Does working memory training change the brain? A simultaneous PET/MRI study

Introduction: Working memory (WM) training has been considered as a potential intervention to delay age- and disease-related cognitive decline. An efficient training is expected to rely on so called neural plasticity (NP). Given a variety of mechanisms underlying NP, we applied a combination of neuroimaging techniques such as PET, fMRI, and DTI in a large group of middle-aged healthy subjects (n=45). Simultaneous acquisition of 18F-fluorodeoxyglucose (FDG) PET and MRI data on a hybrid PET/MR system was conducted before and after an intensive, supervised, online-based WM training. We hypothesized an increase in connectivity of the default mode and central executive networks following training. Comprehensive analyses of the neuroimaging data addressed potential changes in metabolic, functional and microstructural indices of the brain integrity, at a network and a single voxel level. Resting state networks were extracted from PET and fMRI data using independent component analysis. Fractional anisotropy and diffusivity measures of white matter integrity were derived from DTI data using tract-based spatial statistics. In addition, established univariate analyses were applied to all imaging data at a voxel level. Anterior and posterior default mode, central executive, and salience network were identified in resting state FDG-PET and fMRI data. In both modalities no significant changes in network integrity were found. Similarly, we observed no changes in white matter integrity. Univariate analyses did not reveal significant findings, too. In particular, there were no significant changes in relative FDG uptake following WM training. Analyses of behavioral data revealed a significant practice effect, but no transfer effect, i.e. an improvement in tasks other than the trained tasks. In summary, we used a state of the art neuroimaging instrumentation and a WM training paradigm to investigate the phenomenon of NP. A comprehensive analysis of multi-modal imaging data did not reveal any significant changes at a network or single voxel level. Subgroup analyses will follow. The present results may indicate that measurable NP occurs only in the case of significant transfer effects. As such effects were not detected, the concept of WM training should be revised.