Push and Pull: Borrowing Concepts From Control Theory for Seizure Spread

Commentary

Virtual Cortical Resection Reveals Push-Pull Network Control Preceding Seizure Evolution.

Khambhati AN, Davis KA, Lucas TH, Litt B, Bassett DS. Neuron 2016;91:1170–1182.

In ~20 million people with drug-resistant epilepsy, focal seizures originating in dysfunctional brain networks will often evolve and spread to surrounding tissue, disrupting function in otherwise normal brain regions. To identify network control mechanisms that regulate seizure spread, we developed a novel tool for pinpointing brain regions that facilitate synchronization in the epileptic network. Our method measures the impact of virtually resecting putative control regions on synchronization in a validated model of the human epileptic network. By applying our technique to time-varying functional networks, we identified brain regions whose topological role is to synchronize or desynchronize the epileptic network. Our results suggest that greater antagonistic push-pull interaction between synchronizing and desynchronizing brain regions better constrains seizure spread. These methods, while applied here to epilepsy, are generalizable to other brain networks and have wide applicability in isolating and mapping functional drivers of brain dynamics in health and disease.

Our model of focal epileptic seizures has evolved over time. In the past, focal epilepsy was thought to originate at a circumscribed focal cortical site, the distinct and circumscribed seizure focus. Seizures spread from there, characterized by synchronized EEG activity. However, is seizure evolution predetermined by the overwhelming spread of synchronized electrical activity from the seizure focus, or is it a failure of the surrounding neuronal tissues to control the synchronization? Increasingly seizures are thought of as a problem of neuronal cortical and subcortical networks with distributed network nodes (1). Khambhati et al. borrowed concepts from physics and control engineering and describe a push–pull model of seizure dynamics using a computational model of “virtual resection” of network nodes. Applied to seizure dynamics synchronization is the push leading to greater seizure spread, while desynchronization functions as a pull containing the seizure.

In addition, Khambhati et al. examine the so far unanswered question of “why do some seizures electrically spread beyond others?” This ultimately comes down to the question “why do some seizures progress from focal seizures with minimal symptoms to generalized tonic-clonic seizures while others in the same patient don't?” We do know that sleep and AED withdrawal has an impact on secondary generalization, but we know little about network dynamics (2).

The authors examined 10 patients with 18 focally contained seizures and 16 seizures with generalized EEG changes on intracranial EEG. The authors speak about focal networks that are seizures contained by few EEG channels and distributed networks, which are seizures with larger spread, but not necessarily clinical secondary generalization. They examine the seizure and pre-seizure state. The pre-seizure epoch is the time immediately before the clinically marked EEG seizure onset, time matched to the duration of the synchronized seizure discharge. To define a measure of synchronization they use a multitaper coherence estimation between electrode pairs, which they define as a “node.” They estimate node strength for each node in the network by applying a time-varying LaPlacian matrix and assigning eigenvalues. By virtually resecting nodes, they measure node centrality, which means how much is each node contributing to either synchronization or desynchronization within the network.

They find that in distributed seizures as compared with focal seizures, high-gamma activity (95–105 Hz) is more synchronized (≥node strength) in the pre-seizure state, and beta (15–25 Hz) and low-gamma activity (30–40 Hz) is more synchronized in the seizure state. Synchronization in the alpha and theta band (5–15 Hz) shows no differences. This means if there is more synchronized high-gamma activity prior to seizures, it is likely that seizures will exhibit greater seizure spread, which may translate to more significant clinical manifestations. As high-gamma activity was the most significant activity predicting seizure spread, the remainder of the analysis is restricted to the high-gamma band.

The authors examine whether node strength dispersion is important for seizure spread. If some nodes are highly connected and others minimally synchronization is lower as compared if all nodes are connected with equal strength. This means less seizure spread the more heterogenicity exists between nodes.

In addition to examining network topology, such as node strength and dispersion, they also examine network geometry by defining a measure of “node centrality” utilizing virtual resection. Greater node centrality of a node means the node is a greater controller of synchronizability. Node strength was not correlated with node centrality, which means that whether a node is a “controller” or “non-controller” is dependent how the node is connected to others, not only what the strength of the connection is.

The authors divide all nodes into desynchronizing nodes, synchronizing nodes, and bulk nodes Bulk nodes are neither synchronizing or desynchronizing and somewhat “neutral.” In focal seizures, desynchronizing and synchronizing nodes exhibit larger control centrality than in distributed networks during the pre-seizure and seizure epoch. Nodes in the seizure onset zone demonstrated no differences in node centrality between focal and distributed seizures, while synchronizing and desynchronizing nodes outside the seizure onset zone clearly exhibited increased control centrality during focal seizure as compared with distributed seizures. This is an important finding. It suggests that the seizure onset zone has generally little control function over seizure evolution and seizure spread is dependent on the surrounding network and its control properties. Which anatomical networks are vital for specific types of seizures, remains to be studied in the future.

This work nicely demonstrated how concepts from physics and control engineering could be applied to medical applications. Push–pull mechanism may control the spread of seizure, eventually controlling the magnitude of clinical symptoms. If we could find a way to control the “push” by a better “pull,” we may not be able to eliminate the seizure discharge, but contain the irritating spread, and therefore clinical symptoms. This is a fundamentally different concept than our concept of resective epilepsy surgery. Alternating the network properties by shifting balances is an alternative strategy to treat epilepsy. Brain stimulation could be one way to do so (3, 4).

The authors demonstrate that more sophisticated virtual models of epileptic networks are essential to test novel treatment modalities (5). Comprehensively testing the efficacy of multiple brain stimulation paradigms is virtually and realistically impossible, as this would require multiple large and costly patient studies. The parameter space for brain stimulation is nearly infinite and highly variable. Frequency, amplitude, duration, location, pulse width, polarity, number, and duration of trains can all be modulated and one little change in stimulation can have opposite effects (6). Virtual models may help to narrow down parameters for effective brain modulation to influence networks. That does not mean, that from time to time we have to examine whether our models truly model reality.

Unfortunately, the number of seizures available and inter-individual differences between intracranial ECoG arrays often limits research, as in this study. Only collaborations between multiple centers and pooled databases can overcome these limitations as already demonstrated by multiple initiatives (7, 8). We owe it to our patients to pool our resources to understand seizures as much as we can.