Diffusion MRI tracks cortical microstructural changes during the early stages of Alzheimer’s disease

Abstract There is increased interest in developing markers reflecting microstructural changes that could serve as outcome measures in clinical trials. This is especially important after unexpected results in trials evaluating disease-modifying therapies targeting amyloid-β (Aβ), where morphological metrics from MRI showed increased volume loss despite promising clinical treatment effects. In this study, changes over time in cortical mean diffusivity, derived using diffusion tensor imaging, were investigated in a large cohort (n = 424) of non-demented participants from the Swedish BioFINDER study. Participants were stratified following the Aβ/tau (AT) framework. The results revealed a widespread increase in mean diffusivity over time, including both temporal and parietal cortical regions, in Aβ-positive but still tau-negative individuals. These increases were steeper in Aβ-positive and tau-positive individuals and robust to the inclusion of cortical thickness in the model. A steeper increase in mean diffusivity was also associated with both changes over time in fluid markers reflecting astrocytic activity (i.e. plasma level of glial fibrillary acidic protein and CSF levels of YKL-40) and worsening of cognitive performance (all P < 0.01). By tracking cortical microstructural changes over time and possibly reflecting variations related to the astrocytic response, cortical mean diffusivity emerges as a promising marker for tracking treatments-induced microstructural changes in clinical trials.

from the dMRI space to the T1w space (target: T1w image, moving image: first volume from the dMRI series with b-value=0).The registration was implemented using antsRegistration (ANTs version: 2.1).FSL routines were used to fit the DTI model to the dMRI data.After that, the mean diffusivity (MD) map was warped to the space of the T1w scan acquired during the same session using the transformations derived from the non-linear registration and the MD map was projected on the cortical surface by sampling MD values at three equidistant points between white and pial surfaces starting and stopping within 25% from the borders of the cortical ribbon and averaging them to provide a single MD value for each vertex of the subject's cortical surface.These processing steps were performing using FreeSurfer commands as previously reported 2,3 .Finally, MD median values were extracted the 68 cortical regions of the Desikan-Killiany atlas.Critically for this study, the Desikan-Killiany atlas was warped into the individual time points T1w space following the FreeSurfer longitudinal pipeline which includes the creation of within subject templates as a first step of segmentation and reconstruction.Moreover, each time point is initialized within the template to reduce the variability in the optimization process (see supplementary figure 1 for an overview of the processing steps).

Cortical thickness-corrected changes over time in MD values
To account for changes over time in cortical thickness we corrected the regional mean diffusivity values for regional cortical thickness using the covariance method before estimating changes over time in mean diffusivity using LME.To this end, the regional MD values were first regressed against the corresponding cortical thickness (CT) values.After that, the CT-

Regional changes over time in CT-corrected cortical MD differs between biomarker-defined groups
Longitudinal CT-corrected cortical MD still revealed microstructural differences, namely steeper increase in the CT-corrected MD over time, in the Aβ-positive/tau-negative group when compared with the control (Aβ-negative/tau-negative) group.Although the spatial extent of the results is more limited compared to the analysis without accounting for changes over time in CT, regions critically involved in the AD disease process (e.g., isthmuscingulate, the anterior portion of the cingulate cortex, the entorhinal cortex and the parahippocampal gyrus) still showed statistically significant differences in changes over time in MD (standardized-β range: 0.26 -0.44; see supplementary figure 2A).
Difference in longitudinal CT-corrected cortical MD between the Aβ-positive/tau-positive group and the Aβ-positive/tau-negative group were still wide-spread and encompassed several regions in both the temporal and the parietal lobe as well as frontal regions (see supplementary figure 2B; standardized-β range: 0.24 -0.64).

Changes over time in fluid markers of astrocytic activity are associated with changes over time in CT-corrected cortical MD
In the subgroup of participants with available plasma levels of GFAP (N=322), the regression analysis revealed a widespread positive association between longitudinal CT-corrected cortical MD and changes over time in levels of GFAP although the spatial extent was reduced in comparison to the analysis employing the uncorrected MD values (see supplementary figure 2B, standardized-β range: 0.16 -0.29).Similar results, although with lower standardize-β values, were also found when investigating the association with CSF level of YKL-40 (N=292; see supplementary figure 2C, standardized-β range: 0.13 -0.24).

Changes over time in plasma NfL levels are associated with changes over time in CT-corrected cortical MD
Longitudinal CT-corrected cortical MD was also associated with changes over time in plasma levels of NfL although with a reduction in the spatial extent of the results compared to the analysis using uncorrected estimate of changes over time in MD (see supplementary figure 2E, standardized-β range: 0.14 -0.24).

Changes over time in CSF levels of sTREM-2 are not associated with changes over time in cortical MD
In the subgroup of participants with available longitudinal CSF sTREM-2 levels (N=292), the regression analysis revealed no significant association between changes over time in cortical MD and changes over time in sTREM-2.The analysis focusing on a-priori defined ROIs showed also no significant associations [early-Aβ ROI: standardized-β=0.07,p>0.2; temporal ROI: standardized-β=0.09,p>0.1]

Changes over time in CSF level of GFAP are associated with changes over time in cortical MD
In the subgroup of participants with available data on the astrocytic marker GFAP in CSF (N=292), the regression analysis revealed a widespread positive association between longitudinal cortical MD and changes over time in levels of GFAP with higher standardized-β in temporal regions (see supplementary figure 1A, standardized-β range: 0.13 -0.21).

Supplementary references
corrected MD values were obtained with: CT-corrected-MDROIp1 = MDROIp1slope-modelROI * (CTROIp1 -CTROImean), where MDROIp1 = MD values of a specific ROI for the participant p1; slope-modelROI = slope of the model regressing MD values in a specific ROI against the cortical thickness values from the same ROI; CTROIp1 = cortical thickness values of a specific ROI for the participant p1; CTROImean = mean cortical thickness values of a specific ROI across all participants.The CT-corrected MD values were then used to compute CT-corrected estimate of MD changes over time by computing the individual-specific random slopes using the CTcorrected MD values as dependent variable on the LME models.Finally, we repeated the main analyses of the study employing the CT-corrected changes over time in MD instead of the uncorrected changes over time in MD.