A, Results illustrated on the coronal section of the minimal deformation template show baseline hippocampal volume positively regressed at the follow-up positron emission tomographic scan (P ≤ .001, uncorrected; n = 36). B, Glass brain is adjacent for viewing of all significant regions.
A, Results illustrated on the sagittal section of the minimal deformation template depicting baseline white matter hyperintensities negatively regressed with metabolic change [(PET1 − PET2)/time] (P ≤ .001, uncorrected; n = 36). B, Glass brain is adjacent for viewing of all significant regions.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Kuczynski B, Reed B, Mungas D, Weiner M, Chui HC, Jagust W. Cognitive and Anatomic Contributions of Metabolic Decline in Alzheimer Disease and Cerebrovascular Disease. Arch Neurol. 2008;65(5):650–655. doi:10.1001/archneur.65.5.650
Alzheimer disease and cerebrovascular disease affect elderly persons through alterations in brain structure and metabolism that produce cognitive decline. Understanding how each disease contributes to dementia is essential from both a pathophysiologic and diagnostic perspective.
To elucidate how baseline cognitive function (episodic memory and executive function) and brain anatomy (white matter hyperintensities and hippocampal volume) are associated with baseline (positron emission tomography-1 [PET1]) and longitudinal (PET2) glucose metabolism in 38 subjects older than 55 years ranging from normal cognition, cognitive impairment without dementia, and dementia.
Cross-sectional regression analyses across subjects.
Multicenter, university-based study of subcortical vascular dementia.
Main Outcome Measures
Regional cerebral glucose metabolism was the primary outcome, with the major hypotheses that memory and hippocampal volume are related to temporoparietal hypometabolism while executive function and white matter hyperintensities correlate with frontal lobe hypometabolism.
Low baseline hippocampal volume predicted longitudinal development (PET2) of medial temporal hypometabolism. Lower memory was associated with parietal and cingulate hypometabolism at PET1, which increased at the 2-year-follow-up (PET2). Executive function was associated with frontal and temporoparietal hypometabolism at PET1 but only with frontal hypometabolism at follow-up. White matter hyperintensities predicted hypometabolism over time in the frontoparietal regions, predicting a rate of metabolic change (PET1 − PET2/time).
Low baseline episodic memory and hippocampal volume predict the metabolic alterations associated with Alzheimer disease, whereas elevated baseline white matter hyperintensities predict a different pattern of metabolic decline that is plausibly associated with cerebrovascular disease.
Dementia in older adults is frequently caused by the combined pathologic conditions of Alzheimer disease (AD) and cerebrovascular disease (CVD). Since AD and CVD frequently occur together and overlap in their presentations, their differentiation may be difficult or impossible in a given case of dementia. Nevertheless, there is ample evidence that they contribute separately to the development of cognitive decline and dementia.1,2 We, therefore, approached this study with the goal of defining the different pathways through which AD and CVD exert their effects on brain structure and function. While this approach is unlikely to lead to the development of diagnostic tests for the 2 conditions, it may elucidate how dementia occurs through their differential pathological vulnerabilities. Both cross-sectional and longitudinal relationships were of interest. We proposed that hippocampal volume (HV) and memory loss would primarily contribute to temporoparietalhypometabolism, while white matter hyperintensities (WMH) and executive dysfunction would contribute to frontal hypometabolism. Our goal was to define the patterns of association between these anatomic and cognitive variables with cross-sectional and longitudinal changes in glucose metabolism to explain how different brain systems are susceptible to the different pathological processes of AD and CVD.
Thirty-eight subjects recruited from dementia centers at either the University of California, Davis or the University of California, San Francisco between 1997 and 2004 were evaluated under the multi-institutional study of “The Aging Brain: Vasculature, Ischemia, and Behavior,” having completed at least 2 positron emission tomography (PET) imaging sessions. Informed consent to participate in the study was obtained in accordance with the policies of each institutional review board. The recruitment criteria included being older than 55 years and between the normal to moderate dementia range for cognition assessed by the Clinical Dementia Rating Scale score3 of 0 to 1, with scores of 0 indicating normal cognitive function, 0.5, cognitively impaired not demented (CIND), and 1, demented. Exclusionary criteria included diagnosis of cortical stroke or other neurological illnesses (other than AD or subcortical CVD), Mini-Mental State Examination score less than 15, and use of psychoactive drugs (other than stable doses of acetylcholinesterase inhibitors or selective serotonin reuptake inhibitors). All individuals were evaluated by clinicians at the university dementia clinics using standard criteria for vascular dementia4 and AD.5 Diagnostically, this cohort consisted of 14 with no cognitive im pairment (normal) (37%), 13 with dementia (34%), and 11 cognitively impaired but without dementia (29%). All groups contained but were not split by individuals with lacunes. Individuals in the group with dementia included 9 with probable AD (69%) (meeting the National Institute of Neurological and Communicative Disorders and Stroke criteria),5 1 with probable ischemic vascular dementia (8%) (meeting the State of California Alzheimer Disease Diagnostic and Treatment Centers criteria for probable ischemic vascular dementia),4 2 with mixed disease (a combination of AD and ischemic vascular dementia) (15%), and 1 with undetermined disease but not within our exclusionary criteria (8%). Five individuals with CIND (37%) were considered to have CVD (CVD but not meeting the Alzheimer Disease Diagnostic and Treatment Centers' criteria for ischemic vascular dementia).
We used regression analysis including all subjects without regard to diagnostic grouping based on the premise that the pathologies of AD and CVD are likely to be distributed across these elderly subjects. Baseline data included history and physical examination, neuropsychological testing (episodic memory [MEM] and executive function [EXEC]), PET, and magnetic resonance (MR) imaging. Baseline PET (PET1) and MR imaging were performed, on average, 0.88 month apart (median interval, 0.35 month). Follow-up PET scans (PET2) were performed at 2 years (mean [SD] interval, 1.9 [0.68] years). Therefore, the overall design included structural MR imaging and neuropsychological testing as the predictor variables and metabolic alterations (PET) at baseline or follow-up as the outcome or dependent variable. Thirty-eight individuals had full cross-sectional and longitudinal metabolic data sets and are the subject of this article. However, because of incomplete cognitive testing and imaging data, the number of individuals in each analysis varied between 32 and 37.
All individuals underwent a standardized battery of neuropsychological tests. The Mini-Mental State Examination6 was used as a clinical measure of global function. For analysis, scores from the battery were combined to form 2 composite, summary scales measuring MEM and EXEC. The development and characteristics of these scales have been previously described.7 Briefly, item scores from a comprehensive neuropsychological test battery in a sample of 400 elderly persons who varied in cognitive status from cognitively normal to mildly demented served as the basis for scale development. Item response theory analytic methods8 were used to create psychometrically matched scales. Each scale was transformed to have a mean (SD) of 100 (15). Donor items for MEM came from a list-learning task and emphasize delayed recall. Donor items for EXEC consisted of working memory and verbal fluency items and the Initiation-Perseveration subscale from the Mattis Dementia Rating Scale.9
Fluorodeoxyglucose-PET data were acquired using a 47-section scanner (ECAT EXACT, model 921; Siemens/CTI, Inc, Knoxville, Tennessee) in 2-dimensional acquisition mode to image the fluorodeoxyglucose F18 radiotracer. The injected dose for each individual was approximately 370 megabecquerels (MBq) of fluorodeoxyglucose. After injection, the individual was seated in a room for 40 minutes. Subjects were then positioned in the scanner to enable a field of view encompassing the entire brain. Emission data were collected in 2 dimensions for 40 minutes, followed by a 20-minute transmission period using a rotating germanium 68 source consisting of 3 rods of approximately 74 MBq per rod.
All MR imaging data for this study were acquired using a 1.5-T MR imaging system (Vision; Siemens Vision System, Munich, Germany). Sequences included a double spin-echo with repetition time (TR)/echo time (TE)1/TE2 (2500/20/80 ms), 1 signal acquired, 3-mm section thickness with no section gap, and in-plane resolution of 0.94 × 0.94 mm2. The second sequence provided the T1-weighted images via a T1 coronal magnetization prepared rapid gradient echo spin-echo axial design (TR/TE = 10/4 ms; 1 signal acquired), 1 × 1 mm2 in-plane resolution with contiguous 1.4-mm-thick sections.
The MR imaging variables of interest were volumetric measures of HV and WMH normalized to intracranial volume. The intracranial volume was calculated by summing all pixels within the intracranial area. The HV was derived from the T1-weighted image using a semiautomated, high-dimensional brain-warping algorithm (Medtronic Surgical Navigation Technologies, Louisville, Colorado), while the WMH were delineated via a segmentation program. These procedures have been described in depth elsewhere.10,11
All PET scans were partial volume corrected before analyses. The partial volume correction involved creating a brain mask from high-resolution T1-weighted MR imaging consisting of gray and white matter. Convolving this brain mask with the point spread function specific to the PET scanner provides a means for calculating the percentage of brain tissue at each voxel, thus eliminating the effects of cerebrospinal fluid. The PET count for each voxel was thus adjusted based on the percent of brain matter.12 Individual PET images were coregistered and aligned in standardized space, normalized to the mean global activity, and smoothed to 16 mm full-width, half-maximum. Because the anatomy of the aged brain presents normalization problems due to atrophy that distorts topography and makes superpositioning of identical regions difficult, we used a minimal deformation template derived from a group of aged brains13 as the target image. The minimal deformation template was created using the T1-weighted images of 25 normal, older individuals with a mean age of 71 years and is described in detail elsewhere.14
Voxelwise correlational analyses were performed on the PET data using statistical parametric mapping software (available at http://www.fil.ion.ucl.ac.uk/spm/). Regional associations were deemed significant if the t statistic reached P < .001 (uncorrected for multiple comparisons) with a cluster size greater than 100. The resulting significant areas were overlaid on the minimal deformation template for viewing. To find longitudinal metabolic differences, the 2-year follow-up PET (PET2) was subtracted from the baseline PET (PET1) and normalized to the interval between scans to yield a rate-of-change variable [(PET1 − PET2)/Time]. The metabolic pattern (outcome variable) of each PET image [PET1, PET2, and (PET1 − PET2)/Time] was correlated with covariates of interest at baseline—the neuropsychological test scores, HV, and WMH (predictor variables).
All data are presented in radiologic format. Table 1 summarizes the baseline demographic data for each group. As a whole, the entire cohort was well educated and had only moderate cognitive impairment. By classification, those in the group with dementia had significantly lower baseline Mini-Mental State Examination scores compared with those in the CIND group and the cognitively normal group. There were no significant differences between groups for age or education. Tables 2, 3, 4, and 5 give the Montreal Neurologic Institute coordinates for each anatomic area where metabolism significantly correlated with each predictor variable (P < .001; cluster size >100).
Figure 1 shows the positive association between baseline HV and the metabolic patterns at follow-up. The HV (right and left hemispheric volumes combined) is associated with medial temporal lobe hypometabolism at follow-up (in particular, the hippocampus and parahippocampal gyrus). As given in Table 2, the right-sided HV correlated with a few temporal regions at baseline, but the middle temporal regions (mostly bilateral), right parietal and posterior cingulate metabolism correlate at follow-up. Therefore, our data indicate that small HV at baseline predicts lower metabolism in temporal lobe and posterior cingulate metabolism over time. As further given in Table 2, baseline HV (left and right hemispheric volumes) predicted metabolic change [(PET1 − PET2)/time] in the medial temporal and precuneus regions.
Baseline MEM positively correlated with metabolism in the left angular, inferior parietal, and bilateral cingulate (posterior and middle) regions at baseline. At follow-up, MEM positively correlated with bilateral temporoparietal hypometabolism. Thus, lower baseline MEM predicts lower temporoparietal metabolism. There were no significant relationships between metabolic change over time [(PET1 − PET2)/time] and baseline MEM. (See Table 3 for all significant anatomic areas.)
As given in Table 4, baseline EXEC correlated with left temporoparietal metabolism and the left middle frontal lobe at baseline, which decreased in significance at follow-up. At follow-up, only the middle frontal region was significantly associated with EXEC (P = .001). No significant relationships were noted between metabolic change over time [(PET1 − PET2)/time] and EXEC.
At baseline, there was a significant negative correlation between the extent of WMH and the level of glucose metabolism in the right frontal and parietal lobe regions (precentral and postcentral gyri). A similar but more extensive pattern was seen at follow-up (precentral and postcentral gyri and the paracentral lobule; P = .001), which also occurred at follow-up within the left hemisphere (P = .003). Therefore, higher baseline WMH predict lower frontal and parietal lobe metabolism. Finally, our data demonstrate the longitudinal predictive value of baseline WMH. The relationship between baseline WMH and the change in metabolism over time [(PET1 − PET2)/time] within the left frontoparietal regions is given in Table 5 and shown in Figure 2. The longitudinal metabolic change occurred in the left hemisphere. Therefore, higher baseline WMH predict hypometabolism in the frontal lobe and parietal regions longitudinally, indicating a faster metabolic rate of change for individuals with WMH.
We also investigated the relationships between the anatomic and cognitive variables. Only HV and MEM were significantly correlated (P < .001, R2 = 0.31). Note there was no significant relationship between WMH and EXEC (R2 = 0.02). These relationships are similar to our voxelwise data in that WMH are unrelated to the cognitive variables while HV is highly correlated with MEM and moderately correlated with EXEC (P = .09, R2 = 0.1).
This study focused on the relationships and possible contributions of baseline MEM and EXEC, and baseline hippocampal atrophy and WMH with glucose metabolism. Our working hypothesis, that HV and MEM would be similar, as EXEC and WMH would be similar, in their contributions to glucose metabolism, were largely confirmed. First, small HV and decreased MEM at baseline predict diminishing metabolism in temporoparietal brain regions. Most interestingly, these baseline measures correlate with the progression of temporoparietal hypometabolism over time. Second, baseline WMH were associated with frontoparietal hypometabolism at baseline and follow-up, and in the rate of metabolic decline. The extent of frontal lobe hypometabolism associated with baseline WMH increased in size and intensity between baseline and follow-up PET scans. The relationships between WMH and frontal hypometabolism are consistent with those seen in cross-sectional studies15,16 and with a possible underlying cause related to CVD. These results suggest that HV and memory contribute to decreased temporoparietal metabolism while WMH contribute to frontal lobe hypometabolism. While baseline cognitive performance did not predict subsequent declines in glucose metabolism, baseline measures of brain anatomy did.
A finding contrary to prediction was the association between EXEC and baseline temporoparietal hypometabolism, which became less significant at follow-up. Alzheimer disease may be driving this relationship, further supporting the point that individual cognitive domains are not necessarily specific for any disease pathology. It is well known that advancing AD presents extensive EXEC dysfunction as well as frontoparietal brain alterations; however, this was not significantly present in our data.
Our data suggest that metabolic alterations within the temporal and parietal lobes are primarily associated with MEM impairments and hippocampal atrophy, both of which have been associated with AD, whereas, metabolic alterations in the frontal lobes are primarily associated with WMH and, therefore, are putatively associated with CVD and, in particular, not associated with AD-like metabolic change. Importantly, these metabolic patterns are detectable in the context of mixed disease with overlapping symptom patterns.
Although the focus of this study was on the predictability of baseline measurements, the results did not differ much at follow-up. Briefly, most of the hypometabolic areas were larger and of greater significance when correlated with the 2-year follow-up neuropsychological data. More importantly, there were no associations between metabolism and neuropsychological performance over time.
The major limitation of this study is that the true extent and distribution of neuropathology in these subjects are unknown. However, the variables that we chose for examination have been well studied, and their relationship to pathology is relatively well known. Also, the multiple comparisons of this voxel-based study limit interpretation of the results. Although we did not correct for multiple comparisons, we did restrict our findings to a cluster size of 100 or larger. This inhibits finding smaller significant areas, but it minimizes the chance of type I error. Because our results revealed large areas previously associated with AD and CVD, it seems unlikely that they are a consequence of false discovery. Also, owing to the colinearity of many of our variables and the small sample size, we were unable to perform a multivariate analysis.
Taken together, our findings suggest that while tests of individual cognitive domains may not prove useful in defining an etiology of dementia, patterns of cognitive decline, anatomic change, and metabolism together support a conceptualization that different pathological processes affect different brain systems. By evaluating these different modalities simultaneously, an understanding of how cognitive decline might reflect the interactions of AD and CVD can begin to emerge.
Correspondence: Beth Kuczynski, PhD, Helen Wills Neuroscience Institute, University of California, Berkeley, 118 Barker Hall, MC 3190, Berkeley, CA 94720 (email@example.com).
Accepted for Publication: July 30, 2007.
Author Contributions:Study concept and design: Kuczynski, Reed, Mungas, and Jagust. Acquisition of data: Kuczynski, Reed, Mungas, Weiner, Chui, and Jagust. Analysis and interpretation of data: Kuczynski, Reed, Mungas, and Jagust. Drafting of the manuscript: Kuczynski. Critical revision of the manuscript for important intellectual content: Reed, Mungas, Weiner, Chui, and Jagust. Statistical analysis: Kuczynski, Reed, and Mungas. Obtained funding: Kuczynski, Reed, Mungas, and Jagust. Administrative, technical, and material support: Reed, Weiner, Chui, and Jagust. Study supervision: Jagust.
Financial Disclosure: None reported.
Funding/Support: This study was supported by grant AG12435 from the National Institute on Aging.
Additional Contributions: Charles De Carli, MD, and Christine Nordahl, PhD, created the minimal deformation template used in our analyses.