The Virtual Mouse Brain: A Computational Neuroinformatics Platform to Study Whole Mouse Brain Dynamics

Abstract

Connectome-based modeling of large-scale brain network dynamics enables causal in silico interrogation of the brain's structure-function relationship, necessitating the close integration of diverse neuroinformatics fields. Here we extend the open-source simulation software The Virtual Brain (TVB) to whole mouse brain network modeling based on individual diffusion magnetic resonance imaging (dMRI)-based or tracer-based detailed mouse connectomes. We provide practical examples on how to use The Virtual Mouse Brain (TVMB) to simulate brain activity, such as seizure propagation and the switching behavior of the resting state dynamics in health and disease. TVMB enables theoretically driven experimental planning and ways to test predictions in the numerous strains of mice available to study brain function in normal and pathological conditions.

Keywords: connectome; fMRI; modeling; resting state.

Graphical abstract

Figure 1.

A, We used The Allen Connectivity Builder to build the structural connectivity matrix. The color map represents the connections strengths with a base-ten logarithmic scale. The resolution of the grid is 100 μm; the weights of the matrices are defined as the ratio between the projection and the injection density; all the areas in the parcellation have at least one injection experiment that has infected >50 voxels in those areas; the matrix contains only regions with a volume >2 mm³. B, Simulated resting state BOLD time series using the connectome built in A, and the eMFM to model the dynamics of each brain area. C, FCD matrix obtained from the time series. The three black segments (I, II, and III) correspond to epochs of stability of the FCD identified with the spectral embedding technique. D, Functional hubs detected in silico mapped on brain sections using the brain region volume visualizer. Images in the same row represent the plotting of the eigenvectors components, in absolute value, of the FC belonging to the same epoch. Images organized in different columns refer to eigenvectors belonging to different eigenvalues of the matrices. The scale used allows highlighting only the brain area associated to component of the eigenvector greater than the half of the maximum component. Such scale permits to efficiently visualize the relative difference between eigenvectors. According to our definition (see Materials and Methods), the areas with warm colors are the hub regions of the brain network defined by the FC matrices calculated over the relative epoch; the importance of each hub region is proportional to the corresponding eigenvalue. E, Experimental resting state networks and the corresponding functional hubs detected in Mechling et al. (2014). The Ipython scripts to obtain the results in the figure are Figure 1-1 and Figure 1-2 (Extended data).

Figure 2.

A, Connectivity matrix obtained from Calabrese et al. (2015). B, Simulated resting state BOLD time series using the connectome shown in A, and the eMFM to model the dynamics of each brain area. C, FCD matrix obtained from the time series. The black segment identifies the epoch of stability of the FCD identified with the spectral embedding technique. D, Functional hubs detected in silico mapped in mouse brain sections using the brain region volume visualizer, as in Figure 1C.

Figure 3.

A, B, BOLD signals and the corresponding FCD matrix, respectively, obtained by simulating the mouse brain in which some links are removed to mimic epilepsy conditions. C, Functional hubs detected in the epileptic mouse brain after removing links as seen in some forms of epilepsy. The hubs displayed here are extracted from the FC matrix calculated over all the simulated BOLD signals (20 min), i.e., the global FC, since the FCD simulated in the epileptic mouse brain does not present evident sign of nonstationarity and consequently the epoch of stability cannot be detected. The Ipython script to obtain the results in the figure is in Figure 3-1 (Extended data).

Figure 4.

Simulating epileptiform activity in the mouse brain. A, The time series show simulated seizure genesis and propagation (direct current recording) in silico. B, The graph shows the propagation pattern. Time 0 corresponds to seizure onset in the left hippocampus. On the x-axis, regions are ordered as they are progressively recruited in Toyoda et al. (2013). The y-axis shows the average time of recruitment in arbitrary units of these regions after triggering a seizure in the left hippocampus in silico. Note the good match between simulated and experimental data. Extensive names of the region composing each group are illustrated in the table in Table 1. C, The time distance from seizure onset in the left hippocampus is given by the color scale and plotted in the brain volume for each region. The Ipython script to obtain the results in the figure is Figure 4-1 (Extended data).

Figure 5.

The cartoon illustrates how it is possible to use TVMB to do predictions when studying aging. A mouse can be scanned at different times t extracting anatomic and functional brain information. The anatomic information can be processed to obtain a connectome that can be used in TVMB to create a virtual mouse at each time step. The functional experimental information can be compared with the predictions done in TVMB, investigating how, for example, anatomic modifications during aging affect whole-brain dynamics. Multiple other testable predictions can be done. For example, explore in silico which types of neurones can be stimulated (or silenced) to activate specific resting state networks. The predictions can then be tested in ad hoc transgenic mice with optogenetics.