OUP user menu

Diffusion tensor imaging in preclinical and presymptomatic carriers of familial Alzheimer's disease mutations

John M. Ringman, Joseph O’Neill, Daniel Geschwind, Luis Medina, Liana G. Apostolova, Yaneth Rodriguez, Barbara Schaffer, Arousiak Varpetian, Benjamin Tseng, Freddy Ortiz, Jaime Fitten, Jeffrey L. Cummings, George Bartzokis
DOI: http://dx.doi.org/10.1093/brain/awm102 1767-1776 First published online: 23 May 2007


Measures are needed that identify persons that will develop Alzheimer's disease in order to target them for preventative interventions. There is evidence from animal, pathological and imaging studies that disruption of white matter occurs in the course of Alzheimer's disease and may be an early event. Prior studies have suggested that late-myelinating regions or white matter connecting limbic structures are particularly susceptible to degradation. Persons destined to develop the disease by virtue of fully penetrant genetic alterations (familial Alzheimer's disease or FAD) provide a model in which early and even presymptomatic changes of the disease may be identified. In this study we performed diffusion tensor imaging (DTI) on 2 demented and 21 subjects at-risk for inheriting an FAD mutation. We compared global and localized fractional anisotropy (FA) measures in white matter between FAD mutation carriers and non-carriers in the preclinical (clinical dementia rating <1, n = 20) and presymptomatic (clinical dementia rating = 0, n = 15) stages of the disease. There were no significant differences between mutation carriers and non-carriers with regard to absolute age, age relative to the typical age of disease diagnosis in their family, gender or Mini-Mental Status Examination Score. Among preclinical FAD mutation carriers (n = 12), mean whole brain white-matter FA (P = 0.045), FA of the columns of the fornix (P = 0.012), area of the perforant pathways bilaterally (right side: P = 0.028, left side: P = 0.027) and left orbitofrontal lobe (P = 0.024) were decreased relative to that of non-carriers (n = 8). We also found that FA in the columns of the fornix (P = 0.008) and left orbitofrontal lobe white matter (P = 0.045) were decreased in the eight presymptomatic mutation carriers compared to seven non-carriers. Logistic regression demonstrated that FA of the columns of the fornix was a better predictor of mutation status than was cross-sectional area of the fornix, global mean white-matter FA and left frontal lobe white-matter FA. In a linear regression analysis, white-matter volume (P = 0.002), hippocampal volume (P = 0.023) and mutation status (P = 0.032) significantly predicted fornix FA. We conclude that FA is decreased in the white matter in preclinical and even presymptomatic FAD mutation carriers, particularly in the late-myelinating tracts connecting limbic structures. Decreased FA in of the columns of the fornix is particularly robust in early FAD and may provide a biomarker for early disease in sporadic Alzheimer's disease.

  • familial Alzheimer's disease
  • presymptomatic
  • diffusion tensor imaging
  • fractional anisotropy
  • Presenilin-1
  • amyloid precursor protein
  • white matter
  • fornix
  • biomarker


Disruption of white matter in Alzheimer's disease has been demonstrated in studies of post-mortem human brain (Chia et al., 1984; Englund et al., 1988), in animal models (Pak et al., 2003; Wirths et al., 2006) and in vivo in humans (Bartzokis et al., 2003, 2004). The electron microscopic appearance of white matter in Alzheimer's disease was described in 1964 (Terry et al., 1964). In this report, the normal lamellar structure of myelin was noted to be lost in places despite relatively normal-appearing axoplasm. It has been argued that such disruption of white matter reflects the susceptibility of late-myelinating regions to the effects of aging and Alzheimer's disease, and that these changes are primary in causing the clinical manifestations of the disease (Bartzokis, 2004). Whether a primary or secondary event in the aetiology of Alzheimer's disease, a comprehensive explanation of the disease must include an understanding of these changes.

Diffusion-weighted imaging (DWI) takes advantage of the ability of MRI to measure the direction and magnitude of proton diffusion. Various measures of water proton diffusion can be obtained using DWI including the overall magnitude of diffusion (diffusivity) as well as the tendency with which diffusion tends to be directionally dependent. This latter property is measured by calculating a diffusion tensor for each voxel using diffusion tensor imaging or DTI (Basser et al., 1994). As diffusion in white matter tends to be parallel to the direction of fibre tracts and disruption of these tracks can reduce this tendency, DTI provides a measure of white-matter integrity. Fractional anisotropy (FA) is one index of the tendency for water diffusion to occur in a single direction within a voxel. FA is measured on a scale from 0 (random diffusion) to 1 (highly linear diffusion).

Previous studies of changes in white-matter integrity in Alzheimer's disease employing DTI have variably shown relatively increased mean diffusivity and decreased FA in anterior (Bozzali et al., 2002) or posterior (Medina et al., 2006) regions of white matter. Other DTI studies have attempted to elucidate the earliest point at which such changes can be detected by focusing on persons in the earliest clinical stages of the disease (mild cognitive impairment or MCI) or persons at genetic risk for Alzheimer's disease by virtue of carrying an ApoE ε4 allele. A recent study (Rose et al., 2006) found increased diffusivity in the entorhinal and parieto-occipital cortices as well as other regions and decreased FA in the limbic parahippocampal sub-gyral white matter in persons with MCI compared to controls. Another study found that mean diffusivity in the hippocampus in persons with the amnestic sub-type of MCI predicted future progression to dementia (Kantarci et al., 2005). The ApoE ε4 allele, which accounts for about 13% of ApoE alleles in the non-demented elderly caucasian population (Tang et al., 1998), is the most important known genetic risk factor for the development of late-onset Alzheimer's disease. A DTI study demonstrated that non-demented carriers of the ApoE ε4 allele had lower FA in the corpus callosum (particularly in its posterior portion) and in the left posterior hippocampus relative to persons not carrying this allele (Persson et al., 2006).

Though studying persons with MCI and those carrying the ApoE ε4 allele increase our ability to study the earliest stages of Alzheimer's disease, these factors are imperfect predictors of the future development of the disease. Identification of the earliest changes occurring in Alzheimer's disease therefore remains a challenge. Three genes have been identified, alteration of which causes a nearly 100% penetrant, autosomal dominantly inherited form of Alzheimer's disease. Though rare, the study of persons at-risk for inheriting this form of FAD due to mutations in the Presenilin-1 (PS1) or amyloid precursor protein (APP) genes provides a model in which we can use DTI to study white matter in persons in whom the future development of Alzheimer's disease can be reliably predicted. In the current study, we used DTI to quantify white-matter integrity in persons destined to develop Alzheimer's disease by virtue of inheriting mutations causing FAD. We hypothesized that FAD mutation carriers would have decreased FA in white-matter tracts connecting limbic structures (the cingula and columns of the fornix) and in late-myelinating regions (frontal lobe white matter and genu of the corpus callosum) as opposed to early-myelinating regions (inferior splenium of the corpus callosum and corticospinal tracts). We compared mean overall white-matter FA between preclinical (some cognitive impairment present but not demented) and presymptomatic (without identifiable cognitive deficits) FAD mutation carriers and non-carriers as well as FA in these specific areas.

Materials and Methods


Twenty-three persons established to have (n = 2) or be at-risk for (n = 21) known pathogenic PS1 (n = 19) or APP (n = 4) mutations received in-depth clinical, cognitive and imaging assessments. The Clinical Dementia Rating Scale (CDR) was performed by the PI and the Mini-Mental Status Examination (MMSE) by a research assistant. Both investigators performed these assessments blind to subjects’ genetic status in 20 subjects. Mutation status had been clinically established in the two demented patients and one presymptomatic at-risk patient. At-risk persons who tested negative for the FAD mutation present in their family served as controls in this study. All subjects, or their proxies, signed written, informed consent. All study procedures were approved by the Institutional Review Boards at UCLA and the National Institute of Neurology and Neurosurgery in Mexico City.

Magnetic resonance imaging

All images were obtained on the same 1.5 T Siemens Sonata MRI scanner. Four immediately sequential, six-direction diffusion-weighted whole-brain volumes were acquired on all subjects in the plane of the AC–PC line using Echo-Planar Imaging (EPI) with a TR of 6 s, TE of 78 ms. B-values were 0 and 1000 s/mm2. Voxel size was 3 × 3 × 3 mm3. FA maps were then generated using the FMRIB's Diffusion Toolbox (FDT) in FSL software (version 3.2) (Behrens et al., 2003; Smith et al., 2004). The four volumes for each subject were co-registered using FMRIB's Linear Image Registration Tool (FLIRT) to correct for any shift in head position between acquisitions (Jenkinson and Smith, 2001). These four volumes were then averaged together. The averaged B = 1000 s/mm2 volume generated six subvolumes, each representing one of the six diffusion directions. These subvolumes together with the B = 0 (non-diffusion weighted) volume were concatenated into seven volumes in a single file. Eddy current correction was then applied, also using FDT. A diffusion tensor model was fit at each voxel using FDTs DTIFIT, generating fractional anisotropy (FA) maps. Two types of analyses were performed on these FA maps; global and region-of-interest (ROI) analyses of selected areas in white matter.

All subjects also underwent structural MRI in the same session as DTI. Whole-brain T1-weighted images were obtained in the sagital plane using an MP-RAGE sequence (TR = 1900 ms, TE = 4.38 ms, TI = 1100 ms, flip angle 15°). Voxel size was 1 × 1 × 1 mm3. Brain volumes were extracted from the cranium and extracranial tissues using FMRIB's Brain Extraction Tool (Smith, 2002). Gray and white-matter volumes (normalized to total intracranial volume) were calculated using FMRIB's SIENAX (Structural Image Evaluation, using Normalisation, of Atrophy) (Smith et al., 2002). Cross-sectional areas of the columns of the fornix were estimated for each subject using methods similar to Callen et al. (2001). Specifically, the axial slice in which the descending fornices were closest to each other and where the cross-sectional area appeared greatest was chosen and the cross-sectional area measured using FSLView (Fig. 1). Hippocampi were traced by the PI using FSLView according to the protocol of Pantel et al. (2000). Volumes were calculated using FSL software and normalized to intracranial volume. Normalized right and left hippocampal volumes were averaged for the multivariate analysis (see later).

Fig. 1

Axial (left) and sagital (right) views of an exemplary image in which the descending columns of the fornix were localized, delineated and cross-sectional area calculated.

For the global analyses of white matter, brain-extracted whole-brain volumes from T1-weighted images were registered to Talairach space using FLIRT. These registered volumes were segmented into CSF, white and gray matter using FMRIB's Automated Segmentation Tool (Zhang et al., 2001). The FA maps created as above were then also registered to Talairach space using FLIRT. The white-matter volumes from the structural images were then used as masks to delineate the white-matter volumes from the FA maps for each subject (Fig. 2). Average FA values of white matter were then calculated for each subject from these white-matter FA maps.

Fig. 2

Axial (left) and coronal (right) views of a Talairach-registered extracted white-matter FA map.

For all ROI analyses other than the perforant pathway, voxels were selected from the FA maps in native space by two co-investigators blind to subjects’ genetic or clinical status. Specifically, two voxels each representing the splenium and genu of the corpus callosum, the right and left corticospinal tracts at the level of the mesencephalon and bilateral frontal white matter were chosen by one investigator (GB). In regards to the anterior corpus callosum, we desired to obtain a sample that would consistently be in the middle of the structure in order to sample tracts connecting the prefrontal cortices. Therefore, voxels from the genu of the corpus callosum were chosen from the two axial slices on which the angle formed by the left and right sides of the genu appeared the most linear (Fig. 3). For the splenium of the corpus callosum, the second and third lowest slices on which the fibres of the splenium connected in the midline were chosen in order to sample the lower half of the splenium which contains predominantly primary sensory (visual) fibres (Fig. 3, for details see Bartzokis et al., 2006). For both structures two pairs of adjacent voxels were used, one set to the left and one to the right of the structure midline. For analysis of frontal lobe white matter, the voxels chosen were anterior (towards the frontal pole) and superior to the orbitofrontal cortex. To ensure that gray matter was not included in the ROI, the gyrus rectus was identified and the two white-matter voxels chosen were superior to it and inferior and slightly anterior to the cingulate gyrus (i.e. where the frontal white matter still is elongated in the anterior–posterior direction, Fig. 4). For each hemisphere, the averages of the FA values of the two voxels in the splenium, genu and inferior frontal lobes were taken as representative for the respective region.

Fig. 3

Two contiguous corpus callosum voxels were chosen on both sides of the midline in the horizontal planes depicted by arrows. The plane of the genu (right) was defined by its widest part, where axons connecting the prefrontal cortices cross. The plane of the splenium was below its widest part, where axons connecting primary visual cortex cross.

Fig. 4

Two contiguous voxels were chosen in the inferior frontal white matter where it is elongated in the anterior–posterior direction in a slice superior to the orbitofrontal cortex and inferior and slightly anterior to the cingulate gyrus.

The brightest voxel corresponding to the body of the fornices, the right and left cingulum bundle and the area of the perforant pathway were chosen by another investigator (LA). Orthogonal views of the FA maps in native space were employed to pick the brightest voxel corresponding to the descending columns of the fornix (Fig. 5). Again in native space, FA colour maps in which the directionality of diffusion is colour-coded were generated and used to facilitate the selection of posterior cingulum voxels. Representative right and left posterior cingulum voxels were chosen as voxels having blue colour (i.e. preponderence of superior–inferior diffusion) located behind the corpus callosum (identified as a strip of red colour, i.e. preponderance of right–left diffusion). To define FA in the area of the perforant pathways, FA maps and T1-weighted images were co-registered in Talairach space. These images were superimposed and for each subject the sagital slice where the hippocampus was seen in its full length was used to select voxels in the underlying region of the perforant pathway. Two voxels were chosen on each side: the brightest FA voxel underlying the head of the hippocampus (e.g. it's anterior third) and the brightest voxel underlying the anterior part of the body of the hippocampus (or it's middle third, Fig. 6). These were then averaged to provide the value for that side.

Fig. 5

Axial view of an exemplary FA map demonstrating how the brightest voxel corresponding to the descending columns of the fornix was chosen.

Fig. 6

Sagital view of superimposed Talairach-registered T1-weighted image and FA map depicting how voxels in the are of the perforant pathway were selected (see text for details).

Genetic testing

Subjects underwent genetic testing for the FAD mutation for which they were known to be at-risk. They were informed they would be tested but in the context of the research protocol would not be told the result. The option of revealing genetic testing through a genetic counselor outside of the study was offered. Blood samples were coded according to a unique identifier and forwarded to the genetics laboratory. DNA was extracted and ApoE genotyping were performed using standard techniques. The presence of the A431E (number at risk = 11) and L235V (n = 7) substitutions in PSEN1 were assessed using RFLP analyses. The presence of the G206A substitution in PSEN1 (n = 1) and the V717I substitution in APP (n = 4) was assessed directly with bi-directional sequencing.

Statistical analyses

Subjects were categorized as either demented (CDR total scores ≥1), ‘preclinical’ (CDR scores <1) or ‘presymptomatic’ (CDR scores = 0). Within families harbouring PS1 and APP mutations, the age of disease onset tends to be fairly consistent (Fox et al., 1997). Therefore, in order to estimate the subjects’ age relative to the typical age of onset in their families, an ‘adjusted age’ or number of years prior to the typical age of dementia diagnosis in their families was calculated.

Two-tailed independent sample t-tests were performed comparing FA measures, brain volume defined as percent of total intracranial volume, normalized grey and white-matter volumes, cross-sectional area of the columns of the fornix, mean normalized volumes of the right and left hippocampi, MMSE score and adjusted age between pre-clinical FAD mutation carriers (MCs) and non-carriers (NCs). All t-tests were performed twice; once for preclinical subjects and once for presymptomatic subjects.

In order to determine which variables best predicted FAD mutation status in presymptomatic subjects, a logistic regression analysis was performed with mutation status as the dependent variable. Those variables that differed between mutation carriers and non-carriers with total CDR scores of 0 by t-test at the 0.05 significance level were entered as independent variables.

To establish what the strongest determinants of fornix FA were, linear regression analyses were performed with fornix FA as the dependent variable and fornix area, normalized white and gray-matter volumes, normalized mean hippocampal volume and FAD mutation status as predictor variables. Data from all 23 subjects were included in the logistic and linear regression analyses. Because of the relatively low numbers of subjects available for this type of study, P-values of 0.05 were used throughout without adjustments for multiple comparisons. All analyses were performed using the Statistical Package for the Social Sciences, Version 11.0.2.


Fourteen subjects were FAD MCs and nine were NCs (Fig. 7). Of the 14 FAD MCs, two were demented (CDR scores of 2 and 3). All 23 subjects were female except for one NC and two MCs (1 CDR = 0, 1 CDR = 0.5). One NC was approximately 18 years older than the typical age of diagnosis in her family and therefore was excluded from the t-tests comparing preclinical and presymptomatic subjects as she was no longer ‘at-risk’. Of the remaining 20 preclinical at-risk subjects, 12 were MCs (CDR = 0.5 in 4, 0 in 8) and 8 were NCs (CDR = 0.5 in 1, 0 in 7). Therefore, 15 subjects (8 MCs and 7 NCs) were presymptomatic (CDR = 0). There were no significant differences in age, adjusted age or MMSE score between the 12 preclinical MCs and 9 NCs (Table 1) nor between the 8 presymptomatic MCs and 7 NCs (Table 2).

View this table:
Table 1

Demographic data for the 12 preclinical (CDR total score <0.5) FAD MCs (mutation carriers) and 8 non-carriers (NCs)

FAD MCs (n = 12)FAD NCs (n = 8)P-values
Mean age in years (SD)35 (6.4)36 (6.2)0.64
Mean adjusted age (SD)−11 (6.8)−12 (7.2)0.73
Mean MMSE score (range)27.1 (21–30)28.1 (24–29)0.40
Gender (# female)107n.s.
  • Note: MMSE = Mini-Mental Status Examination

View this table:
Table 2

Demographic data for the eight presymptomatic (CDR total score = 0) FAD MCs (mutation carriers) and seven non-carriers (NCs)

FAD MCs (n = 8)FAD NCs (n = 7)P-values
Mean age in years (SD)32 (6.4)36 (6.6)0.34
Mean adjusted age (SD)−13 (6.0)−12 (7.7)0.76
Mean MMSE score (range)28.8 (28–30)28 (24–29)0.29
Gender (# female)76n.s.
  • Note: MMSE = Mini-Mental Status Examination.

Fig. 7

Chart indicating the break-down of subjects by clinical and FAD mutation status. CDR = Clinical Dementia Rating scale.

There was no significant difference between MCs and NCs in regards to overall brain size, white and gray-matter volumes or mean hippocampal volumes in either pre-clinical or presymptomatic subjects. Cross-sectional area of the fornix was smaller in MCs in both the preclinical (16.4 versus 20.5 mm2, P = 0.043, Table 3) and presymptomatic subjects (16.6 versus 21.3 mm2, P = 0.045, Table 4, Fig. 8). In preclinical and presymptomatic subjects, mean FA values of MCs were numerically lower in all ROIs except for the inferior splenium of the corpus callosum and the right and left corticospinal tracts (Tables 3 and 4). Mean overall FA was statistically lower in the preclinical (0.30 versus 0.32, P = 0.045) but not the presymptomatic subjects (P = 0.070). FA in the columns of the fornix was also lower in preclinical (0.53 versus 0.66, P = 0.012) and presymptomatic (0.50 versus 0.66, P = 0.008, Fig. 4) subjects. In the area of the perforant path, FA was lower bilaterally in preclinical MCs (0.28 versus 0.36, P = 0.028 on the right side, 0.29 versus 0.36, P = 0.027 on the left side). FA was non-significantly decreased in the perforant pathway in presymptomatic MCs (P-values of 0.067 on right and 0.076 on the left). FA in MCs was also lower in the white matter of the left frontal lobe (0.50 versus 0.56, P = 0.024 for preclinical subjects, 0.49 versus 0.56, P = 0.045 for presymptomatic subjects).

View this table:
Table 3

MRI variables in the 12 preclinical MCs versus 8 NCs

FAD MCs (n = 12)FAD NCs (n = 8)P-values
Brain size as percent of intracranial volume86%88%0.125
Gray-matter vol (mm3)835 646844 8910.752
White-matter vol (mm3)776 296803 3280.285
Mean R and L hippocampal volume (mm3)328331350.408
Fornix area, mm2 (SD)16.4 (4.2)20.5 (4.0)0.043
Mean whole brain WM FA (SD)0.30 (0.02)0.32 (0.02)0.045
FA fornix (SD)0.53 (0.09)0.66 (0.11)0.012
FA R cingulum (SD)0.49 (0.08)0.56 (0.07)0.065
FA L cingulum (SD)0.49 (0.07)0.52 (0.08)0.262
FA R perforant path area (SD)0.28 (0.03)0.36 (0.08)0.028
FA L perforant path area (SD)0.29 (0.06)0.36 (0.06)0.027
FA genu of c.c. (SD)0.77 (0.06)0.82 (0.05)0.089
FA splenium of c.c. (SD)0.83 (0.09)0.81 (0.10)0.551
FA R frontal white matter (SD)0.51 (0.06)0.54 (0.07)0.389
FA L frontal white matter (SD)0.50 (0.05)0.56 (0.06)0.024
FA R CS tract (SD)0.70 (0.07)0.69 (0.11)0.861
FA L CS tract (SD)0.72 (0.07)0.70 (0.08)0.622
  • Note: FA = fractional anisotropy, R = right, L = left, c.c. = corpus callosum, WM = white matter, CS = corticospinal. P-values represent results of two-tailed independent sample t-tests. Values that are different at the P < 0.05 level are shown in bold italics.

View this table:
Table 4

MRI variables in the 8 presymptomatic MCs versus 7 NCs

FAD MCs (n = 8)FAD NCs (n = 7)P-values
Brain size as percent of intracranial volume87%88%0.257
Gray-matter vol, mm3855 311839 7050.631
White-matter vol, mm3767 920811 0230.182
Mean R and L hippocampal volume, mm3330731030.360
Fornix area, mm2 (SD)16.6 (4.4)21.3 (3.5)0.045
Mean Whole brain WM FA (SD)0.30 (0.02)0.32 (0.02)0.070
FA fornix (SD)0.50 (0.06)0.66 (0.12)0.008
FA R cingulum (SD)0.50 (0.09)0.56 (0.08)0.195
FA L cingulum (SD)0.49 (0.07)0.54 (0.07)0.195
FA R perforant pathway area (SD)0.29 (0.03)0.36 (0.09)0.067
FA L perforant pathway area (SD)0.29 (0.07)0.36 (0.07)0.076
FA genu of c.c. (SD)0.77 (0.07)0.82 (0.05)0.116
FA splenium of c.c. (SD)0.84 (0.08)0.81 (0.10)0.554
FA R frontal white matter (SD)0.52 (0.04)0.53 (0.07)0.763
FA L frontal white matter (SD)0.49 (0.05)0.56 (0.06)0.045
FA R CS tract (SD)0.72 (0.05)0.71 (0.11)0.780
FA L CS tract (SD)0.75 (0.06)0.69 (0.09)0.201
  • Note: FA = fractional anisotropy, R = right, L = left, c.c. = corpus callosum, WM = white matter, CS = corticospinal. P-values represent results of two-tailed independent sample t-tests. Values that are different at the P < 0.05 level are shown in bold italics.

Fig. 8

Scatterplots of the FA (left) and cross-sectional area of the columns of the fornix (right) in 8 FAD MCs and 7 NCs.

In the logistic regression analysis, FA of the fornix was the best predictor of mutation status (P = 0.005), followed by left frontal white FA (P = 0.007), and cross-sectional area of the fornix (P = 0.031). In the linear regression model, white-matter volume was the greatest predictor of fornix FA (P = 0.002), followed by hippocampal volume (P = 0.023), and FAD mutation status (P = 0.032). Cross-sectional area of the fornix (P = 0.127) and gray-matter volume (P = 0.339) were not significant predictors of fornix FA.


In the current study, we found that FA is reduced in specific areas of white matter in persons carrying FAD mutations prior to the development of symptoms of dementia. The global analysis suggests that average whole-brain white-matter FA is reduced in pre-clinical FAD mutation carriers, and the ROI analyses demonstrated that this effect was evident in the fornix and white matter underlying the orbitofrontal region during the presymptomatic stage of the illness. This effect was not present in the early-myelinating regions of splenium of the corpus callosum or the corticospinal tracts at the level of the midbrain. This is consistent with prior studies demonstrating that limbic projections and pathways connecting the frontal lobes are preferentially affected in the course of Alzheimer's disease (Braak and Braak, 1996; Bartzokis et al., 2004). The greatest reduction we found using our methods was in the columns of the fornix and decreased FA in the fornix was a more sensitive measure for predicting FAD mutation status in presymptomatic persons than were cross-sectional area of the fornix, overall brain size, white and grey-matter volumes or hippocampal volumes. Though these measures were not different between preclinical or presymptomatic FAD MCs and NCs, the observation in the linear regression analysis that overall white-matter volume was highly related to fornix FA suggests that low FA in this structure may represent an earlier manifestation of white-matter deterioration than can be detected by comparison of volumes.

Neuronal death occurring secondary to excessive production or decreased degradation of toxic forms of beta-amyloid is the most widely accepted aetiological mechanism for Alzheimer's disease. Nonetheless, multiple studies in both animal models and humans have demonstrated that the integrity of white matter is affected during the course of Alzheimer's disease and indeed can be an early event. Whether myelin pathology is primary or if axonal disruption due to neuronal loss is causative is uncertain but it is clear that changes in white matter occur in Alzheimer's disease and they can be measured using various techniques.

In previous investigations of living subjects, this has been studied by measuring white-matter volumes (Bartzokis et al., 2001), transverse relaxation rates (Bartzokis et al., 2003) and by employing DTI (Medina et al., 2006). The water content of white matter increases as its myelin component decreases in various pathological states. The transverse relaxation rate (R2) provides an index of the water content of a tissue with lower values reflecting higher water concentrations. Decreased R2 values are seen in areas of pathology in demyelinating diseases (Papanikolaou et al., 2004) as well as in normal aging (Bartzokis et al., 2001, 2004), individuals at-risk for Alzheimer's disease (Bartzokis et al., 2006) and Alzheimer's disease (Bartzokis et al., 2003). Prior studies using this technique have demonstrated that R2 is below normal in the frontal lobe white matter of persons with Alzheimer's disease (Bartzokis et al., 2003) and suggest that R2 of the frontal lobes and genu of the corpus callosum decreases with age at a faster rate in carriers of the ApoE ε4 allele (Bartzokis et al., 2006). It has been hypothesized that this greater involvement of late-myelinating areas is due to a selective vulnerability of oligodendrocytes in these regions (Bartzokis et al., 2004). Regardless of the underlying cause, our finding of significantly decreased FA in the white matter of the frontal lobe of FAD MCs and trends towards this in the genu of the corpus callosum (despite our small number of subjects) but not in the splenium or corticospinal tract provides convergent evidence for selective involvement of these late-myelinating areas in early Alzheimer's disease.

Neuronal loss is earliest and most severe in the medial temporal lobe in Alzheimer's disease. It might therefore be expected that the most significant alterations in white matter would be found in the tracts connecting this region and other limbic (Callen et al., 2001) and non-limbic brain regions. A previous pathological study demonstrated a 60% loss of neurons in layer II of the entorhinal cortex in persons dying with a CDR score of 0.5 (Gomez-Isla et al., 1996). As the neurons in this layer give rise to the perforant pathway that projects to the hippocampus, one might expect that the FA in this area might decline as result. Indeed, a prior DTI study demonstrated that the intervoxel coherence in the area of the perforant pathway was decreased in persons with MCI compared to controls (Kalus et al., 2006). Using our methods, we were able to detect significant decreases in FA in the area of the perforant pathway bilaterally in preclinical but not presymptomatic FAD MCs. The fornix is the predominant outflow tract of the hippocampus and connects it with the septal nuclei as well as the mamillary bodies in the hypothalamus. FA of the cingulum bundle, another outflow tract of the medial temporal lobe that connects it with the cingulate gyrus, was also reduced, albeit non-significantly so, in presymptomatic MCs. Our finding of decreased FA in the columns of the fornix of persons inheriting but not yet affected by FAD confirms its early involvement in the disease and suggest that FA of the fornix might be a useful index of early disease in sporadic Alzheimer's disease. We know of one prior abstract in which FA of the columns of the fornix was reported to be decreased in persons with Alzheimer's disease (Bozoki et al., 2004).

A prior report found that non-demented persons with an ApoE ε4 allele had decreased FA in the posterior corpus callosum and the left posterior hippocampus compared to controls without this allele (Persson et al., 2006). It is unclear if this represents a consequence of the early changes of Alzheimer's disease of a trait associated with the ApoE ε4 genotype. In our multiple linear regression analysis, we did not find an robust effect of ApoE genotype on fornix FA (data not shown) but our study was likely underpowered to identify such an effect if present (4 of 23 subjects carried an ApoE ε4 allele.

A possible confounder in this study is that atrophy of the fornix in its intraventricular course would also contribute to decreased FA. That is, the smaller the fornix, the more CSF would be included in a given voxel size (3 mm in our study). Since CSF has an FA near 0, the partial volume effect of increased CSF in the voxel would cause the mean FA within that voxel to be decreased. As the voxels chosen to represent the fornix in our study did not have FAs dramatically lower than those from the parenchymal white matter, we do not feel this effect was pronounced in our study. Also, the regression analyses in our study suggest that cross-sectional area of the fornix influences FA but does not completely explain the decreases seen in FAD MCs. We speculate that the significant drop we see is the composite result of decreased FA within the fornix with a possible smaller contribution from diminished fornix size in these subjects. Replication of our study using higher-resolution DTI would help resolve this question.

An advantage to studying relatively young persons with or at-risk for FAD is the paucity of co-morbid illness (e.g. hypertension) that might contribute to confounding cerebral pathology (e.g. ischaemic changes). None of the subjects in our study had notable non-Alzheimer's cerebral pathology and they thus provide a relatively ‘pure’ model of the disease. However, there may be limitations to the degree to which the findings of our study of FAD are generalizable to sporadic Alzheimer's disease. Other than age of onset, there are additional clinical (Assini et al., 2003) and pathological (Houlden et al., 2000; Takao et al., 2001) differences between FAD and sporadic Alzheimer's disease. Abnormal white matter may represent a quality of FAD not directly related to the incipient pathology of Alzheimer's disease. Persons carrying FAD mutations may have life-long white-matter abnormalities (a trait) rather than acquired changes (a state) of developing Alzheimer's disease characterized by decreased fractional anisotropy. Though there is a suggestion of accelerated decrease of FA with advancing age in MCs (Fig. 9), the data presented in Fig. 9 are also compatible with even earlier decreases in FA in this population.

Fig. 9

Scatterplots of mean white-matter FA in FAD NCs and MCs in relation to absolute age (left and middle) and age relative to the family-specific median age of disease diagnosis (adjusted age or ‘relative age’) in MCs (right).

Evidence for developmental abnormalities in PS1-related FAD comes from both animal and human studies. Embryos of transgenic mice in which both copies of the PS1 gene have been knocked out have abnormalities of neuronal migration and differentiation characterized by disorganization of the cerebral cortex (Handler et al., 2000). Also, a patient with a PS1 mutation who developed young-onset Alzheimer's disease and came to autopsy was noted to have ectopic neurons in the white matter (Takao et al., 2001). Though we do not know of any such structural abnormalities having been reported in the white matter of persons dying with the FAD mutations included in this study (Mullan et al., 1993; Cochran et al., 2001), such a possibility exists. If present, one might expect that this could account for the differences seen.

If PS1 mutations cause aberrant neuronal migration or other white-matter abnormalities, it would be of interest to know the effect of APP mutations on white-matter integrity as to our knowledge white-matter abnormalities have not been described in pathological specimens of persons dying with APP mutations. Notably, four subjects in the current study were at-risk for the V717I substitution in the APP gene and among them the tendency for MCs to have lower mean white-matter FA was maintained. Unfortunately, due to these small numbers and the necessity to maintain subject anonymity with regard to mutation status little more can be said or concluded about this.

Among persons with the A431E substitution in the PSEN1 gene, many develop significant spastic tetraparesis as the dementia progresses (Murrell et al., 2006). In pathological studies, tetraparesis occurring in FAD has been related to disproportionate involvement of the motor cortex with amyloid pathology with presumptive consequent degeneration of the corticospinal tracts. None of the 11 subjects in our study at-risk for this mutation had significant para- or tetraparesis but it is notable that one such MC who had a CDR score of 0.5 was hyperreflexic and had relatively low FA of the corticospinal tract in the cerebral peduncle (data not shown). This suggests that DTI may have utility in detecting and possibly in predicting this complication.

Our data indicate that loss of white-matter integrity, as indexed by FA acquired through DTI, is an early feature in FAD. Furthermore, this loss is somewhat selective, disproportionately affecting tracts that connect limbic structures and the frontal lobes. Though there are limitations to our study regarding the resolution of the DTI technique employed and the ability to generalize from FAD to the more common form of late-onset Alzheimer's disease, we feel that further investigations of early changes in white matter, particularly of the FA of the columns of the fornix, are merited as potential indicators or even predictors of Alzheimer's disease status.


This study was supported by PHS K08 AG-22228, California DHS #04-35522, M01-RR00865, General Clinical Research Centers Program and the Shirley and Jack Goldberg Trust. Further support for this study came from Alzheimer's Disease Research Center Grant AG-16570 from the National Institute on Aging, an Alzheimer's Disease Research Center of California grant, and the Sidell Kagan Foundation.


  • Abbreviations:
    amyloid precursor protein
    clinical dementia rating
    diffusion tensor imaging
    diffusion-weighted imaging
    familial Alzheimer's disease
    mild cognitive impairment


View Abstract