Learning networks in health and Parkinson's disease: reproducibility and treatment effects

Abstract

In a previous H(2) (15)O/PET study of motor sequence learning, we used principal components analysis (PCA) of region of interest (ROI) data to identify performance-related activation patterns in normal subjects and patients with Parkinson's disease (PD). In the present study, we determined whether these patterns predicted learning performance in subsequent normal and untreated PD cohorts. Using a voxel-based PCA approach, we correlated the changes in network activity that occurred during antiparkinsonian treatment and their relationship to learning performance. We found that the previously identified ROI-based patterns correlated with learning performance in the prospective normal (P < 0.01) and untreated PD (P < 0.05) cohorts. Voxel analysis revealed that target retrieval was related to a network characterized by bilateral activation of the dorsolateral prefrontal, premotor and anterior cingulate cortex, the precuneus, and the occipital association areas as well as the right ventral prefrontal and inferior parietal regions. Target acquisition was associated with a different network involving activation of the caudate, putamen, and right dentate nucleus, as well as the left ventral prefrontal and inferior parietal areas. Antiparkinsonian therapy gave rise to changes in retrieval performance that correlated with network modulation (P < 0.01). Increases in network activation and learning performance occurred with internal pallidal deep brain stimulation (GPi DBS); decrements in these measures were present with levodopa. Our findings suggest that network analysis of activation data can provide stable descriptors of learning performance. Network quantification can provide an objective means of assessing the effects of therapy on cognitive functioning in neurodegenerative disorders.

Figure 1

Voxel‐based network analysis of H₂ ¹⁵O/PET data from 18 healthy controls scanned during motor sequence learning: retrieval pattern. A: In the normal cohort, subject scores for the first principal component (PC 1, accounting for 9.2% of the Subject × Voxel variance) correlated with target retrieval (R² = 0.60, P < 0.001). B: This network topography (Table II) was characterized by bilateral learning‐related activations (RTlearn > Mpred) in the dorsolateral prefrontal cortex (DLPFC) and premotor cortex (PMC), and in the left anterior cingulate area and the right inferior parietal cortex. This learning pattern contained additional activations in the ventral prefrontal and occipital association areas, and in the lateral cerebellum. [Positive region weights (red‐yellow) were thresholded at Z = 2 to display clusters contributing significantly (P < 0.01) to the network (see text)].

Figure 2

Voxel‐based network analysis of H₂ ¹⁵O/PET data from 18 healthy controls scanned during motor sequence learning: acquisition pattern. A: In the normal cohort, subject scores for a linear combination of the third and fifth principal components (PC 3 and PC 5, accounting for 9.0% of the Subject × Voxel variance) correlated with target acquisition (R² = 0.31, P < 0.01). B: This network topography was characterized by learning‐related activations (RTlearn > Mpred) in the caudate and ventral prefrontal cortex, as well as in the superior temporal gyrus and posterior parietal regions. [Positive region weights (red‐yellow) were thresholded at Z = 2 to display clusters contributing significantly (P < 0.01) to the network (see text)].

Figure 3

Changes in learning performance and network activity during acute antiparkinsonian intervention with internal pallidal deep brain stimulation (squares) or levodopa infusion (triangles). Treatment‐mediated (ON–OFF) changes in the subject scores for the normal retrieval network correlated significantly with concurrently measured changes in the global retrieval index (R² = 0.40, P < 0.01). Learning performance and network activity tended to increase during pallidal stimulation and to decline with levodopa infusion (see text).