Can Neuroimaging Predict Dementia in Parkinson's Disease?

Juliette H. Lanskey; Peter McColgan; Anette E. Schrag; Julio Acosta-Cabronero; Geraint Rees; Huw R. Morris; Rimona S. Weil

Disclosures

Brain. 2018;141(9):2545-2560. 

In This Article

Connectomics and Graph Theory

Connectomics is an emerging field, where neuroimaging data are used to generate brain networks. Functional and structural brain networks can be constructed using rs-fMRI and diffusion tractography, respectively. The brain is divided into distinct regions based on structural or functional information. Each brain region represents a node that may be connected to other nodes in the brain network. In structural brain networks these connections represent anatomical white matter connections, whereas in functional brain networks, connections represent temporal correlations from functional MRI time series (Fornito and Bullmore, 2015).

The topological characteristics of brain networks are described using a mathematical approach known as graph theory (Rubinov and Sporns, 2010). This quantifies relationships across the brain network. It includes measures of network segregation such as the clustering coefficient, which represents the fraction of a node's neighbours that are connected to each other (Glossary).

Functional Connectomics

Rs-fMRI is commonly used in Parkinson's disease connectomics and cognition. However, network construction methodologies vary, making comparisons across studies difficult (Table 1). One study revealed higher clustering coefficients and modularity in PD-MCI than in patients without cognitive involvement (Baggio et al., 2014), suggesting increased segregation of the functional connectome. Consistent with this, higher clustering coefficients were associated with impaired performance on visuo-spatial tasks.

In contrast, markers of increased segregation: higher clustering coefficient and reduced global efficiency were associated with a 'better' cognitive phenotype in another study (Lopes et al., 2017). Increased segregation may relate to increased connections within specialized brain modules or loss of connections between brain modules, and depending on the balance of these, could be associated with either improved or worsened cognition. Conflicting results may also arise due to methodological differences in calculating graph theory metrics.

A machine learning approach was recently used to classify PD-MCI and Parkinson's disease without cognitive involvement based on functional connectomes (Abós et al., 2017) (Figure 2C). Connections used in the classification procedure correlated with executive and visuospatial scores. A network-based statistics (NBS) analysis revealed reduced connectivity in occipital-temporal and occipital-frontal connections in PD-MCI compared to controls. This suggests that inter-regional connections, particularly those involving the occipital lobe, are associated with cognitive impairment in Parkinson's disease. Correlations between connectivity of visuospatial modules and cognitive performance are even seen in early, drug-naïve Parkinson's disease (Luo et al., 2015).

Another group investigated differences in functional connectomes across cognitive subgroups in Parkinson's disease (Lopes et al., 2017) (Figure 2D). A cluster analysis was performed on a neuropsychological battery splitting patients into five phenotypes, from cognitively intact to severe deficits across cognitive domains. NBS analysis revealed associations between cognitive phenotype and connections involving frontal, temporal, occipital and basal ganglia regions. Separating connections into interhemispheric and intrahemispheric subtypes showed that interhemispheric connections differed between phenotype extremes. This suggests that loss of connections between hemispheres may impact on cognition in Parkinson's disease [similar associations are seen in Huntington's disease (McColgan et al., 2017)].

Structural Connectomics

Global network changes in PD-MCI are also seen using structural connectomics (Galantucci et al., 2017), where higher clustering coefficient and reduced global efficiency were found in PD-MCI compared to Parkinson's patients without cognitive involvement. This is consistent with the model that PD-MCI is associated with increased network segregation and reduced integration and that loss of connections between functional brain modules impairs cognitive functions that require inter-module cooperation (Lopes et al., 2017).

Connectomics and Regional Gene Expression

Brain structure and function at the macrostructural level can now be linked to gene expression at a cellular level using atlases of gene expression microarray data (Hawrylycz et al., 2012, 2015). In Parkinson's disease, regions with the largest reductions in connection strength show highest regional expression of MAPT (Rittman et al., 2016). In health, brain regions with long-range connections are enriched for genes involved in oxidative metabolism and mitochondrial function (Vértes et al., 2016). This is in keeping with the observation that genetic mutations associated with Parkinson's disease frequently affect these pathways (Helley et al., 2017).

Taken together, connectomic studies of cognitive impairment in Parkinson's disease suggest loss of network segregation and integration, affecting hub regions specifically, with prominent loss of connections between hemispheres and specialized functional modules, particularly in posterior brain regions. However, integrating findings across studies is difficult because of methodological differences and the use of unselected Parkinson's disease cohorts where cognitive phenotypes are ill-defined. These studies highlight the potential for connectomics to identify vulnerable networks and connections involved in PDD. As graph-theoretical methodologies are refined and applied consistently across groups, or within large-scale collaborative studies, they will have increasing importance for early neuroimaging detection of cognitive involvement in Parkinson's disease.

processing....