Introduction

Multiple sclerosis (MS) is a chronic inflammatory demyelinating disease of the central nervous system that predominantly affects young adults and causes intermittent (relapsing-remitting MS, RRMS) or cumulative (secondary progressive MS, SPMS and primary progressive MS, PPMS) neurological dysfunction. Pathologically, MS manifests itself in multifocal lesions characterized by myelin destruction and repair, axonal loss and reactive astrogliosis. Magnetic resonance imaging (MRI) is a sensitive modality allowing users to visualize brain lesions located in white matter (WM) and is routinely used in the diagnosis and prognosis of MS. However, this type of lesion is only weakly correlated with clinical symptoms [102]. This well-established clinical/MRI paradox suggests that other types of brain damage may occur during the disease and be responsible for the clinical progression.

Over the last few years, MS has been shown to be characterized not only by focal white matter (WM) damage, but also by diffuse damage which can be detected using non-conventional MRI techniques such as magnetization transfer imaging and diffusion tensor imaging, or estimated by means of brain volume quantification [68, 102]. Brain volume loss, related to both grey matter and white matter atrophy, has been reported in patients with MS. The role of the grey matter (GM) decrease in this pathology has been the focus of increasing attention. First, a number of histopathological and MR studies have demonstrated that cortical lesions are more common than supposed [62]. Second, spectroscopic studies have shown that levels of N-acetylaspartic acid (NAA), a biomarker of neuronal integrity, decrease in the cortical GM of patients with MS [69]. Finally, grey matter atrophy has been found to be associated more closely with clinical disability than WM lesions [84].

A range of analytical tools has been developed to quantify brain volume more accurately. The present study had three goals: (1) to review the approaches currently used to quantify GM and WM atrophy in MS patients and the main results reported so far; (2) to describe the relationship between brain atrophy, other MRI markers and clinical disability (motor or cognitive), in order to discuss the clinical relevance of this new parameter, and (3) to examine differences in brain atrophy according to the main forms of MS. This review also sought to clarify the involvement and possible consequences of brain atrophy in MS disease progression.

Atrophy measurement in patients with MS

Methodological requirements for atrophy measurement

The recent focus on brain atrophy in a bid to improve understanding of the pathophysiology of MS has led to growing interest in measuring tissue loss in patients [68]. Numerous manual, semi-automated and automated methods based on conventional MRI are now available for measuring whole and regional brain volume. The optimum technique for measuring tissue volume is one which is reproducible, sensitive to subtle modifications, practical, fast and accurate. That said, although accuracy can be difficult to verify, it becomes less of a problem when the same method is applied to all the images and to all the sessions in a longitudinal analysis. Beyond the technique used for segmentation, many others parameters can affect the quantification of brain atrophy, like the pulse sequence and the resolution chosen for acquisition [51, 89]. Indeed, better resolution helps to reduce the partial volume effect [68]. Using a thin slice technique (1 mm cubed) helps to reduce the partial volume effect and consequently permits a better estimation of tissue volumes. Moreover, high contrast makes segmentation between the different cerebral compartments easier. According to the compartment of interest, different contrast could be chosen (cerebrospinal fluid/parenchyma or grey/white matter). A recent study pooled brain volume measurement data which had been acquired in five different centers using different acquisition sequences. Intercenter agreement was found for the brain atrophy estimations [54], suggesting that multicenter studies are quite feasible despite variations in sequences.

The first method used to estimate brain atrophy consisted of linear measurement of ventricles or other brain structure dimensions [91]. However, this method depends upon slice selection, slice thickness, the operator and inter-individual variability in cerebral morphology. The manual segmentation of brain structures from 3D acquisitions allows whole brain volume to be estimated, but this is a time-consuming method [39]. Semi-automated methods use algorithms of brain segmentation from 3D volumes and increase processing speed and reproducibility [51]. Nowadays, fully automated methods are most often used. This type of approach precludes operator errors and allows a large cohort of patients to be included, as well as healthy controls, in order to establish a benchmark for brain atrophy in normal aging [4, 92, 97]. Fully automated or semi-automated methods can be applied to a specific region of interest (such as the thalamus or the hippocampus) to obtain a regional brain volume [52]. Another manner of studying regional atrophy is to use an exploratory approach involving a whole brain voxel-to-voxel analysis [83, 88]. This type of approach does not need a priori hypotheses about regions of interest and allows the entire brain to be explored. The difficulty of the regional analyses is to obtain homogeneous GM boundaries despite the variation of the contrast within the image due to the MRI field nonuniformities. This problem makes the measurement of regional volumes difficult but it can be partly reduced using high resolution and high contrast images. Moreover, there are many tools that permit correction of this intensity inhomogeneity in MRI [100].

To assess disease progression more accurately and identify the extent of true inter-individual differences, longitudinal analyses of serial MRIs can be performed. One simple method consists in quantifying brain atrophy rates by segmenting the brain on serially acquired MRIs and calculating the change in volume over the scan interval [56]. Alternative methods for assessing the brain atrophy rate consist of image subtraction and direct quantification of volume change. One described method is based on the registration of serial MRIs and the calculation of the rate of cerebral atrophy directly from the registered scans [41]. Another fully automated method known as structural image evaluation using normalization of atrophy (SIENA) estimates the percentage of brain volume change (PBVC) by registering the baseline and follow-up MR image and subsequently determining the displacement of the brain edge between these two scans [92].

It seems that automated measurements of whole brain atrophy are nowadays preferred because of the inherent efficiency and reproducibility of these methods [73]. However, the operator input required in semi-automated measurements may lead to increase accuracy of the masking and segmentation steps. Thus, when specific anatomical regions are considered, semi-automated methods should be more valid because of the expert interaction ensuring quality control throughout the algorithm.

Finally, different segmentation techniques and acquisition approaches have been used in different quantification studies, thus, hampering the meaningful comparison of brain atrophy in MS patients. A summary of the results detailed below is presented in Table 1.

Table 1 Summary of studies reporting brain atrophy estimates for multiple sclerosis
Full size table

Measures of whole brain volumes

As mentioned above, many different segmentation methods are available for estimating brain volume using manual or automated techniques. Using a whole brain technique, Filippi et al. [39] manually segmented the cerebral hemispheres of 15 patients with MS (RR or SP) and reported a lower whole brain volume compared with that of 15 healthy controls (1,488 ml vs. 1,572 ml). More recently, various image analysis tools have been developed [73], including both automated and semi-automated algorithms, relying on either raw or normalized brain volume assessments. Whatever the technique used to estimate brain parenchymal volume (BPV), all studies report smaller volumes in MS patients than in age-matched control subjects [12, 26, 35, 43, 84, 85, 97, 101]. Ge et al. [43] investigated the question of inter-subject variability and found a difference between patients and control subjects, but only in the brain parenchymal fraction (BPF), as absolute BPVs were similar. The brain parenchymal fraction, defined as the ratio of BPV to total intracranial volume, has been used by many authors to measure whole brain atrophy [12, 25, 58, 84, 85, 89, 97], as it takes inter-individual variability in head size into account. These authors have reported a loss of BPF for MS patients estimated at between 2.7 and 4.8% compared with controls. One study showed that, even in young patients with MS (between 20 and 29 years), BPF is lower than that of age-matched controls [58]. Chard et al. [25] reported a decrease in BPF of about 3.3% in 26 young patients with MS compared with 27 healthy subjects. Using the same segmentation method (SPM99) and the same age range of patients (23–61 and 27–56 years), Sanfilipo et al. [84] revealed a loss of BPF of approximately 4.6% in 40 patients, whereas Tiberio et al. [97] found a smaller loss of 2.7% in 21 patients compared with healthy controls. Patients with SPMS were found to present greater atrophy than RRMS patients [95], and one explanation for this difference may lie in the MS subtypes that were assessed. Tiberio et al. only included patients with RRMS, whereas the group examined by Sanfilipo et al. included patients with either RRMS or SPMS. Another explanation for this discrepancy is the resolution used by authors during images acquisition. Indeed, Sanfilipo et al. used a thick slice of 2.5 mm, whereas Tiberio et al. used a thinner slice of 1.5 mm. As the partial volume effect may affect tissue volume estimation, it is possible that the volume loss of Sanfilpo et al. is overestimated compared to others.

Recent segmentation techniques have made it possible to distinguish between grey matter and white matter in the brain parenchyma [68]. In order to identify the contribution of each tissue type to the atrophy observed in MS, it is necessary to take into account the lesions that are essentially visible in WM on conventional MRI scans. The detection of lesions can be performed using either manual or automated methods and allows the estimation of GM or WM volumes to be adjusted. For most authors, T2-visible lesions essentially depict abnormal appearing WM and are added to normal appearing WM volume to avoid an artificial reduction in volume.

Studies measuring GM volume (GMV) or fraction (GMF, relative to intracranial volume) have all reported the same result: patients with MS present GM atrophy compared with control subjects [25, 48, 85, 87, 95, 97]. For example, Tedeschi et al. [95] studied a large cohort of 597 patients with RR-, SP- and PPMS and found a 3.9% decrease in GM volume in patients compared with 104 control subjects. However, the image resolution used in this study corresponds to a slice thickness of 4 mm that is sensitive to partial volume effect and affects the volume estimation. This GM atrophy was found even in early stages of the pathology [25], as minor tissue loss of 2.5% compared with controls was reported in 26 young patients with a disease duration of less than 3 years. The study by Quarantelli et al. [77] included patients with a mean disease duration of nearly 10 years and revealed a higher loss of GM tissue compared with controls (7.1%). Beyond the difference for resolution between these two studies (slice thickness of 4 vs. 1.5 mm), the difference in atrophy estimation between these two studies can partly be explained by disease duration and also suggests that GM atrophy should be used as a surrogate marker for the progression of the pathology. Very few studies have failed to find any significant difference between patients and controls on GMV [26, 42]. One of these had limitations due to the binarized segmentation method used to estimate GMV and WMV [42]. In the second report [26], the lack of any significant difference in GMF between 13 patients with RRMS and 9 healthy subjects may have been a consequence of the small cohort size. Another study reported that there was no significant difference in total GM volume between patients with RRMS (n = 147) and controls (n = 27), whereas neocortex volume was significantly lower in MS patients [50]. This result suggests that brain atrophy does not affect all brain structures in the same manner.

Concerning WM atrophy, many studies have reported a smaller WM volume for patients with MS compared with controls [25, 42, 53, 87, 95, 97]. The large study by Tedeschi et al. [95] reported a WMF loss of 3.7% for patients compared with controls. Ge et al. [42] quantified the GM and WM fractions in 30 patients with RRMS and found that brain atrophy in these patients compared with control subjects was mainly attributable to WM loss. However, as previously discussed, the method of segmentation used in this study was contestable. On the other hand, some studies have not found any significant difference in WM volumes between patients and controls. Ceccarelli et al. [23] reported that there was no difference in WMV between RRMS patients (n = 32) and controls (n = 16). Similarly, Quarantelli et al. [77] failed to detect any difference between 50 patients with RRMS and 54 controls for WM fraction. Sanfilippo et al. [84] found a 3.6% difference in WM volume between 41 patients with MS (RRMS and SPMS) and 18 healthy subjects, but this difference was not significant.

In summary, while the loss of GM volume is well established in MS patients, WM atrophy seems to be more difficult to detect. The main explanation for these heterogeneous findings is that axonal loss is present but is somewhat compensated for, in terms of volume, by gliosis and edema, two phenomena occurring in areas segmented as plaques and then added to the WM compartment [77]. Other factors, such as image resolution, contrast or slice thickness, may also affect GM/WM segmentation, thereby hampering the detection of WM loss [23].

Measures of brain atrophy progression

The time course of brain atrophy in MS can only be explored in longitudinal studies. The earliest and easiest way of evaluating the evolution of brain atrophy was to use semi-quantitative linear measures consisting of estimating lateral ventricle or third ventricle enlargement in 2D slices [91, 98]. In a prospective study of 301 patients with RRMS [91], third and lateral ventricle widths were determined on axial images, along a plane corresponding to the anteroposterior midpoint of the ventricle. In addition, the size of the corpus callosum was determined by outlining the margins of the structure on the mid-sagittal section. Brain width was taken to be the distance between two points on the cortical surface, measured on the same axial slice as the width of the lateral ventricles. The results for within-subject changes in brain atrophy at 1- and 2-year intervals revealed significant increases in lateral ventricle and third ventricle width and significant decreases in corpus callosum size and brain width. In a 1-year follow-up study of 17 subjects with a clinically isolated syndrome (CIS) suggestive of MS disease, Brex et al. [16] showed that ventricular enlargement had occurred at a significantly greater rate in the 9 patients who went on to develop MS than in those who displayed no further symptoms. More recently, other authors tried to validate these linear measures as surrogate markers of brain atrophy in 171 MS patients and 91 healthy controls, in order to use this method for daily assessment in clinical practice [19]. They measured three linear brain atrophy markers (intercaudate distance, third ventricle width and frontal horn width) and correlated these with brain parenchymal volume measured using a fully automated method of segmentation (SIENAX). They demonstrated that intercaudate distance measurements are the most valid brain atrophy measure for monitoring MS progression. Although the linear method can be used in clinical practice, it is especially sensitive to the selection and thickness of slices and thus prevents the precise quantification of brain atrophy.

Partial segmentation techniques have also been used to gauge the progression of brain atrophy in MS patients. This technique consists of extracting the brain parenchyma from the skull and cerebrospinal fluid (CSF) on a set of contiguous slices. Losseff et al. [67] applied this technique serially to 29 patients with RR- or SPMS who were imaged monthly from month 1 to month 9 and then at 12 and 18 months. The authors reported a decrease in volume beyond the 95% confidence limits for measurement variation in 16 patients after 18 months of follow-up. Stevenson et al. applied a similar method to 137 patients with PPMS who underwent two MRI brain scans 1 year apart and highlighted evidence of significant brain atrophy in these patients over the time of 1 year [93]. On top of the fact that these methods are time-consuming, slice thickness-dependent and operator-dependent, the above-mentioned studies did not compare MS patients with controls, thereby making it impossible to distinguish atrophy due to physiological aging from atrophy specifically due to the pathology.

Using whole brain segmentation, the rate of brain volume atrophy was measured over 2.5 years by Ge et al. [43], who recorded a loss of BPF of approximately 1.6% per year in 36 patients with MS (RR- and SPMS). For Kalkers et al. the rate of brain atrophy measured in 83 patients with different MS phenotypes (RR-, SP- and PPMS) was 0.7% per year [56]. However, neither of these two studies included healthy subjects. Chard et al. [26] compared rates of brain parenchymal atrophy in patients and controls and found a significantly higher rate for patients (1.2% per year vs. 0.2% per year for controls). Fox et al. [41] used a method based on the registration of serial MRI, wherein the rate of cerebral atrophy was calculated directly from registered scans. They revealed a rate of atrophy that was more than twice as high for 26 patients with MS (RR-, SP-, and PPMS) than it was for controls (0.8% per year vs. 0.3% per year). Another study calculated the change in brain atrophy over 3 months (estimated atrophy rate = 1.1% per year) in 128 patients with RRMS, and showed that this atrophy was detectable even within a very short period of time [49].

Several groups have assessed the rate of GM loss compared with that of WM loss and found significant GM atrophy only [33, 50, 86, 97, 99]. For example, over a period of 2 years, it was found that RRMS patients (n = 147) lost significantly more GMV (2.6 vs. 0.72%) than normal controls (n = 27), whereas WMV was stable and similar for both patients and controls [50]. In a 3-year longitudinal study [33] of patients with CIS, patients who developed MS presented a significant decrease in GMF (−3.3% over 3 years) but no changes in WMF.

In summary, all the longitudinal studies found significant loss of GMF with no concomitant loss of WM, although patients’ WM volumes were smaller than those of healthy subjects in the majority of cross-sectional studies. Thus, while WM atrophy is more apparent at baseline, GM atrophy is more dynamic over the period of observation. These results suggest that GM atrophy may be a more sensitive marker of early disease progression in MS and support the notion that WM atrophy could be an early event and could precedes GM atrophy. One possible mechanism suggested by some authors to account for delayed GM atrophy is neuronal degeneration, which occurs secondary to the retrograde or anterograde Wallerian degeneration observed following axonal transection in WM lesions [68]. An alternative explanation is that GM and WM atrophy are both independent, non-linear processes [26]. The two phenomena may either be related, with a variable delay, or else they may be independent. Finally, the absence of WM atrophy during disease progression may be due to the fact that tissue loss in WM associated with myelin and axon loss is present but obscured by inflammation and gliosis, whereas GM loss is more easily detectable, as it occurs without any reactive gliosis or inflammation [69]. Neuropathological studies have found that lesions located in the cortex contain fewer inflammatory cells than white matter lesions [60, 75]. This paucity of inflammatory cells in cortical lesions suggests that the molecules responsible for leukocyte adhesion to endothelial cells are expressed at decreased levels.

Measures of regional atrophy

Little is known about the mechanisms involved in atrophy in MS, and whether atrophy is a diffuse process affecting all brain regions equally or whether there is a preferential pattern of focal atrophy. Thus, several authors have investigated regional brain atrophy in patients with MS.

Recently, lobar atrophy, defined as the regional brain parenchymal (GM + WM) fraction of the frontal, temporal, parietal, and occipital lobes in each hemisphere, was assessed in 31 patients with MS (RR- and SPMS) and 16 healthy subjects [10]. In this study, patients showed significant regional cerebral atrophy in all the lobes and more particularly in the left temporal regional brain parenchyma. Carone et al. [21] combined a SPM99 segmentation process used to obtain GM and WM compartments with a parcellation method called SABRE (semi-automatic brain region extraction). Thirteen hemispheric regions were defined by means of the manual tracing of the main sulci and the Talairach coordinate system. The GM and WM volumes were then quantified for each region. This method was applied to 68 patients with RR- or SPMS and 39 controls. The authors revealed that all the regional GM fractions were significantly lower for patients, except those of the anterior temporal and inferior parietal regions. The largest mean percentage difference between the groups concerned the posterior basal ganglia/thalamus region (−19.3%). In contrast, for regional WM fractions, only 4 of the 13 regions that were tested were significantly lower for patients: the inferior parietal, medial inferior frontal, posterior temporal and inferior frontal regions.

Atrophy of the deep grey nuclei (thalamus and caudate nucleus) has been specifically studied by a number of authors [11, 31, 52, 94]. A neuronal density decrease of 22% in the thalamus of patients with MS has been described [31]. Houtchens et al. performed a specific segmentation of the thalamus of 79 patients with MS (RR- and PPMS) and 16 normal subjects [52]. They found that the thalamic volume was 16.8% lower in the MS group. This result is consistent with previous studies which have reported third ventricle enlargement in MS, suggesting atrophy of central structures such as the thalamus (central atrophy) [91]. Hippocampal volume has also been reported to be reduced in patients with MS (23 RRMS and 114 SPMS) compared with controls (n = 18) [90]. Pelletier et al. [74] focused on WM atrophy and divided the corpus callosum area into 6 subregions for 30 patients with RRMS and 25 healthy subjects. They revealed significant atrophy in the MS group compared with controls for the overall callosal area, as well as for each of the 6 subregions.

In addition to these studies based on the manual or semi-automated outlining of regions of interest, other studies have explored the whole brain using a voxel-based method. This type of method has been widely applied in order to pinpoint brain atrophy in normal aging [47, 57, 64], degenerative pathologies, such as Alzheimer’s disease [6, 29], and semantic dementia [37]. Using this voxel-based morphometry (VBM) approach, several studies have also reported bilateral thalamic atrophy [4, 24, 59, 88] in MS. In addition, most of these studies have revealed atrophy in other cerebral regions. Sepulcre et al. [88] investigated the regional distribution of GM atrophy in 31 patients with PPMS compared with 15 controls. While at baseline, they only showed bilateral thalamic atrophy compared with controls, GM loss had extended to the putamen, caudate and cortical areas of patients at the 1-year follow-up. However, it is important to stress that control subjects were not included in this longitudinal design. Another VBM study was conducted of PPMS patients (n = 46) and 23 controls [59]. Here, the authors reported significant GM atrophy in the right and left postcentral gyri, right middle frontal gyrus, left insula and thalamus bilaterally. For their part, Morgen et al. found that patients with RRMS (n = 19) showed GM volume reduction in the temporal and prefrontal cortex, essentially in the left hemisphere [70]. Another recent study of 51 patients with RRMS and 34 controls confirmed that GM volume is significantly decreased in the left frontotemporal cortex but also in the precuneus, anterior cingulate gyrus, and caudate nuclei bilaterally [76]. However, this study did not use a high resolution of image (slice thickness of 4 mm) that limits the quality of GM boundaries. Cortical thickness was also measured in two other studies. In 20 patients with MS (RR- and SPMS), average cortical width was found to be significantly lower, compared with controls, in the frontal and temporal cortex in early MS and in the motor cortex in patients with more advanced disability [83]. In a study of 115 patients (10 CIS, 32 possible MS, 42 RRMS and 31 SPMS), cortical thinning was found for possible MS, RRMS and SPMS patients in the precentral, postcentral, frontal and some occipital areas [20].

In summary, the thalamus may be particularly vulnerable to the destructive processes of MS, as atrophy of this structure has been found in the majority of studies. One explanation for this result is that this structure has rich reciprocal connectivity with many brain areas and may be susceptible to Wallerian degeneration due to demyelination and axonal loss in cerebral white matter [52]. Surprisingly, a recent study reported that only the neocortex volume was significantly smaller for RRMS patients [50], suggesting that the deep GM structures are not affected by atrophy in MS. However, over a subsequent 5-year period, the authors found that the GM volume loss involved not only the neocortex but also extended to the deep grey nuclei. Recent voxel-by-voxel analyses [59, 70, 88] have confirmed the presence of focal atrophy in the deep grey nuclei but also revealed that MS disease is also characterized by diffuse cortical atrophy, essentially located in the temporal and frontal regions. Thus, VBM analysis could be an important tool for improving our understanding of the topography, time course and clinical relevance of GM atrophy in MS.

Clinical relevance of brain atrophy

Correlations with lesion load

While there is evidence of brain atrophy in patients with MS, the exact pathological mechanisms that contribute to this atrophy are not yet clearly understood. However, some mechanisms can be inferred from the relationship between atrophy and lesion load, defined as the volume of specific WM lesions observed in MRI scans of MS patients. Lesions that are enhanced with intravenous gadolinium-based contrast agents in T1-weighted images represent active areas of inflammation and blood-brain barrier dysfunction [38]. These types of lesions only remain active for an average of 3–6 weeks [32], and thereafter no longer show up in T1-weighted images. Most of them, however, become permanent hyperintense lesions in T2-weighted images (T2 lesions) and represent a range of pathological changes (edema, gliosis, inflammation, demyelination, remyelination and axonal loss). About half of all gadolinium-enhanced lesions will ultimately persist as zones of severe hypointensity on T1-weighted images, characterized as T1 black holes (T1 lesions). These lesions represent irreversible tissue destruction and axonal loss [13]. Since T2 and T1 lesions reflect more definitive tissue damage than transient damage, these persistent lesions have been most intensively investigated in imaging studies of MS (see Table 2).

Table 2 Summary of studies reporting relationship between brain atrophy and other characteristics of multiple sclerosis
Full size table

Cross-sectional studies have sought to define the link between whole brain volume and lesion load. Most authors have shown that whole brain volume, defined as BPF, is significantly negatively correlated with T2 lesion volume [25, 40, 49, 82, 84] and T1 lesion volume [12, 25, 26, 40, 82, 84]. For instance, Chard et al. [25] found significant correlations in 26 patients with RRMS, with r values reaching −0.78 for T2 lesion volume and −0.59 for T1 lesion volume. Another study failed to find any significant correlations between BPF and lesion load [12, 43], but this included patients with both RR- and SPMS, whereas the majority of previous studies included only RRMS patients.

Longitudinal analyses allow us to assess predictors of GM and WM atrophy in order to gain a better understanding of the pathological substrate of these changes. Dalton et al. [33] conducted a study of 58 patients with CIS and reported that changes in T2 lesion and T1 lesion volume over 3 years were correlated with progressive brain atrophy, as measured by the change in BPF over the same period. For Hardmeier et al. [49], the baseline BPF of 138 RRMS patients showed a strong negative correlation with T2 lesion volume, while a weak correlation was found between the change in BPF over 3 months and T2 lesion volume at baseline, suggesting that lesion load is only a weak predictor of progressive brain atrophy. For their part, Rudick et al. [82] included 140 patients with RRMS in their study and reported that the baseline T2 lesion volume was inversely correlated with the percentage of BPF change over 2 years, but accounted for only 8.2% of the variance in progressive brain atrophy. However, two large longitudinal studies revealed strong correlations between brain atrophy and lesion load. In one large longitudinal study lasting 8 years and including 106 RRMS patients, T2 and T1 lesion volumes were both found to be significantly correlated with BPF at baseline and at subsequent time points, with Spearman r values ranging from −0.35 to −0.55 [40]. In the second study, conducted with 147 patients with RRMS over 5 years [50], the change in T2 lesion volume over 2 years was correlated with the percentage of brain volume change over 5 years.

Some authors have reported that it is GM volume, rather than WM volume, that is significantly negatively correlated with total T2 or T1 lesion volume [25, 42, 49, 50, 77, 84, 95, 99], suggesting that only GM reduction can be explained by variations in lesion load, while WM atrophy is partly independent of the genesis of lesions. This result was found in studies including patients with RRMS [25, 42, 49, 50, 77, 99], SPMS [84, 95] and CIS [33]. In these studies, the correlation with WM volume was always weaker than the correlation with GM volume and often below the significance threshold. For example, in a study including 26 patients with RRMS, GMF was correlated with T2 (r = −0.72) and T1 lesion volume (r = −0.52) but not with WMF (r = −0.31 for T2 lesions and −0.26 for T1 lesions) [25]. This relationship between GM volume and lesion load has also been reported for the thalamus: a significant inverse correlation was found between thalamus volume and T2 lesion (r = −0.62) and T1 lesion (r = −0.71) volume in 79 patients with different types of MS [52]. In patients with PPMS (n = 19), cortical atrophy was not significantly correlated with T2 lesion volume [35], whereas a significant relationship was found for RRMS patients (n = 38) in the same study. These data imply that neocortical atrophy is not necessarily dependent upon WM changes for all forms of MS.

Longitudinal studies have revealed that changes in GM volume are poorly predicted, if at all, by lesion load at baseline [26, 33, 86, 99]. For example, in a longitudinal study including 117 patients with RRMS, the change in GM volume over 18 months was not correlated with either T2 lesion or T1 lesion volume at study entry, whereas a significant association (r = −0.40, p < 0.001 for T2 lesions and r = −0.22, p = 0.015 for T1 lesions) was found between GM volume and lesion load at baseline [99].

Some authors have sought to determine whether the relationship between lesion load and cortical atrophy is the same for all cortical regions. In a study carried out with 20 RR/SPMS patients with a high T2 lesion load (>20 cm3), a significant thinning in the motor area was found and patients with a high T1 lesion volume (>3 cm3) showed a significant thinning of cortical areas in the temporal and frontal areas and along the motor cortex [83]. Charil et al. [27] correlated total WM lesion load with total cortical thickness in 425 patients with RRMS and reported that the most significant correlations were observed for the bilateral anterior cingulate gyrus, insula, the transverse temporal gyrus and association areas. Primary sensory, visual, and motor areas showed less significant relationships with total lesion load. As the authors found that lesion density was particularly high in periventricular WM, they concluded that these lesions might interrupt tracts originating from or projecting into cingulate and association areas. This could explain why these connected cortical areas are vulnerable to atrophy in MS. They added that the location of WM lesions might affect the pattern of focal atrophy, as suggested by the relationship they found between the WM lesions in the anterior and posterior parts of the brain (defined as y > 0 and y < 0 in stereotactic space) and preferential thinning of the cortex in anterior and posterior regions. However, in another study including 110 RR- and SPMS patients, the regional lesion volume, calculated for 26 regions, was not correlated with regional GMF [22], although it was significantly correlated with the total GMF in 9/26 regions for T2 lesion volume (largest r = −0.45 for right inferior frontal gyrus) and 5/26 regions for T1 lesion volume (largest r = −0.45 for right inferior frontal gyrus). The correlations between lesion load and overall GMF rather than regional GMF suggest that a distinct generalized disease process accounts for GM atrophy better than regionally retrograde or anterograde degeneration due to transection in WM lesions.

In summary, most studies examining the association of T2 lesions or T1 lesions with whole or regional brain atrophy have reported that lesion load essentially contributes to the development of GM atrophy, rather than WM atrophy, in patients with MS. Chard et al. have proposed the following hypothesis to explain this relationship between T2 lesions located in WM and GM atrophy [25]. First, as mentioned above, the decrease in GM volume may reflect retrograde or anterograde degeneration extending along fiber tracts following axonal transection in WM lesions. Second, axonal damage may be associated with demyelination and leads to axonal and neuronal atrophy. However, an analysis of the literature reveals that lesion load has only a limited predictive value for the subsequent development of GM atrophy, suggesting that GM pathology cannot simply be a consequence of WM damage, and that MS-related damage to these different compartments may follow distinct pathological pathways. GM atrophy may also stem from independent discrete and diffuse GM lesions. While the majority of lesions are found around the ventricles, others are located in or adjacent to cerebral cortical grey matter [62]. These cortical lesions, identified on FLAIR images, have been found to account for more variance in brain atrophy than WM lesions [5]. However, these cortical lesions are less likely to be detected than WM lesions in conventional MRI sequences, compared with histopathological studies [44, 60]. Recently, new MRI sequences have been developed in order to improve the detection of cortical MS lesions and should allow us to better understand its relationship with cortical atrophy [45, 71]. To explain the absence of correlation with the decrease in WM volume, Chard et al. [26] have suggested that gliosis, inflammation and edema of lesions may alter WM volumes and obscure the relationship between lesion load and WM atrophy.

Correlation between atrophy and disability

There is no straightforward link between the focal WM lesions characteristic of MS disease and disability progression [15]. Thus, brain atrophy measures have been made in an attempt to explain disability better than established measures of focal lesions. Disability in MS is generally assessed by means of the expanded disability status scale (EDSS), following a complete neurological examination [61]. More recently, a functional composite score called the multiple sclerosis functional composite (MSFC) has been used in many clinical trials to measure the progression of the disease [81]. The MSFC is a multidimensional clinical outcome measure [80] which, in addition to a test of motor functions, includes a cognitive assessment of attention, working memory and processing speed, through the paced auditory serial addition test (PASAT). A summary of studies examining the relationship between brain atrophy and disability is given in Table 2.

In cross-sectional studies, a significant negative correlation between BPF and the EDSS score [35, 58, 89] or disease duration [12, 43, 49, 58, 84, 89] has been widely reported. In addition, for Fisher et al. [40] who studied a group of 106 patients with RRMS, the change in BPF between year 2 and the year 8 follow-up was negatively correlated with the changes in EDSS and MSFC over the same period. Moreover, the change in BPF from baseline to year 2 was correlated with EDSS and the MSFC at the year8 follow-up. The authors concluded that the change in BPF was a significant predictor of disability status.

According to other authors, relationships between overall atrophy and disability are essentially due to GM atrophy. A study of 57 patients with RR- or PPMS showed that, at baseline, neocortical volume correlated with EDSS scores [35]. Using regression modeling, Sanfilipo et al. [84] demonstrated that GM atrophy was more closely related to neurological disability, as measured by the EDSS score, than WM atrophy and T1 and T2 lesion volumes.

However, many cross-sectional or longitudinal studies have failed to find a significant correlation between overall brain atrophy and any score of clinical disability [25, 39, 41, 42, 56, 77, 97]. Different arguments have been advanced by authors to explain this lack of significance: some of these studies included patients with mild disability and used short spans of scores which limited the potential for detecting meaningful correlations [25, 56, 97]. The heterogeneity of the MS group and the short follow-up may explain the absence of significant correlations in some studies [41], as follow-up has to be sufficiently long to observe significant disability progression over time in longitudinal analyses. Moreover, some authors have suggested that the EDSS score used to measure disability should be associated more closely with the atrophy of the spinal cord [39]. An EDSS score above 3 is more dependent upon locomotor impairment associated with spinal damage than upon cerebral hemisphere dysfunction. Significant negative correlations have already been found between upper cervical cord atrophy and the EDSS score [65]. Some studies have found that the correlation between BPF and MSFC is consistently stronger than the correlation between BPF and EDSS. This result suggests that brain atrophy in MS could be linked not only to disability but also to cognitive deficits. Therefore, more extensive psychometric test batteries, in addition to EDSS and MSFC estimates, would be useful for gauging the consequences of brain atrophy in MS.

Some studies have investigated the correlation between disability and regional brain atrophy. A longitudinal analysis of 79 recently diagnosed MS patients revealed that a worsening EDSS score was significantly correlated to atrophy in periventricular regions and neocortical regions in the parietal, temporal and occipital areas [55]. Locatelli et al. [66] measured regional BPF for each lobe and showed that the regional BPF of the frontal lobes was associated with the EDSS score of 39 patients with RRMS. More specifically, superior frontal gyrus and temporal cortex atrophy was found to be linked with EDSS in a separate study of 35 patients with MS (RR- and SPMS) [7]. Chen et al. [28] reported that MS patients (RR- and SPMS) who displayed progressive disability over 1 year (10 patients) had a higher rate of GM thickening in the precentral gyrus and parietal areas than 20 stable patients. For their part, Charil et al. [27] estimated that the cortical thickness of 425 patients with RRMS was equivalent to 0.08 mm per EDSS unit and was located essentially in the bilateral prefrontal cortex, left inferior temporal gyrus and anterior cingulate region. Sailer et al. [83] divided their patients into three groups according to their EDSS score: they found that patients with more severe disability (EDSS ≥ 6, n = 6) presented additional regions of atrophy located in the motor area compared with patients with a low disability score. In addition, they reported that longer disease duration was associated with additional atrophy in the same regions and found a very strong correlation between the EDSS score and the specific area of the motor cortex (r = −0.69).

Lastly, it seems that clinical disability is only weakly correlated with whole brain atrophy. However, studies investigating regional atrophy have revealed that the EDSS score could be strongly associated with focal atrophy, especially in regions involved in motor functions.

Correlation between atrophy and cognition

Cognitive impairment affects approximately 50% of patients with MS [78] and typically involves memory, attention, and the speed of information processing [14]. Some studies have shown that these deficits may be detectable during the early phase of MS [2, 36]. The frequency of cognitive disturbances for each clinical form has not been entirely established and the natural history of these disorders has yet to be extensively studied [17]. While correlations between cognitive status and WM lesion load have been shown to be only modest [18, 79], recent measures of brain atrophy appear to be more closely associated with cognitive deficits [17]. The assessment of the correlation between neuropsychological deficits and the quantification of brain atrophy may provide new insights into the mechanisms underlying the development of cognitive impairments in patients with MS (see Table 2 for a summary of studies).

In a 2-year follow-up study of 53 patients affected by early RRMS, Zivadinov et al. [103] found that the development of overall cognitive impairment during the study period was predicted by changes in BPV. Concerning specific cognitive functions, another study of 37 patients with MS (RR or SP) found that higher BPF predicted higher scores on the PASAT and SDMT, two tests assessing attention, working memory and processing speed [9]. These researchers also noted that ventricular enlargement was strongly negatively correlated with SDMT scores. An association between ventricular enlargement and SDMT scores was also reported in a different set of 37 patients with MS (RR or SP) [30]. Using a test that probes various domains of attention, Lazeron et al. [63] measured the processing speed and performance accuracy of 32 patients with different subtypes of MS (11 RR-, 13 SP- and 8 PPMS). They revealed that a greater reduction in brain volume was significantly associated with a slower processing speed but not with reduced performance accuracy in MS, as the latter remained stable [63].

Some authors have concentrated their investigations on GM atrophy instead of whole brain atrophy. In a study of 41 patients affected by RRMS, Amato et al. [1] found that cortical atrophy was present only in cognitively impaired MS patients (n = 23) and was correlated with poorer performances on several neuropsychological tests. These authors observed that neocortical atrophy was positively associated with impairment in verbal memory, verbal fluency, and attention/concentration. After following 28 of these 41 patients over 2.5 years [3], they also reported that normalized cortical volume (NCV) changes were significantly higher in cognitively deteriorated patients (n = 12) than in stable patients. Moreover, in the deteriorated patients, NCV changes were negatively correlated with performance on the verbal fluency test. For these authors, progressive neocortical grey matter loss is relevant to MS-associated cognitive impairment and may represent a sensitive marker of deteriorating cognitive performance in RRMS. In a large group of 82 patients with RR- and SPMS, Benedict et al. [8] showed that neocortical volume was significantly correlated with many neuropsychological measures (verbal and visuo-spatial memory, processing speed, working memory), with r values ranging from 0.29 to 0.58. In this study, third ventricular width, reflecting thalamic atrophy, was also retained as a predictor of cognitive impairment. Another study of 79 patients with different forms of MS demonstrated moderate to strong relationships between thalamic atrophy and impairment in numerous cognitive domains, including processing speed and visuospatial memory [52]. In these studies, cognitive impairment in MS would appear to be related to cortical as well as subcortical atrophy.

According to some authors, the correlation between brain volume and cognitive deficits depends on whether GM or WM tissue is considered. For example, GMV was found to be lower for RRMS patients exhibiting cognitive impairment (n = 12) than for cognitively preserved MS patients (n = 16), whereas no significant difference was found for WM volume [46]. A recently published study of 40 patients with MS (RR- and SPMS), suggested that WM volume loss is more closely associated with processing speed and working memory deficits, whereas GM volume loss is more closely linked to verbal memory impairment [85]. WM volume loss was found to be associated with deficits in temporary storage and the manipulation of new information, both tasks requiring rapid communication between neuronal networks. This result supports the notion that, in MS disease, the interruption of fiber tracts contributes to the disconnection of networks engaged in working memory tasks. This hypothesis has also been validated by the study by Pelletier et al. [74] who found that poorer performance on tasks exploring interhemispheric transfer of information was related to atrophy of the corpus callosum in 30 patients with RRMS. Whereas some authors have reported a correlation between neocortical volume (excluding subcortical structures) and tests exploring attention or speed processing [3], GMF was not found to be correlated with this type of test by Sanfilipo et al. [85]. This discrepancy suggests that subcortical structures and the neocortex are not involved in cognitive performances in precisely the same way. Associations between regional measures of cortical atrophy and neuropsychological dysfunction need to be established in order to verify these hypotheses.

In a study which followed a cohort of 79 patients with RR/SPMS for 2.2 years, the regional analysis revealed that poorer PASAT scores at follow-up were significantly correlated with atrophy, but only around the ventricles [55]. However, most of these 79 patients exhibited high PASAT scores and an improvement within the study period, suggesting a learning effect and only minimal cognitive deterioration. Using a cross-sectional design, Locatelli et al. [66] measured regional BPF for each lobe in 39 patients with RRMS and found that the PASAT score was correlated with atrophy in the neocortex, especially the frontal lobes. The involvement of frontal atrophy in cognitive deficits had already been suggested by correlations between bilateral superior frontal gyrus atrophy and impairment in verbal learning, spatial learning, attention and conceptual reasoning in 35 patients with MS (RR- and SPMS) [7]. In a voxel-by voxel analysis, Morgen et al. [70] found that the low PASAT scores of 19 RRMS patients were associated with reduced volume of the bilateral prefrontal, precentral, superior parietal and right cerebellar cortex. WM volume, however, did not show any significant correlation with cognitive performance in this study. These regional analyses suggest that neocortical atrophy plays an important role in cognitive impairment in MS, although some investigations have found that ventricular enlargement, considered to reflect atrophy of central structures, is also associated with neuropsychological deficits [8, 30]. Tekok-kilic et al. [96] explored associations between regional GM atrophy and neuropsychological functions in 59 patients with MS (RR-, SP- and PPMS) after controlling for the influence of third ventricular enlargement. They showed that central and cortical atrophy independently predicts the presence of slower processing and memory impairment in MS.

To conclude, brain atrophy, especially GM atrophy, seems to be involved in cognitive impairments in all subtypes of MS. Both cortical and subcortical atrophy can contribute to these deficits, but as yet very few studies have investigated the relationship between regional brain atrophy and specific neuropsychological deficits.

Atrophy and subtypes of MS

Although many studies have provided consistent evidence that brain atrophy is a relevant feature of MS, the involvement of either GM or WM atrophy in the different clinical forms of MS (RRMS, SPMS and PPMS) has received rather less attention. While the majority of studies have focused on RRMS, others have compared the evolution of whole or regional brain atrophy in the different phenotypes of MS. Some studies have reported significance or a trend for smaller whole brain volumes in patients with SPMS compared with age-matched RRMS patients [12, 43, 89, 101]. Vrenken et al. [101] reported that normalized brain volume (NBV: brain volume normalized to skull size) was significantly lower in 18 patients with SPMS (1,406 ml) than it was in the 36 RRMS patients (1,473 ml). The same study failed to find any significant difference between the NBV of PPMS patients and that of RRMS patients. De Stefano et al. [35] also compared PPMS and RRMS patients (38 RR- vs.19 PPMS) with equivalent disease duration and reported that NCV was similar between these two subtypes of MS.

Tedeschi et al. compared GMF and WMF in a large cohort of patients with different subtypes of disease (n = 597) and found significantly lower GMF (51.4 vs. 50.0%) and WMF (34.2 vs. 32.8%) for SPMS patients (n = 427) compared with RRMS patients (n = 140), with intermediate values for PPMS patients (50.7% for GMF and 33.1% for WMF). The results of the PPMS group may be attributable to the difference in disease duration between PPMS and SPMS patients, but the relevant data are not available. A reduction in GM volume for SPMS patients (n = 11) compared with RRMS patients (n = 23) has been also reported specifically for the hippocampal region [90].

Concerning the evolution of atrophy in the course of the disease, Ge et al. found a similar rate of brain parenchymal loss in a group of 36 patients with RR- or SPMS (−1.5% per year for RRMS vs. −2.0% per year for SPMS). In a cohort of 83 patients with MS (42 RR-, 21 SP-, 20 PPMS), Kalkers et al. [56] analyzed the rate of brain volume loss over time using two markers of atrophy: BPF as a measure of whole brain atrophy and ventricular fraction (VF: ventricular volume/intracranial volume) as a measure of central atrophy. They reported that, at baseline, the patients with RRMS had the highest BPF and the lowest VF. However, the annualized rate of brain parenchymal atrophy did not significantly differ between subtypes of patients (0.7% per year), whereas the annualized rate of ventricular enlargement was higher for SPMS patients, though there was only a trend towards significance. Dalton et al. [34] confirmed the latter result with their investigation of ventricular enlargement over 1 year in 30 patients with early RRMS (duration less than 1 year), 41 patients with established RRMS and 23 patients with SPMS. They reported ventricular enlargement in all MS patients over the 1-year period, but with significantly greater enlargement for the SPMS group. Fox et al. [41] also found a difference in the brain atrophy rate between subgroups of MS, the largest decrease (0.9% per year) being for patients with SPMS and the smallest (0.6% per year) for patients with PPMS. However, the differences were not statistically significant and were very close to the rate reported by Kalkers et al. [56].

The relationship between whole brain atrophy and lesion load reveals that GM volume is correlated with lesion load for RRMS [84, 95] and SPMS patients [25, 42, 49, 50, 77, 99]. For patients with PPMS, it seems that the number of gadolinium-enhanced lesions observed over a 1-year period is predictive of WM atrophy development, whereas GM matter atrophy cannot be predicted by lesion load [86]. The use of proton density-weighted images to detect T2 lesion volumes instead of FLAIR images, plus phenotypic variation, may explain the difference.

Ge et al. found a significant negative correlation between brain atrophy and EDSS scores in patients with SPMS but not in those with RRMS. They concluded that brain tissue loss is only correlated with clinical disease progression for SPMS patients. De Stefano et al. reported that the correlation between NCV and EDSS was much stronger in PPMS patients (r = −0.64) than in RRMS ones (r = −0.27). In contrast, another study, carried out with 43 PPMS patients, reported that EDSS scores were correlated with WM atrophy, but not GM atrophy [87]. These two apparently different results cannot be directly compared and contrasted because one group estimated total GMF (including subcortical GM) while the other group only used cortical GM volume.

This dichotomy between central and peripheral atrophy has already been noted in relation to cognitive impairments. While third ventricle width is negatively correlated with SDMT scores for both RR- and SPMS patients, the positive correlation between NCV and SDMT is stronger for SPMS than for RRMS [8].

Most studies investigating regional atrophy have pooled data from different types of MS, while others have focused on RRMS patients. Only one study directly compared the evolution of regional brain parenchymal atrophy in the three main subtypes (20 RRMS, 19 SPMS, and 31PPMS) with that of a control group [72]. Brain atrophy was found for all subtypes of patients in the lateral fissure, insula, middle temporal gyrus, frontal lobes, and some regions of the parietal and occipital lobes. In the progressive forms, the authors reported that the brain atrophy tended to involve different additional structures compared with the RRMS patients. However, no distinction was made between regional GM and WM atrophy in this study and direct comparisons between different subtypes of patients were not performed.

To conclude, the quantification of brain atrophy shows that SPMS patients have smaller brain volumes than RRMS and PPMS patients. However, many studies have only identified a trend between groups, suggesting that the difference is only subtle. In contrast, longitudinal analyses have revealed a similar rate of atrophy independent of MS subtype. We can suppose that a RRMS patient with a high degree of atrophy at baseline would convert more rapidly to an SP form than other RRMS patients. Moreover, whereas brain atrophy is more closely correlated with WM lesions in the remitting form of MS, it is more closely associated with disability in the progressive form. Finally, the analysis of regional atrophy suggests that phenotypic variation can be associated with damage to different brain structures, although the only study to have made this comparison failed to distinguish between GM from WM atrophy. In future studies, regional analysis of atrophy must be carried out in order to specify the precise nature of the relationship between regional atrophy and disability or cognition, according to the form of the disease.

Conclusion

The emergence of effective MR imaging techniques and methods analysis has helped to increase our knowledge about the pathophysiology of MS. The studies described show that white matter is not the only tissue to be affected during disease progression and highlight the existence of a focal and a diffuse grey matter damage during MS. Moreover, this review emphasizes the relevance of GM atrophy describing its role in clinical and cognitive deterioration. Thus, grey matter atrophy should become a more important parameter in future clinical studies of patients. However, more investigations are still required to identify which cerebral regions are most sensitive to atrophy.