Within the medial temporal lobe (MTL), the entorhinal cortex (ERC) is the earliest site of AD pathology and volume loss (Bobinski et al., 1999; Braak and Braak, 1996; Gomez-Isla et al., 1996), present even in predementia stages of the disease (Gomez-Isla et al., 1996). Very early in the progression of AD, individual layers, rather than whole regions of cortex, show marked loss of neurons (Gomez-Isla et al., 1996). The restriction of AD pathology to specific laminae of the ERC was powerfully demonstrated by Gomez-Isla and colleagues (Gomez-Isla et al., 1996) who revealed that mild AD subjects had a 60% reduction in cell count within layer II of the ERC and 40% reduction in layer IV compared to controls. The authors concluded that even in the mildest stage of AD, the marked decrement of layer II neurons differentiated disease-related pathology from normal aging. High-resolution MRI confirmed the presence of sub-regionally specific neuronal dysfunction that was greatest within the ERC in AD (Small et al., 2000). These studies suggest that cortical thickness, or the distance between inner and outer gray matter boundaries, may be a more sensitive measure of AD-related pathology than traditional volumetric methods that assess total size. Such findings raise the possibility that in-vivo measures of cortical thickness might identify differences in subjects carrying a genetic risk for AD but without current symptoms.

MRI scanning was performed on a 3-Tesla scanner (General Electric, Waukesha, Wis.). Echo-planar shimming sequence achieved approximately 0.3 ppm r.m.s. field inhomogeneity. A Fast Spin Echo (FSE) sagittal sequence [TR 6000/TE 17&85/4 mm thick, 1 mm spacing/FOV 20 cm/1 NEX/ 26 slices] was used for spatial localization. A high-resolution oblique coronal T2 FSE structural sequence was acquired for structural segmentation and unfolding procedures [TR 3000/TE 41/ 3mm thick, 0 mm spacing/FOV 20 cm/ 2 NEX/16 slices, in-plane voxel size 0.39×0.39mm]. These images were acquired perpendicular to the long axis of the hippocampus to minimize slice-to-slice variation in anatomy (Zeineh et al., 2001). A T-1 weighted 3-dimensional volume scan [SPGR, T1 500/TE 3.7/FOV 20/1 NEX] was also acquired to visually serve as a guide in sulcal visualization during segmentation procedures in the same way an atlas is used as a visual reference.

Structural unfolding was used to create a continuous gray matter strip within the MTL using the high-resolution structural images, described in detail by Zeineh and colleagues (Zeineh et al., 2000; Zeineh et al., 2003). The unfolding procedure is designed to take the entire volume of pixels within the MTL defined as gray matter in a 3-D strip and flatten it so the volume is visible as a single 2-D sheet. First, we manually defined white matter and cerebrospinal fluid (CSF) within the MTL (Figure 1A). We segmented the following as “white matter”: (1) white matter in the parahippocampal gyrus throughout the fill rostrocaudal extent, (2) white matter on the medial aspect of the occipitotemporal sulcus, (3) CSF in the inferior horn of the lateral ventricle, and (4) the fornix. We segmented the following as “CSF”: (5) CSF in the collateral sulcus and the hippocampal fissure, (6) the ambient cistern (including the wing) and the adjacent posterior cerebral artery and basal vein, and (7) the boundaries of the segmentation. We next used a region-expansion algorithm to grow 18 connected layers of gray matter from the white-matter edge, stopping at the boundary of CSF and covering all of the gray matter pixels of interest. The final product was a gray matter ribbon that included cornu ammonis (CA) fields 1, 2, and 3, the dentate gyrus (DG), subiculum (Sub), ERC, perirhinal cortex (PRC), parahippocampal cortex (PHC), and the fusiform gyrus (Fus). As in previous studies using the unfolding procedure (Eldridge et al., 2005; Zeineh et al., 2000; Zeineh et al., 2003), we identified these regions using a rule-based decision schema in concert with histological and MRI atlases (Amaral and Insausti, 1990; Duvernoy, 1998; Mai, 1997). Unfolding computations were performed using modified versions of mrUnfold software [available at http://sourceforge.net/projects/mtl-unfolding/] (Engel et al., 1997). As shown in color on the MRI image (Figure 1B) and explained in detail in the figure legend, we delineated the boundaries of the following subregions on the in-plane image based on reference to atlases (Amaral and Insausti, 1990; Duvernoy, 1998; Mai, 1997) : Fus, PHC, PRC, ERC, Sub, CA1 and CA fields 2, 3 and dentate gyrus (CA23DG). Because of the convoluted shape of CA fields 2, 3 and dentate gyrus in the anterior hippocampus the regions were treated as a single entity. The thickness map of one subject in 3-dimensional space, superimposed on an anatomical image is shown in Figure 1C.

All demarcations of structural boundaries were traced on the original coronal images and projected mathematically to the corresponding coordinates in flat map space. We performed reverse transformations for each subject and visually compared this reverse transformation to the original data to confirm that the data was in the correct space (Zeineh et al., 2000). Prior work has confirmed that this algorithm produces topographically correct unfoldings with minimal levels of distortion (Wandell et al., 1996; Zeineh et al., 2000; Zeineh et al., 2003).

Thickness maps were produced in 3-dimensional (3D) space as follows: for each gray matter voxel in each layer of the manifold we computed the distance to the closest non-gray matter voxel (white matter or CSF). The middle layer of any given section of cortex will be equidistant from the adjacent white matter and CSF and will have the greatest distance values to the closest non-gray voxel. For each 2-dimensional (2D) flat voxel, we took the maximum of the distance values of the corresponding 3D voxels across all layers (thus effectively extracting this middle layer which will be the maximum) and multiplied by two to arrive at a thickness value. In the 2D maps, thickness was represented in voxel intensity.

We conducted a region of interest (ROI) analysis of cortical thickness for the following subregions: Fus, PHC, PRC, ERC, Sub, CA1, and CA23DG. We computed the mean thickness in each subregion by averaging the thickness across all 2-D voxels within each ROI. To average across hemispheres in each subject, we weighted the mean thickness according to the number of 2D voxels within that subregion in order to minimize the potentially confounding effects of surface area asymmetries. This weighted mean thickness for each subregion was used in our statistical analyses. We calculated the total hippocampal thickness in each subject by performing a similarly weighted average across all of the subregions. In order to assess volume, we calculated the volume of the voxels for each given region in 3-D space. In addition, the total mean cortical volume was calculated by averaging across all of these subregions within the MTL.

ICCs for cortical thickness measurements in MTL subregions of APOE-4 carriers were calculated for all subregions (CA23DG: 0.973, CA1: 0.981, Sub: 0.988, ERC: 0.992, PRC: 0.970, PHC: 0.994, Fus: 0.986). ICC results from APOE-3 carriers were similar (CA23DG: 0.969, CA1: 0.974, Sub: 0.982, ERC: 0.983, PRC: 0.989, PHC: 0.987, Fus: 0.991). Averaged over all regions, the ICC was 0.983 for APOE-4 carriers and 0.982 for APOE-3 carriers. SEM values were also similar between the two genetic risk groups (APOE-4: 0.044; APOE-3: 0.038). We also determined ICC values for volumetric measurements in APOE-4 carriers (CA23DG: 0.967, CA1: 0.985, Sub: 0.968, ERC: 0.991, PRC: 0.974, PHC: 0.982, Fus: 0.993) and APOE-3 carriers (CA23DG: 0.996, CA1: 0.983, Sub: 0.972, ERC: 0.978, PRC: 0.993, PHC: 0.982, Fus: 0.989). Total ICC and SEM values over all of these subregions for the genetic groups were similar for hippocampal volumetric measurements (APOE-4: 0.980; APOE-3: 0.981). SEM values for both genetic risk groups were also similar (APOE-4: 0.047; APOE-3: 0.045).

Structural MRI studies of hippocampal damage have traditionally used volumetry to analyze cortical structure. The laminar organization of human cerebral cortex, however, and the specificity of neuronal loss within layers II and IV in early AD (Gomez-Isla et al., 1996; Hyman et al., 1984) favors the use of measures of cortical thickness over volume, especially in an asymptomatic population in whom cortical changes, if present, are likely to be extremely subtle. The affected cells in these layers connect the hippocampus with association cortices, thalamus, hypothalamus and basal forebrain – they are crucial to memory formation and the cause of early memory impairment in the disease (Hyman et al., 1984). Dynamic patterns of gray matter loss in AD have been more recently revealed in vivo using neuroimaging (Barnes et al., 2004; Thompson et al., 2003). Patients with AD had a total gray matter loss rate of 5% per year, on average, compared to healthy controls who lost only 1% gray matter per year (Thompson et al., 2003). Abnormalities in cortical thickness might reflect underlying neuropathological abnormalities in the intrinsic structure of cortical laminae (Kuperberg et al., 2003; von Economo, 1929).

Volumetric measurements are the most commonly used measures of structural change in AD. However, our volumetric measurements detected no significant differences between the two APOE groups, either in individual subregions or total volume within the MTL. In studies that have examined the hippocampal volume in cognitively-intact APOE-4 carriers, non-significant gray matter volume changes in the expected direction may suggest that with either more subjects or slightly older subjects significant differences would have been seen. Of the volumetric studies that have not found significant differences in the hippocampal volumes of APOE-4 subjects, many have reported trends towards lower hippocampal volumes in the at-risk group (Jack et al., 1998; Reiman et al., 1998; Schmidt et al., 1996; Tohgi et al., 1997). Schmidt and colleagues reported that while there were no significant differences in the hippocampal volume of APOE-4 carriers, they found the normalized hippocampal volumes in their at-risk group to be 7% smaller (Schmidt et al., 1996). Congruent with these findings, Tohgi and colleagues reported 9% reductions in their at-risk subjects (p=0.06) (Tohgi et al., 1997). There was a non-significant, 8% reduction in non-demented carriers of the APOE-4 allele in a study by Reiman and colleagues (Reiman et al., 1998). The study by Plassman and colleagues reported significantly lower hippocampal volumes in APOE-4 carriers and suggested that the at-risk allele may be associated with morphological differences in the absence of clinical symptoms (Plassman et al., 1997).

Our results also reveal a trend using volumetric measurements, although these differences in the genetic groups were not significant. Hippocampal volumes in our study were, on average, 2.7% lower in APOE-4 carriers. When we evaluated cortical thickness in these same cognitively intact subjects, the average for the entire MTL region was reduced by 6.8% (p=0.0015). However, on a subregional scale cortical thickness was reduced by 14.8% (p=1.0e⁻⁶) in ERC and 12.6% (p=6.83e⁻⁵) in the Sub. Because of the complex curvature of the MTL, which is “rolled,” traditional volumetric methods cannot easily differentiate the hippocampal subregions, which differ in how and when they are affected by incipient AD. Further, volumetric measures may be more variable as they are more sensitive to baseline differences in the size, and may measure the effects of intra-laminar cell loss less directly than cortical thickness. Our measurements of cortical thickness were made solely on the basis of gray matter and did not take into account white matter degeneration. Several studies have evaluated white matter loss versus gray matter loss in AD-related pathology and have reported gray matter to be more sensitive to AD-related pathology (Fotenos et al., 2005; Thompson et al., 2003). While white matter lesions have been found in both normal aging and in AD (Burns et al., 2005), Thompson and colleagues found no significant differences in white matter loss rates between AD patients relative to control subjects (Thompson et al., 2003).

There are several potential explanations for these results that should be addressed. Because volume is a product of thickness by area, there may be a large amount of variation in cortical surface area across subjects that may not be related to AD or genetic risk. This makes overall volumetric measures less sensitive as they include this type of area variation (which may not have a significant effect on cognition). It is worth noting that because the resolution of our images results in anisotropic voxels (0.4×0.4×3mm), it is possible that the highly elongated voxel length results in a lapse in our ability to observe gyral variations that take place within the 3 mm thick slices. However, because reducing slice thickness would require sacrificing in-plane resolution that was critical to our ability to visualize gray-white matter boundaries within the MTL, we advocate our choice of maximizing in-plane resolution as the most beneficial for this segmentation procedure. To compensate for this limitation, we did acquire and do recommend concomitant 3-D isotropic T1-weighted images with lower in-plane but higher through-plane resolution which aid in visualization and segmentation. Also, because thickness is a component of volume, one might expect that, keeping all else constant, a reduction in thickness should result in decreased volume. However, given the relative size of the average thickness within the hippocampus versus the total volume metrics, the impact of cortical thinning on total volume would be expected to be quite minimal. A significant difference in cortical thickness in a subregion would only minimally change the volume as a whole. Additionally, thinning might be quite localized and its effect washed out in a more global measure. This may explain the inconsistent results that volumetric studies have revealed in similar APOE-4 subjects in the past (Jack et al., 1998; Plassman et al., 1997; Reiman et al., 1998; Schmidt et al., 1996; Tohgi et al., 1997). Also, although we corrected for the size of individual subregions by weighting them according to the number of voxels, we did not correct for whole-brain cortical thickness or volume, thus these findings may not be limited to the MTL. Finally, it remains unclear from our data whether these differences reflect a baseline difference in the brains of APOE-4 carriers, and perhaps a risk factor for AD, or whether they actually reflect the beginnings of progressive cell loss that will eventually lead to AD.