Investigating White Matter Neuroinflammation in Alzheimer Disease Using Diffusion-Based Neuroinflammation Imaging

Abstract

Background and Objectives

Alzheimer disease (AD) is primarily associated with accumulations of amyloid plaques and tau tangles in gray matter, however, it is now acknowledged that neuroinflammation, particularly in white matter (WM), significantly contributes to the development and progression of AD. This study aims to investigate WM neuroinflammation in the continuum of AD and its association with AD pathologies and cognition using diffusion-based neuroinflammation imaging (NII).

Methods

This is a cross-sectional, single-center, retrospective evaluation conducted on an observational study of 310 older research participants who were enrolled in the Knight Alzheimer's Disease Research Center cohort. Hindered water ratio (HR), an index of WM neuroinflammation, was quantified by a noninvasive diffusion MRI method, NII. The alterations of NII-HR were investigated at different AD stages, classified based on CSF concentrations of β-amyloid (Aβ) 42/Aβ40 for amyloid and phosphorylated tau181 (p-tau181) for tau. On the voxel and regional levels, the relationship between NII-HR and CSF markers of amyloid, tau, and neuroinflammation were examined, as well as cognition.

Results

This cross-sectional study included 310 participants (mean age 67.1 [±9.1] years), with 52 percent being female. Subgroups included 120 individuals (38.7%) with CSF measures of soluble triggering receptor expressed on myeloid cells 2, 80 participants (25.8%) with CSF measures of chitinase-3–like protein 1, and 110 individuals (35.5%) with longitudinal cognitive measures. The study found that cognitively normal individuals with positive CSF Aβ42/Aβ40 and p-tau181 had higher HR than healthy controls and those with positive CSF Aβ42/Aβ40 but negative p-tau181. WM tracts with elevated NII-HR in individuals with positive CSF Aβ42/Aβ40 and p-tau181 were primarily located in the posterior brain regions while those with elevated NII-HR in individuals with positive CSF Aβ42/Aβ40 and p-tau181 connected the posterior and anterior brain regions. A significant negative correlation between NII-HR and CSF Aβ42/Aβ40 was found in individuals with positive CSF Aβ42/Aβ40. Baseline NII-HR correlated with baseline cognitive composite score and predicted longitudinal cognitive decline.

Discussion

Those findings suggest that WM neuroinflammation undergoes alterations before the onset of AD clinical symptoms and that it interacts with amyloidosis. This highlights the potential value of noninvasive monitoring of WM neuroinflammation in AD progression and treatment.

Introduction

Alzheimer disease (AD) is the most common cause of dementia. The characteristic AD neuropathology includes extracellular β-amyloid (Aβ) plaques composed of Aβ peptide and intraneuronal neurofibrillary tangles primarily composed of hyperphosphorylated tau. It is important to note that the neuropathologic changes of AD accumulate for a decade or more before the onset of dementia symptoms. There is increasing interest in the role of neuroinflammation in AD,¹ and recent evidence supports that AD-related neuroinflammation occurs not only in the gray matter^2,3 but also in the white matter (WM). Raj and colleagues⁴ found increased expression of IBA1, HLA-DR, and CD68, markers of neuroinflammation activities, in the WM of human postmortem early-onset AD brain tissue compared with age-matched controls. Other investigations on cerebral WM neuropathologic abnormalities in patients with AD reported axonal loss, demyelination, death of oligodendroglia cells, and reactive gliosis.^5,6 These findings underline the importance of examining the role of WM injury and neuroinflammation in the pathophysiology of AD.

Recently, advanced diffusion MRI techniques have been used to investigate WM injury and neuroinflammation, including neurite orientation dispersion and density imaging (NODDI),⁷ free water imaging (FWI),⁸ and diffusion basis spectrum imaging (DBSI).^9,10 The NODDI model demonstrated its sensitivity in capturing changes in neuroinflammation in mice.¹¹ As a regularized bi-tensor model, FWI measures isotopically hindered water to reflect inflammation-associated edema in AD and other neurodegenerative disorders.^12,13 As a data-driven method, DBSI uses a linear combination of multiple diffusion tensors to describe heterogeneous pathophysiology components (cellularity, extracellular water/vasogenic edema, axonal injury/loss, and demyelination).^9,10,14,15 The hindered water diffusion derived from DBSI is sensitive to cerebral edema associated with neuroinflammation.^16,17 Recent human studies also found that elevated DBSI hindered water ratio (HR) in obesity, suggesting that WM neuroinflammation–related cerebral edema was associated with obesity-related brain dysfunction.^18,19 The results of these studies demonstrate the great potential for advanced diffusion MRI methods to identify the microstructural alternations associated with neuroinflammation.

Despite the development of these novel techniques to image water diffusion related to WM neuroinflammation, it has been understudied relative to cortical neuroinflammation in AD.^2,20 Taking advantage of DBSI as a noninvasive diffusion MRI technique to image and quantify cerebral extracellular water edema associated with WM neuroinflammation,²¹ we aim to systematically examine WM neuroinflammation and link it to AD hallmarks, such as amyloid and tau, as well as cognitive function. Specifically, we implemented a clinical-compatible DBSI technique, neuroinflammation imaging (NII), in this study to image and quantify cerebral extracellular water edema in older research participants. In our previous investigation, we used NII-derived cell diffusivity, fractional anisotropy, and radial diffusivity to explore WM inflammation–associated cellularity changes and degeneration in individuals with preclinical and early symptomatic AD. However, our earlier work did not address the specific relationships between WM neuroinflammation and the hallmark AD pathologies of amyloid and tau, nor did it evaluate any potential links with cognitive function or the predictive value for cognitive decline. These critical aspects have now been thoroughly examined in this study to expand our understanding of the role of WM neuroinflammation in the context of AD pathology and its potential implications for disease progression and cognitive outcomes. NII-HR was quantified in each participant. CSF measures of amyloid (A, the ratio of Aβ peptide 42–40, Aβ42/Aβ40) and tau (T, tau phosphorylated at position 181, p-tau181) and the presence and/or severity of dementia were used to categorize participants at different stages of the AD continuum. The relationship of NII-HR with CSF Aβ42/Aβ40, CSF p-tau181, and a cognitive composite was examined.

Methods

Study Participants

Participants were recruited in an observational study focusing on memory and aging, conducted at the Knight Alzheimer's Disease Research Center (ADRC) at Washington University School of Medicine (St. Louis, MO). Inclusion criteria were the following: (1) Participants fell within the age range of 55–100 years, (2) participants had undergone a diffusion MRI scan, (3) participants had undergone CSF collection within 1 year of the diffusion MRI acquisition with data available on CSF Aβ42/Aβ40 and p-tau181, and (4) participants had undergone cognitive battery examination within 1 year of and after the diffusion MRI acquisition. A flowchart of diffusion MRI acquisition, CSF collection, and cognitive examination is shown in eFigure 1 (links.lww.com/WNL/D381). Details of recruitment have been published elsewhere.²² Experienced clinicians conducted thorough evaluations of all individuals using a semi-structured interview, which included input from a knowledgeable collateral source. The interview was in accordance with the National Alzheimer's Coordinating Center's Uniform Data Set guidelines.²³ A diagnosis of symptomatic AD was made as appropriate according to previous publications.^24,25 Dementia presence and severity were assessed using the Clinical Dementia Rating (CDR).²⁶ The CDR scale ranges from 0 (cognitively normal) to 0.5 (very mild), 1 (mild), 2 (moderate), and 3 (severe) dementia. Notably, most individuals diagnosed with mild cognitive impairment impairing memory would be assigned a CDR of 0.5 (very mild dementia). Race and sex were self-identified. This study is a cross-sectional study in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology statement.²⁷

Standard Protocol Approvals, Registrations, and Patient Consents

The Human Research Protection Office at Washington University approved all studies. Written informed consent was obtained from all participants in the study. In compliance with the retrospective analysis that is institutional review board–approved, the approval from an ethical standards committee to conduct this study was received.

APOE Genotyping

Standard procedures were used to extract DNA from peripheral blood samples collected from all participants during the enrollment process. The gene encoding APOE was genotyped as previously described.²⁸ APOE ε4 carriers had at least 1 APOE ε4 allele (ε2/ε4, ε3/ε4, or ε4/ε4).

Cognitive Measures

A cognitive battery was administered after the clinical assessment. To measure cognitive performance, we calculated a global composite score from the Knight ADRC cognitive battery for each individual. The composite score was calculated as previously described.²⁹ In brief, each individual's z-score across all common tasks collected by the Knight ADRC's Neuropsychological Battery was averaged.³⁰ The cognitive battery that was performed closest in time (within <1 year) to the NII scans was used to generate the baseline cognitive scores.

CSF Collection and Analysis

The procedures for collecting and handling CSF samples were consistent with those described in a previous publication.³¹ A fully automated platform (LUMIPULSE G1200, Fujirebio, Malvern, PA) was used to measure concentrations of Aβ40, Aβ42, total tau, and p-tau181 by chemiluminescent enzyme immunoassay. The measurements were performed in accordance with the manufacturer's specifications. CSF chitinase-3–like protein 1 (YKL-40) was measured with a commercial ELISA from Quidel. Measurement of CSF soluble triggering receptor expressed on myeloid cells 2 (sTREM2) was performed using an in-house plate-based ELISA.³²

MRI Acquisition

MRI data were acquired using 3T TIM Trio (Siemens, Erlangen, Germany) scanners with a 12-channel head coil equipped with parallel imaging. The acquisition parameters for diffusion MRI, T1-weighted magnetization-prepared rapid acquisition gradient echo, and T2-weighted fast spin echo scans were identical to those detailed in a previous publication.¹⁵

NII Data Processing and Analysis

The analysis of the NII data is the same as described in our previous work.¹⁵ Briefly, thresholds for the isotropic diffusivities between 0.3 and 2.5 were considered as the HR (NII-HR) to reflect the extracellular water edema in WM, based on our previous studies.^9,10 NII-HR was quantified by solving the NII model. Tract-Based Spatial Statistics (TBSS) from the FMRIB Software Library were used to examine the whole-brain, voxel-wise NII index with multiple comparisons.³³ To ensure unbiased results in the voxel-wise statistics, we accounted for potential influences of age, sex, APOE ε4 carrier status, and intracranial brain volume (ICV) by including them as covariates in the analysis. In addition to the voxel-wise analysis, the total WM NII-HR, which is defined as the mean hindered ratio from all TBSS WM tracts, was also generated to indicate the level of WM neuroinflammation for each individual. For regional analysis, the averaged NII-HR for each WM tract was extracted as well.

Statistical Analysis

Nonparametric Kruskal-Wallis and χ² tests were used for comparing continuous and categorical variables of the participants' demographics, respectively. Differences between groups in NII-HR, CSF biomarkers, and the global cognitive composite were evaluated by 1-way analysis of variance. Tukey honest significant differences were used to evaluate the significance of pair-wise differences between group means. Partial Spearman correlations were used to assess the strength of the relationships between NII-HR and CSF biomarkers or the cognitive composite and were adjusted for age, sex, APOE ε4 carrier status, and ICV. Cutoffs for abnormal CSF Aβ42/Aβ40 (<0.0673) and p-tau181 (>51.8 pg/mL) were previously established and represent the value that best distinguished amyloid PET-positive and negative individuals based on the Youden index (combined sensitivity and specificity).³⁴

Spearman correlation coefficients were used to assess correlations between the baseline NII-HR from each WM tract and the baseline global cognitive composite. A random coefficient model was used to examine how NII-HR was associated with rate of change in the global cognitive composite, as described in our previous study.³⁵ By accounting for the heterogeneous number of visits and intervals between visits for each participant, the random coefficient model is able to incorporate all cognitive measures taken within 1 year of the NII scan, along with all available follow-up cognitive measures for each participant in the statistical analysis. The number of visits and intervals of follow-up cognitive tests are summarized in Table 1. The statistical model with time as both a fixed and random effect is the following:

$$\text{Cognition}=\beta 1\times \text{time}+\beta 2\times \text{baseline}\hspace{0.17em}\text{NII}+\beta 3\times \text{time}\times \text{baseline}\hspace{0.17em}\text{NII}+\beta 4\times \text{age}+\beta 5\times \text{sex}+\beta 6\times \text{APOE}\hspace{0.17em}\epsilon 4$$ (1)

Table 1.

Demographic and Clinic Characteristics of the Participants

Variables	A−T−/CDR 0	A+T−/CDR 0	A+T+/CDR 0	A+T+/CDR 0.5	A+T+/CDR 1	Overall p values
n	190	39	40	30	11	—
Baseline CDR	0	0	0	0.5	1	—
Baseline CDR-SB	0.01 (0.07)a,b	0.03 (0.11)a,b	0.05 (0.15)a,b	2.67 (1.04)c,d,e	5.59 (1.34)c,d,e	<0.001
Baseline MMSE	29.29 (1.08)a,b	28.97 (1.35)a,b	28.95 (1.66)a,b	24.93 (3.07)c,d,e	23.36 (3.96)c,d,e	<0.001
Baseline age, y	63.79 (8.63)a,b,c,d	69.21 (6.42)a,b,e	72.45 (7.61)e	74.64 (5.93)c,e	77.16 (7.27)c,e	<0.001
Female	103 (54)	20 (51)	18 (45)	16 (53)	5 (45)	0.712
Education, y	15.98 (2.48)	15.82 (2.48)	16.00 (2.99)	15.40 (2.65)	14.55 (3.11)	0.368
APOE ε4 carriers	48 (25)a,b,d	18 (47)	26 (65)e	23 (77)e	9 (82)e	<0.001
Non-Hispanic White	163 (86)	38 (92)	39 (98)	30 (100)	10 (91)	0.051
Baseline CSF Aβ42/Aβ40	0.0885 (0.00819)a,b.c,d	0.0539 (0.00847)a,b,d,e	0.0427 (0.00812)c,e	0.0426 (0.00702)c,e	0.0425 (0.0646)c,e	<0.001
Baseline CSF Aβ42, pg/mL	861.43 (258.33)a,b,c,d	536.41 (178.55)e	551.92 (160.08)e	474.03 (164.92)e	455.82 (139.46)e	<0.001
Baseline CSF p-Tau181, pg/mL	29.43 (8.46)a,b,d	38.69 (9.33)a,b,d	83.02 (24.76)a,c,e	107.77 (41.35)c,d,e	94.20 (58.04)c,e	<0.001
Baseline CSF normalized sTREM2f	1,616.38 (493.07)a,d	1,735.66 (374.32)d	2,090.32 (908.48)c,e	1,966.23 (476.85)e	1,968.73 (427.01)	<0.001
Baseline CSF YKL-40, pg/mLg	201.84 (76.55)a,d	215.15 (59.71)d	313.07 (97.39)c,e	301.60 (85.44)e	242.20 (—)	<0.001
Baseline global cognitive composite	0.24 (0.30)a,b	0.26 (0.37)a,b	0.08 (0.38)a,b	−0.68 (0.62)b,c,d,e	−1.12 (0.75)a,c,d,e	<0.001
Cognitive follow-up (no. of visits)h	5.21 (2.85)	6.38 (2.67)	6.36 (3.06)	4.15 (2.19)c,d	3.00 (0.87)c,d	<0.001
Cognitive follow-up time, yh	6.43 (2.73)	6.72 (2.52)	6.39 (2.94)	3.74 (2.61)c,d,e	2.57 (1.60)c,d,e	<0.001
Rate of change in global cognitive compositeh	−0.01 (0.07)a,b	−0.02 (0.06)a	−0.03 (0.11)a	−0.26 (0.22)b,c,d,e	−0.11 (0.16)a,e	<0.001

Abbreviations: Aβ = β-amyloid; CDR = Clinical Dementia Rating; MMSE = Mini-Mental State Examination; p-tau181 = tau phosphorylated at site 181; SB = sum of boxes; sTREM2 = soluble triggering receptor expressed on myeloid cells 2; T = tau; YKL-40 = chitinase-3–like protein 1.

Data are presented as mean (SD) or n (%). APOE ε4 carrier, individual carrying at least 1 APOE ε4 allele.

Comparisons between groups on CSF measures and cognitions were adjusted for age, sex, years of education, and APOE ε4 carrier status. When the overall difference between groups was significant, post hoc testing between groups was conducted. Bold values are p < 0.05.

^ap < 0.05 compared with A+T+/CDR 0.5.

^bp < 0.05 compared with A+T+/CDR 1.

^cp < 0.05 compared with A+T−/CDR 0.

^dp < 0.05 compared with A+T+/CDR 0.

^ep < 0.05 compared with A−T−/CDR 0.

^fAvailable in a subgroup: A−T−/CDR 0, n = 185; A+T−/CDR 0, n = 39; A+T+/CDR 0, n = 39; A+T+/CDR 0.5, n = 29; A+T+/CDR 1, n = 11.

^gAvailable in a subgroup: A−T−/CDR 0, n = 183; A+T−/CDR 0, n = 34; A+T+/CDR 0, n = 24; A+T+/CDR 0.5, n = 6; A+T+/CDR 1, n = 1.

^hAvailable in a subgroup: A−T−/CDR 0, n = 173; A+T−/CDR 0, n = 39; A+T+/CDR 0, n = 39; A+T+/CDR 0.5, n = 26; A+T+/CDR 1, n = 9.

The β estimate and corresponding p values for the interaction term were calculated using type 3 sum of squares: The time × baseline NII interaction term was examined to determine whether the baseline NII-HR predicted cognitive decline. All statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC) or R (R Core Team [2020]). A significance level of p < 0.05 was considered statistically significant, and Bonferroni correction was applied to account for multiple comparisons.

Data Availability

The article and supplementary material contain all the data related to this study. Upon request from a qualified investigator, deidentified data will be made available.

Results

Participant Characteristics

The 310 participants were grouped based on CSF Aβ42/Aβ40 (A+, abnormal <0.0673), CSF p-tau181 (T+, abnormal >51.8 pg/mL), and CDR to represent different stages of AD: (1) A−T−/CDR 0 (n = 190); (2) A+T−/CDR 0 (n = 39); (3) A+T+/CDR 0 (n = 40); (4) A+T+/CDR 0.5 (n = 30); and (5) A+T+/CDR 1 (n = 11). There was a significant difference in the ages of the groups (p < 0.0001), with the A−T−/CDR 0 group being the youngest (mean ± SD 63.79 ± 8.63 years) and the average age increasing in each successive group. There were no significant differences in sex, years of education, or race between the 5 groups.

The 3 groups with abnormal CSF p-tau181 (T+) had lower Aβ42/Aβ40 than the group with abnormal Aβ42/Aβ40 but normal p-tau181 (A−T−/CDR 0). However, CSF p-tau concentrations were not different between the A−T−/CDR 0 and A+T−/CDR 0 groups. In a subset of individuals with data available on CSF sTREM2 or YKL-40 (eTable 1 and 2, links.lww.com/WNL/D382), CSF sTREM2 and YKL-40 concentrations were significantly higher in the A+T+/CDR 0 and A+T+/CDR 0.5 groups compared with the A−T−/CDR 0 and A+T−/CDR 0 groups (eFigure 2, links.lww.com/WNL/D381). The baseline global cognitive composite was lower in groups with clinical cognitive impairment (CDR >0) compared with the groups with normal cognition (CDR = 0). Furthermore, groups with clinical cognitive impairment had a faster rate of decline on the global cognitive composite.

WM NII-HR Is Altered Early in AD

Early Elevation of Total WM NII-HR in AD

The NII-derived overall WM NII-HR index was overlaid on the TBSS-generated WM tracts (Figure 1A). The NII-HR on the TBSS-generated WM skeleton was quantified for all participants. The box plot in Figure 1A shows the total WM NII-HR for all WM tracts compared among the 5 groups. Significantly higher total WM NII-HR was observed in the A+T+/CDR 0 and A+T+/CDR 0.5 groups compared with the A−T−/CDR 0 group. There was no difference in the total WM NII-HR for all WM tracts between the A+T+/CDR 0.5 and A+T+/CDR 0 groups. There were no differences between the A+T+/CDR 1 group and other groups, but this may be related to the smaller size of this group (n = 11).

Figure 1. Quantification of WM NII-HR in AD.

WM NII-HR is visualized on the TBSS-generated WM tracts in Panel A–D. The total WM NII-HR in the A+T+/CDR 0 and A+T+/CDR 0.5 groups was significantly higher than the A+T−/CDR 0 and A−T−/CDR 0 groups, as shown in the box plot in Panel A. The regional WM NII-HR in the A+T+/CDR 0 and A+T+/CDR 0.5 groups was significantly higher than the A+T−/CDR 0 and A−T−/CDR 0 groups on the SS tract, as shown in the box plot in Panel B. The regional WM NII-HR in the A+T+/CDR 0.5 group was significantly higher than in groups with normal cognition (CDR 0) on the representative EC tract, as shown in the box plot in Panel C. There was no group difference of the regional WM NII-HR among all 5 groups on the representative PLIC tract, as shown in the box plot in Panel D. *p < 0.05; **p < 0.005; adjusted for age, sex, APOE ε4 carrier status, and ICV. AD = Alzheimer disease; CDR = Clinical Dementia Rating; EC = external capsule; HR = hindered water ratio; ICV = intracranial brain volume; NII = neuroinflammation imaging; PLIC = posterior limb of internal capsule; SS = sagittal stratum; TBSS = Tract-Based Spatial Statistics; WM = white matter.

WM Tracts With Early Changes in NII-HR

WM NII-HR in 3 regions demonstrated early changes in which the still cognitively normal A+T+/CDR 0 group had significantly higher NII-HR compared with the A−T−/CDR 0 and A+T−/CDR 0 groups: the sagittal stratum (SS), posterior thalamic radiation (PTR), and tapetum (Tp) (see red regions in Figure 2). A representative NII-HR for the SS is shown (Figure 1B). The group with very mild dementia (A+T+/CDR 0.5) also had higher NII-HR in these regions compared with the A−T−/CDR 0 and A+T−/CDR 0 groups.

Figure 2. Visualization of WM Tracts With the Alternations of NII-HR at Different Stages of AD.

Three WM tracts with early elevated NII-HR in the A+T+/CDR 0 group are marked in red. Eight WM tracts with late elevated NII-HR in the A+T+/CDR 0.5 group are marked in yellow. A = anterior; AD = Alzheimer disease; CDR = Clinical Dementia Rating; HR = hindered water ratio; NII = neuroinflammation imaging; P = posterior; R = right; L = left; WM = white matter.

WM Tracts With Later Changes in NII-HR

WM NII-HR in 8 regions was higher in the group with very mild dementia (A+T+/CDR 0.5) compared with the cognitively normal A−T−/CDR 0, A+T−/CDR 0, and A+T+/CDR 0 groups: the genu, body, and splenium of corpus callosum; external capsule (EC); superior longitudinal fasciculus; cingulum; anterior corona radiata; and posterior corona radiata (see yellow regions in Figure 2). A representative NII-HR for the EC is shown (Figure 1C). For these regions, there were no significant differences between WM NII-HR for the A−T−/CDR 0 group and the A+T−/CDR 0 or A+T+/CDR 0 group.

WM Tracts With Unchanged NII-HR

There were no group differences in the regional NII-HR among all 5 groups for most WM tracts, including middle cerebral peduncle and pontine crossing tract. The names of the WM tracts with unchanged NII-HR are presented in eFigure 3 (links.lww.com/WNL/D381). A representative NII-HR for the posterior limb of internal capsule (PLIC) is shown (Figure 1D).

WM NII-HR Is Correlated With CSF Aβ42/Aβ40

Total WM NII-HR Is Correlated With CSF Aβ42/Aβ40 in Individuals With Amyloidosis

The association between the total WM NII-HR and CSF markers of amyloid (CSF Aβ42/Aβ40) and tau (CSF p-tau181) was examined. In individuals without amyloidosis, there was no significant correlation between total WM NII-HR and CSF Aβ42/Aβ40 (rho = 0.11, p = 0.11, Figure 3A) or CSF p-tau181 (rho = −0.09, p = 0.20, Figure 3B). However, in individuals with amyloidosis, there was a significant negative correlation between total WM NII-HR and CSF Aβ42/Aβ40 (rho = −0.26, p = 0.004, Figure 3A). By contrast, there was no significant correlation between total WM NII-HR and CSF p-tau181 (rho = 0.13, p = 0.15, Figure 3B).

Figure 3. Total WM NII-HR Is Correlated With CSF Aβ42/Aβ40, but Not CSF p-tau181, in Individuals With Amyloidosis.

There is a significant correlation between the total WM NII-HR and CSF Aβ42/Aβ40 in individuals with amyloidosis (black solid circles in A). There is no correlation between the total WM NII-HR and CSF p-tau181 in individuals without amyloidosis (black solid circles in B). There are no correlations between the total WM NII-HR and CSF levels of Aβ42/Aβ40 (hollow circles in A) and p-tau181 (hollow circles in B) in individuals without amyloidosis. Analyses were adjusted for age, sex, APOE ε4 carrier status, and ICV. HR = hindered water ratio; ICV = intracranial brain volume; NII = neuroinflammation imaging; WM = white matter.

Voxel-Wise Correlations Between WM NII-HR and CSF Aβ42/Aβ40 in Individuals With Amyloidosis

Voxel-wise correlations in participants with amyloidosis demonstrated widespread significant negative correlations between WM NII-HR and CSF Aβ42/Aβ40 (eFigure 4, links.lww.com/WNL/D381). These voxel-wise correlations for representative WM tracts with early (SS), later (EC), and unchanged NII-HR (PLIC) are shown in eFigure 5. The WM tract–based correlation ratio, computed as a ratio between the number of voxels with significant correlation and the total number of voxels on each WM tract, indicates the extent of the WM tract correlating with the CSF biomarker. The higher the WM tract–based correlation ratio, the more voxels contribute to the significant correlation. For WM tracts with early changes in NII-HR, more than 40% of voxels demonstrated a significant negative correlation between NII-HR and CSF Aβ42/Aβ40, and the averaged WM tract-based correlation ratio was 0.42 (eFigure 5A). For WM tracts with later changes in NII-HR, approximately 30% of voxels demonstrated significant negative correlations between NII-HR and CSF Aβ42/Aβ40, and the averaged WM tract–based correlation ratio was 0.28 (eFigure 5B). For WM tracts with unchanged NII-HR, approximately 15% of voxels demonstrated significant negative correlations between NII-HR and CSF Aβ42/Aβ40, and the averaged WM tract–based correlation ratio was 0.15 (eFigure 5C).

No Correlations Between WM NII-HR and CSF Measures of sTREM2 and YKL-40

There were no significant correlations between the total WM NII-HR and CSF sTREM2 and YKL-40 (eFigure 6, links.lww.com/WNL/D381). There were also no significant voxel-wise correlations between WM NII-HR and CSF measures of sTREM2 and YKL-40 in our cohort.

Correlations Between WM NII-HR and the Baseline Cognitive Measure

Total WM NII-HR was significantly correlated with the global cognitive composite (rho = −0.32, p < 0.001, Figure 4A). Voxel-wise correlations demonstrated widespread significant negative correlations between NII-HR and the cognitive composite (eFigure 7, links.lww.com/WNL/D381). Correlations between the cognitive composite and NII-HR for representative WM tracts that change early (SS) and later (EC) are shown in Figure 4, B and C. By contrast, NII-HR was not correlated with the global cognitive composite in the unchanged PLIC tract (Figure 4D). After adjusting for either CSF Aβ42/Aβ40 or CSF p-tau181, correlations between the baseline WM NII-HR and the baseline cognitive composite remained significant.

Figure 4. Correlations Between the Total WM NII-HR and Cognitive Measures of the Global Cognitive Composite.

Scatterplots show the significant negative correlations between the total WM NII-HR and the global cognitive composite (A). The significant negative correlation between the WM NII-HR and the global cognitive composite was also found in the SS (B) and EC (C) tracts. There was no significant correlation between the global cognitive composite and WM NII-HR in the PLIC tract (D). Analyses were adjusted for age, sex, APOE ε4 carrier status, and ICV. HR = hindered water ratio; ICV = intracranial brain volume; NII = neuroinflammation imaging; PLIC = posterior limb of internal capsule; WM = white matter.

Baseline WM NII-HR Predicts Cognitive Decline in AD

The ability of baseline WM NII-HR to predict cognitive decline was evaluated using a random coefficient model. The baseline total WM NII-HR predicted decline on the global cognitive composite (p = 0.001) (Table 2). Regional NII-HR in representative early (SS, p = 0.02) and later (EC, p = 0.02) changing regions also predicted cognitive decline (Table 2). The baseline WM NII-HR did not predict cognitive decline in the unchanged WM tracts.

Table 2.

Random Coefficient Model of Cognitive Decline as a Function of WM NII-HR by WM Tract

Region of interests	Estimate (standard error)	t Value	p Value
Whole WM	−0.48 (0.15)	−3.25	0.001
WM tracts with early elevated NII-HR
SS-L	−0.51 (0.16)	−3.18	0.02
SS-R	−0.42 (0.13)	−3.20	0.02
Tp-L	−0.08 (0.06)	−1.34	0.18
Tp-R	−0.14 (0.07)	−1.98	0.06
PTR-L	−0.39 (0.14)	−2.81	0.10
PTR-R	−0.22 (0.13)	−1.67	0.10
WM tracts with later elevated NII-HR
gCC	−0.45 (0.17)	−2.65	0.15
bCC	−0.67 (0.21)	−3.21	0.02
sCC	−0.82 (0.35)	−2.35	0.38
EC-L	−0.48 (0.14)	−3.34	0.02
EC-R	−0.40 (0.14)	−2.76	0.11
SLF-L	−0.32 (0.16)	−1.98	0.10
SLF-R	−0.21 (0.17)	−1.25	0.21
Ccg-L	−0.41 (0.18)	−2.23	0.38
Ccg-R	−0.61 (0.19)	−3.08	0.38
ACR-L	−0.31 (0.14)	−2.16	0.57
ACR-R	−0.20 (0.14)	−1.44	0.15
PCR-L	0.12 (0.12)	0.98	0.33
PCR-R	0.10 (0.12)	0.80	0.43

Abbreviations: ACR = anterior corona radiata; bCC = body of corpus callosum; Ccg = cingulum (cingulate gyrus); EC = external capsule; gCC = genu of corpus callosum; HR = hindered water ratio; L = left; NII = neuroinflammation imaging; PCR = posterior corona radiata; PTR = posterior thalamic radiation; R = right; sCC = splenium of corpus callosum; SLF = superior longitudinal fasciculus; SS = sagittal stratum; Tp = tapetum; WM = white matter.

Discussion

In this study, cerebral WM extracellular water edema–associated neuroinflammation was investigated using NII over the AD continuum from cognitively normal and AD biomarker negative to mild AD dementia. We found elevated NII-HR in cognitively normal individuals with abnormal amyloid and tau CSF biomarkers, indicating that WM neuroinflammation is detectable before the onset of AD symptoms. Although CSF sTREM2 and YKL-40 are believed to reflect TREM2-mediated microglial activation and astrocytosis, respectively, there were significant correlations of total WM NII-HR with CSF Aβ42/Aβ40, but not with CSF p-tau181, sTREM2, or YKL-40. This suggests that amyloidosis is driving neuroinflammation as measured by WM NII-HR. Baseline NII-HR predicted performance on a cognitive composite as well as cognitive decline, demonstrating the association between neuroinflammation and AD symptoms.

Although deposition of amyloid plaques is believed to be an initiating event that leads to the downstream synaptic dysfunction and neuronal death characteristic of AD, growing interest is being observed in exploring additional potential pathophysiologic mechanisms related to AD, including the accumulation of intracellular tau,³⁶ and, more recently, neuroinflammation.³⁷ In AD brains, the presence of activated microglia and astrocytes, along with the release of various inflammatory cytokines around Aβ plaques, has been documented.³⁸ This immune activation and consequent neuroinflammation are believed to be involved in the progression of AD. For example, reactive astrocytes are critically involved in the formation of cerebral edema through the regulation of the aquaporin-4 (AQP4) water channel, which is abundantly expressed on astrocytic end feet in CNS.³⁹ It has also been hypothesized that AQP4-induced astroglial water flow is a driving force that contributes to the perivascular clearance of interstitial solutes like amyloid and tau.⁴⁰ Because diffusion MRI has demonstrated its sensitivity in identifying cerebral edema in a variety of brain disorders, it is reasonable and feasible to assess neuroinflammation-associated cerebral edema in AD using NII, evaluate its relationship with other AD pathologies of amyloid and tau, and investigate how it affects cognition.

Current fluid biomarkers of neuroinflammation in AD research include CSF measures of sTREM2 and YKL-40. The elevated CSF level of YKL-40 in mild cognitive impairment (MCI) and AD compared with cognitively normal individuals highlight the marker's potential in distinguishing between participants with normal cognition, MCI, and AD, as well as in predicting the progression from cognitively normal status to MCI.⁴¹ The levels of CSF sTREM2 and YKL-40 change relatively early in the AD time course, indicating that neuroinflammation occurs at or shortly after amyloid deposition.⁴² In addition, the increased CSF sTREM2 has also been observed in participants of autosomal dominant AD (ADAD) mutation carriers⁴³ and early symptomatic stages of sporadic AD.⁴⁴ In this study, the A+T+/CDR 0 group had increased levels of CSF sTREM2 and YKL-40 compared with the A−T−/CDR 0 group, suggesting that neuroinflammation changed early before the onset of clinical symptoms in AD.⁶ Similar to the findings of CSF sTREM2 and YKL-40, the increased total WM NII-HR in the A+T+/CDR 0 and A+T+/CDR 0.5 groups compared with the A−T−/CDR 0 group may indicate increased WM neuroinflammation in preclinical and early symptomatic AD. It is interesting that both NII and CSF inflammation measures were not further increased in the group with mild dementia (CDR = 1), suggesting that neuroinflammation may not continue to increase after symptoms are present, which is consistent with previous CSF and PET studies that discovered an increase in neuroinflammation in early AD and a decrease in neuroinflammation in the later AD stage.⁴⁴ More importantly, the noninvasive and nonradioactive quantification of neuroinflammation using NII enables the longitudinal examination of WM neuroinflammation in high spatial resolution at different stages of AD, which could complement the CSF and PET in AD research and improve understanding of the evolution of neuroinflammation in AD.

In this study, the earliest NII-HR alternations have been observed in 3 WM tracts: SS, PTR, and Tp. The WM tract of SS investigated in this study includes the inferior longitudinal fasciculus (ILF) and inferior frontal occipital fasciculus in the JHU white matter tractography atlas. Previous studies have found decreased microstructural integrity of ILF, PTR, and Tp in participants with preclinical AD and AD compared with normal controls.^45,46 In addition, these tracts are located in the medial temporal lobe or are connected with the hippocampus,⁴⁷ one of the earliest affected brain regions in AD. Therefore, the AD neuroinflammation–associated cerebral edema could result from the early AD pathology affecting the medial temporal lobe due to the accumulation of aggregated hyperphosphorylated tau protein and deposition of Aβ in AD brains.⁴⁸

The significant correlation between the total WM NII-HR and CSF Aβ42/Aβ40 was found in individuals with amyloidosis but not in individuals without amyloidosis, suggesting that WM neuroinflammation may be driven by the increase of amyloid deposition in preclinical AD. This is also consistent with our previous findings that WM neuroinflammation was negatively related to the CSF measure of amyloidosis in preclinical AD, suggesting that the WM inflammation is associated with increasing Aβ burden. In addition, the absence of significant correlation between the total WM NII-HR and CSF p-tau181 in individuals with amyloidosis may indicate a weaker association between WM neuroinflammation and tauopathy in preclinical and early AD. Studies have found a clear interaction among neuroinflammation, amyloid, and tau accumulation in human AD brains.⁴⁹ By combining the noninvasive and nonradioactive NII inflammation biomarker with PET imaging of amyloid and tau, the spatial and temporal relationship between neuroinflammation, amyloid, and tau can be systematically evaluated in future studies to better understand the mechanisms underlying AD.

In this study, we tested whether WM neuroinflammation quantified by NII is associated with cognitive impairment. We found that baseline WM NII-HR was negatively correlated with baseline cognitive measures in widespread WM tracts. Furthermore, using a random coefficient model, we found that the baseline WM NII-HR predicted the cognitive decline. These findings imply that cognitive impairment is associated with greater neuroinflammation and that higher WM neuroinflammation could accelerate cognitive decline. Our findings warrant future studies to explore how the accumulation and spread of neuroinflammation contributes to the onset and severity of cognitive changes in AD.

Recently, the interaction between WM integrity and tau spreading has been examined to better understand the underlying mechanisms of WM injury and tau accumulation.⁵⁰ In this study, we found that WM tracts with early changes in NII-HR (Figure 5B.a) were identified in the posterior region of the brain, which interact with the cortical regions affected in Braak stage I/II and/or Braak stage III/IV as visualized by PET tau uptake (Figure 5, A.a and b). In addition, we observed that WM tracts with the later changes in NII-HR were the fibers connecting the posterior and anterior brain regions (Figure 5B.b). The propagation of tau from Braak stage I/II to Braak stage III/IV may be related to altered neuroinflammation of those WM tracts (Figure 5A.c). The WM tracts with unchanged NII-HR in early symptomatic AD may promote tau spreading from the Braak stage III/IV to V/VI in the later stage of AD. Longitudinal studies are needed to evaluate the hypothesis that WM neuroinflammation is linked to tau propagation.

Figure 5. Postulated Relationship Between WM Neuroinflammation and Tau Deposition.

The Braak stages were grouped by reaching the SUVR positivity threshold of 1.22 in FreeSurfer brain regions that correspond to the accumulation areas in each Braak stage group. Three individuals matching each Braak stage were selected from the Knight ADRC cohort. Voxel-wise individual participant PET data were surface-projected onto a 3D mesh of the right hemisphere cortical surface for visualization (Panel A). The tractography of WM tracts with early, later, and unchanged NII-HR was generated and visualized in Panel B using DSI studio. The arrows indicate orientations of WM tracts. ADRC = Alzheimer's Disease Research Center; HR = hindered water ratio; NII = neuroinflammation imaging; P = posterior; S = superior; SUVR = standardized uptake value ratio; WM = white matter.

There are several limitations to this study. First, our cohort included a small number of participants with mild dementia (CDR 1, n = 11) and no participants with moderate-to-severe dementia (CDR >1), which limited our ability to investigate the full spectrum of the alterations of WM neuroinflammation in AD, such as later stage of AD dementia. Future studies to combine imaging data of later stage of AD dementia from other cohorts, such as Alzheimer's Disease Neuroimaging Initiative, are imperative. Second, CSF measures of amyloidosis and tauopathy, which were used to stratify groups, lack spatial information. Future studies comparing NII measures to amyloid and tau PET scans will enable the investigation of the spatial distribution of AD pathologies and their interaction over the course of AD. Third, the cross-sectional approach in this study could not demonstrate the causal pathways between neuroinflammation and amyloidosis, tauopathy, and cognition. Future studies including longitudinal MRI and PET scans are needed to provide more insight of pathogenesis in AD.

In summary, we have investigated the evolution of WM neuroinflammation and the relationship between WM neuroinflammation, CSF biomarkers of amyloid and tau, and cognition over the AD continuum. Using a measure of WM neuroinflammation, NII-HR, we found that (1) WM neuroinflammation was increased in the preclinical AD and early symptomatic AD stages. (2) The spatial changes of WM neuroinflammation begin with the WM tracts in the posterior part of the brain and proceed to those tracts connecting the posterior and anterior brain regions. (3) In the presence of amyloidosis, WM neuroinflammation increased with greater abnormality in CSF amyloid but not with greater abnormality in CSF tau. (4) WM neuroinflammation was correlated with cognitive impairment, and baseline WM neuroinflammation predicted the cognitive decline. The findings from this study suggest that WM neuroinflammation is altered before the onset of AD clinical symptoms and WM neuroinflammation interacts with amyloidosis, which suggests the utility of noninvasive and nonradioactive monitoring of WM neuroinflammation in AD progression and treatment.

Glossary

AD

Alzheimer disease

ADRC

Alzheimer's Disease Research Center

AQP4

aquaporin-4

Aβ

β-amyloid

CDR

Clinical Dementia Rating

DBSI

diffusion basis spectrum imaging

EC

external capsule

FWI

free water imaging

HR

hindered water ratio

ICV

intracranial brain volume

ILF

inferior longitudinal fasciculus

NII

neuroinflammation imaging

NODDI

neurite orientation dispersion and density imaging

p-tau181

phosphorylated tau181

PLIC

posterior limb of internal capsule

PTR

posterior thalamic radiation

SS

sagittal stratum

sTREM2

soluble triggering receptor expressed on myeloid cells 2

TBSS

Tract-Based Spatial Statistics

Tp

tapetum

WM

white matter

YKL-40

chitinase-3–like protein 1

Appendix. Authors

Name	Location	Contribution
Qing Wang, PhD	Mallinckrodt Institute of Radiology, and Knight Alzheimer Disease Research Center, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data; additional contributions: developed the methodology, performed primary data analysis, and wrote the paper
Suzanne E. Schindler, MD, PhD	Knight Alzheimer Disease Research Center, and Department of Neurology, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data; additional contributions: contributed resources
Gengsheng Chen, PhD	Mallinckrodt Institute of Radiology, and Knight Alzheimer Disease Research Center, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Nicole S. Mckay, PhD	Mallinckrodt Institute of Radiology, and Knight Alzheimer Disease Research Center, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data; additional contributions: helped with cognitive test
Austin McCullough, PhD	Mallinckrodt Institute of Radiology, and Knight Alzheimer Disease Research Center, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Shaney Flores, MS	Mallinckrodt Institute of Radiology, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Jingxia Liu, PhD	Department of Surgery, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Zhexian Sun, PhD	Department of Biomedical Engineering, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Sicheng Wang, PhD	Department of Electrical and System Engineering, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Wenshang Wang, MS	Department of Electrical and System Engineering, Washington University School of Medicine, St. Louis, MO	Analysis or interpretation of data
Jason Hassenstab, PhD	Knight Alzheimer Disease Research Center, and Department of Neurology, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data; additional contributions: contributed resources
Carlos Cruchaga, PhD	Department of Neurology, and Department of Psychiatry, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data; additional contributions: contributed resources
Richard J. Perrin, MD, PhD	Knight Alzheimer Disease Research Center, and Department of Pathology & Immunology, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data; additional contributions: contributed resources
Anne M. Fagan, PhD	Knight Alzheimer Disease Research Center, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data; additional contributions: contributed to study conception, resources, and project administration
John C. Morris, MD	Knight Alzheimer Disease Research Center, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data; additional contributions: contributed to study conception, resources, and project administration
Yong Wang, PhD	Mallinckrodt Institute of Radiology, Department of Electrical and System Engineering, and Department of Obstetrics & Gynecology, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data; additional contributions: developed the methodology, performed primary data analysis, and wrote the paper; provided supervision of the study
Tammie L.S. Benzinger, MD, PhD	Mallinckrodt Institute of Radiology, and Knight Alzheimer Disease Research Center, and Department of Neurosurgery, Washington University School of Medicine, St. Louis, MO	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data; additional contributions: contributed to study conception, resources, and project administration; provided supervision of the study

Study Funding

This study was supported, in part, by grants from the NIH including the National Institutes on Aging (NIA) P01AG027276 (Antecedent Biomarkers of AD: The Adult Children Study, PI: J.C. Morris); NIA P01AG003991 (Healthy Aging and Senile Dementia, PI: J.C. Morris); NIA P30AG066444 (Alzheimer Disease Research Center, PI: J.C. Morris); and NIA R01AG054567-01A1(PIs: T.L.S. Benzinger and Y. Wang). Q. Wang is supported by NIA R03AG072375-01 (PI: Q. Wang) and NIA R01AG074909 (PI: Q. Wang). S.E. Schindler is supported by NIA R01AG070941 (PI: S.E. Schindler). Additional support was generously provided by the Charles and Joanne Knight Alzheimer's Research Initiative and by the Fred Simmons and Olga Mohan Fund and the Paula and Rodger Riney Fund.

Disclosure

Q. Wang is the inventor on the patent application PCT/US2017/030161. S.E. Washington, University of Indiana, and St. Luke's Hospital; is a board member of Greater Missouri Alzheimer's Association; and received data on behalf of Washington University from C2N Diagnostics at no cost. G. Chen, N.S. McKay, A. McCullough, S. Flores, J. Liu, Z. Sun, S. Wang, and W. Wang report no disclosures relevant to the manuscript. J. Hassenstab is a paid consultant for Lundbeck, Biogen, Roche, and Takeda. C. Cruchaga has received research support from: GSK and EISAI. The funders of the study had no role in the collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. C. Cruchaga is a member of the advisory board of Vivid Genomics and Circular Genomics and owns Stocks. R. Perrin reports no disclosures relevant to the manuscript. A.M. Fagan received consulting fees from DiamiR and Siemens Healthcare Diagnostics Inc. and participated on an advisory board at Roche Diagnostics, Genentech, and Diadem. J.C. Morris received consulting fees from Barcelona Brain Research Center and TS Srinivasan Advisory Board; received payment or honoraria from Montefiore Grand Rounds and Tetra-Inst ADRC seminar series; and participated on an advisory board at Cure Alzheimer's Fund. Y. Wang is the inventor on the patent application PCT/US2017/030161. T.L.S. Benzinger is the inventor on the patent application PCT/US2017/030161 and has investigator-initiated research funding from NIH, Alzheimer's Association, Barnes-Jewish Hospital foundation, and Avid Radiopharmaceuticals (a wholly owned subsidiary of Eli Lilly); participates as a site investigator in clinical trials sponsored by Avid Radiopharmaceuticals, Eli Lilly, Biogen, Eisai, Janssen, and Roche; serves as an unpaid consultant to Eisai and Siemens; and is on the Speaker's Bureau for Biogen. Go to Neurology.org/N for full disclosures.