Intracellular fluid accumulation underlies brain volume increases in early Alzheimer’s disease

Abstract

In the preclinical stages of Alzheimer’s disease, increased brain volume has been associated with amyloid-beta pathology, particularly in regions that undergo volume reductions as the disease progresses. Glial reactivity and water diffusion alterations have been linked to such macroscopic volumetric changes. Brain volume reductions have also been reported following amyloid-beta removal with anti-amyloid therapies with beneficial clinical effects, but it remains unclear whether these changes result from resolving amyloid-triggered neuroinflammation or neurodegeneration. Intravoxel incoherent motion modelling based on multi-shell diffusion-weighted imaging may provide a better understanding of the processes underlying these paradoxical changes. This study used intravoxel incoherent motion diffusion MRI to examine how alterations in cerebral water pools contribute to increased brain volume linked to amyloid-beta deposition and neuroinflammation in cognitively unimpaired individuals. We developed a three-compartment diffusion MRI model with four parameters of cerebral water diffusion: slow diffusion coefficient, fast diffusion coefficient, slow signal portion, and perfusion fraction. We computed these diffusion parameters in 297 cognitively unimpaired late middle-aged adults, 35% of whom showed evidence of amyloid deposition. We examined their correlation with demographic factors (age, sex, apolipoprotein E status), markers of Alzheimer’s disease pathology, neurodegeneration, neuroinflammation, and mean diffusivity. Then, we identified regions showing grey matter volume increases related to amyloid burden and examined the association between grey matter volume and diffusion parameters within these regions. We did not find evidence of associations between diffusion parameters and amyloid-related biomarkers, whether assessed by PET or cerebrospinal fluid measures. In contrast, the four diffusion parameters showed strong and widespread associations with biomarkers of neuroinflammation and neurodegeneration, particularly in frontoparietal and cingulate regions. Additionally, in grey matter regions where volume increases were related to amyloid levels, volumes were negatively correlated with the slow diffusion coefficient (P = 0.001), perfusion fraction (P = 0.036) and mean diffusivity (P = 0.047). These findings indicate that diffusion-derived measures are more sensitive to neuroinflammatory and neurodegenerative processes than to amyloid pathology in cognitively unimpaired individuals. Furthermore, the observed negative association between grey matter volume and slow diffusion coefficient in amyloid-related regions may reflect increased cellular complexity rather than intracellular water accumulation. This interpretation suggests that glial remodelling or microstructural changes could underlie brain volume increases in amyloid-positive individuals without overt neurodegeneration. These results underscore the value of intravoxel incoherent motion-derived metrics for gaining deeper insights into the pathophysiological mechanisms of Alzheimer’s disease, influencing brain volume changes as well as those resulting from the response to anti-amyloid therapies.

Kassinopoulos et al. examine how amyloid-β pathology in cognitively unimpaired individuals is linked to regional grey matter volume increases. Using diffusion MRI, they show that reduced intracellular water diffusion, likely reflecting neuroinflammation-related increased cellular complexity or glial remodelling, may underlie these volumetric changes.

Graphical Abstract

Graphical Abstract.

Introduction

Alzheimer’s disease (AD) is a neurodegenerative disease characterized by the accumulation of amyloid beta (Aβ) plaques and tau neurofibrillary tangles.¹ Typically, Aβ accumulates in the brain in late-middle-aged individuals who can remain cognitively unimpaired for decades. At these early preclinical stages, increases in brain volume have been linked with a neuroinflammatory response to Aβ accumulation and changes in water diffusion.^2,3 Neuroinflammation is a complex response to cerebral tissue injury involving the activation of immunocompetent cells -microglia, astrocytes- and can lead to brain swelling due to the accumulation of fluid in the tissue. In particular, increases in brain volume have been linked to an increase in CSF neuroinflammation biomarkers: soluble triggering receptor expressed on myeloid cells 2 (sTREM2), S100 calcium-binding protein (S100B), chitinase-3-like protein 1 (YKL-40), interleukin 6 (IL-6), and glial fibrillary acidic protein (GFAP)^4-7 Moreover, an increase in CSF neuroinflammation biomarkers has been associated with reductions in cerebral water diffusion in preclinical and symptomatic AD stages.⁸

Recently, Aβ removal through the use of monoclonal antibodies has been shown to be clinically beneficial in patients with mild cognitive decline and mild dementia due to AD.^9,10 However, paradoxically, such Aβ removal has also been associated with a reduction in brain volume, which is typically interpreted as a sign of accelerated neurodegeneration.¹¹ Yet, previous findings suggest that the reduction in brain volume with amyloid removal could be due to amyloid plaque clearance and associated cerebral fluid shifts,¹² as well as due to the resolution of the neuroinflammatory response to Aβ.¹³ Given that neuroinflammation and cerebral water accumulation are closely tied to AD pathophysiology, exploring the relationships between distinct cerebral water pools, neuroinflammatory activity, and brain volume may offer insights into the mechanisms underlying the paradoxical brain volume reductions observed following anti-amyloid monoclonal antibody treatment in early symptomatic AD.

In this study, we aimed to gain a better understanding of the relationship between cerebral water diffusion, neuroinflammation, and brain volume in a late middle-aged cohort of cognitively unimpaired individuals, many with positive Aβ biomarkers. Using multi-shell Diffusion-Weighted Imaging (DWI) combined with a three-compartment intravoxel incoherent motion (IVIM) model,¹⁴ we derived four diffusion-related parameters that distinguish contributions from the intracellular and extracellular space. Two of these parameters reflect the rate of diffusion in each space, while the other two reflect the relative signal contributions from the intracellular water pool and the microvasculature. We first analysed the possible association of these parameters with (a) participant characteristics such as age, sex, and apolipoprotein E (APOE) genotype; (b) biomarkers of AD (amyloid-PET positivity, and Aβ and phosphorylated tau proteins from the CSF) and neurodegeneration (neurofilament light [NfL] from the CSF, hippocampal volume, and AD signature [average cortical thickness in a set of brain regions known to undergo atrophy in AD])^15,16; (c) biomarkers of neuroinflammation (sTREM2, S100B, YKL-40, IL6, and GFAP from the CSF, and plasma GFAP); and (d) mean diffusivity (MD). Then, we examined regions with GM volume increases related to amyloid load and assessed the associations between IVIM parameters, MD, and GM volume in these regions. These analyses aimed to explore the potential contribution of cerebral water diffusion to the increase in GM volume, with a particular focus on neuroinflammatory processes.

Materials and methods

Study sample

This was an observational, cross-sectional, case-control study that included 297 late middle-aged adults from the ALFA + cohort.¹⁷ All ALFA + participants were cognitively unimpaired with a Mini-Mental State Examination above 26 and a Clinical Dementia Rating of 0. They were tested for the APOE-ε4 allele, and most had a family history of AD.¹⁸ All subjects gave written informed consent according to the Declaration of Helsinki, and the study was approved by the independent Ethics Committee ‘Parc de Salut Mar’ Barcelona (Spain).

Figure 1 summarizes the participant selection process and exclusion criteria of the 370 participants who underwent MRI acquisition for the ALFA + study. Out of these, 73 participants were excluded due to reasons mentioned in Fig. 1, leaving 297 participants in the present study. Aβ PET, APOE genotyping, and fluid sampling and biomarker analyses are described in the Supplementary Materials.

Figure 1.

Participant characteristics flow chart. Flow chart of the participant selection. A total of 370 ALFA + participants underwent multi-shell DWI. Of these, 73 were excluded for the following reasons: (A) unavailability of CSF amyloid-β and p-tau181 biomarkers (n = 21); (B) evidence of Suspected Non-Alzheimer’s disease Pathophysiology (SNAP, i.e. A − T+), incidental findings, or outliers for p-tau181 values exceeding five standard deviations from mean (n = 16); (C) cropped field of view (FOV) in acquired DWI images for more than one slice (n = 18); and (D) poor quality maps related to image artifacts or algorithm failure (n = 18). Based on this selection, 297 participants were considered in the present study.

Structural MRI acquisition

Data were acquired using a 3T Philips Ingenia CX scanner at the neuroimaging unit at the Barcelonaβeta Brain Research Center using a 32-channel head coil. High-resolution structural images were acquired using a 3D T1-weighted (T1w) Turbo Field Echo pulse sequence (TR/TE/TI = 9.9/4.6/900 ms, flip angle = 8°, FOV = 240 × 240 × 180 mm³, sagittal orientation, voxel size = 0.75 × 0.75 × 0.75 mm³).

DWI acquisition

Multi-shell DWI images were acquired using a multi-slice single-shot Spin Echo-Planar Imaging (EPI) sequence (TR = 5000 ms, FOV = 240 × 240 × 128 mm³, acquired voxel size = 2.0 × 2.0 × 4.0 mm³, reconstructed voxel size = 0.94 × 0.94 × 4.0 mm³). The acquisition was split into three sequences with different b-values to evaluate water diffusion within different compartments of the brain. First sequence: TE = 42 ms, b = 0, 150 s/mm²; second sequence: TE = 60 ms, b = 0, 300, 500, 1000 s/mm²; and third sequence: TE = 84 ms, b = 0, 2000, 4000 s/mm². The total acquisition time was around four minutes and 15 s. Additionally, single-shell DWI images were acquired using a pulse sequence optimized for diffusion tensor imaging (DTI) with the following parameters: 64 unique diffusion-encoding directions (b = 1300 s/mm²), TR = 9000 ms, TE = 90 ms, FOV = 230 × 230 × 145 mm³, and voxel size = 2.1 × 2.1 × 2.2 mm. These single-shell images were used to compute MD maps, which are commonly employed in diffusion MRI studies.

Structural MRI quantifications

Hippocampal volume was measured as a marker of neurodegeneration. We segmented the hippocampus on T1w MRI images using FreeSurfer version 7.1.1. We then constructed a bilateral hippocampal volume by summing up the volumes in the left and right hemispheres. Finally, we employed linear regression analysis to remove the effect of total intracranial volume (TIV), derived from FreeSurfer version 7.1.1, and obtain the TIV-adjusted hippocampal volume.¹⁹

AD signature was also measured as a marker of neurodegeneration. We computed the Jack et al. AD-signature composite¹⁵ from T1w images using FreeSurfer version 7.1.1. This measure is based on the surface-area weighted average of the mean cortical thickness in entorhinal, inferior temporal and middle temporal cortices, and the fusiform gyri.¹⁶

Maps of modulated grey matter (GM) volumes were obtained using the DARTEL implementation for voxel-based morphometry in SPM12. These maps were then smoothed with a 12 mm full-width at half maximum (FWHM) Gaussian kernel.

DWI quantifications

We derived four water diffusion parameters from a three-compartment IVIM DWI model, referred to later as IVIM parameters. This model assumes three intravoxel compartments, as described in Equation (1):

$$S=S_0\left[f{e}^{-TE/T{2}_v}{e}^{D^{*}b}+(1-f)e^{-TE/T{2}_t}{F}_t(D_f,D_s)\right]$$ (1)

The first compartment represents the fast motion of the water pool inside blood vessels and is determined by the information provided by diffusion signal decay at small b-values. The proportion of signal drop due to water movement inside the vessels is represented by the parameter f (perfusion fraction) and the diffusion coefficient for this compartment is represented by D*.

The second and third compartments correspond to the extra- and intracellular spaces, respectively. The intracellular water diffusivity is slower than the extracellular one due to cellular organelles. These two compartments are represented in Equation (1) by $F_t(D_f,D_s)$ that is modelled using Equation (2):

$$F_t(D_f,D_s)=[(1-SSP){\mathrm{e}}^{-D_f\mathrm{b}}+SSP{\mathrm{e}}^{-D_s\mathrm{b}}]$$ (2)

The parameter $D_f$ represents the diffusion coefficient for the signal decay due to the diffusion of extracellular water (fast DC), and $D_s$ represents the diffusion coefficient for the signal decay due to the diffusion of intracellular water (slow DC). The SSP parameter represents the slow signal portion inside the voxel. Finally, in Equation (1), $T{2}_v$ and $T{2}_t$ represent the T2 relaxation parameters in the vascular and tissue compartments.

Diffusion signal analysis

Before modelling, the images from the different acquisitions were registered using affine registration to correct signal distortions and/or patient movement during the acquisitions. A first set of initial values for the model was obtained from the different images acquired at b = 0 s/mm², which were used to estimate T2 values. Later, the different diffusion coefficients were estimated in reverse order, from high b-values to estimate $D_s$, to intermediate b-values to estimate $D_f$. For the perfusion component, only f was estimated using a simplified IVIM model.²⁰

Once the initial values were obtained, a final non-linear fitting was performed with the whole model to ensure better accuracy and signal coherence between different compartments. The whole pipeline was implemented using Interactive Data Language (IDL), producing the four IVIM parameters ($D_s,D_f,SSP,$ and $f$) and the corresponding maps. These maps were normalized to the Montreal Neurological Institute (MNI) space through non-linear registration using the SPM12 software.

Correlation between mean diffusivity (MD) and the three-compartment IVIM parameters

To better understand what the four parameters of our three-compartment IVIM model reflect, we compared them with the parameter MD, derived from the single-shell DWI acquisition. The DWI images were corrected for magnetic field inhomogeneity distortions (TOPUP) and eddy current distortions using the FMRIB Software Library (FSL) software package,²¹ and MD maps were obtained with the FSL tool DTIFIT. The MD maps were then normalized to the Montreal Neurological Institute (MNI) space through non-linear registration using the SPM12 software. The correspondence between MD maps and maps of our three-compartment IVIM parameters was assessed within a mask consisting of GM and white matter (WM) voxels using Pearson correlation.

Statistical analysis

We first sought the association of our four IVIM parameters across voxels in the whole brain with age, sex, and APOE-ε4 in individuals with normal core AD biomarkers (i.e. subjects without accumulation of Aβ and tau proteins, A − T − group). We then examined voxel-wise associations between GM volume maps, obtained using DARTEL, and each of the IVIM parameters across all individuals. All analyses were adjusted for age, sex, and APOE-ε4 status. As these analyses revealed strong associations, GM volume was regressed out from all IVIM parameter maps prior to conducting subsequent statistical analyses in the entire cohort, to account for its potential confounding influence.

Following this adjustment, we examined differences in our IVIM parameters between subjects either with an accumulation of Aβ and not tau proteins (A + T − group), or an accumulation of Aβ and tau proteins (A + T + group), and the A − T − control group. We also examined the relationship of the proposed IVIM parameters with binary amyloid-PET positivity (>12CL) and continuous measures of AD biomarkers (Aβ42/Aβ40 and p-tau181/Aβ40 from the CSF), neurodegeneration biomarkers (CSF NfL, AD signature, and hippocampal volume), and neuroinflammation biomarkers (sTREM2, S100B, YKL-40, IL6, and GFAP from the CSF, and plasma GFAP), in all individuals. We used the p-tau181 values for assessing tau positivity based on the cut-offs proposed by Milà-Alomà et al..²² We normalized all CSF biomarkers to Aβ40 as described by Karlsson²³ and Zetterberg and Bendlin.²⁴

Statistical tests of the IVIM parameters were derived through multiple regression in SPM12 under Matlab environment (R2023a), in which the IVIM parameters were used as dependent variables, and age, sex, APOE status, amyloid-PET positivity, and the continuous measures of AD, neurodegeneration, and neuroinflammation biomarkers were considered separately as independent variables. Age, sex, and APOE-ε4 were used as confounding variables unless they were the variables of interest. Clusters smaller than 50 voxels were excluded (k ≥ 50). We examined for both positive and negative associations at a significance level of 0.001 (uncorrected for multiple comparisons), followed by false discovery rate (FDR) correction at the cluster level (P < 0.05). For the analysis of GM volume maps, voxel-wise multiple regression was conducted using an in-house Matlab script. This script generated t-stat maps for each IVIM parameter, which were subsequently processed in SPM12 for correction of multiple comparisons.

Finally, we investigated the association between GM volume and water diffusion in regions showing amyloid-related increases in GM volume. These regions were identified through multiple regression in two separate analyses: (a) using amyloid-PET positivity, and (b) using CSF Aβ42/40 as the dependent variable. The ratio of CSF Aβ42 measured with Elecsys® to CSF Aβ40 measured with the NeuroToolKit was utilized to identify amyloid-related GM changes to align more closely with the analysis in the study by Salvadó et al..⁷ Age, sex, and total intracranial volume (TIV) were used as confounding variables (P < 0.005 uncorrected; k ≥ 100). We then assessed the association between GM volume and each diffusion parameter using multiple regression models, in which each diffusion parameter was treated as the independent variable, GM volume as the dependent variable, and age, sex, and TIV as confounding variables.

Results

Participant characteristics

A total of 297 participants were included and divided into three groups according to their CSF amyloid-β (A) and CSF p-tau181 tau (T) levels, using cut-offs defined in Milà-Alomà et al.²² (Aβ42/40 ≤ 0.071, p-tau181 > 24 pg/ml): A − T − (n = 192), A + T − (n = 81), and A + T + (n = 24) (Table 1). Of these participants, 151 had available Aβ PET scans. Among them, 121 had Centiloid values below 12 CLs (amyloid-PET negative), while 30 had values above 12 CLs (amyloid-PET positive).

Table 1.

Study participants’ demographics and characteristics of Alzheimer’s disease

	A − T − (n = 192)	A + T − (n = 81)	A + T + (n = 24)	A − T − versus A + T − (P value)	A − T − versus A + T + (P value)	A + T − versus A + T + (P value)
Age, years	60.1 (4.4) [50, 68]	61.2 (4.9) [50, 71]	64.0 (4.7) [52, 73]	0.073	0.001	0.016
Females, N (%)	128 (66.7%)	50 (61.7%)	17 (70.8%)	0.434	0.682	0.415
APOE-ε4 carriers, N (%)	76 (39.6%)	67 (82.7%)	14 (58.3%)	<0.001	<0.001	<0.001
Aβ PET, CL	−3.9 (6.5)	12.6 (17.2)	27.9 (22.3)	<0.001	<0.001	0.029
CSF Aβ42/Aβ40	0.087 (0.009)	0.054 (0.011)	0.045 (0.011)	<0.001	<0.001	0.001
CSF p-tau181, pg/ml	14.2 (4.1)	15.2 (4.0)	30.0 (5.1)	0.079	<0.001	<0.001
TIV-adjusted hippocampal volume, mm3	7617.0 (555.6)	7690.8 (529.8)	7501.4 (766.4)	0.320	0.509	0.298
MMSE, score	29.1 (0.9) [27, 30]	29.3 (0.9) [27, 30]	28.8 (1.2) [27, 30]	0.213	0.199	0.076
PACC, z-score	0.0 (0.7)	0.1 (0.7)	−0.4 (0.7)	0.142	0.023	0.004

Data are expressed as mean (SD) for all measurements, and number (percentage) for the variables ‘females’ and ‘APOE-ε4 carriers’. The ranges for age and MMSE score are indicated in square brackets. The P values were computed with a t-test for all the measurements and χ² test for the variables ‘females’ and ‘APOE-ε4 carriers’.

Aβ: amyloid-beta; A−: normal levels of Aβ proteins; A+: altered levels of Aβ proteins; APOE-ε4: apolipoprotein E epsilon 4; CL: centiloid; MMSE: Mini-Mental State Examination; PACC: Preclinical Alzheimer Cognitive Composite; PET: positron emission tomography; T−: normal levels of tau proteins; T+: accumulation of tau proteins; TIV: total intracranial volume.

The mean age was higher in the A + T + group (64.0 years), compared with the A − T − (60.1 years, P = 0.001) and A + T − (61.2 years, P = 0.016) groups. The majority of the participants were females (n = 195, 65.7%), with no significant differences between groups (Table 1).

The remaining description of the participants’ characteristics is in the Supplemental Material and Table 1.

Association of the IVIM parameters with participant characteristics in the control group

IVIM parametric maps averaged across subjects within the A − T − control group showed that, in the GM, slow DC and perfusion fraction values were higher than in other regions; and, in the WM, fast DC and SSP values were higher than in other regions (Supplementary Fig. 1).

Different participant characteristics correlated with the IVIM parameters in the A − T − group (Fig. 2). Age and female sex presented a strong positive association with slow DC. Female sex also showed a strong positive association with SSP, whereas age showed a strong negative association with it. Moreover, age had a strong positive association with perfusion fraction, whereas female sex had a strong negative association with it. Finally, age was positively associated with fast DC, whereas female sex was negatively associated with it. These associations were mainly in the GM of the cingulum. APOE-ε4 was negatively associated with slow DC in the basal ganglia. Given these observed effects, we used age, sex, and APOE-ε4 as confounding variables in our subsequent analyses that included all the subjects of the study.

Figure 2.

Association of the IVIM parameters with demographic variables and the genetic factor APOE-ε4, in the A − T − control group (n = 192). Regions with significant associations were identified through multiple regression considering each IVIM parameter as a dependent variable and age, female sex, and APOE-ε4 as covariates. The statistical maps present t-stats thresholded at a voxel-level significance of P < 0.001 and corrected for multiple comparisons at the cluster level using FDR (P < 0.05), excluding clusters smaller than 50 voxels. Warm and cold colours indicate positive and negative associations, respectively. The axial slices presented across the columns (left-right) are indicated by lines on the brain surface in the rightmost column (bottom-up). APOE-ε4: apolipoprotein E epsilon 4; DC: diffusion coefficient; IVIM: intravoxel incoherent motion; SSP: slow signal portion.

Association of the IVIM parameters with GM volume, AD and neurodegeneration biomarkers

Voxel-wise analyses revealed distinct associations between GM volume and IVIM parameters (Fig. 3). GM volume was positively associated with slow DC in frontal and temporal cortices, as well as in medial occipital and subcortical regions. SSP also showed positive associations in posterior parietal areas, medial temporal structures, and the cingulate cortex. In contrast, GM volume was negatively associated with fast DC in posterior parietal and occipital regions. Perfusion fraction exhibited a mixed pattern, showing negative associations in medial parietal areas and the cerebellum, and positive associations, though limited to smaller areas, in frontal and lateral temporal cortices.

Figure 3.

Association of the IVIM parameters with GM volume and neurodegeneration biomarkers in all subjects (n = 297). Regions with significant associations were identified through multiple regression considering each IVIM parameter as a dependent variable, biomarkers as covariates, and age, sex, and APOE-ε4 as confounding variables. The statistical maps present t-stats thresholded at a voxel-level significance of P < 0.001 and corrected for multiple comparisons at the cluster level using FDR (P < 0.05), excluding clusters smaller than 50 voxels. Warm and cold colours indicate positive and negative associations, respectively. The axial slices presented across the columns (left-right) are indicated by lines on the brain surface in the rightmost column (bottom-up). Aβ40: amyloid beta 40; AD: Alzheimer’s disease; APOE-ε4: apolipoprotein E epsilon 4; DC: diffusion coefficient; GM: gray matter; IVIM: intravoxel incoherent motion; NfL: neurofilament light; SSP: slow signal portion.

We did not find significant differences in IVIM parameters when comparing the A + T − and A + T + groups to the A − T − control group. Similarly, in the 151 participants with available amyloid-PET Centiloid values, no IVIM parameter showed statistically significant associations with amyloid-PET status. In addition, we did not find any associations of the levels of CSF Aβ42/Aβ40 and p-tau181/Aβ40 with the IVIM parameters when pooling data from all three groups (A − T−, A + T−, and A + T+).

In contrast, the biomarkers of neurodegeneration were associated with several IVIM parameters in different brain regions (Fig. 3). CSF NfL/Aβ40 presented a positive association with fast DC in the posterior insula, and with SSP in the precuneus and postcentral gyrus. Furthermore, it presented a negative association with perfusion fraction in the cingulate gyrus, precuneus, and superior parietal lobe.

The AD signature was positively associated with slow DC in the medial frontal gyrus, insula, inferior frontal gyrus, superior temporal gyrus, and precentral gyrus; and with SSP in the medial frontal gyrus, insula, precentral gyrus, medial temporal gyrus, and fusiform gyrus. The hippocampal volume did not show any associations with any of the IVIM parameters.

Association between the IVIM parameters and neuroinflammation biomarkers

The CSF biomarkers of neuroinflammation were associated with several IVIM parameters in both GM and WM (Figs 4 and 5). sTREM2/Aβ40 was negatively associated with fast DC in the cingulate gyrus and positively associated with SSP in the same region (Fig. 4). Additionally, it showed a negative association with SSP in the superior temporal gyrus. Perfusion fraction was negatively associated with sTREM2/Aβ40 in multiple regions, including the cingulate gyrus, right middle and inferior frontal gyrus, right postcentral gyrus, left precuneus, and cerebellar regions.

Figure 4.

Association between the IVIM parameters and neuroinflammation biomarkers in all subjects (n = 297)—part 1. Regions with significant associations were identified through multiple regression, considering each IVIM parameter as a dependent variable, biomarkers as covariates, and age, sex, and APOE-ε4 as confounding variables. The statistical maps present t-stats thresholded at a voxel-level significance of P < 0.001 and corrected for multiple comparisons at the cluster level using FDR (P < 0.05), excluding clusters smaller than 50 voxels. Warm and cold colours indicate positive and negative associations, respectively. The axial slices presented across the columns (left-right) are indicated by lines on the brain surface in the rightmost column (bottom-up). Aβ40: amyloid beta 40; APOE-ε4: apolipoprotein E epsilon 4; DC: diffusion coefficient; IVIM: intravoxel incoherent motion; SSP: slow signal portion; sTREM2: soluble triggering receptor expressed on myeloid cells 2; S100B: S100 calcium-binding protein; YKL-40: chitinase-3-like protein 1.

Figure 5.

Association between the IVIM parameters and neuroinflammation biomarkers in all subjects (n = 297)—part 2. Regions with significant associations were identified through multiple regression, considering each IVIM parameter as a dependent variable, biomarkers as covariates, and age, sex, and APOE-ε4 as confounding variables. The statistical maps present t-stats thresholded at a voxel-level significance of P < 0.001 and corrected for multiple comparisons at the cluster level using FDR (P < 0.05), excluding clusters smaller than 50 voxels. Warm and cold colours indicate positive and negative associations, respectively. The axial slices presented across the columns (left-right) are indicated by lines on the brain surface in the rightmost column (bottom-up). Aβ40: amyloid beta 40; APOE-ε4: apolipoprotein E epsilon 4; DC: diffusion coefficient; GFAP: glial fibrillary acidic protein; IL-6: interleukin-6; IVIM: intravoxel incoherent motion; SSP: slow signal portion.

S100B/Aβ40 was positively associated with fast DC in the posterior insula and transverse temporal gyrus and negatively associated with fast DC in the cingulate gyrus (Fig. 4). Additionally, it showed a negative association with SSP in the transverse temporal and superior temporal gyri. Perfusion fraction was negatively associated with S100B/Aβ40 in the cingulate gyrus, precuneus, and cerebellar regions.

YKL-40/Aβ40 was negatively associated with fast DC in the uncus (Fig. 4). Additionally, it showed a negative association with perfusion fraction in the mid frontal gyrus, paracentral lobule, postcentral gyrus, precuneus, and cingulate gyrus.

IL-6/Aβ40 was positively associated with slow DC in the right frontal WM adjacent to the anterior cingulate cortex, and in the left medial frontoparietal WM (Fig. 5). Further, it showed a negative association with SSP in the right medial frontoparietal WM, as well as in the periventricular WM between the posterior insula and caudate.

GFAP/Aβ40 was negatively associated with SSP in the periventricular white matter, and with perfusion fraction in the cingulate gyrus and cerebellar regions (Fig. 5). Plasma GFAP was positively associated with fast DC in the inferior parietal lobe, supramarginal gyrus, and mid occipital and temporal gyri (Fig. 5). Further, it showed a negative association with SSP in the inferior parietal lobe, in the right mid frontal gyrus, right postcentral gyrus, and the left angular gyrus. Finally, plasma GFAP was negatively associated with perfusion fraction in the right posterior cingulate cortex and right posterior insula.

Association between the IVIM and MD parameter maps

We also studied the correlation between the MD map and the four IVIM parameter maps in all subjects and observed that the MD map negatively correlated with slow DC, fast DC, and SSP maps, whereas it positively correlated with the perfusion fraction map (P < 0.001) (Supplementary Fig. 2).

Association of GM volume with amyloid-PET positivity, CSF Aβ42/40, IVIM parameters, and MD

GM volume positively correlated with amyloid-PET positivity in the parahippocampus, fusiform gyrus, isthmus of the cingulate gyrus, paracentral lobule, and right occipital lobe (Supplementary Fig. 3). Moreover, the average GM volume across regions associated with amyloid-PET positivity showed a negative correlation with slow DC (P = 0.001), perfusion fraction (P = 0.036), and MD (P = 0.047) (Fig. 6). GM volume was also negatively associated with CSF Aβ42/40 levels in the parahippocampus, fusiform gyrus, isthmus of the cingulate gyrus and right occipital lobe, and rostral anterior cingulate and precuneus (Supplementary Fig. 4). Further, the average GM volume across regions associated with reduced CSF Aβ42/40 showed negative correlations with slow DC (P = 0.007), perfusion fraction (P = 0.027), and MD (P < 0.001) (Supplementary Fig. 5).

Figure 6.

Association between GM volume averaged across regions where GM volume was positively associated with amyloid-PET positivity and the diffusion parameters in all subjects (n = 297). Significant associations between GM volume and the diffusion parameters derived from the IVIM model, namely (A) slow DC, (B) fast DC, (C) SSP, (D) perfusion fraction, as well as (E) the MD diffusion parameter, were identified through multiple regression, considering each diffusion parameter as the independent variable, GM volume as the dependent variable, and age, sex, and TIV as confounding variables. Each data point represents a single participant. Interaction terms between diffusion parameters and AT stage were also tested in the regression model, but were not statistically significant. a.u.: arbitrary units; DC: diffusion coefficient; GM: gray matter; intravoxel incoherent motion; MD: mean diffusivity; SSP: slow signal portion.

Discussion

In this study, we derived a three-compartment IVIM model with four parameters to measure distinct pools of cerebral water diffusion in subjects with preclinical AD (amyloid-positive cognitively unimpaired individuals) and healthy individuals. Our results show that, in brain areas showing amyloid accumulation, the observed higher GM volumes were associated with lower slow DC, a marker of intracellular water diffusion. This finding suggests a potential link between amyloid accumulation and altered microstructural properties indicative of intracellular water retention or cell swelling. Such changes may reflect neuroinflammatory processes, possibly driven by astroglial reactivity, although further studies are needed to directly assess this relationship. These results warrant the use of IVIM parameters to study the aetiology of the brain volume reductions associated with amyloid-clearance in anti-amyloid immunotherapies, showing a clinical benefit for the participants.

Cerebral accumulation of amyloid beta in the early preclinical stages of AD is associated with a GM volume increase in brain regions that typically undergo neurodegeneration and volume decrease in later symptomatic stages; however, the underlying mechanisms remain controversial.²⁵ Understanding these mechanisms could shed light on why clinical trials with anti-amyloid drugs with beneficial clinical effects have found a decrease in brain volume in early symptomatic AD.^9,10 The use of multi-compartment models for studying cerebral water diffusion in AD has been limited,^26,27 despite their success in distinguishing between different water pools in individuals with various neurological issues^28-31 Particularly, three-compartment IVIM models^32,33 are believed to better account for the existence of distinct water pools in biological tissues compared to two-compartment models. They have been shown to provide better fit to the data across various organs, including the brain,³⁴ offering richer information and potentially more accurate measures of diffusion-related parameters. Considering this, we developed a three-compartment IVIM model with four parameters to measure cerebral water diffusion in subjects with preclinical AD and healthy individuals. Specifically, we derived two parameters reflecting the water diffusion rate in the intracellular and extracellular spaces, and two additional parameters representing the relative signal contributions from the intracellular water and the microvasculature.

Firstly, we characterized the IVIM parameters against demographic factors in the healthy control group. We found that age and sex were associated with all the IVIM parameters, while APOE status was associated only with slow DC, underscoring the importance of accounting for these factors as confounding variables in the subsequent statistical analyses. Notably, we observed an association of age with an increase in perfusion fraction in the cortical region, which aligns with findings reported by Vieni and colleagues (2020).³⁵ Moreover, we observed widespread positive associations of female sex with SSP and a negative association with perfusion fraction. These patterns were spatially similar to those observed for age, albeit with reversed polarity. Importantly, sex effects were independent of age, as confirmed by matched age distributions and statistical adjustment. Our findings are consistent with previous studies showing that MD increases with age and is higher in males than in females.³⁶ Note, in our data, MD was positively correlated with perfusion fraction and negatively with SSP, reinforcing the interpretation that MD may reflect mainly extracellular and vascular diffusivity. The increased SSP and decreased perfusion fraction in females may reflect greater cell density in grey matter, supporting the notion that sex-related differences in cellular composition influence water diffusion characteristics captured by IVIM parameters.

Furthermore, in the full cohort analysis, we observed robust regional associations between GM volume and the four IVIM parameters. These associations mirrored the trends previously identified in females, namely, positive correlations with slow DC and SSP, and negative correlations with fast DC and perfusion fraction, though the spatial distribution of these effects was not entirely overlapping. One potential explanation is that IVIM quantification may be influenced by CSF contamination at the voxel level, which can reduce the contribution of the GM signal. Alternatively, these patterns may reflect individual variability in regional GM volume beyond sex-related differences. Individuals with greater GM volume may exhibit a higher proportion of diffusion signal originating from intracellular water, characterized by elevated diffusion rates within these compartments.

We also identified connections between biomarkers of neurodegeneration and the IVIM parameters in cognitively unimpaired subjects, either with preclinical AD or healthy. In particular, NfL/Aβ40 and AD signature were associated with SSP in various GM regions, including frontal, parietal, and temporal areas. This supports previous research on the association of plasma NfL³⁷ and cortical thickness³⁸ with water diffusion in the parietal and temporal regions. However, we did not observe any associations between hippocampal volume and IVIM parameters, which contrasts with recent studies reporting links between hippocampal volume and the intermediate component fraction $f_{int}$, a diffusion parameter thought to reflect the volume fraction of the interstitial fluid compartment.³²

Moreover, we observed an association between fluid biomarkers of neuroinflammation related to AD and our IVIM parameters. Similarly, Spotorno et al.³⁷ found a positive association between two of the fluid biomarkers of inflammation that we analysed (CSF YKL-40 and plasma GFAP) and water diffusion. Additionally, another study found that two of the CSF biomarkers of neuroinflammation that we examined (YKL-40 and sTREM2) can predict cerebral volumetric changes over time in cognitively unimpaired individuals.⁶ Taken together, our findings and these previous studies suggest that neuroinflammation is associated with cerebral water diffusion and brain volumetric changes.

Building on this, we identified several recurring patterns across neuroinflammatory biomarkers in relation to IVIM parameters. Most notably, perfusion fraction and SSP showed consistent negative associations, although these effects did not always occur in the same regions. The widespread negative association of perfusion fraction suggests that neuroinflammation may reduce microvascular water contribution, potentially reflecting vascular compression or impaired perfusion. In parallel, reductions in SSP with astrocytic markers (e.g. CSF S100B/Aβ40, CSF GFAP/Aβ40, plasma GFAP) indicate a lower intracellular water signal when astrocytic activity is high, possibly due to increased extracellular space or altered cellular morphology.

Regarding fast DC, we observed negative associations in the cingulate cortex for both sTREM2/Aβ40 and S100B/Aβ40. Interestingly, these two biomarkers also showed negative associations with perfusion fraction in the same regions. This pattern suggests that both astrocytic and microglial activation are linked not only to impaired perfusion but also to decreased extracellular diffusion rates, which may reflect increased complexity or crowding in the extracellular space. Finally, while the overall direction of associations was largely consistent across biomarkers, spatial patterns varied. For example, SSP reductions were observed in GM for CSF sTREM2/Aβ40 and CSF S100B/Aβ40, whereas for CSF IL6/Aβ40, CSF GFAP/Aβ40, and plasma GFAP, decreases were primarily localized to GM. Further studies are needed to clarify the biological mechanisms underlying these spatial differences.

Importantly, we did not observe significant differences in IVIM parameters when comparing CSF-based AT groups (A − T − versus A + T − and A + T+) or amyloid-PET status. We also conducted post-hoc subgroup analyses to examine associations between IVIM parameters and CSF Aβ42/40 within each AT group separately; however, no significant associations were identified. Similarly, no associations were found between IVIM parameters and continuous measures of CSF Aβ42/40 or p-tau181/Aβ40, suggesting that IVIM metrics may be more sensitive to neuroinflammatory processes than to amyloid or tau pathology in cognitively unimpaired individuals. These findings contrast with those of Montal et al. (2018), who reported a biphasic trajectory of MD changes along the AD continuum, with decreased MD in early preclinical stages associated with amyloid deposition, and increased MD in later stages reflecting neurodegeneration. The lack of IVIM sensitivity to amyloid in our cohort may be due to differences in AT staging methodology or cohort characteristics between the two studies.

Interestingly, our voxel-wise analysis revealed widespread positive associations between GM volume and slow DC, suggesting that higher GM volume is generally linked to increased intracellular water diffusivity. This pattern may reflect healthier tissue microstructure or reduced restriction within the intracellular compartment. However, when focusing specifically on regions where GM volume was associated with amyloid-PET positivity, we observed a negative association between GM volume and slow DC. Notably, SSP, reflecting the relative contribution of intracellular water to the diffusion signal, also showed a negative association with GM volume in these regions, albeit not statistically significant. These concurrent reductions in slow DC and SSP suggest that these changes are unlikely to reflect intracellular water accumulation. Instead, they may be driven by mechanisms such as increased intracellular complexity, including glial and cytoskeletal remodelling or organelle crowding, that restrict water mobility and reduce the signal contribution from the intracellular compartment.

Given previous findings linking markers of astroglial activity (e.g. CSF YKL-40 and plasma GFAP) to both cerebral water diffusion and volumetric changes,^6,37 our results suggest that neuroinflammation may contribute to microstructural alterations underlying GM volume increases. These changes may involve increased cellular complexity or glial remodelling that restricts water mobility. This hypothesis is particularly relevant considering that brain volume reductions are often observed following anti-amyloid therapies, which also lead to decreases in plasma GFAP levels.^9,39 Although the precise mechanisms remain to be elucidated, these findings support the use of multi-compartment diffusion imaging to investigate the mechanisms underlying the volumetric decreases observed after clinically beneficial amyloid removal.

Building on these findings, we observed that slow DC demonstrated a stronger association with GM volume than MD, highlighting the added sensitivity of multi-compartment diffusion imaging to microstructural changes. While MD captures average diffusivity across both intra- and extracellular compartments, slow DC specifically reflects intracellular water mobility. This distinction is critical in the context of preclinical AD, where subtle intracellular changes may precede apparent neurodegeneration. By disentangling distinct diffusion components, multi-compartment models offer enhanced specificity and may outperform conventional DWI metrics in detecting early pathological processes.

Our study has some limitations. First, as a cross-sectional study, we did not investigate if DWI can track changes in water pools over time¹⁴; however, this study paves the way for the use of multi-compartmental DWI to better understand the effects of anti-amyloid drugs on brain structure. Second, our cohort, composed of healthy individuals and subjects in the early pre-clinical stage of AD, may explain the absence of association of IVIM parameters with either CSF or PET amyloid biomarkers. Further studies at later stages of the AD continuum are required to uncover the true extent of these associations. Third, given the exploratory nature of the study and limited statistical power in terms of sample size, we did not apply a correction for multiple comparisons across biomarker tests, which may increase the risk of false-positive findings. Therefore, associations reported between IVIM parameters and fluid biomarkers should be interpreted with caution. Future studies with larger sample sizes are needed to apply more stringent correction methods and validate these findings.

Additionally, we did not apply denoising or eddy-current correction to the multi-shell data due to the very low number of diffusion directions, which makes standard correction methods unsuitable or potentially unstable. Advanced denoising techniques such as MPPCA⁴⁰ require high directional redundancy, which our protocol does not provide. Furthermore, susceptibility distortion correction was not performed due to the absence of reversed phase-encoding images. This limitation may particularly affect IVIM estimates in regions near air–tissue interfaces, such as the orbitofrontal and prefrontal cortices, where magnetic susceptibility differences from nasal sinuses can cause local geometric warping. Such distortions may lead to misalignment with anatomical templates and partial volume effects. While affine registration mitigates global misalignment, it cannot fully correct these local distortions. Future work should incorporate optimized acquisition strategies, including reversed phase-encoding images, to minimize potential bias in IVIM parameter estimation.

Lastly, our modelling approach was based on a three-compartment IVIM framework, which, while clinically feasible, does not fully adhere to current principles of biophysical modelling of diffusion MRI. Advanced models leveraging multidimensional diffusion-relaxation encoding (Lampinen et al., 2023; Novikov et al., 2019) can provide richer microstructural information and reduce parameter bias. Future studies should explore these approaches to validate and extend our findings.

Despite these limitations, our study has various strengths. We examined a cohort much larger than those in previous studies,^26,32 which was well-characterised with a range of fluid and MRI biomarkers. Moreover, we developed and utilized a new three-compartment DWI model that distinguishes between intracellular and extracellular water diffusion, providing a more accurate characterisation of cerebral water pools.

Multi-compartment diffusion-weighted imaging enables the measurement of cerebral water IVIM parameters associated with CSF biomarkers of neuroinflammation and cerebral volume alterations. This method may help disentangle the aetiology of cerebral volume reductions in individuals receiving anti-amyloid therapies in early symptomatic AD. Further studies are needed to better understand whether multi-compartment DWI can distinguish between different stages of AD and other neurodegenerative diseases.

Appendix 1

Collaborators of the ALFA study are: Müge Akinci, Federica Anastasi, Annabella Beteta, Raffaele Cacciaglia, Lidia Canals, Alba Cañas, Carme Deulofeu, Maria Emilio, Irene Cumplido-Mayoral, Marta del Campo, Carme Deulofeu, Ruth Dominguez, Maria Emilio, Sherezade Fuentes, Marina García, Laura Hernández, Gema Huesa, Laura Iglesias, Esther Jiménez, David López-Martos, Paula Marne, Tania Menchón, Paula Ortiz-Romero, Eleni Palpatzis, Wiesje Pelkmans, Albina Polo, Sandra Pradas, Mahnaz Shekari, Lluís Solsona, Anna Soteras, Núria Tort-Colet, Marc Vilanova and Natalia Vilor Tejedor.

Supplementary Material

Supplementary material is available at Brain Communications online.

Funding

The research leading to these results has received funding from 'la Caixa' Foundation (ID 100010434), under agreement LCF/PR/GN17/50300004, the Health Department of the Catalan Government (Health Research and Innovation Strategic Plan (PERIS) 2016–2020 grant# SLT002/16/00201), and the Alzheimer's Association and an international anonymous charity foundation through the TriBEKa Imaging Platform project (TriBEKa-17-519007). Additional support has been received from the Universities and Research Secretariat, Ministry of Business and Knowledge of the Catalan Government under the grant no. 2021 SGR 00913. Henrik Zetterberg is a Wallenberg Scholar and a Distinguished Professor at the Swedish Research Council supported by grants from the Swedish Research Council (#2023-00356; #2022-01018 and #2019-02397), the European Union’s Horizon Europe research and innovation programme under grant agreement No 101053962, Swedish State Support for Clinical Research (#ALFGBG-71320), the Alzheimer Drug Discovery Foundation (ADDF), USA (#201809-2016862), the AD Strategic Fund and the Alzheimer's Association (#ADSF-21-831376-C, #ADSF-21-831381-C, #ADSF-21-831377-C, and #ADSF-24-1284328-C), the European Partnership on Metrology, co-financed from the European Union’s Horizon Europe Research and Innovation Programme and by the Participating States (NEuroBioStand, #22HLT07), the Bluefield Project, Cure Alzheimer’s Fund, the Olav Thon Foundation, the Erling-Persson Family Foundation, Familjen Rönströms Stiftelse, Stiftelsen för Gamla Tjänarinnor, Hjärnfonden, Sweden (#FO2022-0270), the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 860197 (MIRIADE), the European Union Joint Programme—Neurodegenerative Disease Research (JPND2021-00694), the National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre, and the UK Dementia Research Institute at UCL (UKDRI-1003). Oriol Grau Rivera receives funding from the Alzheimer’s Association Research Fellowship Program (2019-AARF-644568) and is supported by the Spanish Ministry of Science, Innovation and Universities (IJC2020-043417-I). Kaj Blennow is supported by the Swedish Research Council (#2017-00915 and #2022-00732), the Swedish Alzheimer Foundation (#AF-930351, #AF-939721, #AF-968270, and #AF-994551), Hjärnfonden, Sweden (#FO2017-0243 and #ALZ2022-0006), the Swedish state under the agreement between the Swedish government and the County Councils, the ALF-agreement (#ALFGBG-715986 and #ALFGBG-965240), the European Union Joint Program for Neurodegenerative Disorders (JPND2019-466-236), the Alzheimer’s Association 2021 Zenith Award (ZEN-21-848495), the Alzheimer’s Association 2022-2025 Grant (SG-23-1038904 QC), La Fondation Recherche Alzheimer (FRA), Paris, France, the Kirsten and Freddy Johansen Foundation, Copenhagen, Denmark, and Familjen Rönströms Stiftelse, Stockholm, Sweden. Marc Suárez-Calvet receives funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (Grant agreement No. 948677) and the Instituto de Salud Carlos III through the projects PI19/00155 and PI22/00456 (Co-funded by European Regional Development Fund, FEDER, ‘A way to make Europe’). Moreover, Marc Suárez-Calvet receives the support of a fellowship from ‘la Caixa’ Foundation (ID 100010434) and from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No 847648 (fellowship code LCF/BQ/PR21/11840004).

Competing interests

Gwendlyn Kollmorgen is a full-time employee of Roche Diagnostics GmbH, Penzberg, Germany. Clara Quijano-Rubio is a full-time employee of Roche Diagnostics International Ltd, Rotkreuz, Switzerland. Henrik Zetterberg has served at scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZPath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, LabCorp, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, and Wave, has given lectures sponsored by Alzecure, BioArctic, Biogen, Cellectricon, Fujirebio, Lilly, Novo Nordisk, Roche, and WebMD, and is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program (outside submitted work). Kaj Blennow has served as a consultant and on advisory boards for AbbVie, AC Immune, ALZPath, AriBio, BioArctic, Biogen, Eisai, Lilly, and Moleac Pte. Ltd, Neurimmune, Novartis, Ono Pharma, Prothena, Roche Diagnostics, Sanofi and Siemens Healthineers; has served at data monitoring committees for Julius Clinical and Novartis; has given lectures, produced educational materials and participated in educational programs for AC Immune, Biogen, Celdara Medical, Eisai and Roche Diagnostics; and is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program, outside the work presented in this paper. Marc Suárez-Calvet has received in the past 36 months consultancy/speaker fees (paid to the institution) from Almirall, Eli Lilly, Novo Nordisk, and Roche Diagnostics. He has received consultancy fees or served on advisory boards (paid to the institution) of Eli Lilly, Grifols and Roche Diagnostics. He was granted a project and is a site investigator of a clinical trial (funded by the institution) by Roche Diagnostics. In-kind support for research (to the institution) was received from ADx Neurosciences, Alamar Biosciences, ALZPath, Avid Radiopharmaceuticals, Eli Lilly, Fujirebio, Janssen Research & Development, Meso Scale Discovery, and Roche Diagnostics; Marc Suárez-Calvet did not receive any personal compensation from these organisations or any other for-profit organisation.

Data availability

The data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. Code used for the analyses is available at the following GitHub repository: https://github.com/mkassinopoulos/2026_multib_paper_AlfaPLUS.