Axial diffusivity is the primary correlate of axonal injury in the experimental autoimmune encephalomyelitis spinal cord: a quantitative pixelwise analysis

Abstract

The dissociation between magnetic resonance imaging (MRI) and permanent disability in multiple sclerosis (MS), termed the clinicoradiological paradox, can primarily be attributed to the lack of specificity of conventional, relaxivity-based MRI measurements in detecting axonal damage, the primary pathological correlate of long-term impairment in MS. Diffusion tensor imaging (DTI) has shown promise in specifically detecting axonal damage and demyelination in MS and its animal model, experimental autoimmune encephalomyelitis (EAE). To quantify the specificity of DTI in detecting axonal injury, in vivo DTI maps from the spinal cords of mice with EAE and quantitative histological maps were both registered to a common space. A pixelwise correlation analysis between DTI parameters, histological metrics, and EAE scores revealed a significant correlation between the water diffusion parallel to the white matter fibers, or axial diffusivity, and EAE score. Furthermore, axial diffusivity was the primary correlate of quantitative staining for neurofilaments (SMI31), markers of axonal integrity. Both axial diffusivity and neurofilament staining were decreased throughout the entire white matter, not solely within the demyelinated lesions seen in EAE. In contrast, although anisotropy was significantly correlated with EAE score, it was not correlated with axonal damage. The results demonstrate a strong, quantitative relationship between axial diffusivity and axonal damage and show that anisotropy is not specific for axonal damage after inflammatory demyelination.

Figure 1.

Clinical course of EAE mice. Mean daily EAE score for 32 mice. Error bars represent SD. All mice underwent DTI in the chronic phase (>25 d after immunization).

Figure 2.

Registration of DTI parameter maps. DTI parameter maps were manually registered to a common space using rotations, translations, and global scaling. The average T2-weighted (b = 0) image from eight control mice (A) shows that the spinal cord is well aligned after registration. The average T2-weighted (B), RA (C), axial (D), and radial (E) parameter maps demonstrate the preservation of white matter borders after the registration. The maps of the coefficient of variation for axial (F) and radial (G) diffusivity further demonstrate the reproducibility of the in vivo DTI data and the registration accuracy.

Figure 3.

Registration and quantification of histological sections. Corresponding landmarks were placed on the average RA map (A) and each of the digitized histological images (B; SMI31 shown here). Landmarks were manually placed along the white matter border and along prominent white and gray matter interfaces. A thin-plate spline deformation grid was derived and overlaid on the histological image (C; magnified view) in which each square of the grid corresponds to a single pixel on the MRI image. The histological image was segmented (D) into foreground (red) and background (black) using a threshold value derived from the pixel intensities of the surrounding spinal nerves, which are not affected in MOG-induced EAE. The area fraction of foreground staining for each square of the grid was computed to produce a quantitative histological image (E) that was registered to the MR images.

Figure 4.

Mean DTI parameter maps grouped by EAE clinical score. The spinal cord white matter of mice with increasing severity of EAE (from top to bottom) shows a decrease in axial diffusivity that scales with the degree of impairment. The decrease was prominent in nearly all of the ventrolateral white matter as well as in the most posterior portion of the dorsal white matter. Radial diffusivity was increased in the ventrolateral white matter in mice with EAE, but there was no consistent pattern associated with the severity of EAE. RA decreased with increasing severity, reflecting its dependence on axial diffusivity. Images are from spinal segment L2.

Figure 5.

Pixelwise correlations between DTI parameters and EAE scores. Four adjacent slices of lumbar spinal cord segments L1 and L2 are shown. Axial diffusivity and RA were significantly correlated with EAE score in nearly the entire ventrolateral white matter and the most dorsal portion of the dorsal white matter. Radial diffusivity was significantly correlated with EAE score in only a few regions along the periphery of the ventrolateral white matter. Correlation (Spearman's r) maps were thresholded at p < 0.05 and corrected for multiple comparisons, and significant voxels were overlaid on the average DTI parameter maps from all mice.

Figure 6.

Mean quantitative histological maps grouped by EAE score. SMI31 staining (left) reveals a loss of intact axons that scales with the severity of EAE (from top to bottom). MBP staining for intact myelin is decreased in the EAE spinal cord white matter, but no consistent relationship with EAE score is evident. DAPI staining is indicative of cellularity and is particularly increased in certain regions in mice with clinical scores of 4 (bottom).

Figure 7.

Pixelwise correlations between quantitative histological measures and EAE score. SMI31 staining was significantly correlated with EAE score in the majority of the spinal cord white matter. MBP staining was significantly correlated with EAE score along the white matter periphery. DAPI staining demonstrated only few pixels reaching significance. Correlation (Spearman's r) maps were thresholded at p < 0.05, corrected for multiple comparisons, and overlaid on the mean registered histological maps from all mice.

Figure 8.

Pixelwise correlations between histological and DTI measures. SMI31, MBP, and DAPI were highly correlated with one another (A), reflecting the intimate relationship between inflammation, demyelination, and axonal damage in EAE. Despite the interrelation, only axial diffusivity was highly correlated with SMI31 staining (B). To address the potential confounds affecting the correlation between axial diffusivity and SMI31, MBP and DAPI were included as covariates in the regression model (C). Axial diffusivity was still significantly correlated with axonal damage, although less so. In all panels, significant pixels were thresholded at p < 0.05 (corrected for multiple comparisons) and are shown in color overlaid on the correlation (Pearson's r) maps, except for C, which is overlaid on the average SMI31 map. The figure legend applies to all panels.