Disruption of gray matter microstructure in Alzheimer’s disease continuum using fiber orientation


 There have been several MR imaging biomarkers of Alzheimer’s disease (AD) for early diagnosis. Cortical mean diffusivity (MD) is one of them for the study of the cortical microstructural change in AD. However, the feasibility of MD often remain in doubt as partial volume effects may overestimate the results. This study aims to investigate feasible gray matter microstructural biomarker with higher sensitivity for early AD detection. We propose diffusion tensor imaging (DTI) measure, ‘radiality’, for early AD biomarker. It is a dot product between cortical surface normal vector and primary diffusion direction, which reflects the fiber orientation within the cortical column.
Here, we gathered neuroimages from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database; 78 cognitive normal, 50 early mild cognitive impairment (EMCI), 34 late mild cognitive impairment (LMCI), and 39 AD patients. Then, we evaluated cortical thickness (CTh), MD, amount of amyloid and tau accumulations using positron emission tomography (PET), which are conventional AD biomarkers. Radiality was projected on gray matter surface to compare and validate the changes along other neuroimage biomarkers.
Results showed decreased radiality primarily in entorhinal, insula, frontal and temporal cortex as disease progress onward. Especially, radiality could delineate the difference between cognitive normal and EMCI group while other biomarkers could not. We looked into the relationship between the radiality and other biomarkers to validate its pathological evidence in AD. Overall, radiality showed high association with conventional biomarkers. Additional ROI analysis exhibits dynamics of AD related changes as stages onward.
In conclusion, radiality in cortical gray matter showed AD specific changes and relevance with other conventional AD biomarkers with higher sensitivity. Besides, it could show group differences in early AD changes from EMCI which show advantage for early diagnosis than using conventional biomarkers. We provide the evidence of structure changes with cognitive impairment and suggest radiality as a sensitive biomarker for early AD.

the evidence of structure changes with cognitive impairment and suggest radiality as a sensitive biomarker for early diagnose and progress monitor AD.

Background
Alzheimer's disease (AD) has a long preclinical period where several pathophysiological changes occur before the main symptom. As progress of AD is not completely understood, it makes early diagnosis and intervention difficult [1,2]. Repetitive failures of recent drug trials attribute to applying treatment to patients at the progressed stage [3][4][5]. Thus, identification of people at the earlier stage is critical in clinical trials and may be promising for controlling this devastating disease.
At present, there are several biomarkers to diagnose and monitor disease progression; amyloid and tau deposit through PET imaging or from cerebrospinal fluid (CSF), volumetric and morphology analysis using T1 weighted MR imaging and clinical assessments. The Alzheimer's disease Neuroimaging Initiative (ADNI) proposed two stages-early mild cognitive impairment (EMCI) and late mild cognitive impairment (LMCI)-based on the memory performance [6][7]. Patients with EMCI and LMCI were subdivided solely on the score of the memory scale, but other biomarkers such as the hippocampal volume and CSF biomarkers also showed continuous trajectories suggesting that EMCI could be an earliest stage of AD [8]. Thus, evaluating the differences between EMCI and LMCI might help understanding the early stage disease progression.
Diffusion tensor image (DTI) is sensitized to the motion of water molecules as they interact within tissues, thus reflecting characteristics of their immediate structural surroundings [9,10], thus widely used for studying white matter integrity. Early studies using DTI in AD have focused mostly on white matter using fractional anisotropy. However, since white-matter changes in AD may be the results of Wallerian degeneration, followed by the loss of cortical neurons in gray matter [11,12], the destruction of white matter is a less sensitive change in AD.
The idea of measuring microstructural changes in grey matter using DTI has been demonstrated in both AD and frontotemporal dementia [13][14][15]. These studies showed that gray matter mean diffusivity (MD) is increased in patients compared with healthy control and MD could be a promising imaging biomarker. However, some studies suggested that increased MD could be a spilled over effect from CSF and this effect persisted even with rigorous correction such as partial volume effect correction [16].
To overcome this problem, we used radiality, which is presumably reflecting the integrity of tangential cortical fibers. This parameter has been applied to study neurodevelopment and could distinguish stages of aging [17][18][19]. Moreover, cortical microstructural changes are often observed with aging or neurodegeneration, which can be opposite of neurodevelopment [20][21][22]. Thus, changes of fiber orientation may suggest cortical disruption and feasibility of radiality as a biomarker in the neurodegenerative disorders.
In this study, we hypothesized that the radiality of the gray matter could be a microstructure measure of cortex and used as the early signature of AD. We performed a cross-sectional surface-based cortical analysis approach using DTI, amyloid PET, and Tau PET images to a population of patients with AD continuum. We evaluated whether gray matter radiality shows: i) early mesoscopic changes in regions known to undergo early ADrelated pathological change, and ii) compliment to conventional AD biomarkers while providing a distinct information regarding AD-related pathologies.

Demographics
Data used in preparation of this article were obtained from Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). Among ADNI phase 2 & 3 database, we analyzed the subjects who took both MRI and PET (amyloid, AV 45 and tau, AV 1451); 78 cognitive normal (CN), 51 EMCI, 34 LMCI, and 39 AD. Subjects were sampled with following criteria; age around 60 to 90 year old, education year 12 to 20, and gender match within group. To assess AD continuum, amyloid negative CN and amyloid positive EMCI, LMCI, and AD subjects were selected. A total of 202 subjects' T1 and DTI images were gathered from ADNI. To increase sample size, multi-center approach was used.  Table 1. Note that subjects who underwent AV1451 tau PET imaging were 44 in CN, 9 in EMCI, 5 in LMCI, and 3 in AD. Additional 28 CN subjects who showed amyloid positive were gathered to see earliest AD pathological changes as presented in Supp. Table 1. 2.2 Image processing T1 weighted images were processed with Freesurfer package v6.0 (http://surfer.nmr.mgh.harvard.edu) using procedure as previously reported [13]. Cortical thickness (CTh) maps were registered to Freesurfer average sphere through spherical registration for group comparison. DTI and PET images were registered with their respect to T1 images using boundary-based algorithm for further process. DTI images were processed using FSL package as followed: eddy current correction, rotate gradient vectors from the results of eddy correction, and tensor fitting to produce mean diffusivity map and primary eigenvector map. DTI metrics were further processed to avoid partial volume effect using Koo et al [24]. PET images were partial volume corrected using mri_gtmpvc which is built in Freesurfer package. PET images were normalized by mean signal from whole cerebellum and used as standardized uptake value ratio (SUVR) for amyloid and tau PET, AV45 and AV1451 respectively. Then images were boundary-based registered to respective T1 structural image. To avoid any partial volume effects, the deepest parts of the gray matter were analyzed. Figure 1 shows the overall scheme of the process. Lastly, CTh was smoothed with 10 mm while other modalities were smoothed with 15 mm full width half maximum Gaussian kernel.

Calculation of radiality
A surface normal vector was obtained from individual gray matter surface to define cortical orientation. Vertex-wise dot product between primary eigenvector of diffusion tensor and the surface normal vector was quantified as a radiality index; r: where v represents surface normal vector and e1 represents primary diffusion direction [20].
It ranges from 0 to 1, where r = 0 indicates tangential diffusion and r = 1 indicates radial diffusion to cortex. Subject's principal eigenvector map was projected onto the individual surface reconstruction to calculate vertex-wise radiality as discussed in [20].

Statistical analysis
We first compared the differences between groups for radiality, CTh, MD, AV45 and AV1451 with a two-class general linear model, as implemented in Freesurfer. The results were cluster-wise corrected for family-wise error corrected p-value < 0.05 To assess the relationship between radiality and other neuroimage biomarkers, a vertex by vertex partial correlation was computed between the radiality, CTh, MD, AV45, and AV1451 values. Specifically, a general linear model was created, being radiality the dependent variable of interest, using other biomarkers as the independent variable, and introducing age, gender, and year of education as nuisance variables. 3. Results

Group comparison along AD continuum
We first compared radiality, CTh, and MD differences between groups; CN vs EMCI, CN vs LMCI, CN vs AD as respectively. The results were cluster-wise corrected for family-wise error corrected p-value < 0.05. Figure 2 shows significant group different clusters range from p-value 0.05 to 10 − 5 . Only radiality could delineate the difference from EMCI to CN.
Compared to CN, all groups showed decreased radiality, decreased CTh, and increased MD. There were no group difference of radiality in EMCI and LMCI.

Partial correlation between radiality and other image biomarkers
We then found vertex-wise correlation between image biomarkers and radiality. CTh showed mostly positive correlations that decrease in cortical thickness accompanied with decrease in radiality. MD showed mostly negative correlation that increase in MD accompanied with decrease in radiality. Amyloid and tau levels showed negative correlation with radiality (Fig. 3).

Correlations between image biomarkers and radiality
In order to find progressive changes in radiality as disease progression, AD specific ROIs mask was used to calculate mean biomarker data. Each subject's mean data were scatter plotted and used to calculate Pearson correlation. CTh showed R = 0.641, MD showed R = -0.677, AV45 showed R = -0.490, and AV1451 showed R = -0.412 with radiality (Fig. 4).

Radiality dynamics from AD specific ROIs
To find generative changes in radiality as disease progresses, mean radiality in AD ROIs was calculated for direct comparison among groups. Radiality within AD specific ROIs were plotted in a box and whisker plot. It shows that the radilaity is decreasing linearly with disease progression.
When see the change of radilatiy in each ROI, most of the ROIs showed characteristic of disease progression as decrease in radiality. Significance was tested with one-way ANOVA with P < 0.05, 0.01, 0.001. Insula, middle and superior temporal cortex showed most radiality reduction with disease onward.

Cut-off analysis using Radiality
To further test feasibility of radiality as AD biomarker, we performed cut-off analysis to distinguish CN with other AD stages (Supp.

Discussion
In this study, we tried to identify the early features of EMCI using cortical radiality, which reflects mesoscopic structural changes. By leveraging the radiality in the gray matter, we could detect the changes in EMCI which were not detected by conventional MRI biomarkers. To the best of our knowledge, this is the first study, which applied gray matter radiality in neurodegenerative disease and detected significant mesoscopic changes in EMCI using MRI. In our study, we found progressively larger regions of decreased radiality as disease progresses, starting from medial temporal cortex in EMCI to whole brain in AD. While, CTh or MD of EMCI did not show significant difference with CN.
We investigated the relationship of radiality with other image measures. Association between radiality and CTh showed strong positive correlation on widespread regions of the brain as shown in Fig. 3. It is clear that higher CTh indicate deeper cortical structure and fiber orientation tend to have radial orientation. Cortical depth profile analysis showed that thicker the cortical thickness larger the radiality [25]. In addition, MD showed strong negative correlation on mostly temporal, parietal, and frontal cortices. Radiality may sensitive to CTh but reflecting microstructural feature as well. With AV45 and AV1451, radiality showed association that widely overlapped with both CTh and MD. It should be noted that there were few tau PET images were available. Radiality may also reflect changes due to accumulation of pathologic protein accumulation within the cortex.
Although microstructural changes associated with radiality is unclear, one plausible feature is the disorganization of tangential cortical fibers. It has been reported that the presence of tangential cortical fibers distinguishes stages of neurodevelopment and aging.
There are several events that lead to increase in tangentially oriented fibers including dendritic elaboration [26], formation of local circuits [27], addition of thalamo-cortical fibers [28] and disappearance of radial glia [29,30]. Decrease in radiality may indicate reversal of the events that take place during neurodevelopment. For instance, synaptic loss, neuronal soma changes and neurite disorganization occurs along with the neuronal loss may lead to decrease in radiality. These changes may concurrent with net loss of macromolecule that affect diffusivity, increasing free water in extracellular space.
However, radiality provide evidence of neuronal density that explain concurrent cortical atrophy. Furthermore, accumulation of amyloid or tau proteins may also participate in the disruption of microstructure. Given radiality can delineate the EMCI, we can further speculate that these microstructural changes occur in the earlier stage of AD which are not apparent in macroscopic investigation.
To test the sensitivity of radiality, we sought to find earliest stage of AD. Interestingly, our CN vs EMCI cluster analysis did not show biphasic trajectory as discussed in previous work [31]. Thus, we conducted additional analysis on amyloid negative CN and amyloid positive CN (Supp Fig. 1). We could observe biphasic behavior of CTh and MD where biomarkers showed opposite direction of changes. While CTh showed increased and MD decreased, radiality showed monotonous decrease in amyloid positive CN. This distinct behavior of radiality could characterize the changes in EMCI while CTh and MD could not. Both the CTh increase and MD decrease in early stage of AD was thought to be caused by an amyloidinduced inflammatory response. However, radiality seems to decrease whenever there is microstructural changes in the tissue. From preterm study, occipital cortex showed decrease in radiality as in early development [17][18][19]. At early stage, the extracellular water is hindered orthogonal to the fibers, and thus has a principal direction of diffusivity that augments the radial structure of the intra-cellular compartment.
In short, radiality could discriminate EMCI from CN and exhibited property of reflecting early disease progression. Previous study showed inverse change of CTh and MD in early stage of AD [31]. Such regions of change in radiality were entorhinal cortex, parahippocampal gyrus, middle temporal gyrus, superior and middle frontal gyrus and the supramarginal gyrus bilaterally; a pattern which is similar to that seen in studies of cortical thickness [32]. That is, we could confirm that EMCI may predate LMCI in the perspective of disease progression.
There were several limitations of the current study. First, use of multi-protocol DTI images could influence the observation of progressive changes in MCI. We sought to control age, gender, year of education, and center variate among the group while applying ComBat to minimize the variation between subjects [33]. Second, number of subjects who took AV1451 imaging were not enough to show the relationship with tau pathology. In order to focus on progressive changes, not only showing relationship with AV45 but also with AV1451 is important aspect [34]. However, several subjects in this study underwent screening only once without follow up or only MRI data were available.

Conclusions
In conclusion, we investigated the cortical changes in EMCI using structural MRI and DTI as well as PET imaging markers. Only radiality could delineate the changes in EMCI while cortical thickness and MD could not. In addition, radiality changes in frontal cortex as simultaneously with AV45 in continuum. These results indicate that multimodal approach, atrophy and microstructure, may illuminate early changes in AD. However, further study is needed to support this diffusion orientation change. We also demonstrate that the orientation changes in AD specific ROIs to overcome cluster analysis, indicating possible use of it as a biomarker for AD progression.    insula cortex. Significance was tested with two-sample t-test with * P <0.05, ** P <0.01, *** P <0.001

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download.