Neuritic plaques and neurofibrillary tangles begin accumulating years before the earliest clinical features of Alzheimer’s disease (e.g., memory loss) manifest themselves (Hyman, 1997
; Price, 1997
). The hippocampus is affected early in the disease process by neurofibrillary tangle accumulation, which spreads in a well-defined trajectory. The first tangles are typically seen in the entorhinal cortex; next, they spread to the CA1 and subicular areas, then to the CA2 and 3 and finally the CA4 areas of the hippocampal formation before invading the neocortex (Schonheit et al., 2004
). The subfield-specific variability in tangle density and neuronal loss has been well documented. The subiculum and CA1 were reported to be the most affected and CA3 and CA4 the least affected hippocampal subfields in post-mortem AD brains (Bobinski et al., 1995
). One post-mortem study of 39 nondemented elderly reported that 56% of the subjects had neurofibrillary pathology in the transentorhinal and entorhinal areas (Braak stages I and II), 28% also had moderate involvement of the CA1 and subiculum (Braak stage III) and 13% demonstrated both heavy hippocampal involvement and evidence of neocortical tangle pathology (Braak stage IV and higher) (Knopman et al., 2003
). Braak stages I–IV have been assigned to more than 80% of MCI subjects at autopsy (Petersen et al., 2006
). Prior MRI and PET data also suggest the vulnerability of the hippocampal formation in the early stages of AD. This has been observed in studies predicting the decline from MCI to AD and most recently from NL to MCI and AD (Apostolova et al., 2006b
; de Leon et al., 1993
; den Heijer et al., 2006
; Mosconi et al., 2007
; Rusinek et al., 2003
We found that on average 3 years prior to the diagnosis of amnestic MCI and 6 years prior to the diagnosis of probable AD, subjects who are predestined to become demented already show disease-associated changes. Our study is the first to demonstrate that these changes initially localize to the subiculum and the CA1 subfield—the two hippocampal areas pathologically affected very early in the disease course. Upon imaging 3(±1) years later, subjects normal at baseline but now meeting MCI criteria manifested hippocampal atrophy spread to the CA2 and CA3 subfields. To our knowledge, this is the first longitudinal neuroimaging study of pre-clinical AD that maps this highly specific and orderly progression of disease-associated changes through the hippocampus. Our findings suggest CA1 and subicular involvement is predictive of cognitive decline to MCI while progressive involvement of the CA1 and subiculum and atrophy spread to the CA2-3 subfield in amnestic MCI is predictive of future diagnosis of AD.
The annual atrophy rates in the NL-NL and the NL-MCIAD
group (NL-NL 1.1–2% and NL-MCIAD
3.2–3.7% annually) are comparable to those already reported in the literature (Jack et al., 2000
). Both the baseline and the 3-year follow-up hippocampal comparisons demonstrated restricted areas where the NL-MCIAD
group showed greater radial distance (thicker hippocampus) relative to the NL-NL group ( and , bottom row). These differences were not statistically significant. One plausible explanation is between-subject variability of hippocampal shape (Carmichael et al., 2005
). However, a reassuring observation is that at 3-year follow-up, as the groups become more divergent in their cognitive performance, these areas appear much smaller relative to the baseline comparison where both groups exhibited normal cognition.
In addition we observed restricted areas of apparent gain in radial distance over time that were more pronounced in the NL-NL vs. the NL-MCIAD group (see ). These changes were not statistically significant. There are two possible reasons for the occurrence of apparent positive change in the radial distance over time. Although presently accepted as the gold standard, hand tracing of the hippocampus or any other region of interest is not perfectly reproducible. This could lead to net gains in radial distance even when no change has occurred (i.e., in the absence of true effect). Such gains should not be interpreted as meaningful unless they beat the 0.05 significance level when corrected for multiple comparisons. Furthermore subtle movement artifacts would likewise decrease the signal-to-noise ratio and could result in non-biological effects.
In the current study we came across one occurrence of a hippocampal fissure—in the left hippocampus of one subject from the NL-MCIAD group at baseline and at 3-year follow-up. We decided to use the most conservative approach in respect to our hypotheses and did not extend the trace in the fissure (rather we have rounded the hippocampal contour and included the fissure in the trace). This ultimately leads to an increased left hippocampal volumetric measurement in this specific subject and works against our hypothesis of increased atrophy in the NL-MCIAD relative to the NL-NL group at baseline and follow-up.
Two other hippocampal imaging studies used a different but conceptually related computational anatomy technique for 3D hippocampal modeling to study the effects of normal aging vs. questionable AD (defined as subjects whose Clinical Dementia Rating scale (CDR) scores were 0 and 0.5, respectively, including both MCI and mild AD patients). One of the studies reported that subjects deemed to be cognitively normal over 2 years (i.e., who had CDR = 0 both at baseline and follow-up) showed progressive atrophy of the anterior CA1 region (localized to the hippocampal head) and the subiculum. Subjects with questionable AD at baseline (i.e., CDR = 0.5) showed additional progressive atrophy of the posterior CA1 area (Wang et al., 2003
). Another study followed 49 nondemented subjects with CDR of 0 for an average of 5 years. They reported isolated progressive atrophic changes of the left CA1 subfield in subjects who declined from CDR = 0 to CDR of 0.5 (cognitively normal to questionable AD) (Csernansky et al., 2005
). Our results build on these findings by showing the spread of disease-associated changes over time, through the hippocampal subfields, following the well-documented pathologic progression of AD-type pathology. These results raise hopes for the utility of this mapping technique not only as a sensitive diagnostic and prognostic tool but also as a surrogate marker for future disease-modifying clinical trials.
Two studies to date have segmented out the hippocampal subfields in cognitively normal elderly and AD subjects (Csernansky et al., 2005
; Mueller et al., 2007
) and a third one in young adults (Zeineh et al., 2003
). At 1.5 T there was suboptimal resolution for subfield identification (Csernansky et al., 2005
). In a recent 4 T study using 2 mm thick T2 hippocampal image sections Mueller et al. resolved the CA1, CA2, subiculum and the combined CA3-4 areas and traced them manually (Mueller et al., 2007
). While identification of individual hippocampal layers at 4 T was still not possible, the authors used a set of reliable anatomical landmarks for boundary approximation. The authors did not resolve and trace the subfields of the whole structure; they estimated the subfield volumes from a set of three contiguous slices immediately posterior to the hippocampal head under the assumption that this 3-slice measurement correlates well with the volume of the whole subfield. The study reported that the three AD subjects had smaller CA1 and subiculum measurements relative to their age-matched control counterparts, while advanced age seemed to exert an effect on the CA1 subfield (Mueller et al., 2007
Several strengths and weaknesses of our study need to be recognized. Major strengths of the study are the well-characterized patient population and the longitudinal design. As MCI is heterogeneous, a strength of the study design is the inclusion of cognitively normal subjects who were closely followed-up until they met the stringent NINCDS-ADRDA criteria for probable AD, minimizing possible contamination with other dementing disorders. The use of a state-of-the art computational anatomy technique enabled us to uncover sub-regionally specific hippocampal involvement in subjects with pre-clinical AD and to dissociate different patterns of involvement in cognitively normal subjects who are transitioning to MCI and MCI subjects who are transitioning to AD. The limitations of our study lie in the relatively small sample size restricting our ability to generalize our findings and the strict inclusion of only the amnestic MCI subtype that prevents extension of our conclusions to subjects with nonamnestic MCI. The resolution at1.5 Tdoes not permit accurate subfield delineation. Thus our conclusions in respect to selective subfield involvement are based on assumptions about the location of each subfield boundary made in consultation with two well-established hippocampal histologic atlases (Duvernoy, 1988
; West and Gundersen, 1990
). Although our findings seem consistent with the previously published pathologic evidence of the trajectory of neurofibrillary tangle spread through the hippocampus (Bobinski et al., 1998
), the lack of histology-based subfield definition calls for caution when drawing conclusions from these data. Without pathological validation, both the diagnosis of AD and the inference that there is underlying neurofibrillary tangle pathology in the specific hippocampal subfields remain to be ascertained.
Our algorithm, as well as several approaches developed by other groups (e.g., Csernansky et al., 2005
), measures the extent and severity of hippocampal shape deformations as a proxy for hippocampal atrophy. To estimate the severity of the inward deformations or hippocampal thinning that occurs with hippocampal atrophy we compute the radial distance from the central core of each hippocampal structure to the respective points on the surface. The areas of thinning thus have smaller radial distance to the core. Even so, it cannot be ruled out that hippocampal deformations may have mimicked subfield atrophy, and there are some limitations in using surface mesh models to detect heterogeneous hippocampal atrophy. For example, if atrophy occurred in the dorsal side while volume increase occurred in the ventral side, this could result in a non-significant finding due to the shifting of the central skeleton. This is unlikely, as hippocampal growth or hypertrophy is not expected in aging or AD. Even so, the use of the central axis as a reference to gauge atrophy is a strength in many ways as the radial atrophy measures will be invariant to any overall shifting of the structure in space. In addition, the manual delineations allow highly precise delineation of boundaries for assessing atrophy, whereas a more automated registration method, like voxel-based morphometry, can typically match the hippocampal boundaries across subjects only very approximately.
Another strong advantage of the surface-based mapping technique relative to other voxel-based mapping approaches is its imperviousness to shifts in stereotaxic space. In other voxel-based mapping approaches, if a structure shifts in stereotaxic space, the shift can be recovered using automated nonlinear registration as a deformation, but it is highly unlikely that an automated alignment will register the complex shape boundaries of the hippocampus accurately across subjects, and because the deformation is constrained to be spatially smooth, it is almost inevitable that some changes in the structure would be incorrectly inferred if the structures only shifted. In the medial axis mapping approach we used here, however, if the structure shifts, the medial axis is translated by a corresponding amount, so there is no net alteration in the amount of atrophy. This is an advantage of using the medial axis curve as a reference, as the measures of 3D atrophy relative to this curve are shift invariant, and as intrinsic measures they should also not depend on the specifics of how the images are registered.