Introduction
Cognitive impairment (CI) is a common symptom in multiple sclerosis (MS). The neuropathological substrates of MS-related CI, and hence potential therapeutic targets, are still controversial [
1]. One of the earliest magnetic resonance imaging (MRI) studies on CI in MS [
2] reported a strong relationship between CI and white matter lesions (WML) compatible with the classic idea of a disconnection syndrome [
3]. In later MRI studies, relations of CI to deep and cortical grey matter (GM) atrophy were reported [
4‐
11]. When we, to the best of our knowledge, reviewed the ten structural brain MRI studies on CI in MS, with the highest numbers of CI patients included [
4‐
13], we observed inconsistent results with regard to the contribution of WML load and GM atrophy to CI in MS. This seems well conceivable given the variety of methods used which hampers comparability. Some studies found only a contribution of WML [
12,
13]. In these two studies, high-resolution structural images were not available, so that changes in GM, i.e., atrophy, may have been missed. Other studies identified GM atrophy as the main driver of CI in MS. One study found an association of deep and cortical GM atrophy through voxel-wise analysis of high-resolution T1-weighted images at 3 T [
4], whilst WML volume did not differ between groups. However, patients groups were heterogeneous comprising patients with relapsing–remitting (
n = 22), secondary progressive (
n = 29), and primary progressive (
n = 22) MS, which may have resulted in a high variance with regard to WML volume lowering statistical power. In another two studies, primarily thalamic atrophy [
5,
6] was related to CI in MS. In one of the two studies, the methodology was focused on the thalamus. In a complex statistical model, only thalamus measures explained CI; of note, raw values of normalized brain volume, deep GM, and WML volume were also significantly different between groups [
5]. In the other of the two studies, healthy controls were compared to patients with PPMS which showed lower cognitive test scores. This setting required a correlation of test performance with MRI-based measures only in the patients. Again, in a statistical model including several MRI parameters (stepwise linear regression), only thalamic volume showed a significant effect, whilst raw values of other parameters differed as well [
6]. Further studies reported contributions of both WML load and GM atrophy [
7‐
11]. However, in some of these studies, groups of patients with different degrees of CI also differed in other characteristics such as disease duration and EDSS, which was not accounted for statistically, so that results do not necessarily demonstrate differential contributions to CI [
7,
8]. The same applies to a study on a large group of MS patients in which 43% were classified as CI and in which voxel-wise correlations of GM with scores of different tests were performed across the whole group [
10]. In a study using high-resolution MRI, CI was related to WML and a loss of deep GM and, to a small degree, lower cortical thickness [
9]. Moreover in a study on 1052 patients with MS, WML volume and brain parenchymal fraction could be analyzed. A weak correlation between CI and both MRI-based parameters was found in the early stages (< 2 years) and a strong correlation in later stages (> 15 years), whilst no differentiation between GM structures was possible. Finally, atrophy of the corpus callosum has also been related to CI, although based on a lower number of patients [
14‐
17].
In this study, we retrospectively analyzed data from a large cohort of patients with MS primarily in early stages. All included patients underwent neuropsychological testing and standardized high-resolution MRI including FLAIR and T1-weighted sequences. These data enabled us to search for structural brain changes related to CI both at the global and regional level across the whole brain. Besides identifying brain structures, we could compare the sizes of their contributions, analyze their relation to each other, and search for differential contributions.
Discussion
We aimed to study brain structures critically involved in MS-related CI by investigating a large cohort of MS patients with available cognitive testing, including confirmation of CI in 51 patients by detailed NPA, and high-resolution MRI. This enabled us not only to identify related brain structures but also to estimate their effect sizes and to analyze their differential contributions as well as their relation to each other. We will consider our study design and cohort, then discuss the results focusing on WML and thalamic atrophy, and finally, acknowledge limitations.
This study was part of our TUM-MS in-house observational study on MS planned in 2007. Data collection was performed according to a standardized protocol for regular outpatient visits. CP was assumed in patients without any indication of CI according to the treating physician who routinely considers his and the patient’s impression as well as the result of the screening battery MuSIC. For quality assurance, we further applied the commonly accepted cut-off score of the MuSIC test battery for CP [
18]. Although analyzed and not confirmed in the original publication [
18], this score may still be influenced by the other factors apart from CI in MS, such as educational level. Therefore, this threshold may have introduced a bias towards higher premorbid intellectual capacity in the CP group. In contrast, CI of all included patients was confirmed through detailed NPA by an experienced neuropsychologist. Although in accordance with the Germany guideline on neuropsychological evaluation [
25], our criterion, again introduced in retrospect for quality assurance, of a documented performance score below − 1
z in at least two domains is liberal compared to other studies [
4‐
9,
11]. However, in these studies, cohorts with higher EDSS and longer disease durations were investigated (more than 10 years [
4,
6‐
9,
11] and more than 7 years [
5], respectively). Therefore, we believe that our CI patients were in the early stages of both MS and CI and that CI was milder than in patients of other studies. This may also explain why CI frequency was relatively low in our cohort compared to the frequency of 43–70% reported in the literature [
35], where, again, most cohorts were in later stages of MS than our cohort. To conclude, we could analyze a large group of patients with comparatively mild but thoroughly confirmed CI, whilst we cannot fully exclude that a few patients of the even larger group of patients, classified as CP, were actually CI, which would have decreased statistical power rather than leading to false-positive results.
Comparing CI to CP patients, differences of three parameters stood out—all well in accordance with the literature: overall disability (i.e., EDSS) [
12], WML volume, and thalamic atrophy [
4‐
11]. In addition, voxel-wise analyses confirmed that thalamic atrophy is the most striking difference between CI and CP patients within brain GM, whilst none of the areas of CI-related cortical thinning showed an effect size in this order of magnitude. Although associations of both WML load and thalamic atrophy with CI in MS had been demonstrated before by further evidence [
29,
36‐
40], we were impressed by the robustness of these effects compared to the remaining results. In addition, both parameters were strongly correlated, which had been reported before [
27,
41,
42] although not consistently [
43]. Against this background, we felt that further analyses focusing on these two parameters were justified. In a common model, only WML volume independently explained the occurrence of CI. Of note, the inclusion of the interaction term of WML volume and thalamic atrophy led to a better overall model fit indicating significant contributions of WML volume and its interaction with thalamic volume. In other words, thalamic atrophy seems to be less problematic than WML, whilst co-occurrence of the two is detrimental. In our opinion, the idea that thalamic atrophy is in part driven by WML through axonal transection of connecting fibers best explains our results [
27,
42,
44]. Such an effect of WML on thalamic atrophy is likely to depend on overall volume, eloquence of location, and destructive power (i.e., the extent of axonal transection in a given WML volume). These three are also very likely to increase the occurrence of CI in MS. Hence, we speculate that, in our statistical model, the main effect of WML volume reflects the spatial extent of WML load, whilst the interaction of thalamic and WML volume reflects the destructiveness of WML and their eloquence of location. In contrast, we were unable to demonstrate an independent main effect of thalamic volume in our cohort of patients, who were primarily in the early stages of MS. Yet direct thalamic damage through MS-related pathology exists [
45] and may become prominent in later stages.
With regard to the cerebral cortex, we observed wide-spread thinning in patients with CI compared to patients with CP. Although to a lower extent, we could relate cortical regions (primarily temporal and frontal lobe regions) more specifically to CI by correction for WML volume or overall disability (i.e., EDSS). The resulting effect was very robust but not in the order of magnitude of thalamic atrophy or WML. In MS patients in stages as early as our cohort, cortical atrophy was related to CI in only few studies [
9,
46], whilst this effect seems to become more robust in later stages [
4,
9,
47,
48]. Of note, a recent study demonstrated reduced cortical GM volume as the only significant MRI predictor of cognitive decline over a period of 5 years in a cohort at a later stage (compared to our cohort) with a mean symptom duration of 15 years at baseline [
29]. We, therefore, speculate that the effect of cortical thinning on CI in MS is smaller in the early stages and comes more and more into play in later stages.
We also found atrophy of corpus callosum in MS patients with CI compared to CP. As volumetry was performed from normalized T1w images after filling (inpainting) of WML, this effect cannot be fully explained by volume loss through WML within the corpus callosum. Atrophy of the corpus callosum has been consistently described before but in smaller cohorts and, again, in later stages of MS [
14,
15,
17]. Because of its strongest correlation with overall WM volume, we believe that the corpus callosum volume is an informative marker of WM pathology which is well conceivable given the densely packed long-range WM fibers within this WM structure. In line with this notion, MS-related changes in the tissue structure of the corpus callosum have been demonstrated by several studies using diffusion-tensor imaging techniques [
16,
49‐
52].
We acknowledge limitations of our study. The educational level was not available in our cohort and could hence not be accounted for. The tests used for detailed NPA are not commonly used in MS, which reduces comparability of our cohort with the other cohorts. We could relate our clinical data only to measures based on the conventional structural MRI, so that we were restricted to quantification of MS pathology by volumetry of brain structures and T2-weighted hyperintense WML, which certainly does not cover the whole spectrum of MS pathology. Studies using more advanced techniques have indeed found associations with other MR-based parameters such as cortical lesions detected through double inversion recovery sequences [
53,
54], or WM integrity damage detected through diffusion-tensor imaging [
36,
50,
55]. Furthermore, our region-wise analyses can only detect effects spatially overlapping across subjects; it does not account for network information [
56]. Hence, damage to different brain regions (across subjects but impeding the function of the same network) is not covered by our analysis.
In summary, we found a robust association of regional cortical thinning, corpus callosum atrophy, WML volume, and thalamic atrophy with CI. It seems that, in early MS, brain WML is the main driver of CI, whilst co-occurrence of thalamic atrophy intensifies the effect of WML. Yet, our data also demonstrate that, in the early stage of MS, comparatively mild CI is already associated with tissue damage in different brain compartments underlining the need to take CI as serious as physical symptoms. We believe that, for specific therapies of CI in MS, a deeper understanding is key. Studies on large cohorts, well-characterized (i.e. standardized diagnosis and quantification of CI), examined with multimodal imaging techniques, ideally longitudinally, will be necessary and probably only feasible in multicenter trials.
Compliance with ethical standards