Titel;

mcDESPOT-Derived Demyelination Volume in Multiple Sclerosis Patients

Correlates with Clinical Disability and Senses Early Myelin Loss

H. H. Kitzler1,3* , J. Su2* , M. Zeineh2, C. Harper-Little3, A. Leung4, M.

Kremenchutzky5, S. C. Deoni6, and B. K. Rutt2

1 Department of Neuroadiology, Technische Universitaet Dresden, Dresden, Germany

2 Department of Radiology, Stanford University, Stanford, California, USA

3 Robarts Research Institute, University of Western Ontario, London, Ontario, Canada,

4 Department of Diagnostic Radiology and Nuclear Medicine, University of Western

Ontario, London, Ontario, Canada

5 Department of Clinical Neurological Sciences, University of Western Ontario, London,

Ontario, Canada

6 Department of Engineering, Brown University, Providence, Rhode Island, USA

(*both authors contributed equally to this work)

Running title: Whole-Brain Demyelination Quantification in MS

Total word count (text body): xxxx

Summary (400 words max; current state: 400) LEFT UNTOUCHED, LEAST TO

BE REWRITTEN AS SUGGESTED!

Conventional magnetic resonance (MRMRI) imaging is established as one of the most important surrogate

markers of Multiple Sclerosis (MS) development and treatment outcome. Based on the assumption that the

clinical course of MS is adequately reflected by focal white matter changes, many clinical trials have used lesion

volume as the principal MRMRI-derived measure; however, such measures have been recently criticized as

adding little or no independent information over and above non-imaging disability outcome measurements when

evaluated retrospectively. [reference required here?] This highlights the need to develop and validate new

quantitative WM imaging strategies that aim to characterize the invisible burden of demyelination in the brain and

establish much more sensitive and specific markers of MS that correlate strongly with clinical disability. One of

the most promising of such arising measures is myelin-selective MR imaging MRI that allows the acquisition of

Myelin-Water fraction (MWF) maps, a parameter that is correlated to the brain white matter (WM) myelination. The

aim of our study was to apply the newest myelin-selective MRMRI method, multi-component Driven Equilibrium

Single Pulse Observation of T1 and T2 (mcDESPOT) in a controlled clinical MS pilot trial. This study was designed

to assess the capabilities of this new method to explain differences in disease course and degree of disability in

subjects spanning a broad spectrum of MS disease severity. The whole-brain isotropically-resolved 3D acquisition

capability of mcDESPOT allowed for the first time the registration of 3D MWF maps to standard space, and

consequently a formalized voxel based analysis (VBA) of the data. This VBA approach combined with image

segmentation further allowed the derivation of new volumetric measures of disease severity: total demyelinated

volume (DV) in WM, DV within WM lesions, DV within dirty appearing white matter (DAWM) and DV within normal

appearing white matter (NAWM). The analysis confirmed that neither lesion burden nor lesion demyelination

correlate well with clinical disease activity measured with the extended disability status scale (EDSS) in MS

patients. In contrast, our measurements of demyelination volume in NAWM correlated significantly with the

EDSS score (R2 0.405; p<0.01). [will need to update numbers, plus decide what other data is going into this paper and to

summarize all the key results in this section here][Our most remarkable result is differentiating CIS from normals. There is

some correlation of EDSS with DV but not enough to say “wow.” The other question is whether to include the multi-variate

regression analysis.] The same measurement discriminated Clinically Isolated Syndrome (CIS) patients from a

normal control population (p<0.001), hence the technique senses very early disease-related myelin loss.

Furthermore, the same parameter discriminated patients with the secondary-progressive (SPMS) course from

relapsing-remitting MS (RRMS; p<0.01). Overall, our results demonstrate that mcDESPOT-defined demyelination

measurements show great promise to act as imaging markers of clinical disease activity in MS. Further

investigation will determine if this measure can serve as a risk factor for the conversion into definite MS and for

the secondary transition into irreversible disease progression.

Keywords: multiple sclerosis; demyelination; normal appearing white matter;

quantitative MRIMRI; myelin-selective imaging

Introduction

Multiple Sclerosis and Imaging Conventional MRI

Multiple Sclerosis (MS) is an immunologically mediated demyelinating and

axonal disease of the human central nervous system (CNS) and. It is one of the

most common disabling neurological diseases in young people adults with the

typical age-at-onset being 20 to 40 years and it is approximately twice as

common in women as in men [Platten 2006]. Depending on the location of CNS

lesions, [Hagen or others: I simplified your wording, but need to know if what I

wrote is still strictly correct] MS patients experience diverse neurological

symptoms and impairment of e.g. motor, visual, or, sensory function. Over 80%

of MS patients initially present with a relapsing disease course that eventually

transitions into permanent disability. More than 50% of patients require a walking

aid within 15 years from initial diagnosis [Weinshanker 1989].

The MS etiology [can we just say “the cause of MS”? Or is etiology a standard

term in medicine?] is believed to be the result of a complex combination of

environmental, genetic, and autoimmune factors resulting in an immune-

mediated attack on CNS myelin, [Steinman 2004]. Myelin is the basic structure

of the myelin axonal sheath of , surrounding neuronal axons cells which isand

vital for their appropriate function [Steinman 2004]. The pathology of MS ,

however, lesions is heterogeneous, including inflammatory cell infiltration,

astroglial hypertrophy, axonal loss, but demyelination is the recognized hallmark

of the disease [Lucchinetti 2001].

The application of conventional magnetic resonance imaging (MRI) techniques

has revolutionized the clinical practice in MS [Villenga 2009]. Conventional

magnetic resonance (MRMRI) imaging studies reveal focal signal deviations

lesions, traditionally called lesions or “plaques” within throughout the white matter

(WM) and less frequently also in grey matter (GM) of the brain and spinal cord in

both T2 and T1 weighted scans. This appearance of lesions on top of the

background of an inflammatory reaction throughout the central nervous system

(CNS) defineeds MS early as a multifocal inflammatory demyelinating disease.

Conventional MRI derived lesion numeric and volumetric measures are currently

used as paraclinical markers in standardized diagnostic schemes [Polman,

2005]. However, thee lesion-centered view has been challenged by studies

investigating the relationship between conventional MRI measures with clinical

MS disease severity and neuro-functional scores revealing only dissatisfying and

non-significant correlations [Fulton, 1999, more ref]. Despite the acknowledged

potential application of conventional MRI, presently available MRI technologies

have failed in meeting the critical goal of reflecting MS patient disability status

and predicting disease progression. [ref]

Such conventional MRI measures also have been widely employed as

presumptive surrogates across a broad spectrum of MS studies, ranging from

pilot trials through multi-center pivotal phase III studies. However, the presently

employed metrics fail to meet Prentice’s surrogate endpoint validation criteria for

reliable surrogate markers in predicting downstream disease activity [Prentice,

1989]; and statistically appear to offer no more valid endpoint than that already

offered by clinical outcomes of MS clinical disability as measured by the

extended disability status scale (EDSS) and the relapse rate [Daumer

2009].novel quantitative MR technologies, which have provided insights into

partial pathological aspects [not sure what you mean by this phrase "partial

pathological aspects"] of the disease. It is now thought [known?] that primary

demyelination, i.e. selective myelin destruction, is not restricted to focal MS

lesions but occurs throughout the entire CNS parenchyma. Moreover such

demyelination may be accompanied to a variable degree by remyelination and

repair. [should add strategic or key references to some of the above statements]

The current state-of-the-art treatments are disease-modifying agents that at

present are able to decrease relapse rates by 30% [Weiner 2009]. However,

despite these advances, the field of MS still lacks specific markers to predict

clinical relapses and disease progression. Novel immunotherapies are on the

rise, but to date non-invasive technologies have failed to provide accurate,

reliable tools to assess the state of myelination. Such technologies are

particularly needed for testing drug efficacy or for monitoring treatment.

Imaging technologies provide potential instruments to investigate in vivo, real

time changes that occur within the CNS over the broad spectrum of natural MS

courses as well as during treatment. The application of conventional MR imaging

techniques has already revolutionized the clinical practice in MS [Villenga 2009].

Conventional MR imaging derived numeric and volumetric measures are

currently used as paraclinical markers in standardized diagnostic schemes

[Polman, 2005]. However, despite the widely acknowledged potential application

of MRI, presently available MR technologies have failed in meeting the critical

goal of predicting disease progression and MS patient disability status. [ref?]

Conventional MR imaging measures also have been widely employed as

presumptive surrogates across a broad spectrum of MS studies, ranging

from pilot trials through multi-center pivotal phase III studies. However, the

presently employed metrics fail to meet Prentice’s surrogate endpoint

validation criteria for reliable surrogate markers in predicting downstream

disease activity [Prentice, 1989]; and statistically appear to offer no more

valid endpoint than that already offered by clinical outcomes of MS relapse

rate and clinical disability as measured by EDSS [Daumer 2009].

[Maybe a new sub-heading here?]Quantitative MRI and Imaging Myelin In

Vivo

In conventional MRI Imaging studies of MS patients, the WM tissue compartment

that does not clearly show lesions or abnormalities is referred to as the Normal

Appearing White Matter (NAWM) compartment. Moreover, lesions are not always

well-defined areas of MR signal change with sharp boundaries, but often present

ill-defined surrounding regions of signal deviation, [I don't like this term

circumjacent signal deviation: can you try something simpler / clearer? I've made

one suggestion.] the so called Dirty Appearing White Matter (DAWM). Outside of

the lesions or DAWM, [do you mean in NAWM here? If so, just say so more

clearly] a number of modern Newer, unconventional quantitative MRMRI are

often aimed at the derivation of more specific and quantifiable information about

MS pathology and its distribution. Quantitative in nature, Tthese quantitative MRI

technologies have observed alteration in parameters that may be related to

intrinsic tissue integrity myelination and axonal integrity myelination, indicative of

a process of diffuse myelin damage and axonalneuronal loss not restricted to

lesion tissue but throughout the entire CNS parenchyma [Seewann, 2009;

Vrenken, 2010].

Newer, unconventional MR imaging strategies are often aimed at the derivation

of more specific information about MS pathology and its distribution.

Progressive changes of intrinsic NAWM microstructure related to the tissue water

diffusion characteristics were detected in primary-progressive MS (PPMS)

patients in serial diffusion MR imaging study that quantified the apparent diffusion

coefficient (ADC) [Schmierer 2006]. Werring et al. investigated the dynamic

evolution of water diffusion measurements in pre-lesion NAWM in another serial

diffusion MR imaging study and found a steady and moderate increase in ADC,

followed by a rapid and marked increase at the time of lesion formation, and

even a significant but milder increase in matched NAWM regions [what does this

mean, "matched NAWM regions"?] [Werring, 2000].

Widespread tissue changes are found in NAWM of MS patients by measuring the

magnetization transfer ratio (MTR). Those changes are mainly explained in terms

of axonal damage and loss of one of the major pathological features of multiple

sclerosis [Filippi, 1998]. [I don't understand the last half of this sentence: what is

the major pathological feature you are referring to? Rewrite to make it clear] A

histological analysis of the substrate of those imaging findings revealed that not

only MTR but also T1 contrast ratio correlated strongly with axonal density, even

in NAWM. However, defining T2 lesions revealed no correlation but a range of

pathology, illustrating the low specificity of T2-weighted imaging [van

Waesberghe, 1999]. Early axonal pathology, can also be quantified with Proton

(H+) spectroscopy (S) that provides chemical composition information at the

level of metabolites. Early S studies have noted specific changes in metabolite

signatures, not only within focal T2 lesions but even a deviation from normal in

NAWM areas [Helms 2000]. A measure of ‘whole-brain’ N-acetylaspartate

(WBNAA), a marker of axonal integrity, in particular confirmed widespread axonal

pathology, largely independent of -visible inflammation in MS patients even in

Clinically Isolated Syndrome. No correlation however, was found between the T2

lesion volumes and WBNAA concentrations [is this really a concentration value

or a total integrated NAA value?] [Filippi, 2003].

Axons and their myelin sheath form an individually customized unit. [This English

doesn't make sense. I'm not even sure what you mean here. Do you mean that

the phrase "axonal loss" implies both dymelination and axonal degradation? This

definitely needs to be re-worded.] However, axonal loss is not necessarily

accompanied by demyelination, moreover both histopathologic changes seem to

contribute independently to the appearance in conventional MR imaging scans.

An imaging-histopathology case study confirmed that axonal degeneration could

occur in the absence of myelin loss as a histopathologic correlate to abnormal

MR findings in MS patients [Bjartmar, 2001]. [This last paragraph needs to be

improved. Hard to understand.]

Single component T1 relaxation time was found to be abnormal in NAWM in

established MS. When compared to MTR, quantitative T1 measurement was

more sensitive in detecting subtle pathological change. No correlation was found

between NAWM T1 changes and lesion abnormalities [what do you mean here:

lesion volume? lesion signal characteristics?] suggesting independent underlying

pathologic mechanisms [Griffin, 2002].

Fulton et al. determined the relationship between T2 lesion volume and both

neurocognitive and physical disability in untreated relapsing-remitting multiple

sclerosis. Despite some correlation to information-processing speed and verbal

long-term memory, none of ten other neurocognitive examinations or the physical

disability scales as rated according to the EDSS showed significant correlation

with total lesion volume. This challenges the view that lesion volume

measurement is a robust surrogate marker of impairment in patients with MS

[Fulton, 1999]. [Last sentence is awkward and hard to understand]

MTR vs disability correlation?

These studies point to a new direction for MS MR imaging research: to move

away from the lesion-centered view and to develop highly sensitive MR methods

that accurately and quantitatively reflect the global disease burden even in areas

that are apparently normal. Once such methods are developed, important

hypotheses can be tested; for example, that such measures reflect the subtle

underlying disease-determining pathology and will predict clinical changes in MS

disease development, as well as transition towards chronic progression.

Imaging Myelin In Vivo

It seems obvious that novel MS MR imaging MRI strategies should focus on CNS

tissue properties that are directly involved in specific disease-related pathological

processes, especially the process of demyelination. Direct quantitative myelin

assessment would have important application in a variety of inflammatory,

degenerative, and developmental disorders of the CNS as well as regeneration

from injury and trauma.

Current conventional MR imaging MRI methods do not specifically reflect myelin

content. Within the battery of unconventional MRMRI technologies,

magnetization transfer imaging (MTI), diffusion tensor imaging (DTI), and single-

component T1 and T2 relaxometry, i.e. the precise measurement of intrinsic

magnetic tissue properties, are all thought to provide information related to

myelin content. However, these measures are non-specific towards myelination.

While quantitative MTI provides an estimate of the macromolecule-bound water

fraction, this measure may also reflect inflammation processes [fix] with signal

originating from an aspect of pathology in MS other than demyelination

[Vavasour 1998, Gareau 2000]. Moreover a histological analysis of the substrate

MTI findings revealed a strong correlation with axonal density [van Waesberghe,

1999]. With regards to DTI, significant fractional anisotropy (FA) is observed

even in non-myelinated nerve tissue indicating that axonal structures may at

least in part be responsible for the signal generation [Beaulieu 2002]. [should

also mention the crossing fiber problem of DTI which is obviously unrelated to

myelination Moreover, the quantification of the integrity of myelinated WM fiber

tracts reflected by FA measures may be directly affecteddistorted in regions

where there are un-resolvable crossing fiber structures [Oochy 2007].

Finally, both T1 and T2 are influenced by a number of tissue structure and

biochemical characteristics, including free water content and the presence of

paramagnetic atoms such as iron.

Currently, multi-component relaxometric imaging (MCRI) provides the most direct

means of quantifying myelin volume in vivo. In conventional T2 MCRI, the

measured MRIMRI signal is decomposed into contributions from two or more

water pools, which in brain tissue are attributed to an intra and extra-cellular

water pool and water trapped between the hydrophobic bilayers of the myelin

sheath [Whittall 199789, Menon 1991 -> ref to be fixed]. [is it really true that

there are MCRI methods that use more than a two-pool model?] Through

appropriate data acquisition, typically comprising multiple spin-echo images

acquired over a range of echo times, and multi-exponential data analysis, maps

of the T2 characteristics and volume fractions of each water pool may be

estimated. As these volume fraction estimates show strong correlation with ‘gold-

standard’ histologic assessments [Webb 2003 -> ref to be fixed, Laule 2006],

MCRI has become the de facto standard for non-invasive myelin quantification.

Unfortunately, established MCRI methods require lengthy imaging times while

providing limited volume coverage. For example, the method of Whittall, MacKay

and colleagues [Whittall 1997, Mädler 2006 -> ref to be fixed, ISMRM

Proceedings?] requires approximately 16 minutes to acquire 16 contiguous slices

with a voxel volume of 10 mm3. These volume coverage, / spatial resolution,

and / imaging time characteristics are comparable to more recent alternative

techniques [Oh 2007 -> ref to be fixed] and make high resolution, whole-brain

investigations challenging.

An alternative to MCRI is image combination, in which imaging data acquired

with different acquisition parameters are combined so as to emphasize tissues

with specific T2 relaxation characteristics [Whittall 1991 -> ref to be fixed, Jones

20042004 -> ref to be fixed, Vidarsson 2005]. While multi-slice myelin fraction

maps may be estimated in as little as 5 minutes [Vidarsson 2005], these methods

are sensitive to T1 effects, depend on the short and long T2 selection criteria,

and can suffer from low signal-to-noise ratio (SNR) efficiency.

One of the most promising such applications is myelin-selective MR [Laule, 2007;

MacKay...add], however, these methods did not allow whole-brain high resolution

imaging in clinically practical scan times until very recently. [This paragraph

seems to stand out -- I think you've already mentioned multi-echo spin echo

methods with multi-exponential decomposition to create myelin water fraction

maps. If this is what you mean by "myelin-selective MR" then haven't you already

covered this?]

[ correlation to MTT, to characterize current practise in this area]

Multi-component Relaxometry

Multi-component driven equilibrium single pulse observation of T1/T2

(mcDESPOT) is the most recent MCRI a quantitative MR technique that allows

rapid acquisition of whole-brain data which is then processed to yield quantitative

two-pool parameters which can be used to derived quantitative myelin water

information. A series of spoiled gradient echo (SPGR) and phase-cycled steady-

state free precession (SSFP) volumetric scans are each collected each over a

broad range of flip angles (FA) at constant repetition times (TR) that allow the

extraction of a set of quantitative tissue parameters at every voxel over the whole

brain. One of these quantitative parameters is the Myelin Water Fraction (MWF),

that we and others have hypothesized to be proportional to the fraction of water

trapped between the myelin bilayers [ref-we!, ref-others!]. MWF is theoretically

specific to myelination and hitherto seems no evidence emergedt to bethat this

measure is mixed with other tissue signal components [ref!].

Extensive histologic validation has been performed validating the the method of

Whittall, MacKay and colleagues [Whittall 1997, Mädler 2006].

EXAMPLES TO FOLLOW….

[the word "seems" is weak.

WE NEED TO DISCUSS THE FOLLOWING:

I think you have to somehow sell the idea that the MWF we get from mcDESPOT

is the same quantity that McKay et al get from their T2 MCRI, and then state that

since that group has validated this MWF quantity so thoroughly that we believe

our MWF is therefore going to reflect the same strong correlation to actual myelin

content.]

We could compare ROI measurments in our data in greater fiber tracts/anatomic

regions to the data presented in:

[Is it true that mcDESPOT’s MWF looks more or less the same as T2 MCRI-

based MWF maps?] [I haven’t seen such a comparison, Brian did you?][Based

on our numbers in MSmcDESPOT-Changing to FLAIR, it would seem not. Mean

MWF in WM is around 20% for CIS.]

The novel feature of this acquisition technique is the fact that it covers the entire

brain producing isotropic datasets of the presumed myelin water fraction.

Cite other recent mcDESPOT applications here (Deoni, Kolind!)

To evaluateEvaluating specific white matter MR imaging MRI strategies like

mcDESPOT that selectively quantify myelin is of the major utmost clinical

neurological importance especially in the field of demyelinating diseases.

[“utmost” seems a little too strong for my taste, consider “major”] Therefore, the

specific aims of our study were (1) to derive myelin water fraction (MWF) maps

and our novel measurement of demyelinated volume (DV) using mcDESPOT in a

broad spectrum of MS patients, and (2) to test the hypothesis that MWF and/or

DV in normal appearing white matter (NAWM) correlates with disability in MS

thus reflecting non-lesional MS pathology that may determine disease. severity

[perhaps a mention of DV as well] Moreover, the study was driven by the

hypothesis that our new technique can quantify myelination at every voxel

throughout whole brain tissue, and therefore allows us to measure the degree of

demyelination anywhere in the brain, not just at lesion locations visible in

conventional MR data. [Somehow I think you need to state this a bit differently:

The novel feature of the mcDESPOT hereby acquisition technique is the fact that

it covers the entire brain producing isotropic datasets of the presumed myelin

water fraction. theThis whole-brain high-resolution isotropic 3D nature of the

mcDESPOT acquisition meansallows that myelin quantification canto be done

much more rigorously, using formalized voxel based analysis methods that have

found tremendous value in various voxel based morphometry applications, yet

which previous T2 MCRI / myelin mapping methods have really not been able to

achieve.

Thus rather subtle non-lesional involvement may determine disease activity and

may re-define MS as a rather global disease state that questions focal lesions in

their capability to reflect disease activity [severity?].

Materials and Methods

MS Patients & Healthy Controls

The MS group cohort (n=26) we assembled for this controlled clinical trial was

almost equally distributed across the different clinically definite MS types and the

accepted MS-precursor Clinically Isolated Syndrome (CIS): n=16 patients with

definite MS (relapsing-remitting [RRMS] n=5; secondary-progressive [SPMS]

n=6; and primary-progressive Multiple Sclerosis [PPMS] n=5) as well as n=10

patients with CIS (CIS (low risk [lr-CIS], n=5, and, high risk Clinically Isolated

Syndrome [hr-CIS], n=5).) The patient group age was (mean 49, min 19/max 68,

std. dev. ~12) Give male/male ratio (18/26 female). We also scanned an age-

matched group (n=26) of healthy controls [(HC]; n=26) (mean 42, min 23, max

66, std dev. ~12.6, 16/26 female) [don't think it was precisely age-matched] to be

used for formal statistical comparison (See demographic data for patients and

healthy controls at Tab. 1). All patients and healthy controlsparticipants were

recruited in accordance to local ethics board requirements of the University of

Western Ontario, London, Ontario, Canada. In all patients, we measured the MS

Extended Disability Status Scale (EDSS) score [ref! Kurtzke 1983], an average

scoring number derived from measures of various functions of the central

nervous system, using a scale from 0 to 10, with 10 representing greatest

disability . [is it a problem that we didn't measure EDSS in healthy controls?]

[Kurtzke 1983]. The average EDSS score for the patient group was 3.74.0 (max

8.50; min 0; std dev. 2.46). Within the entire group of patients only one lr-CIS

patient was treated with copaxone and one SPMS patient was treated with

copaxone or avonex, respectively. All other patients did not receive any immune

modulating treatment. [note that we didn’t use all the patients listed in the chart,

also be careful to use the “corrected” set of EDSS values where P001 = 2.0].

( ADD mean duration of disease, min/max, treatment, … ) [There should be a

Table that shows all relevant patient information.]

MRMRI Data Acquisition

Image acquisition used a clinical 1.5 T MRMRI scanner (GE Signa HDx, General

Electric Healthcare, Waukesha, WI) equipped with an 8-channel receive-only

radio-frequency (RF) brain array coil. We acquired isotropic nearly isotropic [1.7 x

1.7 x 2.0 mm] 3D whole-brain mcDESPOT [nearly isotropic 1.7x1.7x2mm] data

using the following imaging parameters: FOV = 22 cm, matrix = 128 x 128, slice

thickness = 1.7 mm; SPGR parameters: TE / TR = 2.1 / 6.7ms, α = {3, 4, 5, 6, 7,

8, 11, 13, 18}°; bSSFP parameters: TE / TR = 1.8 / 3.6 ms, α = {11, 14, 20, 24,

28, 34, 41, 51, 67}°, two phase cycles acquired per bSSFP flip angle. The total

mcDESPOT acquisition time was ~13 min allowing imaging within clinically-

relevant scan times. For anatomical reference and lesional tissue analysis, an

additional 2D-FLAIR sequence (TE / TR = 125 / 8800 ms, TI = 2200 ms, FOV =

22 cm, matrix = 256 x 256, slice thickness = 3 mm), as well as T1-MPRAGE

sequence (TE / TR = 3.8 / 9 ms, TI = 600 ms, FOV = 24 cm, matrix = 256 x

256, slice thickness = 1.2 mm). before and after the administration of contrast

media (Gadolinium-DTPA) was acquired. [If you aren't going to use the post-Gad

images for anything in this paper, then maybe don't bother even mentioning the

Gad injection.]

MRMRI Data Postprocessing

The processing of mcDESPOT data was accomplished using a custom MacPro

4.1 Xeon 64 Bit workstation [Jason: spell out the MacPro model correctly here]

[(2.66 GHz Quad-Core Intel Xeon, 6GB RAM) , MacPro 4.1 for macDESPOT,

2x2.8GHz Quad-Core Intel Xeon, 2GB RAM, MacPro3.1 for octopus] and

specialized in-house Python scripts to automate usage of the FMRIMRIB

Software Library (FSL) for brain extraction and intra-subject co-registration of

scans. [citeSmith et al.] [what’s the 2nd part about CNS parench. water dist?]. [not

just python scripts, since there are some underlying C programs right? Jason,

please write this up to correctly describe the software][how detailed should I be,

e.g. registration target is SPGR FA 18 deg., trilinear interpolation, specific

parameters used for FSL programs?] [maybe not down to specific parameters

used for FSL programs, but some kind of "block diagram" description.] what’s the

2nd part about CNS parench. water dist?]. The registered images are then

processed with the mcDESPOT multi-exponential fitting code, which uses

stochastic region of contraction to search for the model parameters [cite!]. This

produces 7 maps corresponding to T1 and T2 maps in the fast and slow relaxing

pools, off-resonance, residence time, and MWF. The myelin water fraction

(MWF) was obtained as the fractional short T2 component of the total T2

distribution and Myelin water fraction (MWF ) maps were derived from the data

for each subject from the mcDESPOT data [Deoni 2008].

Since the mcDESPOT method produces nearly isotropic 3D data over the entire

brain, any resulting quantitative map can be subsequently be warped to a brain

standard space. Thus, aAll MWF maps were non-linearly registered using FSL

to the MNI (Montreal Neurological Institute, Montreal, QC, Canada) [cite!]

standard brain space as defined by the non-linear MNI152 1 mm3 isotropic

resolution brain using FSL. A linear image registration with a common

framework of an iterative transformation process was used to overlay a base

image by a distorted floating image accurately. We used … [Jason][???, I have

no clue what this is about. A description of how FLIRT works?]

Subsequently, voxel-based analysis (VBA) was applied to these maps, whereby

mean and standard deviation volumes were computed from the healthy controls

MWF datamaps. On a patient-by-patient basis, each voxel's MWF value was

statistically compared to the healthy control distribution, to produce a z-score

value for that voxel. Those voxels that fell in the range of z-score < -4, i.e. that

had a MWF value at least 4 standard deviations below the mean healthy control

value, were marked and defined as significantly demyelinated .[Deoni 2008].

(more background VBA/VBM theory?) [this is a fairly intuitive idea proposed by

Sean, I’ve not seen it used anywhere else] Finally, these demyelinated voxels

were summed and scaled by the voxel volume to produce a total demyelinated

volume for each patient. We selected this measure as the main outcome

measure derived from our mcDESOT approach, and termed this quantity the

Demyelinated Volume (DV). This measure was computed for each subject,

regardless of whether they were an MS patient or normal control.

Tissue Compartment Segmentation

For subsequent voxel by voxel comparisons the conventional MRI data was

divided into fundamental WM tissue compartments for further analysis. Initially

bBrains were segmented into gray matter (GM), WM and cerebrospinal fluid

(CSF) mapps. First a selective WM map was obtained through probabilistic

segmentation from T1-weighted MPRAGE data via the Statistical Parametric

Mapping software package (SPM8; Welcome Department of Imaging

Neuroscience, UCL, London, UK) [cite!]. The probabilistic maps were converted

into binary masks by first median filtering to alleviate some errors due to grainy

noise in the MPRAGE scan and then thresholding at the 0.5 level. [is

“thresholding” too obscure a term?][no I think it is fine]… [Jason]. The resulting

WM masks were inspected by a trained radiologist and all selection below

pontine levels was excluded to standardize the analyzed brain parenchyma since

medulla oblongata and cervical spinal cord were unequally included in the

scanned volume between subjects. Some minor manual correction was applied

in areas of mixed GM and WM like the basal ganglia, in cerebrospinal plexus,

cortical vessels and the sellar region to exclude false positively selected areas

based on neuro-anatomical considerations.

Normalized Brain Volume (NBV) was also achieved as GM+WM volume

normalized by GM+WM+CSF volume [cite!]. [I thought we were going with the

term Parenchymal Volume Fraction? Still need to decide if volume loss is an

accepted clinical measure. If not, probably shouldn’t include.] We computed

NBV for each patient.Subsequently, all WM hyperintensities determined to be MS

lesions were identified identified as well-defined focal areas of elevated MRI

signal intensity in the FLAIR imagesdata, a T2-weighted MRI sequence type with

CSF water signal suppression, known to be favorable to detect MS lesions [cite!].

[Jason]

We applied a semi-automatic segmentation approach similar to the process used

to derive demyelinated volume from MWF maps. The FLAIR volume was linearly

registered to the mcDESPOT data volume and then the previously calculated

non-linear warp parameters were applied to bring this FLAIR volume into

standard space. Then Vvoxel-wise mean and standard deviation maps for

healthy control subjects were calculated, on the basis of FLAIR signal intensity.

However, conventional MRI images have arbitrary intensity scales with intensity

values that depend on various acquisition factors. Therefore a correction for

signal intensity inconsistency between subjects was applied to prepare data for

quantitative image analysis. This was accomplished by dividing all voxel

intensities by the “robust maximum” (the signal value at the 98%th percentile ) in

each skull-extracted brain for normal controls and patients. The robust maximum

is the signal value at the 98% percentile. With this correction, we then computed

z-score maps for each MS patient based on the healthy control population mean

and standard deviation maps. Lesions were then identified as clusters of voxels

whose intensities were abnormally brightelevated, greater than 4four>+4

standard deviations above the mean among normals. [not really sure I'm

understanding this as I read this: was the "robust maximum" normalization step

not applied to the healthy control volumes? It doesn't sound like it. If not, why

not?]

The technique spares hyperintense caps around the anterior and posterior horn

and trigonum of the lateral ventricles even if lesions are present. Those changes

are known as subependymal gliosis related to normal aging, i.e are not MS

lesions, and therefore our method dealt with these hyperintensities correctly.

False positive selection in non-brain tissue occurred mainly in the skull marrow

which as higher signal variation across the group, especially within the

subarachnoid vessels (flow void) at the brain surface and in the choroid plexus

and intra-ventricular vessels (flow void). (- details necessary?)

[Sorry I wrote things that kind of overlap with what you say below. Feel free to

keep whatever works best.]

TheA subsequent comparative analysis of individual patient data with healthy

control data means produced lesion segmentation masks depicting voxels

deviating from the population signal intensity by a z-score of +4 [image].

Subsequently a manual adjustment by an experienced MS neuroradiologist

([HHK)] cleared the preselected lesion masks from false positively selected

voxels in non-brain tissue, at the brain surface, and intra-ventricular locations

based on anatomical considerations. The ITK-Snap [Yushkevich 2006cite!]

software package provided a 3D seeding function to eliminate adjacent voxel

clusters of ambiguous origin when at least one featured a distinct localization

beyond brain tissue. [I think this is a little too wordy: ambiguous origin, distinct

localization. I’m unsure of the intended meaning.] Additionally the same

experienced later accomplished a conservative selection of MS lesion tissue in

individual patient space [Not sure what this means either.] (any comparison?).

[Is this last paragraph necessary, or is the previous text sufficient?] This

approach allowed producing maps of well-defined WM T2 lesions.

This technique selected the core lesion. DAWM areas surrounding lesions with

slight increases of WM signal were not selected by the method. IIn a second

step, another set of segmentation masks was produced depicting the regions

with ill-defined borders of intermediate signal intensity between that of lesions on

T2-weighted imaging and that of normal WM. This defines a tissue compartment

of non-lesional pathology, called Dirty-Appearing White Matter (DAWM). The z-

score maps were selected using a threshold of >+2. For each patient, selected

areas in this resulting mask were compared against the edited core lesion masks

using MATLAB [MathWorks, Natick, MA, USAcite]. Those 3D regions that

contained a core lesion were kept as DAWM, the rest were rejected as false

positives. (figure X)[image]. Thus only DAWM surrounding a core lesion was

included. This segmentation was reviewed by an experienced neuroradiologist.

No further correction editing was had to be applied to this unbiased automated

technique. [Not sure what this last sentence means -- sounds like you are saying

that the radiologist did not actually edit anything even though he reviewed it.]

Explain WM-DAWM-core lesions=NAWMIn addition, the WM compartment that

does not harbor any lesions or non-lesional abnormalities on T2 weighted

imaging referred to as the Normal Appearing White Matter (NAWM)

The normal-appearing white matter (NAWM) tissue compartment, which should

describe white matter tissue that looks ordinary on a FLAIR scan, was then

defined as all WM voxels minus core lesion plusand DAWM voxels (WM – core

lesions – DAWM = NAWM) for each patient (figure X). Subsequently tThe WM,

NAWM, DAWM, and lesion segmentations map registration to the standard MNI

space masks were applied to the quantitative maps in patient space to allowed

further compartment-specific study of MWF (Table x) and DV (Table x).

Brain Volume Measurements

The brain parenchymal volume fraction (BPFPVF) is a measure of global

atrophy. It is achieved as GM+WM (brain parenchymal) volume normalized by

GM+WM+CSF (intracranial) volume [Kalkers 2002]. We computed NBV for each

patient using MATLAB [Jason! Correct?]. Additionally the ventricular fraction (VF)

defined as ventricular volume/intracranial volume as a measure of central, mainly

WM atrophy [not sure if we’ll be able to get this measure, not sure what approach

to take to get it, maybe take the CSF mask and keep only the CSF regions

surrounded by WM]

Voxel Based Statistical Analysis

Performing all of the processing above in standard space allowed formalized

statistical testing on a voxel by voxel basis. The age-matched group of healthy

controls was used for direct statistical comparison using a z-score analysis.

Using segmentation results as a matrix for specific quantitative analysis allowed

defining demyelinated volumes within different WM tissue compartments being

NAWM, DAWM, and core lesion. [this last sentence very awkward and difficult to

understand]

[alternate paragraph]

Moreover, tThe above processing enabled us to develop our new measure of

demyelinated volume as the sum of voxels in patient space with a significantly

lower MWF than normal controls in a variety of tissue compartments including

total WM, NAWM, DAWM, and core lesions. The tissue compartment masks

were aligned to each subject’s MWF map using a linear registration and

thresholding at the 0.5 levelnearest neighbor interpolation. Now, with multiple

DV scores for each WM tissue compartment and for each subject, [you should

spell out exactly what you mean by the multiple DV scores] we performed a

battery of Wilcoxon rank-sum tests to examine the ability of this measure to

differentiate between healthy controls normals and different classes of MS. We

also correlated EDSS with DV after a log transformation in an attempt to see if it

is indicative of disability.

(What about T1-Lesions???)

[I was under the impression that T1-lesions show up on T2 FLAIR as well, is this

incorrect?] [correct, it is assumed that T1 hypointense lesions indicate

myelin destruction AND axonal decay, hence not all T2 lesion are also seen

in T1, and that creates a subpopulation T2+/T1+, as well as the Gad

enhancing ‘acute’ T1 lesions, Maybe not in this paper, we should think

about completing MWF and DV measurments in those two lesion

types/compartments also, open for discussion]

Multiple linear regression models an outcome as the combination of

several predictors under some assumptions. These include that the

outcome is linear in any of the predictors after accounting for the others,

the residuals are normally distributed, and theyalso are constant variance

with zero mean. Using a measure like R2, these types of models will always

appear better the more predictors that are included. In fact, even adding

random noise as predictor would seem to improve the model. A better

criterion for evaluating models would be one that not only rewards good

predictions of the outcome but also penalizes for using a large number of

predictors. Mallow’s Cp is such a criterion. The goal in model selection is

to find the model that best explains the outcome parsimoniously. It

attempts to determine the true underlying factors behind an outcome.

Using the R software environment [cite], an exhaustive search was

conducted for all possible combinations of the following predictors: age,

disease duration, PVF, log-DV in whole brain, log-DV in WM, log-DV in

NAWM , log-DV in lesions, T2 lesion load, mean MWF in whole brain, mean

MWF in WM, mean MWF in NAWM, mean MWF in lesions, and gender.

Results

The mean myelin water fraction was initially compared in different white matter

compartments in both the MS patients and the healthy controls (Fig. xx). A

constant drop of MWF in CIS, RRMS and SPMS in global WM, NAWM and

DAWM and MS lesions was noted descending from MS precursor CIS via RRMS

to the progressive courses SPMS and PPMS. Further, the WM lesion MWF in all

courses was lower than in non-lesional WM tissue defined by conventional MRI

confirming lesions to be the focus of inflammatory demyelination. However, the

drop of MWF was non-signifcantlysignificantly higher in CIS lesion compared to

RRMS lesions. Discrimination analysis for MWF in WM was done using

Wilcoxon rank sum testing for each MS subclass vs. normal controls.

Additionally, RRMS vs. SPMS was tested with the somewhat small sample size

of 5 and 6 patients in each group respectively.(… Jason: discrimination of

different classes)[will fill in when WM masks complete]

Consecutively oOur newly introduced voxel-basedquantitative measure,

Demyelinated Volume, was investigated in the same WM compartments (Fig.

xx). Here, the expected increase in demyelination was not uniformly increasing

over the different MS courses, rather a similar amount of demyelinated volume in

lesional and non-lesional pathologic WM tissue, as well as normal appearing WM

was found in both CIS and RRMS. In contrast progressive MS patients revealed

a much higher amount of demyelinated brain tissue volume with even higher

values in PPMS compared to SPMS patients in all tissue classes with a constant

increase of DV from lesions to DAWM and NAWM. However, the DAWM

demyelination in CIS and RRMS was lower than their lesion tissue

demyelination, with RRMS DAWM DV even significantly lower in comparison to

CIS patients. The mean DV in controls however was non-zero value. This is due

to the nature of DV being defined as the sum of voxels with a MWF z-score <-4.

By random chance, normals will have a limited amount of demyelinated voxels.

reflecting not disease related demyelination to be found in healthy individuals

(Necessarly true???). (Would’nt it make sense to compare the percentage of DV

in the WM compartment volume to compare the proportional demyelination to

estimated the severity of tissue degradation in compartment of different sizes

NAWM >>>lesions???)[Yes, this would partially explain why DV increases from

lesions to DAWM and NAWM. It does make more sense intuitively to discuss

these numbers instead but we probably won’t be using them for our regressions

since they correlate poorly.]

Rank sum testing for significance for DV in WM compartments revealed that all

classes cancould be distinguished from healthy controls with p < 0.001 for DV in

global WM (here still in whole brain). Notably, PVF fails to distinguish CIS (p =

0.68) and RR patients (p = 0.76) from controls while DV can. RRMS patients

had significantly lower DV than SPMS patients with p < 0.05 and PVF with p <

0.01.

Notably, PVF fails to distinguish CIS and RR patients. Progressive patients were

also significantly different from normals in PVF with p < 0.01. (Do we display this

somewhere else?)[Since PVF only applies to whole brain, we can’t really

generate another bar chart for it. Perhaps a table of values.]

Next a straightforwardsimple linear regression analysis of the clinical MS

disability score EDSS with MRI retrieved measures was executed . We found an

expected low Pearson correlation of EDSS with the white matter T2 lesion load

the ‘burden of disease’ (R2 = 0.28) reflecting its criticized poor predictive value of

disease related disability (Fig. xx). In contrast, demyelinated volume in normal

appearing WM revealed (have to explain the nawm hypothesis here again) a

much higher correlation to EDSS (R2 = 0.40) suggesting a more direct causative

association to disease-defining neurological impairment (Fig. xx). However, the

parenchymal volume volume fraction correlated highest with EDSS (R2 = 0.56)

indicating the accepted association of brain tissue loss with functional decline

depicted by this brain atrophy indicator (Fig. xx).

Because the concept of the linear regression analysis (R2 ) does not assume a

linearity of the correlation and because EDSS is not a linear scale, alternative

statistics that measures how well the relationship between two variables can be

described by a monotonic function and assuming a monotonic correlation may be

more appropriate to the data presented because of its charactersitics. Such

measure is tThe Spearman Rank Correlation Coefficient (Rrs) is such a measure.

We repeated our regression analysis using this a non-parametric testingmethod.

Hereby w W e found the lowest correlation of EDSS with lesion load (rs = 0.49; p

0.012) and, again, repeatedly a much higher correlation of greater significance

of EDSS with demyelinated volume in normal appearing WM (rs = 0.6059; p

0.0013) and againthe highest correlation EDSS with PVF (rs = 0.6073; p <

0.0001).

A better method to predict EDSS would be to incorporate multiple predictors

rather than just one. We used an exhaustive search to find the best multiple

linear regression model evaluated by Mallow’s Cp. The best model for EDSS

wais determined to contain disease duration (β = 0.0957, p << 0.001),

parenchymal volume volume fraction (β = -19.498, p < 0.001 ), and mean myelin

water fraction in lesions (β = -12.076, p = 0.102). This model hads the lowest

value for Mallow’s Cp among all the possibilities. It explaineds 80.7% of the (R2 )

variance in EDSS and hads an adjusted R2 of 78.1%. However, there were a few

other models ranked closely behind this one that switchedexchanged MWF in

lesions with MWF in WM or log-DV in NAWM. Disease duration and PVF were

still kept. This suggests that these two predictors present vital information for

determining EDSS. It should be noted that all of the third predictors failed an F-

test at the 5% significance level. [these numbers will change]

However Several influencing factors

Criterion helps to choose larger smaller models

importance or not

constant offset = intercept

choose any possible correlation

How large are residuals how trade off between values

Model selection testing fitting to predict EDSS

Linear model for EDSS measured by Mallow’s Cp

Assumption what the model is? ->Diagnostics quality testing

(Combinatorial ?)

best model with lowest mallows Cp

what is improven predictive value of our model

predictive power and fewer terms

mallows cp measuring the fit

full model means all terms tested

The 3 lowest points in descending order are:

Duration, PVF

Duration, PVF, log(DVnawm)

Duration, PVF, MWFwm

model with and w/o DV

analysis of variance (ANOVA) nested models

significantly better fit WITH DV nawm

An exhaustive search for the best linear model for EDSS as measured by Mallow’s CpFull model has terms for:Age, Disease Duration, PVF, log(DVbrain), log(DVwm), log(DVnawm), DVlesion, Lesion Load, MWFbrain, MWFwm, MWFnawm, MWFlesion, Male

The best model is chosen to include:Disease Duration, PVF, MWFwm

Disease Duration, PVF, , log(DVnawm) is a very close second

pvf vs DVnawm 0,24/rs -0.62

dv predicting one type to the other but not for predicting edss development

The correlation [what kind of correlation?] between measurements of

demyelinated volume in various tissue compartments and the MS clinical disease

score EDSS was calculated. Regions of significantly demyelinated voxels were

extracted from the MWF fraction map by computing with the above-mentioned

procedure, visualized and overlayed on an anatomical reference as shown in

representative result images in Fig. 1.

The plain myelin water fraction distribution measured is shown in Table 1.

Initially the traditional measurement of the burden of disease, i.e. the total volume

of hyperintense lesions seen in T2 FLAIR images (Fig.3) was quantified. [should

we include normalized lesion volume measures?] [If it’s a standard metric, then

yes.]

This T2 lesion load did not show a relevant correlation to the EDSS (r2 0.26) and

was subsequently compared to the measurement of demyelination volume within

the same lesions compartment (Fig. 4). This lesion demyelination (DVLesions)

showed an even weaker correlation to EDSS (r2 0.13; p < 0.073) [p-value

indicates significance, easy to misinterpret non-significant].

Before selectively analyzing different WM compartments the total brain

demyelinated volume (DVtotal) was found (Fig.2). Every MS type category differed

significantly in the amount of demyelinated volume compared to the healthy

controls. In the progressive MS-courses (SPMS, and PPMS; mean volume, SD)

a much larger demyelinated volume was observed compared to the non-

progressive categories RRMS and CIS; mean volume, SD). A series of Wilcoxon

rank sum tests showed that the demyelinated volume in any of the MS sub-

classes was significantly higher than the healthy control population, testing at the

5% level. In other words, all MS and CIS patients appear different from normals

in the DV metric.

Particular pairs of patient sub-groups revealed significant differences in

demyelination tested with Wilcoxon rank sum test for significance. The group of

RRMS patients was significantly less demyelinated than the SPMS patients (p <

0.01). The groups of high risk and low risk Clinically Isolated Syndrome were not

significantly separated from each other (p = XXX, do we need to mention?),

which questions the importance of this clinical categorization.[that's a pretty

strong statement to make] However, the entire CIS-group of patients, i.e. low

risk and high risk CIS patients (n=10) can be clearly discriminated from the

healthy control group (p << 0.001).

Within Normal Appearing White Matter compartment demyelinated volume

(DVNAWM), we see a higher correlation with EDSS. These results confirm previous

findings by other investigators that lesions are not the most important disease–

determining marker in Multiple Sclerosis, and provides intriguing evidence that

the invisible burden of demyelination in NAWM is a much more important disease

marker.

Dirty Appearing White Matter compartment demyelinated volume (DVNAWM)

results (to write more about this).

Within the study cohort, correlations were also found between NBV and EDSS (r²

= 0.51, p<0.001), and between DV and NBV (r² = 0.51, p<0.001). NBV was not

significantly different for the CIS subgroup than for the age-matched normal

control group, which was the case for the DV measure. NBV is not able to

distinguish between normals and CIS patients.

Regions of significantly demyelinated voxels were extracted from the MWF

fraction map with the previously described procedure. The DV pattern

substantially deviates from thethr conventional T2 lesion pattern (Fig. xx).

andImportantly, it progresses into the NAWM, showing that it is sensitive to the

invisible burden of disease. [we should probably describe this qualitative part

earlier than the correlation analyses]

MZ suggested: volume vs myelin / BR suggested: pat only vol vs myel meas

MZ suggested: multivariate analysis Rosenberg 3 columns for independent

contribution? NBV/PVF - MWF – Lesion

BR suggested:

Mean/SD across normals, any

correlation to edds all patients together

HK: To be discussed:

(normalized brain volume (NBV), … correlations? )

correlation DV to EDSS subscores, sub-combinations!!

Is Demyelinated Volume whole brain corrected for NBV?

Correlation total brain DV vs lesion load?

Conclusion

[ highlight mcDESPOT as new acquisition technique!]

We have demonstrated that myelin-selective mcDESPOT allows the assessment

of whole-brain isotropic tissue volumes in clinically relevant scan times.

Subsequently, this enabled the quantification of whole-brain and compartmental

demyelination using voxel-based analysis with MWF measurements. The

compartments of interest were WM, NAWM, DAWM, and lesions. Moreover, this

enabled allowed the use ofsubsequent formalized statistics for group comparison

using the Wilcoxon rank sum test. This has not been possible with other

quantitative myelin selective imaging MRI techniques up to date. For the first

time mcDESPOT was used in a controlled clinical study with a cohort of different

disease courses of MS. This current pilot study sought to determine the

prevalence and severity of subtle demyelination in NAWM in different courses of

MS and its relationship to the severity of clinical symptoms and disease related

disability. The mcDESPOT-defined demyelination measurements showed great

promise to act as new markers of clinical disease activity in MS.

[ highlight image processing differences compared to other studies!]

Based on the important mcDESPOT feature of a whole-brain isotropic

acquisition, we were able, for the first time, to combine VBA with whole-brain

MWF measurements. Meyers et al. had already compared traditional

quantitative imaging MRI with region of interest (ROI) and voxel-based analysis

methods to determine the optimal method of analysis of MWF derived from

alternative multi-echo T2 data. Their results of scan-rescan reproducibility

indicated that MWF VBA was the most consistent between scans. For the VBA

method, mean MWF was found to be more reproducible than median MWF

[Meyers, 2009].

Another important quality criteria of quantitative imaging MRI measurements are

their variability. We found (more).

We also have applied a novel strategy to segment MS lesions and important WM

compartments. The standardized segmentation of "FLAIR hyperintensities"

reduced a priori bias of the selecting neuradiologist in comparison to manual only

segmentation. Based onWith this study population based core lesion

segmentation we were subsequently able to provide important WM

compartments for selective MWF and demyelination analysis (more). The

classification of both DAWM and lesions made it possible to define NAWM.

Although subtle demyelination is recognized as pathological features of MS, it is

less clear how early this occurs and how it correlates with MRI-visible lesion

burden. The analysis of the traditional MS surrogate marker T2 lesion volume or

even if specifically myelin is quantifiedlesion demyelination confirmed the known

fact that neither lesion burden nor lesion demyelination correlate well with clinical

disease activity in MS patients. The correlational study results more or less imply

that brain atrophy (PVF) and the invisible burden of disease areis more important

factors towards disability than the lesion tissue.

Measurements of demyelinated volume in NAWM discriminated CIS-patients

from a control population. This highly significant difference represented one of

the most important findings of this study, since many other quantitative imaging

metrics would not show the same sensitivity to early MS. [I love these lines!]

This fact in particular demonstrated the potentially high sensitivity of mcDESPOT

in detecting early and conventionally invisible disease-related myelin loss. It will

need further longitudinal investigation to determine if this measure will serve as a

risk factor for conversion into definite MS in a follow-up study.

Demyelinated volume measurements in NAWM can also discriminate patients

with a secondary-progressive course from relapsing-remitting MS though with our

sample, PVF appears better at this. (more) These findings suggest that these

metrics have merit in predicting conversion into secondary progressive MS.

[ highlight new scientific finding!]

The high correlations to clinical and atrophy measures means that this new

imaging method has strong potential to act as a surrogate measure of disease

severity. Moreover, our results show that mcDESPOT is sensitive to brain tissue

changes even at the pre-MS stage, and well before established volumetric

measures register significant changes. Brain volume loss is an established

clinical marker in MS, however MWF VBA analysis provides some more specific

information over that.

The mcDESPOT method may discern the relationship between multiple sclerosis

and subtle micro-structural demyelinating changes that likely occur in brain tissue

well before lesions can be detected with conventional MRI. Furthermore, it

possibly may directly quantify and visualize the substrate of disability in MS. It is

our specific hypothesis that the development and validation of this new MRI

method designed to measure myelin content over the whole brain and the use of

this method to quantify myelin content in brain regions outside of lesions will yield

more accurate prognostic markers of disease progression in further studies.

We hereby postulate a new MRI derived measure that is able to discriminate

fdifferent MS courses by their amount of demyelination.

T2 weighting is particularly sensitive, but unspecific, to a wide range of brain

pathologies.

MTR shows many aspects of lesional and non-lesional pathology in MS but non-

lesional activity may not be evident.

MTR is sensitive to physiological changes to myelin induced by

inflammation,while the short T2 component is a more specific indicator of myelin

content in tissue. But in MTR non-lesional activity may not be evident

[citations needed]

[Discussing multi-center setting aspects]

An important aspect is the ability to use a prospective myelin imaging method

robustly in a multi-center setting. Besides the lack of specificity (discussed

above) both MTI and DTI feature significant technical limitations. MTI

measurments are semiquantitative, sensitive to machine performance, scanner

and sequence variations. DTI, however, suffers from inherently low signal

intensity (or signal-to-noise ratio [SNR]), and image distortion [Bodini 2009]. The

hereby applied mcDESPOT technique however allows …

[Discussing reproducibility aspects]

Vavasour_2006_NeuroIma_MWFReproducability

[Discussing results in DAWM]

DAWM was addressed to result mainly from Wallerian degeneration originating

from lesions [Kutzelnigg 2005]. Compared to NAWM, DAWM showed reduction

in MWF in MS brain tissue specimen, corresponding to a reduction of Luxol fast

blue (LFB) for myelin phospholipids staining and Bielschowsky silver

impregnation for axons. This suggests that DAWM is characterized by loss of

myelin phospholipids and axonal reduction [Moore 2008].

[Discussing outlook towards remyelinating therapies]

The current state-of-the-art treatments are disease-modifying agents that at

present are able to decrease relapse rates by 30% [Weiner 2009]. However,

despite these advances, the field of MS still lacks specific markers to predict

clinical relapses and disease progression. Novel immunotherapies are on the

rise, but to date non-invasive technologies have failed to provide accurate,

reliable tools to assess the state of myelination. Such technologies are

particularly needed for testing drug efficacy or for monitoring treatment. Imaging

technologies provide potential instruments to investigate in vivo, real time

changes that occur within the CNS over the broad spectrum of natural MS

courses as well as during treatment. ]]]

Discussing Model Selection and Statistical Analysis

Other papers have used non-exhaustive search methods like stepwise

regression when it was not necessary [Korteweg 2008]. These are prone to

finding local minima rather than the globally optimum model. It should be noted

that Mallow’s Cp is itself a random variable dependent on the data. Based on

our limited sample size, it is unclear which model is the most accurate but it is

reasonable to conclude that both disease duration and PVF are significant

players in predicting EDSS since they are used in all of the contending best

models. Less is known about a third predictor as is evident in its high p-value,

which indicates that we cannot reject the null hypothesis that it has no

contribution to EDSS.

BR: critical n=5, subgroup analysis cautious!

1 reas high corr edss

2 early meas dist normals to early MS

3 non versus progressive

Acknowledgements

Thank you to all our participants and their families. Thank you to all of our

research support staff; in particular D. Greer for intensive recruitment and C.

Harper-Little for scanning support.

Funding

XXX

Tables and Figures

Demographic Data lr-CISn=5

hr-CIS

n=5

RRMSn=5

SPMSn=6

PPMSn=5

MS (all)n=26

Mean age, yr (SD) 47 (11) 35 (10) 48 (12) 58 (8) 55 (7) 49 (12)

male/female ratio 2/3 1/4 1/4 0/6 4/1 8/18

Mean disease duration, yr (SD) 2 (2) 3 (2) 15 (10) 18 (15) 12 (9) 10 (11)

Mean EDSS score (SD) 2.0 (1.0) 1.4 (0.8) 2.0 (1.7) 6.4 (1.1) 5.6 (1.1) 4.0 (2.4)

Table 1 Demographic Data of Patients with MS and Reference Subjects: MS =

multiple sclerosis; lr-CIS = low risk Clinically Isolated Syndrome; hr-CIS = high

risk Clinically Isolated Syndrome; RRMS = relapsing-remitting MS; SPMS =

secondary progressive MS; SD = standard deviation; EDSS= Expanded

Disability Status Scale.

Mean MWF Mean (SD)

Normals All Pat lrCIS hrCIS RRMS SPMS PPMS

Total WM MWF0.216

(0.031)

0.178 (0.068)

0.194 (0.065)

0.200 (0.061)

0.185 (0.070)

0.162 (0.075)

0.153 (0.068)

WM w/o lesions0.179

(0.068)

0.194 (0.065)

0.200 (0.061)

0.185 (0.070)

0.164 (0.074)

0.156 (0.067)

NAWM0.180

(0.068)

0.194 (0.065)

0.200 (0.061)

0.185 (0.070)

0.165 (0.074)

0.158 (0.067)

DAWM 0.160

(0.053)

0.177 (0.045)

0.181 (0.059)

0.171 (0.041)

0.152 (0.057)

0.123 (0.064)

Core T2 lesions0.069

(0.058)

0.079 (0.063)

0.074 (0.065)

0.089 (0.069)

0.066 (0.059)

0.038 (0.036)

Table 21 Myelin Water Fraction Distribution: Total WM MWF drops continuously

over the disease spectrum. Between groups we found two important facts.

RRMS and SPMS differed significantly (statistics?) from each other. In PPMS the

MWF drop in lesions is greater than in SPMS. (to be edited)

[Do we need WM w/o Lesions (lesions extracted)?]

Mean DV (SD) Normals All Pat lrCIS hrCIS RRMS SPMS PPMS

Total WM MWF

WM w/o lesions

NAWM

DAWM

T2 lesions

Table 3 Demyelated Volume Distribution

ADDITIONAL ANALYSIS [Suggestion: we should insert a table displaying DV in

different compartments, I assume we might see important differences in DV lesion

in different MS types ]

Table 2 Demyelinated Volume Distribution: The MWF

Percentage of DV in WM sub-compartments in different MS types

Figure 1 WM Segmentation and Voxel-based Analysis: Exemplary display of the

retrieval of WM segmentation, MWF maps, and VBA results [A] Probabilistic WM

map, [B] WM compartments resulting from segmentation, [C] MWF map resulting

from mcDESPOT analysis, [D] Demyelination map resulting from study

population defined VBA, and, [E] selective compartment-specific demyelination

map addressing VBA results to specific compartments.

Figure XX Mean myelin water fraction in different white matter compartments in

different MS patients and the healthy controls (dashed line needs to be included

for HC mean MWF)

Figure XX Demyelinated Volume in different WM compartments. Demyelinated

Volume is plot as log(DV) on the base of 10.

CHANGE new editing WM in the normals!

Figure 2 Demyelination Group Comparison: Total brain Demyelinated Volume

(DVtotal) in the different MS subtypes in comparison to the healthy control group

results, tested with Wilcoxon’s rank sum test for significance. Healthy controls

and all CIS patients, and RRMS and SPMS reveal significant group differences

noted above graph bar.

Figure 3 EDSS Correlation Analysis: T2-lesion-load (volume) vs EDSS. This

confirms again that lesion metrics such as volumetric measurement of affected

tissue conventional MRI techniques only weakly correlate with MS disease

activity. [Did we correct this for brain volume???]

Figure 4 EDSS Correlation Analysis: Lesion Compartment Demyelination

(DVLesions) vs EDSS

Figure 5 EDSS Correlation Analysis: NAWM Compartment Demyelination

(DVNAWM)

Figure 6 EDSS Correlation Analysis: DAWM Compartment Demyelination

(DVDAWM)

References (´Brain´ template: alphabetical order, six authors max et al.)

Bjartmar C, Kinkel RK, Kidd G, Rudick RA, Trapp BD. Axonal loss in normal-

appearing white matter in a patient with acute MS. Neurology. 2001; 57: 1248-

1252.

Bodini B, Khaleeli Z, Cercignani M, Miller DH, Thompson AJ, Ciccarelli O.

Exploring the relationship between white matter and gray matter damage in early

primary progressive multiple sclerosis: an in vivo study with TBSS and VBM.

Hum Brain Mapp. 2009; 30(9): 2852-61.

Daumer M, Neuhaus A, Morrissey S, Hintzen R, Ebers GC. MRI as an outcome

in multiple sclerosis clinical trials. Neurology 2009; 72(8): 705-11.

Deoni SCL, Rutt BK, Arun T, Pierpaoli C, Jones DK. Gleaning Multi-Component

T1 and T2 Information from Steady-State Imaging Data. Magn Reson Med 2008;

60, 1372-1387.

Filippi M, Bozzali M, Rovaris M, Gonen OCK, Ghezzi A, Martinelli V, et al.

Evidence for widespread axonal damage at the earliest clinical stage of multiple

sclerosis. Brain. 2003; 126: 433-437.

Filippi M, Rocca MA, Martino G, Horsfield MA, Comi G. Magnetization transfer

changes in the normal appearing white matter precede the appearance of

enhancing lesions in patients with multiple sclerosis. Ann Neurol. 1998; 43: 809-

814.

Fulton JC, Grossman RI, Udupa J, Mannon LJ, Grossman M, Wei L, Polansky M,

et al. MRI lesion load and cognitive function in patients with relapsing-remitting

multiple sclerosis. AJNR Am J Neuroradiol. 1999; 20(10): 1951-5.

Gareau PJ, Rutt BK, Karlik SJ, Mitchell JR. Magnetization transfer and

multicomponent T2 relaxation measurements with histopathologic correlation in

an experimental model of MS. J Magn Reson Imaging. 2000; 11(6): 586-95.

Griffin CM, Chard DT, Parker GJ, Barker GJ, Thompson AJ, Miller DH. The

relationship between lesion and normal appearing brain tissue abnormalities in

early relapsing remitting multiple sclerosis. J Neurol. 2002; 249: 193-199.

Helms G, Stawiarz L, Kivisäkk P, Link H. Regression analysis of metabolite

concentrations estimated from localized proton MR spectra of active and chronic

multiple sclerosis lesions. Magn Reson Med. 2000; 43(1):102-10.

In vivo measurement of T2 distributions and water contents in normal human

brain. Whittall KP, MacKay AL, Graeb DA, Nugent RA, Li DK, Paty DW. Magn

Reson Med. 1997 Jan;37(1):34-43.

Jones CK, Xiang QS, Whittall KP, MacKay AL. Linear combination of multiecho

data: short T2 component selection. Magn Reson Med. 2004; 51(3): 495-

502.Jones 2004

Kalkers NF, Ameziane N, Bot JC, Minneboo A, Polman CH, Barkhof F.

Longitudinal brain volume measurement in multiple sclerosis: rate of brain

atrophy is independent of the disease subtype. Arch Neurol. 2002; 59(10): 1572-

6.

Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an expanded

disability status scale (EDSS). Neurology. 1983; 33(11): 1444-52.

Laule C, Vavasour IM, Kolind SH, Traboulsee AL, Moore GR, Li DK, et al. Long

T2 water in multiple sclerosis: what else can we learn from multi-echo T2

relaxation? J Neurol 2007; 254(11): 1579-87.

Lucchinetti C, Brück W, Parisi J, Scheithauer B, Rodriguez M, Lassmann H.

Heterogeneity of multiple sclerosis lesions: implications for the pathogenesis of

demyelination. Ann Neurol. 2000; 47(6): 707-17.

Lucchinetti C, Brück W, Parisi J, Scheithauer B, Rodriguez M, Lassmann H.

Heterogeneity of multiple sclerosis lesions: implications for the pathogenesis of

demyelination. Ann Neurol. 2000; 47(6): 707-17.

Mädler B, MacKay AL. In-vivo 3D multi-component T2-relaxation measurements

for quantitative myelin imaging at 3T. Proc ISMRM 2006;13:2112.Madler 2006

Menon RS, Allen PS. Application of continuous relaxation time distributions to the

fitting of data from model systems and excised tissue. Magn Reson Med. 1991;

20(2): 214-27.Menon 1991

Meyers SM, Laule C, Vavasour IM, Kolind SH, Mädler B, Tam R, et al.

Reproducibility of myelin water fraction analysis: a comparison of region of

interest and voxel-based analysis methods. Magn Reson Imaging. 2009;

27(8):1096-103.

Moore GRW, Laule C, MacKay A, Leung E, Li DKB, Zhao G. Dirty-appearing

white matter in multiple sclerosis : Preliminary observations of myelin

phospholipid and axonal loss. J of Neurol. 2008; 255(11): 1802-1811.

Myelin water imaging in multiple sclerosis: quantitative correlations with

histopathology. Laule C, Leung E, Lis DK, Traboulsee AL, Paty DW, MacKay AL,

Moore GR. Mult Scler. 2006 Dec;12(6):747-53.

Oh J, Han ET, Lee MC, Nelson SJ, Pelletier D. Multislice brain myelin water

fractions at 3T in multiple sclerosis. J Neuroimaging. 2007; 17(2): 156-63.Oh

2007

Oouchi H, Yamada K, Sakai K, Kizu O, Kubota T, Ito H, et al. Diffusion

anisotropy measurement of brain white matter is affected by voxel size:

underestimation occurs in areas with crossing fibers. AJNR. 2007; 28(6):1102-6.

Platten M, Ho PP, Youssef S, Fontoura P, Garren H, Hur EM, et al. Treatment of

autoimmune neuroinflammation with a synthetic tryptophan metabolite. Science

2005; 310 (5749): 850-5.

Polman CH, Reingold SC, Edan G, Filippi M, Hartung HP, Kappos L, et al.

Diagnostic criteria for multiple sclerosis: 2005 revisions to the "McDonald

Criteria". Ann Neurol. 2005; 58(6): 840-6.

Prentice RL. Surrogate endpoints in clinical trials: definition and operational

criteria. Stat Med. 1989; 8: 431-440.

Smith, SM, Jenkinson, M, Woolrich, MW, Beckmann, CF, Behrens, TEJ,

Johansen-Berg, H, et al. Advances in functional and structural MR image

analysis and implementation as FSL. NeuroImage, 2004; 23: 208-219.

Schmierer K, Altmann DR, Kassim N, Kitzler H, Kerskens CM, Doege CA, et al.

Progressive change in primary progressive multiple sclerosis normal-appearing

white matter: a serial diffusion magnetic resonance imaging study. Mult Scler.

2004; 10(2): 182-7.

Seewann A, Vrenken H, van der Valk P, Blezer EL, Knol DL, Castelijns JA, et al.

Diffusely abnormal white matter in chronic multiple sclerosis: imaging and

histopathologic analysis. Arch Neurol. 2009; 66(5): 601-9.

Steinman L. Elaborate interactions between the immune and nervous systems.

Nat Immunol. 2004; 5(6): 575-81.

van Waesberghe JH, Kamphorst W, De Groot CJ, van Walderveen MA,

Castelijns JA, Ravid R, et al. Axonal loss in multiple sclerosis lesions: magnetic

resonance imaging insights into substrates of disability. Ann Neurol. 1999; 46(5):

747-54.

Vavasour IM, Whittall KP, MacKay AL, Li DK, Vorobeychik G, Paty DW. A

comparison between magnetization transfer ratios and myelin water percentages

in normals and multiple sclerosis patients. Magn Reson Med. 1998

Nov;40(5):763-8.

Vellinga MM, Geurts JJ, Rostrup E, Uitdehaag BM, Polman CH, Barkhof F, et al.

Clinical correlations of brain lesion distribution in multiple sclerosis. J Magn

Reson Imaging. 2009; 29(4): 768-73.

Vidarsson L, Conolly SM, Lim KO, Gold GE, Pauly JM. Echo time optimization for

linear combination myelin imaging. Magn Reson Med. 2005; 53(2): 398-407.

Vrenken H, Seewann A, Knol DL, Polman CH, Barkhof F, Geurts JJ. Diffusely

abnormal white matter in progressive multiple sclerosis: in vivo quantitative MR

imaging characterization and comparison between disease types. AJNR Am J

Neuroradiol. 2010; 31(3): 541-8.

Webb S, Munro CA, Midha R, Stanisz GJ. Is multicomponent T2 a good measure

of myelin content in peripheral nerve? Magn Reson Med. 2003; 49(4): 638-

45.Webb 2003

Weiner HL. The challenge of multiple sclerosis: How do we cure a chronic

heterogeneous disease? Ann Neurol. 2009; 65(3): 239-248.

Weinshenker BG. The natural history of multiple sclerosis: update. Semin Neurol.

1998; 18(3): 301-7.

Werring DJ, Brassat D, Droogan AG, Clark CA, Symms MR, Barker GJ, et al.

The pathogenesis of lesions and normal-appearing white matter changes in

multiple sclerosis: a serial diffusion MRI study. Brain. 2000; 123: 1667-1676.

Whittall 1989

Whittall KP, MacKay AL, Graeb DA, Nugent RA, Li DK, Paty DW. In vivo

measurement of T2 distributions and water contents in normal human brain.

Magn Reson Med. 1997; 37(1): 34-43.Whittall 1991

Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, et al. User-guided 3D

active contour segmentation of anatomical structures: Significantly improved

efficiency and reliability. Neuroimage. 2006; 31(3): 1116-28.

Excluded modules

Diffusion imaging allows retrieving information about the intrinsic WM

microstructure related to the tissue water diffusion characteristics. Werring et al.

investigated the dynamic of the evolution of water diffusion measurements in pre-

lesion WM in relapsing-remitting MS (RRMS) in a serial diffusion MRI study and

found a steady and moderate increase in the apparent diffusion coefficient

(ADC),, a measure of tissue water diffusion restriction, followed by a rapid and

marked increase at the time of lesion formation, and even a significant but milder

increase in WM regions not harboring lesions [Werring, 2000]. Such progressive

changes of normal appearing white matter (NAWM) were also detected in

primary-progressive MS (PPMS) patients in a longitudinal diffusion MRI study

that also quantified ADC [Schmierer 2006].

Widespread tissue changes are found in NAWM of MS patients by measuring the

magnetization transfer ratio (MTR), . Those changes are mainly explained in

terms of axonal damage and axon loss as one of the major pathological features

accompanying demyelination in MS [Filippi, 1998]. A histological analysis of the

substrate of those imaging findings revealed that not only MTR but also T1

contrast ratio correlated strongly with axonal density, even in NAWM [van

Waesberghe, 1999].

Early axonal pathology, can also be quantified with Proton (H+) MRI

spectroscopy (MRIS) that provides chemical composition information at the level

of metabolites. Early MRIS studies have noted specific changes in metabolite

signatures, not only within focal T2 lesions but even a deviation from normal in

NAWM areas [Helms 2000]. A measure of ‘whole-brain’ N-acetylaspartate

(WBNAA), a marker of axonal integrity, in particular confirmed widespread axonal

pathology, largely independent of MRI-visible inflammation in MS patients even

in Clinically Isolated Syndrome. No correlation however, was found between the

T2 lesion volumes and WBNAA concentrations [is this really a concentration

value or a total integrated NAA value?] [Filippi, 2003].

Axons and their myelin sheath form an individually customized unit.. However,

axonal loss is not necessarily accompanied by demyelination, moreover both

histopathologic changes seem to contribute independently to the appearance in

conventional MRI scans. An imaging-histopathology case study confirmed that

axonal degeneration can occur in the absence of myelin loss as a histopathologic

correlate to abnormal MRI findings in MS patients [Bjartmar, 2001]. [This last

paragraph needs to be improved. Hard to understand.]

Single component T1 relaxation time was found to be abnormal in NAWM in

established MS. When compared to MTR, quantitative T1 measurement was

more sensitive in detecting subtle pathological change. No correlation was found

between NAWM T1 changes and lesion signal characteristics suggesting

independent underlying pathologic mechanisms [Griffin, 2002].

MTR vs disability correlation?

These studies point to a new direction for MS MRIresearch: to move away from

the lesion-centered view and to develop highly sensitive MRI methods that

accurately and quantitatively reflect the global disease burden even in areas that

are apparently normal. Once such methods are developed, important hypotheses

can be tested; for example, that such measures reflect the subtle underlying

disease-determining pathology and will predict clinical changes in MS disease

development, as well as transition towards chronic progression.

In conclusion novel quantitative MRI technologies, have provided in vivo insights

into the pathology of the disease and revealed that primary demyelination, i.e.

selective myelin destruction, is not restricted to focal MS lesions but occurs

throughout the entire CNS parenchyma [key references]. Moreover such

demyelination may be accompanied to a variable degree by remyelination and

repair [key references].