Longitudinal patterns of cortical thinning in multiple sclerosis

Abstract In multiple sclerosis (MS), cortical atrophy is correlated with clinical and neuropsychological measures. We aimed to examine the differences in the temporospatial evolution of cortical thickness (CTh) between MS‐subtypes and to study the association of CTh with T2‐weighted white matter lesions (T2LV) and clinical progression. Two hundred and forty‐three MS patients (180 relapsing–remitting [RRMS], 51 secondary‐progressive [SPMS], and 12 primary‐progressive [PPMS]) underwent annual clinical (incl. expanded disability status scale [EDSS]) and MRI‐examinations over 6 years. T2LV and CTh were measured. CTh did not differ between MS‐subgroups. Higher total T2LV was associated with extended bilateral CTh‐reduction on average, but did not correlate with CTh‐changes over time. In RRMS, CTh‐ and EDSS‐changes over time were negatively correlated in large bilateral prefrontal, frontal, parietal, temporal, and occipital areas. In SPMS, CTh was not associated with the EDSS. In PPMS, CTh‐ and EDSS‐changes over time were correlated in small clusters predominantly in left parietal areas. Increase of brain lesion load does not lead to an immediate CTh‐reduction. Although CTh did not differ between MS‐subtypes, a dissociation in the correlation between CTh‐ and EDSS‐changes over time between RRMS and progressive‐MS was shown, possibly underlining the contribution of subcortical pathology to clinical progression in progressive‐MS.

most likely reflects a diffuse reduction in cortical neuronal density, axonal density, and neuronal size (Popescu et al., 2015). This process has been shown to be associated with cortical lesion volume (Calabrese et al., 2010) and is thought to be, at least in part, the consequence of a pathogenic process driven from pial lesions (Mainero et al., 2015). Cross-sectional analyses point to an additional contribution of demyelination in white matter (WM) tracts connecting the brain cortex with subcortical structures (e.g., deep GM) to this disease component (Kolasinski et al., 2012;Steenwijk et al., 2014); however, currently there is a lack of longitudinal data supporting this pathomechanism.
Cortical GM atrophy is prevalent in all MS subtypes as early as at the stage of clinically isolated syndromes and becomes more widespread and severe with increasing disease duration (Eshaghi et al., 2018;Fisher et al., 2008;Roosendaal et al., 2011). Despite the fact that whole GM volume loss does not differ between MS disease subtypes, a recent multicenter study showed accelerated GM atrophy rates in the temporal lobe in secondary progressive (SPMS) compared to relapsing remitting MS (RRMS) (Eshaghi et al., 2018). This suggests that, the rate of GM volume loss in the temporal lobe may be a surrogate for transition from the RRMS to the SPMS phase of the disease. However, it is unclear if this observed between-group difference concerns a specific temporal lobe GM region or diffuse temporal cortical volume loss and whether the observed between MS disease subtype difference may be ascribed to the higher age of the SPMS patients.
Increasing evidence from both cross-sectional and longitudinal analyses points towards an association of cortical GM pathology such as cortical lesions and atrophy with physical and especially with cognitive dysfunction (Bergsland et al., 2017;Eijlers et al., 2018;Fisniku et al., 2008). However, the exact anatomical substrate of the cortical GM loss driving disease progression is still poorly understood.
In this study, we examined the cortical thickness (CTh) changes in a large cohort of MS patients over 6 years. We aimed at localizing differences in the reduction of cortical GM between disease phenotypes and to study the contribution of CTh changes to the progression of physical and cognitive disability progression. We also examined the effect of WM lesions on cortical GM changes over time.

| Study design and participants
We analyzed clinical and MRI data of RRMS, SPMS, and primary progressive (PPMS) patients (Table 1) from an ongoing large scale cohort study (GeneMSA) at a single center (Multiple Sclerosis Center, University Hospital, Basel, Switzerland) (Tsagkas et al., 2018;Tsagkas et al., 2019). Patients were followed over a maximum of 6 years (7 annual time points). Both clinical assessments and the MRI examinations were performed at least 1 month after the occurrence of a clinical relapse or treatment with glucocorticosteroids. The diagnosis of MS was made in accordance with international panel established criteria (McDonald et al., 2001). The study was approved by the local ethics committee. All patients included in this work have been previously reported in former studies (Tsagkas et al., 2018;Tsagkas et al., 2019).

| Clinical assessment
All patients underwent a standardized neurological examination including the Expanded Disability Status Scale (EDSS; www. neurostatus.org) by trained and certified examiners, Timed 25-foot walk test (T25fwt) and Paced Auditory Serial Addition Test (PASAT) annually. Patients also underwent an annual Symbol Digit Modalities Test (SDMT) starting at the fourth follow-up time (or at the third year of monitoring). No parallel test versions were used for the SDMT, whereas two versions of the PASAT were deployed for annual neuropsychological tests. Relapses that occurred 12 months prior to each follow-up were also recorded. All clinical and neuropsychological metrics were recorded as cross-sectional measurements at each follow-up, whereas longitudinal changes of those metrics were not documented prospectively.

| MRI analysis
All brain WM lesions were segmented on the PD-weighted images by trained expert observers according to the standard operating procedures used at the local institution for the analysis of clinical phase II and phase III trial. T2-weighted lesion volume (T2LV) was calculated for the whole brain as well as for each lobe as segmented by the "Automatic Nonlinear Image Matching and Anatomical Labeling" algorithm (ANIMAL) (Collins, Holmes, Peters, & Evans, 1995) at all available time points.
In order to avoid misclassification of lesions as GM, lesions masks generated on PD-weighted images were used to fill the lesions on T1-weighted images with the intensity of the surrounding white matter tissue . CTh was estimated on the lesion-filled T1-weighted images using the fully automated CIVET 1.1.10 pipeline (Collins, Neelin, Peters, & Evans, 1994;Lyttelton, Boucher, Robbins, & Evans, 2007). Summarizing this process, the T1-weighted images were linearly registered to the standard stereotaxic space defined by the MNI ICBM 152 model (Mazziotta et al., 2001). The images were then corrected for intensity nonuniformity using N3 (Sled, Zijdenbos, & Evans, 1998) and a nonlinear registration to the model (Collins et al., 1994) was applied. The tissue classification was performed using INSECT, whose output was then fed to a Partial Volume Estimator, which in turn is used for the actual surface fitting (Tohka, Zijdenbos, & Evans, 2004). Each voxel was classified as WM, GM, or CSF. The images were then mapped to a probabilistic atlas using the ANIMAL algorithm. Finally, the WM surface was generated by using a deformable ellipsoid polygonal model that shrinks until it fits the WM/GM interface. To generate the GM surface, the WM surface was expanded until the GM/CSF interface (or pial surface) is reached using a Laplacian approach in order to find the best fit (Jones, Buchbinder, & Aharon, 2000;Kim et al., 2005). Specifically, to adequately estimate the CTh, the Laplace's equation describes a smooth trajectory between the WM and GM surfaces defining a layered set of surfaces (Jones et al., 2000). Thus, each vertex on the WM surface maps

| Statistical analysis
Comparisons of demographic factors, clinical measurements, and number of follow-ups between MS subtypes were made using Welch's and Pearson's chi-squared test with Yate's continuity correction. A logarithmic transformation of the EDSS was performed in order to correct for its nonlinearity in representing physical disability, as conducted in previous studies Tsagkas et al., 2018;Tsagkas et al., 2019). The annualized relapse rate was calculated for each patient.
Vertex-wise longitudinal analysis was performed using a linear mixed effect model (LMER) in order to explore longitudinal correlations between the patients' CTh and demographic, clinical and T2LV measurements. LMER was also used to examine the trends of PASAT and SDMT changes over time in our cohort after a square transformation for PASAT in order to approximate a normal distribution. This was done using a random intercept and a random time slope for each subject to allow for within-subject and between-subject variance. CTh was always used as the dependent variable in our analysis. For the investigation of the association between CTh and clinical outcomes or T2LV, the independent variables were entered blockwise keeping the following sequence: first demographics and then clinical variables or T2LV respectively. Separate analyses were conducted for the whole brain T2LV as well as for the left and right T2LV of the frontal, parietal, temporal, and occipital WM. Each variable was tested both for its correlation to the CTh intercept as well as to the CTh slope over time. All independent variables without statistical significance were excluded from the final model. In order to reduce the risk of type I errors the results were corrected for multiple comparisons by using the False Discovery Rate approach set at q < 0.05.
In order to assess between-group CTh differences of RRMS and SPMS with PPMS patients, we performed propensity-score matching baseline covariates, including sex, age and disease duration as described in a previous study (Tsagkas et al., 2019). RRMS and SPMS were matched with PPMS patients, based on high similarity of propensity scores, on a 2:1 basis for each group and all groups had a similar follow-up time. Comparisons of the RRMS and SPMS CTh with PPMS were done using vertex-wise LMER using the False Discovery Rate approach set at q < 0.01 (instead of 0.05) in order to correct for multiple comparisons between the patient groups.
Beside the multiple comparison correction approaches discussed above, no other approach was used in the rest of the analysis, since the models used for the examination of the correlation between CTh and demographical/clinical data are independent from each other.

| RESULTS
A total of 243 MS patients (180 RRMS, 51 SPMS, and 12 PPMS) were monitored yearly over an average time span of 4.36 ± 2.03 years.
Ninety-three patients, completed all 7 scans, whereas another 35 completed 6 scans and 29 completed 5 scans. The rest of the patients (86) completed four scans or less. Demographics and clinical characteristics of our cohort are described in Table 1.

| Reduction of cortical thickness over time
Reduction of CTh in the whole cohort and each individual MS subtypes are graphically displayed in Figure 1. In the whole cohort as well

| Association of CTh with demographic factors
Associations between demographic factors and CTh are graphically displayed in Figure 2 and Table 2. Age at baseline was associated with a reduction of the average CTh predominantly in the parietal, prefrontral, and frontal cortex bilaterally, while being F I G U R E 1 CTh reduction over time in the whole cohort and individual subgroups of disease subtypes. The gradient from yellow to red indicates a lower to higher negative reduction respectively, as shown by the t-values extracted from our linear mixed effect models. In each graph, the highest (or less negative) gradient value represents the threshold of the respective t-values after correction with the false discovery rate approach for multiple comparisons set at q < 0.05. Up left: CTh reduction over time in the whole cohort. Up right: CTh reduction over time in the relapsing remitting multiple sclerosis (RRMS). Down left: CTh reduction over time in the secondary progressive multiple sclerosis (SPMS). Down right: no statistically significant CTh reduction over time was shown in the primary progressive multiple sclerosis (PPMS) slightly more extended in the right hemisphere (q < 0.05). Age at baseline was also negatively associated with CTh changes over time in extended cortical regions mostly involving the bilateral prefrontal cortex, bilateral parieto-occipital regions, and the superior temporal gyri (q < 0.05). Disease duration at baseline was also associated with extended cortical thinning of the bilateral frontal and prefrontal cortex as well as large parietal, temporal, and occipital CTh reduction-more extended in the left hemisphere, but was not correlated with the CTh changes over time (q < 0.05). Sex was not correlated with CTh or its changes over time. In RRMS, the annualized relapse rate was not associated with CTh or its changes over time.
F I G U R E 2 Effect of age and disease duration (DD) on the cortical thickness (CTh) of all multiple sclerosis patients. The gradient from yellow to red indicates a weaker to stronger negative correlation respectively, as shown by the t-values extracted from our linear mixed effect models. In each graph, the highest (or less negative) gradient value represents the threshold of the respective t-values after correction with the false discovery rate approach for multiple comparisons set at q < 0.05. Up left: correlation of the average CTh with age at baseline.

| CTh differences between groups
CTh and its changes over time did not differ between RRMS and SPMS patients after correcting for age and disease duration at baseline.
In sex-, age-and disease duration-matched subgroups of 60 RRMS, SPMS, and PPMS patients (24 RRMS, 24 SPMS and 12 PPMS, mean age at baseline 48.9 ± 8.2 years, mean disease duration 10.2 ± 6.6 years, 32 female), also no differences in CTh and its changes over time were found among groups.

| Association of CTh with WM lesion load
After correcting for age and disease duration, a negative correlation was found between whole brain T2LV and the average CTh in regions extending symmetrically in nearly the whole cortex bilaterally (overall right: MTV −4.00 ± 1.36; left: MTV −4.24 ± 1.35, q < 0.05). Whole brain T2LV changes were not associated with the CTh changes over time. These results are also graphically displayed in Figure 3.
When examining the relation between regional T2LV (left and right frontal, parietal, temporal and occipital WM lesions) and CTh, all regional T2LV were associated with a reduction of CTh in extended bilateral cortical regions, analogous to the whole brain T2LV (q < 0.05). In addition, left temporal T2LV changes were nega- and occipital T2LV and CTh is shown in Figure 4, whereas the correlation between occipital T2LV and CTh is displayed in detail in Table 3.

| Whole cohort
In the whole cohort, the log(EDSS) was not associated with the average CTh, after correcting for age and disease duration. However, log(EDSS) changes were negatively correlated with the CTh changes over time in large extended bilateral cortical regions ( Figure 5), predominantly in the right temporal and left frontal and parietal lobes (q < 0.05). These results are also shown in detail in Table 4.

| RRMS
In the RRMS group, the log(EDSS) was not associated with the average CTh, after correcting for age and disease duration. However, log(EDSS) changes were negatively correlated with the CTh changes over time in large, extended bilateral cortical regions (q < 0.05) ( Figure 5). These results are also shown in detail in Table 4.
F I G U R E 3 Effect of the whole brain T2w-lesion (T2LV) on the cortical thickness (CTh) of all multiple sclerosis patients. The gradient from yellow to red indicates a weaker to stronger negative correlation respectively, as shown by the t-values extracted from our linear mixed effect models. In each graph, the highest (or less negative) gradient value represents the threshold of the respective t-values after correction with the false discovery rate approach for multiple comparisons set at q < 0.05. Left: correlation of the average CTh with the average T2LV. Right: no statistically significant correlation was shown between CTh and T2LV changes over time

| SPMS
In the SPMS group, the log(EDSS) was not associated with the CTh, after correcting for age ( Figure 5).

| PPMS
In the PPMS group, the log(EDSS) was not correlated with the average CTh. However, log(EDSS) changes were negatively correlated with F I G U R E 4 Effect of the left temporal and left occipital T2w-lesion (T2LV) on the cortical thickness (CTh) of all multiple sclerosis patients. The gradient from yellow to red indicates a weaker to stronger negative correlation respectively, as shown by the t-values extracted from our linear mixed effect models. In each graph, the highest (or less negative) gradient value represents the threshold of the respective t-values after correction with the false discovery rate approach for multiple comparisons set at q < 0.  (Table 1).

| DISCUSSION
This is the first longitudinal study examining the relationship of CTh in a vertex-wise manner with clinical-and lesion load measurements in a large cohort of different MS phenotypes over 6 years. Our work demonstrated similar temporospatial cortical changes over different disease subtypes, which -howeverwere related to disease progression in a disease-type-specific manner. We also showed an association of T2LV and CTh, although the effect of T2LV changes to longitudinal CTh changes was shown to be only marginal.
Our study demonstrated a significant CTh reduction over time in large prefrontal, frontal, parietal, and temporal cortical areas in all MS patients. These results are similar to a large cross-sectional multicenter study comparing RRMS patients and healthy controls (Narayana et al., 2012). Therefore, it can be hypothesized that the observed atrophy demonstrated in these large cortical areas may represent a disease-specific effect rather than the impact of aging, although our study could not confirm this hypothesis due to the absence of healthy controls. Similar results were shown in separate analysis for RRMS and SPMS patients, whereas no significant CTh reduction over time was shown in the PPMS. The latter finding should be considered with caution, since the sample size of our PPMS group was rather small (n = 12) and therefore this analysis may have lacked in power. This may be also supported by the findings of a recent large longitudinal volumetric study showing significant cortical atrophy in this group (Eshaghi et al., 2018).

In our patients, a clear effect of aging on CTh was demonstrated,
which is in line with previous cross-sectional and longitudinal studies of healthy individuals (Fjell et al., 2015;Thambisetty et al., 2010). In particular -as also seen in Figure 2, older patients were found to have reduced frontotemporal CTh as well as an accelerated cortical  thinning in large cortical areas involving the prefrontal cortex, parietooccipital regions, and the superior temporal gyri. Aging-related patterns of CTh have been shown to be driven by both genetic factors (Matsushita et al., 2015) and functional relationships of converging regions (Fjell et al., 2015).
Our analysis also revealed a correlation between CTh and disease duration with diffuse cortical thinning being apparent in later stages of the disease. The reported association between the loss of cortical GM and increasing disease duration is independent of normal aging, since disease duration was added after age in our LMER models.
F I G U R E 5 Correlation between EDSS and CTh changes over time in the whole cohort and individual subgroups of disease subtypes. The gradient from yellow to red indicates a weaker to stronger negative correlation respectively, as shown by the t-values extracted from our linear mixed effect models. In each graph, the highest (or less negative) gradient value represents the threshold of the respective t-values after correction with the false discovery rate approach for multiple comparisons set at q < 0.05. Up left: correlation between EDSS and CTh changes over time in the whole cohort. Up right: correlation between EDSS and CTh changes over time in the relapsing remitting multiple sclerosis (RRMS). Down left: no statistically significant correlation was shown between EDSS and CTh changes over time in the secondary progressive multiple sclerosis (SPMS). Down right: correlation between EDSS and CTh changes over time in the primary progressive multiple sclerosis (PPMS) However, the rate of CTh reduction over time was not a function of disease duration, suggesting a steady cortical thinning throughout the course of the disease in patients during the monitoring time of our study.

T A B L E 4 Association of cortical thickness changes over time with EDSS changes by disease subtypes
As opposed to previous cross-sectional studies and one recent large-scale longitudinal volumetric studies of cortical GM in MS (Eshaghi et al., 2018;Fisher et al., 2008;Roosendaal et al., 2011 Our work demonstrated a correlation of widespread CTh reduction with larger whole brain T2LV. This is in line with previous crosssectional studies suggesting that focal inflammatory events in the WM may-at least partially-"drive" cortical atrophy (Bergsland et al., 2015;Bodini et al., 2009;Henry et al., 2009). However, whole brain T2LV did not contribute to the temporal evolution of CTh over 6 years, suggesting that focal inflammatory events do not lead to an immediate loss of cortical GM. Similarly, the annualized relapse rate was not associated with CTh in RRMS. Further, exploration of a potential effect of the regional T2LV changes over time on CTh reduc- showing that other structures such as the spinal cord correlate better with progression of physical disability than brain metrics in progressive MS patients (Tsagkas et al., 2018;Tsagkas et al., 2019).
Surprisingly T25fwt, PASAT and SDMT did not correlate with the CTh neither in the whole cohort nor in the different MS-subtypes.
Concerning the T25fwt, it could be hypothesized that the large between-patient variability could partly be responsible for the lack of association with CTh changes, even with motor-related cortical areas.
Moreover, other structures such as the spinal cord have been also shown to be better explanatory variables for T25fwt compared to brain metrics (Tsagkas et al., 2018;Tsagkas et al., 2019). Furthermore, in contrast to previous literature (Calabrese et al., 2009;Steenwijk et al., 2016), cognitive performance-as measured by PASAT and SDMT-in our cohort was not associated with CTh. However, a paradoxic significant improvement of those scores was evident in our patients in both scores, which may be attributed to a learning effect through repetition. This is in line with a number of studies including healthy controls and MS patients showing improved cognitive performance through practice or repetitive testing, even when testing was performed with relatively long intervals between follow-up, similarly to our study (Baird, Tombaugh, & Francis, 2007;Bartels, Wegrzyn, Wiedl, Ackermann, & Ehrenreich, 2010;Basso, Bornstein, & Lang, 1999;Johnen et al., 2019;Roar, Illes, & Sejbaek, 2016).
There are some limitations of our study that have to be men- While injectables also show an effect on brain gray matter atrophy, based on previous studies, we believe that this effect is rather negligible (Favaretto, Lazzarotto, Margoni, Poggiali, & Gallo, 2018). During the collection of data in this study, there was no approved treatment for PPMS, so that no patient received treatment in this patient group.
However, the distribution of disease modifying agents in our RRMS and SPMS patients did not significantly differ. Finally, it has to be noted that we did not examine the association between cortical lesions and CTh over time, since the sensitivity of T2-and PD-weighted sequences for cortical lesions is known to be very low.

| CONCLUSION
In conclusion, our study demonstrated a more prominent diffuse CTh reduction with increasing lesion load. However, only a marginal focal effect of regional T2LV changes to CTh changes over time was shown in neighboring and anatomically connected cortical areas, thus suggesting that GM atrophy progresses -at least partiallyindependent from focal inflammatory events. MS-subgroups did not differ in terms of CTh. However, a clear dissociation in the correlation between CTh and EDSS changes over time between RRMS and SPMS patients was shown. Based on this finding we can hypothesize, that other CNS structures such as the spinal cord may be more relevant with regard to disability progression in SPMS patients.

ACKNOWLEDGMENTS
The authors are very grateful to all participants and medical staff involved in the GeneMSA cohort study, in particular Alain Thoeni, who collected all MRI data. C.T. was financially supported by the

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.