Investigating demyelination, iron accumulation, and synaptic loss in Alzheimer’s disease using multimodal imaging techniques

Alzheimer’s disease (AD), the most common type of dementia, is associated with neuronal death and synaptic loss [1], [2]. Pathological aggregation of amyloid-beta and tau protein are key elements of AD pathophysiology. Myelin loss and iron accumulation in the brain are also fundamental features of aging and dementia [3], [4], but are less frequently investigated. Quantitative MRI (qMRI) enables us to determine the brain tissue parameters such as magnetization transfer (MT) and effective transverse relaxation (R2*), which leads to the detection of microstructural tissue-related alterations in aging and neurodegenerative diseases [5].
Here we investigate the association of neurodegeneration (as indexed by loss of synaptic density), increased iron accumulation, and decreased myelination in Alzheimer's disease in cohorts of 24 amyloid-positive patients (AD, 11 males and 13 females) and 19 healthy controls (HC, 9 males, and 10 females). All participants underwent a multiparameter qMRI protocol, which was processed to generate probability maps for MTsat and R2* [5]. Synaptic density was evaluated by the total volume distribution (Vt)maps, representing the distribution of the [18F] UCB-H PET radiotracer in the brain [6].
The data is organized according to the Brain Imaging Data Structure (BIDS) [7]. MRI data processing was performed in MATLAB (The MathWorks Inc., Natick, MA, USA) using the SPM12 framework (www.fil.ion.ucl.ac.uk/spm) and the hMRI toolbox [8] after modifications to make MR data compatible with the BIDS format [9]. Each multi-parameter map presents a different tissue-related (semi-)quantitative property, and therefore the qMRI maps have specific units. Therefore, all maps were z-transformed to ensure the comparability of the maps in a multivariate model. Then, we used General Linear Model (GLM) to test the groups against each other using age and sex as the covariates. Also, a multivariate GLM (mGLM) was performed on all modalities using the MSPM toolbox (https://github.com/LREN-CHUV/MSPM) to test differences in groups controlling for the age and sex of the participants [10].
Univariate group analysis of MTsat data resulted in a significant difference at the cluster level in the right hippocampus with p_cluster<0.05 FWE corrected and p_voxel<.001 uncorrected as cluster forming threshold (Figure1.A). In contrast, the same analysis for R2* modality reveals no difference between the groups. PET_Vt maps showed a difference between AD and HC at p_voxel<0.05 (FWE corrected) in the right amygdala and hippocampus (Figure1.B), which agrees with previously reported results in [6]. See table.1 for more information.
Multimodal analysis combining R2*, MTsat, and PET_Vt shows a bilateral difference in hippocampus between patients and healthy controls for voxel-wise analysis with corrected FWE P-voxel < 0.05 (Figure1.C). The canonical analysis suggests that AD patients had combined decreased myelination, decreased synaptic density, and increased iron in the hippocampus compared to controls.
To conclude, in the case of AD, there is an interaction between neuropathological risk factors, therefore, to restrain the true multivariate nature of the data and better control for the false positive rate, one should use the multivariate model over multiple univariate models.