Prognostic value of white matter lesion shrinking in early multiple sclerosis: An intuitive or naïve notion?

Abstract Background and purpose New or enlarging T2‐hyperintense white matter lesions (WML) are associated with clinical disease progression in multiple sclerosis (MS). The prognostic value of WML shrinking is unclear. Assuming that waning of acute inflammation and repair processes would be the main drivers of WML shrinking, we aimed to assess the prognostic value of WML shrinking in early MS. Methods We retrospectively analyzed a cohort of 144 early MS patients with three brain MRI scans at baseline and after 1 and 3 years available. All patients were therapy naïve at baseline and 70.5% of them treated with disease modifying drugs at year 1. We determined the volume of WML shrinking between MRI scans, total WML volumes, number of gadolinium‐enhancing and new WML, white matter (WM) and gray matter volumes at each MRI scan. Clinical disability was measured by Expanded Disability Status Scale. We performed the correlation analyses of WML shrinking with other MRI parameters and clinical outcome. Results White matter lesions shrinking was highly variable between patients and correlated with the initial number of gadolinium‐enhancing WML and with WM volume decrease. WML shrinking was not associated with clinical outcome. Conclusion We found no indication of a prognostic value of WML shrinking in early MS patients. WML shrinking seems to be related to waning of acute inflammation.

it a good sign. However, the prognostic value of WML shrinking is unclear.
We assumed that a newly appearing WML first rapidly increases in size and then slowly decreases (Figure 1a; Meier & Guttmann, 2003;Reich et al., 2015). WML appearance on MRI is accompanied by gadolinium enhancement, which usually lasts for 2-8 weeks (Absinta, Sati, & Reich, 2016;Cotton, Weiner, Jolesz, & Guttmann, 2003;Guttmann et al., 2016;Lai et al., 1996). Determination of WML shrinkage may be erratic in case of a newly appearing WML ( Figure 1b). As the first MRI scan may capture different phases of the initial WML growing, timing of the initial scan may decide whether analysis of the same WML after 1 year demonstrates shrinkage In our study, we addressed the phenomenon of WML shrinking as observed in daily clinical practice. We investigated a cohort of 144 early MS patients which had received brain MRI scans at baseline (MRI 0), after 1 (MRI 1), and after 3 years (MRI 3). We determined WML shrinking by the longitudinal pipeline of lesion segmentation tool (LST) which measures changes of each individual WML between the consecutive MRI scans. We correlated WML shrinking (MRI 0-1; MRI 1-3) with basic and other MRI parameters to understand its relation to other aspects of MS pathology and to identify potential confounders. Finally, we related WML shrinking to clinical outcome and therapeutic success. To exclude the confounding effect of newly appearing WML as described above, we repeated analyses with clinical parameters in the subgroup of patients without gadolinium-enhancing WML.

| Subjects and study design
This study was performed in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans and was approved by the local ethics committee. We retrospectively analyzed the data that were collected in an observational study (TUM-MS) (Polman et al., 2011) or clinically isolated syndrome (CIS), which was defined as first demyelinating event suspicious of MS accompanied by at least two WML typical of MS detected by MRI; patients had three consecutive brain MRI scans at baseline (MRI 0) and after about 1 (mean 11.6 months; range 7-16 months) and 3 years (mean 35.78 months; range 30-42 months); patients were not treated with disease modifying drugs (DMD) at baseline. 70.5% of them were treated with DMDs at MRI 1 according to their individual preferences and suggestion by their treating physician. Exclusion criteria were an immunomodulatory treatment prior to the baseline scan and corticosteroid treatment within 30 days prior to any MRI scan. The full dataset was available in 146 MS patients. One patient was withdrawn from analysis due to an artifact in WML segmentation and another due to several tumefactive demyelinating WML, leading to exceptionally high shrinking of WML between con-

| T2-hyperintense white matter lesion segmentation
T2-hyperintense WML were segmented from FLAIR and T1-w images by the lesion growth algorithm as implemented in the lesion segmentation tool (LST) toolbox (Schmidt et al., 2012) version 2.0.15 (http://www.appli ed-stati stics.de/lst.html) for SPM12 (http://www. fil.ion.ucl.ac.uk/spm). The same initial threshold (κ = 0.3) was used for all images. This value has been proven to be useful in previous studies (Muhlau et al., 2013;Rissanen et al., 2014) and was confirmed by visual inspection.
Lesion segmentation tool's longitudinal pipeline ) was used to assess WML decrease and increase between MRI 0-1, and 1-3, based on changes of each individual WML. The longitudinal pipeline comprises the following steps:

| Longitudinal segmentation
A joint lesion map was created. This is a binary mask including all voxels that were segmented as a WML in at least one time point.  Figure 2. All analyses of WML changes were checked by thorough visual inspection, which confirmed the performance of LST as expected (Schmidt et al., 2012. Two patients were excluded (one with artifact, one with tumefactive WML).

| New and gadolinium-enhancing T2hyperintense white matter lesions
The number of new and gadolinium-enhancing WML was extracted from the radiology report.

| Brain imaging parameters
Imaging parameters of all time points are summarized in Table 2.
Mean WML volume did not change significantly between MRI 0 and

| Correlation of white matter lesion shrinking with demographic parameters
White matter lesion shrinking between MRI 0 and 1 was associated with DMD treatment at MRI 1 (r = −.169, p .043), WML shrinking between MRI 1 and 3 was associated with DMD treatment at MRI 3 (r −0.230, p .006). More effective DMDs were associated with more pronounced shrinking of WML. Age, sex, disease duration, and time interval between MRI scans were not significantly associated with WML shrinking.

| Correlation of white matter lesion shrinking with markers of acute inflammation
White matter lesion shrinking between MRI 0 and 1 correlated with WML volume at MRI 0 (r −.471, p < .001; Figure 3a). Highest shrinking of WML volume was found in patients with gadolinium-enhancing WML at MRI 0 (yellow dots). WML shrinking between MRI 0 and 1 correlated with the number of gadolinium-enhancing WML at MRI 0 (r −.530, p < .001).
White matter lesion shrinking between MRI 1 and 3 (Figure 3b) was less pronounced than between MRI 0 and 1 (p < .001) and cor-

| Correlation of white matter lesion shrinking with brain volume changes
WML shrinking between MRI 0 and 1 was related to a reduction of WM volume between MRI 0 and 1 (r = .290, p < .001). WML shrinking between MRI 1 and 3 was related to a reduction of WM volume between MRI 1 and 3 (r .447, p < .001). No association between WML shrinking and changes of GM volume was found (all p-values > .1).

| Multiple linear regression model to explain variance of white matter lesion shrinking
In a linear regression model, WML shrinking between MRI 0 and 1 was explained by age (younger MS patients had more pronounced WML shrinking), WML volume at MRI 0, the number of gadoliniumenhancing WML at MRI 0 and WM volume change between MRI 0 and 1 (R 2 0.362, Table 3). WML shrinking between MRI 1 and 3 was explained by the number of gadolinium-enhancing WML at MRI 1 and WM volume change between MRI 1 and 3 (R 2 0.284, Table 4).

| Correlation of white matter lesion shrinking with clinical parameters
Expanded Disability Status Scale was low at all three time points (  Figure 4).

| D ISCUSS I ON
In the study at hand, we aimed to assess the prognostic value of WML shrinking in early MS patients as observed in clinical practice.  with markers of acute inflammation (number of new and gadolinium-enhancing WML). Since WML shrink most in the first years after their appearance, we believe that early WML shrinking is primarily due to waning of edema and acute inflammation.
In addition, we found a correlation of WML shrinking with reduction of WM volume, which might also be explained by the waning of initial inflammation/edema. This "pseudoatrophy" particularly in the white matter has been described to accompany initiation of several disease modifying drugs (Dwyer et al., 2015;Vidal-Jordana, Sastre-Garriga, & Pérez-Miralles, 2013Zivadinov et al., 2008). In addition, the association of WML shrinking with reduction of WM volume could in part reflect degradation of some irreversibly destroyed tissue within the WML and in related regions as had been demonstrated by the novel technique of voxel-guided morphometry (Fox et al., 2016). In contrast, a long-term study, investigating WML changes in 22 MS patients over a mean follow-up period of 16.4 years, found no association of WML shrinking with changes in supratentorial brain volume (Sethi et al., 2016). However, just one WML per patient was considered and related to changes in whole brain but not WM volume. Correspondingly, we found an association of WML shrinking with changes in WM but not GM volumes. This is well conceivable with the notion that GM pathology develops partly independent from WM pathology in MS (Calabrese et al., 2015). We therefore do not assume that the association of WML shrinking with WM volume reduction observed in our study results from global degenerative processes.
We found higher WML shrinking in patients treated with more effective DMDs. This correlation, however, lost significance in the linear regression model after correction for baseline inflammatory activity. We assume that therapy is more likely to be escalated in patients with higher inflammatory disease activity which also shows the higher shrinking of WML.
Although WML shrinking was highest in patients with gadolinium-enhancing WML, some patients did not show substantial WML shrinking even if gadolinium-enhancing WML were found at the respective baseline scan (Figure 3). In these cases, first MRI scan might have captured WML in their initial growing phase as

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. F I G U R E 4 Box plot of WML shrinking between MRI 0 and 1 for patients with therapy switch between MRI 1 and 3 due to ongoing radiological or clinical disease activity (left) and other patients (right; no therapy switch or switch within baseline therapies due to side effects) is shown. Patients are numbered with respect to the extent of white matter lesion shrinking (1 = highest). High WML shrinking between MRI 0 and 1 had no significant effect on therapeutic success (binary logistic regression model, p .948). Five patients with pronounced WML shrinking between MRI 0 and 1 were switched between MRI 1 and 3 due to ongoing clinical or radiological disease activity. Patients under escalation therapies at MRI 1 (N = 3) are not included other patients therapy switch due to ongoing disease activity WML shrinking MRI 0 -1 (ml)