Reorganization of cortical oscillatory dynamics underlying disinhibition in frontotemporal dementia

Abstract

The distribution of pathology in frontotemporal dementia is anatomically selective, to distinct cortical regions and with differential neurodegeneration across the cortical layers. The cytoarchitecture and connectivity of cortical laminae preferentially supports frequency-specific oscillations and hierarchical information transfer between brain regions. We therefore predicted that in frontotemporal dementia, core functional deficits such as disinhibition would be associated with differences in the frequency spectrum and altered cross-frequency coupling between frontal cortical regions. We examined this hypothesis using a 'Go-NoGo' response inhibition paradigm with 18 patients with behavioural variant frontotemporal dementia and 20 healthy aged-matched controls during magnetoencephalography. During Go and NoGo trials, beta desynchronization was severely attenuated in patients. Beta power was associated with increased impulsivity, as measured by the Cambridge Behavioural Inventory, a carer-based questionnaire of changes in everyday behaviour. To quantify the changes in cross-frequency coupling in the frontal lobe, we used dynamic causal modelling to test a family of hierarchical casual models, which included the inferior frontal gyrus, pre-supplementary motor area (preSMA) and primary motor cortex. This analysis revealed evidence for cross-frequency coupling in a fully connected network in both groups. However, in the patient group, we identified a significant loss of reciprocal connectivity of the inferior frontal gyrus, particularly for interactions in the gamma band and for theta to alpha coupling. Importantly, although prefrontal coupling was diminished, gamma connectivity between preSMA and motor cortex was enhanced in patients. We propose that the disruption of behavioural control arises from reduced frequency-specific connectivity of the prefrontal cortex, together with a hyper-synchronous reorganization of connectivity among preSMA and motor regions. These results are supported by preclinical evidence of the selectivity of frontotemporal lobar degeneration on oscillatory dynamics, and provide a clinically relevant yet precise neurophysiological signature of behavioural control as a potential pharmacological target for early phase experimental medicines studies.

Figure 1

Illustration of the experimental background and principal hypothesis. (A) Voxel-based morphometry of grey and white matter loss in the patient versus control group. The areas in red confirm the expected reductions in grey and white matter tissue in frontal and temporal cortex. The areas coloured blue had strong evidence for normal cortical volume (Bayesian probability of the null > 0.7). While prefrontal regions are particularly atrophic (red), the precentral gyrus, including the primary motor cortex, had evidence of normal grey matter volume (blue). (B) GFAP (left) and haematoxylin and eosin (right) slices from frontal cortex in a patient with bvFTD, which demonstrate a clear outer layer emphasis to the pathology (as indicated by the brackets). (C) A model of the regions included in the DCM analysis (IFG, preSMA and motor cortex), and the specified connectivity. (D) Schematic illustration of the framework for the experimental motivation: superficial layers of prefrontal cortex generating gamma oscillations have reciprocal connections with the deeper layers of motor regions which support beta oscillations. (E) Hypothesis: the layer-specific burden of pathology is predicted to disrupt the cross-frequency coupling, attenuate the beta desynchronization and consequently impair movement control.

Figure 2

Time frequency spectra and relationship with behaviour. (A) Time frequency spectra for controls and bvFTD patients for successful Go and NoGo trials. A clear ERD/ERS in the beta/alpha bands and an early increase in theta are evident, which are diminished in patients. Contrasts between conditions (vi–vii) and between groups (iii, vi) are plotted with significant statistical thresholds outlined in black (F tests, P < 0.05 clusterwise corrected after P < 0.001 voxelwise threshold). The interaction (Go versus NoGo × Patients versus Controls, ix) reveals two windows of significance, used for further analyses. [B(i)] Time of peak beta desynchronization plotted against reaction times, revealing a tight link between desynchronization and movement. [B(ii)] Plot of relative beta power (the difference between Go and NoGo trials), at the peak of the significant interaction (18 Hz at 540 ms), which correlates with behavioural measures of disinhibition from the CBI. The correlation is positive, indicating that patients who are more behaviourally disinhibited have less desynchronization during successful NoGo trials.

Figure 3

Details of dynamic causal models. (A) DCM model space with seven different architectures. (B) Bayesian model selection reveals the non-linear family best fitted the data, which allows for cross-frequency coupling. Of the seven models within the non-linear family, Model 1 was the winning model, with task modulation of reciprocal connections between all regions. For both controls and bvFTD patients, Model 1 had the greatest exceedance and posterior probabilities and a difference in log evidence between the winning and second best model > 3. (C) Time-frequency spectra for each region, for each condition and group.

Figure 4

Statistical parametric maps of the frequency to frequency estimations for each interregional connection of the winning model, plotted on a log scale in Hz. Negative (blue) values represent a suppression effect: a power increase in the origin frequency decreases power in the target region, and positive (red) values represent an increase in origin frequency that increases power in the target region. In B and D the differences between the bvFTD and control groups are outlined in black (P < 0.05 FWE cluster-wise corrected after P < 0.001 voxel-wise correction).

Figure 5

Statistical parametric maps of the frequency to frequency estimations for each of the self-connections of the winning model. Negative (blue) values represent a suppression effect: a power increase in the origin frequency decreases power in the target region, and positive (red) values represent an increase in origin frequency that increases power in the target region. The significant cluster differences between the bvFTD and control groups are outlined in black (P < 0.05 FWE cluster-wise corrected after P < 0.001 voxel-wise correction). The self-connections for all trials (A matrix) show within-frequency couplings that are strongly negative on the diagonal. This feature of cortical networks is included in dynamic causal models, in which estimations of intrinsic connections depend on self-inhibition for stability, and this constraint establishes a negative prior on the coupling parameter (Friston et al., 2003).