Dynamical mechanisms of epileptic transitions

 

Epilepsy is dynamical disease of the brain. Such diseases are characterized by a qualitative changes from normal behavior to abnormal dynamics. We may assume that epileptic brain features two states: the interictal state characterized by a normal electroencephalography (EEG) ongoing activity, and the ictal state, that is characterized by paroxysmal occurrence of synchronous oscillations and is generally called, in neurology, a seizure. An example of EEG recording during epileptic seizure of the absence type in human patient is shown in Fig. 1.


sw pattern
Fig. 1. An example of the EEG recording of absence epileptic seizure. Paroxysmal 3 Hz spike and wave pattern emerges abruptly out of normal background and suddenly ceases after few seconds.

The question that we asked ourselves was: why a seizure starts and why does it stop? Rephrasing this question in a more formal manner: what are the dynamical mechanisms of transitions between normal and abnormal state? To answer this question we developed a computational model of the thalamocortical network. This network is considered to be responsible for generation of the absence epileptic seizures. The model diagram is shown in Fig. 2.


model

Fig. 2. Diagram of the model of the thalamocortical network consisting of four neuronal populations: PY - pyramidal cells, IN - cortical interneurons, TC - thalamocortical relay cells, RE - thalamic reticular cells. The Simulink code of the model is available at: https://senselab.med.yale.edu/ModelDB/showmodel.asp?model=111880

Model’s activity corresponds to local field potentials or, in general, to EEG signals. An example of model simulation is shown in Fig. 3. One can see that model’s activity undergoes a change from normal behavior (small amplitude) to seizure-like activity (high amplitude oscillation) and back to normal state. Such transitions in the model occur spontaneously, i.e., they are not induced by model parameter changes. Understanding mechanisms of these transitions in the model may help to understand what happens in a real epileptic brain.

signals
Fig. 3. An example of seizure-like activity in the model (upper panel) and and seizure in an epileptic rat. The model parameters were based on rat data, therefore the dominant frequency of the modelled seizure signal is 9 Hz, comparable to the rat, as shown by power spectra on the right side.


System analysis showed that the model possesses bistable dynamics. In other words, for the same set of parameters the model may exhibit two types of activity: normal and seizure-like. Stochastic fluctuations present in the network may flip the system across the separatrix separating the two coexisting states (attractors) as illustrated, by means of phase-space reconstruction, in Fig. 4.

phase space

Fig. 4. Two dimensional reconstruction of the model's phase-space showing two coexisting  attractors  - normal and ictal, separated by a separatrix (red line). If the trajectory is confined within the separatrix the system exhibits normal activity. After crossing the separatrix, the system switches to ictal attractor and seizure activity is generated. Due to fluctuations present in the system, the trajectory switches between the two attractors, what results in randomly occurring seizure onsets and offsets. This figure appeared in Computer modelling of epilepsy, W.W. Lytton, Nature Reviews Neuroscience, 2008.


Experimental validation of  model’s prediction is essential part of the modeling process. In the model, the transitions between the two states occur randomly in time with constant probability. Accordingly, the
distributions of durations of normal and seizure-like epochs are exponential. (Derivation of exponential distribution of inter-event durations from constant probability rate of event occurrence can be found here). Analysis of long-term experimental recordings in human subjects and animal models confirmed model hypothesis in many cases but also refuted it in some other cases. Fig. 5 shows model predictions (upper panel), examples of positive experimental validation (middle panel) and examples clearly deviating from model predictions (lower panel).


Model predictions
inter ictal model

Positive validation - human data

ictal2
inter2

Negative validation - human data
ictal3
inter3
Fig. 5. Distributions of lengths of seizure and seizure-free epochs fitted with exponential function (red line). Upper panel: simulated data exhibiting exponential distribution. Middle panel: experimental human data that confirm model predictions (exponential function fits well the histogram scores). Lower panel: experimental human data from another patient, where model predictions are not fulfilled - the histograms deviate from exponential fit.

We may conclude that our modeling work brought some new insights into fundamental laws governing the generation of epileptic seizures. At the same time new questions arose such as: which physiological mechanisms modulate the transition probabilities and that are responsible for deviation from the exponential law? New,  hypothesized mechanism can be included in the model and predictions of the extended model can be tested again against experimental data. This is how a progress can be made!

Finally, we should note that here we considered mainly seizures of the absence type originated in the thalamocortical network. Epilepsy is a heterogeneous disease that includes many seizure types. Different scenarios of epileptic transitions are presented and discussed in the review article Dynamical Diseases of Brain Systems: Different
Routes to Epileptic Seizures,
Lopes da Silva et al., IEEE TBME 2003.