ORIGINAL ARTICLES Imaging Correlates of Axonal Swelling in Chronic Multiple Sclerosis Brains Elizabeth Fisher, PhD,1 Ansi Chang, MD,2 Robert J. Fox, MD,3 Jean A. Tkach, PhD,4 Therese Svarovsky, BS,2 Kunio Nakamura, BS,1 Richard A. Rudick, MD,3 and Bruce D. Trapp, PhD2 Objective: T2-weighted magnetic resonance imaging is a sensitive tool for monitoring progression of multiple sclerosis, but it does not provide information on the severity of the underlying tissue damage. Measurement of T1 hypointensities and magnetization transfer ratio (MTR) can potentially distinguish lesions with more severe tissue damage. The objective of this study was to use image-guided pathology to determine histological differences between lesions that are abnormal only on T2-weighted images versus lesions that are abnormal on T2-weighted, T1-weighted, and MTR images. Methods: A total of 110 regions were selected from postmortem magnetic resonance images of 10 multiple sclerosis patients. Regions were classified into three magnetic resonance imaging–defined categories: normal-appearing white matter; abnormal on T2-weighted image only (T2-only); and abnormal on T2-weighted, T1-weighted, and MTR images (T2T1MTR). Myelin status, lesion activity, astrocytosis, serum protein distribution, axonal area, and axonal loss were evaluated histopathologically. Results: Comparisons between groups showed that T2T1MTR regions were more likely to be demyelinated (83% compared with 55% of T2-only regions) and more likely to be chronic inactive lesions (68% compared with 0% of demyelinated T2-only regions). There was no difference between T2-only and T2T1MTR regions in axonal area, but there was a significant difference in axonal count, indicating that axons in the T2T1MTR regions were enlarged relative to those in T2-only regions. Interpretation: Axonal swelling and axonal loss were major pathological features that distinguish T2T1MTR regions from T2-only regions. Ann Neurol 2007;62:219 –228 Magnetic resonance imaging (MRI) is an objective and sensitive tool for diagnosing and monitoring multiple sclerosis (MS) patients.1–3 MRI has also been proposed as a tool for understanding MS pathogenesis in vivo. However, its usefulness in this regard is limited because conventional MRI is inherently nonspecific for distinct pathological processes. The nonspecific nature of MRI lesions is also one likely explanation for the weak correlations between lesion volumes and the progression of clinical disability.4 – 8 MRI signal contrast arises from differences in water and lipid content and the macromolecular environment of adjacent tissues. In theory, all focal MS pathological processes, including blood– brain barrier (BBB) breakdown, inflammation, demyelination, axonal loss, and gliosis, lead to an alteration of water and lipid content, and hence conspicuous lesions on T2-weighted MRI (provided the affected region is large enough to be detected). Additional MR sequences often are applied to distinguish underlying pathology, or at least to determine lesion severity. For example, lesions that are hypointense on T1-weighted images, or “black holes,” are generally considered to have more severe tissue destruction than lesions that are isointense on T1-weighted images, based on correlations with axonal loss,9 –11 stronger correlations to disability,12,13 and association with chronic stage of lesion development.14 T1 hypointense regions have also been shown to have lower N-acetylaspartate compared with T1 isointense lesions, and normal white matter using magnetic resonance spectroscopy.15,16 Similarly, decreased magnetization transfer ratio (MTR) in lesions has been shown to be correlated to increased disability,17,18 reduced N-acetylaspartate,19 demyelination,20 and axonal loss.11 Previous postmortem and in vivo (biopsy) studies have identified the histopathological characteristics of different MRI features, such as T2 hypointensity,14,21 gadolinium enhancement,9,22,23 T1 hypoin- From the Departments of 1Biomedical Engineering and 2Neurosciences, Lerner Research Institute; 3Mellen Center for Multiple Sclerosis Treatment and Research, Cleveland Clinic; and 4Department of Radiology, Case Western Reserve University/University Hospitals Cleveland, Cleveland, OH. This article includes supplementary materials available via the Internet at http://www.interscience.wiley.com/jpages/0364-5134/ suppmat Received Oct 10, 2006, and in revised form Jan 31, 2007. Accepted for publication Feb 2, 2007. Current address for T. Svarovsky: Department of Pathology, Wisconsin National Primate Research Center, Madison, WI. Published online April 11, 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/ana.21113 Address correspondence to Dr Fisher, Department of Biomedical Engineering ND20, Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, OH 44195. E-mail: email@example.com © 2007 American Neurological Association Published by Wiley-Liss, Inc., through Wiley Subscription Services 219 tensity,9,11,14,24 and decreased MTR.11 Most of the evidence to date supports the idea that one can pathologically subclassify lesions in vivo using a single MRI data set according to the following rules25: (1) T2 hyperintensities are pathologically nonspecific, and therefore represent MS lesions in all phases of pathogenesis; (2) gadolinium-enhancing lesions are regions with breakdown of the BBB and possibly infiltration of hematogenous leukocytes; (3) T1 hypointensities are MS lesions with axonal loss and severe tissue destruction; and (4) lesions with decreased MTR represent MS lesions with demyelination and axonal loss. In support of this general concept, correlations between lesion volumes and disability have been found to be stronger for T1 lesion volume than for the total T2 lesion volume12; however, these correlations are still quite modest in most studies, ranging from 0.22 to 0.54.12,13 Notably, the lack of strong correlations is also related to difficulties with measurement of clinical disease progression, in addition to limitations of cranial MRI. However, the correlations are not much stronger when an objective measure of disease severity, such as brain atrophy, is used as the dependent variable. Correlations between T1 lesion volume and brain atrophy are about the same as or lower than correlations between T2 lesion volume and atrophy, in the range of 0.4 to 0.5, and T1 lesion volume is not an independent predictor of subsequent brain atrophy.26 –28 The findings for lesion MTR measurements are similar. MTR is decreased in MS brains and varies widely in MS lesions, from extremely low to nearly normal values.29 MTR is reduced in gadolinium-enhancing lesions because of acute inflammation, and it subsequently recovers over time in a subset of lesions, but remains low in others.30 Variations in lesion MTR also relate to degree of demyelination,20 remyelination,31 and degree of axonal loss.11 Correlations between mean lesion MTR and disability vary widely between studies, from nonexistent19,32 to relatively strong (⫺0.4 to ⫺0.7).17,18 In general, the expectations that measurements of the more severe lesions, such as T1 hypointensities with decreased MTR, would be strongly correlated with disease progression have not been met. These weaker than expected correlations led to persistent questions on the ability to differentiate lesions based on MRI. Such a capability would be highly beneficial in the evaluation of new neuroprotective therapies, as a way to distinguish potentially reparable lesions from irreparable lesions. The purpose of this study was to use image-guided pathology to directly determine the major histological differences between lesions that are abnormal only on T2-weighted images versus lesions that are abnormal on T2-weighted, T1-weighted, and MTR images. 220 Annals of Neurology Vol 62 No 3 September 2007 Subjects and Methods Overview MS brain tissue was acquired through the tissue procurement program established at the Cleveland Clinic Foundation. Patients and their families signed an institutional review board– approved advance directive to donate brain and spinal cord tissue for research purposes. Demographic and disease-related information for each patient was entered into a database. The same protocol was followed for each patient: Within a few hours of death, the cadaver was brought to the imaging facility for an MRI examination followed by rapid autopsy. The brain and spinal cord were removed, and one cerebral hemisphere was immediately fixed in 4% paraformaldehyde for at least 4 weeks. After fixation, the hemisphere was reimaged in a custom-designed slicing box for coregistration purposes. The brain tissue was sliced coronally in the box, and each slice was numbered, stored, and subsequently photographed. The cadaver images were registered to the postfixation images so that the image planes corresponded with the tissue slices. Lesions were segmented and classified in the cadaver images using automated image analysis programs. Using the registered MRIs, we generated region maps to indicate regions of interest (ROIs) for histological analysis. Magnetic Resonance Imaging Details Imaging was performed on a 1.5-Tesla MR scanner (VISION; Siemens, Erlangen, Germany). The MRI protocol consisted of a three-dimensional T1-weighted magnetization prepared rapid acquisition gradient-echo image, a T2weighted fluid-attenuated inversion recovery image (FLAIR), a T2-weighted fast spin-echo image, a T1-weighted spinecho image, and an MTR image calculated from proton density–weighted three-dimensional gradient-echo images acquired with and without a magnetization transfer pulse. The magnetization prepared rapid acquisition gradient-echo image was acquired with 1 ⫻ 1 ⫻ 1mm resolution, and all the other images had 0.9 ⫻ 0.9 ⫻ 3mm resolution. Detailed sequence parameters are provided in Supplementary Table 1 (see supplementary online materials). For the postfixation image, the fixed cerebral hemisphere was placed in a customdesigned slicing box before imaging to aid in image to tissue colocalization. The precision-machined polycarbonate box has sliding walls that lock in place to securely hold the tissue during imaging and slicing. Two walls of the box have vertical knife-guide slots positioned 10mm apart with 2mm gelfilled cylindrical wells above and below each slot. When fixed tissue is imaged in the box before slicing, the gel markers are detectable on the MRI and indicate the orientation and position of each slice plane. Image analysis was performed using software developed in-house (BIP; Department of Biomedical Engineering, Cleveland Clinic Foundation, Cleveland, OH). The analysis for the cadaver images consisted of brain segmentation, brain parenchymal fraction (BPF) calculation, lesion segmentation, and lesion classification. Brain segmentation and BPF calculation were performed on the FLAIR image using a fully automated program, as described previously.28,33 T2 lesions were segmented in the brain-masked FLAIR image using a modified version of the iterated conditional modest (ICM) algorithm.34 The segmented T2 lesion regions were used to guide the automated segmentation of lesions in the T1 and MTR images. First, the T2 lesion regions were overlaid on coregistered T1 and MTR images. For each image type, the intensity statistics of the normal-appearing brain tissue in the neighborhood surrounding each lesion were calculated, and thresholds for abnormal tissue within each lesion region were determined. Abnormal was defined as any voxel in the T1 or MTR image that was colocalized within a T2 lesion and had an intensity less than the mean intensity minus one standard deviation of the local nonlesion tissue. Therefore, classification of regions as abnormal on T1 and MTR images was based on low intensity relative to neighboring normalappearing white matter (NAWM), not by global thresholding. Lesions were subsequently classified into two types: regions of T2 hyperintensity that were normal appearing on T1 and MTR (T2-only), and regions of T2 hyperintensity that were hypointense on T1 and MTR (T2T1MTR). The first type represented regions hypothesized to be of lesser severity, whereas the second type represented regions hypothesized to be of greater severity. For each brain, ROIs within each type of lesion and within NAWM were delineated for the generation of region maps (see description later in this article). FLAIR, T1, and MTR contrast ratios were calculated for each ROI as the mean intensity within the region divided by the mean intensity within all of the NAWM ROIs for the same brain. Image analysis for the postfixation images consisted of detection of the markers in the slicing box, localization of the cut planes, and registration with the cadaver images. Thresholding and connected-components labeling were used to identify the cerebral hemisphere and the markers. The bestfit plane equations were determined from sets of markers using a program encoded with the dimensions of the slicing box. The postfixation image was registered to the cadaver images using the Iterative Closest Point algorithm to determine the optimal three-dimensional affine transform (three rotations ⫹ three translations ⫹ uniform scaling).35 The equations of the cut planes were then used to extract the individual planes from the analyzed cadaver image volumes that corresponded to each tissue slice. Lesion maps were generated for each tissue slice from the corresponding MRI planes. A 10mm grid was overlaid on the image planes to provide a frame of reference. The outlines of ROIs that corresponded to each type of lesion (T2only and T2T1MTR) and NAWM were transferred to the region maps. These maps were then used to guide tissue sampling for the histological analysis. Immunocytochemistry Details ROIs were removed, cryoprotected, and sectioned (30m thick) on a freezing-sliding microtome. The free-floating sections (30m thick) were microwaved in 10mM citric acid buffer (pH 6.0) for 5 minutes, incubated in 3% hydrogen peroxide and 10% Triton X-100 (Sigma, St. Louis, MO) in phosphate-buffered saline for 30 minutes, and immunostained by the avidin-biotin complex procedure with diaminobenzidine tetrahydrochloride as a chromagen, as described previously.36 Sections stained for serum protein were not microwaved. Sections stained for IgG were processed using the avidin-biotin complex procedure with diaminobenzidine tet- rahydrochloride, whereas horseradish peroxidase–conjugated fibrinogen and albumin were visualized with diaminobenzidine tetrahydrochloride. Sections were also double labeled for axons and myelin utilizing immunofluorescent procedures, as described previously.36 A list of antibodies is provided in Supplementary Table 2. MRI-defined regions were histologically classified based on the presence or absence of myelin staining proteolipid protein (PLP), activity of the lesion, and the serum protein detection. The activity of lesions were subdivided into active, chronic active, and chronic inactive. Active lesions were defined as hypercellular throughout the lesion area, chronic active lesions had a hypercellular border of major histocompatibility complex (MHC) class II–positive cells and a hypocellular center, and chronic inactive lesions were defined as hypocellular throughout.37 If myelin was present in the ROI, the microglia were examined for activation markers and changes in shape. Microglia exhibiting both MHC class II expression and retracting and thickening of processes were noted as “activated.” The presence of serum proteins in each region was categorized as either diffuse or cellular staining, specific to axons and glial cells. Sections that were positive in at least one serum protein stain were considered positive. These sections were compared with non-MS control tissue (to distinguish lesions and myelinated white matter, because many of the tissue samples contained gliosis). Sections stained for SMI 32 and 31 were examined for axonal measurements. Axonal area was estimated using a computeraided technique, similar to that described previously.38 Multiple confocal micrographs for each region were digitized (40⫻ magnification), transferred to a workstation, automatically thresholded, and quantified. Axonal area was calculated in red-channel images as the number of thresholded pixels multiplied by 100% and divided by the total number of pixels in the fixed-size image (250 ⫻ 250m). Results from six nonoverlapping areas were averaged to give the final axonal area for each region. The measurement error, as determined by the mean standard deviation of repeated analyses on the same area, was 1.7%. Axonal count was determined automatically in the same thresholded images as the number of objects greater than four pixels in size. For each region, an axonal diameter index was calculated as the mean axonal area divided by the axonal count. Data Analysis Each region was categorized based on MRI characteristics (NAWM, T2-only, or T2T1MTR) and on histopathological features. For each MRI-based group, the number of regions with each specific histopathological feature was determined and compared between MRI groups to identify any major differences. The 2 test was used to test for differences between MRI groups for lesion activity (active vs chronic active vs chronic inactive). Fisher’s exact test was used to test for differences between MRI groups for myelin (myelinated vs demyelinated) and serum proteins (diffuse vs cellular). Kruskal–Wallis one-way analysis of variance with Dunn’s test for multiple comparisons was applied to test for differences in axonal measurements among the three MRI groups. Contrast ratios and axonal measurements were compared between MRI and histopathology groups using the nonparametric Fisher et al: MRI of Axonal Swelling in MS 221 Mann–Whitney U test. Spearman’s rank correlation coefficients were calculated to determine the relation between T1 and MTR contrast ratios and axonal measurements in all MRI lesions and in the T2-only and T2T1MTR lesion groups separately. Results Tissue Acquisition This report includes data from 10 confirmed MS cases that were imaged and processed through the brain donation protocol. Demographic and disease-related details are provided in Supplementary Table 3. All 10 patients (6 female and 4 male patients) had secondaryprogressive MS. The mean (standard deviation) age was 56 (9.7) years, disease duration was 31 (11) years, Extended Disability Status Scale score at the time of death was 8.6 (1.0), BPF was 0.77 (0.06), and brain weight was 1,151.1 (145) gm. The mean time between death and brain removal was 5.4 hours. Imaging was performed immediately before autopsy. Comparison of the lesion maps and photographs confirmed that good coregistration was achieved between the MRIs and tissue slices for each case. The Supplementary Figure shows an example lesion map/ tissue slice pair and the MR image with the lesion types superimposed. A total of 110 MRI-defined regions, which were mapped onto coregistered tissue slice images, were sectioned and stained for histological analysis: 29 NAWM regions, 40 T2-only regions, and 41 T2T1MTR regions. The MRI region types were distributed across all 10 brains. All regions identified radiologically were evaluated histologically. The mean contrast ratios and histological characteristics for each MRI group are summarized in the Table. Comparison of image contrast ratios between the MRI groups confirmed that each group was significantly different from the other two. Examples of each MRI region type and corresponding histology are shown in Figures 1 (NAWM), 2 (T2-only), and 3 (T2T1MTR). Pathological Features of Different Magnetic Resonance Imaging Region Types All NAWM regions were myelinated on histological inspection. There was no evidence of serum proteins in the NAWM regions, but 90% of NAWM regions contained activated microglia. Axonal area was highly variable in NAWM, ranging from 7 to 29%. The mean (standard deviation) percentage of axonal area was 17.3% (4.2%). Axonal count in NAWM ranged from 3,751 to 8,056, with a mean of 6,119 (936). The T2-only group was heterogenous, with only 55% of these regions showing demyelination. Activated microglia were associated with all of the myelinated Table. Magnetic Resonance Imaging and Histological Characteristics for Each Magnetic Resonance Imaging Region Type Characteristics NAWM (n ⴝ 29) T2-only (n ⴝ 40) T2T1MTR (n ⴝ 41) Mean T2 contrast ratio (SD) 1.0 (0.11) 1.3 (0.21) 1.4 (0.27) Mean T1 contrast ratio (SD) 1.0 (0.04) 0.96 (0.06) 0.80 (0.10) Mean MTR contrast ratio (SD) 1.0 (0.03) 0.91 (0.06) 0.67 (0.13) Myelinated Activated microglia Demyelinated 100% (29/29) 45% (18/40) 90% (26/29) 100% (18/18) 0 17% (7/41) 100% (7/7) 55% (22/40) 83% (34/41) Active 27% (6/22) 6% (2/34) Chronic active 73% (16/22) Chronic inactive 0 Astrocytosis 48% (14/29) Serum proteins 0 26% (9/34) 68% (23/34) 93% (37/40) 98% (40/41) 100% (40/40) 100% (41/41) Diffuse 72% (29/40) 37% (15/41) Cellular 28% (11/40) 63% (26/41) Axons Mean % area (SD) 17.3% (4.2) 12.0% (4.3) 11.9% (4.4) Mean count (SD) 6,119 (936) 3,930 (902) 2,711 (815) NAWM ⫽ normal-appearing white matter; T2T1MTR ⫽ T2-weighted, T1-weighted, and magnetization transfer ratio; SD ⫽ standard deviation; MTR ⫽ magnetization transfer ratio. 222 Annals of Neurology Vol 62 No 3 September 2007 Fig 1. Magnetic resonance images (MRIs) and histology for a typical normal-appearing white matter (NAWM) region. (A) T2weighted fluid-attenuated inversion recovery MRI; arrow marks the region of interest used for the histology sections in E, F, and G. (B) T1-weighted MRI. (C) Magnetization transfer ratio image. (D) Corresponding tissue slice with region of interest outlined in yellow. (E) Myelin-stained section (PLP) showing no loss of myelin. (F) Major histocompatibility complex class II staining for lesion activity showing activated microglia. (G) Serum protein staining (IgG) showing no evidence of serum protein staining. Scale bars ⫽ 1mm (E–G). T2-only regions. Nearly all of the T2-only regions, both myelinated and demyelinated, had astrocytosis (93%), and all had either diffuse (72%) or cellular (28%) serum proteins. When the demyelinated T2only regions were subcategorized based on lesion activity, 27% were active and 73% were chronic active, but none was chronic inactive lesions. The mean (standard deviation) percentage of axonal area for T2-only regions was 12.0% (4.3%), which was significantly less than for NAWM regions ( p ⬍ 0.001). Mean axonal count was 3,930 (902), which was also significantly less than in NAWM ( p ⬍ 0.001). Mean axonal diameter index was not significantly different from NAWM (3.2 compared with 2.8, p ⫽ 0.52) In comparison with the T2-only regions, a significantly larger percentage (83%) of the T2T1MTR regions were demyelinated (Fisher’s exact test, p ⫽ 0.008). All of the myelinated T2T1MTR regions had Fig 2. Magnetic resonance images (MRIs) and histology for a typical T2-only region. (A) T2-weighted fluid-attenuated inversion recovery MRI; arrow marks the region of interest used for the histology sections below. (B) T1-weighted MRI; (C) Magnetization transfer ratio image. (D) Corresponding tissue slice with region of interest outlined in yellow. (E) Myelin-stained section (PLP) showing loss of myelin. (F) Major histocompatibility complex class II staining for lesion activity showing staining at lesion edge (ie, a chronic active lesion). (G) Serum protein staining (IgG) showing diffuse staining throughout the lesion. Scale bars ⫽ 1mm (E–G). V ⫽ ventricle. Fisher et al: MRI of Axonal Swelling in MS 223 Fig 3. Magnetic resonance images (MRIs) and histology for a typical T2-weighted, T1-weighted, and magnetization transfer ratio abnormal (T2T1MTR) region. (A) T2-weighted fluid-attenuated inversion recovery MRI; arrow marks the region of interest used for the histology sections below. (B) T1-weighted MRI. (C) MTR image. (D) Corresponding tissue slice with region of interest outlined in yellow. (E) Myelin-stained section (PLP) showing loss of myelin. (F) Major histocompatibility complex class II staining for lesion activity showing hypointense staining throughout the lesion (ie, a chronic inactive lesion). (G) Serum protein staining (IgG) showing cellular staining throughout the lesion. Scale bars ⫽ 1mm (E–G). V ⫽ ventricle. activated microglia. Demyelinated T2T1MTR regions were repopulated by activated microglia, which had shorter and fewer processes than those in NAWM regions. T2-only demyelinated regions contained phagocytic macrophages if acute or activated microglia and occasional macrophages if chronic active. As in the T2only regions, nearly all of the T2T1MTR regions, both myelinated and demyelinated, had astrocytosis (98%), and all had either diffuse (37%) or cellular (63%) serum proteins. There was a significantly greater proportion of serum proteins found in the cellular compartment in the T2T1MTR regions compared with the T2-only regions ( p ⫽ 0.002). When the demyelinated T2T1MTR regions were subcategorized based on lesion activity, only 6% were active, 26% were chronic active, and 68% were chronic inactive lesions. This distribution of lesion activity was significantly different from that of the T2-only regions (2 test ⫽ 24.8; p ⬍ 0.0001). The mean percentage of axonal area for the T2T1MTR regions was 11.9% (4.4%), which was less than the NAWM regions ( p ⬍ 0.001), but not significantly different from the T2-only regions. However, mean axonal count was significantly less in the T2T1MTR regions compared with both the NAWM and T2-only groups ( p ⬍ 0.001), with a mean of 2,711 (815). Correspondingly, the mean axonal diameter index was significantly greater in T2T1MTR regions compared with the other two groups ( p ⬍ 0.001). Figure 4 is a graph of the axonal area, count, and diameter measurements relative to NAWM. Images depicting axonal morphology in the three MRI groups are provided in Figure 5. 224 Annals of Neurology Vol 62 No 3 September 2007 Correlations between MRI contrast ratios (FLAIR, T1, or MTR) and percentage axonal area ranged from weak for T1 and MTR (Spearman’s rank correlation coefficient [SRCC] ⫽ 0.25 and 0.27; p ⬍ 0.01) to moderate for T2 (SRCC ⫽ ⫺0.39; p ⬍ 0.0001). In contrast, correlations between MRI contrast ratios and axonal count were strong for all three parameters, particularly for MTR (SRCC ⫽ ⫺0.52, 0.71, and 0.79 for FLAIR, T1, and MTR, respectively; p ⬍ 0.0001 for all). Correlations between MRI contrast ratios and Fig 4. Plot of percentage axonal area, axonal count, and swelling index in each magnetic resonance imaging group (gray bars denote T2-weighted imaging only; black bars denote T2-weighted, T1-weighted, and magnetization transfer ratio abnormal [T2T1MTR]) relative to the means for normal-appearing white matter (NAWM; hatched bars) regions. sd ⫽ standard deviation. Fig 5. Confocal images of sections stained with polyclonal myelin basic protein antibody (green) and axonal markers SMI 31 and 32 (red). (A) Normal-appearing white matter (NAWM) region showing normal myelination. (B) Demyelinated T2-only region. (C) Demyelinated T2-weighted, T1-weighted, and magnetization transfer ratio abnormal (T2T1MTR) region. (D) Red-channel image of NAWM region in (A) with percentage axonal area ⫽ 18.0% and axonal count ⫽ 6,420, or 105% of the mean for all NAWM regions. (E) Red-channel image of T2-only region in (B) with percentage axonal area ⫽ 13.8% and axonal count ⫽ 3,665, or 60% of the mean for NAWM. (F) Red-channel image of T2T1MTR region with percentage axonal area ⫽ 13.0% and axonal count ⫽ 1,748, or 29% of mean for NAWM. Scale bar ⫽ 20m (A–F). axonal measurements were also determined in the demyelinated regions separately (n ⫽ 56) to identify the potential contribution of axonal swelling or loss to MRI characteristics independent of myelin. In the demyelinated subset, which included 22 T2-only and 34 T2T1MTR regions, there were no correlations at all between MRI contrast ratios and percentage of axonal area. There was also no correlation between FLAIR contrast ratio and axonal count. However, the correlations between T1 and MTR contrast ratios and axonal count were still significant (T2: SRCC ⫽ 0.44, p ⫽ 0.0009; MTR: SRCC ⫽ 0.41, p ⫽ 0.002). MRI contrast ratios and axonal measurements were not correlated to postmortem time. When the regions were subdivided based on the histopathological features, FLAIR, T1, and MTR contrast ratios were all significantly different in the myelinated regions as compared with the demyelinated regions (1.1 vs 1.3 for median FLAIR contrast ratio; 0.97 vs 0.86 for median T1 contrast ratio; and 0.97 vs 0.77 for median MTR contrast ratio; p ⬍ 0.001 for all comparisons). The demyelinated regions were then subdivided by lesion activity. Median FLAIR contrast ratios for active, chronic active, and chronic inactive lesions were 1.26, 1.27, and 1.33, respectively; median T1 contrast ratios were 0.96, 0.93, and 0.76, respectively; and median MTR contrast ratios in these three groups were 0.90, 0.86, and 0.59, respectively. The chronic inactive regions had significantly lower T1 contrast ratio and MTR contrast ratios compared with the active and chronic active groups, but there were no other significant differences between activity-based groups. Other Observations This study was primarily focused on pathological changes that were most likely to contribute to the variable MRI characteristics in the three types of regions. However, we also examined several other cellular components that showed little correlation in distinguishing between T2-only and T2T1MTR lesions. For example, general T-cell density as identified by CD3 staining made up less than 0.5% of the total parenchymal area in both T2-only and T2T1MTR regions. Perivascular cuffs represented less than 1% of T2-only or T2T1MTR lesion areas. Other cellular components such as phagocytic and foamy macrophages were abundant only in active lesions, which only made up 7.3% of the total regions in this study, and thus could not contribute to either T1 or MTR differences observed overall. Activated microglia were present in varying degrees, but were most abundant in T2-only regions and least abundant in T2T1MTR regions. Discussion To investigate the validity of MRI-based lesion differentiation, we imaged postmortem MS brains in situ and applied a new image-to-tissue coregistration method for localization of tissue samples for histopathological analyses based on MRI characteristics. Although no single histological features clearly distinguished T2-only regions from T2T1MTR regions, several important differences were observed. Compared with the T2-only regions, the T2T1MTR regions were more likely to be demyelinated (83% of the T2T1MTR group compared with 55% of the T2-only Fisher et al: MRI of Axonal Swelling in MS 225 group). Demyelinated regions from both MRI groups had significantly greater FLAIR contrast ratio and lower T1 and MTR contrast ratios than myelinated regions. Demyelinated T2T1MTR regions were much more likely to be chronic inactive lesions, whereas two thirds of the regions in this group were chronic inactive; none of the T2-only lesions was classified as such. T2T1MTR regions were also more likely to have serum proteins located intracellularly rather than diffusely, as in the majority of T2-only regions. Acute or recent breakdown of the BBB, therefore, is a consistent feature of T2-only regions. Demyelination does not appear to be a consistent consequence of BBB breakdown because only 55% of the T2-only regions were demyelinated. Surprisingly, there was no difference in the area occupied by axons between the two groups, even though there were significantly fewer axons in T2T1MTR regions. In addition to demyelination and axonal loss, axonal swelling is a significant determinant of T1 hypointensity and MTR abnormality. An important observation of this report is the identification of swollen axons as a marked feature in the T2T1MTR regions (see Fig 5F). Based on visual inspection of axonal diameters in T2T1MTR lesions and adjacent NAWM, the enlarged axonal diameters did not result from selective loss of smaller diameter axons. Rather, axons were abnormally swollen and much larger than axons in NAWM. In a previous postmortem study of MS spinal cords, swollen axons were found within demyelinated lesions in patients with long-standing MS.39 These lesions had high signal intensity on T2-weighted MRIs, but T1-weighted and MTR images were not acquired in that study. An important consequence of axonal swelling is that the space occupied by axons is maintained despite the destruction of substantial numbers of axons, potentially confounding measurements of brain atrophy. Thus, axonal swelling could partially explain why correlations between T2T1MTR lesion volumes and BPF are weaker than expected. From this postmortem study, it is not possible to determine whether all the axonal swelling occurred in vivo. If postmortem changes contribute to axonal swelling, these changes operate in demyelinated axons and in more chronic lesions identified histologically and by T2T1MTR changes. Axonal diameters were not increased in demyelinated T2-only lesions. Furthermore, because there were no correlations between postmortem time and any of the MRI contrast ratios or axonal measurements, it is likely that the axonal swelling occurred in vivo and is indicative of a slowly progressing necrotic death that eventually affects most chronically demyelinated axons. Current concepts,40 supported by experimental animal models and molecular and morphological analysis of chronic MS brains, implicate malfunction of ion channels and pumps that 226 Annals of Neurology Vol 62 No 3 September 2007 maintain axoplasmic Na/K gradients during nerve conduction.41,42 Specifically, an imbalance of Na/K exchange will eventually lead to swelling and death of chronically demyelinated axons. Increased axoplasmic Ca and reduced adenosine triphosphate production appear to be essential contributors to the process of axonal swelling and axonal death.42 When all of the histological differences between MRI groups are considered, it is clear that there is a greater likelihood for a T2T1MTR region to have more severe tissue damage than a T2-only region; however, the T2T1MTR regions exhibited a high degree of pathological heterogeneity. The observation that there is considerable overlap in the histopathological features underlying different MRI lesion types confirms previous reports.9 –11,14,20,23,31,43 It is also consistent with the weak correlations reported in most MRI-clinical correlation studies. In an analysis of MRI characteristics of MS biopsy specimens, Brück and colleagues9 found that MRI features were not specific for distinct pathology, but that T1 hypointensity was related to several different factors, including extracellular edema, axonal loss, and degree of demyelination. A more recent report noted that T1 hypointensity at the time of biopsy did not correlate with axonal loss or demyelinating activity in a group of relapsing-remitting and primary progressive patients.24 However, these findings may not be representative of MS lesions in general because of the atypical MRI appearance of biopsied MS lesions. Postmortem MRI-histopathological studies have demonstrated strong correlations between T1 contrast ratio and residual axonal density,10,11 MTR and axonal density,14 and a significant association between myelin status and MRI characteristics on T1 and MTR images.31 Schmierer and colleagues20 report a strong correlation between MTR and quantitative measurements of myelination and a moderate correlation between T1 relaxation time and axonal count, but no significant difference in axonal counts between lesions that were hypointense versus those that were isointense on T1-weighted MRI. As in van Waesberghe and colleagues’ study,11 we found that both MTR and T1 contrast ratios were highly correlated to axonal loss, as was FLR contrast ratio, to a lesser extent. There was also a significant correlation between MTR and axonal area, but only in the T2T1MTR regions and not in the set of all lesions combined. This finding suggests that variance in MTR in the older, more chronic lesions may be determined mainly by axonal water, but that variance in MTR in the less severe lesions may be dominated by myelin content or extracellular water. Evidence of more extracellular water in the T2-only lesions as compared with the T2T1MTR lesions (as determined by the presence of diffuse serum proteins) supports this explanation. Another interesting observation in this study is that 45% of T2-only regions and 17% of T2T1MTR regions were histologically classified as myelinated. Myelin sheath thickness was appropriate for axonal diameters in these regions. Therefore, they were classified as normally myelinated. None of the regions contained shadow plaques, although remyelination was often found at the edge of some chronic active and chronic inactive lesions. The finding that such a high proportion of MRI-detected lesions are actually myelinated appears counterintuitive, particularly about the T2T1MTR group, but not surprising because it is consistent with previous studies.11,14,20 De Groot and coworkers14 sampled postmortem brains based on MRI abnormalities and found that 48% of T2 lesions had no apparent loss of myelin. They also found that 30% of the lesions that were classified as either mildly or severely hypointense on T1 MRI had no apparent loss of myelin, similar to this study. In Schmierer and colleagues’ study,20 44 of 98 ROIs selected by T2 hyperintensity were discarded, mainly because of technical problems, but six regions were not studied because they were found to be normally myelinated. We analyzed such regions to determine the potential source of MRI signal change. Unlike NAWM, all of the myelinated T2-only regions had serum proteins and many had significantly reduced axonal area and axonal count, indicating MS pathology in the absence of colocalized demyelination. Differences in the findings reported here and in previous MRI-pathology correlation studies are likely due to differences in techniques. In contrast with most other studies, we performed postmortem imaging in situ before autopsy and used different MRI lesion detection, image-to-tissue coregistration, and histology methods than those described for other studies. Schmierer and colleagues20 applied a surgical stereotaxic frame system after brain removal on individual tissue slices to accurately mark ROIs based on MRI coordinates for subsequent histological sampling.44 van Waesberghe and colleagues11 also imaged single brain slices immediately after autopsy. To report axon density, they used a visual ranking system from 0 to 100% based on Bodian staining, whereas we used a computer-aided analysis method to determine the total percentage area occupied by axons based on SMI 32 and 31 staining. We did not use Bodian staining in this study because it binds to all filaments, including those in astrocytes, which creates a problem in the automated thresholding step used for quantification. Conclusion Currently, T1 hypointensity and reduced MTR are considered the most practical way to distinguish MS lesions with more severe tissue destruction in vivo. This postmortem study of patients with long-standing MS confirms that although not all T1 hypointensities with low MTR represent areas of severe tissue destruction, there is a greater likelihood that they do correspond to demyelinated, chronic inactive lesions with significantly fewer and more swollen axons, consistent with irreversible damage. Conversely, MRI-defined regions that are abnormal only on T2-weighted images are more likely to correspond to regions with less severe tissue damage, breakdown of the BBB, and relatively less axonal loss. Axonal swelling, a major distinguishing feature between T2-only and T2T1MTR regions, is an indication of the axonal changes that result from chronic demyelination and precede axonal degeneration. Refinement of widely available MRI acquisition sequences, like the conventional T1-weighted and MTR images used for this study, may provide a means of distinguishing demyelination, axonal loss, and axonal pathology in MS brains. This study was supported by the NIH (National Institute of Neurological Disorders and Stroke, PO1 NS38667, E.F., A.C., R.J.F., J.A.T., R.A.R., B.D.T.) and the National Center for Research Resources (Public Health Service, M01-RR018390, E.F., R.A.R., B.D.T.). We thank R. Klinkosz for technical assistance. References 1. McDonald WI, Compston A, Edan G, et al. Recommended diagnostic criteria for multiple sclerosis: guidelines from the International Panel on the diagnosis of multiple sclerosis. Ann Neurol 2001;50:121–127. 2. Filippi M, Dousset V, McFarland HF, et al. Role of magnetic resonance imaging in the diagnosis and monitoring of multiple sclerosis: consensus report of the White Matter Study Group. J Mag Reson Imaging 2002;15:499 –504. 3. McFarland HF, Barkhof F, Antel J, Miller DH. The role of MRI as a surrogate outcome measure in multiple sclerosis. Mult Scler 2002;8:40 –51. 4. Miller DH. Magnetic resonance in monitoring the treatment of multiple sclerosis. Ann Neurol 1994;36(suppl):S91–S94. 5. Filippi M, Paty DW, Kappos L, et al. Correlations between changes in disability and T2-weighted brain MRI activity in multiple sclerosis: a follow-up study. Neurology 1995;45: 255–260. 6. Mammi S, Filippi M, Martinelli V, et al. Correlation between brain MRI lesion volume and disability in patients with multiple sclerosis. Acta Neurol Scand 1996;94:93–96. 7. Weiner HL, Guttmann CR, Khoury SJ, et al. Serial magnetic resonance imaging in multiple sclerosis: correlation with attacks, disability, and disease stage. J Neuroimmunol 2000;104: 164 –173. 8. Rovaris M, Comi G, Ladkani D, et al. European/Canadian Glatiramer Acetate Study Group. Short-term correlations between clinical and MR imaging findings in relapsing-remitting multiple sclerosis. AJNR Am J Neuroradiol 2003;24:75– 81. 9. Brück W, Bitsch A, Kolenda H, et al. Inflammatory central nervous system demyelination: correlation of magnetic resonance imaging findings with lesion pathology. Ann Neurol 1997;42:783–793. 10. van Walderveen MA, Kamphorst W, Scheltens P, et al. Histopathologic correlate of hypointense lesions on T1-weighted spin-echo MRI in multiple sclerosis. Neurology 1998;50: 1282–1288. Fisher et al: MRI of Axonal Swelling in MS 227 11. van Waesberghe JH, Kamphorst W, De Groot CJ, et al. Axonal loss in multiple sclerosis lesions: magnetic resonance imaging insights into substrates of disability. Ann Neurol 1999;46: 747–754. 12. Truyen L, van Waesberghe JH, van Walderveen MA, et al. Accumulation of hypointense lesions (“black holes”) on T1 spinecho MRI correlates with disease progression in multiple sclerosis. Neurology 1996;47:1469 –1476. 13. Simon JH, Lull J, Jacobs LD, et al. A longitudinal study of T1 hypointense lesions in relapsing MS: MSCRG trial of interferon beta-1a. Neurology 2000;55:185–192. 14. De Groot CJ, Bergers E, Kamphorst W, et al. Post-mortem MRI-guided sampling of multiple sclerosis brain lesions: increased yield of active demyelinating and (p)reactive lesions. Brain 2001;124(pt 8):1635–1645. 15. van Walderveen MA, Barkhof F, Pouwels PJ, et al. Neuronal damage in T1-hypointense multiple sclerosis lesions demonstrated in vivo using proton magnetic resonance spectroscopy. Ann Neurol 1999;46:79 – 87. 16. Li BS, Wang H, Gonen O. Brain metabolite profiles of T1hypointense lesions in relapsing-remitting multiple sclerosis. AJNR Am J Neuroradiol 2003;24:68 –74. 17. Gass A, Barker GJ, Kidd D, et al. Correlation of magnetization transfer ratio with clinical disability in multiple sclerosis. Ann Neurol 1994;36:62– 67. 18. Grimaud J, Barker GJ, Wang L, et al. Correlation of magnetic resonance imaging parameters with clinical disability in multiple sclerosis: a preliminary study. J Neurol 1999;246:961–967. 19. Pike GB, de Stefano N, Narayanan S, et al. Combined magnetization transfer and proton spectroscopic imaging in the assessment of pathologic brain lesions in multiple sclerosis. AJNR Am J Neuroradiol 1999;20:829 – 837. 20. Schmierer K, Scaravilli F, Altmann DR, et al. Magnetization transfer ratio and myelin in postmortem multiple sclerosis brain. Ann Neurol 2004;56:407– 415. 21. Stewart WA, Hall LD, Berry K, Paty DW. Correlation between NMR scan and brain slice data in multiple sclerosis. Lancet 1984;2:412. 22. Katz D, Taubenberger J, Raine C, et al. Gadolinium-enhancing lesions on magnetic resonance imaging: neuropathological findings. Ann Neurol 1990;28:243 (Abstract). 23. Nesbit GM, Forbes GS, Scheithauer BW, et al. Multiple sclerosis: histopathologic and MR and/or CT correlation in 37 cases at biopsy and three cases at autopsy. Radiology 1991;180: 467– 474. 24. Bitsch A, Kuhlmann T, Stadelmann C, et al. A longitudinal MRI study of histopathologically defined hypointense multiple sclerosis lesions. Ann Neurol 2001;49:793–796. 25. Wayne Moore GR. MRI-clinical correlations: more than inflammation alone—what can MRI contribute to improve the understanding of pathological processes in MS? J Neurol Sci 2003;206:175–179. 26. Kalkers NF, Bergers E, Castelijns JA, et al. Optimizing the association between disability and biological markers in MS. Neurology 2001;57:1253–1258. 228 Annals of Neurology Vol 62 No 3 September 2007 27. Fisher E, Rudick RA, Simon JH, et al. Eight-year follow-up study of brain atrophy in patients with MS. Neurology 2002; 59:1412–1420. 28. Rudick RA, Fisher E, Lee JC, et al., Multiple Sclerosis Collaborative Research Group. Use of the brain parenchymal fraction to measure whole brain atrophy in relapsing-remitting MS. Neurology 1999;53:1698 –1704. 29. Dousset V, Grossman RI, Ramer KN, et al. Experimental allergic encephalomyelitis and multiple sclerosis: lesion characterization with magnetization transfer imaging. Radiology 1992; 182:483– 491. 30. Richert ND, Ostuni JL, Bash CN, et al. Interferon beta-1b and intravenous methylprednisolone promote lesion recovery in multiple sclerosis. Mult Scler 2001;7:49 –58. 31. Barkhof F, Bruck W, De Groot CJ, et al. Remyelinated lesions in multiple sclerosis: magnetic resonance image appearance. Arch Neurol 2003;60:1073–1081. 32. Rovaris M, Bozzali M, Rodegher M, et al. Brain MRI correlates of magnetization transfer imaging metrics in patients with multiple sclerosis. J Neurol Sci 1999;166:58 – 63. 33. Fisher E, Cothren RM, Tkach JA, et al. Knowledge-based 3D segmentation of MR images for quantitative MS lesion tracking. SPIE Med Imaging 1997;3034:19 –25. 34. Besag J. On the statistical analysis of dirty pictures. J R Stat Soc 1986;48:259 –302. 35. Besl PJ, McKay ND. Method for registration of 3-D shapes. IEEE Trans Pattern Anal Mach Intell 1992;14:239 –255. 36. Trapp BD, Peterson J, Ransohoff RM, et al. Axonal transection in the lesions of multiple sclerosis. N Engl J Med 1998;338: 278 –285. 37. Bo L, Mork S, Kong PA, et al. Detection of MHC class IIantigens on macrophages and microglia, but not on astrocytes and endothelia in active multiple sclerosis lesions. J Neuroimmunol 1994;51:135–146. 38. Bjartmar C, Kidd G, Mork S, et al. Neurological disability correlates with spinal cord axonal loss and reduced N-acetyl aspartate in chronic multiple sclerosis patients. Ann Neurol 2000;48: 893–901. 39. Bergers E, Bot JC, De Groot CJ, et al. Axonal damage in the spinal cord of MS patients occurs largely independent of T2 MRI lesions. Neurology 2002;59:1766 –1771. 40. Stys PK. White matter injury mechanisms. Curr Mol Med 2004;4:113–130. 41. Craner MJ, Newcombe J, Black JA, et al. Molecular changes in neurons in multiple sclerosis: altered axonal expression of Nav1.2 and Nav1.6 sodium channels and Na⫹/Ca2⫹ exchanger. Proc Natl Acad Sci USA 2004;101:8168 – 8173. 42. Dutta R, McDonough J, Yin X, et al. Mitochondrial dysfunction as a cause of axonal degeneration in multiple sclerosis patients. Ann Neurol 2006;59:478 – 489. 43. Barnes D, Munro PM, Youl BD, et al. The longstanding MS lesion. A quantitative MRI and electron microscopic study. Brain 1991;114(pt 3):1271–1280. 44. Schmierer K, Scaravilli F, Barker GJ, et al. Stereotactic coregistration of magnetic resonance imaging and histopathology in post-mortem multiple sclerosis brain. Neuropathol Appl Neurobiol 2003;29:596 – 601.