Early and extensive spinal white matter involvement in neuromyelitis optica

Abstract

Objectives

Studies of longitudinally extensive spinal cord lesions (LESCLs) in neuromyelitis optica (NMO) have focused on gray matter, where the relevant antigen, aquaporin‐4 (AQP4), is abundant. Because spinal white matter pathology in NMO is not well characterized, we aimed to clarify spinal white matter pathology of LESCLs in NMO.

Methods

We analyzed 50 spinal cord lesions from eleven autopsied NMO/NMO spectrum disorder (NMOSD) cases. We also evaluated LESCLs with three or fewer spinal cord attacks by 3‐tesla MRI in 15 AQP4 antibody‐positive NMO/NMOSD patients and in 15 AQP4 antibody‐negative multiple sclerosis (MS) patients.

Results

Pathological analysis revealed seven cases of AQP4 loss and four predominantly demyelinating cases. Forty‐four lesions from AQP4 loss cases involved significantly more frequently posterior columns (PC) and lateral columns (LC) than anterior columns (AC) (59.1%, 63.6%, and 34.1%, respectively). The posterior horn (PH), central portion (CP), and anterior horn (AH) were similarly affected (38.6%, 36.4% and 31.8%, respectively). Isolated perivascular inflammatory lesions with selective loss of astrocyte endfoot proteins, AQP4 and connexin 43, were present only in white matter and were more frequent in PC and LC than in AC (22.7%, 29.5% and 2.3%, P ^corr = 0.020, and P ^corr = 0.004, respectively). MRI indicated LESCLs more frequently affected PC and LC than AC in anti‐AQP4 antibody‐seropositive NMO/NMOSD (86.7%, 60.0% and 20.0%, P ^corr = 0.005, and P ^corr = 0.043, respectively) and AQP4 antibody‐seronegative MS patients (86.7%, 73.3% and 33.3%, P ^corr = 0.063, and P ^corr = 0.043, respectively). PH, CP and AH were involved in 93.3%, 86.7% and 73.3% of seropositive patients, respectively, and in 53.3%, 60.0% and 40.0% of seronegative patients, respectively.

Conclusions

NMO frequently and extensively affects spinal white matter in addition to central gray matter, especially in PC and LC, where isolated perivascular lesions with astrocyte endfoot protein loss may emerge. Spinal white matter involvement may also appear in early NMO, similar to cerebral white matter lesions.

Introduction

Neuromyelitis optica (NMO) is an autoimmune inflammatory demyelinating disease of the central nervous system characterized by recurrent attacks of optic neuritis and transverse myelitis, in which NMO‐IgG targeting aquaporin‐4 (AQP4) on perivascular astrocyte endfeet plays a major role in producing severe inflammatory lesions. Longitudinally extensive spinal cord lesions (LESCLs) extending three or more vertebral segments are regarded as the most specific imaging feature of NMO 3, 44.

LESCLs were reported to preferentially involve the central gray matter, where AQP4 is more abundantly expressed compared with the white matter 23, 26, 31, 33. Consequently, attention has been focused on spinal gray matter pathology whereas white matter pathology in LESCLs remains to be elucidated. Few neuropathological studies have focused on the horizontal extension patterns of spinal NMO lesions, although one case report described inflammatory lesions existing in both the central gray and peripheral white matters of the spinal cord 45. Recent neuropathological studies demonstrated NMO lesions had pathological diversity, suggesting various pathomechanisms were involved 5, 25, 32. It is well known that extensive cerebral white matter lesions occasionally appear in NMO, as shown by magnetic resonance imaging (MRI) 26, 29 and neuropathological studies 1, 2, 5, 15, 21, 39. Therefore, we aimed to clarify the spinal white matter pathology in NMO, especially the fine distribution and extension patterns, by combined neuropathological and neuroimaging studies of LESCLs in a series of NMO/NMO spectrum disorder (NMOSD) cases.

Methods

Patients

For neuropathological studies, archival autopsied brain, optic nerve, and spinal cord tissues from 10 NMO cases, including one anti‐AQP4 antibody‐seropositive case, and an additional case with NMOSD were used. NMO/NMOSD diagnosis was based on Wingerchuk's criteria 42, 43, 44. The clinical findings of the autopsied cases are summarized in Table 1. For the retrospective neuroimaging study, 30 patients (27 females and three males) with LESCLs extending three or more vertebral segments and examined by 3‐tesla MRI during the early course of the spinal cord disease, were enrolled. These patients had suffered three or fewer spinal cord attacks and were admitted to the Neurology ward of Kyushu University Hospital during 2007–2014 and thoroughly examined for spinal cord relapses. All patients were subjected to brain and spinal cord MRI and an anti‐AQP4 antibody assay, as described below 12, 28. NMO/NMOSD was diagnosed based on the 2015 diagnostic criteria for NMOSD 41. Fifteen NMO/NMOSD patients were seropositive for anti‐AQP4 antibodies and fifteen multiple sclerosis (MS) patients who met the revised McDonald criteria 37, but did not fulfill the NMOSD criteria, were anti‐AQP4 antibody‐seronegative 41. Patient disability status was scored according to the Kurtzke Expanded Disability Status Scale (EDSS) 19 at relapse and at the time of MRI scans.

Table 1.

Summary of clinical and pathological findings in autopsied NMO or NMOSD cases.

Autopsy	Age (yrs)	Sex	Disease duration (yrs)	Annualized relapse rate	Clinically estimated sites of lesions*	Pathologically determined sites of lesions	Investigated number of spinal cord segments	Cause of death	Neurological symptom possibly associated with death
NMO‐1	44	F	3.8	1.6	O2, S6	O, S, Bs, Cr, Cl	1	Pneumonia	Quadriplegia
NMO‐2	44	F	1.8	2.8	O2, Bs3, S2	O, S, Bs, Cr	7	Pneumonia	Bulbar palsy
NMO‐3	48	F	0.5	4.0	O2, Bs1, S1	O, S, Bs, Cr	4	Aspiration pneumonia	Quadriplegia
NMO‐4	32	M	6.3	1.1	O1, Bs2, S7	O, S, Bs, Cl, Cr	9	Bronchial pneumonia	Quadriplegia, bulbar palsy
NMO‐5	28	F	4.7	0.6	O3, Bs2, S2	O, S, Bs, Cl, Cr	4	Bronchial pneumonia	Paraplegia, bulbar palsy
NMO‐6	35	F	7.0	1.4	O4, Bs4, S4	O, S, Bs, Cr	3	Respiratory failure	Paraplegia, bulbar palsy
NMO‐7	37	F	10.8	1.5	O9, Bs2, S14	O, S, Bs, Cl, Cr	11	Respiratory failure	Quadriplegia, bulbar palsy
NMO‐8	47	F	8.3	0.1	O2, Bs1, S2	O, S, Bs, Cr	3	Bronchial pneumonia	Bulbar palsy
NMO‐9	54	F	4.0	1.3	O1, S6	O, S	10	Respiratory failure	Bulbar palsy
NMO‐10†	37	F	17.0	1.1	O8, Bs1, S9, Cr3	Bs, S, Cr	2	Sudden death due to suspected respiratory failure	Quadriplegia
NMOSD	88	M	0.4	0.0	S1	O, S	8	Pneumonia	Paraplegia

*Numbers indicate exacerbations at each lesion site (for example, O2 represents two episodes of optic neuritis).

^†An NMO case seropositive for anti‐AQP4 antibody.

Abbreviations: Bs = brainstem; Cl = cerebellum; Cr = cerebrum, F = female; M = male; NMO = neuromyelitis optica; NMOSD = neuromyelitis optica spectrum disorder; O = optic nerve; S = spinal cord; yrs = years.

Anti‐AQP4 antibody assay

Anti‐AQP4 antibodies were measured using a cell‐based assay and flow cytometry using GFP‐AQP4 fusion protein‐transfected HEK‐293 cells, as described previously 12, 28. The examiners were blind to the origin of the specimens and the anti‐AQP4 antibody assay was performed at least twice for each sample. Samples that gave a positive result twice were deemed to be positive.

Immunohistochemistry

All autopsied cases were obtained from the Department of Neuropathology, Kyushu University, with the exception of an anti‐AQP4 antibody‐seropositive NMO case (NMO‐10) from Tenri Hospital. Because the remaining ten cases died prior to the introduction of systematic anti‐AQP4 antibody analysis at our institution, the serostatus of these cases was not assessed. The median age at autopsy was 44.0 (range 28–88) years in NMO/NMOSD cases (nine females and two males). Disease durations ranged from 0.4 to 17.0 years (median 4.7 years). All patients suffered from severe paraplegia, quadriplegia, or bulbar palsy prior to death caused by respiratory diseases. LESCLs were confirmed in all patients by neuroimaging and/or pathological examinations. Autopsy specimens were fixed in 10% buffered formalin and processed into paraffin sections (5 µm thick). The sections were routinely subjected to hematoxylin and eosin, Klüver–Barrera, and Bodian or Bielschowsky silver impregnation staining. The primary antibodies used for immunohistochemistry were rabbit polyclonal anti‐human connexin 43 (Cx43, Abcam, Cambridge, UK; 1:1000), rabbit polyclonal anti‐human AQP4 (Santa Cruz Biotechnology, Santa Cruz, CA, USA; 1:500), rabbit polyclonal anti‐human glial fibrillary acidic protein (GFAP, DakoCytomation, Glostrup, Denmark; 1:1000), mouse monoclonal anti‐human GFAP (DakoCytomation; 1:100), rabbit polyclonal anti‐human myelin‐associated glycoprotein (Sigma Aldrich, St Louis, MO, USA; 1:400), rabbit polyclonal anti‐human myelin oligodendrocyte glycoprotein (Sigma Aldrich; 1:1000), mouse monoclonal anti‐human myelin basic protein (Novocastra, Newcastle upon Tyne, UK; 1:50), mouse monoclonal anti‐human CD68 (DakoCytomation, 1:200), rabbit polyclonal anti‐human C3d complement (DakoCytomation, 1:1000), mouse monoclonal anti‐human C9neo (Abcam; 1:1000), rabbit polyclonal anti‐human IgG (DakoCytomation, 1:10,000) and rabbit polyclonal anti‐human IgM (DakoCytomation; 1:10 000). Antigen retrieval and primary antibody incubation were performed as previously described 25. Antigen retrieval in paraffin sections was performed by autoclaving in citrate buffer (pH 6.0) at 121°C for 15 minutes. All sections were deparaffinized in xylene and rehydrated through an ethanol gradient. Endogenous peroxidase activity was blocked with 0.3% (v/v) H₂O₂/methanol. Sections were then incubated with primary antibodies at 4°C overnight. After rinsing, sections were subjected to either a streptavidin‐biotin complex or an enhanced indirect immunoperoxidase method using Envision (DakoCytomation). Immunoreactivity was detected using 3,3′‐diaminobenzidine and sections were counterstained with hematoxylin. Elastica van Gieson stain was used for selected sections. Using the same set of paraffin sections as described above, double immunofluorescence staining was performed using the following combinations of antibodies: rabbit polyclonal anti‐human Cx43 and mouse monoclonal anti‐human GFAP. Sections were deparaffinized in xylene and rehydrated through an ethanol gradient. Sections were then incubated with primary antibodies overnight at 4°C. After rinsing, sections were incubated with Alexa 488‐conjugated goat anti‐rabbit IgG or Alexa 546‐conjugated goat anti‐mouse IgG (Life Technologies) and then counterstained with 4′,6‐diamidino‐2‐phenylindole. Images were captured using a confocal laser microscope system (Nikon A1, Nikon, Japan). We classified inflammatory lesions as active, chronic active or chronic inactive, based on the density of macrophages phagocytosing myelin debris, as previously described 20, 22. Active lesions were characterized as destructive lesions densely and diffusely infiltrated with macrophages phagocytosing myelin debris, as identified by Luxol fast blue staining and immunohistochemistry for minor myelin proteins. Chronic active lesions were defined as those with a hypercellularity of macrophages restricted to the periphery of lesions. Chronic inactive lesions showed no increase in macrophage numbers throughout the lesion. As we regarded isolated perivascular lesions to be an early stage phenomenon, they were excluded from stage classification based on the density of macrophages phagocytosing myelin debris.

Neuropathological analysis of spinal cord lesion distribution

We characterized the inflammatory lesions according to their immunoreactivity to astrocyte, myelin and macrophage markers, and deposition of activated complement or immunoglobulin, and then counted the spinal cord lesions in the axial sections with careful consideration of longitudinal consecutiveness. The lesion distribution of LESCLs was investigated macroscopically and 62 myelomeres (segments of spinal cord) from all cases, from one to eleven per case, were studied histologically (Table 1). Secondary degenerative lesions were excluded. We evaluated the localization of each lesion according to the following anatomically classified regions: the posterior (PC), lateral (LC), and anterior columns (AC) of the white matter, and the posterior (PH) and anterior horns (AH), and the central portion (CP) of the gray matter. A lesion involving more than one region was counted as occupying each region (eg, if a lesion involved all regions except for AC, it was classified for each of the five regions AH, CP, PH, LC and PC). We classified the levels of AQP4 relative to the intensity of myelin staining into preferential AQP4 loss and relatively preserved AQP4 types. AQP4 loss was defined as a loss or reduction of AQP4 immunoreactivity extending beyond or conforming to myelin loss, and the predominantly demyelinating type was defined as preserved AQP4 immunoreactivity relative to myelin loss 25, 27. Furthermore, AQP4 loss was sub‐classified into two types according to the presence or absence of the perivascular deposition of complement and immunoglobulin (AQP4 loss with C and AQP4 loss without C) 25, 27. Regarding the heterogeneity of NMO lesions, we defined “AQP4 loss cases” as those where most lesions showed AQP4 loss and “predominantly demyelinating cases” as those exhibiting relatively preserved AQP4 lesions without exception.

MRI

All MRI studies were performed using 3‐ or 1.5‐tesla systems (Achieva, Philips, Best, The Netherlands). To evaluate LESCLs, only 3‐tesla T2‐weighted axial and sagittal plane scans were used and the most extensive lesions on the axial plane were analyzed. Brain lesions were analyzed by 3‐ or 1.5‐tesla T2‐weighted imaging. For the 3‐tesla system, the typical imaging parameters for the brain were as follows: sagittal turbo fluid‐attenuated inversion recovery (FLAIR) imaging using repetition time (TR)/echo time (TE)/inversion time (TI) = 10,000/120/2700 ms, field‐of‐view (FOV) = 240 × 228 mm, and slice thickness = 5 mm, axial T2‐weighted turbo spin‐echo imaging using TR/TE = 3000/80 ms, FOV = 230 × 230 mm, and slice thickness = 5 mm, axial T1‐weighted spin‐echo imaging using TR/TE = 450/9 ms, FOV = 230 × 183 mm, and slice thickness = 5 mm, axial turbo FLAIR imaging using TR/TE/TI = 10 000/120/2700 ms, FOV = 230 × 196 mm, and slice thickness = 5 mm, axial diffusion‐weighted imaging using b‐values of 0 and 1000 s/mm², TR/TE = 4000/59 ms, FOV = 230 × 230 mm, and imaging time = 48 s. The contrast material (gadopentetate dimeglumine, (Magnevist) Bayer, Osaka, Japan; gadodiamide (Omniscan), Daiichisankyo, Tokyo, Japan; or gadoteridol (Prohance), Eisai, Tokyo, Japan) was injected at 0.1 mmol/kg and axial post‐contrast T1‐weighted spin‐echo imaging performed using TR/TE = 465/21 ms, FOV = 230 × 183 mm, and slice thickness = 5 mm. The typical imaging parameters for the cervical spine were as follows: sagittal T2‐weighted turbo spin‐echo imaging using TR/TE = 2700/69 ms, FOV = 160 × 250 mm, and slice thickness = 4 mm, sagittal T1‐weighted turbo spin‐echo imaging using TR/TE = 450/5 ms, FOV = 160 × 251 mm, and slice thickness = 4 mm, axial T2‐weighted turbo spin‐echo imaging using TR/TE = 3000/62 ms, FOV = 160 × 160 mm, and slice thickness = 5 mm, axial T1‐weighted turbo spin‐echo imaging using TR/TE = 406/6 ms, FOV = 160 × 160 mm, and slice thickness = 5 mm. The typical imaging parameters for the thoracic spine were as follows: sagittal T2‐weighted turbo spin‐echo imaging using TR/TE = 2500/71 ms, FOV = 200 × 319 mm, and slice thickness = 4 mm, sagittal T1‐weighted turbo spin‐echo imaging using TR/TE = 400/6 ms, FOV = 200 × 319 mm, and slice thickness = 4 mm, axial T2‐weighted turbo spin‐echo imaging using TR/TE = 5411/66 ms, FOV = 160 × 160 mm, and slice thickness = 5 mm, axial T1‐weighted turbo spin‐echo imaging using TR/TE = 511/8 ms, FOV = 160 × 160 mm, and slice thickness = 5 mm. The typical imaging parameters for the lumbar spine were as follows: sagittal T2‐weighted turbo spin‐echo imaging using TR/TE = 2500/70 ms, FOV = 200 × 319 mm, and slice thickness = 4 mm, sagittal T1‐weighted turbo spin‐echo imaging using TR/TE = 400/6 ms, FOV = 200 × 319 mm, and slice thickness = 4 mm, axial T2‐weighted turbo spin‐echo imaging using TR/TE = 3266/70 ms, FOV = 160 × 160 mm, and slice thickness = 5 mm, axial T1‐weighted turbo spin‐echo imaging using TR/TE = 593/7 ms, FOV = 160 × 160 mm, and slice thickness = 5 mm. For the 1.5‐tesla system, the typical imaging parameters for the brain were as follows: sagittal turbo FLAIR imaging using TR/TE/TI = 9000/110/2400 ms, FOV = 240 × 229 mm, and slice thickness = 5 mm, axial T2‐weighted turbo spin‐echo imaging using TR/TE = 4505/100 ms, FOV = 230 × 207 mm, and slice thickness = 5 mm, axial T1‐weighted spin‐echo imaging using TR/TE = 592/12 ms, FOV = 230 × 183 mm, and slice thickness = 5 mm, axial turbo FLAIR imaging using TR/TE/TI = 9000/110/2400 ms, FOV = 230 × 206 mm, and slice thickness = 5 mm, axial diffusion‐weighted imaging using b‐values of 0 and 1000 s/mm², TR/TE = 3700/72 ms, FOV = 230 × 230 mm, imaging time = 44 s. The contrast material for the 1.5‐tesla system was the same as that for the 3‐tesla system, and axial post‐contrast T1‐weighted spin‐echo imaging was performed using TR/TE = 400/20 ms, FOV = 230 × 183 mm, and slice thickness = 5 mm. MRI scans used in the present study were taken at the acute to subacute stages (within 2 months after relapse onset) in fourteen patients and at the convalescent stage (more than 2 months after relapse onset) in sixteen patients. MRI scans at the convalescent stage were performed on significantly more anti‐AQP4 antibody‐seronegative patients than on seropositive patients (80.0% vs. 26.7%, P = 0.009). The frequency of steroid treatment at sampling was significantly higher in anti‐AQP4 antibody‐seropositive patients than in seronegative patients (73.3% vs. 20.0%, P = 0.009). MRI films were retrospectively studied by examiners blinded to the anti‐AQP4 antibody status (S.H. and T.Y.). The localization of each lesion was evaluated using same method as in the neuropathological study.

Statistical analysis

For the neuropathological studies, comparisons of the frequency of any given observation between AQP4 loss cases and predominantly demyelinating cases, and among cases of AQP4 loss with C, AQP4 loss without C and predominantly demyelinating lesions were performed using Fisher's exact test. For neuroimaging studies, comparisons of the frequency of any given observation between anti‐AQP4 antibody‐seropositive and ‐seronegative patients was compared using Fisher's exact test and Mann–Whitney U‐test, as appropriate. In both studies, the frequencies of lesions between anatomical regions were compared using the one sample chi‐square test, excluding overlapping data. When multiple comparisons were performed, uncorrected P values were corrected by multiplying them by the number of comparisons (Bonferroni–Dunn correction) to determine the corrected P values (P ^corr). In all assays, statistical significance was set at P < 0.05.

Ethics statement

The study was approved by The Kyushu University Institutional Review Board for Clinical Research.

Results

Pathological lesion distribution profiles in AQP4 loss cases and predominantly demyelinating cases

Of the 50 spinal cord lesions assessed, 18 (36.0%) were classified as AQP4 loss with C type lesions, 25 (50.0%) as AQP4 loss without C type lesions, and seven (14.0%) as predominantly demyelinating type lesions (a representative lesion of each is shown in Figure 1). Based on the most predominant pathology type, we first classified the NMO cases into AQP4 loss cases (7 cases; NMO‐2, 3, 4, 7, 9, 10, and NMOSD) and predominantly demyelinating cases (4 cases; NMO‐1, 5, 6, 8) (Table 2). All predominantly demyelinating cases had only predominantly demyelinating type lesions while all AQP4 loss cases but one (NMO‐2) showed AQP4 loss in all lesions with or without complement deposition. Because NMO‐2 had three AQP4 loss type lesions and one predominantly demyelinating type lesion, NMO‐2 in the AQP4 loss cases was classified based on the predominant pathology type. Among the AQP4 loss cases, NMO‐4 and NMO‐7 had only AQP4 loss with C type lesions, and NMO‐3 and NMOSD had only AQP4 loss without C type lesions. However, NMO‐9 and NMO‐10 had both AQP4 loss with and without C type lesions, and NMO‐2 had all three types of lesions, reflecting NMO lesion heterogeneity (Table 2).

Figure 1.

Various extension patterns of extensive and isolated perivascular lesions in the spinal cord. In the active (B–D) and isolated perivascular (A, E) lesions from NMO‐9, the most extensive lesion mainly affects the lateral and anterior columns and the adjacent gray matter at T8 (D). Perivascular lesions run from the peripheral white matter centrally and merge together (arrow in D, and Figure 3O). Isolated perivascular lesions are present in the lateral column at C4 (A) and T10 (E) (arrowheads). (B) In the active lesion at T4 of NMO‐9, isolated perivascular lesions (arrowheads) are scattered in the posterior column whereas the main lesions involve the lateral columns and the central gray matter. (C) The active lesion at T7 of NMO‐9 demonstrates near holocord involvement with spared peripheral white matter rim. The area of most pronounced AQP4 loss shows a ring‐shaped band because of the relative preservation of AQP4 in both the central portion and the peripheral white matter rim. An isolated perivascular lesion is also present in the right lateral column (arrowhead). The holocord active lesions at T4 (H) and L3 (F–G) of NMO‐7 show spared peripheral white matter rim, and are similar to, but more extensive than, those of NMO‐10 (I). (F–G) The white matter shows more severe impairment than the gray matter. (I) In NMO‐10, the active lesion at T3 shows extensive lesions in the central gray matter and adjacent white matter with spared peripheral white matter rim. Many linear white matter lesions along the blood vessels appear to fuse into (or radiate from) the partially necrotic lesion center. (J) In NMO‐10, the chronic active lesion at T8 shows isolated perivascular lesions (arrowheads) with complement deposition (Figure 3E) and a linear appearance along the blood vessels, some of which fuse with each other and extend toward the central gray matter lesions. Peripheral white matter rims are mostly spared. (K–N) In NMO‐3, the chronic active lesions at L3 show AQP4 loss without complement deposition type lesions that mainly involve bilateral posterior columns extending toward the posterior horns. Immunoreactivities for AQP4 (K, N) and GFAP (L) are markedly diminished relative to myelin loss (M). Isolated perivascular lesions with (outlined arrowheads) or without (filled arrowhead) disruption of the glia limitans in the lateral columns are present. (N) Higher magnification (arrow in K) shows an isolated perivascular lesion that extends nearly toward the anterior horn. (O) In NMO‐3, the most extensive lesion among the chronic inactive lesions at T3 involves the posterior and anterior columns and the central gray matter bilaterally. An isolated perivascular lesion (arrowhead) and fused perivascular lesions are present in the lateral columns. (P) In an NMOSD case, the active extensive lesion at T6 affects the right posterior column and ipsilateral posterior horn extending to the central portion. An isolated perivascular lesion is also present in the left posterior column (arrowhead). (Q) In an NMOSD case, the chronic active lesions at T12 preferentially involve right posterior and lateral columns and left anterior column, extending to the adjacent gray matter to some extent. (R–U) In NMO‐6, the chronic active lesions at T4 show myelin staining (T–U) that is diminished over a wider area relative to immunoreactivities for AQP4 (R) and GFAP (S). Immunoreactivities for AQP4 and GFAP are diminished at bilateral lateral columns, the central portion, right posterior horn, and ipsilateral posterior column with spared peripheral white matter rim. Scale Bar = 1 mm (A–M, O–U), 200 µm (N). AQP4 = aquaporin‐4; GFAP = glial fibrillary acidic protein; KB = Klüver–Barrera; MOG = myelin oligodendrocyte glycoprotein.

Table 2.

Summary of lesion subtypes found in eleven cases of NMO and NMOSD.

Classification	NMO case number	Lesion subtype seen in each case			Frequency of lesion subtype, n lesions/total (%)
		AQP4 loss with C type	AQP4 loss without C type	Predominantly demyelinating type	AQP4 loss with C type	AQP4 loss without C type	Predominantly demyelinating type
AQP4 loss cases	NMO‐2, 3, 4, 7, 9, 10, NMOSD	NMO‐4, 7	NMO‐3, NMOSD	None	18/44 (40.9)	25/44 (56.8)	1/44 (2.3)
		NMO‐9, 10		None
		NMO‐2
Predominantly demyelinating cases	NMO‐1, 5, 6, 8	None	None	NMO‐1, 5, 6, 8	0/6 (0)	0/6 (0)	6/6 (100)

Abbreviations: AQP4 = aquaporin‐4; NMO = neuromyelitis optica; NMOSD = neuromyelitis optica spectrum disorder.

Regarding the sagittal lesion distribution, cervical, thoracic and lumbosacral lesions were present in 7/11 (63.6%), 10/11 (90.9%) and 4/11 (36.4%) cases, respectively. Regarding the axial lesion distribution, AQP4 loss cases containing lesions with complement deposition type, the presence of an extensive lesion at the T7 level in NMO‐9 (Figure 1C) demonstrated near holocord involvement with spared peripheral white matter rim. The area containing the most pronounced AQP4 loss showed a ring‐shaped band because of the preservation of AQP4 in both the central portion and the peripheral white matter rim. In both rostral (C4) (Figure 1A) and caudal (T10) (Figure 1E) ends of the lesions, isolated linear perivascular lesions were present with the radial vessels in the LC, where spared spaces existed between such isolated lesions and the spinal cord surface. Between the ends and center (at T4 and T8), scattered perivascular lesions in PC and LC were fused with each other and projected toward the extensive lesions involving both gray and white matter (Figure 1B,D). In NMO‐10, isolated perivascular lesions with complement deposition showed a linear appearance along the vessels, fusing with each other at T8 (Figures 1J and 3E) and projecting from the white matter toward the partially necrotic gray matter lesion center at T3, while sparing the subpial white matter areas (Figure 1I). Overall, three cases (NMO‐7, NMO‐9 and NMO‐10) had holocord lesions with spared peripheral white matter rim (Figure 1C,F,H,I).

Figure 3.

Profiles of isolated perivascular lesions and inflammatory infiltrate patterns. (A–L) Higher magnification of isolated perivascular lesions at T8 of NMO‐10 (A–F) and at T10 of NMO‐9 (G–M). (A–F) Isolated perivascular lesions with complement deposition around the vessels (E), perivascular lymphocytic infiltrates (A) and extensive loss of AQP4 (D) and Cx43 (F) with the relative preservation of myelin (B) are present in NMO‐10. Immunoreactivity for GFAP is preserved (C). Perivascular AQP4 expression (G) is markedly diminished whereas GFAP (K) is relatively preserved. Minimal vacuolation of the spinal cord tissue without overt demyelination is present around the affected vessel (H). Immunoreactivity for Cx43 (L) is markedly diminished relative to that for GFAP (K) around the affected vessel. Elastica van Gieson staining shows an absence of internal elastic lamina of the affected vessel (I) and the presence of elastic lamina (arrow) in a small artery (J) from the same section as (I). Higher magnification of an active lesion at T4 (N) and T8 (O) of NMO‐9. (N) Inflammatory infiltrates are present in the meninges with disrupted glia limitans, which topographically connect with extensive lesions (corresponding to the arrow in Figure 1B). (O) Tissue rarefaction around vessels projecting from peripheral white matter centrally (arrowheads). Inflammatory cells aggregate around the central end of the affected vessel but are absent in the meninges. Scale Bar = 200 µm (A–F, O), 100 µm (G, H, K–N), 50 µm (I, J). HE = hematoxylin and eosin; KB = Klüver–Barrera; AQP4 = aquaporin‐4; GFAP = glial fibrillary acidic protein; EVG = Elastica van Gieson.

Among the cases showing only AQP4 loss without C type lesions, the most extensive lesion at T3 in NMO‐3 involved PC and AC and bilateral central gray matter (Figure 1O) whereas at L3 the lesion mainly involved bilateral PC, extending toward PH (Figure 1K–M). In addition, isolated perivascular lesions with (outlined arrowheads) or without (filled arrowhead) disrupted glia limitans and fused perivascular lesions were visible in LC (Figure 1K–M,O), one of which was running next to the AH (Figure 1N). In an NMOSD case, the main lesion at T6 affected the right PC and ipsilateral PH extending toward the CP, while at T12 the right PC and LC and left AC were mainly involved, extending to the adjacent gray matter (Figure 1P–Q).

Among the predominantly demyelinating cases, as shown in a representative NMO‐6 case (Figure 1R–U), myelin loss was prominent relative to the immunoreactivity for AQP4 and GFAP, showing spared peripheral white matter rim.

Figure 2A summarizes the overall axial lesion distribution according to the classified cases, including the isolated perivascular lesions. Overall, the white matter was more frequently involved than the gray matter in total cases, AQP4 loss cases, and predominantly demyelinating cases (100% vs. 42.0%, 100% vs. 38.6%, and 100% vs. 66.7%, respectively). In AQP4 loss cases, the white matter lesions were more frequently located in PC (59.1%) and LC (63.6%) than in AC (34.1%) (P ^corr = 0.035 in PC vs. AC, and P ^corr = 0.009 in LC vs. AC), whereas in the gray matter there was no difference in the lesion distribution among PH, CP and AH (38.6%, 36.4%, and 31.8%, respectively) (Figure 2A). In the predominantly demyelinating cases, similar trends in axial lesion distribution were observed (similar lesion distribution among AH, CP, and PH in the gray matter but higher lesion distribution in PC and LC than AC); however, the difference was not significant in part because of the small sample size used (Figure 2A).

Figure 2.

Axial distribution frequencies of spinal cord lesions in defined cases and lesion subtypes. Axial distribution of spinal cord lesions according to defined cases (A) and lesion subtypes (B), compared between anatomically classified regions. (C) Axial distribution of isolated perivascular lesions according to similarly defined subtypes. “Total” means any gray/white matter involvement. AQP4 = aquaporin‐4; AH = anterior horn; CP = central portion; PH = posterior horn; AC = anterior column; LC = lateral column; PC = posterior column. **Statistical significance between the linked groups (P ^corr < 0.05).

Pathological lesion distribution profiles according to lesion subtypes in autopsied NMO/NMOSD cases

Figure 2B summarizes the overall axial lesion distribution according to lesion subtypes, including isolated perivascular lesions. Overall, the white matter was more frequently involved than the gray matter in AQP4 loss with C type lesions, AQP4 loss without C type lesions, and predominantly demyelinating type lesions (100% vs. 44.4%, 100% vs. 32.0% and 100% vs. 71.4%, respectively). AQP4 loss with C type lesions in the white matter were more frequently located in PC (72.2%) and LC (61.1%) than in AC (38.9%) (not statistically significant, in part because of the small sample size used), whereas in the gray matter lesions they were almost evenly distributed among PH, CP, and AH (44.4%, 44.4% and 33.3%, respectively). In AQP4 loss without C type lesions, LC (64.0%) and PC (48.0%) were more frequently affected than AC (28.0%) (P ^corr = 0.038 in LC vs. AC). Similar trends were also observed in the predominantly demyelinating lesions (PC 85.7%, LC 71.4%, and AC 57.1%). Even when isolated perivascular lesions exclusively present in the white matter were excluded, similar trends were observed (AC, LC, and PC involvement in 77.8%, 77.8%, and 88.9% of those with AQP4 loss with C type lesions, in 60.0%, 70.0% and 70.0% of those with AQP4 loss without C type lesions, and in 57.1%, 71.4% and 85.7% of those with predominantly demyelinating type lesions, respectively). Regarding the lesion staging, where isolated perivascular lesions were excluded, active lesions were more frequently found in AQP4 loss with C type lesions (7/9, 77.8%) than in AQP4 loss without C type lesions (3/10, 30.0%, P ^corr = 0.209) and predominantly demyelinating type lesions (0/7, 0%, P ^corr = 0.009) whereas AQP4 loss without C type lesions (6/10, 60%, P ^corr = 0.172) and predominantly demyelinating type lesions (6/7, 85.7%, P ^corr = 0.026) had more frequent chronic active lesions than AQP4 loss with C type lesions (1/9, 11.1%) (Supporting Information Table S1). There was no significant difference in the lesion staging between AQP4 loss without C type lesions and predominantly demyelinating type lesions. Moreover, the combined frequencies of active and chronic active lesions did not significantly differ among the three types (AQP4 loss with C type lesions 8/9, 88.9%; AQP4 loss without C type lesions 9/10, 90.0%; predominantly demyelinating type lesions 6/7, 85.7%).

Characterization of isolated perivascular lesions

We frequently observed isolated perivascular lesions exclusively in the white matter (representative lesions from NMO‐3, 9, 10, and NMOSD are shown in Figures 1 and 3). These isolated perivascular lesions were present in five of eleven cases (45.5%), of which most were AQP4 loss cases (5/7, 71.4%), but none were predominantly demyelinating cases (0/4, 0%). In isolated perivascular lesions from NMO‐9, immunoreactivities for AQP4 and Cx43 were markedly diminished in highly degenerated GFAP‐positive astrocytes and minimal tissue vacuolation around the affected vessels was detected (Figure 3G–H,K–M). In NMO‐10, immunoreactivities for AQP4 and Cx43 were decreased but those for GFAP and myelin were preserved (Figure 3A–F). These lesions were accompanied by perivascular lymphocyte infiltrates (Figure 3A). In the isolated perivascular lesions demonstrating loss of AQP4 and Cx43 (24/24, 100%), minimal vacuolation of the spinal cord tissue was observed in 10/24 (41.7%) lesions whereas immunoreactivity for GFAP was diminished in 20/24 (83.3%) and preserved in 4/24 (16.7%) lesions; all four of these lesions were from an anti‐AQP4 antibody‐positive case. Elastica van Gieson staining revealed a lack of internal elastic lamina in the affected vessels of the isolated perivascular lesions (Figure 3I), suggesting these vessels were post‐capillary venules. These isolated perivascular lesions represented 48.0% of the total lesions (24/50): 50.0% (9/18) in the AQP4 loss with C type lesions, 60.0% (15/25) in the AQP4 loss without C type lesions, and 0% (0/7) of predominantly demyelinating type lesions. Overall, isolated perivascular lesions were more frequent in PC (41.7%) and LC (54.2%) than in AC (4.2%) (P ^corr = 0.020 in PC vs. AC, and P ^corr = 0.004 in LC vs. AC) (Figure 2C). The same finding was observed for both AQP4 loss with C type lesions (PC 55.6%, LC 44.4% and AC 0%, not statistically significant in part because of the small sample size used in either PC vs. AC or in LC vs. AC) and AQP4 loss without C type lesions (PC 33.3%, LC 60.0%, and AC 6.7%, P ^corr = 0.308 in PC vs. AC, and P ^corr = 0.034 in LC vs. AC).

Meningeal inflammation adjacent to the spinal cord lesions

Lymphocyte infiltrates were not present in the nearby meninges of 24 isolated perivascular lesions, whereas among 26 other lesions, five (19.2%), including one with a disruption of the glia limitans, were accompanied by meningeal lymphocyte infiltrates, all of which connected with extensive lesions (Figure 3N–O).

Demographic features of patients with LESCLs assessed by neuroimaging

Table 3A and Supporting Information Table S2 summarize the clinical and neuroimaging findings of 30 patients with LESCLs according to their anti‐AQP4 antibody status. Clinical features were similar, irrespective of anti‐AQP4 antibody serostatus. Anti‐nuclear antibody was detected more frequently in anti‐AQP4 antibody‐seropositive patients than in seronegative patients (P = 0.046). Cerebrospinal fluid (CSF) oligoclonal IgG bands were significantly more frequent in anti‐AQP4 antibody‐seronegative patients than in seropositive patients (P = 0.001), whereas other CSF parameters did not differ significantly between the two groups. MS‐like brain lesions fulfilling the Barkhof criteria 4 were detected significantly more frequently in anti‐AQP4 antibody‐seronegative patients than in seropositive patients (P = 0.002).

Table 3.

Clinical, laboratory and neuroimaging findings in patients with LESCLs.

	AQP4 Abs (+) (n=15)	AQP4 Abs (−) (n=15)	P value
(A) Clinical and laboratory findings of patients with LESCLs
Female, n (%)	15 (100)	12 (80.0)	N.S.
Age, y, median (range)
at first attack	39 (23–52)	30 (27–39)	N.S.
at sampling	41 (29–52)	37 (32–43)	N.S.
Total no. of attacks, median (range)	3 (3–4)	2 (2–4)	N.S.
EDSS score, median (range)
at peak	3.5 (3–4)	3 (2–3.5)	N.S.
at sampling	3 (2.5–4)	2 (2–3.5)	N.S.
CSF analysis, n/total (%)
Elevated WBC count (>5/µL)	8/13 (61.5)	6/11 (54.5)	N.S.
Elevated Protein (>40 mg/dL)	7/13 (53.8)	3/14 (21.4)	N.S.
IgG index, median (range)	0.53 (0.49–0.61)*	0.68 (0.47–0.88) †	N.S.
Oligoclonal IgG bands	0/12 (0)	7/11 (63.6)	0.001
2010 McDonald criteria ‡ , n (%)	3 (20.0)	15 (100.0)	<0.0001
Immunological abnormalities, n/total (%)
ANA positivity	7/13 (53.9)	2/14 (14.3)	0.046
SS‐A antibody positivity	6/15 (40.0)	1/15 (6.7)	N.S.
Other autoimmune diseases, n (%)	5 (33.3)	1 (6.7)	N.S.
(B) Spinal cord MRI findings of patients with LESCLs.
Spinal MRI lesion distribution on sagittal planes, n (%)
Cervical	10 (66.7)	14 (93.3)	N.S.
Thoracic	8 (53.3)	2 (13.3)	0.050
Lumbosacral	0 (0)	1 (6.7)	N.S.
T2‐lesion length, vertebral segments, median (range)	4.5 (3‐6.5)	3 (3‐4.5)	N.S.
Spinal cord swelling, n (%)	3 (20.0)	3 (20.0)	N.S.
T1 hypointense foci, n (%)	1 (6.7)	2 (13.3)	N.S.
Gadolinium‐enhancing lesion, n (%)	5 (33.3)	3 (20.0)	N.S.
Bright spotty lesion § , n (%)	4 (26.7)	0 (0)	0.099
Holocord lesion with spared peripheral white matter rim, n (%)	1 (6.7)	1 (6.7)	N.S.
Characteristic gray matter lesion, n (%)
H‐shaped lesion¶	2 (13.3)	0 (0)	N.S.
I‐shaped lesion**	4 (26.7)	0 (0)	0.099
Snake‐eye‐like lesion ††	5 (33.3)	0 (0)	0.042
Tram‐track‐like lesion ‡‡	3 (20.0)	1 (6.7)	N.S.
Isolated white matter lesion §§ , n (%)	1 (6.7)	1 (6.7)	N.S.

Data indicate the median (interquartile range) unless otherwise indicated.

*n = 11.

^†n = 10.

^‡Fulfillment of the 2010 McDonald criteria for MS 37.

^§Bright spotty lesions are defined as very hyper‐intense spotty lesions on T2‐weighted axial images that are at least as high as the surrounding CSF 10, 47.

^¶,**H‐shaped or I‐shaped lesions on T2‐weighted axial images are defined as bilateral or unilateral central gray matter lesions, based on the anatomical structure.

^††Snake‐eye‐like lesions on T2‐weighted axial images are defined as bilateral anterior horn lesions, based on the anatomical structure.

^‡‡Tram‐track‐like lesions on T2‐weighted axial images are defined as unilateral linear posterior horn and posterior column lesions running in parallel.

^§§An isolated white matter lesion is defined as a small white matter lesion, which is less than 2 mm at its longest diameter.

Abbreviations: Abs = antibodies; ANA = anti‐nuclear antibody; AQP4 = aquaporin‐4; CSF = cerebrospinal fluid; EDSS = Kurtzke Expanded Disability Status Scale; N.S. = not significant; NMO = neuromyelitis optica; SS‐A = anti‐Sjögren's syndrome‐A.

Comparison of LESCLs between anti‐AQP4 antibody‐positive and ‐negative patients

MRI sagittal imaging of spinal cord showed that thoracic spinal cord lesions tended to be more frequent in anti‐AQP4 antibody‐seropositive patients than in seronegative patients (P = 0.050) (Table 3B). In addition, cervical spinal cord lesions were significantly more frequent in seronegative patients than thoracic cord lesions (P = 0.001). However, there were no significant differences in lesion length or frequencies of spinal cord swelling or contrast‐enhanced lesions (Table 3B). In the axial plane, characteristic gray matter lesions, such as H‐shaped, I‐shaped, snake‐eye‐like and bright spotty lesions 10, 47, were found exclusively in anti‐AQP4 antibody‐seropositive patients (13.3%, P = 0.483, 26.7%, P = 0.099, 33.3%, P = 0.042, and 26.7%, P = 0.099, respectively, for lesion type) (Figure 4). Peculiar tram‐track‐like lesions and holocord lesions with spared peripheral rim were found in a few patients, irrespective of their anti‐AQP4 antibody status (Figure 4).

Figure 4.

Representative MRI findings of anti‐AQP4 antibody‐positive and ‐negative patients. (A) Spinal cord MRIs of a 20‐year‐old female with anti‐AQP4 antibody‐seropositive NMO at the acute stage. A LESCL with spinal cord swelling is present at the lower medulla through thoracic levels on the sagittal plane (Aa). On axial planes, tram‐track‐like lesions, two linear lesions located in parallel in the posterior column and ipsilateral posterior horn (arrows), are present at C1 (Ab and the uppermost arrowhead in Aa). At C3 (Ac and the second arrowhead in Aa), the lesions involve the bilateral posterior horns and large portions of the posterior column, and further extend to occupy most parts of the spinal cord, sparing the peripheral white matter rim at C4 (Ad and the third arrowhead in Aa). Thoracic cord lesion has an H‐shaped appearance at T3 (Ae and the lowermost arrowhead in Aa) and a holocord pattern with spared peripheral white matter rim at T9 (Af). (B) Spinal cord MRI of a 47‐year‐old female with anti‐AQP4 antibody‐seropositive NMOSD at the subacute stage shows a LESCL from medulla to C6 on the sagittal plane (Ba), and tram‐track‐like lesions at C3 (Bb and the arrowhead in Ba). (C) Spinal cord MRIs of a 52‐year‐old female with anti‐AQP4 antibody‐seropositive NMOSD at the subacute stage shows a LESCL from T2 to T8 on the sagittal plane (Ca), snake‐eye‐like lesions at T2 (Cb and the uppermost arrowhead in Ca), a bright spotty lesion at T4 (Cc and the middle arrowhead in Ca), and an I‐shaped lesion at T7 (Cd and the lowermost arrowhead in Ca). (D) Brain and spinal cord MRIs of a 31‐year‐old female with anti‐AQP4 antibody‐seronegative MS at the subacute stage. A LESCL is visible at C3‐6 (Da), and the lesion at C4 shows a near holocord involvement with a tendency of sparing peripheral white matter (Dc and the arrowhead in Da). Many periventricular ovoid lesions are present in the cerebrum (Db). (E) Brain and spinal cord MRIs of a 79‐year‐old female with anti‐AQP4 antibody‐seronegative MS at the convalescent stage. A LESCL involves the C1 to C4 spinal cord (Ea). At C2 (Ec and the arrowhead in Ea), the lesion almost occupies the entire spinal cord. A few periventricular ovoid lesions are present in the cerebrum (Eb). The sagittal and axial spinal cord MRIs are T2‐weighted images (A–C, Da, Dc, Ea, and Ec) and the brain MRIs are fluid‐attenuated inversion recovery images (Db and Eb).

The lesion distribution of LESCLs in the axial plane of all patients was determined. PC and LC were more frequently affected than AC (86.7%, 66.7% and 26.7%, respectively, P ^corr = 0.0003 in PC vs. AC, and P ^corr = 0.002 in LC vs. AC), whereas in the gray matter there was no difference in distribution among PH, CP, and AH (73.3%, 73.3% and 56.7%, respectively) (Figure 5). In the white matter, PC and LC were more frequently affected compared with AC in both seropositive patients (PC 86.7%, LC 60.0% and AC 20.0%, respectively, P ^corr = 0.005 in PC vs. AC, and P ^corr = 0.043 in LC vs. AC) and seronegative patients (PC 86.7%, LC 73.3% and AC 33.3%, respectively, P ^corr = 0.063 in PC vs. AC, and P ^corr = 0.043 in LC vs. AC). There was no significant difference in the frequency of involvement of each white matter region between the seropositive and seronegative patients. In the gray matter, PH, CP and AH were involved in 93.3%, 86.7%, and 73.3% of seropositive patients, respectively, and in 53.3%, 60.0% and 40.0% of seronegative patients, respectively, indicating a more frequent involvement of PH in seropositive patients compared with seronegative patients (P = 0.035).

Figure 5.

Axial distribution of LESCLs by neuroimaging. Axial distribution of LESCLs in all patients (A), and in those with and without anti‐AQP4 antibodies (B, C), compared according to anatomically classified regions. AQP4 = aquaporin‐4; AH = anterior horn; CP = central portion; PH = posterior horn; AC = anterior column; LC = lateral column; PC = posterior column. Statistical significance between the linked groups is indicated as (P < 0.05=*, P ^corr < 0.05=**).

Discussion

The present combined neuropathological and neuroimaging study revealed that NMO spinal cord lesions preferentially affected the white matter in addition to the gray matter, especially the PC and LC in the early course of disease. This study also demonstrated that isolated perivascular lesions with selective loss of astrocyte endfoot proteins, such as AQP4 and Cx43, frequently exist exclusively in the PC and LC of the white matter.

LESCLs in anti‐AQP4 antibody‐seropositive NMO cases were preferentially present in the central gray matter on MRI 26, 33, which is explained by the greater abundance of AQP4 antigen in the gray matter compared with the white matter 31. However, diffusion tensor imaging of the spinal cord demonstrated white matter damage 35, 38. Our 3‐tesla MRI study of the fine distribution of early LESCLs demonstrated that the spinal white matter, especially the PC and LC, can be involved as frequently as the gray matter in the relatively early stages of disease, regardless of anti‐AQP4 antibody status, whereas gray matter involvement was more common in anti‐AQP4 antibody‐seropositive LESCLs than in seronegative LESCLs. The latter observation is in accord with the rich localization of AQP4 in the central gray matter 31, while the early extensive involvement of the spinal white matter is hard to explain solely based on the localization of this antigen.

Pathologically, the spinal white matter, especially in the PC and LC, was more frequently affected than the gray matter in AQP4 loss cases, which included a proven anti‐AQP4 antibody‐positive case. In the leading edge areas of LESCLs, we frequently detected isolated perivascular lesions with inflammatory infiltrates and complement deposition but without obvious demyelination. In these lesions (including an anti‐AQP4 antibody‐positive case), GFAP‐positive astrocytes were still present, whereas astrocyte endfoot proteins, such as AQP4 and Cx43, were lost completely. Intriguingly, all isolated perivascular lesions were localized exclusively in the white matter but not in the gray matter. By contrast, in the predominantly demyelinating cases, no isolated perivascular lesions were observed.

The observations that all isolated perivascular lesions were AQP4 loss type lesions and that all lesions in the predominantly demyelinating cases showed predominantly demyelinating type lesions but not AQP4 loss type lesions suggest that the different lesion type may reflect distinct mechanisms of lesion formation. In this study, isolated perivascular lesions with selective astrocyte endfoot protein loss might be early lesions of the AQP4 loss type, and not the predominantly demyelinating type, reflecting distinct mechanisms. However, the finding that NMO‐2 had a preferentially demyelinating type lesion in addition to the predominant AQP4 loss type lesions and that 3/7 AQP4 loss cases had both AQP4 loss with complement deposition type and AQP4 loss without complement deposition type lesions are compatible with the previously reported heterogeneous NMO pathology 5, 25, 32 and may support the idea that different lesion types could in part be a stage‐dependent phenomenon with the predominantly demyelinating lesion as the end‐stage. When excluding isolated perivascular lesions, active lesions were detected significantly more frequently in AQP4 loss with C type lesions than in predominantly demyelinating type lesions, suggesting that lesion heterogeneity is a stage‐dependent phenomenon. However, there was no significant difference in the lesion staging between AQP4 loss without C type lesions and predominantly demyelinating type lesions, and the frequency of chronic inactive lesions did not differ significantly among AQP4 loss with C type lesions, AQP4 loss without C type lesions, and predominantly demyelinating type lesions. These findings suggest that all lesion heterogeneity cannot be explained solely by a stage‐dependent phenomenon.

Isolated perivascular lesions were observed to run in parallel from PC and LC toward extensive gray and white matter lesions, fusing with each other to become holocord lesions with spared peripheral rim. On MRI, holocord involvement with spared peripheral rim was reported in NMO but was rare in MS 16, 26, 34. Although this study was cross‐sectional, we propose the following lesion extension pattern in NMO based on the pathological findings from the lesion center to the leading edge and combined with the corresponding MRI findings (Figure 6). First, isolated lesions with selective astrocyte endfoot protein loss emerge around blood vessels in the PC and LC. Second, such isolated perivascular lesions fuse with each other, presenting extensive PC and LC lesions, and extend toward the central gray matter. Once the inflammatory lesions reach the gray matter, anti‐AQP4 antibodies extensively damage all gray matter astrocytes, which highly express AQP4, producing I‐ or H‐shaped lesions in the axial plane and LESCLs in the sagittal plane. Third, holocord lesions develop sparing the peripheral white matter rim. Finally, whole spinal cord destruction culminates in a markedly atrophied spinal cord. Intriguingly, our observations are supported by recent experimental animal models of NMO showing perivascular and confluent spinal cord lesions around radial vessels in the white and gray matter junction 18, 36, 49. Such lesions are topographically and morphologically similar to our proposed lesion extension patterns in the early course of disease.

Figure 6.

Lesion extension pattern hypothesis. (A–J) Extension patterns of spinal cord lesions in NMO based on pathological (A–E) and neuroimaging findings (F–J) are shown. (A, F) Isolated white matter lesions emerge in the lateral and posterior columns. (B, G) Extensive lesion involving the posterior column and posterior horn. (C, H) When the white matter lesion reaches the central gray matter, it extends to the entire gray matter, according to the abundant localization of AQP4 and typically shows an H‐shape. (D, I) Holocord involvement showing spared peripheral white matter rim. (E, J) Whole spinal cord destruction culminating in marked atrophy of the spinal cord. (K–O) Hypothetical schematic representation of the lesion extension pattern. In the early disease course, isolated perivascular lesions arise in the posterior and lateral columns (K). Subsequently, these lesions fuse with each other to form extensive white matter lesions, involving the adjacent gray matter (L). Gray matter lesions diffusely extend into the entire gray matter (M), resulting in holocord involvement with spared peripheral rim (N). Ultimately the spinal cord is totally destroyed, leaving necrotic or atrophic changes (O). (P, Q) A hypothetical scheme of the lesion extension patterns based on circulatory disturbance and AQP4 localization. The spinal cord circulatory system is divided into two main systems: central (orange) and peripheral (green) (shown on the left of panel P). The central system consists of branches of the anterior spinal artery (ASA) and vein (ASV) in the anterior median fissure while the peripheral system consists of branches of the posterior spinal arteries (PSA) and posterolateral (PLV) and posteromedial veins (PMV). (Right of panel P) The regions close to the anterior median fissure (orange area) and white matter surface (green area) are presumed to have a rich circulatory reserve. (Left of panel Q) In NMO, perivascular astrocytes at post‐capillary venules in the white matter are first affected by autoantibody‐ and complement‐mediated mechanisms with or without accompanying cellular mechanisms 14, 30, 46, 49. Loss of AQP4 and Cx43 may cause the disruption of water homeostasis and glial energy transfer, thereby potentiating tissue damage. Perivascular lesions fuse with each other and expand to the white matter and adjacent gray matter areas. Gray matter lesions extend according to the AQP4 distribution, and involve the entire gray matter on the axial plane (arrows in left of panel Q) and even extend longitudinally toward the adjacent upper and lower spinal cord levels, producing LESCLs. (Right of panel Q) The peripheral white matter rim may be spared because of its proximity to the circulation system. (A) GFAP, (B–E) AQP4. Scale Bar = 1 mm (A–E). AQP4 = aquaporin‐4; Cx43 = connexin 43; GFAP = glial fibrillary acidic protein; ASA = anterior spinal artery; ASV anterior spinal vein; PSA = posterior spinal artery; PLV = posterolateral vein; PMV = posteromedial vein. The axial spinal cord MRIs are T2‐weighted images (F–J).

Spinal arteries and veins form the central and peripheral circulatory system 50. The former consists of branches of the anterior spinal artery and vein while the latter consists of branches of the posterior spinal arteries and posterolateral and posteromedial veins. Both the central and peripheral circulatory systems operate in the cervico‐thoracic cord. The AC is closer to the central circulatory system than the PC and LC whereas the parenchymal volume of the AC is smaller than that of the PC and LC, which provides the AC with a more favorable circulation compared with the PC and LC. Isolated perivascular lesions may, therefore, emerge in areas of poor circulatory reserve, similar to the preferential development of MS plaques in hypo‐perfused white matter 8, 24, 40. Disruption of perivascular astrocyte endfeet by autoantibody‐mediated mechanisms may disturb water and energy maintenance via the loss of AQP4 water channels and Cx43 gap junction channels that are critical for energy transfer 7, 25, 32, causing extensive parenchymatous damage. In contrast, the subpial regions of the spinal cord have the anatomical advantage of sufficient drainage because of the proximity of the peripheral circulatory system and, therefore, in congestive myelopathy 9, 13, 17 and spinal cord infarction 6, 11, 48 the subpial peripheral rim tends to be spared despite holocord involvement. The same may be true for extensive NMO lesions.

A fraction of extensive spinal cord lesions were accompanied by meningeal lymphocyte infiltrates, whereas no isolated perivascular lesions had meningeal infiltrates. This suggests meningeal inflammation may not be a primary contributor to the formation of early spinal cord lesions in NMO but rather that parenchymal inflammation may secondarily induce meningeal infiltration through glia limitans disruption.

There were some limitations in our study. First, AQP4 antibody serostatus was unknown in all but one case because these were archival cases from before the discovery of anti‐AQP4 antibodies. However, we believe that a comparison of the lesion distribution between AQP4 loss cases and preferentially demyelinating cases is useful, providing insight to the pathomechanisms of these two conditions. It is possible that the former may correspond to an AQP4 antibody‐positive condition while the latter may correspond to an AQP4 antibody‐negative condition. Second, only a few sections were used in some cases, because only limited numbers of sections were available from the archival materials. This may have restricted the extensive investigation of lesion heterogeneity in individual cases. Therefore, in this study we presented the axial distribution patterns by pathological lesion type of each lesion, and by categorized cases classified according to the dominant pathology type. Third, the sample size of the neuroimaging study was small because the inclusion criteria, that is, three or fewer spinal cord attacks, LESCLs and 3‐tesla MRI, were stringent, especially for anti‐AQP4 antibody‐seronegative patients. However, we believe it is valuable to investigate the distribution and extension pattern of LESCLs by prospective high‐field MRI because the neuroimaging findings in this study were compatible with the neuropathological results.

In conclusion, isolated perivascular lesions and extensive white matter lesions are important components of NMO and need to be considered when elucidating the mechanisms of LESCL formation.

Disclosure Statement

S. Hayashida, K. Masaki, T. Yonekawa, SO. Suzuki, A. Hiwatashi, M. Watanabe, T. Suenaga and T. Iwaki report no disclosures. T. Matsushita received a grant and payment from Bayer Schering Pharma and Takeda Pharmaceutical Company for manuscript preparation and for development of educational presentations, and has also received speaker honoraria payments from Mitsubishi Tanabe Pharma, Bayer Schering Pharma and Biogen Idec Japan. R. Yamasaki has received research support from Bayer Schering Pharma, Biogen Idec Japan, Novartis Pharma, and Mitsubishi Tanabe Pharma. H. Murai is a consultant for Novartis Pharma and received speaking fees from Astellas Pharma and the Japan Blood Products Organization. J. Kira serves as an editorial board member of the Multiple Sclerosis Journal, Multiple Sclerosis and Related Disorders, BMC Medicine, PLOS ONE, Expert Review of Neurotherapeutics, Intractable and Rare Diseases Research, Neurology and Clinical Neuroscience, Acta Neuropathologica Communications, Bio Med Research International, the Journal of Neurology and Psychology, the Scientific World Journal and the Journal of the Neurological Sciences. He is a consultant for Biogen Idec Japan and Medical Review. He has received honoraria from Bayer Healthcare, Mitsubishi Tanabe Pharma, Nobelpharma, Otsuka Pharmaceutical and Medical Review. He is funded by a research grant for Nervous and Mental Disorders from the Ministry of Health, Labour and Welfare, Japan, and grants from the Japan Science and Technology Agency and the Ministry of Education, Culture, Sports, Science and Technology, Japan.

Supporting information

Additional Supporting Information may be found in the online version of this article at the publisher's web‐site:

Table S1. Lesion stage according to defined lesion subtypes.

Table S2. Clinical, laboratory, and neuroimaging findings in patients with LESCLs.