Early DAT is distinguished from aging by high-dimensional mapping of the hippocampus

Methods.

Subject selection and assessment. All elderly subjects were living in the community and enrolled in longitudinal studies of DAT and healthy aging at the AD Research Center at Washington University School of Medicine. Members of AD families with known AD genetic mutations were excluded from this study. Also, subjects were excluded if they presented with symptoms of other neuropsychiatric disorders that could have confounded the diagnosis of DAT.

Clinical Dementia Rating (CDR) scale assessments were performed annually in the elderly subjects by experienced clinicians without reference to neuropsychological tests or in vivo neuroimaging data. These assessments were performed using interviews with collateral sources for all subjects; the final scoring reflected information derived from both the subject assessment and the collateral source interview. CDR scores were compiled from six “box scores,”18 each of which rates a category of behavioral functions. CDR scores of 0 indicate no dementia, and CDR scores of 0.5, 1, 2, and 3 indicate the presence of very mild, mild, moderate, and severe dementia. Interrater reliability for the CDR is high.19,20⇓

Clinical assessments indicating the presence of AD (i.e., CDR > 0) have been confirmed at autopsy with 93% accuracy; elderly control subjects (i.e., CDR 0) have been confirmed with normal brains at autopsy with 80% accuracy.21 CDR 0.5 subjects may have only subtle impairments of cognition and behavior, but progress to more severe stages of illness (i.e., CDR > 0.5) over variable intervals of time,22 and the great majority of CDR 0.5 subjects show neuropathologic signs of AD on postmortem examination, in contrast to subjects rated as CDR 0.23

The elderly subjects recruited for the study met the following requirements: CDR 0.5 with a diagnosis of DAT (n = 18) or CDR 0 (n = 18), no contraindications to MRI scanning, and willingness to cooperate (informed consent was obtained after the nature and risks of the study were explained). The sample sizes were selected based on the magnitude of hippocampal volume reductions previously observed in similar groups of DAT and control subjects.3,5,8,9⇓⇓⇓ Sixteen of the 18 CDR 0.5 subjects met National Institute of Neurological and Communicative Disorders and Stroke/AD and Related Disorders Association criteria for probable AD. The subjects also had demographic characteristics similar to larger cohorts of CDR 0 and CDR 0.5 subjects previously recruited at our center for clinical studies not requiring an MRI scan19,20⇓ ( table 1). Annual CDR assessments as close as possible to the time of MRI scanning were used in categorizing these subjects; the mean (SD) duration of time between the time of clinical assessment and MRI scanning was 53 (44) days.

MRI scans from a control group of healthy younger subjects (n = 15) were obtained from a prior study of schizophrenia conducted by our research group.24 These younger control subjects were selected from the same community as the elderly subjects specifically recruited for this study and met similar exclusion criteria for confounding neurologic and medical illnesses.

Table 1 summarizes the general characteristics of the three groups of subjects. All subjects were right-handed. The two older subject groups were similar in age and gender. The two groups of elderly subjects had similar total cerebral and total intracranial volumes, but these volumes were smaller in the elderly subjects than in the younger control subjects. The presence of similar between-group differences in both total cerebral and total intracranial volumes suggests an overall difference in head size, perhaps attributable to cohort differences related to nutrition and height (data not available). Nonetheless, total cerebral volume and total intracranial volume were included as covariates in statistical analyses of hippocampal volume. Because the gender distribution of the younger control subjects also differed from the two groups of elderly subjects, all analyses were checked for the influence of this variable; however, no significant gender effects were found.

APOE gene allele status was determined in 30 of the 36 elder subjects. Two CDR 0.5 subjects were homozygous for the APOE ε4 allele, and nine CDR 0.5 subjects were heterozygous for the APOE ε4 allele. Nine CDR 0 subjects were also heterozygous for the APOE ε4 allele.

Results.

Analysis of hippocampal volume. A comparison of hippocampal volumes in the CDR 0 and CDR 0.5 subjects and the younger control subjects showed that the three groups differed with respect to hippocampal volumes (group effect: F = 20.0; df = 2,48; p = 0.0001; figure 1). In post-hoc between-group comparisons, the CDR 0.5 subjects showed volume decreases in both the left and right hippocampus compared with the CDR 0 subjects (F = 19.4; p = 0.0001) and the younger control subjects (F = 37.1; p = 0.0001). However, the comparison of hippocampal volumes in the CDR 0 subjects and the younger control subjects did not reach significance (F = 5.38; p = 0.065). In addition, hemispheric asymmetries were observed in hippocampal volumes across all groups (hemisphere effect [left < right]: F = 180; df = 1,48; p < 0.0001). The asymmetry of hippocampal volumes across all subject groups was also observed when men and women were separately considered in each group.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Hippocampal volumes are compared in elderly CDR 0.5 subjects, elderly CDR 0 subjects, and younger control subjects. Left and right hippocampal volumes were significantly smaller in elderly CDR 0.5 subjects than in elderly CDR 0 subjects or younger control subjects. The left hippocampus was significantly smaller than the right hippocampus across all three groups of subjects.

The mean (SD) hippocampal volumes for subjects classified as CDR 0.5 was 1,729 (269) mm³ in the left hemisphere and 2,158 (362) mm³ in the right hemisphere, whereas the mean (SD) hippocampal volumes for the CDR 0 subjects was 2,189 (375) mm³ in the left hemisphere and 2,774 (498) mm³ in the right hemisphere, and the mean (SD) hippocampal volumes for the younger control subjects was 2,603 (364) mm³ in the left hemisphere and 2,845 (393) mm³ in the right hemisphere. A comparison of hippocampal volumes among the CDR 0.5 subjects, CDR 0 subjects, and younger control subjects, after covarying with the values for total cerebral volume (group effect: F = 7.9; df = 2,47; p = 0.001) and total intracranial volume (group effect: F = 7.8; df = 2,47; p = 0.001), produced equivalent results.

Finally, a comparison of the left and right hippocampal volumes in CDR 0 subjects who were heterozygous for the APOE ε4 allele (n = 9) versus the CDR 0 subjects that had no APOE ε4 allele (n = 7) showed no group effect (F = 2.77; df = 1,14; p = 0.12) or group × hemisphere interaction (F = 2.81; df = 1,14; p = 0.12). In fact, there was no suggestion of even a trend in the direction of the expected hypothesis, because CDR 0 subjects with APOE ε4 alleles had hippocampal volumes that were slightly larger than CDR 0 subjects without APOE ε4 alleles.

These results suggest that if one carefully selects only truly nondemented elderly subjects for study, excluding cases of very mild DAT that might otherwise contaminate the putative “control” sample, brain structure volume losses associated with healthy aging may be less than previously reported by some27,28⇓ but not all29 investigators. Differences in hippocampal volume differentiated CDR 0 subjects from CDR 0.5 subjects. However, although elderly CDR 0 subjects had somewhat smaller hippocampal volumes than younger control subjects, this difference did not quite reach significance; the correlation between age and hippocampal volumes within the two control groups was modest but did reach significance (r = −0.44, p = 0.01). This finding is consistent with the results of postmortem stereologic studies at our center, which show no substantial difference in neuronal number within the entorhinal cortex of CDR 0 subjects.13

Analysis of hippocampal shape.

An analysis of eigenvectors derived by reducing the dimensionality of a matrix of transformation vector fields derived from all three subject groups (i.e., CDR 0, CDR 0.5, and younger control subjects) revealed that the first and second eigenvectors distinguished among the three groups of subjects (F = 87.0; df = 4,94; p < 0.0001). As described above, these particular eigenvectors were selected using a multivariate ANOVA model with appropriate contrasts. In contrast to the results obtained using hippocampal volumes to compare the three groups, post-hoc testing of between-group differences in this analysis showed that all groups could be readily distinguished from each other (i.e., CDR 0.5 versus CDR 0: F = 5.60, p = 0.006; CDR 0.5 versus younger control subjects: F = 313, p < 0.0001; CDR 0 versus younger control subjects: F = 303, p < 0.0001). When hippocampal volume values for the left and right hippocampus were included with the selected eigenvectors in the analysis, nearly identical results were obtained. The three groups were again distinguished (F = 63.1; df = 6,92; p = 0.0001), and post-hoc between-group comparisons showed that all groups were distinguished from each other (i.e., CDR 0.5 versus CDR 0: F = 6.33, p = 0.001; CDR 0.5 versus younger control subjects: F = 221, p < 0.0001; CDR 0 versus younger control subjects: F = 204, p < 0.0001). Finally, we examined the possibility that the inclusion of total brain volume or total intracranial volume as covariates might alter group discrimination. However, highly similar results were again obtained when total brain volume values (three-group discrimination: F = 54.5, df = 8,90, p < 0.0001; CDR 0.5 versus CDR 0: F = 4.76, p = 0.003; CDR 0.5 versus younger control subjects: F = 206, p < 0.0001; CDR 0 versus younger control subjects: F = 198, p < 0.0001) or total intracranial volume values (three-group discrimination: F = 53.9, df = 8,90, p < 0.0001; CDR 0.5 versus CDR 0: F = 4.75, p = 0.003; CDR 0.5 versus younger control subjects: F = 202, p < 0.0001; CDR 0 versus younger control subjects: F = 194, p < 0.0001) were included along with the selected eigenvectors in the analysis.

Between-group comparisons of the CDR 0.5 and CDR 0 subjects and of the CDR 0 and younger control subjects were performed to further characterize and visualize the patterns of hippocampal deformation that distinguished these groups. For CDR 0.5 and CDR 0 subjects, a stepwise logistic regression was performed on the selected eigenvectors based on χ² scores. Eigenvectors one (χ² = 8.0) and five (χ² = 9.3) were selected as most able to distinguish the CDR 0.5 and CDR 0 subject groups (F = 11.4, df = 2,33, p = 0.0002). In figure 2A, log-likelihood ratios for each subject in these two groups are plotted. However, when left and right hippocampal volumes were included with the shape eigenvectors in the analysis, only eigenvector five was selected in addition to the hippocampal volumes as informative (F = 12.2; df = 2,33; p = 0.0001) (see figure 2B), suggesting that eigenvector one was correlated with hippocampal volume values. In a jackknife analysis utilizing eigenvectors one and five alone, 12 of the 18 subjects in the CDR 0.5 group (67%) and 14 of the 18 subjects in the CDR 0 group (78%) could be categorized successfully. When the jackknife analysis included left and right hippocampal volumes and eigenvector five, 15 of the 18 subjects in the CDR 0.5 group (83%) and 14 of the 18 subjects in the CDR 0 group (78%) could be categorized successfully. Finally, the addition of total brain volumes or total intracranial volumes along with eigenvector five in these analyses yielded highly similar results (i.e., in both cases, 16 of the 18 CDR 0.5 subjects and 14 of the 18 CDR 0 subjects were successfully categorized).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Log-likelihood ratio values derived from the linear combination of eigenvectors representing hippocampal shape are compared in CDR 0.5 subjects and CDR 0 subjects. Eigenvectors one and five were selected as informative in distinguishing these two subject groups (A) (p = 0.0002). However, when left and right hippocampal volume values were included in the stepwise logistic regression procedure (B), only eigenvector five was selected in addition to the volume values as informative in distinguishing the two groups (p = 0.0001).

A comparison of CDR 0 and younger control subjects showed that eigenvectors one and two were selected as most informative in distinguishing these two groups (F = 348; df = 2.30; p < 0.0001) ( figure 3A). However, these same two eigenvectors were selected even when left and right hippocampal volumes (F = 225; df = 2,30; p < 0.0001) (see figure 3B), total brain volumes (F = 224; df = 4, 28; p < 0.0001) or total intracranial volumes (F = 215; df = 4,28; p < 0.0001) were included in the analyses. In a jackknife analysis using eigenvectors one and two, with or without the inclusion of left and right hippocampal volumes, total brain volumes, or total intracranial volumes as additional variables, all subjects in the CDR 0 group (100%) and all subjects in the younger group (100%) were successfully categorized.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Likelihood ratio values derived from the linear combination of eigenvectors representing hippocampal shape are compared in elderly CDR 0 subjects and younger control subjects. Eigenvectors one and two were selected as informative in distinguishing these two subject groups (A) (p < 0.0001). Even when left and right hippocampal volume values were included in the stepwise logistic regression procedure (B), these same two eigenvectors were again selected as informative (p < 0.0001).

Finally, we sought to determine whether log-likelihood ratio values could distinguish among CDR 0 subjects with (n = 9) and without (n = 7) an APOE ε4 allele. There was no separation in the log-likelihood ratios between these two groups of subjects.

The pattern of hippocampal surface deformations that occurred in the CDR 0.5 subjects as compared with the CDR 0 subjects are visualized in figure 4. As described above, deformation maps are displayed using both point-by-point rank order tests (see figure 4A) and reconstructions from the linear combination of eigenvectors one and five (see figure 4B). In both cases, a similar pattern of neuroanatomic deformation of the hippocampal surface is suggested. Regardless of the method of visualizing hippocampal surface deformations, the CDR 0.5 subjects show inward deformities in the head of the hippocampus (right > left) and along the lateral surface of the left and right hippocampal bodies.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4. Hippocampal surface deformations characteristic of early DAT are visualized. (A) Deformations of the hippocampal surface that differentiated CDR 0.5 subjects and CDR 0 subjects are displayed using statistical thresholds (inward and outward deformations at the p < 0.05 level) derived from rank order tests at every point on the hippocampal surface. The direction of surface displacement for the point-by-point p values is coded by different colors. (B) Between-group deformations of the hippocampal surface are displayed by reconstructing the hippocampal surfaces from the linear combination of eigenvectors one and five, which significantly distinguished among the three groups of elderly subjects. The magnitudes (mm) and direction of surface displacement are indicated using a color flame scale. The left and right hippocampi are shown from a perspective slightly above and to the left and right of a midline plane to highlight deformities along the lateral body of the left and right hippocampi.

Visualization of the hippocampal surface deformations that differentiated the CDR 0 and younger control subjects suggests a sharply different pattern of change in comparison with the pattern observed in CDR 0.5 and CDR 0 subjects. This pattern suggests approximately equal areas of inward and outward deformation ( figure 5). Furthermore, the distribution of these shape changes (i.e., inward and downward deformation of the head and tail, but outward and upward deformation of the body) suggests a general flattening of the structure, not accompanied by any substantial loss of volume. The difference in the pattern of hippocampal deformation that differentiated CDR 0 subjects from healthy younger control subjects and CDR 0.5 subjects from CDR 0 subjects suggests that healthy aging is a phenomenon distinct from early AD.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5. Hippocampal surface deformations characteristic of normal aging are visualized. (A) Deformations of the hippocampal surface that distinguished between CDR 0 subjects and younger control subjects are displayed using a statistical threshold (inward and outward deformations at the p < 0.05 level) derived from point-by-point rank order tests. The direction of surface displacement for the point-by-point p values is coded by different colors. (B) Aging-related deformations of the hippocampal surface are displayed by reconstructing the hippocampal surfaces from the linear combination of eigenvectors one and two, which significantly distinguished the two groups. The magnitude (mm) and direction of surface displacement are indicated using a color flame scale. The left and right hippocampi are shown from a perspective slightly above and to the left and right of a midline plane to highlight deformities along the lateral body of the left and right hippocampi.

Discussion.

The results of this study suggest that hippocampal shape eigenvectors yielded by HDBM used in combination with hippocampal volumes may be helpful in distinguishing early DAT from healthy aging. Using volume and shape metrics together, specific patterns of hippocampal deformity were identified that distinguished these two neurobiological phenomena from each other. The lack of a substantial loss of hippocampal volume in the healthy elderly control subjects as compared with the younger control subjects was especially notable. Healthy elderly subjects could only be distinguished from their younger counterparts by an analysis of hippocampal shape.

Table 2 summarizes effect sizes for comparisons between subject groups using hippocampal shape eigenvectors and volumes. Although hippocampal volumes did differentiate the CDR 0.5 and CDR 0 subjects, including the information represented by the shape eigenvectors allowed for further distinction of the two groups. The remaining overlap between CDR 0.5 and CDR 0 subjects should be interpreted with caution, because clinical assessments cannot always identify individuals with the earliest (i.e., preclinical) forms of AD. For example, recent postmortem studies from our center have described plaques characteristic of AD in a minority of individuals categorized as healthy elderly subjects.21,23⇓ Unfortunately, APOE allele status among the CDR 0 subjects was not useful in distinguishing truly normal elderly subjects and elderly subjects with preclinical disease. However, this result is consistent with the results of other recent MRI studies using this type of information.30

This study had three features that improved the chances of distinguishing very mild DAT from healthy aging. First, the healthy elderly control subjects selected for this study were screened using detailed evaluations to exclude subjects with detectable symptoms of dementia. Second, these clinical evaluations also allowed us to identify and assess the severity of illness in those subjects who did manifest detectable symptoms of DAT. Third, our newly developed tools for computational anatomy (i.e., HDBM) gave us access to higher dimensional metrics related to neuroanatomic shapes for use in statistical comparisons among the subject groups. However, because of the slowly evolving nature of AD, uncertainty must remain attached to the clinical diagnosis of normal versus DAT and the precision and interpretation of our neuroanatomic assessments. Further reduction of this uncertainty would require postmortem confirmation of these antemortem assessments.

The pattern of hippocampal deformities observed in the subjects with very mild DAT (i.e., CDR 0.5) suggests a loss of tissue in the head and on the lateral surface of the body of the hippocampus. The lateral surface of the body of the hippocampus is nearest to the CA1 hippocampal subfield, and neuronal damage in this neuroanatomic subregion has been observed in postmortem studies of early AD.10,12,31⇓⇓ Unfortunately, localization of hippocampal cell subfields in the head of the hippocampus is far more difficult, because the CA1 subfield wraps around to cover the head of the hippocampus on both the lateral and medial surfaces. We observed a similar pattern of hippocampal deformation in CDR 0.5 subjects in the left and right hippocampus, which suggests that the loss of hippocampal tissue in DAT is largely symmetric. This pattern of hippocampal degeneration is also consistent with the results of postmortem studies of AD, in which the densities of plaques and tangles have been largely found to occur in a symmetric pattern.12,32,33⇓⇓

Finally, our results suggest a substantial asymmetry of hippocampal volumes (left < right) across all groups of subjects. This result is consistent with the results of prior postmortem studies34-36⇓⇓ and our own prior neuroimaging studies.24,25⇓ However, the functional consequences of this apparent asymmetry of hippocampal volumes are not known, both for normal individuals and for individuals with early DAT.

We demonstrated and quantified changes in the shape of the hippocampus associated with healthy aging. Our results suggest that these changes, although involving a substantial portion of the hippocampal surface, are not associated with a substantial loss of hippocampal volume. In addition, the pattern of shape changes in the healthy elderly subjects as compared with the younger control subjects could be clearly distinguished from the hippocampal deformities found in subjects with the earliest detectable symptoms of DAT. Visualization of this shape change suggested a general flattening of the structure, with the head and tail of the hippocampus appearing to bend downward in relationship to the body. Whereas shape changes accompanied by evidence of substantial volume losses suggest a neurodegenerative process intrinsic to the brain structure of interest, shape changes not accompanied by evidence of volume losses or increases could occur because of a process intrinsic or extrinsic to the structure. In other words, changes in brain structures immediately in the vicinity of the hippocampus could impose a shape change upon it without substantially altering its volume. Such findings suggest the need to apply our tools of HDBM to structures surrounding the hippocampus, such as the parahippocampal gyrus, which have also been implicated in the early pathogenesis of AD.13 Nonetheless, different patterns of hippocampal shape and volume changes in early DAT and healthy aging are consistent with the results of postmortem and other neuroimaging studies, which suggest that there are critical differences in the way these two phenomena affect limbic brain structures.23,37,38⇓⇓

HDBM applied to high-resolution MRI scans now affords scientific investigators the opportunity to examine both general (i.e., volume) and local (i.e., shape) features of neuroanatomy with a high degree of spatial resolution and precision. The results of this study, along with our prior studies of schizophrenia using the same technology,24 suggest that HDBM is capable of identifying subtle neuroanatomic deformities in subjects with neuropsychiatric diseases. The visualization and quantitative analysis of neuroanatomic shapes through HDBM may help to coordinate in vivo neuroimaging techniques with cellular approaches for studying neuropsychiatric diseases, because highly localized areas of tissue deformity (i.e., the CA1 hippocampal subfield) can be selected for combined study.

After additional study, HDBM might also become a useful tool for enhancing the diagnosis of AD at a stage when clinical symptoms are intermittent or equivocal. HDBM can be applied to MRI scanning sequences routinely available at most academic centers, and conventional microcomputers are capable of running HDBM algorithms. However, the results of this preliminary study with a relatively small number of highly selected subjects do not yet show that assessment of the hippocampus by HDBM could identify preclinical cases of AD or differentiate AD from other dementing disorders in clinical settings. Any method of brain structure assessment, including HDBM, has limitations, such as measurement error and a dependency on the high quality of available MRI data sets, which could undermine its utility in everyday clinical settings. Prospective, longitudinal studies of larger and less selected cohorts of elderly subjects are needed to determine the feasibility of using HDBM in such clinical settings. Brain structures other than the hippocampus, including regions of cerebral cortex recently implicated in neuroimaging studies of elderly subjects at risk for developing DAT,29,39⇓ should also be assessed.