Seven-week-old C57BL/6J female mice were purchased from Charles River Japan. Mice were born and raised under specific pathogen-free (SPF) conditions. Mice were housed in an air-conditioned room at 24 ± 2°C with a 12-hour light–dark cycle and had free access to water and food.

Targeted EAE was induced in a manner similar to that previously described (14). Briefly, mice were immunized with MOG_35-55 (200 μg, Sigma Genosys) and complete Freund’s adjuvant (2.5 mg, Chondrex). Three weeks later, after dorsal laminectomy at Th8, a 1.5 μL of cytokine mixture containing TNF-α (750 ng, R&D Systems) and IFN-γ (1 μg, Peprotech) was injected into the spinal cord at a depth of 0.5 mm. Pertussis toxin (List Biological Laboratories, Inc.) was administered intravenously 0 and 2 days post cytokine injection. Humanized anti-RGMa antibody (10 mg/kg) and an isotype control antibody (palivizumab, 10 mg/kg) were intravenously injected twice a week from 9 days post cytokine injection (DPI) onwards. Clinical scores were assessed based on the following criteria: 0, no abnormalities noted; 0.5, weak tail reflex; 1, loss of tail reflex; 1.5, loss of tail reflex with hind leg inhibition without gait abnormality; 2, partial hind limb paralysis with gait abnormality; 3, complete hind limb paralysis; 4, front and hind limb paralysis; and 5, moribund state. The assessor was blinded to the treatment groups.

MR imaging were performed at 7, 14, 21 DPI ( Figure 1A ). A BioSpec 117/11 system (Bruker, 11.7 Tesla) equipped with Quadrature Volume Coils (Rapid Biomedical) was used for MR imaging. The fast spin echo protocol (RARE) was used for T1-weighted and diffusion tensor imaging (DTI). Sagittal T1-weighted imaging (TR/TE = 600/16 ms, 40 × 20 mm FoV, 256 × 128 acquisition matrix, 5 × 0.8 mm slices, 0.8 mm interslice distance, NEX = 8), axial T1-weighted imaging (TR/TE = 500/18 ms, 26 × 26 mm FoV, 200 × 200 acquisition matrix, 11 × 0.8 mm slices, 0.8 mm interslice distance, NEX =4), and DTI [single-shot echo-planar imaging with 6 diffusion directions (b = 4900 s/mm) and 1× b0 acquisitions, TR/TE = 2750/18 ms, 26 × 26 mm FoV, 128 × 128 acquisition matrix, 11 × 0.8 mm slices, 0.8 mm interslice distance] were conducted. The spinal segmental level was determined based on the vertebral landmarks, by using sagittal T1-weighted images (22).

DCE-MRI was performed using axial T1-weighted imaging. Following a baseline scan, Gd (Omniscan, Daiichi Sankyo, Japan) was injected as a bolus at 0.25 mmol/kg via catheter. A total of six axial images were obtained throughout 10 min following Gd injection. In both DTI and DCE-MRI, axial images were acquired to set the center of slice at 8th thoracic vertebra ( Figure 1B ).

The acquired images were analyzed using Fiji (http://fiji.sc/). Parametric images of DTI, fractional anisotropy (FA), axial diffusivity (AD), and radial diffusivity (RD) maps, which are considered as indicators of white matter, axon, and myelin integrity (23, 24), were used: especially, in EAE models, AD and RD maps were reported to correlates with axon integrity (confirmed by SMI staining), and myelin integrity (confirmed by MBP staining), respectively (23, 25, 26). Images were generated using built-in software (Paravision 5.1). The regions of interest (ROIs) in the slices were drawn around the spinal cords ( Figure 1C ). The FA, AD, and RD maps were binarized, and the area ratio to spinal cords was calculated as %FA positive area, %AD positive area, %RD negative area, respectively. The FA, AD, and RD binarized thresholds were determined to be ≥ 0.55 (unitless), ≥ 1.4 μm²/ms, and ≦0.5 μm²/ms, respectively. These thresholds were calculated so that the ratio of the binarized area to the whole spinal cord area was the same as that of white matter area to the whole spinal cord area, approximately 55% in the cytokine injection site, as shown in Figure 1D .

In DCE-MRI, ROIs were drawn in the spinal cords ( Figure 1E ). The extent of Gd leakage into the spinal cord was quantified as the sum of the Gd influx rate in each slice. The influx rate, Kp-sc ( Figure 1E ), was calculated as a method previously reported by Tatar et al. (27).

Transcriptional analysis was performed at 14 DPI. Spinal cords were sampled with a width of 4.5 mm (centered on cytokine injection site) after transcardial perfusion with ice-cold PBS, then immediately frozen with liquid nitrogen and stored at −80°C until extraction. Total RNA extraction was performed using the Maxwell^® RSC simplyRNA Tissue Kit (Promega) according to the manufacturer’s protocol. Microarray analysis was performed using Mouse Genome 430 2.0 Array (Thermo Fisher Scientific) according to the manufacturer’s protocol. Single array analysis was performed using Microarray Suite version 5.0 (MAS5.0) and global scaling as the normalization method. Differential expression analysis and gene set enrichment analysis (GSEA) were performed using R-packages limma (28) and clusterProfiler (29) respectively. The gene set libraries for GSEA ( Supplementary File 1 ) were prepared from Enrichr (30), CellMarker (31), and Munji et al. reported genes (32).

Deeply anesthetized mice were transcardially perfused with ice-cold PBS, followed by 4% ice-cold paraformaldehyde (PFA) in PBS, then post-fixed overnight in 4% PFA in PBS. For Luxol Fast Blue staining, 10-μm transverse sections were incubated in 0.1% Luxol Fast Blue (Sigma) overnight at 56°C. Images were captured using an Olympus SZX16 bright-field microscope.

For immunohistochemistry, the spinal segmental level was determined based on the 8th thoracic vertebra, and the extracted spinal cord were divided into rostral, epicenter, and caudal segment of 3 mm each before post-fixation. Post-fixed samples were cryoprotected with 30% sucrose. Afterwards, spinal cord was serially sectioned into 20-μm transverse sections. Goat anti-P-selectin (1:1,000, AF737; R&D Systems), rabbit anti-collagen 1 (1:100, ab34710; Abcam), goat anti-collagen 4 (1:50, AB769; Sigma-Aldrich), rat anti-CD31 (1:25, 550274; BD Pharmingen), and goat anti-RGMa (1:100, AF2458; R&D Systems) were used as primary antibodies. For RGMa staining, antigen retrieval was performed by incubating the sections with 10 mM citrate buffer (pH 6.0) at 80°C for 10 min. Afterwards, sections were incubated with appropriate Alexa Fluor secondary antibodies (1:500, Invitrogen). Images were captured using an Olympus IX83 fluorescence microscope equipped with the CellSens software (Olympus).

To evaluate abnormal vascular lesions, CD31+, P-selectin+, perivascular collagen 1+, and collagen 4+ areas were measured. CD31+ vascular area was extracted using the canny edge detection and fill hole method, and the P-selectin high-intensity region within the CD31+ vascular area was calculated as the P-selectin+ area. The perivascular collagen area was calculated by subtracting the CD31+ area from the collagen+ area in vessels. All analyses were performed using MATLAB (MathWorks), and regions of interest were drawn using bright-field images. Three images were quantified for each spinal cord region, and the mean value of those was analyzed as individual animal data.

Evans Blue (Wako) dissolved in PBS (4% w/v) was intravenously injected at a total volume of 150 μL. After 1 h, the mice were transcardially perfused with ice-cold PBS, followed by 4% ice-cold paraformaldehyde (PFA) in PBS, under anesthesia. The leakage area at 7 DPI was evaluated to confirm its relationship with Gd leakage.

Dunnett’s multiple comparison test, Mann−Whitney test, one-way analysis of variance (ANOVA) followed by the Tukey–Kramer test, two-way ANOVA followed by the Bonferroni test, Pearson correlation analysis, Spearman rank correlation analysis, standardized partial multiple regression analysis, and simple linear regression analysis were performed using GraphPad Prism 9.0.2 (GraphPad Software). For transcriptional analysis, p-adjust value was calculated by the R-packages limma (28) and clusterProfiler (29). Error bars represent the standard error of the mean. p < 0.05 was considered to indicate statistical significance. For correlation analyses, the absolute value of correlation coefficients (r) was classified as follows: 0.0–0.2, very weak to negligible correlation; 0.2–0.4, weak correlation; 0.4–0.7, moderate correlation; 0.7–1.0, strong correlation. For regression analyses, the absolute value of R-squared (r ²) was classified as follows: 0.0-0.2, none or very weak effect size; 0.2-0.4, weak effect size; 0.4-0.7, moderate effect size; 0.7-1.0, strong effect size.

To evaluate the extent of BSCB disruption and myelin/axon integrity in the spinal cord of targeted EAE mice, we performed DCE-MRI and DTI by using high-field (11.7 Tesla) MRI ( Figures 1A–E ). To assess the validity of the DCE-MRI and DTI protocols, we compared fractional anisotropy (FA) maps with LFB staining for white matter integrity ( Figure 1F ), and the extent of Gd leakage with Evans blue leakage for BSCB disruption ( Figure 1G ). Both DTI and DCE-MRI reflected white matter integrity and the extent of BSCB disruption. We then performed longitudinal analysis of DCE-MRI and DTI at 7, 14, and 21 days post-induction (DPI) of targeted EAE. The extent of Gd leakage showed a peak in the acute phase (7 DPI) and then gradually decreased ( Figure 1H ). %FA positive area, %AD positive area, and %RD negative area exhibited a gradual decrease, suggesting the progression of white matter disintegration toward the chronic phase ( Figures 1I–K ; Supplementary Figures S1A–D ).

To evaluate the therapeutic effect of anti-RGMa, antibody administration was initiated after the onset of motor symptoms (9 DPI), while longitudinal DCE-MRI and DTI were performed at 7, 14, and 21 DPI ( Figure 2A ; Supplementary Figures 1E–H ). In the preliminary experiments, we observed a tendency of worse outcome in individuals with a strong Gd leakage at 7 DPI. Therefore, to assess the effect of anti-RGMa antibody treatment, we divided mice into two groups (anti-RGMa antibody group and isotype control antibody group) based on the extent of Gd leakage at 7 DPI to avoid deviation between the groups. Anti-RGMa treatment had a prompt effect on clinical symptoms ( Figure 2B ). MRI parameters were also attenuated: at 14 DPI, the extent of Gd leakage was drastically reduced by the treatment, and the decreases in %FA positive area and %AD positive area were also attenuated ( Figures 2C–E ). Furthermore, both percentage areas were improved in the chronic phase (21 DPI) ( Figures 2D, F ). These results demonstrated that anti-RGMa antibody treatment had a strong and prompt therapeutic effect on BSCB disruption and white matter disintegration in targeted EAE mice.

To evaluate the correlation between each MRI parameter and neurological disability in targeted EAE mice at each time point, we performed Spearman’s correlation analysis ( Figure 3 ). We also performed correlation analysis using pooled data from all timepoint ( Supplementary Figures 1I–L ). The extent of Gd leakage strongly correlated with the clinical score in the acute phase (7 DPI, r = 0.831, p < 0.001) and moderately correlated with that at the subacute phase (14 DPI, r = 0.549, p = 0.018). %RD negative area strongly correlated with the clinical score at the chronic phase (21 DPI, r = -0.748, p < 0.001). Furthermore, in order to evaluate the effect size of these MRI parameters on clinical scores, we calculated standardized partial regression coefficients (β) by using multiple linear regression analysis ( Table 1 ). The extent of Gd leakage had a greater effect on the clinical score in the acute phase (7 DPI, β = 0.682, p < 0.001), while the %RD area had a greater effect on the clinical score in the chronic phase (21 DPI, β = -0.662, p = 0.039). Taken together, the extent of Gd leakage was highly correlated with neurologic disability in the acute phase, while the increase in RD was well correlated with that in the later phase.

To evaluate the strength of the relationship between the extent of BSCB disruption and neurological symptoms as well as white matter, axon, or myelin integrity, we performed correlation analysis between Gd leakage and the other parameters (%FA, %AD positive area, %RD negative area, and clinical score) at later time points based on data from the isotype control antibody group ( Table 2 ). The extent of Gd leakage strongly correlated with %RD negative area 7 days later (Day 7 Gd leakage – Day 14%RD, r = 0.699, p = 0.036; Day 14 Gd leakage – Day 21%RD, r = 0.877, p = 0.002). Furthermore, in order to evaluate the predictive power of BSCB disruption with regard to myelin integrity after 7 days, we applied a prediction model based on BSCB disruption and demyelination. The extent of Gd leakage was used as the input value, %RD negative area at 7 days later was used as the output value, and the effect size (r² ) of regression analysis was considered as predictive power ( Figure 4A ). In the isotype control group, the extent of Gd leakage predicted %RD negative area (moderately at 14 DPI, r² = 0.489, p = 0.036; strongly at 21 DPI, r² = 0.769, p = 0.002) ( Figures 4B, C ). However, anti-RGMa antibody treatment abolished its predictive power (14 DPI, r² = 0.105, p = 0.361; 21 DPI, r² = 0.045, p = 0.582) ( Figures 4D, E ). These results suggest that the extent of BSCB disruption can be used as a predictive biomarker for demyelination severity in the later phase.

Given that anti-RGMa effectively ameliorated BSCB disruption 5 days after treatment initiation, we sought to explore the molecular profile involved in this effect. We performed microarray analysis of mouse spinal cords at 14 DPI (5 days after the first anti-RGMa antibody injection, Figure 5A ). Using pre-processed microarray data ( Supplementary File 2 ), principal component analysis was performed, and the result shows clear difference among naïve, control, and anti-RGMa antibody groups ( Figure 5B ). The differentially expressed genes (p-value < 0.05 and fold change > 1.5, isotype control group v.s. anti-RGMa antibody group) were presented in volcano plot and MA plot ( Figure 5C ), and the ranked gene list were used for GSEA. The GSEA results were summarized in Supplementary File 3 . GSEA analysis of CNS cell markers showed significant enrichment of marker genes in neurons, oligodendrocytes, astrocytes, macrophages, mural cells, smooth muscle cells, and endothelial cells ( Figures 5D, E ; Supplementary Figure 2 ). In particular, endothelial cell and mural cell (pericyte) markers were negatively enriched in anti-RGMa treated group. Therefore, we further analyzed the gene sets with negative enrichment in GSEA ( Figures 5F-H ), and revealed that gene sets related to extracellular matrix production and epithelial mesenchymal transition (EMT) were strongly regulated by anti-RGMa antibody treatment. Further, we performed GSEA by utilizing endothelial BBB-related gene sets reported by Munji et al. (32), and revealed that anti-RGMa treatment significantly altered BBB dysfunction modules ( Figure 5I ; Supplementary Figure 3 ).

Alteration of gene expression in each gene set among the naïve, control antibody treated, and RGMa antibody treated groups is shown in Figure 5J . In addition to BBB dysfunction-related genes and ECM production-related genes, endothelial mesenchymal transition (EndMT)-related genes were also analyzed based on its similarity with EMT. Most genes in each group were upregulated by EAE induction, and this upregulation was ameliorated by anti-RGMa antibody treatment. Because EndMT is critically involved in endothelial dysfunction (33) including BBB pathophysiology (34, 35), these results support the therapeutic effect of anti-RGMa antibody on vascular pathology.

To further confirm the effect of anti-RGMa on vascular pathology, we performed immunohistochemical analysis of vascular endothelial cell marker CD31 and basement membrane marker type IV collagen, both of which are BSCB components ( Figure 6A ). Endothelial cells exhibited abnormal thickening at 14 DPI ( Figure 6B ). Type IV collagen staining revealed a decrease in staining at 8 DPI, perivascular deposition at 14 DPI, and fibrous mesh formation at 21 DPI ( Figure 6B ). We also observed dense RGMa immunoreactivity around endothelial cells within the lesion site at 14 and 21 DPI ( Figure 6C ). Given the presence of remarkable vascular pathology at 14 DPI, we evaluated the effects of anti-RGMa antibody treatment (started at 9 DPI) by using P-selectin and type I collagen, which were included in the BBB dysfunction and EndMT-related genes and whose expression was altered by anti-RGMa based on transcriptome analysis ( Figure 7A ). As shown in Figure 7B , these proteins were confirmed as components of abnormal vessels in the spinal cord of targeted EAE mice. Subsequent analysis indicated that vascular abnormalities were prominent in the white matter, especially within the epicenter region ( Figure 7C ). Quantitative analysis revealed that perivascular collagen deposition in the white matter tended to be suppressed by anti-RGMa antibody treatment ( Figure 7D ). Moreover, endothelial cell abnormality and P-selectin upregulation were significantly attenuated by antibody treatment ( Figure 7D ).