White matter hypointensities and hyperintensities have equivalent correlations with age and CSF β‐amyloid in the nondemented elderly

Abstract Introduction T1‐ and T2‐weighted sequences from MRI often provide useful complementary information about tissue properties. Leukoaraiosis results in signal abnormalities on T1‐weighted images, which are automatically quantified by FreeSurfer, but this marker is poorly characterized and is rarely used. We evaluated associations between white matter hyperintensity (WM‐hyper) volume from FLAIR and white matter hypointensity (WM‐hypo) volume from T1‐weighted images and compared their associations with age and cerebrospinal fluid (CSF) β‐amyloid and tau. Methods A total of 56 nondemented participants (68–94 years) were recruited and gave informed consent. All participants went through MR imaging on a GE 1.5T scanner and of these 47 underwent lumbar puncture for CSF analysis. WM‐hypo was calculated using FreeSurfer analysis of T1 FSPGR 3D, and WM‐hyper was calculated with the Lesion Segmentation Toolbox in the SPM software package using T2‐FLAIR. Results WM‐hyper and WM‐hypo were strongly correlated (r = .81; parameter estimate (p.e.): 1.53 ± 0.15; p < .0001). Age was significantly associated with both WM‐hyper (r = .31, p.e. 0.078 ± 0.030, p = .013) and WM‐hypo (r = .42, p.e. 0.055 ± 0.015, p < .001). CSF β‐amyloid levels were predicted by WM‐hyper (r = .33, p.e. −0.11 ± 0.044, p = .013) and WM‐hypo (r = .42, p.e. −0.24 ± 0.073, p = .002). CSF tau levels were not correlated with either WM‐hyper (p = .9) or WM‐hypo (p = .99). Conclusions Strong correlations between WM‐hyper and WM‐hypo, and similar associations with age, abnormal β‐amyloid, and tau suggest a general equivalence between these two imaging markers. Our work supports the equivalence of white matter hypointensity volumes derived from FreeSurfer for evaluating leukoaraiosis. This may have particular utility when T2‐FLAIR is low in quality or absent, enabling analysis of older imaging data sets.


| INTRODUC TI ON
Chronic vascular insult is an important biomarker for cognitive impairment (Gorelick et al., 2011) where damage accumulates silently for decades before the onset of clinically identifiable dementia symptoms (Brown & Thore, 2011). Early detection of microvascular disease is therefore critical to identify and guide efforts to prevent the onset of dementia. Neuroimaging is an essential tool in the evaluation of age-related accumulation of small vessel disease that contributes to vascular insult in the brain, which shows up as white matter lesions. The term "leukoaraiosis" was used to describe such phenomenon by Hachinski et al. to these white matter lesions as a descriptive placeholder to prevent "premature" assumption of pathology and to "encourage the search for causes" (Hachinski, Potter, & Merskey, 1986); the Greek root leuko-means white and -araiosis is a rarefication, meaning a reduced density. On computed tomography (CT) and T1-weighted magnetic resonance imaging (MRI), white matter lesions appear dark and have been termed white matter hypointensities (WM-hypo), while on fluid-sensitive MRI sequences such as fluid-attenuated inversion recovery (FLAIR), they appear bright and are referred to as white matter hyperintensities (WMhyper) as shown in Figure 1. Most research has utilized FLAIR in particular because of the increased conspicuity of WM lesions after the suppression of signal from the cerebrospinal fluid. Previously, the presence of abnormalities on spin-echo T1-weighted sequences which had lower sensitivity for lesion detection was used as a marker of more severe insult (Sinnecker et al., 2012). Continued sequence development in MRI led to the advent of ultrafast gradient-echo sequences, such as T1 magnetization prepared rapid acquisition gradient echo (MPRAGE), which increased conspicuity of WM-hypo on T1 (Mistry et al., 2011). Currently, there is limited understanding of similarities or differences between WM-hypo derived from modern 3D gradient-echo sequences and WM-hyper derived from T2-FLAIR (Dadar et al., 2018).
Assessment of T1 and T2 values may provide additional specificity about the underlying pathology and severity of white matter lesions.
For example, during development, the myelinating white matter undergoes temporally distinct phases of signal intensity changes on T1and T2-weighted images (Ashikaga, Araki, Ono, Nishimura, & Ishida, 1999), reflecting the complex microstructural origin of MRI signals.
Early in development, myelination results in an increased signal on T1 approaching the adult state, while the signal on FLAIR remains bright, reflecting persistent elevations in free water (Holland, Haas, Norman, Brant-Zawadzki, & Newton, 1986). Changes on T1 and FLAIR with leukoaraiosis may also represent distinct, though related, changes in the white matter as is seen during development. Thus, WM-hyper and WM-hypo volumes might be closely despite actually measuring something different (Erkinjuntti et al., 1987). Though age-related WM lesions are commonly attributed to chronic ischemic cerebrovascular disease, they are nonspecific. A growing number of studies (King, 2014;Lee et al., 2016;Pietroboni et al., 2018) are finding direct correlations of WM disease with β-amyloid and tau accumulation, F I G U R E 1 White matter lesions appear bright (WM-hyper) on T2-FLAIR (a) and dark (WM-hypo) on T1-FSPGR (c) from a representative individual. Automatically identified WM lesions are shown overlaid on anatomic images with a yellow mask for WM-hyper in (b) and with a blue mask for WM-hypo in (d) supporting significant overlap between WM lesion development and neurodegeneration (Erten-Lyons et al., 2013;McAleese et al., 2015).
Ischemic white matter lesions may indicate a vascular susceptibility to AD (Hughes et al., 2013;King, 2014;King et al., 2014). Newer evidence points to other pathologic pathways whereby AD itself causes damage to the white matter tracts or results in secondary axonal degeneration due to gray matter damage (Amlien & Fjell, 2014).
Automated tools may provide significant benefits in generating reproducible and accurate assessments of white matter disease burden.
Methods for automated assessment of FLAIR WM-hyper (Hulsey et al., 2012) have been developed for research, but no single approach has achieved widespread adoption. Recently, FDA-approved programs like LesionQuant began offering an automatic evaluation of WM-hyper.
High contrast and high-resolution T1-weighted images are the primary input requirement for automated brain segmentation programs such as FreeSurfer (Bruce Fischl, 2012)  In this study, we sought to assess whether WM-hypo and WMhyper are equivalent markers of WM damage in normal aging and early neurodegenerative disease by evaluating their associations with age and cerebrospinal fluid (CSF) levels of β-amyloid and tau.
We hypothesized that WM-hypo and WM-hyper volumes would be highly correlated, but that volumes of WM-hyper would be more extensive over a larger region when compared with WM-hypo. We also hypothesized that WM-hypo and WM-hyper would be increased with age and would reflect neurodegeneration as revealed by abnormal CSF β-amyloid and tau levels.

| Sample population
In this IRB-approved study with written informed consent, 56 nondemented participants were examined. All participants were recruited through newspaper article, local senior centers visit, and word of mouth. All participants underwent MR imaging and cognitive testing but only 47 consented to lumbar puncture for CSF collection (Harrington et al., 2013).

| Cognitive assessment
Study participants were included if they were classified as cognitively healthy (CH) or had mild cognitive impairment (MCI) after medical and neuropsychological assessment using the Uniform Data Set-3 criteria (Weintraub et al., 2018) of the National Alzheimer's Coordinating Center after consensus clinical conference as previously described for this study cohort (Harrington et al., 2013

| WM lesion load on Standard Space
WM-hyper and WM-hypo masks of these 56 participants were transposed onto the standard space using SPM12 with a voxel size of 1 mm. The WM-hyper and WM-hypo masks in standard space were added together to create a group distribution mask showing all WM-hyper lesions in the cohort. The group mask intensity thresholds were set to greater than 1 to limit noise.

| CSF data acquisition
After an overnight fast, lumbar CSF was obtained between 8:00 a.m. and 10:00 a.m. and immediately examined for cells and total protein.

| RE SULTS
Demographic information is shown in   Figure 2. WM-hyper occurs at locations of WM-hypo but also involves additional brain regions, such as the fornix and internal capsule.

| WM-hyper and WM-hypo association with CSF β-amyloid and tau
Correlations of WM-hyper and WM-hypo with CSF markers were evaluated and if significant were further evaluated in a linear model adjusted for age and cognitive status. Relationships of WM-hyper and WM-hypo with β-amyloid are shown in Figure 5. β-Amyloid accumulation in the brain coincides with a decrease in CSF amyloid.
Lower CSF β-amyloid concentrations, the more pathological condition, were predicted by greater WM-hyper (Figure 5a; r = .33, p.e. −0.11 ± 0.044, p = .013) and WM-hypo (Figure 5b; r = .40, p.e. −0.24 ± 0.073, p = .002) volumes. Adjusting for age and cognitive status did not significantly alter prediction of β-amyloid by WM-hyper (−0.12 ± 0.05, p = .016) and WM-hypo (−0.26 ± 0.08, p = .003). There was no correlation of CSF tau with WM-hyper (Figure 6a; p = .9) or WM-hypo (Figure 6b; p = .99). be clinically significant as impairment is mostly associated with large lesion volumes . As WM lesion volumes increase, we observe a better fit between WM-hypo and WM-hyper as shown by less deviation of points from the regression line. We also observe that WM-hyper increase in volume is greater than the corresponding value of WM-hypo. This supports our assertion that for a given lesion volume, WM-hypo corresponds to more advanced disease as compared to WM-hyper. This further corroborates with the spatial distribution of WM-hyper and WM-hypo as shown in Figure 2. The area of WM-hypo involvement is almost entirely within the region where WM-hyper occurs in the periventricular white matter. WM-hyper has additional involvement within the deep white matter, which may correspond to an earlier stage of disease development (Schmahmann, Smith, Eichler, & Filley, 2008). In future longitudinal studies, we will assess whether WM-hypo later spread to involve areas that currently only demonstrate WM-hyper. Though WM-hyper is nonspecific and associated with many factors, age is consistently the strongest predictor of lesion accumulation in community-based studies (King, Chen, et al., 2013;Zhong & Lou, 2016).

| D ISCUSS I ON
We found the expected linear correlation between larger WM-hyper and WM-hypo ( Figure 4) volume with greater age. It suggests that evaluation of T1 has utility in demonstrating the presence of age-related white matter changes in the absence of a T2-FLAIR sequence.
Both WM-hypo and WM-hyper were shown to be increased among those with lower CSF β-amyloid in our cohort of nondemented individuals ( Figure 5). The presence of this association has been interpreted by some as suggestive that damage to small blood vessels within the brain may promote risk for Alzheimer's disease (Iadecola, 2010). Recent work has suggested that amyloid accumulation associated with AD neurodegeneration may also contribute to the formation of WM lesions independent of vascular diseases (Scott et al., 2016). WM lesions are present in autosomal dominant familial Alzheimer's disease (Lee et al., 2016) even at younger ages when microvascular disease is not generally prevalent . ADrelated damage may cause white matter damage through Wallerian degeneration occurring secondary to neuronal insult (McAleese et al., 2017). A recent study by Pietroboni et al. (2018) found abnormal β-amyloid to be the best predictor of WM lesion load within a cohort F I G U R E 4 Increase in WM-hyper (a) and WM-hypo (b) shows a significant correlation with age F I G U R E 5 Increased WM-hyper (a) and WM-hypo (b) were strongly correlated with low levels of CSF β-amyloid, supporting a link between white matter damage and changes associated with early Alzheimer's disease pathology with AD. Low CSF β-amyloid levels indicate impaired clearance from, and accumulation within, the brain and may detect abnormalities earlier than positron emission tomography (PET) (Palmqvist, Mattsson, & Hansson, 2016). An important difference in our current study is that we excluded individuals with dementia, demonstrating association of WM lesion with earlier stages of AD pathology in a cohort consisting mostly of cognitively intact individuals. Consistent results produced by comparing WM-hypo and WM-hyper directly denote them to be comparable markers of leukoaraiosis and overall suggest tight pairing between distinct changes occurring in the white matter driving T1-and T2-weighted signal changes (Simpson et al., 2007).
Despite showing statistically significant associations with β-amyloid, we observed no associations with tau for either WM-hyper or WM-hypo ( Figure 6), but work by McAleese et al. (2015) found a closer independent correlation between WM lesion load in AD with tau than β-amyloid. This difference may be related to a lower prevalence of tau in our cohort, which includes primarily cognitively intact, nondemented individuals (Harrington et al., 2013). CSF tau increases as cognitive health declines and tends to be a more accurate marker of later AD stage (Shim & Morris, 2011). It is therefore possible that tau may become more closely correlated with WM lesion volume as our cohort ages.
A recent study by Dadar et al. (2018) provided the initial proof of T1-weighted assessment of WM-hypo lesions in comparison with hyperintense lesions identified on T2-weighted, proton density (PD), or FLAIR. We build upon this work here in several respects. First, the prior study used an in-house Random Forest classifier detection program for WM-hypo and manual segmentation for WM-hyper, and it is unclear how these findings will generalize to other methods. Here, we use freely available automated segmentation tools for the identification of both WM-hypo and WM-hyper. Second, in our current work, we provide more information about the correlation of white matter lesions with the presence of neurodegeneration as indicated by CSF levels of βamyloid and tau.

| Limitations
There are several limitations to this study. First, we only have a cohort of 56 nondemented subjects. The small sample size may not fully represent the general population; however, the result is the first step toward analyzing a larger data set. Second, while we compared WM-hypo and WM-hyper, differences in the underlying algorithms for detecting lesions may also underlie differences we observed rather than actual differences in tissue pathology.
We used the Lesion Segmentation Tool, an automated method for quantifying FLAIR-hyperintense WM lesions which first relies on tissue segmentation performed by SPM using the intensity values in T1-weighted anatomic images to identify tissues as gray matter, white matter, and CSF. WM-hypo is generated by FreeSurfer which uses a more sophisticated algorithm guided by an anatomic atlas for identifying gray matter, WM, and CSF. Third, we utilize cross-sectional comparisons of WM-hyper and WM-hypo with other markers and cannot assess causation. Finally, in this initial work, we compare only global WM-hyper and WM-hypo volumes.
Future studies are needed to take into account spatial distribution and determine if overlapping or nonoverlapping WM-hyper and WM-hypo lesions may provide additional information about lesion severity or specific associations with outcomes.

| CON CLUS ION
Automated assessment of WM-hypo volume is closely associated with WM-hyper and both provide equivalent measures of leukoaraiosis progression with aging and prediction of abnormal CSF β-amyloid. WM-hypo volumes generated by FreeSurfer, which are currently rarely employed, can serve as a meaningful measure of white matter insult. This may facilitate retrospective analysis of older imaging data sets or serve as a standardized technique for comparison of WM lesion severity in existing studies using a widely F I G U R E 6 WM-hyper (a) and WM-hypo (b) showed a trend toward higher values among those with greater CSF Tau concentrations available automated technique with a corresponding FDA-approved version available for clinical use.

CO N FLI C T O F I NTE R E S T
There are no conflicts of interest to disclose.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request. Data collected at Huntington Medical Research Institutes is a contributor to ADNI's data collection.