Spatial Similarity of MRI-Visible Perivascular Spaces in Healthy Young Adult Twins


 Purpose

This study aims to determine whether genetic factors affect the location of dilated perivascular spaces (dPVS) by comparing healthy young twins and non-twin (NT) siblings.
Methods

A total of 700 healthy young adult twins and NT siblings (138 monozygotic (MZ) twin pairs, 79 dizygotic (DZ) twin pairs, and 133 NT sibling pairs) were collected from the Human Connectome Project dataset. dPVS was automatically segmented and normalized to standard space. Then, spatial similarity indices (mean squared error [MSE], structural similarity [SSIM], and dice similarity [DS]) were calculated for dPVS in the basal ganglia (BGdPVS) and white matter (WMdPVS) between paired subjects before and after propensity score matching of dPVS volumes between groups. Within-pair correlations for the regional volumes of dVPS were also assessed using the intraclass correlation coefficient (ICC).
Results

The spatial similarity of dPVS was significantly higher in MZ twins (higher DS [median, 0.382 and 0.310] and SSIM [0.963 and 0.887] and lower MSE [0.005 and 0.005] for BGdPVS and WMdPVS, respectively) than DZ twins (DS [0.121 and 0.119], SSIM [0.941 and 0.868], and MSE [0.010 and 0.011]) and NT siblings (DS [0.106 and 0.097], SSIM [0.924 and 0.848], and MSE [0.016 and 0.017]). No significant difference was found between DZ twins and NT siblings. Similar results were found even after subjects were matched according to dPVS volume. Regional dPVS volumes were also more correlated within pairs in MZ twins than DZ twins and NT siblings.
Conclusion

Our results suggest that genetic factors affect the location of dPVS.


Introduction
Cerebral perivascular spaces (PVS) are cerebrospinal uid (CSF)-lled structures that wrap around the arteries, arterioles, veins, and venules as the vessels enter and exit the brain. After their structure was rst described by Virchow and Robin, much research was done to further understand their anatomy, function and dysfunction 1 . Recently, PVS were found to belong to the glymphatic system, a recently discovered macroscopic waste clearance system, and to function as part of waste cleaning, energy substrate delivery and blood ow regulation 2 as well as act as an early imaging marker for cerebral small vessel disease [3][4][5] . PVS may be dilated and visible on MRI under certain situations, such as aging, cerebral small vessel disease 3 , hypertension 6 , intracranial hemorrhage 7 , multiple sclerosis 8,9 , Alzheimer's disease 10 , and Parkinson disease. 11 The exact mechanism of this dilatation of PVS is still unclear, but probably in uenced by both genetic and environmental factors. 12 . A study with a large elderly population found that dilated PVS (dPVS) burden assessed with a 4-point visual scale was highly heritable 13 . Another recent study with a healthy young population replicated this result by showing that dPVS volume measured using an automated segmentation method was highly heritable as well 14 . Both studies consistently found dPVS in white matter (WMdPVS) to be a more heritable trait than dPVS in the basal ganglia (BGdPVS), suggesting a distinct genetic impact on dPVS burden according to regional location. However, we do not yet know if the location of each dPVS itself is also genetically determined.
Therefore, in this study, we aimed to evaluate whether dPVS location is in uenced by genetic factors by comparing spatial similarity indices in healthy young adult twins and non-twin (NT) siblings.

Materials And Methods
This retrospective study was approved by our institutional review board (The Catholic University of Korea Seoul St. Mary's Hospital). The requirement for informed consent was waived as we used a publicly available dataset for this study. The methods and reporting of results are in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines.

Study Population
dPVS were segmented using structural images and demographic information obtained from the WU-Minn Human Connectome Project dataset which enrolled healthy adult twins and NT siblings between the ages of 22 to 35 to identify relationships between brain circuits, genetics and behavior 15 .
Of 1206 subjects included in the March 2017 (S1200) release, 1113 subjects who underwent 3T MRI to obtain 0.7 mm isotropic 3D T1-and T2-weighted images were initially enrolled. The exclusion criteria were as follows: high blood pressure, diabetes mellitus or signi cant cardiovascular disease; severe neurodevelopmental, neurological or documented neuropsychiatric disorders; zygosity not examined by genotyping; NT siblings without respective pairs; and birth before 34th weeks of gestation for twins and before 37 weeks of gestation for non-twins. More information on recruitment and the inclusion and exclusion criteria of the Human Connectome Project was described in a previous study 15

Imaging acquisition
All MRI data were acquired using a 3T MR scanner (MAGNETOM Skyra CONNECTOM, Siemens Healthcare) customized with a 100mT/m gradient coil, inner bore diameter of 56 cm, and a standard 32channel head coil at Washington University in St. Louis, MO, USA.
More details on the imaging protocols are described in the WU-Minn Human Connectome Project S1200 Release Reference Manual 15

Spatial similarity assessment
The segmentation of dPVS was a fully automated process that was described in detail in a previous study 14 . It entailed the extraction of potential voxels for dPVS using a 3D Frangi lter after signal normalization of 3D T2-weighted images. To reduce false positives outside the brain parenchyma, potential dPVS voxels only inside the BG and WM masks of the Freesurfer segmentation were selected. In addition, we trained and applied a 3D deep convolutional neural network to distinguish dPVS from the false-positive voxels. Based on the nal output of the 3D deep learning algorithm, dPVS masks for each BG and WM were obtained.
To compare dPVS locations between pairs, we assessed the similarity of their dPVS images. T1-weighted images on which dPVS had been de ned were used to rst estimate a deformation eld as it was needed for the spatial transformation between each subject's brain images and the template brain images in SPM12 (https://www. l.ion.ucl.ac.uk/spm/software/spm12/). The deformation eld was then used to transform dPVS images to standard space, and the locations of dPVS in the standard space were compared using the following three similarity indices: mean squared error (MSE), structural similarity (SSIM), and dice similarity (DS). MSE and SSIM have been proposed as metrics for image quality assessment in prior studies. While MSE is a metric simply computed by averaging the squared intensity differences between two images, SSIM is a metric for comparing local patterns of intensities 16 . DS is a metric for gauging the similarity of two sets based on their cardinalities 17 , and can be applied to two binary images to assess the commonality between them with values ranging between 0 and 1. We compared the locations of dPVS separately for BG and WM.

Cognitive function assessment
The well-validated NIH Toolbox Cognition Battery was used to assess cognitive function 18

Statistical Analysis
After normality tests were performed, age, brain regional and dPVS volumes, and all similarity indices were compared between the three groups using the Kruskal-Wallis test and then the post-hoc Dunn's test with Bonferroni adjustment. The frequency of sex was compared between the groups using the chisquared test with Bonferroni adjustment.
As spatial similarity indices could be affected by the total volume of dPVS, we used propensity score matching separately for volumes of BGdPVS and WMdPVS to balance this confounding factor between groups using the nearest matching method with a 1:1:1 ratio 19 The spatial similarity indices were also compared between groups in the matched subjects.
To de ne genetic in uence on the regional location of dPVS, we rst divided WM into four (i.e., frontal, parietal, temporal, and occipital) lobar subregions using the Freesurfer results available in the Human Connectome Project dataset. Then, we performed an intraclass correlation (ICC) analysis within twin or NT pairs for dPVS volumes in each of the BG and WM subregions. The ICC for twin data was calculated as: where MS between and MS within are the mean-square estimate of between-and within-pair variance, respectively 20 .
To assess the clinical implications of dPVS according to location, a correlation analysis was performed between regional dPVS burden and the cognitive function test results.
A P value of <0.05 was considered statistically signi cant. All statistical analyses were performed using R Statistical Software (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria).

Demographics characteristics of the study population
In this study, a total of 700 subjects including 138 monozygotic (MZ) twin pairs, 79 dizygotic (DZ) twin pairs, and 133 NT sibling pairs were nally included. The baseline characteristics of all participants are summarized in Table 1. Both MZ and DZ twins were older (mean age, 29 years old vs. 27 years old, P< 0.001) and consisted of more male participants (58.7% and 62% vs. 46.2%, P=0.015 and P=0.007, respectively) than NT siblings. Group Comparisons Of Spatial Similarity Indices In All Subjects All spatial similarity indices were different among the three groups for both BGdPVS and WMdPVS ( There was no signi cant difference in spatial similarity indices between the DZ and NT groups for both BGdPVS and WMdPVS. There were more overlapping dPVS voxels between paired subjects across all subjects in MZ twins than DZ twins and NT siblings (Fig. 1). Group comparisons of spatial similarity indices in subjects matched for dPVS volumes Higher volumes of dPVS might result in spuriously higher spatial similarity due to the overlapping of adjacent dPVSs. Therefore, to avoid this possibility, we matched dPVS volumes between the groups for BGdPVS and WMdPVS separately. 72 subjects were selected from each group for BGdPVS and 51 subjects were selected from each group for WMdPVS. Spatial similarity indices still indicated that dPVS were most similarly located within the MZ twin pairs for both BGdPVS and WMdPVS, although some of the similarity indices no longer showed statistical signi cance (Table 3). Within-pair Correlations For The Regional Volumes Of Dpvs MZ twin pairs showed higher ICCs for dPVS volumes than DZ twin pairs and NT sibling pairs regardless of the location of dPVS. In all regions, no signi cant difference was found between DZ twin pairs and NT sibling pairs (Table 4).  Data are expressed as intraclass correlation coe cients with 95% con dence intervals in parentheses.
All correlations were signi cant with P < 0.05.
Correlation analysis between regional volumes of dPVS and cognitive function There were no signi cant correlations between regional dPVS volumes and cognitive function test results (Table S1).

Discussion
This study evaluated whether the location of dPVS was affected by genetic factors using high quality images of healthy young twins and NT siblings from a large dataset. We demonstrated that the location of individual dPVS was most similar within pair members in MZ twins. The within-pair spatial similarity between DZ twins and NT siblings showed no signi cant difference. After matching dPVS volume, MZ twin pairs still showed the highest spatial similarity, although statistical signi cance did decrease. The regional volumes of dPVS were more highly correlated in MZ twin pairs than DZ twin pairs and NT sibling pairs, while the within-pair correlations of regional dPVS volumes were not signi cantly different between DZ twin pairs and NT sibling pairs. Therefore, our results suggest that genetics in uence the location of dPVS.
Hitherto, a few studies have explored the heritability of dPVS burden in large populations. Duperron et al.
found that dPVS burden was highly heritable in an elderly population 7 . Choi et al. replicated this result but in a healthy young adult population 14 . These two studies consistently reported WMdPVS as being more heritable than BGdPVS, suggesting that genetic contributions to dPVS burden differed by location.
However, research on whether genetics affect the location of dPVS, instead of its burden, is lacking. A previous study assessed the burden of WM hyperintensities (WMH), which are a well-known imaging marker for cerebral small vessel disease [3][4][5] . The study reported WMH burden as being highly heritable for each of the cerebral lobes as well as for the whole brain 21 , suggesting that location as well as overall burden of WMH might be affected by genetic factors. Cerebral blood ow per each vascular territory, which is closely related with small vessel disease 22,23 , was also reported to be affected by genetic factors 24 . Therefore, although further study is needed, we can postulate that the location of dPVS as well as its overall burden might be genetically determined to some extent.  [29][30][31] , recurrent ischemic 32 and hemorrhagic stroke 33 , and progressive deterioration of cognitive and motor symptoms in Parkinson's disease 11,34 according to the location of PVS. Besides the distinction between BGdPVS and WMdPVS, decreased neuronal and axonal densities with reactive gliosis adjacent to dPVS 35 and increased CSF markers of neurodegeneration associated with higher burden of dPVS in past studies suggest that a speci c location of dPVS might cause differential clinical outcomes just as the lobar location of WMH differentially affects cognition and behavior 36,37 . However, in our study, we failed to nd any signi cant associations between regional dPVS volumes and speci c cognitive functions. This might be because our study population was made up of healthy young adults whose normal range of cognition is not broad enough to su ciently show any signi cant correlation with dPVS volume. Another possible explanation might be that dPVS is an early imaging marker for small vessel disease or neurodegeneration, so, clinical manifestations of dPVS might not develop by the date of the MRI. To elucidate differential clinical outcomes per dPVS location, a longitudinal study on a large population with a wide range of cognitive and behavioral characteristics should be conducted.
There were several limitations to this study. First, this study was performed retrospectively with publicly available data. Therefore, longitudinal follow-up data was not available for the study population, which limits our understanding of the differential clinical implications of dPVS according to location. Further longitudinal study is required. Second, the similarity measurement methods of this study are not yet regularly used to compare medical images between different subjects and thus, the reliability of the similarity indices has not yet been su ciently demonstrated. Third, although propensity score matching can help balance the volume differences between the three groups, biases caused by yet unknown factors might still remain.  Figure 1 Likelihood map of overlapping dPVS voxels between paired subjects across all subjects for each group

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryTable.docx