OUP user menu

Brain atrophy in clinically early relapsing–remitting multiple sclerosis

D. T. Chard, C. M. Griffin, G. J. M. Parker, R. Kapoor, A. J. Thompson, D. H. Miller
DOI: http://dx.doi.org/10.1093/brain/awf025 327-337 First published online: 1 February 2002


Brain atrophy measured by MRI is a potentially useful tool for monitoring disease progression in multiple sclerosis. The location, extent and mechanisms of brain atrophy in early disease are not well documented. Using quantitative MRI, this study investigated whole brain, grey and white matter atrophy in clinically early relapsing–remitting multiple sclerosis and its relationship to lesion measures. Data came from 27 normal control subjects (14 females and 13 males, mean age 36.1 years) and 26 subjects with clinically definite multiple sclerosis (18 females and eight males, mean age 35.1 years, mean delay from first symptom to scan 1.8 years, median Expanded Disability Status Scale score 1.0). All had three‐dimensional fast spoiled gradient recall (3D FSPGR), T1‐weighted pre‐ and post‐gadolinium‐enhanced and T2‐weighted scans. The 3D FSPGR images were automatically segmented into grey and white matter and cerebrospinal fluid using SPM99. 3D FSPGR hypo‐intense, T2 hyper‐intense, T1 hypo‐intense and T1 post‐gadolinium‐enhancing lesion volumes were determined by semi‐automatic lesion segmentation. The SPM99 output was combined with the 3D FSPGR lesion segmentations to quantify tissue volumes as fractions of total intracranial volumes, producing values for the brain parenchymal fraction (BPF), white matter fraction (WMF) and grey matter fraction (GMF). Comparing multiple sclerosis with control subjects, BPF, GMF and WMF were significantly reduced (P < 0.001 for all tissue fractions). Using Pearson correlations, T2 hyper‐intense and T1 hypo‐intense lesion volumes were inversely related to BPF (T2 r = –0.78, P < 0.001; T1 r = –0.59, P = 0.002) and GMF (T2 r = –0.73, P < 0.001; T1 r = –0.53, P = 0.006), but not WMF (T2 r = –0.30, P = 0.134; T1 r = –0.26, P = 0.199). T1 post‐gadolinium‐enhancing lesion volumes were not correlated with any fractional volumes. These results indicate that significant brain atrophy, affecting both grey and white matter, occurs early in the clinical course of multiple sclerosis. The lack of correlation between lesion load measures and WMF suggests that pathological changes in white matter may occur by mechanisms which are at least partly independent from overt lesion genesis in early multiple sclerosis.

  • Keywords: multiple sclerosis; normal controls; MRI; brain; atrophy
  • Abbreviations: BPF = brain parenchymal fraction; 3D FSPGR = three‐dimensional fast spoiled gradient recall; EDSS = Expanded Disability Status Scale; GM = grey matter; GMF = grey matter fraction; SPM = Statistical Parametric Mapping; WM = white matter; WMF = white matter fraction


There has been considerable work quantifying the severity and extent of brain tissue damage in multiple sclerosis using MRI and relating this to clinical outcome. This has included the use of brain atrophy techniques, which have been employed in relapsing–remitting (Rudicket al., 1999; Simonet al., 1999; Lukset al., 2000), progressive (Stevensonet al., 2000) and both (Losseffet al., 1996; Liuet al., 1999; Foxet al., 2000; Geet al., 2000) clinical subtypes of multiple sclerosis. These studies have shown that atrophy occurs at a significantly faster rate in multiple sclerosis subjects compared with a normal healthy population.

It is less clear how early in the course of multiple sclerosis, atrophy appears and whether it affects both white matter (WM) and grey matter (GM). In addition, measures of atrophy and estimates of lesion loads are only weakly correlated in established multiple sclerosis (Paolilloet al., 2000). Whether this atrophy is the result of earlier lesion‐induced damage, raising the possibility that it is more directly related to lesion measures in the early stages of multiple sclerosis, remains to be determined.

In order to address these issues, we studied a group of patients with clinically definite relapsing–remitting multiple sclerosis of short duration. A method using SPM99 (Statistical Parametric Mapping; Wellcome Department of Cognitive Neurology, Institute of Neurology, Queen Square, London) (Ashburner and Friston, 1997, 2000) was developed to quantify brain volume as a fraction of total intracranial volume, along with the WM and GM tissue fractions. There is marked intersubject variability in brain volume in addition to age and gender effects (Pfefferbaumet al., 1994; Blatteret al., 1995; Passeet al., 1997; Xuet al., 2000), all of which could mask disease effects, particularly in cross‐sectional studies. However, inter‐subject variability is markedly reduced by adjusting for variation in the total intra‐cranial volume, and this enables the effects of age and gender to be accounted for better in statistical modelling (Pfefferbaumet al., 1994; Blatteret al., 1995; Jenkinset al., 2000). We explored age and gender effects in normal healthy young to middle‐aged controls before looking for the presence of any additional disease effect and any relationship to either disease duration or lesion volume parameters in the multiple sclerosis cohort.



The normal control data came from a cohort of 27 subjects (14 females and 13 males; mean age 36.1 years at volume scan, range 23.2–55.2 years) with no previous history of neurological disease or other medical conditions. The multiple sclerosis data came from a cohort of 26 subjects (18 females and eight males; mean age 35.1 years when first scanned, range 24.1–48.4 years). The patients were required to have a disease duration <3 years when recruited for the study. At the time of scanning, the mean delay from first symptom onset was 1.8 years (range 0.5–2.8 years). All patients had clinically definite (Poseret al., 1983) relapsing–remitting multiple sclerosis (Lublin and Reingold, 1996) and an Expanded Disability Status Scale (EDSS) (Kurtzke, 1983) score of <3 (median EDSS at first assessment and scan 1.0, range 0–2.5). None of the multiple sclerosis subjects had received interferon β at any stage prior to scanning, nor had they been treated with corticosteroids within the previous month. The project had approval from the ethics committee of the National Hospital for Neurology and Neurosurgery, Queen Square, London, UK. All subjects gave informed consent.

Scan acquisition

All scans were performed on a GE Signa 1.5 tesla scanner (General Electric Medical Systems, Milwaukee, Wis., USA). Data from four scans were included in the study. The first was a 3D inversion‐prepared fast spoiled gradient recall (3D FSPGR) sequence [TR (repetition time) 16 ms, TE (echo time) 4.2 ms, inversion time 450 ms, matrix 256 × 160, FOV (field of view) 300 × 225 mm (interpolated during reconstruction to a final in‐plane resolution 1.2 × 1.2 mm), NEX (number of excitations) 1, with 124 × 1.5 mm slices covering the whole brain]. The second was a dual echo FSE (fast spin echo) sequence (TR 2000 ms, TE 14/98 ms, NEX 2, in‐plane resolution 0.9 × 0.9 mm with 28 × 5 mm slices covering the whole brain). The third and fourth were pre‐ and 20 min post‐gadolinium‐enhanced [0.3 mmol/kg body weight of Magnevist (Schering, Berlin, Germany)] T1‐weighted spin echo sequences (TR 540 ms, TE 20 ms, NEX 1, in‐plane resolution 0.9 × 0.9 mm with 28 × 5 mm slices covering the whole brain). The second, third and fourth scans were acquired during the same scanning session; the first (3D FSPGR) scan was acquired during an earlier, separate session. Sessions were separated by a mean of 9 days (range 2–60 days), during which time subjects did not report any additional clinical events. All data were acquired using a standard quadrature head‐coil.

Image analysis

Tissue segmentation

Images were transferred to a network of Sun workstations (Sun Microsystems, Palo Alto, Calif., USA) for further processing. The 3D FSPGR were automatically segmented into images representing the probability of any given voxel containing GM, WM, CSF and other tissues using SPM99 with maximum image inhomogeneity correction applied.

Lesion segmentations

Lesions were segmented on the 3D FSPGR using a semi‐automatic local thresholding technique, part of the Dispimage image display package (D. L. Plummer, Department of Medical Physics and Bioengineering, University College London Hospitals NHS Trust, London, UK) (Plummer, 1992). Lesions were seen as hypointensities on this sequence; their presence was confirmed by reference to the proton density‐weighted FSE images, where they appeared as hyperintensities. Lesions were also segmented using the same technique applied to the proton density‐weighted FSE images and T1‐weighted pre‐ and post‐gadolinium‐enhanced images, which provided T2 hyperintense, T1 hypo‐intense and gadolinium‐enhancing total lesion volumes (loads) respectively.

Volume determination

The spinal cord cut‐off location was determined visually on the 3D FSPGR image as the most rostral slice not containing cerebellum. The outputs from SPM99 and 3D FSPGR lesion segmentations were processed using in‐house software along with the user‐specified cord cut‐off level to quantify tissue volumes. The lesion mask over‐rode all SPM99 tissue classifications, otherwise a voxel was classified as GM, WM, CSF or other tissue, dependent on which mask had the greatest probability (maximum likelihood) at that location. This generated mutually exclusive masks for each tissue, i.e. a given voxel was partitioned to one mask only. Results were assessed as fractions of total intracranial (TI) volume as determined by adding the GM, WM, lesion and CSF volumes. Brain parenchymal fraction (BPF) was calculated as GM, WM plus lesion volume divided by TI volume. The white matter fraction (WMF) was calculated as WM plus lesion volumes (virtually all lesions were located in the WM) divided by TI volume. The grey matter fraction (GMF) was calculated as the GM volume divided by the TI volume. The lesion fraction was calculated as the lesion volume divided by the TI volume.

Statistical analysis

Statistical analyses were performed using SPSS 9.0 (SPSS, Chicago, Ill., USA). The extent of lesion mis‐classification by SPM99 was assessed by comparing GM, WM and CSF volumes with and without lesion masks in all 26 multiple sclerosis subjects (GM, WM and CSF volumes with lesions included as WM were used for the definitive analyses of BPF, GMF and WMF).

The relationships between all lesion volumes, as estimated from the different scan acquisitions, age and disease duration (both estimated in days) were assessed by Pearson correlations. The associations between EDSS and tissue fractional volumes, lesion volumes, age and disease duration were investigated using Spearman correlations. Spearman correlations were chosen to assess the relationships between EDSS and other parameters, as the former is a non‐linear measure of disability and as such may not be truly considered to be parametric.

The effects of age and gender on BPF, GMF and WMF in both control and multiple sclerosis subjects were investigated using general linear models with gender as a fixed factor, age as a covariate and an interaction term between gender and age. The effect of subject type (multiple sclerosis or control) was assessed using two models. The first included gender and age as a covariate within gender. The second included gender and subject type (multiple sclerosis or control) as fixed factors, age as a covariate within gender and within subject type, and T2 lesion volume and disease duration as covariates. The differences between these two models’ fit to the data were assessed to estimate the overall significance of disease effects.

For multiple sclerosis subjects, the relative effects of T2 hyper‐intense, T1 hypo‐intense and T1 post‐gadolinium‐enhancing lesion volumes and disease duration on fractional tissue volumes were assessed using Pearson correlations. 3D FSPGR lesion volumes were not investigated because they were used to correct fractional volumes and were therefore not independent. Two‐tailed significance values were estimated for correlations. A P‐value <0.05 was regarded as significant.


Lesion volumes

In the 26 multiple sclerosis subjects, the mean total lesion volumes were as follows: 3D FSPGR hypo‐intense lesion volume 6.73 ml (range 0.47–39.88 ml); T2 hyper‐intense lesion volume (on the proton density‐weighted FSE sequence) 6.67 ml (range 0.31–29.02 ml); T1 hypo‐intense lesion volume (on the pre‐contrast T1‐weighted SE sequence) 0.86 ml (range 0.00–6.36 ml); and gadolinium enhanced lesion volume 0.79 ml (range 0.00–8.65 ml). The 3D FSPGR lesion volume constituted 0.5% of the average TI volume (lesion fraction 0.005), of which 52% was classified by SPM99 as GM, 41% as WM and 7% as CSF. Subsequent results use tissue fractions with all the lesions reclassified as WM. Figure 1 shows an example of the output masks from SPM99 with lesions displayed separately. Virtually all lesions on the 3D FSPGR and T2‐weighted scans were located in WM.

Fig. 1 Slices from a multiple sclerosis subject’s T1‐weighted 3D FSPGR scan and the corresponding segmentation masks derived from SPM99 and lesion contouring. Clockwise from top left: original 3D FSPGR image; GM mask; WM mask; lesion mask; CSF mask; and ‘other’ mask.

Correlations between lesion volumes, brain tissue fractions and clinical parameters

There was a very strong correlation between the 3D FSPGR and T2 lesion volumes (r = 0.965, P < 0.001). Moderately strong correlations were found between 3D FSPGR and T1 hypo‐intense lesion volumes (r = 0.601, P = 0.001) and between T2 and T1 hypo‐intense lesion volumes (r = 0.753, P < 0.001). Gadolinium‐enhancing lesion volumes did not correlate significantly with any other lesion measure. None of the lesion volume measurements were correlated with age or disease duration. No significant correlations between EDSS and any lesion or fractional tissue volume or age were found. A modest correlation between EDSS and disease duration was found (rs = –0.424, P = 0.031).

Fractional tissue volumes: age and gender effects in control and multiple sclerosis subjects (Tables 1 and 2)

Table 1 gives the mean values for each fractional volume in males and females in both control and multiple sclerosis subjects. Table 2 presents results of modelling age and gender effects in normal controls. Age and gender effects were not found to be significant for multiple sclerosis subjects and are not shown. Average fractional volume reduction in normal controls per year (averaged between genders and proportional to normal control values at age 36.1 years, the mean age of our control cohort) were –0.2% for BPF, –0.3% for GMF and –0.2% for WMF. Figures 2, 3 and 4 show fractional tissue volumes in both control and multiple sclerosis subjects plotted against age.

Fig. 2 BPF plotted against age for 27 control (NC) subjects and 26 subjects with multiple sclerosis (MS).

Fig. 3 GMF plotted against age for 27 control (NC) subjects and 26 subjects with multiple sclerosis (MS).

Fig. 4 WMF plotted against age in 27 control (NC) subjects and 26 subjects with multiple sclerosis (MS).

View this table:
Table 1

Mean (standard deviation) fractional tissue volume by gender in normal controls and subjects with multiple sclerosis

GenderSubject type
Normal controlMultiple sclerosis

Results are from 13 male (mean age at scanning 36.3 years, range 27.2–52.7 years) and 14 female (mean age at scanning 35.8 years, range 23.2 to 55.2 years) control subjects and eight male (mean age at scanning 34.7 years, range 24.8–48.1 years) and 18 female (mean age at scanning 35.4 years, range 24.1–48.4 years) subjects with multiple sclerosis.

View this table:
Table 2

Age and gender effects in normal controls

Tissue fractionOverall factor P valuesFactor parameter estimates
GenderAgeAge × genderGender*Age
GMF0.0160.0010.0500.070 –0.002–0.001

Age × gender represents the interaction between these factors, i.e. the difference in age effects between genders. Parameter estimates are given with standard error in parentheses. Age parameter estimates are change per year. BPF model, r2 = 0.560; GMF model, r2 = 0.573; WMF model, r2 = 0.428. *Estimates for females compared with males; a negative parameter implies a smaller value in females.

Fractional tissue volumes: absolute disease, T2 lesion volume and disease duration effects (Table 3)

Table 3 shows the results of modelling for absolute disease effects per se along with disease duration and T2 lesion volumes, while allowing for both age and gender effects. The inclusion of multiple sclerosis factors in addition to age and gender effects significantly improved the fit of the model to the data for BPF, GMF and WMF (all P < 0.001), indicating the overall significance of multiple sclerosis disease effects. For the average multiple sclerosis subject in the cohort (age 35.1 years, disease duration 1.8 years, T2 lesion load 6.67 ml), the following estimates for a disease effect were obtained: – 0.030 for BPF, a reduction of ∼3.5% compared with age‐matched normal controls; –0.016 for GMF, a reduction of ∼2.8%; and –0.014 for WMF, a reduction of ∼4.9%. The Pearson correlation between disease duration and GMF was also found to be of borderline significance (r = –0.390, P = 0.049) but correlations of disease duration with BPF and WMF were not significant.

View this table:
Table 3

Multiple sclerosis disease effects

Tissue fractionBPFGMFWMF
Factor P values
 Composite MS effect*<0.001<0.001<0.001
 MS × age0.0120.2830.038
 T2 lesion load<0.001<0.0010.043
 Disease duration0.0230.0300.790
Factor parameter estimates
 MS–0.047 (0.028)–0.019 (0.026)–0.045 (0.021)
 MS × age0.002 (0.001)0.001 (0.001)0.001 (0.001)
 T2 lesion load–0.003 (0.001)–0.003 (0.001)–0.001 (0.000)
 Disease duration–0.014 (0.006)–0.012 (0.006)0.001 (0.004)

Parameter estimates are given with standard error in parentheses. *Estimates for the overall significance of multiple sclerosis (MS) disease effects, i.e. the effects of multiple sclerosis per se, T2 lesion load and disease duration, allowing for age and gender. Estimates were made by comparing models containing all disease‐related parameters with those that did not. Estimates for multiple sclerosis compared with control subjects; a negative parameter implies a smaller value in multiple sclerosis subjects. The row ‘MS’ represents the effect of the disease per se without adding the effects of other disease‐related parameters, such as T2 lesion load and disease duration. Multiple sclerosis × age represents the interaction between multiple sclerosis and age, i.e. the relative effects of multiple sclerosis at different ages. Multiple sclerosis × age and disease duration parameters are change per year. T2 lesion loads are change per ml. Age and gender effects were allowed for in the model but are not presented. BPF model, r2 = 0.709; GMF model, r2 = 0.618; WMF model, r2 = 0.433.

Fractional volumes: comparative lesion volume effects (Table 4)

Table 4 shows the correlations of tissue fraction measures with lesion volumes in multiple sclerosis subjects. BPF and GMF were both strongly correlated with T2 hyper‐intense and moderately with T1 hypo‐intense, but not with T1 post‐gadolinium enhancing lesion volumes. WMF was not correlated with any measure of lesion volume. Figure 5 shows tissue fractions plotted against T2 lesion loads.

Fig. 5 BPF, GMF and WMF plotted against T2 lesion volume (ml) for subjects with multiple sclerosis.

View this table:
Table 4

Pearson correlations of tissue fractions with total lesion volumes in multiple sclerosis subjects

Tissue typeTotal lesion volumes
T2 hyper‐intenseT1 hypo‐intenseT1 Gd‐enhancing
r P r P r P


This study reveals significant white and grey matter brain atrophy in a cohort of clinically definite multiple sclerosis subjects with the shortest clinical disease duration studied to date. Whereas the present study population had a mean disease duration of 1.8 years, previous studies have investigated clinically definite multiple sclerosis populations with a mean duration from first symptoms from 4.0 (median from 3.3) years and above (Losseffet al., 1996; Liuet al., 1999; Rudicket al., 1999; Simonet al., 1999; Foxet al., 2000; Geet al., 2000; Molyneuxet al., 2000; Paolilloet al., 2000). Before considering the significance of the atrophy findings in multiple sclerosis, the lesion load results will be reviewed, and technical, age and gender aspects relevant to the measurement and analysis of brain volumes will be discussed.

Comparison of lesion load measures

As reported by other investigators (Shahet al., 1992), the lesion volume values obtained by 3D FSPGR and T2 lesion volumes correlated very strongly, and with very similar absolute values. The 3D FSPGR sequence used in this study was T1‐weighted but, unlike the standard 2D T1‐weighted spin echo sequence, produced much larger lesion volumes, which closely matched those found on T2‐weighted scans, the traditional gold standard sequence for depicting lesions. The increased resolution (Filippiet al., 1998; Molyneuxet al., 1998) and the different contrast mechanisms of 3D FSPGR versus standard 2D T1‐weighted spin echo probably account for this result (Rovariset al., 1999).

The more modest correlation between standard 2D T1 hypo‐intense and T2 lesion volumes (compared with that observed between 3D FSPGR and T2 lesion volumes) suggests that the former is identifying a subgroup of lesions with potential to provide complementary information. Given that such lesions have a greater degree of axonal loss (van Walderveenet al., 1998), there is a good case for collecting this data, in addition to total T2 load, in natural history and treatment trial studies.

Technical aspects of the brain volume measurement and analysis

Previous work has established the reproducibility of the SPM99‐based technique, applied to the same type of T1‐weighted 3D images acquired in the present study, for measuring tissue fractions in normal controls (Chardet al., 2001), estimating the coefficient of variation for BPF to be 0.5%, for GMF 0.7% and for WMF 1.2%. Lesion segmentations in the current study were not as reproducible as this (coefficient of variation was 7.2% in a measure–remeasure study of 3D FSPGR lesion volumes on eight of the multiple sclerosis subjects). However, given the very small proportion of intracranial volume classified as lesion tissue (lesion fraction 0.005), it is not surprising that this additional variability was not great enough to have prevented the detection of disease‐related effects upon the much larger tissue fractional volumes.

The general linear models presented in this work made use of interaction terms. As previous work has indicated that age and gender may interact to affect brain volume measures (Coffeyet al., 1998; Xuet al., 2000), an interaction term between age and gender was included. Disease and age interactions were also explored, along with T2 lesion volume and disease duration. This was predominantly to ensure the robustness of our overall disease effect conclusions rather than provide a definitive investigation of contributing factors. Their significance in this study indicates that they cannot be ignored, but they should be explored again in larger, more heterogeneous cohorts.

Age and gender effects of brain volume measures

Within the age range of the cohort studied, significant GM and WM atrophy occurred with ageing in normal controls, and GM atrophy predominated over WM (Table 2). Several previous studies have also detected predominantly GM loss with age (Limet al., 1992; Pfefferbaumet al., 1994; Passeet al., 1997; Brunettiet al., 2000), although some have shown both GM and WM atrophy, the latter predominating (Harriset al., 1994; Guttmannet al., 1998). The present work also indicated significant gender differences in the proportion of each tissue type. It can be concluded that both age and gender should be allowed for before looking for a disease effect.

Brain tissue fraction findings in multiple sclerosis

After allowing for age and gender effects, BPF was significantly reduced in the multiple sclerosis group, with a greater proportional reduction in WMF (mean ∼–4.9%) than GMF (mean ∼–2.8%). In controls, age‐related changes were more apparent for GMF (mean –0.3% per year) than WMF (mean –0.2% per year). Therefore, atrophy early in the clinical course of multiple sclerosis does not appear to represent a simple acceleration of the usual age‐related changes.

The reduction in BPF in the present early relapsing–remitting multiple sclerosis study accords with a recent preliminary report of a cohort of subjects with clinically isolated syndromes consistent with demyelination who had ventricular volumes measured over a period of 1 year (Brexet al., 2000). That study found that the nine subjects who went on to develop multiple sclerosis, compared with the eight who did not, had significantly greater increases in ventricular size over 1 year. It therefore seems that brain atrophy occurs in the early stages of multiple sclerosis.

There has been little previous work investigating WM and GM separately for the presence of atrophy in multiple sclerosis. Liu and colleagues found that WM tissue loss was present in a cohort of multiple sclerosis patients with a median disease duration of 7 years and that this correlated with disability (r = –0.37, P = 0.018) (Liuet al., 1999). There were no significant correlations between EDSS and any of the lesion or fractional tissue volume parameters in the present cohort. However, the study was confined to patients with a short disease duration and mild disability, which limited the potential for detecting meaningful correlations. This negative result may also be partially due to the limitations of this measure of disability and it may prove fruitful to explore others which may be more sensitive to small differences in clinical status. The possibility for early atrophy to predict future clinical outcome needs to be investigated with prospective follow‐up.

Relationship between brain tissue fractions and lesion measures

There was no correlation between the enhancing lesion loads and measures of atrophy. Enhancing lesions are correlated with active inflammation on pathological studies (Katzet al., 1993; Brucket al., 1997); however, only a small proportion of lesions will display inflammation at a single time point. A realistic evaluation of the relationship between this MRI marker of inflammation and atrophy would require an assessment of serial data obtained from multiple enhancing scans obtained at frequent intervals (noting that new lesions enhance on average for only 1 month) (Harriset al., 1991; Thompsonet al., 1991). There is little such data available in patients with early disease; data from cohorts with longer disease durations have revealed only modest correlations (Coleset al., 1999; Simonet al., 1999; Saindaneet al., 2000). Long‐term follow‐up of the present cohort will characterize the evolving relationships between lesions and atrophy.

Inspection of Fig. reveals a robust relationship between T2 hyper‐intense lesion volumes and both BPF and GMF, but not WMF. The Pearson correlations (Table 4), confirm the association for both T2 hyper‐intense and T1 hypo‐intense lesion volumes. The stronger relationship was with T2 volume, which gave an r2 of 0.61 for BPF and 0.53 for GMF, suggesting that >50% of the BPF and GMF reduction can be explained by variations in T2 load. These lesion load effects were confirmed after allowing for disease duration by the model presented in Table 3.

There are several potential explanations for this relationship between T2 lesions, which are virtually all located in WM, and atrophy located in the GM. First, the decrease in GMF may reflect both retrograde degeneration to the cell body and Wallerian (predominantly anterograde) degeneration extending along fibre tracts (Simonet al., 2000) following axonal transection in WM lesions (Trappet al., 1998). Secondly, axonal damage without transection but associated with demyelination per se (Fergusonet al., 1997) may lead to axonal and neuronal atrophy (Yinet al., 1998). Thirdly, GM lesions, though rarely seen on MRI, are commonly found at post‐mortem (Brownell and Hughes, 1962; Kiddet al., 1999; Boet al., 2000); such lesions are associated with demyelination and possibly with local axonal and neuronal damage, and their total volumes may be correlated with WM lesion loads. They might also cause subtle changes in signal intensity which affect segmentation between the cortex and CSF such that GM volume appears reduced. Finally, both the genesis of overt lesions and tissue atrophy may be manifestations of some other common pathogenic mechanism which has yet to be elucidated.

The absence of a clear correlation between lesion volumes and WMF, with a non‐significant Pearson correlation (Table 4), and the limited trend observed in the model presented in Table 3, despite the greater extent of WMF reduction when compared with GMF, could be interpreted as an early global WM disease process resulting in atrophy that is at least partly independent of the genesis of overt lesions. Alternatively, water content changes due to low‐grade inflammation or oedema, along with glial proliferation in the normal‐appearing WM (Tourtellotte and Parker, 1968; Allen and McKeown, 1979), might alter WM volumes and obscure a relationship between lesion volume and WMF loss attributable to similar mechanisms proposed for the GMF (see above). It is also worth recalling that WMF reproducibility was not as good as that for GMF, and this may contribute to the masking of relationships. Studies making use of a larger sample with greater lesion volume heterogeneity are needed.

Paolilloet al. (2000) explored the relationship between atrophy and lesions in a cohort with relapsing–remitting multiple sclerosis and a mean disease duration of 5.6 years. They found a correlation between a regional cerebral hemisphere volume measure and T1 hypo‐intense (r = –0.48, P < 0.001), but not T2 load. T1 and T2 load correlated equally with corpus callosum area (r = –0.53, P < 0.001 and r = –0.52, P < 0.001, respectively). The present study indicates a stronger relationship between atrophy and T2 rather than T1 hypo‐intense lesion load in a cohort who have earlier disease. The subgroup of T1 hypo‐intense lesions is associated with greater axonal loss (van Walderveenet al., 1998) and should correlate more strongly with atrophy if the link mechanism is secondary tract degeneration. However, the T1 load was relatively small in this early disease cohort and the stronger correlation of GM atrophy with T2 lesion volumes suggests that, in early multiple sclerosis, the link between WM lesions and GM atrophy may have additional mechanisms (see earlier discussion).

Relationship between brain tissue fractions and disease duration

Disease duration was found to have a significant additional effect upon GMF but not WMF. This was both observed as a Pearson correlation and confirmed after allowing for additional factors (Table 3). A possible explanation for this result is that there is on‐going GM atrophy that is initiated during acute lesion genesis but is not dependent on further lesion activity. The observation that on‐going axonal loss occurs in chronically demyelinated plaques may be relevant here (Korneket al., 2000). The disparity between GMF and WMF has been considered above, with all the same reasons applying to these findings as we have discussed in relation to lesion volume and tissue fraction correlations.

Atrophy and cell loss

The relationship between atrophy and the loss of oligodendrocytes, axons and neurones may not be linear. One post‐mortem study (Pakkenberg and Gundersen, 1997) in previously healthy subjects reported that an age‐related 9.5% reduction in neurone number was associated with a 12.3% reduction in neocortex volume and 28.0% reduction in WM volume. Whereas such a reduction in WM volumes with age is not universally supported by in vivo MRI studies and fixation artefacts may modify in vitro findings (Milleret al., 1980), the study highlights that there may be tissue‐specific differences in atrophy and that atrophy might not reflect truly the extent of axonal and neuronal loss: the degeneration of other cell types may also contribute to volume loss. In addition, as noted above, in multiple sclerosis volume loss may be masked by oedema, cellular infiltration and proliferation. Furthermore, axonal and neuronal atrophy associated with demyelination has been reported (Yinet al., 1998) and, given that remyelination occurs (Prineas and Connell, 1978; Korneket al., 2000), this effect may be reversible. Serial studies would help to clarify the temporal variability in atrophy, including the degree of reversibility.

The disease‐associated volume reduction in our subjects, of ∼2.8% in GMF and ∼4.9% in WMF, compares with a larger reduction in the neuronal marker N‐acetyl‐aspartate reported in a magnetic resonance spectroscopy study of the normal‐appearing WM (mean 7%) and GM (mean 11%) in a subset of the same early multiple sclerosis cohort included in the present study (Kapelleret al., 2001). Other investigators have also reported proportionately greater reduction in N‐acetyl‐aspartate than in brain volumes in relapsing–remitting multiple sclerosis (Collinset al., 2000). These discordant findings could indicate that axonal and neuronal loss is more marked than is indicated by fractional tissue volume loss alone (for reasons already discussed) or that the more marked reduction in N‐acetyl‐aspartate may indicate a transient effect of axonal and neuronal dysfunction rather than of absolute axonal and neuronal loss on the concentration of this metabolite (Narayanaet al., 1998; Maderet al., 2000).

Study limitations

While this study clearly indicates that atrophy can be detected early in the clinical course of multiple sclerosis, it is based upon cross‐sectional data; prospective follow‐up of the present cohort is proceeding in order to characterize the subsequent evolution of atrophy. Further, while age, disease duration and lesion volumes were all found to influence disease‐related tissue specific atrophy significantly, the trends detected need to be explored again in more heterogeneous and larger samples. It is important not to extrapolate the results outside the parameter ranges covered by the present study: while the assumptions of linear relationships between parameters, such as age and tissue fractions, appear satisfactory for the present cohort, this cannot be assumed to be true for a greater range of values. In addition, when interpreting the magnitude of disease effects, it should also be remembered that various methods of assessing tissue volume may yield different absolute values depending on the techniques used for segmentation and scan acquisition.

Brain tissue segmentation in any disease processes may be complicated by both changes in tissue signal intensity characteristics and, in the case of multiple sclerosis, the presence of lesions. There are a number of further segmentation strategies that may be able to improve upon that used in this study, although each method will be associated with differing susceptibilities to disease‐related segmentation bias. In addition, there is no accepted standard technique for measuring brain tissue volumes, so it is not possible to validate and calibrate any segmentation method fully. Given this, it is important that multiple segmentation methods are implemented independently and applied to different scan acquisition methods as each will have its own advantages and disadvantages.

While the multiple sclerosis and control subjects were well matched for age, there was some imbalance in gender. However, this should not have affected our results markedly as gender and age effects were allowed for in the models used. Indeed, neither age nor gender effects were detected in the multiple sclerosis cohort, probably because they were obliterated by overlying disease‐related effects (as seen by inspection of Figs –4).


The present study shows that both GM and WM fractional volume loss occurs with age, to a greater degree in GM than in WM, and that there are significant gender differences in the proportional amounts of each tissue. After allowing for these effects, there is evidence that significant fractional brain volume loss has already occurred early in the clinical course of relapsing–remitting multiple sclerosis and that, unlike age‐related changes, this is proportionally more evident in WM. Significant relationships were found between lesion volumes and the degree of tissue atrophy in GM but not WM. This suggests that WM pathology may occur by mechanisms which are at least partly independent of overt lesion genesis.


We thank the subjects who kindly agreed to take part in this study, Dr Martin King for valuable statistical advice, the Multiple Sclerosis Society of Great Britain and Northern Ireland for programme grant support to the NMR Research Unit and Schering AG for sponsorship of D.C.


View Abstract