White Matter Integrity on DTI, Amyloid Load, and Neurodegeneration in Non-demented Elderly

Abstract

Importance

Pathophysiologic mechanisms leading to loss of white matter (WM) integrity and the temporal positioning of biomarkers of WM integrity relative to the biomarkers of gray matter (GM) neurodegeneration and amyloid load in the course of AD are poorly understood.

Objective

To investigate the effects of Alzheimer’s disease (AD)-related GM neurodegeneration and high β-amyloid on white matter (WM) microstructure in non-demented older adults.

Design

Longitudinal cohort study

Setting

Population-based Mayo Clinic Study of Aging.

Participants

Participants (n=701) with MRI/DTI and PET studies diagnosed as cognitively normal (CN; n=570) or mild cognitive impairment (MCI; n=131) were included. CN and MCI subjects were divided into biomarker-negative, amyloid- positive only, neurodegeneration- positive only, and amyloid plus neurodegeneration-positive groups based on their amyloid load on 11C-Pittsburgh compound-B PET, AD hypometabolic pattern on 18F-fluorodeoxyglucose PET and/or hippocampal atrophy on MRI.

Main Outcome Measure

Fractional anisotrophy (FA) from diffusion tensor imaging (DTI)

Results

No FA alterations were observed in biomarker-negative MCI, and amyloid-positive only CN and MCI groups. Conversely, neurodegeneration-positive only and amyloid plus neurodegeneration- positive CN and MCI groups consistently had decreased FA in the fornix, which correlated with cognitive performance (Rho=0.38; p<0.001). Patients with MCI had more extensive WM involvement than CN subjects, and greatest FA decreases were observed in the amyloid plus neurodegeneration-positive MCI group.

Conclusions and Relevance

High amyloid load does not influence DTI-based measures of WM integrity in the absence of co-existent GM neurodegeneration in non-demented older adults.

Introduction

Diffusion tensor imaging (DTI) shows the profound loss of WM integrity in Alzheimer’s disease (AD) starting from the prodromal stages^1–7. Decrease in the directionality of water diffusion measured with fractional anisotropy (FA) on DTI has been linked to loss of myelin and axons in the white matter (WM).⁸ According to the biomarker model of AD,⁹ and the pre-clinical staging of AD derived from this model,¹⁰ alterations in biomarkers of gray matter (GM) neurodegeneration follow biomarkers of amyloid deposition during the course of disease progression from pre-clinical to clinical dementia. However, the AD-related pathophysiologic mechanisms leading to loss of WM integrity and the temporal positioning of biomarkers of WM integrity relative to the biomarkers of GM neurodegeneration and amyloid load in the course of AD are unknown.

In this study, we investigated the FA alterations in cognitively normal older adults (CN) and subjects with mild cognitive impairment (MCI) from a population-based study on non-demented older adults, who were classified as neurodegeneration and/or amyloid-positive or -negative. Our objective was to determine the effects of Alzheimer’s disease (AD)-related GM neurodegeneration and high β-amyloid on white matter (WM) microstructure in non-demented older adults.

Methods

Subjects

Older adults (n=701; age=70–89) who participated in the Mayo Clinic Study of Aging (MCSA) MRI and PET studies from November 2009 to August 2013 were included. The MCSA is a prospective population-based study of non-demented older adults in Olmsted County Minnesota¹¹. To be included in the current study, non-demented subjects should have participated in MRI, amyloid PET with 11C-Pittsburgh compound-B (PiB), and 18F-fluorodeoxyglucose (FDG) PET studies during the same cycle of clinical evaluation. The neuropsychological test scores were scaled such that they had a mean of 0.0 and a standard deviation of 1.0 among all MCSA full participants and averaged to obtain a global cognitive z-score.¹¹ The diagnosis of MCI was based on the published criteria: cognitive complaint, cognitive function not normal for age, decline in cognition, essentially normal functional activities, and not demented¹². Diagnosis of dementia was based on the DSM-IV criteria¹³, and patients with dementia were excluded. Diagnosis was determined by a consensus committee including the neurologist, neuropsychologist and the nurse who evaluated each participant, shielded from prior diagnosis. Subjects who had a contraindication for MRI such as a pacemaker, or who were unable to participate in imaging studies because of severe illness were excluded. However, subjects were not excluded due to neurological, psychiatric, or systemic illnesses to preserve the representativeness of the study sample as much as possible.

Classification of subjects into biomarker groups with structural MRI and PET

MRIs were performed at 3 Tesla using an eight-channel phased array coil (GE, Milwaukee, WI). A 3D high resolution MPRAGE acquisition was performed for hippocampal volume measurements and for anatomic segmentation and labeling of the DTI and PET scans. Hippocampal volume was measured with FreeSurfer software (version 5.3).¹⁴ We calculated an adjusted hippocampal volume as the residual from a linear regression of hippocampal volume versus total intracranial volume.¹⁵

PET images were acquired using a PET/CT scanner (DRX; GE Healthcare) operating in 3D mode. After a 40-min uptake period, a 20-min PiB scan was obtained. The amyloid PET acquisition consisted of four 5-min dynamic frames, acquired from 40 to 60 min after injection. FDG PET images were obtained 1 hour after the PIB scan. PiB PET quantitative analysis was performed using the fully automated image processing pipeline which was previously described in detail¹⁶. Briefly, a cortical global amyloid PET standardized uptake value ratio (SUVR) was obtained by combining the prefrontal, orbitofrontal, parietal, temporal, anterior cingulate, and posterior cingulate/precuneus ROI values normalized by the cerebellar GM ROI of an atlas¹⁷ modified in-house. FDG PET scans were analyzed in a similar manner using angular gyrus, posterior cingulate, and inferior temporal cortical ROIs to define an Alzheimer signature composite by Landau et al.¹⁸ normalized to pons and vermis.

Hippocampal atrophy on MRI and/or hypometabolism in the Alzheimer signature composite on FDG PET was used to classify subjects into the neurodegeneration-positive group, and high amyloid load on PET was used to classify subjects into the amyloid-positive group. Cut-points for amyloid positivity, hippocampal atrophy and Alzheimer signature hypometabolism were determined from the 10^th percentile of the measurement distributions in clinically diagnosed AD patients as previously described.¹⁵

DTI Methods

DTI acquisition was a single-shot echo-planar pulse sequence performed in axial plane using parallel imaging with a SENSE factor of two; TR= 10,200 ms; an in-plane matrix of 128/128; field of view of 35 cm. The DTI volumes consisted of 41 diffusion-encoding gradient directions and five volumes of non-diffusion T2-weighted images. The slice thickness was 2.7mm, corresponding to 2.7 mm. isotropic resolution. We used a previously tested and validated method to process DTI scans.¹⁹ Briefly, DTI images were corrected for subject motion and residual eddy current distortion by affine-registering each volumes to the first image volume, which had no diffusion-weighting. Diffusion tensors were fit for extracted voxels using linear least squares optimization and FA images were calculated from the eigenvalues of the tensors using FSL version 4. Advanced normalization tools-Symmetric Normalization (ANTS-SyN) version 1.9.y²⁰ algorithm was used for generating a study-specific template from all FA images and for non-linear registration of a subject’s FA images to the template, and smoothed with a 8mm full-width at half-maximum Gaussian kernel. All CN and MCI biomarker groups were compared to the biomarker-negative CN group using voxel-based analysis (VBA) on SPM5.²¹ A secondary quantitative analysis was performed with the single-subject John’s Hopkins University (JHU) DTI atlas²² by registering the WM ROIs to the study-specific template using ANTS-SyN and displayed on a rendered single-subject transparent brain for visualization using the MRIcroGL program (http://www.mccauslandcenter.sc.edu/mricrogl/).

Neuropsychological Testing

The neuropsychological battery of MCSA was previously described.¹¹ Briefly, all raw neuropsychological test scores were scaled such that they had a mean of 0.0 and a standard deviation of 1.0 among all MCSA participants.¹² We obtained individual domain scores by averaging and scaling the tests for each cognitive domain (memory, language, attention/executive function, visual-spatial processing). A global cognitive function standard score was derived by averaging and scaling the four standardized cognitive domain scores.

Statistical Analysis

Subject characteristics were compared using chi-squared test for differences in proportions or the non-parametric Kruskal-Wallis test on the ranks in the CN and MCI biomarker groups separately. The significance level cut-off for the VBA analysis comparing CN and MCI biomarker groups to the biomarker-negative CN group adjusting for age was set at p<0.05 corrected for multiple comparisons using family-wise error (FWE). FA values derived from right and left hemispheric WM JHU atlas ROIs were averaged and the areas under the receiver operating curve (AUROC) for distinguishing the CN and MCI biomarker groups from the biomarker-negative CN group were calculated for each individual ROI. The JHU atlas-based FA values were compared among the CN and MCI biomarker groups to the biomarker-negative CN control group and ranked according to AUROC values. Regions with AUROC >0.70 were displayed on a rendered single-subject transparent brain for visualization.

Results

Subject Characteristics

Subject characteristics classified according to the biomarker positivity is listed in (Table 1) for the CN group and (Table 2) for the MCI group. Out of 570 CN subjects, 258 (45%) were classified as biomarker-negative and 77 (14%) were classified as amyloid and neurodegeneration-positive. In contrast, a smaller proportion of MCI subjects [21 out of 131 (16%)] were classified as biomarker- negative and a larger proportion of MCI subjects were classified as amyloid and neurodegeneration- positive [55 out of 131 (42%)]. Biomarker-negative subjects were younger than the biomarker- positive subjects in both the CN (p<0.001) and MCI (p=0.002) groups. Because of age differences among the biomarker groups, we used age as a covariate in all DTI analysis. APOE ε4 was more frequent in the amyloid-positive compared to amyloid-negative CN and MCI groups (p<0.001). Cognitive performance measured with the global cognitive z-score declined from biomarker-negative to only amyloid-positive to only neurodegeneration-positive to both amyloid and neurodegeneration-positive groups in the CN (p<0.001) and MCI (p<0.03) subjects. Similar frequencies of amnestic and nonamnestic MCI subtypes were observed among the four biomarker groups of MCI (p=0.91).

Table 1.

Characteristics of cognitively normal subjects classified by biomarker abnormality

	Neurodegeneration-Negative (−)		Neurodegeneration-Positive (+)		P-value
	Amyloid −	Amyloid +	Amyloid −	Amyloid +
No. of subjects	258	113	122	77	---
No. of females (%)	127 (49)	53 (47)	48 (39)	32 (42)	0.28
No of ε4 carriers (%)	47 (18)	44 (39)	18 (15)	33 (43)	<0.001
Age, yrs	76 (73, 79)	79 (74, 83)	79 (76, 84)	81 (77, 84)	<0.001
Education, yrs	14 (12, 16)	14 (12, 16)	14 (12, 16)	14 (12, 16)	0.84
CDR sum of boxes	0 (0, 0)	0 (0, 0)	0 (0, 0)	0 (0, 0)	<0.001
Global cognitive z-score	0.93 (0.44, 1.47)	0.72 (0.13, 1.23)	0.36 (−0.25, 1.02)	0.24 (−0.33, 0.75)	<0.001
Cortical Global PIB SUVR	1.33 (1.29, 1.38)	1.79 (1.64, 2)	1.36 (1.30, 1.40)	1.98 (1.69, 2.28)	---
FDG PET Alzheimer composite	1.48 (1.40, 1.55)	1.44 (1.37, 1.51)	1.27 (1.23, 1.30)	1.26 (1.20, 1.31)	---
Adjusted hippocampal volume	0.38 (−0.05, 0.87)	0.21 (−0.11, 0.67)	−0.50 (−1.04, −0.01)	−0.79 (−1.19, −0.00)	---

Median (interquartile range) are reported for continuous variables. P-values are from a chi-squared test for differences in proportions or the non-parametric Kruskal-Wallis on the ranks.

Table 2.

Characteristics of subjects with mild cognitive impairment classified by biomarker abnormality

	Neurodegeneration-Negative (−)		Neurodegeneration-Positive (+)		P-value
	Amyloid −	Amyloid +	Amyloid −	Amyloid +
No. of subjects	21	18	37	55	---
No. of females (%)	4 (19)	9 (50)	10 (27)	24 (44)	0.08
No of ε4 carriers (%)	2 (10)	11 (61)	5 (14)	30 (55)	<0.001
Age, yrs	77 (74, 83)	77.5 (75, 81)	80(77, 85)	84 (80, 87)	0.002
Education, yrs	12 (12, 14)	12 (12, 14)	14 (12, 16)	15 (12, 17)	0.08
CDR sum of boxes	0.5 (0.0, 1.0)	1.0 (0.5, 1.5)	0.50 (0.0, 1.0)	1.0 (0.5, 2.0)	0.02
Global cognitive z-score	−0.41 (−0.88, 0.32)	−0.50 (−1.23, −0.25)	−0.62 (−1.16, −0.19)	−0.90 (−1.63, −0.48)	0.03
Cortical Global PIB SUVR	1.36 (1.33, 1.38)	2.00 (1.86, 2.33)	1.35 (1.31, 1.40)	2.31 (1.89, 2.48)	---
FDG PET Alzheimer composite	1.48 (1.42, 1.55)	1.42 (1.36, 1.51)	1.27 (1.20, 1.32)	1.21 (1.11, 1.30)	---
Adjusted hippocampal volume	0.05 (−0.54, 0.63)	−0.25 (−0.44, 0.10)	−0.93 (−1.31, −0.38)	−1.29 (−1.72, −0.88)	---
MCI Subtype					0.91
Amnestic	18 (86)	14 (78)	30 (81)	46 (84)
Non-amnestic	3 (14)	4 (22)	7 (19)	9 (16)

Median (interquartile range) is reported for continuous variables. P-values are from a chi-squared test for differences in proportions or the non-parametric Kruskal-Wallis on the ranks.

Voxel-based Analysis (VBA)

VBA did not reveal any differences in FA values when the amyloid-positive only CN group was compared to the biomarker-negative CN control group adjusted for age (p>0.05; FWE corrected). However, lower FA was observed in both of the CN neurodegeneration-positive groups. CN subjects classified as neurodegeneration-positive only had decreased FA in the fornix and focal areas in the corpus callosum and occipital WM compared to the biomarker-negative CN controls. CN subjects classified as neurodegeneration plus amyloid-positive had similarly decreased FA in the fornix and slightly greater involvement in the corpus callosum and occipital lobe WM. Involvement of the right parahippocampal WM was also observed in CN subjects classified as neurodegeneration plus amyloid-positive. (Figure 1a)

Figure 1.

Voxel-based analysis of white matter FA in cognitively normal and mild cognitive impairment biomarker groups compared to the biomarker negative cognitively normal subjects. T-values are displayed in color bars.

VBA findings adjusted for age in subjects with MCI showed similarities to the CN group. There were no differences in FA values when the biomarker-negative and the amyloid-positive only MCI group was compared to the biomarker-negative CN control group (p>0.05; FWE corrected). However, MCI subjects classified as neurodegeneration-positive had decreased FA in the fornix, corpus callosum, focal areas in the cingulate gyrus and occipital lobe WM compared to the biomarker-negative CN controls. MCI subjects classified as neurodegeneration plus amyloid-positive had similarly decreased FA in the fornix, corpus callosum, cingulate gyrus and occipital lobe WM but in addition they had decreased FA in the precuneus, basal frontal, and temporal lobe WM. (Figure 1b)

Atlas-based Analysis

To determine the specific WM tracts that were involved in the neurodegeneration-positive CN and MCI biomarker groups, we performed a secondary JHU atlas-based analysis on subject FA maps and reported the individual regional WM FAs that distinguished the neurodegeneration-positive (amyloid-positive or -negative) CN and MCI groups from the biomarker-negative CN group with an AUROC of >0.70. Right and left hemispheric tract FA values, three sections of the corpus callosum (Genu, body and splenium) and the two sections of the cingulum tract (hippocampal and cingulate gyrus) in the JHU atlas were averaged

The only tract that distinguished CN subjects in the neurodegeneration-positive groups from biomarker-negative CN group was fornix (AUROC=079 for amyloid-positive; AUROC=0.74 for amyloid-negative). Similarly, fornix was the only tract that distinguished neurodegeneration-positive MCI group from the biomarker-negative CN group with the highest AUROC (AUROC=0.87 for amyloid-positive; AUROC=0.83 for amyloid-negative). There were additional tracts and WM regions that distinguished the neurodegeneration-positive MCI group from the biomarker-negative CN group with AUROC >0.70 which are displayed on a single –subject rendered brain in (Figure 2). Similar to our findings in VBA, there were more extensive FA decreases in the WM in neurodegeneration-positive and amyloid-positive MCI subjects compared to the neurodegeneration-positive only MCI subjects.

Figure 2. Region of interest analysis.

White matter regions of interest with decreased FA in the neurodegeneration-positive cognitively normal and mild cognitive impairment biomarker groups are displayed on a rendered transparent brain from a single subject. FA values from regions of interest that distinguished the biomarker positive groups form the biomarker negative cognitively normal subjects with an area under the receiver operating characteristic curve (AUROC) of >80 are colored in red. FA values from regions of interest that distinguished the biomarker-positive groups form the biomarker-negative cognitively normal subjects with an AUROC of 71–80 are colored in yellow.

Fornix fractional anisotrophy (FA) and correlations with cognitive function

Fornix was the only tract that consistently showed lower FA values in neurodegeneration-positive CN and MCI groups compared to the biomarker-negative CN group regardless of amyloid biomarker status. Because positivity for GM neurodegeneration was determined based on the presence of hippocampal atrophy on MRI and/or hypometabolism in the Alzheimer signature composite on FDG PET, we further investigated the GM neurodegeneration biomarkers that were associated with decreased fornix FA. We found that subjects with hippocampal atrophy only, and subjects with hypometabolism in the Alzheimer signature composite only, had decreased FA in the CN and MCI groups compared to biomarker-negative CN controls (p<0.001). In subjects who had both hippocampal atrophy and hypometabolism in the Alzheimer signature composite, the FA was even lower compared to biomarker-negative CN controls (p<0.001) as demonstrated in Figure 3.

Figure 3.

Fractional anisotropy of the fornix in the cognitively normal and mild cognitive impairment in biomarker-negative and neurodegeneration-positive and groups classified according to neurodegeneration biomarker positivity: hypometabolism on Alzheimer signature composite on FDG PET (FDG+) and hippocampal atrophy on MRI (aHV+).

Lower FA values were associated with lower cognitive performance in the whole group with Spearman’s rank correlation (rho) = 0.38 (95% confidence interval [CI]=0.31, 0.45); p=<0.01 and in CN subjects with rho= 0.30 (95% CI=0.22, 0.38); p=<0.01. This correlation between lower FA values and lower cognitive performance was weaker in subjects with MCI with rho= 0.15 (95% CI= −0.04, 0.33); p=0.11. (Figure 4)

Figure 4. Fractional anisotropy of the fornix and global cognitive Z-scores.

The scatter plot displays the relationship between Fractional anisotropy of the fornix and global cognitive Z-scores. Cognitively normal group is represented with the blue triangles and the mild cognitive impairment group is represented with the red circles. The least squares fit for the entire group is shown with a black line. The least squares fit for the cognitively normal group is shown with a blue dashed line, and for the mild cognitive impairment group with a red dashed line.

Discussion

In a cohort of non-demented older adults from the community, classified according to the status of neurodegeneration and amyloid biomarkers, loss of WM microstructural integrity on DTI was associated with biomarkers of GM neurodegeneration but not with amyloid biomarker positivity. As expected, MCI subjects had more extensive WM involvement compared to CN subjects. We found consistent decreases in fornix FA both in CN and MCI neurodegeneration-positive groups, which correlated with cognitive performance in the entire cohort.

High amyloid load by itself did not have an effect on the microstructural integrity of the WM in the absence of GM neurodegeneration both in the CN and MCI subjects. Evidence from prospective cohort studies indicate that brain amyloidosis is not a benign process. Those with high amyloid load are at an increased risk for cognitive decline, MCI or dementia^23–27. However the effects of β-amyloid on WM appear to be associated with GM neurodegeneration. We found more extensive WM FA decreases in neurodegeneration-positive MCI compared to neurodegeneration-positive CN subjects. In addition, MCI subjects with GM neurodegeneration and amyloid biomarker positivity had more widespread WM FA decreases than MCI subjects with GM neurodegeneration alone. The interaction between biomarkers of neurodegeneration and β-amyloid load and their association with adverse cognitive outcomes in the course of AD have been observed and discussed^28–31, and are consistent with our observations. There was more WM involvement with high β-amyloid load only in neurodegeneration –positive subjects with MCI compared to the biomarker-negative CN group. However, in the absence of cognitive impairment, high β-amyloid load did not have any additional effect on WM integrity. In preclinical AD, integrity of WM is associated with GM neurodegeneration rather than β-amyloid, however β-amyloid appears to be related to WM integrity as the disease progresses and individuals develop cognitive impairment.

Fornix FA was consistently decreased in the neurodegeneration-positive CN and MCI groups compared to the biomarker-negative CN group. Decreased fornix FA is one of the earliest MRI abnormalities observed in cognitively normal individuals who are at an increased risk for AD. Decreases in fornix FA have been observed in pre-symptomatic carriers of familial AD mutations³² and in patients with MCI ^5,33,34, which predicted the decline in memory function ⁵. Fornix carries the efferent projections from the CA1 and CA3 pyramidal neurons of the hippocampus and subiculum, connecting these structures to the septal nuclei, anterior thalamic nucleus, mammillary bodies and medial hypothalamus. Fornix also carries the afferent cholinergic and GABAergic projections from the medial septal nuclei and the adjacent diagonal band back to the medial temporal lobe, interconnecting the core limbic structures³⁵. FA measurements from the body of fornix are further protected from the noise of crossing fibers, which makes fornix an ideal anatomic structure for assessing microstructural changes with DTI. Because fornix carries the axons projecting from the CA1 and CA3 pyramidal neurons in the hippocampus, integrity of the fornix is in-part linked to the integrity of the hippocampus³⁶. In keeping with that, we found significantly reduced fornix FA in subjects with hippocampal atrophy. But in addition, fornix FA was decreased in subjects with reduced metabolism in the AD signature composite regions on FDG PET, even in the absence of hippocampal atrophy. The association of DTI-based WM integrity biomarkers with GM neurodegeneration biomarkers is consistent with previous reports on the association of decreased WM FA with GM hypometabolism and atrophy. ^{37, 38,39}

We found correlations between lower fornix FA and lower cognitive performance in the entire cohort and in the CN subjects, but this relationship was weaker in MCI subjects. The neuropathology underlying MCI is heterogeneous. While the most common etiology underlying MCI is AD, vascular and Lewy body disease pathologies in addition to AD are common. ^40–42 These additional pathologies may have a significant impact on cognitive performance in MCI independent of AD-related neurodegeneration. Thus, the weaker relationship between fornix FA and cognitive performance that we observed in MCI compared to CN subjects may be related to a greater pathologic heterogeneity in MCI that impacts cognitive performance.^42,43 Similarly, WM integrity may be affected by other pathologies commonly found in older adults such as Lewy body disease, cerebrovascular disease, hippocampal sclerosis, TDP43, and argyrophilic grain disease along with AD-related pathology. The interaction of these other pathologies with AD-related pathology and their influence on WM integrity requires further investigation.

We did not analyze the data on mean diffusivity, which is significantly effected by partial volume averaging of CSF especially in the fornix that is surrounded by CSF. Although the influence of partial volume averaging of CSF is less on FA, up to 16% of the difference in fornix FA among MCI and CN subjects has been attributed to macrostructural changes in the fornix and associated partial volume averaging of CSF.⁴⁴ Therefore a small percentage of the differences in fornix FA we observed among the biomarker groups may be attributed to macrostructural changes. Although we used an optimized DTI sequence for clinical applications,⁴⁵ higher angular and spatial resolution that allows effective partial volume correction within clinically applicable time frames may potentially improve DTI’s sensitivity to alterations in WM microstructure in pre-clinical AD and MCI.

Data from this study indicate that loss of WM integrity measured with DTI accompanies GM neurodegeneration biomarker abnormalities and not amyloid biomarker positivity in the course of preclinical AD. Putting our findings into context with the preclinical staging of AD, loss of WM integrity on DTI should be observed at Stage 2 along with GM degeneration and amyloid biomarker positivity. However, 21% of the CN subjects did not fit into the preclinical stages of AD because they were classified as neurodegeneration-positive but amyloid-negative, and had FA reductions mostly confined to the fornix. Whether these CN subjects have non-AD related pathology, which we labeled as “suspected non-AD pathology (SNAP)”¹⁵ or a “neurodegeneration-first pathway to AD”⁴⁶ requires further investigation with pathologic confirmation.