. Blood
Flow. Metab. 35, 131–139 (2014).
34. Rocca, M. A. et al. Clinical and imaging assessment
of cognitive dysfunction in multiple sclerosis. Lancet
Neurol. 14, 302–317 (2015).
35. Roosendaal, S. D. et al. Grey matter volume in
a large cohort of MS patients: relation to MRI
parameters and disability. Mult. Scler. 17,
1098–1106 (2011).
36. Fisher, E., Lee, J.-C., Nakamura, K. & Rudick, R. A.
Gray matter atrophy in multiple sclerosis: a
longitudinal study. Ann. Neurol. 64, 255–265
(2008).
37. Filippi, M. et al. Gray matter damage predicts the
accumulation of disability 13 years later in MS.
Neurology 81, 1759–1767 (2013).
38. Eijlers, A. J. C. et al. Predicting cognitive decline in
multiple sclerosis: a 5-year follow-up study. Brain 141,
2605–2618 (2018).
39. Barkhof, F., Haller, S. & Rombouts, S. A. Resting-state
functional MR imaging: a new window to the brain.
Radiology 272, 29–49 (2014).
40. Alonso-Nanclares, L., Gonzalez-Soriano, J.,
Rodriguez, J. R. & DeFelipe, J. Gender differences
in human cortical synaptic density. Proc. Natl Acad.
Sci. USA 105, 14615–14619 (2008).
41. Carp, J. The secret lives of experiments: methods
reporting in the fMRI literature. Neuroimage 63,
289–300 (2012).
42. Puce, A. & Hämäläinen, M. A review of issues related
to data acquisition and analysis in EEG/MEG studies.
Brain Sci. 7, 58 (2017).
182 | March 2021 | volume 17 www.nature.com/nrneurol
43. Rubinov, M. & Sporns, O. Complex network measures
of brain connectivity: uses and interpretations.
Neuroimage 52, 1059–1069 (2010).
44. Schmierer, K. et al. Diffusion tensor imaging of post
mortem multiple sclerosis brain. Neuroimage 35,
467–477 (2007).
45. Schmierer, K., Scaravilli, F., Altmann, D. R.,
Barker, G. J. & Miller, D. H. Magnetization transfer
ratio and myelin in postmortem multiple sclerosis
brain. Ann. Neurol. 56, 407–415 (2004).
46. Thiebaut de Schotten, M. et al. From Phineas
Gage and Monsieur Leborgne to H.M.: revisiting
disconnection syndromes. Cereb. Cortex 25,
4812–4827 (2015).
47. Friston, K. J. et al. Psychophysiological and
modulatory interactions in neuroimaging. Neuroimage
6, 218–229 (1997).
48. McIntosh, A. R. & Gonzalez-Lima, F. Structural
modeling of functional neural pathways mapped with
2-deoxyglucose: effects of acoustic startle habituation
on the auditory system. Brain Res. 547, 295–302
(1991).
49. McIntosh, A. R. & Gonzalez-Lima, F. Structural
equation modeling and its application to network
analysis in functional brain imaging. Hum. Brain
Mapp. 2, 2–22 (1994).
50. Buchel, C. & Friston, K. J. Modulation of connectivity
in visual pathways by attention: cortical interactions
evaluated with structural equation modelling and
fMRI. Cereb. Cortex 7, 768–778 (1997).
51. Penny, W. D., Stephan, K. E., Mechelli, A. &
Friston, K. J. Modelling functional integration:
a comparison of structural equation and dynamic
causal models. Neuroimage 23 (Suppl. 1),
S264–S274 (2004).
52. Tijms, B. M., Series, P., Willshaw, D. J. & Lawrie, S. M.
Similarity-based extraction of individual networks
from gray matter MRI scans. Cereb. Cortex 22,
1530–1541 (2012).
53. Avena-Koenigsberger, A., Misic, B. & Sporns, O.
Communication dynamics in complex brain networks.
Nat. Rev. Neurosci. 19, 17–33 (2018).
54. Enzinger, C. et al. Longitudinal fMRI studies: exploring
brain plasticity and repair in MS. Mult. Scler. 22,
269–278 (2016).
55. Fleischer, V. et al. Increased structural white and
grey matter network connectivity compensates
for functional decline in early multiple sclerosis.
Mult. Scler. 23, 432–441 (2017).
56. Cope, E. C. & Gould, E. Adult neurogenesis, glia, and
the extracellular matrix. Cell Stem Cell 24, 690–705
(2019).
57. Pardini, M. et al. Motor network efficiency and
disability in multiple sclerosis. Neurology 85,
1115–1122 (2015).
58. Steenwijk, M. D. et al. Cortical atrophy patterns
in multiple sclerosis are non-random and clinically
relevant. Brain 139, 115–126 (2016).
59. Cercignani, M. & Gandini Wheeler-Kingshott, C. From
micro- to macro-structures in multiple sclerosis: what
is the added value of diffusion imaging. NMR Biomed.
32, e3888 (2019).
60. Chen, J. E., Rubinov, M. & Chang, C. Methods and
considerations for dynamic analysis of functional
MR imaging data. Neuroimaging Clin. N. Am. 27,
547–560 (2017).
61. Tewarie, P. et al. Disruption of structural and
functional networks in long-standing multiple sclerosis.
Hum. Brain Mapp. 35, 5946–5961 (2014).
62. Pantano, P., Petsas, N., Tona, F. & Sbardella, E.
The role of fMRI to assess plasticity of the motor
system in MS. Front. Neurol. 6, 55 (2015).
63. Roosendaal, S. D. et al. Resting state networks
change in clinically isolated syndrome. Brain 133,
1612–1621 (2010).
64. Faivre, A. et al. Assessing brain connectivity at
rest is clinically relevant in early multiple sclerosis.
Mult. Scler. 18, 1251–1258 (2012).
65. Rocca, M. A. et al. Functional and structural
connectivity of the motor network in pediatric and
adult-onset relapsing-remitting multiple sclerosis.
Radiology 254, 541–550 (2010).
66. Rocca, M. A. et al. Functional network connectivity
abnormalities in multiple sclerosis: Correlations with
disability and cognitive impairment. Mult. Scler. 24,
459–471 (2018).
67. Liu, Y. et al. Functional brain network alterations in
clinically isolated syndrome and multiple sclerosis:
a graph-based connectome study. Radiology 282,
534–541 (2017).
68. Faivre, A. et al. Depletion of brain functional
connectivity enhancement leads to disability
progression in multiple sclerosis: a longitudinal
resting-state fMRI study. Mult. Scler. 22, 1695–1708
(2016).
69. Eijlers, A. J. C. et al. Reduced network dynamics on
functional mri signals cognitive impairment in multiple
sclerosis. Radiology 292, 449–457 (2019).
70. Bisecco, A. et al. Fatigue in multiple sclerosis: The
contribution of resting-state functional connectivity
reorganization. Mult. Scler. 24, 1696–1705 (2018).
71. Schoonheim, M. M., Meijer, K. A. & Geurts, J. J. G.
Network collapse and cognitive impairment in multiple
sclerosis. Front. Neurol. 6, 82 (2015).
72. Kipp, M. et al. Thalamus pathology in multiple
sclerosis: from biology to clinical application.
Cell Mol. Life Sci. 72, 1127–1147 (2014).
73. Castellazzi, G. et al. Functional connectivity alterations
reveal complex mechanisms based on clinical and
radiological status in mild relapsing remitting multiple
sclerosis. Front. Neurol. 9, 690 (2018).
74. Schoonheim, M. M. et al. Changes in functional
network centrality underlie cognitive dysfunction and
physical disability in multiple sclerosis. Mult. Scler. 20,
1058–1065 (2014).
75. Schoonheim, M. M. et al. Thalamus structure and
function determine severity of cognitive impairment
in multiple sclerosis. Neurology 84, 776–783
(2015).
76. Tona, F. et al. Multiple sclerosis: altered thalamic
resting-state functional connectivity and its effect
on cognitive function. Radiology 271, 814–821
(2014).
77. Hidalgo de la Cruz, M. et al. Abnormal functional
connectivity of thalamic sub-regions contributes
to fatigue in multiple sclerosis. Mult. Scler. 24,
1183–1195 (2018).
78. d’Ambrosio, A. et al. Structural connectivity-defined
thalamic subregions have different functional
connectivity abnormalities in multiple sclerosis
patients: Implications for clinical correlations.
Hum. Brain Mapp. 38, 6005–6018 (2017).
79. Jaeger, S. et al. Multiple sclerosis-related fatigue:
altered resting-state functional connectivity of the
ventral striatum and dorsolateral prefrontal cortex.
Mult. Scler. 25, 554–564 (2019).
80. Meijer, K. A., Eijlers, A. J. C., Geurts, J. J. G. &
Schoonheim, M. M. Staging of cortical and deep
grey matter functional connectivity changes in
multiple sclerosis. J. Neurol. Neurosurg. Psychiatr.
89, 205–210 (2018).
81. Lansley, J., Mataix-Cols, D., Grau, M., Radua, J. &
Sastre-Garriga, J. Localized grey matter atrophy in
multiple sclerosis: A meta-analysis of voxel-based
morphometry studies and associations with functional
disability. Neurosci. Biobehav. Rev. 37, 819–830
(2013).
82. Eshaghi, A. et al. Progression of regional grey matter
atrophy in multiple sclerosis. Brain 141, 1665–1677
(2018).
83. Rimkus, C. M. et al. Gray matter networks and
cognitive impairment in multiple sclerosis. Mult. Scler.
25, 382–391 (2019).
84. Tur, C. et al. Clinical relevance of cortical network
dynamics in early primary progressive MS. Mult. Scler.
26, 442–456 (2020).
85. Shu, N. et al. Diffusion tensor tractography reveals
disrupted topological efficiency in white matter
structural networks in multiple sclerosis. Cereb. Cortex
21, 2565–2577 (2011).
86. Pardini, M. et al. Cingulum bundle alterations underlie
subjective fatigue in multiple sclerosis. Mult. Scler. 21,
442–447 (2015).
87. Ciccarelli, O. et al. Optic radiation changes after
optic neuritis detected by tractography-based
group mapping. Hum. Brain Mapp. 25, 308–316
(2005).
88. Gabilondo, I. et al. Retrograde retinal damage after
acute optic tract lesion in MS. J. Neurol. Neurosurg.
Psychiatry 84, 824–826 (2013).
89. Rombouts, S. A. et al. Visual activation patterns in
patients with optic neuritis: an fMRI pilot study.
Neurology 50, 1896–1899 (1998).
90. Gareau, P. J. et al. Reduced visual evoked responses
in multiple sclerosis patients with optic neuritis:
comparison of functional magnetic resonance imaging
and visual evoked potentials. Mult. Scler. 5, 161–164
(1999).
91. Toosy, A. T. et al. Adaptive cortical plasticity in higher
visual areas after acute optic neuritis. Ann. Neurol.
57, 622–633 (2005).
92. Korsholm, K. et al. Recovery from optic neuritis:
an ROI-based analysis of LGN and visual cortical
areas. Brain 130, 1244–1253 (2007).

93. Jenkins, T. et al. Dissecting structure–function
interactions in acute optic neuritis to investigate
neuroplasticity. Hum. Brain Mapp. 31, 276–286
(2010).
94. Backner, Y. et al. Anatomical wiring and functional
networking changes in the visual system following
optic neuritis. JAMA Neurol. 75, 287–295 (2018).
95. Gallo, A. et al. Visual resting-state network in
relapsing-remitting MS with and without previous
optic neuritis. Neurology 79, 1458–1465 (2012).
96. Koini, M. et al. Correlates of executive functions in
multiple sclerosis based on structural and functional
MR imaging: insights from a multicenter study.
Radiology 280, 869–879 (2016).
97. Meijer, K. A. et al. Is impaired information processing
speed a matter of structural or functional damage in
MS? Neuroimage Clin. 20, 844–850 (2018).
98. Liu, Y. et al. Disrupted module efficiency of structural
and functional brain connectomes in clinically
isolated syndrome and multiple sclerosis. Front. Hum.
Neurosci. 12, 138 (2018).
99. Dineen, R. A. et al. Disconnection as a mechanism for
cognitive dysfunction in multiple sclerosis. Brain 132,
239–249 (2009).
100. Mesaros, S. et al. Diffusion tensor MRI tractography
and cognitive impairment in multiple sclerosis.
Neurology 78, 969–975 (2012).
101. Tewarie, P. et al. Explaining the heterogeneity of
functional connectivity findings in multiple sclerosis:
an empirically informed modeling study. Hum. Brain
Mapp. 39, 2541–2548 (2018).
102. Rocca, M. A. et al. Abnormal connectivity of
the sensorimotor network in patients with MS:
a multicenter fMRI study. Hum. Brain Mapp. 30,
2412–2425 (2009).
103. Sumowski, J. F. et al. Brain reserve and cognitive
reserve protect against cognitive decline over
4.5 years in MS. Neurology 82, 1776–1783 (2014).
104. Cordani, C. et al. Imaging correlates of hand motor
performance in multiple sclerosis: a multiparametric
structural and functional MRI study. Mult. Scler. 26,
233–244 (2020).
105. Goodman, A. D. et al. Sustained-release oral
fampridine in multiple sclerosis: a randomised,
double-blind, controlled trial. Lancet 373, 732–738
(2009).
106. Mainero, C. et al. Enhanced brain motor
activity in patients with MS after a single dose of
3,4-diaminopyridine. Neurology 62, 2044–2050
(2004).
107. Cader, S., Palace, J. & Matthews, P. M. Cholinergic
agonism alters cognitive processing and enhances
brain functional connectivity in patients with multiple
sclerosis. J. Psychopharmacol. 23, 686–696 (2009).
108. Fuchs, T. A. et al. Preserved network functional
connectivity underlies cognitive reserve in multiple
sclerosis. Hum. Brain Mapp. 40, 5231–5241 (2019).
109. van Geest, Q. et al. The importance of hippocampal
dynamic connectivity in explaining memory function
in multiple sclerosis. Brain Behav. 8, e00954 (2018).
110. van Geest, Q. et al. Information processing speed in
multiple sclerosis: Relevance of default mode network
dynamics. Neuroimage Clin. 19, 507–515 (2018).
111. Lin, S.-J. et al. Education, and the balance between
dynamic and stationary functional connectivity jointly
support executive functions in relapsing-remitting
multiple sclerosis. Hum. Brain Mapp. 39, 5039–5049
(2018).
112. Bosma, R. L. et al. Dynamic pain connectome
functional connectivity and oscillations reflect
multiple sclerosis pain. Pain 159, 2267–2276
(2018).
113. Llufriu, S. et al. Structural networks involved in
attention and executive functions in multiple sclerosis.
Neuroimage Clin. 13, 288–296 (2017).
114. Pagani, E. et al. Structural connectivity in multiple
sclerosis and modeling of disconnection. Mult. Scler.
26, 220–232 (2020).
115. Prosperini, L., Piattella, M. C., Giannì, C. & Pantano, P.
Functional and structural brain plasticity enhanced
by motor and cognitive rehabilitation in multiple
sclerosis. Neural Plast. 2015, 481574 (2015).
116. Filippi, M. et al. Multiple sclerosis: effects of cognitive
rehabilitation on structural and functional MR imaging
measures–an explorative study. Radiology 262,
932–940 (2012).
117. Gaede, G. et al. Safety and preliminary efficacy of
deep transcranial magnetic stimulation in MS-related
fatigue. Neurol. Neuroimmunol. Neuroinflamm. 5,
e423 (2017).
118. Hulst, H. E. et al. rTMS affects working memory
performance, brain activation and functional
NaTure RevIeWS | NEuROlOGy
connectivity in patients with multiple sclerosis.
J. Neurol. Neurosurg. Psychiatry 88, 386–394
(2017).
119. Boutière, C. et al. Improvement of spasticity following
intermittent theta burst stimulation in multiple
sclerosis is associated with modulation of resting-state
functional connectivity of the primary motor cortices.
Mult. Scler. 23, 855–863 (2017).
120. van den Heuvel, M. P., Bullmore, E. T. & Sporns, O.
Comparative connectomics. Trends Cogn. Sci. 20,
345–361 (2016).
121. Sha, Z. et al. Meta-connectomic analysis reveals
commonly disrupted functional architectures in
network modules and connectors across brain
disorders. Cereb. Cortex 28, 4179–4194 (2018).
122. Covey, J. T. et al. Improved cognitive performance
and event-related potential changes following working
memory training in patients with multiple sclerosis.
Mult. Scler. J. Exp. Transl. Clin. 4,
2055217317747626 (2018).
123. D’Angelo, E. & Gandini Wheeler-Kingshott, C.
Modelling the brain: elementary components to
explain ensemble functions. Riv. Nuovo Cimento 40,
297–333 (2017).
124. Kiljan, S. et al. Structural network topology relates to
tissue properties in multiple sclerosis. J. Neurol. 266,
212–222 (2019).
125. Chard, D. T. & Miller, D. H. What lies beneath grey
matter atrophy in multiple sclerosis? Brain 139,
7–10 (2016).
126. Warren, J. D. et al. Molecular nexopathies:
a new paradigm of neurodegenerative disease.
Trends Neurosci. 36, 561–569 (2013).
127. Benedict, R. H. B., Amato, M. P., John DeLuca, J.
& Geurts, J. J. G. Cognitive impairment in multiple
sclerosis: clinical management, MRI, and therapeutic
avenues. Lancet Neurol. 19, 860–871 (2020).
128. Wegner, C. et al. Relating functional changes during
hand movement to clinical parameters in patients
with multiple sclerosis in a multi-centre fMRI study.
Eur. J. Neurol. 15, 113–122 (2008).
129. Manson, S. C. et al. Impairment of movement-
associated brain deactivation in multiple sclerosis:
further evidence for a functional pathology of
interhemispheric neuronal inhibition. Exp. Brain Res.
187, 25–31 (2008).
130. Mancini, L. et al. Short-term adaptation to a simple
motor task: a physiological process preserved in
multiple sclerosis. Neuroimage 45, 500–511
(2009).
131. Colorado, R. A., Shukla, K., Zhou, Y., Wolinsky, J. S.
& Narayana, P. A. Multi-task functional MRI in
multiple sclerosis patients without clinical disability.
Neuroimage 59, 573–581 (2012).
132. Rocca, M. A. et al. Abnormal adaptation over time
of motor network recruitment in multiple sclerosis
patients with fatigue. Mult. Scler. 22, 1144–1153
(2016).
133. Rocca, M. A. et al. Large-scale neuronal network
dysfunction in relapsing-remitting multiple sclerosis.
Neurology 79, 1449–1457 (2012).
134. Rocca, M. A. et al. Hippocampal-DMN disconnectivity
in MS is related to WM lesions and depression.
Hum. Brain Mapp. 36, 5051–5063 (2015).
135. Rocca, M. A. et al. Impaired functional integration in
multiple sclerosis: a graph theory study. Brain Struct.
Funct. 221, 115–131 (2016).
136. Eijlers, A. J. et al. Increased default-mode network
centrality in cognitively impaired multiple sclerosis
patients. Neurology 88, 952–960 (2017).
137. Meijer, K. A. et al. Increased connectivity of hub
networks and cognitive impairment in multiple
sclerosis. Neurology 88, 2107–2114 (2017).
138. Tommasin, S. et al. Relation between functional
connectivity and disability in multiple sclerosis:
a non-linear model. J. Neurol. 265, 2881–2892
(2018).
139. Thirion, B. et al. Analysis of a large fMRI cohort:
statistical and methodological issues for group
analyses. Neuroimage 35, 105–120 (2007).
140. Chen, X., Lu, B. & Yan, C.-G. Reproducibility of
R-fMRI metrics on the impact of different strategies
for multiple comparison correction and sample sizes.
Hum. Brain Mapp. 39, 300–318 (2018).
Acknowledgements
This work arose out of a Magnetic Resonance Imaging in MS
(MAGNIMS) workshop on brain connectivity and networks in
multiple sclerosis. The workshop was supported by the
Multiple Sclerosis Society UK and Novartis, but they had no
involvement in the workshop programme or in the writing of
this manuscript.
Reviews
Author contributions
D. T. C., A. A. S. A., B. A., T. C., C. E., H. E. H., M.A.R., J. S.-G.,
B. T., C. T. and A. M. W. researched data for the article.
All authors made a substantial contribution to discussion of
content, wrote the article, and reviewed and/or edited the
manuscript before submission.
Competing interests
Over the last 3 years, D. C. has received honoraria (paid to
his employer) from Excemed for faculty-led education work.
He is a consultant for Biogen and Hoffmann-La Roche. He has
received research funding from the International Progressive
MS Alliance, the MS Society UK, and the National Institute
for Health Research (NIHR) University College London
Hospitals (UCLH) Biomedical Research Centre. A. B. reports
travel grants from Biogen France SAS, Genzyme, Novartis
Pharma SAS, Teva Santé SAS. C. E. has received funding for
travel and speaker honoraria from Biogen, Bayer Schering,
Celgene, Genzyme, Merck, Novartis, Roche, and Teva
Pharmaceutical Industries Ltd/Sanofi-Aventis, received
re s e a rc h s u p p o r t f ro m B i o g e n , M e rc k , a n d Teva
Pharmaceutical Industries Ltd/Sanofi-Aventis, and serves on
scientific advisory boards for Bayer, Biogen, Genzyme, Merck,
Novartis, Roche, and Teva Pharmaceutical Industries Ltd/
Sanofi-Aventis. H. E. H. has received compensation for con-
sulting services or speaker honoraria from Biogen Idec,
Celgene, Merck Serono, and Sanofi Genzyme and serves on
the editorial board of the Multiple Sclerosis Journal. M. A. R.
has received speaker’s honoraria from Bayer, Biogen Idec,
Calgene, Genzyme, Merck Serono, Novartis, Roche and Teva,
and receives research support from the Italian Ministry of
Health and Fondazione Italiana Sclerosi Multipla. A. R. serves
as Editorial Board member of Neuroradiology and American
Journal of Neuroradiology, on scientific advisory boards for
Bayer, Biogen, Novartis, OLEA Medical, Sanofi Genzyme,
SyntheticMR and Roche, and has received speaker honoraria
from Bayer, Biogen, Merck Serono, Novartis, Roche, Sanofi
Genzyme and Teva Pharmaceutical Industries Ltd. J. S.-G.
reports grants and personal fees from Genzyme received over
the last 36 months and personal fees from Almirall, Bial,
Biogen, Celgene, Merck, Novartis, Roche and Teva; he is
Director of Revista de Neurologia, for which he does not
receive any compensation, and serves as member of the
Editorial Board of Multiple Sclerosis Journal, for which he
receives compensation. M. M. S. serves on the Editorial
Board of Frontiers in Neurology and has received compensa-
tion for consulting services or speaker honoraria from Biogen,
ExceMed and Genzyme. B. T. received funding from the
ZonMW Memorabel grant programme #73305056. C. T. has
received a postdoctoral research ECTRIMS fellowship (2015);
she has also received honoraria and support for travelling
from Bayer, Biogen, Ismar Healthcare, Merck Serono,
Novartis, Roche, Sanofi and Teva Pharmaceuticals and pro-
vides consultancy services to Roche. C. G. W.-K. reports
receiving research funding from the International Spinal
Research Trust, the Craig H. Neilsen Foundation (the
INSPIRED study), the MS Society (#77), Wings for Life
(the INSPIRED study, #169111) and Horizon 2020
(CDS-QUAMRI, #634541). A. M. W. receives funding from
the European Prevention of Alzheimer’s Dementia consor-
tium, the Amyloid Imaging to Prevent Alzheimer’s Disease
initiative (Innovative Medicines Initiative grants 115736 and
115962) and the European Progression of Neurological
Disease Initiative (Horizon 2020 grant 666992). O. C. serves
as a consultant for Merck, Novartis, and Roche; she receives
an honorarium from the American Academy of Neurology as
Associate Editor of Neurology. F. B. serves as an Editorial
Board member of Brain, European Radiology, Neurology,
Multiple Sclerosis Journal and Radiology; he has accepted
consulting fees from Apitope Ltd, Bayer-Schering Pharma,
Biogen-IDEC, GeNeuro, Sanofi Genzyme, IXICO Ltd, Jansen
Research, Merck Serono, Novartis, Roche, and TEVA and
speaker fees from Biogen-IDEC and IXICO. He has received
grants from the Amyloid Imaging to Prevent Alzheimer’s
Disease Initiative (Innovative Medicines Initiative), the
European Progression of Neurological Disease Initiative
(H2020), UK MS Society, Dutch MS Society, NIHR University
College London Hospital Biomedical Research Centre, the
European Committee for Treatment and Research in Multiple
Sclerosis and the Magnetic Resonance Imaging in MS
network. The other authors declare no competing interests.
Peer review information
Nature Reviews Neurology thanks B. Weinstock and the
other, anonymous, reviewer(s) for their contribution to
the peer review of this work.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
© Springer Nature Limited 2021
volume 17 | March 2021 | 183
Reviews

the MAGNIMS Study Group

F. Barkhof2,16,17, O. Ciccarelli1,2, N. De Stefano18, C. Enzinger6,7, M. Filippi9, M. A. Rocca9, J. L. Frederiksen19, C. Gasperini20, L. Kappos21, J. Palace22, A. Rovira10,
J. Sastre-Garriga11, T. Yousry23,24 and H. Vrenken16

Department of Neurological and Behavioural Sciences, University of Siena, Siena, Italy. 19Department of Neurology, Glostrup University Hospital Copenhagen, Copenhagen, Denmark.
18

Department of Neurosciences, San Camillo Forlanini Hospital, Rome, Italy. 21Department of Neurology, University Hospital, Kantonsspital, Basel, Switzerland. 22Nuffield Department of Clinical
20

Neurology, University of Oxford, Oxford, UK. 23Lysholm Department of Neuroradiology, National Hospital for Neurology and Neurosurgery, London, UK. 24Neuroradiological Academic Unit,
Department of Brain Repair and Rehabilitation, UCL Queen Square Institute of Neurology, University College London, London, UK.

184 | March 2021 | volume 17 www.nature.com/nrneurol