Correlations between lower regional cerebral metabolic rates for glucose levels and higher rates of subsequent decline in verbal long-term memory recall scores in subsequent converters to amnestic pre–mild cognitive impairment (MCI) (A) and in nondecliners (B), as well as greater correlations in the amnestic pre-MCI converters than in the nondecliners (C). For each comparison, statistical maps are superimposed on magnetic resonance imaging sections 48- to 24-mm superior to a horizontal plane through the anterior and posterior commissures in the atlas coordinates of Talairach and Tournoux.21 The left side of each section corresponds to the left hemisphere. P values are uncorrected for multiple comparisons, and maximal P values are given in Table 3.
Caselli RJ, Chen K, Lee W, Alexander GE, Reiman EM. Correlating Cerebral Hypometabolism With Future Memory Decline in Subsequent Converters to Amnestic Pre–Mild Cognitive Impairment. Arch Neurol. 2008;65(9):1231-1236. doi:10.1001/archneurol.2008.1
Before symptomatic memory loss, healthy apolipoprotein E ε4 (APOE ε4) (OMIM 104310) carriers demonstrate accelerated longitudinal decline on memory tests, suggesting the existence of a transitional state between normal aging and mild cognitive impairment (MCI), which we have called amnestic pre-MCI.
To support our neuropsychological construct of pre-MCI by characterizing and comparing the relationship between measurements of baseline regional hypometabolism and subsequent rates of memory decline in a group of individuals with neuropsychologically defined asymptomatic memory decline (pre-MCI group) and in nondecliners after controlling for APOE ε4 gene dose.
Academic medical center.
Of 139 healthy individuals in the Arizona APOE Cohort aged 50 to 69 years who underwent longitudinal neuropsychological testing and fludeoxyglucose F 18–positron emission tomography (FDG-PET) since 1994, 10 met our criteria for amnestic pre-MCI, and 15 showed no decline.
Main Outcome Measures
Correlations between lower regional cerebral metabolic rates for glucose (CMRgl) and rates of verbal memory test decline that occurred at a mean of 41 months after baseline FDG-PET using an automated brain mapping algorithm (SPM5).
The pre-MCI and nondecliner groups did not differ in mean (SD) age (56.8 [4.8] years), education (16.5 [2.3] years), sex (19 women [76%]), or APOE ε4 carrier status (12 ε4 carriers [48%)]. After controlling for APOE ε4 gene dose, the pre-MCI group had significant correlations between lower baseline CMRgl in the posterior cingulate, bilateral parietal, and left prefrontal regions (known to be preferentially affected by Alzheimer disease) and subsequent verbal memory decline. Nondecliners had significant correlations bilaterally in the posterior and midcingulate cortices. Correlations in the left parietal, left temporal, and bilateral frontal regions were significantly greater in the pre-MCI group than those in the nondecliner group.
Individuals with amnestic pre-MCI showed significantly greater correlations between cerebral hypometabolism and subsequent long-term memory decline than nondecliners in Alzheimer disease–affected brain regions.
Mild cognitive impairment (MCI) was originally conceived as a functionally nondisabling amnestic disorder.1 The concept has been expanded to include essentially any form of cognitive complaint,2 but the greatest correlation between MCI and Alzheimer disease (AD) remains with the amnestic form.3- 5 Operationally, MCI is a useful construct to distinguish individuals with a diagnostically ambiguous disorder from those with clinically probable AD. The effect of an intervention (or comorbidity) on the conversion rate from MCI to AD has become a paradigm for assessing possible neurobiologic modifiers of disease course. Early intervention should enhance the likelihood of therapeutic efficacy so that MCI has become the preferred platform for disease modification trials.6 However, a problem with MCI in this context is that in patients who subsequently “convert” to meet the clinical criteria for probable AD, MCI was in most cases an early stage of AD or other degenerative dementia.7 Therefore, there is interest in moving back the diagnostic line of scrimmage to a stage that is earlier than MCI and captures the inflection point at which normal and pathologic aging first diverge, when AD may not yet be irreversibly established.
Is it possible to identify an earlier stage of AD than MCI? Applying previously defined criteria for cognitive domain decline, abnormally declining memory scores on longitudinal neuropsychological tests were demonstrated that preceded symptoms of memory loss in a group of healthy volunteers.8 Consistent with original predictions by Corder et al,9 it was found that APOE genotype influenced the age at onset and pattern of this asymptomatic decline (termed pre-MCI), suggesting that it may be an early stage of AD that precedes MCI. We sought biologic support for the neuropsychological construct of pre-MCI by using fludeoxyglucose F 18–positron emission tomography (FDG-PET) to characterize and compare the correlations between lower cerebral metabolic rates for glucose (CMRgl) and subsequent rates of memory decline in a group of asymptomatic subjects who met our operational criteria for amnestic pre-MCI and a group of asymptomatic subjects without such decline. Using FDG-PET, it was previously reported that asymptomatic apolipoprotein E ε4 (APOE ε4) carriers have similar (but milder) metabolic reductions in the same brain regions as those found in patients with amnestic MCI and AD.8 We now hypothesize that lower CMRgl in brain regions affected by AD will be correlated with subsequent memory decline in subjects who develop amnestic pre-MCI, independent of APOE genotype, and that these correlations will be significantly greater than those in subjects who did not demonstrate subsequent memory domain decline.
Twenty-five participants (10 with amnestic pre-MCI and 15 nondecliners) in this study were a subset of the members of the Arizona APOE Cohort, a group of healthy individuals from Maricopa County (Arizona) selected on the basis of APOE genotype who undergo longitudinal neuropsychological assessment (a subset of whom also undergo FDG-PET) every 2 years.10 In a previous study8 of 214 cohort members aged 50 to 69 years at enrollment who completed at least 2 epochs of neuropsychological testing (139 of whom also underwent FDG-PET), operational criteria were generated for abnormal neuropsychological decline, and 5 categories of decline were defined as follows: (1) no decline on any test, (2) test decline in 1 domain (nonspecific), (3) test decline in more than 1 domain (nonspecific), (4) single cognitive domain–specific decline (domains included executive, memory, language, spatial, and behavioral), and (5) multiple cognitive domain decline.8 Domain decline required a decline of more than 2 SDs compared with the entire cohort on at least 2 separate tests sensitive to that domain. There were 4 tests administered for each domain. For the memory domain, these were the Auditory Verbal Learning Test (AVLT), Selective Reminding Test, Rey-Osterrieth Complex Figure Test, and Visual Retention Test.11 In all, there were 10 amnestic pre-MCI members and 15 nondecliners who met these criteria and underwent FDG-PET, and all were included in this study.
All Arizona APOE Cohort members were originally identified through local newspaper advertisements, reported a family history of AD, and understood that they would receive no information regarding their APOE genotype. APOE allelic status was performed using a polymerase chain reaction–based assay.12 Screening tests included a complete medical history, neurologic examination, Mini-Mental State Examination, Hamilton Depression Rating Scale, and Structured Psychiatric Interview for DSM-III-R.13 No subjects met criteria for MCI,1,2 AD,14 any other form of dementia, or major depressive disorder.13
Our primary goal was to correlate the impending change in memory with baseline FDG-PET. Although our definition used patterns of test decline based on 4 memory tests to categorize patterns of decline and to define our pre-MCI and nondecliner groups, we needed a continuous variable for correlation with CMRgl. Our strategy was to use the single most sensitive memory test score,10 rather than a derived composite score, to be consistent with known clinical validity.11 We used the memory decline epoch in the pre-MCI group that was, by definition, the epoch in which pathologic decline from the preceding epoch occurred. The AVLT long-term memory (LTM) scores from the first (predecline) and second (decline) epochs were used to calculate a memory change score that was then correlated with baseline FDG-PET. Because there was no similar memory decline epoch for subjects in the nondecliner group, the 2 consecutive epochs for each nondecliner were chosen so that, on average, the interval between the first epoch and the time of baseline FDG-PET was statistically no different from that of the pre-MCI group.
Imaging by FDG-PET was performed using a scanner (951/31 ECAT; Siemens, Knoxville, Tennessee), a transmission scan, intravenous injection of 10 MCi of fludeoxyglucose F 18, and a 60-minute dynamic sequence of emission scans as the subjects, who had fasted for at least 4 hours, lay quietly in a darkened room with their eyes closed and directed forward.15 Regional (voxel-by-voxel) analyses were performed by using the FDG-PET images (in counts) acquired during the past 30 minutes. An automated brain mapping algorithm (Statistical Parametric Mapping, SPM5 [http://www.fil.ion.ucl.ac.uk/spm/]) was used to characterize and compare correlations between lower baseline measurements of regional CMRgl and subsequent rates of decline in AVLT LTM in the pre-MCI and nondecliner groups. The FDG-PET images were spatially normalized using linear and nonlinear warping algorithms into a common stereotactic space, smoothed using a 12-mm full-width-at-half-maximum gaussian kernel, and normalized for the individual variation in whole-brain FDG-PET counts using proportionate scaling. Univariate regression analyses were then used to compute statistical maps of the correlations between lower CMRgl and subsequent rates of decline in AVLT LTM in the overall subject group and to compare these correlations in the pre-MCI and nondecliner groups (P < .001, uncorrected for multiple comparisons). The correlations were recomputed after covarying out the effects of baseline AVLT LTM scores to determine the extent to which baseline differences in performance were driving the results. Finally, the correlations were recomputed after covarying out the effects of APOE ε4 gene dose (the number of ε4 alleles in a subject's APOE genotype) to determine the extent to which the observed findings were solely attributable to the effects of this well-established genetic risk factor.
Correlations were based on known patterns of CMRgl reductions in patients with AD16,17 and on previous observations of CMRgl reductions in presymptomatic APOE ε4 carriers.18- 20 We hypothesized that lower baseline CMRgl measurements in the posterior cingulate, parietal, temporal, and prefrontal cortices at baseline would correlate with subsequent memory decline in the pre-MCI group.
Enrollment demographics are summarized in Table 1. At enrollment, no differences were noted between the pre-MCI and nondecliner groups in mean (SD) age (56.8 [4.8] years; P =.49, t test), education status (16.2 [2.3] years; P =.38, t test), sex (19 women [76%]; P =.65, Fisher exact test), APOE genotype (12 ε4 carriers [48%]; P =.69, Fisher exact test), or Mini-Mental State Examination score (29.7 [0.6]; P =.19, t test). Of 4 memory tests, the mean (SD) Complex Figure Test–recall score was lower (although normal) in the pre-MCI group (15.2 [5.4]) than that in the nondecliner group (21.1 [5.4]) (P =.01), but there were no significant differences between the groups on any other memory score. There were no significant differences on any test scores in the remaining domains (Table 2).
The mean (SD) time from baseline FDG-PET to the first AVLT test epoch did not differ between the pre-MCI group (41  months) and the nondecliner group (33 [18.7] months) (P =.34). Longitudinally, the mean (SD) AVLT LTM score declined in the pre-MCI group (2.4 [1.2]) but slightly improved in the nondecliner group (−0.53 [2.1]) (P < .001). The pre-MCI group had significant correlations between lower baseline CMRgl in the posterior cingulate, bilateral parietal, and left prefrontal regions (known to be preferentially affected by AD) and subsequent verbal memory decline. Nondecliners had significant correlations bilaterally in the posterior and midcingulate cortices. Correlations in the left parietal, left temporal, and bilateral frontal regions in the pre-MCI group were significantly greater than those in the nondecliner group (Figure and Table 3). Each of the correlations and interactions was found after controlling for the effects of baseline AVLT LTM score and APOE ε4 gene dose (P < .001, uncorrected for multiple comparisons).
Seven of 10 members of the pre-MCI group had further epochs of neuropsychological testing following the decline epoch. All 7 have experienced further decline after the decline epoch, and 1 has clinically converted to amnestic MCI at 24 months following the decline epoch (49 months after his baseline FDG-PET).
It has previously been shown that cognitively normal late–middle-aged and young adult APOE ε4 carriers as a group have reduced CMRgl in regions known to be preferentially affected by AD.18- 20,22- 24 Our goal in the present study was to show that a neuropsychological construct of pre-MCI defined in individuals8 correlated with an imaging endophenotype of AD. Although previous studies8- 20,22- 24 have shown that APOE ε4 is a powerful predictor of preclinical imaging change, the present study ties similar changes to cognitive outcome independent of APOE gene dose. We now find that lower CMRgl in these brain regions correlates with subsequent rates of decline in verbal memory in persons who subsequently met our criteria for amnestic pre-MCI and that these correlations are independent of APOE genotype. Ongoing evaluation of this cohort is necessary to determine the likelihood and time frame of subsequent symptomatic conversion to MCI and AD so that the implied relationship of these findings to AD should be regarded as preliminary. Nevertheless, our findings lend biologic support to the neuropsychological construct of pre-MCI and suggest that pre-MCI may represent an early stage of AD that is the transition point between the asymptomatic memory decline of normal aging and the symptomatic memory decline of MCI.
It is important to define and distinguish asymptomatic decline on neuropsychological tests scores that may be truly pathologic because identification of such a change may be crucial to our understanding the time course of disease that ultimately results in the symptomatic presentation of MCI and AD. Cognitive data from 20-year-old women in the Nun Study25 and neuropathologic26 and brain imaging27 findings in 30-year-olds have shown changes that were predictive or compatible with AD, yet there is no evidence (to our knowledge) for progressive dementia in individuals so young (apart from autosomal dominant familial AD and rare exceptional cases). There is also some evidence to support the notion of a plateau phase of decline, perhaps buffered by cognitive reserve, that may last for years.28,29 Distinguishing whether AD is a lifelong process with a protracted nonprogressive prodromal phase indistinguishable from (or perhaps synonymous with) normal aging, or a disease in which onset occurs at or a few years before the symptomatic onset of memory (or other cognitive) loss, requires a biologically validated operational definition of abnormal decline. The imaging correlations reported in this study lend further support to our operational definition of amnestic pre-MCI and its relationship to AD that was suggested by its genetic association with APOE ε4 homozygosity as previously reported.8
An important limitation of our study is the few individuals who have converted to MCI or AD to date, although ongoing evaluation of this cohort has shown continued test score decline among the individuals with pre-MCI.8 Another important limitation of this study is our cumbersome definition of pre-MCI that requires more testing than may be practical on a large scale. However, at this stage our focus has been on establishing the existence of this putative transitional state, and as further insights are gained, more streamlined operational criteria may become possible. Studies such as ours that use a change measure may be inherently more likely to find a stronger correlation in the greater change group. In this study, the pre-MCI group by definition showed greater decline than the nondecliners, but both groups showed a similar degree of AVLT-LTM score variability, so we do not believe that this accounted for our results. Finally, our cohort is highly enriched with the APOE ε4 allele and so is not characteristic of a community-based sample.
Our present findings support findings from previous studies18- 20,22- 27 suggesting that AD pathogenesis begins well before the onset of symptoms but refine this concept further by suggesting a pre-MCI transitional state that is measurably distinct from normal age-related decline and that represents the onset of progressive disease, distinguishing it from earlier clinically nonprogressive changes. A previous pre-MCI study8 found that neuropsychologically defined cognitive domain decline preceded clinical conversion to MCI and AD by approximately 2 years. In the present study, our FDG-PET findings correlate with this subsequent memory domain decline at a mean of 41 months before it actually occurred. APOE ε4 genotype influences age and risk of AD onset,30,31 as well as presymptomatic neuropsychological decline.8,10 Small et al23 previously demonstrated that baseline CMRgl in several cortical regions may predict subsequent memory performance decline in APOE ε4 carriers. The present study adds to this finding by correlating presymptomatic memory decline as a marker for AD susceptibility that is independent of APOE genotype.
These and previous findings8,10,18- 20,32 suggest that primary prevention therapeutic strategies should be targeted toward individuals no later than in their mid-50s (and preferably earlier), cerebral metabolic changes already reflect impending neuropsychological decline in vulnerable individuals. This is a much younger age than that usually targeted, and current trial paradigms may not be appropriate for such a study. Enrolling asymptomatic individuals, rather than oligosymptomatic patients with MCI, (even if enriched for APOE ε4) would entail thousands of individuals followed up for 20 years or longer, which is impractical. An alternative test paradigm using an imaging endophenotype would require fewer than 100 individuals per treatment arm17,32 and could target asymptomatic individuals for true primary prevention, rather than those with early-stage symptomatic disease. Given the ongoing number of longitudinal cognitive aging studies enrolling healthy individuals, such a trial is plausible. However, an open question is the validity of the pre-MCI construct. Despite published criteria1 and consensus conferences,2 implementation of the MCI construct has led to considerable discussion about how to define it in the most reliable and valid way, and we expect that our pre-MCI construct may promote further discussion. Our proposed amnestic pre-MCI concept is based on specified criteria for the assessment of an acceleration in longitudinal memory decline, initially supported by its association with persons at highest genetic risk for late-onset AD and now supported by a distinctive association between FDG-PET measurements of cerebral hypometabolism in AD regions and subsequent rates of memory decline. Alternative (and potentially complementary) pre-MCI constructs emphasize symptomatic complaints in the absence of objectively documented cognitive loss.33,34 Further study is needed to determine whether objective or subjective pre-MCI models have greater predictive power for subsequent clinical conversion to MCI and AD, but both represent an opportunity for interventions at an earlier disease stage than MCI if an appropriate outcome measure or biomarker can be identified.
In summary, CMRgl reductions in AD-susceptible brain regions correlate with neuropsychologically defined memory domain decline in presymptomatic individuals. This lends further biologic support to the existence of pre-MCI, a transitional state between normal aging and AD-mediated cognitive decline.
Correspondence: Richard J. Caselli, MD, Department of Neurology, Mayo Clinic, 13400 E Shea Blvd, Scottsdale, AZ 85259 (Caselli.Richard@Mayo.edu).
Accepted for Publication: March 11, 2008.
Author Contributions: All authors had full access to the data. Study concept and design: Caselli and Reiman. Acquisition of data: Caselli, Chen, and Reiman. Analysis and interpretation of data: Caselli, Chen, Lee, Alexander, and Reiman. Drafting of the manuscript: Caselli. Critical revision of the manuscript for important intellectual content: Caselli, Chen, Lee, Alexander, and Reiman. Statistical analysis: Caselli, Chen, Alexander, and Reiman. Obtained funding: Caselli and Reiman. Administrative, technical, and material support: Caselli, Chen, Lee, Alexander, and Reiman. Study supervision: Caselli and Chen.
Financial Disclosure: None reported.
Funding/Support: This study was funded by grant R01 MH57899-01 from the National Institute of Mental Health and by the Arizona Alzheimer's Research Consortium (Dr Caselli).
Additional Contributions: Dan Bandy, MS; Justin M. Venditti, BA; and Sandra Yee-Benedetto, BA, provided excellent technical assistance.