Dynamics underlying epileptic seizures:

(1)

Dynamics underlying epileptic seizures:

insights from a neural mass model

Thesis presented by Xiaoya FAN

with a view to obtaining the PhD Degree in Engineering Sciences and

Technology (“Docteur en Sciences de l’

Ingénieur et Technologie”)

Academic year 2018-2019

Supervisor: Professor Antoine NONCLERCQ

BEAMS (Bio-, Electro- And Mechanical Systems)

Thesis jury:

Olivier DEBEIR (Université libre de Bruxelles, Secretary) Riëm EL TAHRY (Université catholique de Louvain) Nicolas GASPARD (Université libre de Bruxelles)

Michel KINNAERT (Université libre de Bruxelles, Chair)

(2)

I

Dynamics underlying epileptic seizures:

insights from a neural mass model

A dissertation presented by

Xiaoya FAN

This dissertation is submitted for the degree of

Doctor of Engineering Sciences and Technology

OCTOBER 4, 2018

(3)

II

“Logic will get you from A to B. Imagination will take you everywhere.”

(4)

III

Acknowledgements

The pursuit of a PhD is never easy. This thesis is the result of painstaking efforts of, not only mine, but also of many collaborators and remarkable individuals whom I wish to thank. They certainly made this journey more enjoyable and fruitful. Now I finally come to the end of it. At this very moment, I look back to this long journey and cannot help letting out the gratitude from my deepest heart, to all that guided me, supported me, and encouraged me.

First and foremost, I wish to express my sincerest gratitude to my supervisor, Antoine NONCLERCQ, for his patience, motivation, encouragement and ready availability. He has been supportive both academically and emotionally through the entire road. He treated me with all his respect and responsibility. I enjoyed working with him as one of his PhD students and gained a lot from his words and deeds. He parted the clouds above my head when I lost my direction and always encouraged me. Not only is he a responsible mentor, but also a caring friend. He gave me a helping hand when I expressed my concern about the unexpected violence happened to my friends. He cares about my future career in the academy and offered a great recommendation. I feel very lucky to have him as my promoter.

I would like to thank the rest members in my accompanying committee: Michel KINNAERT and Olivier DEBEIR, for their insightful comments, valuable advices and warm encouragement. Many thanks to the collaborators from Erasmus hospital, Nicolas GASPARD, Benjamin LEGROS and Xavier DE TIEGE. As epileptologists, their expertise and knowledge in neuroscience and clinics are indispensable for the accomplishment of this thesis. They provided unique insights, making this thesis complete and profound. I am particularly grateful for the assistance given by Rudy ERCEK, especially in the release of our toolbox. He has been participating and providing his expertise consistently, in every aspect of this thesis. Thank my dear colleague, Federico LUCCHETTI, for his precious advice and help in the carrying out of this project. I would like to offer my special thanks to Gatien HOCEPIED for his previous work on this project, which laid the foundation of this thesis. I thank Francis GRENEZ and Jean SCHOENTGEN, for their professional suggestions on analyzing the route to epilepsy. As a matter of fact, the entire faculty and colleagues of BEAMS department has been supportive and helpful to, not only my research, but also to my life as a foreigner in Belgium. I am grateful for everything they have done to make my stay here much more enjoyable. I shall thank Chinese Scholarship Council and Fonds David et Alice Van Buuren & Fondation Jaumotte-Demoulin for their financial support.

(5)

IV

my lift. They have absolutely made this journey colorful and unforgettable. Thanks to Xavier I appreciated the extraordinary beauty and charm of Belgium and lived my life here to the fullest. My special thank you goes to Jinghong LIU, my dearest friend, roommate during the first two years, from whom I learned to cook and finally prevented me from starving myself. I thank particularly Xing LIU, who is like a loving brother, and literally fed me during the first day when I landed a foreign country for the very first time. I would also like to offer my sincere gratitude to my landlord, also my roommate, Antoinette BROUYAUX. I am grateful that she has been so warm and caring, making the apartment a lovely home. I will never forget my terrific experience of the Christmas Eve with her loving family.

I shall thank my boyfriend, who encouraged me to pursue a PhD, gave me courage to start a long-distance relationship and had faith in our love. Thanks to him I had the chance to know these fabulous people and a whole different culture. I wish to thank my family, my parents and brother, for their unconditional support and tremendous love. They made what I am. I also want to thank myself for being brave, strong, determined and motivated to explore and embrace the unknowns.

(6)

V

Abstract

In this work, we propose an approach that allows to explore the potential pathophysiological mechanisms (at neuronal population level) of ictogenesis by combining clinical intracranial electroencephalographic (iEEG) recordings with a neural mass model. IEEG recordings from temporal lobe epilepsy (TLE) patients around seizure onset were investigated. Physiologically meaningful parameters (average synaptic gains of the excitatory, slow and fast inhibitory population, Ae, B and G) were identified during interictal to ictal transition.

We analyzed the temporal evolution of four ratios, i.e. Ae/G, Ae/B, Ae/(B + G), and B/G. The excitation/inhibition ratio increased around seizure onset and decreased before seizure offset, suggesting the disturbance and restoration of balance between excitation and inhibition around seizure onset and before seizure offset, respectively. Moreover, the slow inhibition may have an earlier effect on the breakdown of excitation/inhibition balance. Results confirm the decrease in excitation/inhibition ratio upon seizure termination in human temporal lobe epilepsy, as revealed by optogenetic approaches both in vivo in animal models and in vitro.

We further explored the distribution of the average synaptic gains in parameter space and their temporal evolution, i.e. the path through the model parameter space, in TLE patients. Results showed that the synaptic gain values located roughly on a plane before seizure onset, dispersed during ictal and returned when the seizure terminated. Cluster analysis was performed on seizure paths and demonstrated consistency in synaptic gain evolution across different seizures from individual patients. Furthermore, two patient groups were identified, each one corresponding to a specific synaptic gain evolution in the parameter space during a seizure. Results were validated by a bootstrapping approach based on comparison with random paths. The differences in the path revealed variations in EEG dynamics for patients despite showing an identical seizure onset pattern. Our approach may have the potential to classify the epileptic patients into subgroups based on different mechanisms revealed by subtle changes in synaptic gains and further enable more robust decisions regarding treatment strategy.

(7)

VI

excitation/(slow + fast inhibition) ratio, allowed the best performance and that the algorithm best suited TLE patients. Leave-one-out cross-validation showed that the algorithm achieved 92.98% sensitivity for TLE patients. The median false positive rate was 0.16 per hour, and median detection delay was 14.5 s. Of interest, the values of the threshold determined by leave-one-out cross-validation for TLE patients were quite constant, suggesting a general excitation/inhibition balance baseline in background iEEG among TLE patients. Such a model-based seizure detection approach is of clinical interest and could also achieve good performance for other types of epilepsy provided that more appropriate model, i.e. better describe epileptic EEG waveforms for other types of epilepsy, is implemented.

(8)

VII

Scientific communications

Publications in international peer-reviewed journals

Fan, X., Gaspard, N., Legros, B., Lucchetti, F., Ercek, R. & Nonclercq, A.*

Dynamics underlying interictal to ictal transition in temporal lobe epilepsy: insights from a neural mass model.

Eur J Neurosci., 47, 258-268 (2018).

Seizure evolution can be characterized as path through synaptic gain space of a neural mass model.

Eur J Neurosci., 48, 3097-3112 (2018)

Publications submitted

Epileptic seizure automated detection based on break of excitation/inhibition balance. Comput Biol Med. (Under revision)

Congress and conference abstract

Poster communications

Fan, X., F., Ercek, Gaspard, X. Tiege, N., Legros, B. & Nonclercq, A.*

Study of dynamics underlying epileptic seizures based on a neural mass model.

12th_{European Congress on Epileptology, September 2016, Prague, the Czech Republic.}

Fan, X., F., Ercek, Gaspard, X. Tiege, N., Legros, B. & Nonclercq, A.*

Study of transition from interictal to ictal: insight from a computational model. 15th_{Belgian Day on Biomedical Engineering. November 2016, Brussels, Belgium.}

Oral communications

Fan, X., F., Ercek, Gaspard & Nonclercq, A.*

Analysis of a neural mass model that simulates epileptiform/background EEG activities.

Joint meeting of the IEEE-EMBS Benelux chapter and the 14th_{Belgian National Day on}

(9)

VIII

List of Abbreviations

AMPA: α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid CNN: Convolutional neural network

DBSCAN: Density-based spatial clustering of applications with noise ECoG: Electrocorticogram

EEG: Electroencephalography

EPSP: Excitatory postsynaptic potential EZ: Epileptogenic zone

fMRI: Functional magnetic resonance imaging FLE: Frontal lobe epilepsy

GABA: Gamma-Aminobutyric acid (γ-Aminobutyric acid) HFO: High frequency oscillation

iEEG: Intracranial electroencephalography IBE: International Bureau for Epilepsy IES: Interictal epileptic spikes

ILAE: International League Against Epilepsy IPSP: Inhibitory postsynaptic potential LFP: Local field potentials

LVFA: Low-voltage fast activity MEG: Magnetoencephalography MTLE: Mesial temporal lobe epilepsies NFM: Neural field model

NMDA: N-Methyl-D-aspartic acid NMM: Neural Mass Model

PCA: Principal Component Analysis PSP: Postsynaptic Potential

PV+_{: Parvalbumin-positive}

SD: Standard deviation

SEEG: Stereoelectroencephalography SOM+_{: Somatostatin-positive}

SPM: Statistical Parametric Mapping

SR: Single-unit recording

(13)

XII

List of figures

Figure 1.1. Framework for the classification of epileptic seizures, proposed by the International

League Against Epilepsy (ILAE) [16] ... 4

Figure 1.2. Regions of the human brain [17] ... 4

Figure 1.3. A typical neuron [17]... 6

Figure 1.4. Electrical synapse (A) versus chemical synapse (B) [26] ... 7

Figure 1.5. Illustration of different types of EEG recordings, adapted from [31] ... 8

Figure 1.6. Simultaneous intracranial and scalp EEG recording of a left temporal lobe seizure [39] ... 9

Figure 1.7. Illustration of two types of spikes. (A) type 1, spike with slow wave; (B) type 2, spike without slow wave ... 11

Figure 1.8. Illustration of high frequency oscillations. FR represents for fast ripples [46]. .... 11

Figure 1.9. Illustration of synchronous generalized 3 Hz spikes-and-waves [60] ... 12

Figure 1.10. Illustration of fast oscillations at seizure onset ... 13

Figure 2.1. (A) Confocal image of the cerebral cortex showing the pyramidal cells and interneurons [159], and structure of the model (+/− represents excitatory/inhibitory); (B) Detailed schematic representation of the model (EPSP/IPSP denotes excitatory/inhibitory postsynaptic potential), adapted from [68]. ... 30

Figure 2.2. (A) Illustration of the signal transmission among neurons. (B) The two blocks representing one neuronal subset. ... 31

Figure 2.3. Each population is modeled by two blocks. (A) Sigmoid function: convert average postsynaptic potential 𝑣 into the average pulse density of potential fired by a population, shown in equation 2.1. (B) The impulse responses of the transfer functions that transfers the average pulse density of afferent action potential into an average excitatory or fast/slow inhibitory postsynaptic membrane potential, given by equation (2.2). ... 32

Figure 2.4. Illustration of different EEG activities reproduced by the model. They were taken from the 1000 simulations labeled by an EEG expert (see section 2.2.2). ... 34

Figure 2.5. Illustration of the diagram for obtaining the activity map ... 35

Figure 2.6. Selective results of activity maps obtained in (B, G) plane, given specific values of Ae (Ae, B and G refer to average excitatory, slow inhibitory and fast inhibitory synaptic gains, respectively) ... 37

Figure 2.7. Selective results of stability maps obtained in (B, G) plane, given specific values of Ae (Ae, B and G refer to average excitatory, slow inhibitory and fast inhibitory synaptic gains, respectively) ... 39

(14)

XIII

Figure 3.1. Illustration of determination of 𝐼𝑟𝑖𝑠𝑒 and 𝐼𝑓𝑎𝑙𝑙, as well as calculation of time delay to seizure onset and seizure offset. (A) iEEG recording; (B) Ratio (solid blue curve) and filtered ratio (dashed dark curve). Ratio denotes, in general either of the four ratios, that is, Ae/B, Ae/G, Ae/(B + G), or B/G. The horizontal solid blue and dashed blue line denote mean + SD and mean + 2SD of the filtered index, respectively. The 𝐼𝑟𝑖𝑠𝑒 and 𝐼𝑓𝑎𝑙𝑙 are represented by the red dot and green circle, respectively. (A) and (B) are aligned with seizure onset and offset, which are indicated by two vertical dashed lines. ... 47 Figure 3.2. Illustration of results. (A) iEEG recording. (B), (C), (D) and (E) Time evolution of

Ae/G, Ae/B, Ae/(B+G) and B/G respectively. The raw ratios are shown by solid blue curves while the filtered ratios are dashed dark curves. The two vertical dashed lines correspond to seizure onset and offset marked by expert, respectively. The horizontal dash-dotted lines represent mean + SD of the filtered indices. The red dot and green circle indicate 𝐼𝑟𝑖𝑠𝑒 and 𝐼𝑓𝑎𝑙𝑙, respectively. (F) An example of real EEG segment during a seizure and the simulated signal. (G) The corresponding normalized power of signals illustrated in (F). ... 49 Figure 3.3. Another illustration of results. (A) iEEG recording. (B, C, D, and E) time evolution

of Ae/G, Ae/B, Ae/(B+G) and B/G, respectively. The raw ratios are shown by solid blue curves, while the filtered ratios are dashed dark curves. The two vertical dashed lines correspond to seizure onset and offset marked by the expert, respectively. The horizontal dash-dotted lines represent mean + SD of the filtered ratios. The red dot and green circle indicate 𝐼𝑟𝑖𝑠𝑒 and 𝐼𝑓𝑎𝑙𝑙, respectively. (F) An example of real EEG segment during a seizure and the simulated signal. (G) The corresponding normalized power of the signals illustrated in (F). ... 50 Figure 3.4. Time delay to seizure onset calculated from four ratios, Ae/G, Ae/B, Ae/(B + G),

and B/G, respectively. Results from different patients are indicated by the shaded boxes. The solid red line corresponds to 30 s before seizure onset, that is, time delay to seizure onset is -30 s. The dashed green line refers to seizure offset, that is, time delay to seizure onset equals to seizure duration and is not constant because different seizures have different durations. ... 51 Figure 3.5. Time delay to seizure offset calculated from four ratios, Ae/G, Ae/B, Ae/(B+G) and

B/G, respectively. Results from different patients are indicated by the shaded boxes. The solid red line corresponds to 30 s after seizure offset, i.e. time delay to seizure offset is 30 s. The dashed green line refers to seizure onset, i.e. time delay to seizure offset equals minus seizure duration and it is not constant because different seizures have different durations. ... 51 Figure 4.1. Illustration of periods of interest ... 62 Figure 4.2. Distribution of identified synaptic gains in the parameter space. It shows the

synaptic gains collected from the recordings of all patients. Each subfigure corresponds to one period of interest. The planes are identical and obtained by linear regression on data from IntIctal period. ... 66 Figure 4.3. The boxplot of contribution ratios of the first two principal components during the

(15)

XIV

Figure 4.4. Clusters showing the highest average silhouette coefficient. (A), (B), (C) and (D) correspond to different number of clusters, i.e. 2, 3, 4, 5, respectively. For each of the four subfigures, the cluster structure is shown on the left and the silhouette coefficients on the right. Each circle represents one seizure. Seizures recorded from different patients are separated by dashed lines and indicated by patient ID in subfigure (D). In the silhouette coefficients, values for all the seizures of the corresponding cluster structure are shown with horizontal bars (seizures grouped in the same cluster are shown together and clusters are separated). ... 68 Figure 4.5. Clusters showing the highest average silhouette coefficient (analysis over 26

seizures, after removing 4 outliers). (A) Cluster structure. Each circle represents one seizure. Seizures recorded from different patients are separated by dashed lines and indicated by patient ID. (B) Silhouette coefficients of all seizures of the corresponding cluster structure (seizures grouped in the same cluster are shown together and the two clusters are separated). ... 68 Figure 4.6. Illustration, for the patient B, of paths through synaptic gain parameter space in

cluster 1. The three sub-figures of each row represent one seizure. Each corresponds to a 2-D projection. Seizures start in red and end in blue. ... 69 Figure 4.7. Illustration, for the patient G, of paths through synaptic gain parameter space in

cluster 2. The three sub-figures of each row represent one seizure. Each corresponds to a 2-D projection. Seizures start in red and end in blue. ... 70 Figure 4.8. Boxplots of bootstrap values (%) for seizure path pairs from five groups, i.e. intra-patient path pairs, inter-intra-patient path pairs, cluster 1 path pairs, cluster 2 path pairs and inter-cluster path pairs (obtained by the “boxplot” function in MATLAB). Bootstrap value < 5% indicates significant consistency between two paths. The whiskers extend to the most extreme data points not considered anomalies (approximately +/–2.7σ and 99.3 percent coverage if the data are normally distributed) and the crosses in red represent anomalies. ... 71 Figure 5.1. Seizure detection algorithm overview ... 83 Figure 5.2. ROC curves obtained for the three patient groups, i.e. TLE, FLE and others. TLE:

temporal lobe epilepsy; FLE: frontal lobe epilepsy ... 88 Figure 5.3. The values of the threshold determined by the leave-one-out cross-validation

procedure, for each patient group, when Ae/(B+G) was used as the cue ... 89 Figure 5.4. The histogram of Ae/(B+G) during interictal periods for temporal lobe epilepsy

(16)

XV

List of tables

Table 1.1. Different model strategies. NMM: Neural Mass Model; NFM: Neural Field Model; EEG: electroencephalogram; LFP: Local Field Potential; MEG: Magnetoencephalography; SR: Single-cell recording; fMRI: functional magnetic resonance imaging ... 16 Table 1.2 Periods of interest during the occurrence of a seizure, typical EEG activity and key

findings. GABA: Gamma-Aminobutyric acid (γ-Aminobutyric acid); LVFA: low voltage fast activity. ... 22 Table 2.1. Model parameters, their physiological meanings and standard values. The standard

values were established in previous studies [67, 68, 76]. ... 33 Table 3.1. Information about patients ... 46 Table 3.2. Variations (%) of 𝑇𝐼𝑟𝑖𝑠𝑒and 𝑇𝐼𝑓𝑎𝑙𝑙by varying ad hoc parameter values. The

sliding window size was set to 1 s, 2 s, 5 s and 10 s. The sliding window step was set to 0.1 s, 0.5 s, 1 s, 1.5 s and 2 s. The number of candidates varied from 50 to 200 with a step of 25. The moving average length varied from 15 s to 50 s with a step of 5 s. ... 53 Table 4.1. Basic information about the patients and cluster results. N/A, not available; NL,

neuronal loss; NI, normal MRI; PNH, periventricular nodular heterotopia; HD, hippocampal dysplasia; MTS, mesial temporal sclerosis; LVFA, low-voltage fast activity. ... 61 Table 4.2. Results of the bootstrapping test to validate cluster results. The bootstrap value was

computed as the percentage of random path pairs that are equally close or closer than real seizure evolution path pairs. Bootstrap value < 5% indicates significantly consistent path pairs. Wilcoxon rank sum tests were performed to compare the statistical difference in bootstrap values between groups and significance was indicated by * (p<0.05). N represents no significant difference between groups. The values in the parenthesis in the last and second last columns were computed after ruling out the four outliers. ... 72 Table 5.1. iEEG data and leave-one-out cross-validation results for each patient. The total

recording time, total number of seizures, total number of detected seizures, overall sensitivity, median false positive rate (FPR), median mean false detection duration (MFDD), median average detection delay and mean threshold for each patient group were provided separately (in bold). Patients 12, 13 and 14 were diagnosed with FTLE, PTOLE and OLE, respectively. TLE: Temporal Lobe Epilepsy; FLE: Frontal Lobe Epilepsy; FTLE: Fronto-Temporal Lobe Epilepsy; PTOLE: Parieto-Temporal-Occipital Lobe Epilepsy; OLE: Parieto-Temporal-Occipital Lobe Epilepsy; N/A: not applicable. The data in parentheses represents results after excluding patient 18 (section 5.3.3). ... 82 Table 5.2. Area under the ROC curve (AUC): comparison between fixed threshold and patient-specific threshold, and among the three cues for different patient groups separately and altogether. ... 87 Table 5.3 Comparison of the performance of our algorithm for TLE group with previously

(17)

XVI

Chapter 1. Background

(18)

2 | P a g e

1.1. General Introduction and Motivations

Epilepsy is one of the most common neurological disorders in the world population and affects over 65 million people worldwide (between 0.7% and 1% of the population) [1]. It is a chronic disorder characterized by recurrent and unpredictable interruptions of normal brain function, called epileptic seizures, which can cause notable deterioration of the patients’ quality of life. The mechanisms behind seizure genesis (i.e. ictogenesis) are complex and currently largely unknown.

Electroencephalography (EEG) records the electrical activity from the cerebral cortex using electrodes placed either on the scalp, i.e. scalp EEG [2], or directly on the exposed surface of the brain, i.e. intracranial EEG [3]. It has been widely used in epilepsy research [4-7]. The study of neurophysiological mechanisms underlying EEG activities could facilitate our understanding of the interictal to ictal transition, and yet is challenging. Computational models are efficient tools for studying underlying mechanisms, as they can relate EEG activities recorded during a seizure to the pathways and mechanisms of seizure evolution [8-10] through physiological relevance of the model parameters.

Physiological-based models are of great help as they link the model parameters to neurophysiology. Temporal evolution of important parameters in such model can be estimated from clinical data. Their analysis provides an opportunity of revealing patient-specific mechanisms underlying ictogenesis1. This is of significant interest because patients with epilepsies which seem identical might have different underlying dynamic mechanisms [11] and may require different treatment.

Advances about the dynamics underlying interictal to ictal transition could guide the development of an algorithm for detecting epileptic seizures at an early stage (i.e. close to the onset) [12]. Such an algorithm could be implemented in an automatic warning system that alerts the patient of the occurrence of a seizure. Alternatively, better understanding of ictogenesis would give valuable help to develop a responsive therapy (i.e. based on a closed-loop device). Thus, a treatment, such as deep brain stimulation or optogenetic control, can be delivered on demand to stop the seizure or even to prevent it. Clinicians can also benefit from

(19)

3 | P a g e

it because such a system could provide a more efficient and objective manner for reviewing EEG data and reduce their workload.

1.2. Epilepsy and Epileptic Seizures: Definition and Classification

The definition of the term “Epilepsy” and “Epileptic seizures” was proposed by the International League Against Epilepsy (ILAE) and the International Bureau for Epilepsy (IBE) [13, 14] for diagnosis and communication among medical professionals and also others. Epileptic seizures is defined as “a transient occurrence of signs and/or symptoms due to abnormal excessive or synchronous neuronal activity in the brain” [13]. Epilepsy, on the other hand, is a disease of the brain “characterized by an enduring predisposition to generate epileptic seizures and by the neurobiologic, cognitive, psychological, and social consequences of this condition” [13, 14]. Practically, the diagnosis of epilepsy is made by any of the following conditions: (1) At least two unprovoked (or reflex/photosensitive, i.e. provoked by flashing lights) seizures occurring more than 24 hours apart; (2) one unprovoked (or reflex) seizure with a probability of having more seizures similar to the general recurrence risk after two unprovoked (or reflex) seizures (at least 60%), occurring over the next 10 years [14]; (3) diagnosis of an epilepsy syndrome, such as benign epilepsy with central-temporal spikes [14]. In brief, epilepsy refers to the disease, whereas an epileptic seizure indicates an event, which is caused by transient disturbance in the electrical activity of the brain.

The classification of epileptic seizures proposed by the ILAE has been updated based on the scientific advances and reflects our gain in understanding of epilepsy and underlying mechanisms. It plays an important role, not only in the diagnosis of patients, but also in the development of anti-epileptic therapies, basic epilepsy research, and communication around the world. Notably, it often guides the selection of anti-epileptic therapies.

The ILAE 2017 classification of seizure types is illustrated in Figure 1.1. The classification of epileptic seizures is based on three main features: 1) where do seizures begin in the brain; 2) level of awareness during a seizure; 3) other features related to seizures, for instance motor or non-motor.

(20)

4 | P a g e

limited to one hemisphere” [15]. Our brain is divided into two hemispheres, each comprised of four sections (Figure 1.2), i.e. the temporal lobe, frontal lobe, parietal lobe and occipital lobe. Focal seizures can originate in a whole hemisphere or part of a lobe and symptoms vary according to where the seizure occurs. They can be further classified according to the discrete area of the brain that is involved in focal seizures. The most common type of focal epilepsy is temporal lobe epilepsy (TLE), which corresponds to 60% of people living with epilepsy.

Figure 1.1. Framework for the classification of epileptic seizures, proposed by the International League Against Epilepsy (ILAE) [16]

Figure 1.2. Regions of the human brain [17]

(21)

5 | P a g e

called focal to bilateral seizures, whereas the classification is based on unilateral onset. Whether epilepsy is the result of focal pathology or not can have implications for surgical options [18]. In some cases, distinction between a focal and generalized onset is not obvious due to inadequate information from current examinations (termed “Unknown” in Figure 1.1).

Focal seizures can be classified as focal aware or focal impaired awareness based on the level of awareness during a seizure. A seizure called aware seizure (or, previously, “simple partial”) indicates that the awareness of the patient remains intact during the seizure, even if the patient is not able to talk or respond. On the other hand, if the awareness is impaired or affected at any time during a seizure, even if the patient has a vague idea of what happened, the seizure is called focal impaired awareness seizure (called previously “complex partial seizure”) [19]. Generalized seizures are all foreseen to affect the patient’s consciousness or awareness. Seizures can be either motor or non-motor, depending on whether visible physical movement is involved during the event.

1.3. EEG in Epilepsy

1.3.1. Basic Physiology

Neurons

(22)

6 | P a g e

Figure 1.3. A typical neuron [17]

Synapses

The neurons communicate with each other via synapses. Suppose one neuron sends a signal to another neuron through a synapse. The former is often called the presynaptic cell and the latter postsynaptic cell. Two types of synapses have been identified in the nervous system, the chemical synapses and the electrical synapses.

(23)

7 | P a g e

potential, EPSP) through activation of specific receptors, primarily N-Methyl-D-aspartic acid (NMDA), kainite, and α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors. The inhibitory neurotransmitters, primarily γ-Aminobutyric acid (GABA) [22], inhibit the postsynaptic neuron (tend to inhibit the generation of an action potential in a postsynaptic neuron) by exerting a hyperpolarizing effect of the postsynaptic membrane potential (inhibitory postsynaptic potential, IPSP) through activation of corresponding receptors, primarily subtype GABAA (ligand-gated with fast and brief response) and subtype

GABAB (G protein-coupled receptors with slow and long-lasting response) [23, 24]. Multiple

neurotransmitters can produce excitatory or inhibitory responses on individual postsynaptic cells. The speed of responses also differs, allowing control of electrical signaling over different time scales.

Electrical synapses are far less common than chemical synapses. In the case of electrical synapses (Figure 1.4 A), the presynaptic and postsynaptic neuron membranes are connected by special channels called gap junctions [25], through which the ions flow passively, promoting or inhibiting the generation of postsynaptic action potentials. Electrical synapses permit faster, bidirectional and more reliable responses. The direct electrical coupling between neurons causes a signal with the same sign as in the presynaptic neuron and usually smaller in amplitude. The electrical transmission is less modifiable because no neurotransmitter is involved.

(24)

8 | P a g e

1.3.2. EEG: Scalp versus Intracranial

The discovery of electrical activity of the brain (the EEG) can date back to 1875 when Richard Caton reported the electrical phenomena of the exposed cerebral hemispheres of rabbits and monkeys [27]. The human EEG was first recorded in 1929, by the German psychiatrist, Hans Berger [28]. It measures the electrical activity in the brain and represents the collective behavior of cortical neurons. It reflects summation of excitatory and inhibitory postsynaptic potentials in uniformly oriented apical dendrites of groups of pyramidal neurons, in the superficial dendritic layers [29, 30]. The potentials can be recorded by electrodes at a short distance from the sources, i.e. neurons, that is, the local EEG or local field potentials (LFPs), or from the surface of the brain cortex (use strip or grid electrodes), that is, the electrocorticogram (ECoG or intracranial EEG), or from the scalp, that is the scalp EEG. Figure 1.5 illustrates various types of electrodes for EEG recording.

Figure 1.5. Illustration of different types of EEG recordings, adapted from [31]

The potential application of EEG in epilepsy rapidly became recognized, when Gibbs and colleagues demonstrated the 3 Hz spike wave discharges in what was then termed petit mal (absence [32] now) epilepsy [29]. Currently, EEG remains a “golden standard” and themost prominent as a tool for the investigation of epilepsy. It plays an essential role in the diagnosis, classification and management of patients with epilepsy. It helps to determine the type of seizure and epilepsy syndrome, and therefore to select anti-epileptic medication and to predict surgical outcome.

(25)

9 | P a g e

Scalp EEG is non-invasive and relatively inexpensive, and therefore is the most widely used in diagnosis and management of epilepsy. IEEG, on the other hand, is an invasive procedure and implantation of electrode arrays on the cortical surface or within the brain is needed. However, scalp EEG suffers from relatively low signal-noise-ratio since the electric potentials, especially at high frequencies, attenuate heavily due to the presence of the dura matter, cerebrospinal fluid, skull and the scalp. They may be overshadowed by artifacts, mainly from electro-oculographic or electromyographic activity from cranial muscles [33]. The frequency bands of typical scalp EEG limit in the range from 0.1 Hz to 30 Hz. Low-voltage fast activity (LVFA), which has been recognized as the most common seizure onset pattern in focal epilepsy [34], cannot be seen easily in scalp EEG. For the same reason, the scalp EEG has relatively poor spatial sensitivity. It is difficult to locate the source of the activities recorded from the scalp, since the cerebrospinal fluid, skull and scalp smear the electrical potentials and the recorded activities reflect the summation of influences from many sources of electric fields [35]. IEEG can offer a higher spatial resolution [36], down to < 5mm2 [37], and capture more directly the neuronal activities. The frequency of iEEG recordings can go beyond 9 kHz [38], depending on the acquisition system. An illustration of simultaneous recording of intracranial and scalp EEG is shown in Figure 1.6.

(26)

10 | P a g e

Simultaneous scalp EEG and iEEG show that the majority of epileptiform discharges recorded directly at the cortical surface cannot be captured by electrodes on the scalp or at least do not show their epileptic nature with sufficient clarity [39]. IEEG is considered as the golden standard for evaluating neuronal activities in patients with epilepsy, since it allows better spatial resolution, wider frequency ranges and higher signal-noise-ratio. It is widely used for presurgical planning [40] to guide the surgical resection by providing precise information for localizing the epileptogenic zone (EZ) and determining its precise boundary [41, 42]. The EZ refers to the part of cortical that is responsible for initiating seizures and necessary to be removed for complete abolition of seizures [42, 43]. Besides its widespread use in the clinical field, the EEG is a tool of major importance in the study of neurophysiological mechanisms underlying the neural activity and in the understanding of the transition from the interictal state to a seizure.

1.3.3. Epileptiform Activities

Epileptiform activities refer to the brain activities recorded during seizures, as well as the abnormal transient events between seizures.

Interictal Epileptiform Discharges

The interictal epileptiform discharges include interictal epileptic spikes (IES), sharp waves and interictal high frequency oscillations, i.e. ripples (120-250 Hz) and fast ripples (250-600 Hz) [44].

(27)

11 | P a g e

Figure 1.7. Illustration of two types of spikes. (A) type 1, spike with slow wave; (B) type 2, spike without slow wave

Sharp waves are transient activities that are clearly distinguished from background activities with pointed peak, with a duration of 70-200 ms.The distinction of durations between spikes and sharp waves is arbitrary and has no clinical purpose [45].

Interictal high frequency oscillations (HFOs, Figure 1.8), i.e. ripples (120-250 Hz) and fast ripples (250-600 Hz), are distinct burst-like events (a few tens of milliseconds) composed of more than 4 oscillations observed in both experimental in vivo models of epilepsy [52] and in human epileptic brain [38, 53, 54]. It has been shown that the HFOs occur in a restricted area [55] and have a higher occurrence rate within the seizure onset zone [56], suggesting that HFOs are good markers of the EZ. This was soon confirmed by subsequent studies demonstrating that the surgical removal of brain tissue-generating HFOs favorably associated with chance of seizure freedom [57, 58].

(28)

12 | P a g e

Ictal Discharges

The ictal discharges mainly include spike-wave discharges in generalized epilepsy, and fast oscillations in beta and low gamma range in focal epilepsy.

The spike-wave, or spike-and-wave (Figure 1.9), is a regular, symmetrical and generalized EEG pattern observed particularly during absence seizures. Two patterns of spike-and-wave were identified in absence seizures according to the frequency [59]: Type 1, at 2.5 Hz or greater, typically 3-4.5 Hz, lasting ≥ 3s; Type 2, at lower frequency (1.5-2.5 Hz).

Figure 1.9. Illustration of synchronous generalized 3 Hz spikes-and-waves [60]

(29)

13 | P a g e

seizure activity. Since they represent the transition from the interictal state to the seizure itself, better understanding of their generators could offer great opportunities in identifying the fundamental causes involved in the occurrence of a seizure. In mesial temporal lobe epilepsy (MTLE), rapid discharges with beta and low gamma range (15-40 Hz) at seizure onset can be observed in the hippocampus, amygdala and entorhinal cortex [61]. The fast oscillations in human neocortex epilepsy, on the other hand, are associated with higher frequency (70-120 Hz) [62].

Figure 1.10. Illustration of fast oscillations at seizure onset

1.4. EEG Modeling

The past a few decades have witnessed the blossom of computational models in the field of epilepsy [46] as they have their unique abilities to explain the mechanisms underlying the generation of EEG activities by linking the key variables to physiological behavior [63]. Their role as a bridge between observation and pathophysiological interpretation has been well recognized [46, 64]. Numerous models have been proposed to investigate the complex pathophysiological drivers leading to ictogenesis and/or epileptogesis2 _{[46]. Another}

advantage of computational models lies in their capacity to propose hypotheses that can be tested experimentally.

One big challenge of modeling approach is to mimic the behavior of the system under study with appropriate simplification [63]. For the generation of electrical activities in the brain, the level of modeling highly depends on the type of behavior to be reproduced by the model and

(30)

14 | P a g e

the observational data it relates to, from single unit recordings, EEG, LFPs to magnetoencephalographic (MEG) signals and functional magnetic resonance imaging (fMRI). In the field of epilepsy, models have been established at nearly all levels, from sub-cellular (neurotransmitter receptors, membrane ion channels), cellular (neurons), to large-scale systems involving cortical and/or subcortical regions [46]. At the microscopic level, structure and function of principal neurons and interneurons are accurately modeled, as well as their interconnection, by synapses and/or gap junctions, resulting in a detailed network model [46, 64]. In a detailed model, the structure and function of a single neuron are represented by compartments and different types of neurons are represented based on their unique morphology (soma, axon and dendrites) and electrical properties (voltage-dependent ion channels, leak conductance, membrane capacitance, axial resistivity, etc.). Nowadays, a considerable number of neuron models is available [65], of which most are built on the basis of the Hodgkin–Huxley formalism [66]. One drawback of these detailed models is that the number of variables increases rapidly with the number of neurons included, along with it tremendous demand for computing capabilities.

(31)

15 | P a g e

the most common pattern observed at the onset of focal seizures, and became quickly popular in epilepsy research [72, 73].

Another approach used in epilepsy research is neural field models (NFMs), which can be seen as an extension of NMMs [74]. These models allow for the study of spatiotemporal patterns of brain activity either at the level of one or multiple region(s) of the cortex or larger scale systems like the entire neocortex [46]. Therefore, they have the unique ability to represent the features of epileptogenic networks that often involve distant and extended brain areas [7] at the cost of a larger number of variables and higher computational load.

The study of dynamics underlying epilepsy using computational models can be achieved in a purely theoretical way using bifurcation theory, which is a powerful tool for systematically studying the relationship between the output of the model and physiological based parameters [11, 72, 73, 75]. It can also be achieved by fitting the model with recorded data [76-78], exploring this way the path through parameter space [79]. Both approaches have the potential to gain a better understanding of the physiology occurring in conjunction with seizure onset and evolution. For instance, Kramer et al. explored the pathways to seizure regions identified with bifurcation analysis on a mesoscale model to reveal potential mechanisms in the causation of seizures [80, 81]. Alternatively, Dadok et al. proposed a Bayesian framework to produce a probabilistic pathway of the temporal evolution of physiological state in the cortex over the course of individual seizures and allow the comparison among different hypotheses [82].

(32)

16 | P a g e

consistency in parameter evolution across different seizures from a given patient [79, 83]. The aforementioned models are physiologically inspired models established based on neurophysiology findings. Therefore, they are easier to interpret and confront with experimental data due to the direct link between the model variables and experimental measurements or perturbations induced by electrical stimulations or drugs in the considered experimental preparation. An alternative is mathematical models, which are phenomenological models that intend to capture certain dynamical properties of neural systems without attempting to mimic the underlying neurophysiological mechanisms. These models offer a number of advantages for formal study, i.e. all possible states as well as transitions (bifurcations) can be determined and theoretically investigated. The limitation is that interpretation is less straightforward as the links between computational and experimental variables are relatively vague and require some extrapolation.

Modeling approaches at different levels provide distinct insights at a given spatial scale or level of neurophysiological detail. Each modeling approach has its unique advantages and limitations (Table 1.1). Taking together findings from these different types of models, a coherent, global perspective of the mechanisms underlying epilepsy and seizures can be obtained.

Table 1.1. Different model strategies. NMM: Neural Mass Model; NFM: Neural Field Model; EEG: electroencephalogram; LFP: Local Field Potential; MEG:Magnetoencephalography;

SR: Single-cell recording; fMRI: functionalmagnetic resonance imaging

Model Data Advantages Limitations Selected studies

NMM (i)EEG, MEG

Small number of variables and low computational load

No spatial features are included

Lopes da silva et al. [71]; Jansen and Rit [67]; Wendling et al. [68]; NFM (i)EEG, LFP, MEG Allow representation of spatiotemporal dynamics of the brain Large number of variables; high computational load Cosandier-Rimele, D. et al. [7]; Moran et al. [84]; Freestone et al [85] ; Detailed models SR, LFP Allow investigating

mechanisms at both cellular and network level

Computational intensive Traub et al. [86]; Demont-Guignard et al. [87]; EI Houssaini et al. [88]; Formal models SR, LFP, (i)EEG, MEG, fMRI

Allow formal study of the model and allow simulation of large scale brain networks

(33)

17 | P a g e

1.5. Excitation and Inhibition in Epilepsy

1.5.1. Excitation and Inhibition Balance

Excitation and inhibition are inseparable events [93]. The interactions between inhibitory interneurons and excitatory principal cells are reciprocal and ubiquitous: the principal cells excite interneurons and are inhibited by them. The glutamatergic excitatory synapses propagate neural firing and the GABAergic inhibitory synapses are responsible for the spatial and temporal pattern of the firing [94]. Together, they regulate the neuronal output and govern cortical/subcortical activity. Their co-occurrence exists from sensory-evoked activity [95] to spontaneous cortical/subcortical activity or oscillations [96, 97].

Previous studies have shown that increases in excitation are invariably accompanied by increases in inhibition [98]. The balance between the two opposing forces has been found experimentally both in vivo [98] and in vitro [99]. It is now recognized as the functional cornerstone in physiological activities in the brain. This balance is maintained dynamically in a normally function brain at multiple scales, from individual neurons [100], ensembles [101] to networks [102].

Epileptic seizures have been historically considered as hyperexcitable phenomena, where excitability of the network is unregulated, possibly due to deficits of GABAergic inhibitory function [103-105]. Indeed, impairment in GABAergic functions can lead to a hyperexcitable epileptic state, and enhancement of GABAergic functions is the underlying mechanisms of many anti-epileptic drugs.

1.5.2. Excitation and Inhibition in IES

(34)

18 | P a g e

inhibition (GABAergic postsynaptic amplitude amplitude), led to the generation of IES [68]. The morphological changes of these spikes during the process of epileptogenesis in a mouse model in vivo were reproduced by decreasing GABAergic inhibition in an NMM [107]. Detailed network models have also been developed to explore more fundamental mechanisms underlying the generation of IES [87]. The amplitude of IES was linked to the AMPAergic synaptic excitatory current while the negative component i.e. the slow wave following fast spike, was attributed to increasing AMPA/decreasing GABA conductance, i.e. increased excitation/inhibition ratio. Taken together, the generation of sharp initial component (spike) of the IES involves increased excitation and the following slow wave is a result of feedback inhibition from GABAergic interneurons on strongly excited principal cells [70].

1.5.3. Excitation and Inhibition in HFOs

HFOs (see section 1.3.3 and Figure 1.8) has been recognized as a biomarker of the EZ. Understanding the mechanisms underlying the generation of HFOs could provide valuable insights into the epileptogenesis because of the close association of HFOs with the epileptogenic tissue [108]. In vitro studies have showed that HFOs arise when neuronal networks are hyperexcitable, which may be caused by several distinct cellular mechanisms [109]. Karlocai and colleagues found increased cellular excitability and enhanced excitatory transmission, along with compromised inhibitory transmission in various in vitro models of pharmacologically induced epilepsy [110]. They found that the firing rate of dendritic interneurons increased, while perisomatic inhibition collapsed due to depolarization block of parvalbumin-positive (PV+) neuron (GABAergic) [110]. Computational studies, combining computational model with experimental data both in vivo and in vitro, showed that fast ripples and IESs share certain common mechanisms (i.e. altered synaptic transmission, particularly increase of excitatory conductance associated with both AMPA and NMDA currents, as well as decrease of inhibitory conductance associated with GABAergic currents) but also show critical differences in terms of the number of pyramidal cells involved and level of synchronization, etc. [111].

1.5.4. Excitation and Inhibitin in Ictogenesis

(35)

19 | P a g e

considered as the first unequivocal iEEG change visually distinguishable from background activity that followed by clear seizure discharges. The most common EEG pattern observed at the onset of focal epilepsy is LVFA (see section 1.3.3 and Figure 1.10). Such fast discharges were not given much attention before because they were difficult to record with conventional EEG systems due to technical limitations [112]. It lasts only a few seconds and transients to high-amplitude ictal discharges afterwards, which lasts from tens of seconds to a few minutes. In this section, we first review the classical view of seizures and then recent advances in ictogenesis, with a particular focus on the brief onset pattern observed at the first a few seconds. Seizures were hypothesized to be hypersynchronous events featured by enhanced excitation and decreased inhibition [113] based on the observation of high-amplitude ictal discharges both in animal models of epilepsy [114] and patients with focal epilepsy [104]. It was proposed that the interictal to ictal transition involved weakening of the inhibitory force leading to excessive excitation [115]. The excessive excitation transforms the transient interictal event, i.e. IES, into a prolonged, high-amplitude ictal discharge.

The classical view that seizures are initiated by increased excitation has been challenged. It has been argued that the presence of synchronous activity during a seizure involves the maintenance rather than the initiation of a seizure and the underlying mechanisms are not necessarily the same [116]. Previous studies have reported the reduction of neuronal firing at seizure onset, possibly due to the inhibition of principal cells [117]. Single-unit recordings performed in the hippocampus of epileptic animal models revealed a decreased firing of glutamatergic principal neurons along with the enhanced inhibition [118]. Results suggested that the initial several seconds of seizure might correlate with an arrest of principal cell firing along with increased inhibitory interneuron firing associated with a powerful GABAergic signaling. In addition, in vitro 4-aminopyridine model study showed that pharmacologic manipulations of antagonize GABAA receptors or decreased presynaptic GABA release abolish

this initial activity and prevent ictal discharges [119]. Similar findings have also been obtained in other models of focal seizures induced in brain slices, for instance by electrical stimulation [120].

(36)

20 | P a g e

of certain neuronal populations via light-activated channels, providing an unprecedented opportunity to investigate the contribution of certain types of neurons in the occurrence of a seizure and/or to seek for anti-epileptic therapy [123, 124]. This control can either depolarize or hyperpolarize, and respectively generate or inhibit action potentials in selective populations of neurons [125]. Shiri and colleagues successfully induced ictal discharges with LVFA onset pattern by optogenetic activation of PV+ fast-spiking inhibitory interneurons in the entorhinal cortex in the in vitro 4-aminopyridine model [126]. Their results revealed the promoting role of fast-spiking GABAergic mediated interneurons in seizure genesis. Besides PV+ fast-spiking somatic interneurons, activation of somatostatin-positive (SOM+_{) regular-spiking interneurons}

targeting dendritic domains (also inhibitory), can also trigger ictal-like activity with LVFA onset in the same model [127]. Therefore, both types of GABAergic inhibitory interneurons contribute to LVFA onset pattern. Evidence favors the link between activation of GABAergic interneurons and epileptiform event, and the blockade of GABAA receptors would cut this link

[128]. The results described above are obtained in in vitro 4-aminopyridine model, which can generate spontaneous seizures. This model represents a hyperexcitable cortical network, which might be necessary for seizure initiation. The transition to ictal events initiated by enhanced inhibition was attributed to post-inhibitory rebound3 excitation [127, 128]. Post-inhibitory rebound [129] is an intrinsic cellular property of thalamic and cortical neurons, whose mechanisms remain poorly understood [130]. It may result in spike-bursts directly after an inhibitory synaptic input, and thus may contribute to the maintenance of oscillations [130, 131]. The authors reached the conclusion that synchronous activation of GABAergic interneurons initiated ictal event onset in hyperexcitable cortical networks [128]. Interestingly, it was found that optogenetic activation of glutamatergic principal cells (excitatory) also favored the generation of seizure-like activity, not with LVFA onset pattern, but with hypersynchronous onset pattern, which consists of high-amplitude spikes that occur at a frequency of <2 Hz [127]. To conclude, ictal-like activities with LVFA onset pattern involve increased inhibition (somatic and/or dendritic) whereas ictal-like activities with hypersynchronous onset pattern involves

(37)

21 | P a g e

increased excitation, and possibly a progressive weakening of inhibition [108]. These findings revealed distinct mechanisms underlying different onset patterns and could provide insight for better therapeutic targets in the treatment of epilepsy depending on seizure onset patterns.

Evidence has also been provided by computational model-based approach. Wendling and colleagues built a model that successfully reproduced LVFA [68]. Their results provided evidence for the link between the generation of LVFA and impaired GABAergic slow inhibition [68]. More precisely, they found that slow inhibition first increased from interictal to preonset and then decreased abruptly from preonset to onset. On the other hand, fast synaptic inhibition mediated by GABA was shown to promote synchronized gamma oscillations in hippocampal interneuron networks in a mouse model in vitro [132]. Indeed, the population of slow inhibitory interneurons alone was not sufficient to generate LVFA in the model, a fast feedback inhibition loop was necessary [68]. This was further investigated in detail in human TLE by combining this model with clinical data [76]. The evolution of key model parameters (involving excitation, slow and fast inhibition) with respect to interictal to ictal transition was analyzed. Sustained increase in excitation was found during the transition from interictal to ictal activity. This is contradictory with previous in vivo studies where arrest of neuronal firing was demonstrated at seizure onset [117], possibly because they used a rather crude temporal evolution [76]. The variation in inhibition was not as simple. Slow dendritic inhibition mediated by GABAergic interneurons first increased from interictal to preonset period, followed by an abrupt decrease at seizure onset. The initial increase was interpreted as a compensatory mechanism to increased excitation [76]. The fast somatic inhibition, on the other hand, stayed constant or increased during this transition. Increased fast somatic inhibition and decreased slow dendritic inhibition were also observed in an experimental model of TLE [133].

(38)

22 | P a g e

receptor signaling in seizure initiation are unexpected and paradoxical to the common belief that seizures are initiated because of the enhancement of glutamatergic excitation [127, 134]. However, a causal relation between seizure initiation and enhanced interneuron activity has not been identified in vivo [116].

The key findings are summarized in Table 1.2.

Table 1.2 Periods of interest during the occurrence of a seizure, typical EEG activity and key findings. GABA:Gamma-Aminobutyric acid (γ-Aminobutyric acid); LVFA: low voltage fast

activity. Periods of

interest

Typical EEG

activity Key findings

Interictal

Background activity with or without spikes and spike-waves

- Synchronized firing of a hyperexcitable subset [51]; - Decreased GABAergic inhibition [106];

- Enhanced excitation and/or decreased inhibition [68]; - Amplitude of spikes is linked to excitation whereas the slow wave is related to increased excitation/decreased inhibition [87]; Onset LVFA preceded by high-amplitude spikes or not

- Brief decrease/increase in firing of excitatory principal cells/inhibitory interneurons in vivo in human [117] and in epileptic pilocarpine-treated rate model [118];

- Decreased inhibition manipulated pharmacologically [119] or caused by neurostimulation [120] abolish initial activity and prevent ictal discharges;

- Optogenetic activation of fast or/and slow inhibitory interneurons induce ictal activity with LVFA onset [126-128, 135] whereas optogenetic activation of excitatory cells induces ictal activity with hypersynchronous onset [127]; - Fast inhibitory interneurons play essential role in the generation of fast discharges at onset [68, 132, 135]; - Enhanced excitation, constant/enhanced fast inhibition and decreased slow inhibition at seizure onset [76];

Ictal High-amplitude rhythmic activity slower than LVFA (typically 4-10 Hz)

- Hypersynchronous events feature by enhanced excitation and decreased inhibition [113];

- Abrupt decrease in slow inhibition [76];

- Increased firing of excitatory principal cells [136];

- Post-inhibitory rebound excitation causes transition to ictal events [127, 128, 136, 137];

1.5.5. Excitation and Inhibition in Seizure Termination

(39)

23 | P a g e

for development of new strategies against epilepsy.

Optogenetic silencing of excitatory principal cells was reported to suppress epileptiform activity in organotypic hippocampal mouse slice cultures [138]. The anti-epileptic effect possessed by optogenetic activation of PV+ inhibitory interneurons was also reported [139]. It has been demonstrated that either optogenetic activation of fast-spiking PV+ inhibitory interneurons or inhibition of excitatory principal cells stop the seizures rapidly in a mouse model of TLE in vivo [123]. In vitro study demonstrated the effectiveness of SOM+ inhibitory interneurons activation in suppressing ongoing seizures in acute hippocampal slices. Global optogenetic activation of mixed inhibitory interneurons was more effective due to a more general GABA release [140], suggesting the participation of both types of inhibitory interneurons at seizure suppression. Indeed, functional upregulation of inhibitory interneurons, particularly fast-spiking ones, was reported at seizure termination in vitro in TLE [141].

New advances showed that optogenetic activation of inhibitory interneurons appear to have both anti-epileptic and ictogenesis effect, depending on the timing of light delivery [139]. More precisely, activation of inhibitory interneurons during ictal period would terminate the seizure. This was attributed to the suppression of principal cell firing caused by enhanced inhibition. On the other hand, interictal activation of the same population of interneurons led to seizure initiation, probably due to increased post-inhibition rebound firing of principal neurons [139].

1.6. Objectives and Outline of the Thesis

In section 1.5, the role played in the generation of IES and HFOs as well as ictogenesis, by the excitatory and inhibitory signaling was discussed. Taken the evidence together, the excitation and inhibition play an essential and complex role in the generation of epileptiform activities, from interictal, onset, ictal and termination. Simply enhanced excitation and/or decreased inhibition cannot fully explain the dynamics underlying epileptic seizures. The two opposing forces interact dynamically during different stages of a seizure. However, existing studies mostly focus on one specific stage of seizure occurrence. The global view of epileptic seizures in its temporal course is still rather vague.

(40)

24 | P a g e

based approach. More specifically, the objectives of this thesis are as follows:

- Propose a methodology that allows for better understanding of the dynamics underlying epileptic seizures through combining a physiological model with clinical data.

- Explore the temporal evolution of excitation/inhibition balance from a global perspective, i.e. from a few minutes before seizure onset to a few minutes after seizure offset, with a fine resolution.

- Characterize the seizure occurrence as a path through synaptic gain parameter space that evolves with time, exploring this way, more details about evolution of excitation and inhibition during the occurrence of a seizure. Such a path allows comparison between patients and seizures, and thus enables patient-specific treatment.

- Provide a model-based clinical tool for revealing EEG waveform dynamics in a more objective manner, which can supplement expert interpretation.

- Propose a model-based algorithm for automatic epileptic seizure detection at an early stage.

(41)

25 | P a g e

Chapter 2. Analysis of the Neural Mass Model

(42)

26 | P a g e

(43)

27 | P a g e

2.1. The NMM for EEG modeling

The choice of the model type used in this thesis is obviously of major importance. Two major decisions were taken to choose the right type of model.

The first decision is related to the choice between physiological based models and mathematical (or phenomenological) models. Although mathematical models enable a complete analytical investigation of the model and can be used to gain detailed understanding on the different types of dynamic regimes, the link between the parameters and physiology is vague and it can be challenging to translate the revealed dynamics into neurophysiological interpretation [70]. Therefore we chose physiological based models over mathematical models.

The second decision is related to the choice among different levels of modeling, describing the neuronal behavior at different spatial scales with different levels of neurophysiological detail. Microscopic models allow for investigation of mechanisms of epilepsy at cellular even subcellular and network level. However, the computational load increases rapidly with the number of neurons and number of details considered, thus could be analytically intractable. The NMMs seem more realistic with reasonable simplification. In addition, they allow for direct comparison of the temporal dynamics of model outputs to those reflected in clinically recorded iEEGs. The NFMs were not chosen because they are considered as extension of NMMs. We believe it is necessary to first validate our methods with a simpler model before moving to a more complex one.

The theoretical basis of NMMs was established in 1950s [142], further developed by Wilson and Cowan [143]. A neuronal population is described as subsets of neurons, i.e. excitatory principal neurons and inhibitory interneurons, interacting through synaptic projections. It has been shown previously that the interaction of excitatory and inhibitory cells is universal in all nervous processes of any complexity [93, 144, 145]. The dynamic stability of the brain was demonstrated to be paradoxical when a neural population was composed entirely of excitatory cells [146]. Therefore, both excitatory and inhibitory cells were included within any local neural population.

(44)

28 | P a g e

existence of spatially localized neural populations has been justified by physiological evidence [147, 148]. Their findings indicate that many cells with relatively small volumes of cortical tissue show nearly identical responses to identical stimuli. Therefore, the spatially localized neural populations can be modeled by a single variable. The average behavior of each subset of neurons is described by lumped parameters without explicit representation of specific dynamics of ionic channels present in the neuron membrane. Therefore, investigation of changes in ionic channels kinetics, which is of interest in epilepsy research, is impossible. Despite this drawback, NMMs allows for study of dynamics underlying EEG waveforms at a population level and thus provide insights into epileptic seizures.

2.1.1. The State-of-the-art

NMMs have been increasingly popular in computational science in general, particularly in the field of epilepsy since they are able to reproduce accurately both ictal and interictal EEG signals.

Dynamics underlying epileptic seizures:

Dynamics underlying epileptic seizures:

insights from a neural mass model

Thesis presented by Xiaoya FAN

with a view to obtaining the PhD Degree in Engineering Sciences and

Technology (“Docteur en Sciences de l’

Ingénieur et Technologie”)

Academic year 2018-2019

Supervisor: Professor Antoine NONCLERCQ

BEAMS (Bio-, Electro- And Mechanical Systems)

Thesis jury:

Dynamics underlying epileptic seizures:

insights from a neural mass model

A dissertation presented by

Xiaoya FAN

This dissertation is submitted for the degree of

Doctor of Engineering Sciences and Technology

OCTOBER 4, 2018

Acknowledgements

Abstract

Scientific communications

Table of contents

List of Abbreviations

List of figures

List of tables

Chapter 1.

Background

1.1. General Introduction and Motivations

1.2. Epilepsy and Epileptic Seizures: Definition and Classification

1.3. EEG in Epilepsy

1.4. EEG Modeling

1.5. Excitation and Inhibition in Epilepsy

1.6. Objectives and Outline of the Thesis

Chapter 2.

Analysis of the Neural Mass Model

2.1. The NMM for EEG modeling