Research Projects Description

Epilepsy is a serious neurological disease that affects a large population. To understand the mechanisms underlying this disease we apply an interdisciplinary approach that includes clinical studies from patients with epilepsy, experimental models, computer simulations and mathematical modeling.



This approach enables us to generate and test hypotheses efficiently by overcoming the caveats of each approach individually. Our studies include the following topics.

Human Tissue and Clinical Recording Studies

Using human recordings and resected tissue allows for the most direct investigation of epileptic network activity. Using slices of human brain from patients with focal epilepsy, we can directly study the neural activity and discover cellular and network differences in human epileptic brain. (Related work - Marcuccilli et al., 2010)

StudyOfHumanTissue_0.pngCharacterization of human cortical neurons.  (A)  A cortical pyramidal neuron. (B)  A regular spiking neuron following injection of a depolarizing pulse.  (C)  Example of a voltage-dependent burster following an injection with a depolarizing pulse.

Likewise, clinical recording of EEG and cortical activity are an essential part of the surgical evaluation and treatment of patients with epilepsy. These recordings represent the activity of huge networks. Therefore, by applying signal processing techniques to clinical recordings, we have a direct approach to study of macroscopic epileptiform (e.g. van Drongelen et al., 2005; Lee et al., 2007; Rana et al., 2012). 

Paroxysmal Depolarizations in Human Focal Seizures

Paroxysmal depolarizing shifts (PDSs) are a cellular marker of epilepsy in which a neuron is depolarized to such an extent that there is reduced firing due to the neuron saturating. While this phenomenon has been documented in experimental models of epilepsy, it has not been confirmed in humans nor has the relationship between PDSs and pathological network activity been studied. We have been studying the presence and role of PDSs in human seizure activity using in vivo microlectrode arrays and resected brain tissue from the focal area of patients with epilepsy. 

Simultaneous extracellular and intracellular recording of neural activity with a paroxysmal depolarization shift (PDS)


Ictal High Gamma Activity

In collaboration with the team at Columbia University, High gamma activity (80-150 Hz) has been related to areas of brain tissue involved in seizure onset, termed the ictal core. We are interested in how neuronal networks generate this high gamma activity. As can be seen in the figure below, core network activity shows significantly more high gamma power than networks in the un-recruited penumbra, but the mechanisms that cause this discrepancy are largely unknown. Therefore, to determine the role of high gamma activity during seizures, we are investigating network activity across multiple scales. Using resected brain tissue from the focal area of patients with epilepsy, we have found that high gamma activity correlates with pathological cellular activity known as paroxysmal depolarizing shifts (shown in the Figure above). We also employ detailed recordings of human brain activity from both microelectrode arrays (giving us mesoscale resolution) and clinical ECoG recordings (at the macroscale) to study how high gamma activity changes across different sized networks. 

High gamma activity during seizures.
For details see Eissa et al. 2016, eNeuro


Long-Range Connections During Focal Seizures

Pathological neuronal firing during focal seizures have been shown to begin in one general area before propagating to surrounding tissue; however, clinical recordings show low frequency seizure activity at distances far from the initiating source. We investigate how this low frequency activity can be evoked by action potentials at the focus. As part of this effort, we determine the long range network interactions using human microelectrode array and ECoG recordings in conjunction with computational models. 

Spike Triggered Average (STA) showing effects at 1cm from the wavefront of action potentials.



Animal Models of Epilepsy

Acute cortical slices provide an efficient model to study single neuron and local network activity (Related work - Martell et al. 2012; Dwyer et al. 2012). Part of our work has focused on the NMDA receptor that possesses both synaptic and voltage-sensitive properties. Based on the networks’ behavior in the presence of NMDA, two types of cortical neurons were found. One type displayed NMDA-driven oscillations that could not be abolished by TTX. This same neuron type was associated with a region of negative conductance (RNSC) in the I-V curve.



The following movie shows a 3D reconstruction of a neo-cortical pyramidal cell.


Synaptic Mechanisms Underlying Development of Seizure-Like Activity

Multiple network patterns emerge as synapses connect during the development and maturation of dissociated cortical cell cultures. These network patterns range from sparse, randomly-occurring spikes to synchronous, intense network-wide bursts that often resemble epileptiform activity. We are interested in how the underlying synaptic connectivity contributes to the development of seizure-like activity at the network level. The figure below depicts network structure (confocal imaging, top row) and function (recordings with multi-electrode arrays, bottom row) of four developmental stages ranging from 5-20 days in vitro (DIV).    





The Role of Synaptic and Intrinsic Currents in Shaping Network Activity

We are interested in understanding the contributions of intrinsic and synaptic mechanisms towards spontaneous seizure-like activity observed in cortical cell cultures. This network behavior is typically characterized by short-duration bursts (~1-2s), separated by longer interburst intervals (>10s).  We hypothesize that while the intraburst properties and burst propagation is modulated by short-timescale synaptic processes, the interburst intervals are governed by intrinsic membrane properties that have much longer time scales, i.e. persistent-sodium (Nap) currents. To test this hypothesis, we used synaptic receptor antagonists PTX, CNQX and CPP to selectively block GABAA, AMPA and NMDA receptors respectively and riluzole to selectively block slow voltage-sensitive Nap channels. We found that bursting activity could be completely abolished by blocking the intrinsic Nap currents, thereby demonstrating the membrane property’s critical role in burst onset within the network. Contrastingly, by blocking different combinations of synaptic receptors, spectro-temporal burst properties were uniquely associated with synaptic functionality. Specifically, we determined that network-wide bursting requires intact excitatory connectivity.

Suresh et al. 2016, J. Neurophysiol
Effect of synaptic transmission captured during intracellular and MEA recordings
Sample intracellular and MEA recordings observed in five out of eight pharmacological conditions that exhibited bursting activity: A) No drugs added (GABAA+NMDA+AMPA receptors functional) B) PTX added (NMDA+AMPA receptors functional), C) CPP added (GABAA+AMPA receptors functional), D) PTX+CPP added (only AMPA receptors functional), E) PTX+CNQX (only NMDA receptors functional). Remaining conditions i.e., CNQX added (GABAA+NMDA receptors functional), CNQX+CPP added (only GABAA receptors functional) and CNQX+CPP+PTX added (none of the three receptors functional) are omitted because no bursting activity occurred. The inset depicts a detail with paroxysmal depolarization.



Suresh et al. 2016, J. Neurophysiol
Network connectivity analysis - Effect of synaptic transmission on correlations and delays
(A) Representative examples of connectivity graphs characterizing networks with AMPA (top row) and NMDA (bottom row) only connections. Each of the 60 MEA channels is depicted as a vertex in the graph and the delay associated with the maximum cross-correlation was employed to determine each of the edges. It can be seen that networks governed by AMPA are predominantly determined by fast (lags between 2-10ms) correlations, while the networks with NMDA receptors show a dominance of slow (lags between 20-150ms) correlations. (B) Normalized sum of maximal correlations (NSMC) between all pairs of the n active channels; the histogram shows the mean NSMC across the eight conditions. (C) Mean ratio of maximum correlations with fast/slow delays (indicated as black bars); Mean ratio of maximum correlations with slow/fast delays (indicated as grey bars). Error bars indicate SEM.



Pathological Network Architecture

One of our principal interests is to determine the role of network architecture in generating pathological activity. In order to study this role, we use a novel method to manipulate the connectivity in dissociated neuronal cultures and create biomimetic connections at a micrometer and sub-millisecond precision. These connections are made by stimulating the target area of a culture when activity is recorded at a source point, i.e. we introduce network motifs, feedback loops, and feedforward effects to the system. The image below features a 3D model of part of the setup we are building to stimulate the cultures optically.



Our goal is to manipulate network architecture and create / destroy pathological activity to better understand the effects of specific architectures. The figure below illustrates how introducing even a single connection between network locations (between electrodes 26 and 77; red vertical arrows) can dramatically alter the network state in the target area.




Computational and Mathematical Modeling

Given the complexities of network activity, theoretical models are occasionally better options for the study of epileptic networks. These models allow for easy hypothesis testing and can provide more flexibility and control. We use two main types of theoretical modeling: mathematical modeling and computational modeling. 

Mathematical models are abstractions of a real system. These simplifications allow us to create an analytic approach or can help formulate insightful simulations. (e.g. Benayoun et al., 2010; Wallace et al.,2011; van Drongelen, 2013).



Alternatively, computational approaches include more realistic, conductance-based models of nerve cells that are used as the nodes in a network. These models cannot be analyzed mathematically but can be simulated. At the scale we are interested (100,000 cells and up) such simulations require powerful parallel computers. (Related work - van Drongelen et al., 2005, 2006; Visser et al., 2010) The diagram below depicts the components of the scalable computational model we used to study the generation of seizure-like oscillations. The model contains two types of (excitatory) pyramidal cells and four types of (inhibitory) interneurons. (Related Work - van Drongelen et al., 2005, 2006).



This movie shows the result of a simulation of a network of 100,000 neurons (to visualize at this scale, only a subset is depicted). Each neuron consists of multiple compartments with Hodgkin & Huxley type ion channels. After an initial stimulus, waves propagate through the network. We used parallel computing to create this result.