Longitudinal diffusion tensor imaging correlates with amyloid burden in Down syndrome

Abstract

INTRODUCTION

Adults with Down syndrome (DS) accumulate amyloid beta (Aβ) plaques faster and earlier on average than neurotypical adults with sporadic Alzheimer's disease (AD). White matter (WM) microstructure characterized with diffusion tensor imaging (DTI) can indicate underlying architectural changes in longitudinal studies, suggestive of neurodegeneration. This study investigated relationships between DTI and Aβ in DS along the AD continuum.

METHODS

Using longitudinal amyloid Pittsburgh compound B positron emission tomography, Centiloid (CL) and DTI parameters were examined in 35 adults with DS ages 25 to 57. DTI measures of anisotropy and diffusivity were analyzed using tract‐based spatial statistics and permutation analysis of linear models, testing for significant correlation between the rates of change for CL and DTI.

RESULTS

All rates of DTI and Aβ changes were significantly related. Significant regions included the corpus callosum, corona radiata, and long‐association fibers.

DISCUSSION

Aβ burden is associated with widespread longitudinal WM changes in DS. This suggests WM microstructure alterations accompany amyloid accumulation.

Highlights

• A Down syndrome–specific template was created.

• Longitudinal diffusion tensor imaging (DTI) and amyloid burden rates of change correlate.

• Longitudinal results show more significant regions than cross‐sectional results.

• DTI and amyloid changes were found over two timepoints, 3.7 years apart on average.

• DTI and amyloid‐PET offer greater sensitivity when tracking microstructural changes.

1. BACKGROUND

Individuals with Down syndrome (DS) have a higher risk and earlier age of onset for Alzheimer's disease (AD) than the neurotypical (NT) population. ¹ Indeed, individuals with DS have an estimated 95% lifetime risk of AD due to genetic predisposition. ² , ³ Chromosome 21 contains the gene encoding for amyloid precursor protein (APP); when triplicated, the APP gene leads to a 50% increase in amyloid beta (Aβ) plaque production and accumulation. ⁴ Therefore, informed methods for AD early intervention, disease tracking, and treatment are crucial for the DS population. Longitudinal positron emission tomography (PET) scans reveal brain regions with elevated Aβ and neurofibrillary tau tangles, ⁵ indicating possible neurodegeneration in adults with DS beginning in the late thirties and early forties. These regions of neurotoxic protein aggregates are prime targets for quantitative magnetic resonance imaging (qMRI), which is well suited to detect structural and microstructural changes, advance understanding of AD pathology in DS, and inform clinical trials.

Diffusion MRI (dMRI), particularly diffusion tensor imaging (DTI), is a tool that probes water diffusion to detect microstructural changes. ⁶ , ⁷ , ⁸ DTI parameters model directional water displacements in the brain. ⁹ Typically, microenvironments with higher cellular density, like axon bundles, exhibit greater restriction and anisotropic diffusion, while free water regions, like cerebrospinal fluid (CSF), exhibit isotropic diffusion. Key DTI metrics include mean diffusivity (MD) and fractional anisotropy (FA), which represent average diffusion and diffusion anisotropy, respectively. ¹⁰ Although DTI is sensitive to neural changes, it lacks specificity to a single neuropathology and appears in various neurodegenerative and cerebrovascular diseases. Thus, its role as an AD biomarker has not yet been fully characterized. ¹¹ Investigating the relationship between DTI and amyloid PET may fill a knowledge gap for tracking microstructural changes in neurodegeneration, neuroinflammation, and demyelination. ¹²

Specific to the adult DS population, post mortem studies of brain tissue reveal more severe white matter (WM) degeneration, Aβ fibrils, and plaque lesions compared to NT individuals, ¹³ particularly in the frontal cortex. ⁴ These changes are speculated to alter water diffusion, reflecting increased extra‐axonal space, and serve as a potential indicator of reduced axonal density or neuroinflammation. ¹⁴ , ¹⁵ DTI is well suited to correlate with global PET Aβ burden because Aβ contributes to fiber degradation. ¹⁶ , ¹⁷ Longitudinal studies should better highlight pathological interactions between rapid Aβ burden and fiber degradation in DS. To our knowledge, this is the first longitudinal DTI study in DS and the first to examine the relationship between DTI outcomes and longitudinal Aβ. ¹⁸ , ¹⁹

Among DTI and Aβ cross‐sectional studies in DS, increased Aβ correlates with widespread decreased FA, including in the genu of the corpus callosum, superior and inferior longitudinal fasiculi, ²⁰ cingulum‐cingulate gyrus, ²¹ and fornix. ²⁰ , ²¹ It also correlates with increased MD across WM association pathways. ²⁰

Several investigations of longitudinal DTI and Aβ in sporadic AD have also been reported. ²² , ²³ , ²⁴ , ²⁵ , ²⁶ With increased Aβ burden, studies have shown increased axial and radial diffusivities (AxD, RD) in the left hippocampus; ²⁷ decreased restricted diffusion among Aβ‐negative subgroups; ²⁸ decreased fornix FA accelerating episodic memory decline; ²⁹ and stronger associations with hippocampal cingulum diffusivity, tau, and memory decline. ³⁰

With the near‐complete prevalence of elevated Aβ pathology and accelerated pace of AD‐related neuropathological changes, the DS population offers a unique opportunity to inform the DTI–Aβ relationship. ³¹ The current study examines within‐individual change in DTI and Aβ longitudinally, drawing on two data collection cycles, while controlling for between‐subject and age‐related variability. Normal aging can induce changes in both DTI measures and Aβ burden. Further, there is also between‐subject heterogeneity of DTI measures, which may obscure the ability to detect relationships with Aβ burden. By leveraging the rates of change in these measures, the confounds of age‐related effects and population heterogeneity are mitigated. This approach focuses on the relationships of intra‐individual changes as the primary objective. The driving hypothesis is that DTI metrics are associated with global Aβ burden, and their rates of change reflect WM neurodegeneration in DS. By addressing age and population variability, longitudinal DTI may emerge as a valuable, non‐invasive biomarker complementing current AD diagnostic methods.

2. METHODS

2.1. Participants

Participant data were from the Alzheimer Biomarkers Consortium–‐Down Syndrome (ABC‐DS), a longitudinal study of adults with DS to obtain and assess early biomarkers of potential AD. ³² Data from 47 participants at the University of Wisconsin–Madison (UW) site were used to focus on a single site to eliminate variance from the effects of scanner, radiofrequency coil, and protocol.

Sample sizes for different analyses and steps varied based on image data availability between the sub‐studies and are described in the following section. Overall, participants were adults with DS, ranging in age from 25.6 to 57.3 years. This study received institutional review board approval, and each participant, or their legally authorized representative, provided informed consent. ³³ Individuals with DS had to be at least 25 years of age, have confirmed DS via genetic testing, have a mental age of at least 3 years based on the Stanford‐Binet Intelligence Scale, and not have untreated medical or psychiatric conditions or any MRI incompatibilities (e.g., metal implants). ³¹ , ³⁴ Cognitive status was determined following the AAMR‐IASSID Working Group for the Establishment of Criteria for Diagnosis of Dementia in Individuals with Intellectual Disability and assigned to one of four categories: cognitively stable (CS), mild cognitive impairment (MCI‐DS), having dementia, or undetermined. ³³ , ³⁵

2.2. Imaging

MRI and amyloid PET images were acquired at month 0 as a baseline, and then again at month 32. ³³ These images were obtained from the online Laboratory of NeuroImaging image repository, which underwent additional quality control (QC) and preprocessing steps.

RESEARCH IN CONTEXT

1. Systematic review: To our knowledge, no longitudinal diffusion tensor imaging (DTI) studies have been performed in adults with Down syndrome (DS), a population that exhibits early‐onset Alzheimer's disease (AD) neuropathology and rapid amyloid beta (Aβ) plaque accumulation. In sporadic AD, longitudinal DTI, which is sensitive to microstructural changes, has shown associations between lower restricted diffusion, lower cognition, and greater Aβ. To improve the sensitivity of DTI measures, this study evaluated the relationship between rates of DTI change and amyloid positron emission tomography.

2. Interpretation: Findings indicate widespread associations between higher anisotropies and lower diffusivities and higher Aβ. Regions included bilateral long‐association fibers, corona radiata, thalamic radiations, and the corpus callosum. Significant relationships were observed for measures separated by 3.7 years on average.

3. Future directions: Longitudinal DTI, in complement with other AD biomarkers, can provide insight into changes in brain microstructure. For greater specificity, future assessments should consider the effects of tau and other diffusion models.

2.2.1. PET imaging

Amyloid PET used an intravenous injection of [C‐11]Pittsburgh compound B (PiB) to measure Aβ burden with a target injected dose of 15mCi. PET scans used to measure Aβ were acquired 50 to 70 minutes post‐injection. All scans were done on a Siemens ECAT HR+ scanner for this cohort. Motion correction and 3D image generation were done with Statistic Parametric Mapping 12 (SPM12) software by re‐aligning and averaging image frames. All PET images used in this study had two separate PET scans to provide longitudinal Aβ information. Based on limitations from MRI image exclusion described below, this left a total of 44 PET scans in the cross‐sectional analysis with coinciding acceptable quality MRI, and 70 in the longitudinal analysis (35 per session).

2.2.2. MRI imaging

T1‐weighted (T1w) and diffusion‐weighted imaging (DWI) MRIs were acquired with a 3‐Tesla GE Signa 750. T1w magnetization‐prepared rapid gradient echo images were collected at 1.1 mm × 1.1 mm × 1.2 mm resolution with a repetition time (TR) of 7300 ms, and an echo time (TE) of 3.0 ms. DWIs were collected with an echo planar imaging (EPI) sequence at 2.0 mm³ isotropic resolution with a TR of 7800 ms and a TE of 60.4 ms. b values, or the amount of diffusion weighting, were collected at 0 and 1000 s/mm² using 6 and 48 directions, respectively. Most baseline scan DWIs were collected with one phase‐encoding polarity, while most 32 month DWIs were collected with two reverse phase‐encoding polarities to aid susceptibility distortion correction needed for EPI. With two reverse phase‐encoding polarities, b values were at 0 and 1000 s/mm² using 6 and 48 respective directions in one polarity, and 6 and 6 respective directions in the other polarity.

MRI images were excluded in cases of severe artifacts, incomplete brain coverage, and/or failures in image processing. Images were visually inspected by neuroimaging experts for major motion artifacts, signal loss, and/or excessive distortion. In addition, significant failures in FreeSurfer ³⁶ segmentation of T1w images were a problem in some cases. DWIs should represent the underlying anatomy, so incomplete brain coverage, frequent motion artifacts, and signal loss were used to exclude cases.

2.3. PET image processing for Aβ quantification

PET processing and quantitative values were obtained following previously described methods, ³¹ separate from DWI analysis. Briefly, after obtaining an averaged static image of 50 to 70 minutes, images were spatially normalized to the Montreal Neurological Institute 152 (MNI152) space with a DS‐specific PiB PET template ³⁷ in SPM12. Standardized uptake value ratio (SUVR) images were derived using voxel normalization to the gray matter cerebellum reference region on the summed PiB PET images. This essentially provided a map of voxels containing Aβ that can be used to calculate the global Aβ burden for each participant. This global burden value was represented with Aβ load (Aβ_L) ³⁸ per subject, which was then converted to equivalent CL (SUVR/Centiloids) with a previously derived linear relationship, ³⁹ because CLs are more commonly used. Global, brain‐wide CL values were used in subsequent analyses. For further description, see Zammit et al. ⁴⁰

2.4. MRI image processing for DTI quantification

T1w images were corrected for intensity variation (N4BiasFieldCorrection), upsampled to 1.0 mm isotropic resolution, and processed using FreeSurfer v7.1.1. DWI images were processed using a local custom pipeline, QMRI‐neuropipe, ⁴¹ consisting of tools from FSL and MRTrix3 for corrections for image noise, Gibbs ringing, EPI distortion (either FSL top‐up or Synb0‐DisCo ⁴² when only one phase encoding available), and eddy currents and motion (FSL eddy).

Preprocessed DWIs were co‐registered to the respective T1w image and upsampled to 1.0 mm voxels using boundary‐based registration (BBR). ⁴³ The co‐registered DWIs were fit to a DTI model using the DiPy python package ⁴⁴ for quantitative DTI maps, including FA, MD, RD, and AxD. Voxelwise DTI values were obtained separately from global CL values.

2.5. Creation of the phenotypic‐specific DS template

As with most populations, there is considerable anatomical variation in DS. Further, current literature suggests that, compared to NT cohorts, individuals with DS tend to have larger ventricles, smaller cerebellums, brainstems, and pontes, and smaller overall brain sizes and weights. ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ Thus, it is important to perform brain image analyses of DS populations using a brain template space that is representative of DS. This study initially investigated the alignment to the MNI152 space; however, the morphology was no longer preserved to a noticeable degree. For example, the cerebellum was stretched further out compared to alignment with the DS‐specific templates, which may increase uncertainty in DTI parameter values and spatial location. Similar findings have been observed in older adults with significant brain atrophy that experience overexpansion during registration to atlases made from younger cohorts, ⁵¹ highlighting the need for population‐based templates in studies of heterogeneous or atypical morphology. ⁵² Furthermore, disease‐specific templates offer significantly better mapping ⁵³ and benefit in voxel‐based analyses, ⁵⁴ especially in populations with motion, ³⁷ motivating the choice for DS‐specific templates.

DS‐specific templates were made using MRI data only for this study. The T1w‐diffusion fused (TiDi‐Fused) framework ⁵⁵ was used, consisting of co‐registration of individual DWI and their respective T1w images using BBR, which takes advantage of segmented tissue boundaries between WM, gray matter, and CSF from FreeSurfer. ⁴³ The final transformation is then applied to the DWIs for each participant.

Next, a T1w template for cross‐sectional analyses was created using ANTs with non‐linear transformations. The non‐linear deformable process was repeated 10 times. ⁵⁶ This model has been shown to perform well for human brain registration compared to other transformation models, ⁵⁷ and has even been shown to be the best option for spatially normalizing DS MRIs. ⁵⁸

Finally, the ANTs‐derived final transforms were applied to individual participant DTI maps to warp them to the DS‐specific T1w template space using ANTs cubic B‐spline interpolation up‐sampled to the T1w image's resolution. DWI encoding directions were transformed with the rigid body transform's rotational component. ⁵⁵ These maps were averaged over all participants for each DTI parameter to generate robust population‐based DTI templates with enhanced spatial resolution. They were also AC‐PC (anterior commissure–posterior commissure) aligned, as is typically used in PET studies, using open‐source AFNI tools, ⁵⁹ and segmented with FreeSurfer for regional analyses to create a usable, pseudo‐atlas without manual delineation.

A separate template was created for the longitudinal analyses. First, within‐individual alignment of the T1w and DWI was performed for each longitudinal measurement pair to create individual midpoint templates. These midpoint templates were made using ANTs ⁶⁰ with the same options and technique as was done for the DS‐specific template creation, but using the “unbiased_pairwise_registration” script. ⁵¹ A new DS population–specific midpoint template was made from all individual midpoint templates using the same methodology described above. Individual images were transformed to this space for subsequent analyses. For further clarification and concise steps for the longitudinal DS‐specific template creation, see Figure 1.

FIGURE 1.

A flowchart with corresponding images to visually describing the creation of the longitudinal DS‐specific templates. LEFT, Template creation steps are written out, with arrows showing which components are transformed to a different space or incorporated. Colors represent which space the image is in (i.e., native, template). RIGHT, A visual representation of the key longitudinal DS‐specific template creation steps corresponding to the written steps on the left. DS, Down syndrome; DTI, diffusion tensor imaging; DWI, diffusion‐weighted imaging; T1w, T1 weighted.

2.6. Cross‐sectional analysis

A cohort of 44 participants with acceptable image quality DWI and T1w images was used for an initial cross‐sectional analysis. Tract‐based spatial statistics (TBSS) were used for a voxel‐based analysis. ⁶¹ The DS‐specific T1w template served as the common space, and FA maps were averaged to make a DS‐specific FA template. A WM skeleton was extracted from the average FA map. ⁶¹ Individual FA data were then projected onto the skeleton for statistical analyses. ⁶¹ For non‐FA measures (MD, RD, AxD), the FSL script “tbss_non_FA” was used. ⁶²

Voxelwise statistics were performed using Permutation Analysis of Linear Models (PALM). ⁶³ Given a generalized linear model (GLM), non‐parametric permutation analysis was performed at each voxel on the WM skeleton to test for voxelwise significant relationships, providing a region where they occur. Permutation analysis was chosen for robustness to heteroscedasticity and to prevent assuming the data's underlying distribution. Three analyses were done to evaluate the model in Equation 1 across all DTI values. The first analysis tested whether DTI was significantly related to age, which is expected. The second investigated whether there were significant relationships between DTI and either age or global CL. The third investigated the same model with an age‐by‐CL interaction. Previous work has already shown that CL increases monotonically with age, as shown for this population in Figure 2, and many other studies have reported the impact of age on FA and other DTI parameters. ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷

$$\mathrm{DTI}={\beta }_0+({\beta }_{\mathrm{Age}}\ast \mathrm{Age})+({\beta }_{\mathrm{CL}}\ast \mathrm{CL})+({\beta }_{\mathrm{Age}\ast \mathrm{CL}}\ast \left(\mathrm{Age}\ast \mathrm{CL}\right))$$ (1)

FIGURE 2.

A graph of Aβ, measured in CL, versus age in years (top, graph A) and mean FA versus age (bottom, graph B) for each participant in this cohort, represented by a different color. First timepoint CL and FA values are shown as circles, while second timepoint values are shown as triangles. This indicates CL's progression as a function of age in adults with DS, as well as the simultaneous progression of FA, sensitive to WM, and potentially highlighting neurodegeneration in the presence of Aβ. Here, CL monotonically increases with age while mean FA decreases. Aβ, amyloid beta; CL, Centiloids; DS, Down syndrome; FA, fractional anisotropy; PET, positron emission tomography; WM, white matter.

These analyses were tested with a voxelwise one‐sample t test for positive and negative associations. Significance was assessed with 500 permutations with the recommended tail approximation, ⁶⁸ allowing for fewer permutations when using family‐wise error (FWE) correction on p values. It fits the tail of permuted distributions to the generalized Pareto distribution (GPD), computing p values from there. ⁶⁹ Significant regions of association were identified using FWE correction and threshold‐free cluster enhancement (TFCE) ⁷⁰ at a threshold of p < 0.05 for correction of multiple comparisons. Significance maps were then transformed to the Johns Hopkins University (JHU) atlases for WM regions and tracts to localize significant regions. ⁷¹ , ⁷² , ⁷³

2.7. Longitudinal analysis: rates of DTI changes

The same techniques as the cross‐sectional analyses were applied with a few critical differences. For each participant, voxelwise maps of rate of change were generated for each DTI measure (FA, MD, RD, AxD).

$$R_{\mathrm{DTI}}=\frac{\mathrm{Session}2\mathrm{DTI}-\mathrm{Session}1\mathrm{DTI}}{\mathrm{T}2-\mathrm{T}1}=\frac{\mathrm{\Delta }DTI}{\mathrm{\Delta }T}$$ (2)

Here, ∆DTI is the difference in each respective diffusion parameter between scanning sessions (e.g., Session 2 DTI – Session 1 DTI), and ∆T is the difference in time between scanning sessions (e.g., T2 – T1). The average ∆T across all participants was 3.7 ± 0.8 years. The DS‐specific midpoint template was used as the common space. Similarly, the rate of change for global CL for each individual was calculated ($R_{\mathrm{CL}})$. 

FIGURE 3.

LEFT, Structural, T1w images in the coronal, sagittal, and axial views highlighting the benefit of using a DS‐specific template for inter‐participant comparisons in a cohort of adults with DS instead of standard NT templates. Individuals with DS consistently have different neuroanatomy, on average, including larger ventricles and smaller cerebellums, pontes, and overall brain size. This is illustrated in the first two rows, showing T1w slices for the MNI152 template and this study's DS‐specific template. Using a standard atlas as a common space, like the MNI152, for individuals with DS could induce errors from misalignment and false neuroanatomical representation when performing population‐based analyses. The third row shows T1w slices from a randomly chosen participant with DS in this cohort. Quantitative comparison shows that the participant's neuroanatomy is more similar to the DS‐specific template. Finally, the final two rows show the participant with DS warped into the MNI152 and DS‐specific template space, respectively. Visual inspection shows less stretching and compression of anatomy when warping to the DS‐specific template space compared to the MNI152 space, making downstream analyses more robust. RIGHT, Axial slices of the generated DTI DS‐specific templates for diffusion MRI analyses. These include templates for FA, MD, RD, and AxD. These were used to more accurately represent DS neuroanatomy for voxelwise and statistical analyses. AxD, axial diffusivity; DS, Down syndrome; DTI, diffusion tensor imaging; FA, fractional anisotropy; MD, mean diffusivity; MNI, Montreal Neurological Institute; NT, neurotypical; RD, radial diffusivity; T1w, T1 weighted.

Models relating the longitudinal rates of change for DTI measures ($R_{\mathrm{DTI}})$ and CL were evaluated.

$$R_{\mathrm{DTI}}={\beta }_0+\left({\beta }_{R_{\mathrm{CL}}}\ast R_{\mathrm{CL}}\right)$$ (3)

The age effects on these parameters were already accounted for in the model, and no other covariates were necessary. Separate models were tested for all DTI parameters: $R_{\mathrm{FA}}$, $R_{\mathrm{MD}}$, $R_{\mathrm{AxD}}$, $R_{\mathrm{RD}}$.

3. RESULTS

Table 1 summarizes all demographic information for the participants in the results of the following subsections. Age, sex, global CL, amyloid PET positivity, cognitive status, and apolipoprotein E status are included. Figure 2 shows how this cohort's CL and mean FA change with age and highlights that Aβ CL monotonically increases with age, while mean FA decreases.

TABLE 1.

Participant demographics broken down for each study performed.

Demographic per study type	Template	Cross‐sectional	Longitudinal
Female	25	23	16
Male	22	21	19
Total	47	44	35
Mean age ± SD [min–max] (yrs)	37.4 ± 8.1 [25.6–57.3]	36.6 ± 7.6 [25.6–55.6]	37.7 ± 7.7 [25.6–53.6]
Mean CL ± SD [min–max]	17.9 ± 20.8 [0.5–101.1]	14.6 ± 13.4 [0.5–60.3]	16.9 ± 14.1 [0.5–60.3]
Amyloid positive PET (CL > 18.0)	14 (50% female)	12 (50% female)	13 (38% female)
Cognitive status (CN/MCI/dementia)	(43/3/1)	(42/1/1)	(33/1/1)
APOE status
APOE ε4 carriers (%)	19.1	20.5	17.1

Note: Information about sex, age, CLs, amyloid positive PET, cognitive status, and APOE status is present.

Abbreviations: APOE, apolipoprotein E; CL, Centiloid; CN, cognitively normal; MCI, mild cognitive impairment; PET, positron emission tomography; SD, standard deviation.

From the entire cohort, 47 T1w images from the baseline scan (Session 1) were used to create a DS‐specific anatomical T1w template. From this cohort, 44 participants had good‐quality baseline DWI with calculated Aβ burden. The DTI data from these individuals were used to generate a DTI template for the cross‐sectional analyses. For the longitudinal analyses, only 35 participants had acceptable image quality baseline and month 32 (Session 2) DWIs with accompanying amyloid PET scans. Demographic distributions between each study were similar (Table 1).

3.1. Template results

The DS‐specific templates shown in Figure 3 were visually inspected. Upon warping participant data to the DS‐specific template, there was a profound reduction in altered anatomy, particularly regarding the enlarged ventricles and smaller brainstem and cerebellum, hallmark features of DS. However, when participant scans were warped to the MNI152 template, visual inspection revealed less preservation of these regions. This suggests potential improvements in spatial alignment and anatomical preservation in these parameter maps across participants.

3.2. Cross‐sectional results

Significant associations with age were found for all DTI parameters, including almost all voxels on the WM skeleton (Figure 4). Similarly, in Equation 1, the age‐by‐CL interaction term also has widespread regions of significance, indicative of the age effect on CL, but in an inverse relationship with FA and a direct relationship with MD, RD, and AxD. This emphasizes the need for an improved model that can disentangle the impact of age on DTI and CL. Doing this would allow for direct analysis of the relationship between DTI and CL, the true interest of this study.

FIGURE 4.

Widespread significant regions of negative association between FA and age, shown in blue in the top row. WM skeleton tracts are shown in green on axial slices of the T1w DS‐specific template. This highlights the impact of age on DTI measures like FA, and the need to correct for it when looking for relationships with other parameters of interest. The bottom row shows significant regions of negative association between FA and CL while controlling for age in the cross‐sectional analysis. Regions included the left and right inferior and superior longitudinal fasciculi, thalamic radiations, sagittal stratum, and inferior fronto‐occipital fasciculi. CL, Centiloids; DS, Down syndrome; DTI, diffusion tensor imaging; FA, fractional anisotropy; T1w, T1 weighted; WM, white matter.

Regions are shown for Equation 1 GLM in Figure 4 throughout multiple axial brain slices for FA. No other DTI parameters had significant associations. Significant regions included the left and right inferior and superior longitudinal fasciculi, thalamic radiations, sagittal stratum, and inferior fronto‐occipital fasciculi for FA only in relation to CL and with age as a covariate (p < 0.05, FWE corrected).

3.3. Longitudinal results: rates of DTI change

Widespread, bilateral regions of significance, illustrated in Figures 4, 5, 6, remained after permutation testing (p < 0.05, FWE corrected) of the GLM in Equation 3 for all $R_{\mathrm{DTI}}$ parameters (e.g., $R_{\mathrm{FA}}$, $R_{\mathrm{MD}}$, $R_{\mathrm{AxD}}$, $R_{\mathrm{RD}}$). $R_{\mathrm{FA}}$ showed an inverse relationship with $R_{\mathrm{CL}}$, while $R_{\mathrm{MD}}$, $R_{\mathrm{AxD}}$, and $R_{\mathrm{RD}}$ showed directly positive relationships with $R_{\mathrm{CL}}$ (Figure 5). Given that the model is inherently normalized by time, accounting for age (Equation 2), no interaction terms were necessary to test.

FIGURE 5.

Graphs presenting the discovered trends in rates of change for DTI ($R_{\mathrm{DTI}}$) parameters as a function of the rate of change of global amyloid burden measured in CL ($R_{\mathrm{CL}}$). All trends are linear, with red regression lines and confidence intervals. Values were extracted for each participant from significant regions and then averaged. Significant regions were based on permutation analysis of linear models. The rate of change of FA ($R_{\mathrm{FA}}$) correlates inversely with $R_{\mathrm{CL}}$ (Graph A), while the rates of change for MD ($R_{\mathrm{MD}}$, Graph B), axial diffusivity ($R_{\mathrm{AxD}}$, Graph C), and RD ($R_{\mathrm{RD}}$, Graph D) correlate directly with $R_{\mathrm{CL}}$. This suggests a relationship between CL and DTI metrics and the potential impact of amyloid burden on white matter. CL, Centiloids; DTI, diffusion tensor imaging; FA, fractional anisotropy; MD, mean diffusivity; RD, radial diffusivity.

FIGURE 6.

Widespread significant regions of negative association between rates of change for FA ($R_{FA}$) and CL ($R_{CL}$), shown in blue. WM skeleton tracts are shown in green on axial slices of the T1w DS‐specific template. Clusters were predominantly in the temporal, frontal, occipital, and parietal lobes. Some include the right anterior and superior corona radiata, the left and right posterior and optic thalamic radiations, the left and right sagittal stratums, the left and right inferior fronto‐occipital longitudinal fasciculi, the right superior longitudinal fasciculus, and the right and left inferior longitudinal fasciculi. CL, Centiloids; DS, Down syndrome; FA, fractional anisotropy; T1w, T1 weighted; WM, white matter.

Most significant regions were in the temporal lobe for FA (Figure 6). However, there were still many significant clusters in the frontal, occipital, and parietal lobes, with roughly equal amounts in those three. Localized WM regions and tracts include the right anterior and superior corona radiata, the left and right posterior and optic thalamic radiations, the left and right sagittal stratums, the left and right inferior fronto‐occipital longitudinal fasciculi, the right superior longitudinal fasciculus, and the right and left inferior longitudinal fasciculi.

Most significant regions were in the parietal lobe for $R_{\mathrm{MD}}$ and $R_{\mathrm{AxD}}$ (Figure 7). Each had clusters extended toward the frontal and occipital lobes, but had fewer than $R_{\mathrm{FA}}$. Specific regions for $R_{\mathrm{MD}}$ include the body and splenium of the corpus callosum, the right superior corona radiata, the left and right posterior corona radiata, the right posterior and optic thalamic radiations, the left and right tapetums, the right corticospinal tract, the right inferior fronto‐occipital fasciculus, and the right superior and inferior longitudinal fasciculi. Specific regions for $R_{\mathrm{AxD}}$ include the same regions as $R_{\mathrm{MD}}$ along with the right retrolenticular part of the internal capsule, the left superior corona radiata, the cingulum, the left corticospinal tract, the forceps major, and the left superior longitudinal fasciculus. $R_{\mathrm{AxD}}$ had small clusters only in the inferior fronto‐occipital fasciculi and the inferior longitudinal fasciculi, unlike $R_{\mathrm{MD}}$.

FIGURE 7.

Widespread significant regions of positive association between rates of change for MD ($R_{\mathrm{MD}}$) and CL ($R_{\mathrm{CL}}$) (top), between rates of change for axial diffusivity ($R_{\mathrm{AxD}}$) and $R_{\mathrm{CL}}$ (middle), and between rates of change for RD ($R_{\mathrm{RD}}$) and $R_{\mathrm{CL}}$ (bottom), shown in red. WM skeleton tracts are shown in green on axial slices of the T1w DS‐specific template. Clusters were mostly in the parietal lobes. $R_{\mathrm{MD}}$ had regions in the body and splenium of the corpus callosum, the right superior corona radiata, the left and right posterior corona radiata, the right posterior and optic thalamic radiations, the left and right tapetums, the right corticospinal tract, the right inferior fronto‐occipital fasciculus, and the right superior and inferior longitudinal fasciculi. $R_{\mathrm{AxD}}$ included the same regions as $R_{\mathrm{MD}}$ along with the right retrolenticular part of the internal capsule, the left superior corona radiata, the cingulum, the left corticospinal tract, the forceps major, and the left superior longitudinal fasciculus. $R_{\mathrm{AxD}}$ did not share major significance in the inferior fronto‐occipital fasciculi and the inferior longitudinal fasciculi seen in $R_{\mathrm{MD}}$. Clusters were predominantly in the frontal and the parietal lobes for $R_{\mathrm{RD}}$. Regions included the genu, body, and splenium of the corpus callosum; the left and right anterior corona radiata; the right superior and posterior corona radiata; the right posterior and optic thalamic radiations; the right sagittal stratum; the forceps major and minor; the right and left inferior fronto‐occipital fasciculi; and the right inferior and superior longitudinal fasciculi. CL, Centiloids; DS, Down syndrome; MD, mean diffusivity; RD, radial diffusivity; T1w, T1 weighted; WM, white matter.

Most significant regions were in the parietal and frontal lobes for $R_{\mathrm{RD}}$, with some branching into the caudate, occipital lobe, temporal lobe, and the putamen (Figure 7). Specific WM regions included the genu, body, and splenium of the corpus callosum, the left and right anterior corona radiata, the right superior and posterior corona radiata, the right posterior and optic thalamic radiations, the right sagittal stratum, the forceps major and minor, the right and left inferior fronto‐occipital fasciculi, and the right inferior and superior longitudinal fasciculi.

4. DISCUSSION

The use of DTI to study WM and AD pathology in adults with DS has been limited in scope and application, particularly compared to research on sporadic AD development. ¹⁹ To date, no known longitudinal DTI studies have been performed in DS or on AD in DS. ¹⁸ The current study explored the relationship between Aβ burden and DTI parameters, starting with a cross‐sectional baseline and extending to a longitudinal study of the rate of DTI change in the same cohort. Two timepoints (spaced ≈ 3.7 years apart) of DWI and global Aβ CL data were used for the analysis. A DS‐specific template was made to preserve anatomical variability unique to this population during warping and registration, while addressing differences from NT atlases. Additionally, a midpoint template was created for each participant to account for intra‐participant variability and enable comparisons in a DS‐specific midpoint template space. TBSS and PALM were performed for voxelwise statistics on an extracted WM skeleton, and significant regions were localized using the JHU WM atlases.

The cross‐sectional results showed a negative association between FA and Aβ CL in adults with DS, consistent with prior studies using TBSS, which found lower FA in the left inferior longitudinal fasciculus linked to higher Aβ burden. ²⁰ , ²¹ Other studies comparing DS and AD in DS to healthy controls also found that adults with DS had reduced FA in this long‐association fiber using TBSS. ⁷⁴ , ⁷⁵ Though there are few DTI‐based studies, it is reassuring to see replicated results. According to a review by Saini et al., a common region of significance is the long‐association fibers across DTI studies in DS. ¹⁸ Compared to other studies, however, this cross‐sectional analysis found relatively few regions of significance. It has been noted in this population that, due to heterogeneity in brain morphology, neurodevelopment, behavior within the MRI scanner, small sample sizes, and standard references used, replicability across studies has been limited thus far. ¹⁸ This remains a key motivating factor for the subsequent longitudinal analyses in this work.

The longitudinal rates of DTI change models found widespread regions where DTI significantly correlated with global Aβ CL, including the superior longitudinal fasiculi, inferior longitudinal fasciculi, and inferior fronto‐occipital longitudinal fasciculi; the corpus callosum; the thalamic and optic radiations; and the corona radiata. Compared to cross‐sectional results, the rate of DTI change model showed greater sensitivity, revealing that higher FA was associated with lower Aβ CL in overlapping and additional regions. Scatterplots of the rates of mean DTI changes as a function of the rate of Aβ CL changes across all significant regions reveal a consistent finding, emphasizing the trend (Figure 5). $R_{\mathrm{FA}}$ decreases with $R_{\mathrm{CL}}$, while $R_{\mathrm{MD}}$, $R_{\mathrm{RD}}$, and $R_{\mathrm{AxD}}$ increase simultaneously. This could indicate reduced WM tract coherence from lower fiber density, as it coincides with greater Aβ burden. Other possible interpretations include increased neuroinflammation, decreased myelination, or overall neuronal degeneration. Though DTI does not provide specific results of the underlying biological mechanisms, all participants have DS with AD pathology, but without a clinical AD diagnosis. Therefore, these changes likely correspond to AD pathology rather than an outside factor.

The rate of change model minimizes age‐related effects on DTI and Aβ burden by incorporating time, emphasizing the strengths of longitudinal data, and by controlling for differences in baseline age. Furthermore, by using participants as their own controls, longitudinal designs reduce inter‐participant variability, focusing on individual changes and group trends. This study also leveraged a DS‐specific template to preserve individual neuroanatomy, TBSS to mitigate WM tract misalignment, and participant‐specific midpoint spaces to minimize anatomical changes and misregistration, making standard reference regions less critical. These methods enabled the characterization of AD pathology in DS through DTI and amyloid PET, capturing temporal relationships and supporting analyses of causality and change. As the first longitudinal DTI study in adults with DS, this work highlights the value of longitudinal approaches in studying progressive conditions like AD and hopefully motivates the use of this methodology in future research of both DS and sporadic AD.

A limitation of this study is that participants came from one ABC‐DS site and are largely homogeneous regarding race/ethnicity and geographical region, reducing generalizability. Ongoing harmonization efforts aim to integrate data across ABC‐DS sites, improving representation and diversity while addressing inter‐scanner and protocol variability. Another limitation to the current study's resulting interpretation is DTI's inherent lack of specificity to biological mechanisms. While highly sensitive to microstructural changes, the DTI model cannot identify the cause of diffusivity and anisotropy or distinguish between neurodegeneration and inflammation. However, repeated DTI can reveal progressive patterns in these microstructural changes across multiple participants. When DTI is used in combination with other commonplace AD biomarkers like amyloid PET, its association with AD becomes more promising, making the multimodal approach another strength in this investigation. With enough replication of this trend, DTI's utility as an indicator of AD pathology in DS seems plausible. Furthermore, DTI is an in vivo approach that is more accessible than other scans in standard clinical practice.

Other, more advanced diffusion models could also be used for greater specificity. Hence, another limitation of this work is the lack of consistent, advanced MRI scan protocols. Most baseline scans had a single phase‐encoding DWI, making EPI distortion correction more challenging. In contrast, the repeat scans often had two phase‐encoding DWIs. Single phase‐encoding polarity DWIs were standard practice until relatively recently, though, so variability in protocols is common and manageable. Additionally, the use of only one non‐zero diffusion weighting (b value) limited the application of advanced diffusion models requiring higher angular resolution. It will be informative to see how these trends compare to more biophysical models.

Finally, though age‐related effects were accounted for in the longitudinal model, the influence of age on the trends found in this work between ${\mathrm{R}}_{\mathrm{DTI}}$ and ${\mathrm{R}}_{\mathrm{DTI}}$ remain unclear. It is reasonable to speculate that the relationship between amyloid accumulation and WM decline might vary with age, similar to other AD biomarkers. Additional investigation to interpret this observation will be explored in future studies.

This study highlights the potentially extensive impact that Aβ and AD pathology have on brain microstructure, and the ability of DTI to detect this in a multimodal framework. Significant relationships were observed for measures over two timepoints, separated by 3.7 ± 0.8 years on average, indicating a short time for a lot of change. Future work will expand the sample size and assess more advanced qMRI models in DS. Analyses characterizing the time course of DTI changes in relation to tau and Aβ will also be performed to place the DTI outcomes within the AD‐related biomarker framework, as applicable to DS.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to disclose. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All participants provided informed consent and assent prior to conducting any study activities.

Supporting information

Supporting information

1. The ABC‐DS Investigators

Ben Handen, PhD (MPI); Brad Christian, PhD (MPI); Elizabeth Head, PhD (MPI); Mark Mapstone, PhD (MPI); Diana Rosas, MD (MGH Site Co‐PI); Flo Lai, MD (MGH Site Co‐PI); Joe Lee, PhD (Columbia Site Co‐PI); Sharon Krinsky‐McHale, PhD (Columbia Site Co‐PI); Fred Schmitt, PhD (Kentucky Site Co‐PI); Jordan Harp, PhD (Kentucky Site Co‐PI); Christy Hom, PhD (UCI Site Co‐PI); Ira Lott, MD (UCI Site Co‐PI); Sigan Hartley, PhD (Wisconsin Site Co‐PI); Shahid Zaman, MD, PhD (Cambridge Site PI); Beau Ances, MD, PhD (Washington University Site PI); Lauren Ptomey, PhD (Kansas University Medical Center Site Co‐PI); Jeff Burns, MD (Kansas University Medical Center Site Co‐PI); Adam Brickman, PhD (Columbia Core Co‐Lead)

LeMerise LG, Guerrero‐Gonzalez J, McVea A, et al. Longitudinal diffusion tensor imaging correlates with amyloid burden in Down syndrome. Alzheimer's Dement. 2025;21:e70572. 10.1002/alz.70572