In the present report, we identified a subset of regions in the AD-related parietotemporal hypometabolic pattern that are associated with impairment in executive function in MCI and AD patients, as well as memory in MCI patients. Interestingly, the metabolic correlates of memory among AD patients was quite different and involved primarily frontal and orbitofrontal areas, possibly suggesting at a shift from the disease-affected parietotemporal areas to more frontal areas. No significant associations were observed in HC participants between brain metabolism and cognition.
Brain-cognition associations in MCI and AD patients
The observed associations between executive function and brain metabolism in the medial and lateral parietal and temporal lobes in patients with MCI and AD is similar to results found in previous studies in smaller samples (Bracco et al., 2007
; Collette et al., 1997
; Kalpouzos et al., 2005
; Lee et al., 2008
; Nishi et al., 2010
; Yun et al., 2011
). Furthermore, the observed associations of memory with frontal brain metabolism in patients with AD and parietal and temporal metabolism in MCI are also similar to previous reports (Chetelat, Desgranges, de la Sayette, Viader, Berkouk, et al., 2003
; Desgranges et al., 1998
; Desgranges et al., 2002
; Edison et al., 2007
; Eustache et al., 2001
; Nishi et al., 2010
; Perani et al., 1993
; Schonknecht et al., 2011
; Schonknecht et al., 2009
; Slansky et al., 1995
; Teipel et al., 2006
). These associations suggest a functional network of frontal, temporal and parietal regions that are involved in memory and executive function. Extensive atrophy and reductions in neuronal function in MCI and AD in these regions may be responsible for the declines in memory and executive function observed in these patients.
Frontal metabolic activity associated with better memory performance among people with AD: re-allocation or selective disruption?
The metabolic association profiles of executive function in both MCI and AD, as well as the profile of memory in MCI, are located within widespread parieto-temporal cluster that has been demonstrated to show disease-related hypometabolism in numerous previous reports. On the other hand, the association profile of memory in AD patients is dramatically different and exclusively involves frontal and orbitofrontal regions. This observation may suggest a neuronal re-allocation of memory-related metabolic activity from the posterior cortex to the frontal cortex in AD in response to increasing disease severity. Since we did not consider repeated evaluations of the same individuals in the present analysis, disease severity was only an across-subjects factor and the potential observed re-allocation is only a cross-sectional phenomenon. To establish a true re-allocation independent confirmation in longitudinal data are obviously needed.
Compensatory re-allocation implies a shift of memory processing from posterior to anterior areas within subject; for a memory decrease from time 1 to time 2 this dictates a positive correlation with the metabolic signal in the area that is the origin point of the shift and a negative correlation with the metabolic signal in the destination point of the shift. An alternative model would be a selective disruption of memory-processing neural circuits with worsening disease severity: in a disease-free stage the implementation of memory processing might be redundant and involve a multitude of circuits that are employed in a subject-specific manner. As observed in our results, across people this heterogeneity in the profile of circuit utilization might preclude the emergence of focal areas of significance when testing the brain-cognition relationship. After the onset of symptoms with MCI, selected circuits might gradually be more disrupted and affected, uncovering brain-cognition relationships that were hidden because of super-imposed variability before.
These two conceptual models, re-allocation vs. selective disruption, cannot be disentangled in a relative sense: every re-allocation is a selective disruption if the absolute magnitude of metabolic signals cannot be discerned. Absolute quantification of metabolic signals will thus be necessary to distinguish between the two scenarios: in contrast to selective disruption, compensatory re-allocation would imply an absolute increase in the metabolic signal in the areas to which the re-allocation occurs. A consequence that both models have in common is a change in the rank order of cognitive performance for an ideal cohort of disease patients who start from similar disease stages and undergo disease changes at a similar pace. Patients who started with worse performance might end up in a better position relative to their peers, if the latter are affected with more disruption or less ability for successful re-allocation.
Longitudinal follow-up and data from the study arms ADNI GO and ADNI2 present an exciting test bed for these ideas in the near future.
Lack of findings in healthy controls
No significant associations between resting brain metabolism and cognition were observed in HC in the present study. As can be discerned from , within-group variability of both ADNI-Mem and ADNI-Exec is comparable across the three diagnostic groups, so the lack of a brain-cognition correlation cannot be attributed to a restricted range of the behavioral scores in the healthy controls.
We can thus ask further whether the controls fail to exhibit sufficient neural variability to manifest a correlation and whether adding more observations would likely increase the statistical power, based on the sub-threshold observed for the healthy controls. We performed 2 supplementary analyses to address these questions.
(1) We took the voxel location that showed the most reliable correlation between metabolism and ADNI-Mem (p(unc)= 0.0019, MNI=[24,−92,−10], Fusiform gyrus, BA 18). We generated Gaussian-distributed values of all independent variables and residuals based on observed sample means and variances, and took the regression weights from the point estimate for this particular voxel location for 1,000 iterations, varying the numbers of people (from 10 to 500 in increments of 10). For the data that were constructed in these simulations, we then generated a power curve and plotted the fraction of iterations that gave a significant correlation with ADNI-Mem as a function of sample size (graph not shown). A power of 0.8 was obtained for 270 people. (Adding additional observations from the new diagnostic category “Early MCI” from study extensions ADNI GO and ADNI 2 might present an opportunity to substantiate this extrapolation).
(2) We checked all 89 voxel locations reported in to , and computed the ratio of metabolic sample variances of the healthy control participants versus the combined pool of MCI and AD participants. We performed a permutation version of an F-test for the equality of variance with 10,000 iterations for each location. None of the 89 locations yielded significant differences between healthy controls and the MCI/AD participants, confirming the impression from visual inspection that variability is well matched across diagnostic groups.
The results of these additional analyses argue against notable brain-cognition correlations that are obscured because of a lack of variability. However, substantially boosting sample sizes by a factor 3 or more might achieve sufficient power to detect brain-cognition correlations in the healthy controls. (In this light, the significant brain-cognition relationships in the AD group, which had equal numeric strength as the HC group, are even more impressive.)
Unique contributions and limitations
This study provides a unique contribution to the current literature by assessing the relationship between brain metabolism and cognition in the largest sample of AD, MCI, and HC participants evaluated to date. Further, we used psychometrically sophisticated composite measures of memory and executive function that combine the results from multiple assessments, which may provide a better estimate of actual cognitive status than the results from a single test. Finally, the use of voxel-wise analysis techniques allow analysis of brain metabolism across the entire brain without any a priori assumptions, which may allow for identification of potentially novel regions associated with memory and executive function. The fact that parieto-temporal regions were identified in this analysis provides further evidence regarding the importance of metabolic reductions in these regions as mediators of AD-related cognitive dysfunction. Furthermore, the relatively novel identification of the association of hypometabolism in frontal regions with memory dysfunction in AD may not have been identified with region of interest or whole brain metabolic pattern techniques.
Despite the significant contributions of the present study, a few limitations are notable. First, this analysis used only cross-sectional FDG PET and psychometric data. Therefore, the assessment of brain-behavior relationships was limited to between-subject comparisons. Any conclusions drawn about changes in metabolism in relation to increasing disease severity and cognitive decline were preliminary, as other factors could explain between-subject differences in metabolism and/or cognition. Future studies utilizing longitudinal FDG PET and neuropsychological test data will help to elucidate relationships between changes in brain metabolism and progressive decline in memory and executive function within subjects. The present study also limited the disease-related differences in metabolism to a simple multiple regression model that does not address the possible unique contributions of each disease stage. Further, quadratic or u-shaped associations with disease severity could not be assessed. For our discovery of brain-cognition relationships we performed linear regression within each diagnostic group separately. This is a compromise that offers analytical tractability, while allowing for disease-stage specific changes in the associations between cognition and brain metabolism. Future studies could evaluate more sophisticated and complex models that incorporate metabolic changes in a longitudinal framework. Ideally this would be performed in larger samples with sufficient power for estimation of non-linear and higher dimensional models. Finally, a number of other variables of interest, which may modulate the relationship of brain metabolism and cognition, were not included in the presented analyses. For example, genetic variation may be an important factor mediating brain-cognition relationships. In fact, a previous study demonstrated a significant effect of apolipoprotein E (APOE) genotype on brain metabolism in the ADNI cohort (Langbaum et al., 2009
). Therefore, future studies could evaluate the role of APOE and other genetic variants in modulating relationships between brain metabolism and memory and executive function. Future studies could also exploit the multimodal nature of ADNI and use information about brain structure as an additional covariate (Kanda et al., 2008
; Samuraki et al., 2007
). The goal of such a study would be to identify metabolic correlates of disease severity, memory and executive function above and beyond
the effects explained by atrophy and cortical thinning, which can be expected to occur in the course of the disease. Such `metabolic density' profiles or, in other words, topographic patterns of metabolic activity per conserved unit of gray matter, might yield valuable knowledge about AD-related neural changes and/or provide additional diagnostic and predictive information for disease identification and monitoring. Other variables have also been shown to affect brain metabolism and/or cognition, including the presence of amyloid deposition (Li et al., 2008
; Mormino et al., 2009
) and cerebrovascular changes (Brickman et al., 2011
; Carmichael et al., 2010
). Future studies could evaluate the role of other independent variables in mediating the relationship between brain metabolism and cognition.
Summary and conclusions
In conclusion, we observed significant positive associations between brain metabolism measured using FDG PET and executive function and memory in patients with MCI and AD, but not in HC. Impairments in executive function were associated with parietotemporal hypometabolism in both MCI and AD. On the other hand, impaired memory was associated with reduced metabolism in parietotemporal regions in MCI patients and frontal regions in AD patients. Overall, the results of the present study underscore the importance of changes in brain metabolism in the cognitive impairment seen in the prodromal and early stages of AD.