Metabolic brain networks in neurodegenerative disorders: a functional imaging approach

Abstract

Network analysis of functional brain imaging data is an innovative approach to study circuit abnormalities in neurodegenerative diseases. In Parkinson's disease, spatial covariance analysis of resting-state metabolic images has identified specific regional patterns associated with motor and cognitive symptoms. With functional imaging, these metabolic networks have recently been used to measure system-related progression and to evaluate novel treatment strategies. Network analysis is also being used to characterize specific functional biomarkers for Huntington's disease and Alzheimer's disease. These networks have been particularly helpful in uncovering compensatory mechanisms in genetically at-risk individuals. Ongoing developments in network applications are likely to enhance the role of functional imaging in the investigation of neurodegenerative disorders.

Figure 1. Parkinson's disease-related pattern: validation and correlates

(a) Parkinson's disease-related pattern (PDRP). This motor-related metabolic spatial covariance pattern is characterized by hypermetabolism in the thalamus, globus pallidus (GP), pons, and primary motor cortex, associated with relative metabolic reductions in the lateral premotor (PMC) and posterior parietal areas [13]. [In the representative slices, relative metabolic increases are displayed in red; relative metabolic decreases are displayed in blue. Slices were overlaid on a standard MRI brain template.] (b) Cluster analysis of PD-related regional topographies identified in seven independent combined groups of PD patients and healthy subjects (Populations A-G) who underwent metabolic imaging studies on different tomographs. Region weights for the disease-related covariance patterns (rows) were similar across populations (r>0.7, p<0.001). Consistent increases in network-related metabolic activity (red) were present in the putamen/globus pallidus, thalamus, and cerebellum; relative decreases (green) were present in the lateral premotor/prefrontal cortex and parietal association regions. Of note, the pattern topographies for the various populations (columns) clustered according to the resolution of the respective tomographs, such that similar PDRPs were discerned for instruments of low (Populations B, F: 12 mm FWHM), medium (Populations A, C, E: 8 mm FWHM), and high (Populations D, G: 4 mm FWHM) resolution. [A region-of-interest (ROI) approach was used in which a standard brain template was used to measure local metabolic activity in 30 anatomically pre-specified areas. For each population, a significant disease-related pattern was identified using subject scores for the first two principal components (singly or in linear combination) such that the two groups were separated at p<0.001, discriminant analysis. In this display, the region weights for each significant pattern, reflecting the contribution of local metabolic activity in each ROI to overall pattern expression [4, 9, 13], were displayed on a heat map based on hierarchical clustering analysis [60]. Cluster trees were created using a complete linkage method with the Pearson correlation coefficient (uncentered) as the similarity measure. Region weights for the PDRPs from each population were presented elsewhere [20]]. (c) PDRP expression correlates with Unified Parkinson's Disease Rating Scale (UPDRS) motor scores from a combined group of three independent PD populations (n=65; r=0.68, p<0.001) [21, 22, 61]. These correlations were also significant within each of the individual patient cohorts (circles: n=27; r=0.66; p<0.001; squares: n=15; r=0.65, p<0.01; triangles: n=23; r=0.76, p<0.001). (d) PDRP expression correlates with intraoperative measurements of spontaneous subthalamic nucleus (STN) firing rate (n=17; r=0.76, p<0.007) recorded in PD patients undergoing the implantation of deep brain stimulation (DBS) electrodes [32]. The regression line includes corrections for individual differences in disease duration and motor ratings across the subjects. [Brain 131 (Pt 5), 1373-1380 Copyright © 2008 of Oxford University Press]

Figure 2. Parkinson's disease-related cognitive pattern: validation and correlates

(a) Parkinson's disease-related cognitive pattern (PDCP). This cognition-related metabolic spatial covariance pattern is characterized by hypometabolism of dorsolateral prefrontal cortex (PMC), rostral supplementary motor area (preSMA), precuneus, and posterior parietal regions, associated with relative metabolic increases in the cerebellum [37]. [In the representative slices, relative metabolic increases are displayed in red; relative metabolic decreases are displayed in blue. Slices were overlaid on a standard MRI brain template.] (b) PDCP expression correlates with performance on neuropsychological tests of memory and executive functioning in non-demented PD patients. For the California Verbal Learning Test: Sum 1 to 5 (CVLT sum), this correlation was significant for the entire cohort (n=56; r=−0.67, p < 0.001), as well as for the original group used for pattern derivation (circles; n=15: r=−0.71, p=0.003) and in two prospective validation groups (squares; n=25: r=−0.53, p=0.007; triangles; n=16: r=−0.80, p<0.001) [37]. [Revised version of Neurology 34(2):714-723 Copyright © 2007 by AAN Enterprises, Inc.] (c) Bar graph of PDCP expression (mean ± SE) in PD patients with dementia (PDD), multiple domain mild cognitive impairment (MCI(m)), single domain mild cognitive impairment (MCI(s)), PD patients without mild cognitive impairment (MCI(-)), and in healthy control subjects. There was a significant difference in PDCP expression across the patient and control groups (F(4,70)=8.87, p<0.001; one-way ANOVA) and among the PD groups (F(3,56)=4.84; p<0.005), with higher expression in the PDD and MCI(m) cohorts compared to the MCI(-) cohort (p<0.03; Tukey-Kramer HSD). For each PD group, PDCP expression was separately compared to healthy control values using Student t-tests. The asterisks denote significant increases in network activity relative to controls (*p<0.05, **p<0.005, ***p<0.0001) in all PD categories including MCI(-). [Revised version of Neurology 70(16 Pt 2):1470-1477 Copyright © 2008 by AAN Enterprises, Inc.] (d) Progressive increases in PDRP (circles) and PDCP (triangles) expression as a function of disease duration. Regression analysis showed that activity of each of the two networks increased linearly over time (p<0.0001). However, the rate of increase in pattern expression differed over time for the two networks (p<0.02, interaction effect), with slower progression of PDCP activity relative to the PDRP.

Figure 3. PD-related metabolic networks: treatment effects

(a) Bar graph illustrating treatment-mediated changes (mean ± SE) in PDRP and PDCP expression measured in scans of cerebral metabolic rate for glucose (CMR; filled bars) and cerebral blood flow (CBF; open bars) [26]. Top: There was a significant dissociation in PDRP expression between the CMR and CBF treatment responses (p<0.001) with levodopa (LD; left), but not with subthalamic nucleus deep brain stimulation (STN DBS; middle) or in PD controls (PD CTRL; right) undergoing repeat imaging on stable doses of dopaminergic medication [26]. Bottom: No significant CMR or CBF responses in PDCP expression were found with either treatment (levodopa or STN DBS) in the same patients. [J. Neurosci. 28, 4201-4209 Copyright © 2008 by the Society for Neuroscience] (b) Changes in PDRP and PDCP expression over time for the operated (filled circles) and unoperated (open circles) hemispheres following subthalamic (STN) gene therapy [33]. Top: There was a significant difference (p<0.002) in the time course of PDRP expression between the two hemispheres. Network activity declined during the first six months and then increased over the subsequent six months in the operated hemisphere. By contrast, network activity in the unoperated hemisphere increased continuously over the entire 12 months following surgery. Bottom: There was no change in PDCP activity over time in either of the two hemispheres following gene therapy (p=0.72). [The dashed line represents one standard error above the normal mean value of zero for each pattern]. [Proc. Natl. Acad. Sci. U. S. A. 104, 19559-19564 Copyright © 2007 National Academy of Sciences, U.S.A.]

Figure 4. Spatial covariance patterns for Huntington's and Alzheimer's diseases

(a) Huntington's disease (HD)-related spatial covariance pattern. This pattern was characterized by relative metabolic decreases in the striatum and cingulate cortex, associated with relative increases in the ventral thalamus, cerebellum, motor cortex, and occipital lobe [50]. [In the representative slices, relative metabolic increases are displayed in red; relative metabolic decreases are displayed in blue. Slices were overlaid on a standard MRI brain template. [Brain 130 (Pt 11), 2858-2867 Copyright © 2007 of Oxford University Press] (b) Alzheimer's disease (AD)-related spatial covariance pattern identified using continuous arterial spin labeling (CASL) MRI [57]. Areas displayed with network-related CBF reductions in AD relative to healthy controls were found to be reliable by bootstrap resampling. Most significant areas within the network were located in the vicinity of the parahippocampal gyrus. No areas with positive region weights on the pattern (i.e., relative CBF increases in the AD subjects) were discerned [Reprinted by permission from Macmillan Publishers Ltd: J Cereb Blood Flow Metab 28(4): 725-736 Copyright © 2007].

Figure I (Box 1)

Metabolic Network Analysis: Computational Diagram

Figure I (Box 3)

Metabolic Network Expression: Computational Diagram