The effect of white matter signal abnormalities on default mode network connectivity in mild cognitive impairment

Abstract Regions within the default mode network (DMN) are particularly vulnerable to Alzheimer's disease pathology and mechanisms of DMN disruption in mild cognitive impairment (MCI) are still unclear. White matter lesions are presumed to be mechanistically linked to vascular dysfunction whereas cortical atrophy may be related to neurodegeneration. We examined associations between DMN seed‐based connectivity, white matter lesion load, and cortical atrophy in MCI and cognitively healthy controls. MCI showed decreased functional connectivity (FC) between the precuneus‐seed and bilateral lateral temporal cortex (LTC), medial prefrontal cortex (mPFC), posterior cingulate cortex, and inferior parietal lobe compared to those with controls. When controlling for white matter lesion volume, DMN connectivity differences between groups were diminished within bilateral LTC, although were significantly increased in the mPFC explained by significant regional associations between white matter lesion volume and DMN connectivity only in the MCI group. When controlling for cortical thickness, DMN FC was similarly decreased across both groups. These findings suggest that white matter lesions and cortical atrophy are differentially associated with alterations in FC patterns in MCI. Associations between white matter lesions and DMN connectivity in MCI further support at least a partial but important vascular contribution to age‐associated neural and cognitive impairment.

White matter damage is common in older adults and in most cases, this represents tissue compromise of a presumed vascular origin that is often visualized on MRI as white matter signal abnormalities (WMSA; Canu et al., 2012;Carmichael et al., 2010;Maillard et al., 2012). Although the clinical manifestations of WMSA are often associated with vascular conditions such as small cerebrovascular disease (Kalheim, Bjornerud, Fladby, Vegge, & Selnes, 2017;Wirth et al., 2017), recent work has also suggested a mechanistic role of WMSA in the development of MCI and AD (Brickman et al., 2012;Coutu et al., 2016;. Notably, WMSA burden has been shown to precipitate the clinical manifestation of MCI (Luchsinger et al., 2009;Provenzano et al., 2013), as well as contribute to the conversion from MCI to AD (Lindemer et al., 2015). Increasing WMSA burden within DMN regions has been shown to contribute to frontal-subcortical pathway disruption and associated cognitive impairment (Habes et al., 2016;Luchsinger et al., 2009;Pugh & Lipsitz, 2002), and decreased functional connectivity in MCI is further associated with reduced white matter structural connectivity (Zhou et al., 2008).
On the other hand, cortical thinning is considered a marker of neurodegenerative change in AD that is linked to amyloid/tau pathological burden independent of vascular conditions and white matter lesion burden (Coutu et al., 2017;Wirth et al., 2017;Zheng et al., 2016). Cortical atrophy patterns have been shown to distinguish MCI from normal aging, where thinning is first observed in medial temporal regions, and then spreading to association areas within medial parietal, lateral temporal, and frontal regions in tandem with AD pathological progression (Driscoll et al., 2009;McEvoy et al., 2009McEvoy et al., , 2011Wirth et al., 2017). Importantly, these cortical atrophy patterns in MCI largely overlap with key DMN regions, with particular emphasis on temporal cortex which is shown to be the most vulnerable to cortical thinning in MCI (Driscoll et al., 2009;Fjell et al., 2009;Pfefferbaum et al., 2013;Raz, Ghisletta, Rodrigue, Kennedy, & Lindenberger, 2010). Thus, abnormal DMN connectivity among individuals with MCI may be related to cortical thinning found in regions most vulnerable to AD pathology (Dickerson et al., 2004(Dickerson et al., , 2009). However, it is unclear whether the effects of white matter lesions on DMN connectivity are distinct from those of cortical atrophy among those with MCI.
The present study aimed to examine alterations in seed-based DMN functional connectivity between individuals with neuropsychologically defined MCI and controls, and whether functional alterations were associated with two distinct measures of brain structural integrity that may be impacted by differing presumed pathological origins: white matter lesions and cortical thickness.

| Participants
A total of 37 older adults aged 60-80 years were enrolled in this study conducted within the Brain Aging and Dementia (BAnD) Laboratory at Massachusetts General Hospital (MGH). Participants were referred for the study through the MGH Alzheimer's Disease Research Center (MGH ADRC), enrolled from a local longitudinal cohort, or through community outreach. All participants were generally healthy and free of atypical disease processes for their age, with exclusion criteria consisting of: a history of heart disease, stroke, or kidney disease, the presence of acute or chronic major neurological conditions, uncontrolled diabetes (based on glucose levels <200 mg/dL when medicated), current use of psychoactive drugs, and contraindications to the MRI environment. All participants were fully independent in activities of daily living-and thus above clinical diagnostic thresholds for major neurocognitive disorder. This study was approved by Partners Healthcare institutional review board (IRB) and was in accordance with the Declaration of Helsinki. Informed consent was obtained from each participant. Test. Each participant's raw neuropsychological data was first converted to standard scores based on population-based norms published for each test, and then transformed to z-scores. The MCI group was operationally defined using previously published criteria that classifies MCI by a noted impairment in two or more cognitive domains, with domain-level impairment defined by two or more tests within that domain falling at least one SD below published normative values (Bondi et al., 2008(Bondi et al., , 2014Jak et al., 2009;Stricker et al., 2013). Based on this operationalization, 20 participants were determined to be cognitively intact and formed our control group (CON), and 17 participants met criteria for MCI-all of which had noted impairment in the memory domain, suggesting an amnestic subtype of MCI.

| MRI acquisition
All structural and functional MRI data was acquired on a 3-Tesla Siemens Trio scanner (Erlangen, Germany) with a 12-channel phasedarray head coil. High-resolution T1-weighted images were obtained 2.4 | T1-weighted high-resolution structural imaging T1-weighted anatomical images were automatically processed to reconstruct cortical surfaces and to segment volume region-of-interests (ROIs) using the standard FreeSurfer processing stream (http://surfer.nmr.mgh. harvard.edu/; Fischl, Sereno, Tootell, & Dale, 1999). The technical details of these procedures are described in prior publications (Fischl et al., 2002;Fischl, Liu, & Dale, 2001) but briefly, this process entails removal of nonbrain tissue, automated Tailarach transformation, gray/white tissue segmentation, intensity normalization, parcellation of the gray/white matter boundaries, topology correction, surface deformation, and parcellation of the cortical surface based on gyral and sulcal structures. Cortical thickness was defined as the shortest distance between vertices comprising the inner surface (gray/white boundary) and outer surface (gray/pial boundary; Fischl & Dale, 2000). All T1-weighted images were registered onto a common surface template using a surfacebased averaging technique reliant on cortical folding patterns. All raw imaging data were inspected for artifacts and accuracy of surface boundary placement (which were manually corrected if needed) prior to F I G U R E 1 The Yeo 7-network parcellation atlas was created using independent component analysis from over 1,000 young healthy adults and is paired with a corresponding confidence map providing an estimate for each vertex across the cortical mantle as belonging to each of seven distinct rsfMRI networks. To create each DMN seed region in standard space (using the FSaverage template), the surface-based parcellation of the precuneus was used as a mask region to constrain the search for the vertex with the peak confidence value of belonging to the DMN within the Yeo 7Network Confidence map. Next, for each hemisphere separately, the selected precuneus vertex with the highest DMN confidence rating was dilated by a factor of 10 on the cortical surface resulting in a circular seed region, and then warped to each participant's native fMRI space to extract the seed time course for each hemisphere separately analysis. Total WMSA volumes were quantified for each hemisphere using T1/T2 image data based on previously described procedure (Lindemer et al., 2015Lindemer, Greve, Fischl, Salat, & Gomez-Isla, 2018).
Briefly, this procedure registers T2-weighted images to an individual's T1-weighted image after it has been processed through FreeSurfer's recon-all stream. It then performs intensity normalization of modalities using a multimodal atlas and segments WMSA from normal-appearing white matter (NAWM) using a spatial array of multimodal Gaussian classifiers as well as individual-based heuristics . WMSAs were defined based on consensus guidelines for measurement of WM hyperintensities of presumed vascular origin (Wardlaw, Smith, Biessels, et al., 2013).
Using these new WMSA labels in conjunction with all standard FreeSurfer labels, a multimodal Gaussian classifier array (MMGCA) was created that contained a matrix (T1 and T2) for each structure at each voxel in addition to spatial and neighborhood prior information. Then follow up the MMGCA procedure with several refinements designed to catch unlabeled WMSAs. These refinements rely heavily on the Mahalanobis distance (MD) of a WMSA voxel from normal-appearing white matter (Lindemer et al., 2015).

| Resting state fMRI processing
Resting state function MRI data were processed using FreeSurfer   Because our MCI group is of the amnestic subtype with a high rate of conversion to AD, we chose the precuneus as our seed region for functional connectivity analysis given that this brain area is a key anatomical hub of the DMN, and is presumed to play an important role in episodic memory, a cognitive function invariably disrupted in Alzheimer's disease (Weiler et al., 2014). As visualized in Figure 1 Resultant surface-based correlation coefficient maps reflecting precuneusseed connectivity across the cortex were concatenated across subjects in standard space and entered into general linear models as described below.

| Statistical analysis
All demographic and clinical data were assessed for normality and the presence of outliers prior to statistical analysis using IBM SPSS Statistics version 24. Surface-based neuroimaging data was analyzed using  Table S1). Vertex-wise statistical results were thresholded at p < .05 and corrected for multiple comparisons using a cluster-based procedure adapted for cortical surface analysis with a cluster-wise alpha level of p < .05. The data that support the F I G U R E 2 Results of whole-brain vertex-wise cortical thickness analysis between healthy older adult control (CON) and mild cognitive impairment (MCI) groups. Cool colors correspond to regions where the MCI had thinner cortex compared to the CON group. Warm colors correspond to the opposite contrast. Data were thresholded at p < .05, with a saturation of p < .001, green outlines indicate regions that survived after multiple comparison correction, with a cluster-wise statistical threshold set to p < .05 findings of this study are available from the corresponding author upon reasonable request.

| Participants
Group characteristics are summarized in Table 1. As expected, those with MCI scored significantly lower on the MMSE (p < .05) and MoCA (p < .001) compared to controls. The two groups did not significantly differ in regard to age, years of education, or sex distribution (all p > .05). The group with MCI also demonstrated significantly greater total WMSA volume compared to controls (p < .05). Because cortical thickness was entered into models as a per-vertex regressor, group differences in cortical thickness are demonstrated as surface models in Figure 2. Overall, individuals with MCI had significantly reduced cortical thickness relative to the CON group in bilateral lateral and medial temporal areas, visual cortex, primary somatosensory, and pCC F I G U R E 3 One-sample group mean (OSGM) functional connectivity maps are shown for (1) cognitively healthy controls (CON): (a) without any regressors, (b) regressing out the effects of white matter signal abnormalities (WMSA) volume only, (c) regressing out cortical thickness (CTH) only, and (d) regressing out both WMSA and CTH.
(2) Mild cognitive impairment (MCI) group: (e) without any regressors, (f) regressing out WMSA volume only, (g) regressing out cortical thickness (CTH) only, and (h) regressing out both WMSA and CTH. For the within-group analyses (a-h), warm colors indicate significant regional positive correlations with the precuneus-seed, and cool colors indicate regions of anticorrelation. Between-group analyses comparing precuneus-seeded DMN functional connectivity maps between MCI and CON groups: (i) without any regressors, (j) regressing out WMSA only, (k) regressing out CTH only, and (l) regressing out both WMSA and CTH. Warm colors indicate regions of stronger DMN connectivity in CON compared to MCI, whereas cool colors indicate regions where the MCI group showed stronger DMN connectivity compared to the CON group. For all analyses shown, statistical thresholds were set to p < .05, saturation to p < .001, with green outlines indicating regions that survived after multiple comparison correction using a cluster-wise statistical threshold set to p < .05 regions, as well as the left mPFC. The result kept stable in bilateral visual cortex and right medial temporal areas after multiple comparisons.

| DMN connectivity
As shown in Figure Figure 3. To assess the degree to which regional differences in precuneus-seeded functional connectivity between groups (HC vs. MCI) associated with neuropsychological performance, we performed follow-up analyses within clusters that showed a significant difference in functional connectivity between groups that survived multiple com-

| DISCUSSION
The present study confirms prior findings of reduced functional connectivity in disparate regions of the DMN network in MCI relative to cognitively healthy controls Chhatwal & Sperling, 2012;De Vogelaere, Santens, Achten, Boon, & Vingerhoets, 2012;Jin, Pelak, & Cordes, 2012;Li et al., 2016;Sorg et al., 2007;Weiler et al., 2014), while extending this line of work to demonstrate that regional connectivity patterns differed both within and between groups as a function of WMSA volume and cortical thickness. Both structural measures differentially influenced group differences in functional connectivity between the precuneus-seed region and the LTC, IPL, and pCC, with WMSA exhibiting the strongest effect on DMN connectivity within mPFC regions. These findings support multiple pathologic processes contributing to network dysfunction in MCI, although the preferential effects of white matter lesion volume on medial prefrontal connectivity patterns suggests a possible vascular etiology.
To determine the relative contributions of white matter lesion burden and cortical thickness to precuneus-seeded functional connectivity profiles, these metrics of interest were systematically added as covariates to both within-and between-subject models. When WMSA volume was included as a regressor, we observed additional between-group differences in functional connectivity metrics within the left mPFC (see Figure 3). When functional connectivity maps were derived for each group separately, the addition of WMSA as a regressor minimally altered the functional connectivity profile in the control group, although drastically decreased mPFC functional connectivity strength in those with MCI. Greater white matter lesion burden has been previously associated with impaired resting state brain connectivity within medial frontal regions in those with MCI (Zhou et al., 2015), and tract-specific white matter damage subserving then mPFC has been related to functional connectivity disruption in patients across the continuum of AD (Taylor et al., 2017;Tullberg et al., 2004).
Although WMSA are not specific to AD pathology, our prior work has shown that the regional distribution of WMSA lesions may be a critical component of AD development and progression given that regions showing a statistically significant relationship between WMSA and time-to-AD-conversion are limited to the temporal and the frontal white matter . Given that the accumulation of white matter lesions are often attributed to the ischemic vulnerability of sparsely perfused vascular end zones of perforating arteries with limited collateral blood supply (Holland et al., 2008), it is likely that the observed influence of WMSA volume on DMN connectivity in MCI may be due to a presumed vascular origin (Braak & Braak, 1991 Biessels, et al., 2013;. White matter lesion distribution tends to spatially overlap with hubs of the DMN, presumably contributing to frontal-subcortical pathway disruption and associated cognitive impairment (Habes et al., 2016;Luchsinger et al., 2009;Pugh & Lipsitz, 2002). Frontal brain areas are particularly vulnerability to vascular insult and the accumulation of WMSA burden in frontal regions is more closely tied to cognitive functioning compared to local patterns of cortical atrophy (Tullberg et al., 2004).  (Fjell et al., 2014;Greicius et al., 2004;Jones et al., 2011;Walhovd et al., 2010). One possible explanation for this similar decline in functional connectivity across both groups after regressing cortical thickness is that regions within the DMN tend to consist of higher-level association areas that are most vulnerable to age-related decline and atrophy (Fjell et al., 2014;Lustig et al., 2003), such that age-related cortical thinning likely exerts a more global impact on DMN functional connectivity compared to the regional specificity of WMSA accumulation. Our study is limited by its cross-sectional design, which precludes inference that the functional connectivity changes we found would necessarily predict progression toward AD. Furthermore, MCI is a heterogeneous condition that likely reflects a range of underlying pathology aside from AD and it is possible that some subjects in our cohort had cognitive impairments due to other etiologies.

CONFLICT OF INTEREST
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.