Regional White Matter Hyperintensity Influences Grey Matter Atrophy in Mild Cognitive Impairment

Participants

Between April 2013 and April 2016, 165 participants were recruited from tertiary neurology centers in Singapore [34, 35]. Inclusion criteria included diagnosis of MCI based on the NIA-AA criteria [36] or AD based on the NIA-AA criteria [37]. Subjects with MCI were required to have cognitive symptoms, deficits on neuropsychological evaluation, a clinical dementia rating (CDR) score of 0.5, and to not meet criteria for dementia. AD subjects were required to have a CDR of 1. Healthy control subjects were required to have a Mini-Mental State Examination (MMSE) score of 27 or above and a CDR of 0. Exclusion criteria included: 1) a history of alcohol or drug abuse; 2) a current or known history of major depression; 3) comorbid neurodegenerative disease such as Parkinson’s disease; 4) CeVD such as cerebral amyloid angiopathy and prior stroke; 5) neuropsychiatric conditions such as psychosis; 6) presence of contraindications to MRI. Of the 165 participants, 49 of them did not have a FLAIR image. As such, in this study, we only included participants with both T1-weighted and FLAIR images (n = 116). Subjects’ demographics and cognitive characteristics are summarized in Table 1. A total of 116 subjects were included in the study, i.e., healthy controls (n = 51), MCI (n = 35), and AD (n = 30). Subjects were well matched in their demographics, such as age, gender, ethnicity, education, and handedness. This study was approved by the SingHealth institutional ethics review board and written informed consent was obtained for all participants.

Neuropsychological assessments

All patients underwent clinical, neuropsychological, and MRI assessments. Clinical evaluations were also conducted, including comprehensive physical examination by neurologists and medical history from the patient and reliable informant [37]. The neuropsychological battery was performed by trained psychologists who evaluated global cognition as well as specific cognitive domains, namely episodic memory, executive function, attention, visuospatial abilities, and language abilities. Global cognition was assessed using the MMSE [38] and Montreal Cognitive Assessment [39]. Episodic memory was assessed using 10-word delayed recall [40] and WAIS-IV delayed visual reproduction [41]; executive function was assessed using color trails 2 [42] and frontal assessment battery [43]; attention/working memory was assessed using coding [44] and digit span forward tests [42]; language was assessed using verbal fluency and the 30-word Boston naming test [45]; and visuospatial was assessed using the Alzheimer’s Disease Assessment Scale–Cognitive (ADAS-Cog) constructional praxis test [40] and block design test [42]. Performance on the individual tasks was transformed into z-scores based on normative scores [46–51]. A composite summary index for each cognitive domain was then derived from the corresponding averages of the tests.

Image acquisition

MRI scans were performed at Duke-NUS Medical School, Singapore on a 3T Tim Trio System (Siemens, Erlangen, Germany) (n = 54) or a 3T Prisma Fit System (Siemens, Erlangen, Germany) (n = 62) after scanner upgrade. We obtained high-resolution T1-weighted MPRAGE (Magnetization-Prepared Rapid Gradient Echo) sequences (192 continuous sagittal slices, TR/TE/TI = 2300/2.28/ 900 ms, flip angle = 9°, FOV = 256×240 mm², matrix = 256×240, isotropic voxel size = 1.0× 1.0×1.0 mm³, bandwidth = 240 Hz/pixel) and FLAIR (Fluid Attenuated Inversion Recovery) sequences (192 continuous sagittal slices, TR/TE/ TI = 5000/387.0/1800 ms, flip angle = 15°, FOV = 256×256 mm², matrix = 256×256, isotropic voxel size = 1.0×1.0×1.0 mm³) on both scanners using the same parameters. Scan images were reviewed at acquisition and subjects with motion artifacts and gross pathological findings were excluded.

Following procedures reported by Weir and Zuo [52, 53], we quantified the test-retest reliability for MPRAGE on Tim Trio and Prisma Fit in an independent cohort of 23 young healthy controls (11 females, 28.7±6.0 years old, 94.5±24.0 days between scans), and achieved a voxel-correspondent intra-class correlation of 0.96±0.04 for GM, 0.97±0.04 for white matter (WM), and 0.88±0.10 for cerebrospinal fluid (CSF).

White matter hyperintensity derivation

From the FLAIR images, WMH was quantified using an in-house automatic procedure. First, SPM8 (http://www.fil.ion.ucl.ac.uk/spm/) was used to segment GM, WM, and CSF of the T1-weighted images. Combining GM and WM images, we obtained subject-specific brain masks which were used to remove non-brain regions from the FLAIR images. Finally, WMH segmentation and quantification were performed, which involved: 1) modal pixel intensity determined by FLAIR image and 2) threshold-based segmentation using a threshold of 1.45 times the modal pixel intensity. Regional WMH volume for the frontal, parietal, temporal, and occipital lobes were also calculated [54]. Manual visual inspection was performed following the quantification of WMH volume. The total and regional WMH volumes were log transformed to ensure normality.

Volume-based morphometry

We applied the voxel-based morphometry protocol using the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm8/) in Statistical Parametric Mapping (SPM12) (http://www.fil.ion.ucl.ac.uk/spm/). This SPM voxel-based morphometry analysis [55] was used to derive subject-level GM, WM, and CSF masks using SPM tissue probability maps from T1-weighted structural data, similar to our previous approach [56], including 1) segmentation of individual T1-weighted images into GM, WM, and CSF using an adaptive Maximum A Posterior (MAP) technique that does not require a priori tissue probabilities [57]; 2) creating a study-specific template using nonlinear DARTEL registration [58]; 3) normalization of each GM/WM segmented map to the customized template in MNI space; 4) modulation by multiplying voxel values by only the nonlinear components of the Jacobian determinants derived from the spatial normalization step, to account for individual brain sizes; 5) smoothing the normalized GM and WM maps with an 8 mm isotropic Gaussian kernel.

Group differences in GMV: To examine group differences, we built a voxel-wise Analysis of Covariates (ANCOVA) general linear model with GMV as the dependent variable, group membership, and total and lobe-wise WMH as the independent variables-of-interest and scanner type, education and ethnicity as covariates. In a secondary analysis, we further controlled for age, gender, and handedness covariates in addition to ethnicity, education, and scanner covariates. The GM regions which showed significant group differences were examined using a threshold of uncorrected p < 0.001 and a minimum cluster size of 100 voxels [30, 32]. Significant clusters in the GM were anatomically identified using the Automated Anatomical Labelling atlas labels.

Associations between GMV and WMH: We also assessed the differential effects of regional WMH on GMV across groups. For this analysis, we set up contrasts to assess which regions showed associations between GMV and total WMH and lobe-wise WMH in AD, MCI, and control groups separately. Scanner type, education, and ethnicity were included as nuisance covariates. The GM regions that showed significant association with total/regional WMH were reported using a threshold of uncorrected p < 0.001 and a minimum cluster size of 100 voxels.

Statistical analyses

Statistical analyses were performed using Statistical Package for Social Sciences (SPSS) version 24.0 (SPSS, Inc; Chicago, IL, USA). Normality for continuous variables was examined using Skewness-Kurtosis and visual inspection of histogram. One-way analysis of variance (ANOVA) was used to examine the differences between groups and chi-square was used to assess differences between categorical variables. Significance was set at a two-tailed probability value of p < 0.05.

To investigate the associations between GMV, regional WMH, and cognition, we performed stepwise multiple linear regression analysis. The independent variables comprised GMV from regions that showed significant group differences in AD compared to controls, total or regional WMH volume while cognitive scores were the main dependent variable. The tested cognitive measures included global cognition (MMSE, Montreal cognitive assessment) and five domain-specific cognition scores (episodic memory, executive function, attention, language, and visuospatial function). The multiple regression analysis was controlled for age, gender, education, handedness, ethnicity, and scanner type. Multiple comparisons correction was carried out across these seven models with a multiple comparisons corrected p-value of p < 0.0071.

AD subjects show less widespread reductions in temporal GMV after controlling for WMH

Total WMH volume was highest in AD subjects compared to both MCI (p = 0.030) and controls (p = 0.003). Before controlling for total WMH, AD subjects showed significantly reduced GMV compared to healthy controls and MCI in frontal, temporal, and parietal regions: right middle frontal gyrus, right superior temporal gyrus, bilateral middle temporal gyrus, bilateral hippocampus and amygdala, left parahippocampus, and right precuneus (Table 2). After controlling for WMH, AD subjects showed less widespread GM atrophy; temporal regions including the left hippocampus, left amygdala, and right superior and middle temporal gyrus showed reductions in GMV (Table 2 and Fig. 1). MCI subjects did not show any difference in GMV compared to controls.

OPEN IN VIEWER

Fig.1

Group differences in grey matter volume after controlling for total white matter hyperintensity volume. AD subjects showed reduced grey matter volume (purple) in the bilateral hippocampus, middle temporal and superior temporal cortices following correction for total WMH volume. Clusters in blue represent areas of grey matter atrophy in AD subjects prior to controlling for WMH volume. Results are shown at the uncorrected p < 0.001 height threshold with an extent threshold of 100 voxels. Results are displayed on representative sections of the MNI template brain. In coronal and axial slices, the left side of the image corresponds to the left side of the brain. R Mid Temp, right middle temporal cortex; lHIP, left hippocampus.

Total WMH volume has the greatest negative impact on frontal and temporal grey matter in MCI

Total WMH volume had the largest negative influence on GMV in MCI subjects, specifically in temporal and frontal regions (p < 0.001; minimum cluster size = 100 voxels; Fig. 2). We found an overlap between regions atrophied in AD and those showing association between total WMH and reduced GMV in MCI (Supplementary Figure 1).

OPEN IN VIEWER

Fig.2

Impact of total white matter hyperintensity volume on grey matter volume in AD, MCI and healthy controls. Total WMH volume showed an association with grey matter volume in MCI (green) and healthy controls (blue) after controlling for age, gender, handedness, ethnicity, education and scanner covariates. MCI subjects showed the most widespread decreases in grey matter volume across the frontal and temporal lobes while controls showed reduced grey matter volume mainly in frontal and middle temporal regions. Subjects with AD did not show any reduced grey matter volume associated with total WMH. Results are displayed on representative sections of the MNI template brain. The left side of the image corresponds to the left side of the brain. WMH, white matter hyperintensity; AD, Alzheimer’s disease; MCI, mild cognitive impairment; HC, healthy controls.

Frontal and parietal WMH have the largest negative impact on voxel-wise GMV in MCI

Frontal and parietal WMH had the largest negative impact on GMV in MCI (Supplementary Figure 2). Specifically, frontal WMH showed associations with frontal and temporal GMV reductions: bilateral middle frontal gyrus, left precentral gyrus, left superior, middle and inferior temporal gyri, bilateral superior medial frontal gyrus, bilateral parahippocampus, bilateral hippocampus, left insula, and right middle and inferior orbitofrontal gyri (Table 3). After further controlling for age, gender, and handedness, frontal WMH showed associations with frontal and temporal GMV reductions: bilateral middle frontal gyrus, left superior, middle and inferior temporal gyri, bilateral superior medial frontal gyrus, left parahippocampus, left hippocampus (Table 3A, Fig. 3). Furthermore, parietal WMH was associated with frontal and temporal GMV reductions: bilateral superior medial frontal gyrus, bilateral medial and right superior orbitofrontal gyri, right middle frontal gyrus, left precentral gyrus, left inferior and superior temporal gyri, left hippocampus, and left amygdala (Table 3). After further controlling for age, gender, and handedness, parietal WMH was associated with mainly frontal GMV reductions: bilateral superior medial frontal gyrus, right medial orbitofrontal gyri, left precentral and postcentral gyri (Table 3B and Fig. 4).

OPEN IN VIEWER

Fig.3

Impact of frontal lobe white matter hyperintensity volume on grey matter volume in AD, MCI, and healthy controls. Greater frontal WMH related to lower grey matter volume in MCI (green) and healthy controls (blue) after controlling for age, gender, handedness, ethnicity, education, and scanner covariates. MCI subjects showed the most widespread WMH-associated decreases in grey matter volume of the bilateral frontal and temporal lobes while controls also showed WMH-related grey matter volume reductions in the bilateral frontal and right temporal regions. Subjects with AD did not show any reduced grey matter volume associated with total WMH. Results are displayed on representative sections of the MNI template brain. In the axial slices, the left side of the image corresponds to the left side of the brain. WMH, white matter hyperintensity; AD, Alzheimer’s disease; MCI, mild cognitive impairment; HC, healthy controls.

OPEN IN VIEWER

Fig.4

Impact of parietal lobe white matter hyperintensity volume on grey matter volume in AD, MCI and healthy controls. Parietal WMH volume showed an association with grey matter volume in MCI (green) and healthy controls (blue) after controlling for age, gender, handedness, ethnicity, education, and scanner covariates. MCI subjects showed changes in grey matter volume across the bilateral frontal, left temporal regions while controls showed reduced grey matter volume mainly in left precuneus and left middle cingulum regions. Subjects with AD did not show any reduced grey matter volume associated with total WMH. Results are displayed on representative sections of the MNI template brain. In the axial slices, the left side of the image corresponds to the left side of the brain. WMH, white matter hyperintensity; AD, Alzheimer’s disease; MCI, mild cognitive impairment; HC, healthy controls.

In healthy controls, frontal and parietal regional WMH was associated with reductions in frontal, temporal, and parietal GMV. On the one hand, frontal WMH was associated with GMV reductions in bilateral middle and right superior frontal gyri, right superior and middle orbitofrontal gyri, right superior temporal gyrus, and right supramarginal gryus before and after controlling for age, gender, and handedness (Table 3A, Fig. 3). On the other hand, parietal WMH was associated with left supplementary motor area and left superior frontal gyrus atrophy (Table 3B). After further controlling for age, gender, and handedness parietal lobe WMH was associated with GMV reduction in left precuneus and left middle cingulum regions in healthy controls (Table 3B, Fig. 4).

In AD subjects, frontal and parietal regional WMH were associated with right middle and inferior orbitofrontal gyrus GMV reductions but only prior to additionally controlling for age, gender and handedness covariates (Table 3A, B).

Occipital and temporal WMH show smaller influence on voxel-wise GMV in healthy controls and MCI compared to frontal and parietal WMH

Although occipital and temporal WMH showed a negative influence on GMV, the effect of temporal WMH was not as widespread as that of frontal and parietal regional WMH.

Occipital WMH was associated with GMV reductions in right superior and middle temporal gyri, right supramarginal gyrus, right medial orbitofrontal gyrus, and right superior frontal gyrus in healthy controls and remained largely the same after additionally controlling for age, gender and handedness in the model; and bilateral superior medial frontal and orbitofrontal gyri, bilateral middle and inferior triangular frontal gyri, left inferior temporal gyrus, left parahippocampus and hippocampus in MCI (Table 3C). After further controlling for age, gender, and handedness, occipital WMH negatively influenced GMV in bilateral superior medial frontal and orbitofrontal gyri, right middle frontal and inferior triangular frontal gyri, left parahippocampus, and hippocampus in MCI (Table 3C).

Temporal WMH had the largest negative impact on right inferior temporal GMV in healthy controls; and right medial orbitofrontal gyrus, right medial superior frontal gyrus, and right middle and superior frontal gyri in MCI (Table 3D). These results remained after further controlling for age, gender and handedness covariates. There was no impact of occipital and temporal regional WMH on GM structure in AD.

Parietal WMH and hippocampal volume predict global cognition

Of all the features including GMV, regional WMH, and cognition that were included in the step-wise regression model, only parietal lobe WMH was selected as a significant predictor of MMSE scores in MCI (p = 0.025, R² = 0.145) (Fig. 5A) whereas hippocampal GMV significantly predicted MMSE scores in AD (p = 0.003, R² = 0.213) (Fig. 5B). There was no association between lobar WMH and global cognition in AD.

OPEN IN VIEWER

Fig.5

AD and MCI subjects showed distinct associations with global cognition. In MCI subjects only, higher parietal lobe WMH was associated with reduced MMSE scores while in AD subjects only, lower hippocampal grey matter volume was associated with decline in MMSE scores. MMSE, Mini-Mental State Examination; WMH, white matter hyperintensity; GMV, grey matter volume; R, right; AD, Alzheimer’s disease; MCI, mild cognitive impairment.

On the other hand, we found no association between lobar WMH or GMV and domain-specific cognition in MCI. In contrast, AD subjects showed positive associations between episodic memory and hippocampal GMV (p = 0.029, B = 0.449) as well as executive function and precuneus GMV (p = 0.020, B = 0.464). There were no associations between domain-specific cognition and lobe-wise WMH volume in AD.

The influence of cerebrovascular disease on neurodegeneration

Previous studies indicate that WMH volume accounts for 23% of the variance of cortical GM atrophy [59] and has the greatest contribution to AD [60]. Similarly, we found that total WMH volume was highest and associated with reduced GMV in AD. Additionally, corroborating with our findings, negative associations between WMH and GMV have been previously observed in the healthy elderly [20]. More importantly, WMH may be associated pathologically to AD since total WMH volume was associated with reduced GMV in the hippocampus, amygdala, and temporal regions in AD in our study. Eminently, neuropathological evidence does show diffuse and focal WM lesions, including WMH, in AD [61]. Although the direction of this association between WMH and cortical degeneration is under conjecture, our findings showed less extensive atrophy when total WMH was controlled for. It is likely that WMH, a marker of CeVD pathology, has additive effects with AD pathology resulting in greater GMV loss. Indeed, concurrent AD and CeVD pathology in prodromal stages has been shown to indicate accelerated disease progression [62–65]. Additionally, WM disruption within areas of WMH may interrupt axonal projections to cortical regions resulting in cortical neuronal loss by retrograde degeneration [59]. Thus, when GMV is corrected for WMH, the amount of GMV is lower because both GMV and WMH are on the same pathological continuum. Therefore, the presence of WMH may result in more pronounced detrimental effects of underlying AD pathology.

We also observed that regions showing atrophy in AD overlapped with regions where reduced GMV associated with total WMH in MCI and healthy controls including the bilateral hippocampus and right middle temporal cortex (Supplementary Figure 1). This offers some insights regarding the early influence of CeVD such that structures that are affected by WMH in prodromal disease stages could provide information on regions that later show the greatest atrophy in AD or regarding further progression to AD [15, 31]. Our findings highlight the importance of distinguishing subjects with high and low WMH load in order to better ascertain extent and pattern of atrophy in AD.

Region-specific WMH influence GMV in prodromal AD

Previous studies have shown that the majority of WMH occur in the frontal lobe and the parietal lobe, similar to the deposition pattern observed in our study [66]. While some studies show the largest effect of WMH on GM atrophy in AD subjects [10, 59], others show that vascular pathology alongside AD pathology is associated with greater hippocampal atrophy in MCI [14, 67]. Such mixed results could be due to hypothesis-driven region-of-interest approaches where all voxels in a region-of-interest are averaged, possibly causing essential signals, especially at early disease stages, to also be averaged out [68]. Here, using a whole-brain voxel-wise approach, we observed that WMH-related effects on atrophy were more profound in MCI and revealed a novel finding that frontal and parietal WMH had the largest negative influence on GMV compared to temporal and occipital WMH. Whole-brain voxel-wise analysis is more sensitive to alterations in GM regions specific to the different disease course [69]. For instance, in our study, both frontal and parietal lobe WMH volume influenced hippocampal, frontal, and temporal GMV in MCI. Similarly, recent studies exhibit associations between high WMH burden, especially in frontal and temporal lobes, and more severe atrophy in AD-related regions. This may be explained as the influence of both vascular pathology and disruptions in WM pathways that link such distant regions in the brain [33, 70]. One postmortem study has also shown that greater axonal and myelin loss were found in WMH in both AD and healthy controls [66]. In addition, previous studies have shown that parietal lobe WMH is associated with more severe disease progression and predictive of MCI conversion to AD over time [3, 71]. On the other hand, studies also point towards the involvement of neurodegenerative factors such as cortical atrophy as well as tau and amyloid-β deposition in the accumulation of WMH via Wallerian degeneration of anatomically connected WM bundles [72–75]. In this regard, postmortem studies do indicate that parietal lobe WMH in AD are associated with demyelination and axonal loss rather than ischemic damage in aged controls [74]. However, due to the lack of amyloid and tau imaging in our current study, no directionality can be presently inferred regarding the association between cortical atrophy and WMH accumulation. Furthermore, although we did not observe significant differences in GMV between MCI and controls, unlike previous studies [76, 77], we have established patterns of GMV reductions (similar to atrophy patterns in AD) as an effect of region-specific WMH, thus indicating that WMH-related derogatory effects on structure appear to be evident early in disease.

Relationship between WMH volume and cognition

We observed that parietal WMH volume was associated with global cognition as measured by MMSE in MCI subjects only. However, such a relationship between WMH and domain-specific cognition was not observed. Such mixed associations between WMH and cognition have also been observed previously in the healthy elderly and MCI [13, 27]. Indeed, studies do indicate that the presence of WMH can itself cause disruptions in fibers connecting spatially distant cortical areas, in turn leading to greater cognitive impairment in concurrent AD and CeVD pathology [78, 79]. Furthermore, such WMH-cognition associations were missing in AD patients. In contrast, consistent with previous findings, hippocampal volume was related to global cognition in AD [80, 81]. This suggests that region-specific WMH volume may contribute to cognitive impairment in MCI rather than in the dementia stage, where atrophy might play a greater role.

Strengths and limitations

The strengths of this study include comprehensive assessments of structural brain measures and neuropsychological assessments and voxel-wise GMV analysis that allows detection of subtle changes in brain volume. Additionally, we assessed the influence of regional WMH on both brain structure and cognition to help provide greater insights into the influence of CeVD on the brain at the healthy control, MCI, and AD stages. However, our study was limited by the cross-sectional study design which prevented us from investigating the effect of WMH on the progression of atrophy patterns. Given the moderate sample size, our voxel-wise GMV findings did not survive corrections for multiple comparisons though stringent height and cluster-level thresholding were performed to minimize false positives. Moreover, we did not have information regarding the amyloid status of our subjects, thus we were unable to assess the influence of such AD risk factors on the relationship between WMH and brain structure. In addition, the MCI group had a mixture of amnestic and non-amnestic MCI subjects; such heterogeneity might confound the relationship between WMH, GMV and cognition. Thus, future work will involve understanding the interaction between AD risk factors, WMH and GMV and age-dependent associations with cognition in prodromal and clinical AD [82].