Neuroimaging Correlates of Cognitive Deficits in Wilson's Disease

Abstract Background Cognitive impairment is common in neurological presentations of Wilson's disease (WD). Various domains can be affected, and subclinical deficits have been reported in patients with hepatic presentations. Associations with imaging abnormalities have not been systematically tested. Objective The aim was to determine the neuroanatomical basis for cognitive deficits in WD. Methods We performed a 16‐item neuropsychological test battery and magnetic resonance brain imaging in 40 patients with WD. The scores for each test were compared between patients with neurological and hepatic presentations and with normative data. Associations with Unified Wilson's Disease Rating Scale neurological examination subscores were examined. Quantitative, whole‐brain, multimodal imaging analyses were used to identify associations with neuroimaging abnormalities in chronically treated stable patients. Results Abstract reasoning, executive function, processing speed, calculation, and visuospatial function scores were lower in patients with neurological presentations than in those with hepatic presentations and correlated with neurological examination subscores. Deficits in abstract reasoning and phonemic fluency were associated with lower putamen volumes even after controlling for neurological severity. About half of patients with hepatic presentations had poor performance in memory for faces, cognitive flexibility, or associative learning relative to normative data. These deficits were associated with widespread cortical atrophy and/or white matter diffusion abnormalities. Conclusions Subtle cognitive deficits in patients with seemingly hepatic presentations represent a distinct neurological phenotype associated with diffuse cortical and white matter pathology. This may precede the classical neurological phenotype characterized by movement disorders and executive dysfunction and be associated with basal ganglia damage. A binary phenotypic classification for WD may no longer be appropriate. © 2022 The Authors. Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society

Partial Fourier 6/8 Monopolar readout Flow compensation for first echo.
2D multi-slice GRAPPA, generalised autocalibrating partial parallel acquisition; IR-SPACE, inversion recovery -sampling perfection with application optimised contrast using different flip angle evolutions; MPRAGE, magnetisation prepared rapid gradient echo; PGSE-EPI, pulsed-gradient spin echo -echo planar imaging Imaging acquisition, processing and analysis

T1-weighted (structural) imaging
Voxel-based morphometry (VBM) was performed using Statistical Parametric Mapping (SPM12,version 7771,http://www.fil.ion.ucl.ac.uk/spm) to identify clusters where decreasing grey matter (GM) volume were associated where decreasing cognitive performance. 33 T1weighted (structural) images were segmented into GM, WM and cerebrospinal fluid (CSF) using standard procedures and spatially normalised using the fast-diffeomorphic image registration algorithm. 34 GM and WM segments were transformed into MNI152 space (Montreal Neurological Institute, McGill University, Canada), modulated and smoothed using a Gaussian kernel with 8 mm full-width at half maximum to create pre-processed GM tissue maps. All segmentations were visually checked for quality. The pre-processed tissue maps were fitted to multiple regression analyses to identify associations with neuropsychological test scores. Total intracranial volume (TIV), calculated in SPM, was included as a nuisance covariate, in addition to age and sex. 35 Statistical thresholds were set at P < 0.05 for familywise-error (FWE) correction and then lowered to uncorrected P < 0.001. A minimum cluster size of 20 voxels was set and thresholded statistical maps were overlaid onto the study-wise mean template.
We conducted a separate region-of-interest (ROI) analysis to assess atrophy in specific subcortical structures. T1-weighted images were bias-corrected and parcellated using the geodesic information flow (GIF) pipeline, 36 based on atlas propagation and label fusion. The brainstem was subsequently segmented using a customized version of a FreeSurfer module. 37 We did not manually correct the automatic segmentation of any ROI. The volume of eight subcortical ROI including the caudate, putamen, pallidum, thalamus, amygdala, midbrain, pons and cerebellum, were extracted and expressed as a percentage of total intracranial volume (TIV), calculated in SPM. All segmentations were visually checked for quality. Linear regression was used to identify associations with neuropsychological test scores. P values for coefficients of interest both with and without false discovery rate (FDR) correction were calculated in R (version 3.6.0, http://www.R-project.org).

FLAIR imaging
WMHs were segmented using Bayesian model selection, an automated lesion segmentation tool applied to rigidly co-registered T1-weighted (structural) and FLAIR sequences. 38 A Gaussian mixture model with dynamically evolving number of components was fit to the data, modelling simultaneously healthy and non-expected observations. WMH-related measures were introduced to the model through subject-specific statistical atlases obtained using the GIF pipeline. After convergence, the model was used to select candidate lesion voxels whose aggregation in connected components was automatically classified as lesion or artefact. WMH segmentations were then visually inspected and flagged if there were significant segmentation errors. This quality control stage was used to make improvements to the automated WMH segmentation, thereby maximising the number of usable segmentations.
The volume of WMHs within 40 anatomically-defined regions were calculated for each participant. 39 WM was separated into four equidistant layers between the ventricular surface and the cortical GM/WM interface. These were then divided into left and right frontal, temporal, parietal and occipital lobes using the GIF parcellation. The basal ganglia and infratentorial regions were considered separately. The volume of WMHs within each region was loge-transformed to reduce skewness. A linear regression model was used to identify associations with neuropsychological test scores. TIV was included as a covariate of no interest, in addition to age and sex. P values for coefficients of interest were calculated with FDR correction in R and these were summarised in bullseye plots to illustrate their anatomical distribution. 39

Diffusion-weighted imaging
The Functional MRI of the Brain Software Library (FSL, version 6.0.3, https://fsl.fmrib.ox.ac.uk/fsl) was used to pre-process DWI data prior to fitting the single tensor model, resulting in volumetric diffusion tensor imaging (DTI) data. DTI datasets were then analysed using tract-based spatial statistics (TBSS). 40 Pre-processing included EDDY to correct for motion and eddy-currents with outlier replacement enabled. FUGUE was applied to correct for distortions using fieldmaps. Tensors were fitted using DTIFIT and fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD) maps were generated, skeletonised and aligned using TBSS. Design matrices for identifying associations with neuropsychological test scores were generated using the general linear model.
Finally, RANDOMISE was used to perform nonparametric permutation analyses based on each design matrix. Covariates were mean-centred and 10,000 permutations of the data were carried out. The threshold-free cluster enhancement algorithm was used to identify clusters of voxels with a FWE corrected P value < 0.05. 41 Clusters of increased or decreased FA, MD, AD and RD were then overlaid onto a mask of the WM skeleton (created using the mean skeletonised FA map) and the MNI152 template.

Susceptibility-weighted imaging
Quantitative susceptibility maps (QSM) were reconstructed from susceptibility-weighted images using a Multi-Scale Dipole Inversion (MSDI)-based pipeline for coil-combined, multigradient echo data in QSMbox (https://gitlab.com/acostaj/QSMbox). 42 Pre-processing steps included unwrapping of complex 3D phase data using a discrete Laplacian method followed by background field removal using Laplacian boundary extraction and variable spherical mean filtering. All steps were applied using default settings. Whole-brain analyses were performed with the QSMexplorer pipeline (https://gitlab.com/acostaj/QSMexplorer). 43 A study-wise space was created from T1-weighted sequences using Advanced Normalisation Tools (ANTs).
Bias-corrected magnitude images were then used to transform the quantitative susceptibility maps to the study-wise space. Absolute susceptibility maps smoothed with a 3 mm standard deviation 3D Gaussian kernel were used to identify associations with neuropsychological test scores in stable patients. RANDOMISE was used to perform nonparametric permutation analyses based on each design matrix. Covariates were mean-centred and 10,000 permutations were performed. The GM segment generated in SPM12 was used to mask the absolute maps.
Threshold-free cluster enhancement was enabled to identify clusters of voxels with a familywise error corrected P value < 0.05. Clusters were then overlaid (for result visualisation) onto the study-wise template. scores. Tissue maps show correlations between neuropsychological tests scores and mean diffusivity/fractional anisotropy in white matter tracts for FWE-corrected P values < 0.05. Tracts where diffusion parameters increase (red) or decrease (blue) with worsening cognitive performance are overlaid onto the white matter skeleton (green). Axial slices at z = -34, -12, 10 and 32 are shown.