MULTIELECTRODE ARRAYS: A TOOL FOR MODELLING GENETIC

A DISSERTATION SUBMITTED TO THE FACULTY OF MEDICINE, DENTISTRY & HEALTH SCIENCES & THE FLOREY INSTITUTE OF NEUROSCIENCE AND MENTAL HEALTH, UNIVERSITY OF MELBOURNE IN TOTAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

Emma Morrisroe

626433

ORCID ID: 0000-0002-7145-1787

June 2019

Page 1

ABSTRACT

Background: Dravet Syndrome (DS) is classified by severe seizures and progressive developmental delay. In 80% of cases, DS is caused by mutations in the SCN1A gene.

Despite a growing understanding of the underlying pathomechanisms of DS thanks to animal and heterologous expression models, clinical outcomes for DS remain poor.

This has led to the investigation of new high throughput models for diagnosis and drug discovery as an avenue for precision medicine. Multielectrode arrays (MEAs) allow dissociated to form networks on a surface embedded with microelectrodes that record their extracellular activity. This model is promising in the field of precision medicine due to a greater complexity than heterologous expression systems and potential for higher throughput drug screening than can be achieved using animal models. Following this premise, this thesis addresses the following aims.

Aim 1: Establish a workflow for culturing primary cortical neurons on MEAs and determine the ideal culture conditions to rapidly generate stable networks for further high-throughput disease modelling and drug screening.

Aim 2: Evaluate the potential of the MEA system in modelling DS. Sensitivity of the system will be determined by comparing activity from networks sourced from animal models that exhibit a severe or mild phenotype as a result of SCN1A haploinsufficiency due to their genetic background.

Aim 3: Determine the predictive validity of MEAs for screening therapeutic compounds by applying known anti-epileptic drugs to the DS model networks.

Page 2

Method: Primary cortical mouse neurons were cultured on multiwell MEA plates.

Various seeding densities and lengths of maturation were tested to determine the optimal parameters for generating stable and reproducible networks. Stability was assessed by comparing a variety of features describing spiking, bursting, and network connectivity. Genetic epilepsy was assessed by culturing neurons derived from mouse models with SCN1A haploinsufficiency. These mice are of different strains and have genetic backgrounds that influence the severity of the SCN1A+/- phenotype (severe or mild expression). Once again spiking, bursting and network connectivity parameters were compared. Finally, stiripentol, lamotrigine and cannabidiol were applied to DS networks and the discriminatory effects between the DS and normal networks were compared.

Results: A stable baseline was generated at DIV21 using seeding density 11,646 cells/mm2. SCN1A haploinsufficiency in the C57BL/6 and SV129 backgrounds mimicked what is observed in vivo, with a significant epilepsy phenotype observed in the C57BL/6 background that was not seen in the SV129 background. While stiripentol, lamotrigine and cannabidiol all had significant effects on the overall activity profile of the networks, the response to the drugs was largely indiscriminate between normal and

SCN1A+/- networks.

Conclusion: A workflow to generate stable networks was achieved and was successfully implemented when characterizing the genetic DS network models. The MEAs captured the DS phenotype and discriminated between severe and mild phenotypic expression.

While there were occasional significant differences between normal and SCN1A+/-

Page 3 network responses to known AEDs, these small differences alone were not enough to confirm that this assay showed predictive validity for drug screening. Further studies examining different drugs and concentrations on the networks, as well as more sensitive analysis methods should be performed to further understand the potential of this assay.

Page 4

DECLARATION

I, EMMA MORRISROE, declare that this thesis entitled “MULTIELECTRODE

ARRAYS: A TOOL FOR MODELLING GENETIC EPILEPSY” and the work presented in it are my own and have been generated by me as the result of my own original research.

I confirm that:

1. This work was done wholly or mainly while in candidature for a research degree at

The Florey Institute of Neuroscience and Mental Health, Department of Medicine,

Dentistry and Health Sciences, The University of Melbourne;

2. This thesis has not been previously submitted for a degree or any other qualification

at this University or any other institution;

3. Where I have consulted the published work of others, this is always clearly

attributed;

4. Where I have quoted from the work of others, the source is always given;

5. I have acknowledged all main sources of help;

6. Where the thesis is based on work done by myself jointly with others, I have made

clear exactly what was done by others and what I have contributed myself;

7. The thesis is less than 100,000 words in length.

Emma Morrisroe

Page 5

PREFACE

This thesis comprises only my original work towards the PhD except for materials, information and experiments as stated below:

All software used to record and extract multiwell MEA data was provided by

Multichannel Systems, Reutlingen, Germany.

Network feature extraction was performed using MATLAB running custom scripts modified by Dr. Dulini Mendis.

All work included in this thesis is unpublished material, not submitted for publication

I have been funded through an Australian Postgraduate Award and received the

Caitlin’s Fund Travel Scholarship.

Emma Morrisroe

Page 6

TABLE OF CONTENTS ABSTRACT ...... 2 DECLARATION ...... 5 PREFACE ...... 6 LIST OF FIGURES ...... 9 ACKNOWLEDGEMENTS...... 11 CHAPTER 1| LITERATURE REVIEW ...... 12 1.1 Epilepsy ...... 12 1.2 Classifications: ...... 12 1.2.1 Seizures ...... 12 1.2.2 Epilepsy Type & Syndromes ...... 13 1.2.3 Etiology ...... 14 1.3 Genetic Epilepsy ...... 14 1.4 Epileptic Encephalopathy ...... 15 1.6 Dravet Syndrome ...... 16 1.5 Sodium Channels ...... 16 1.6 Precision Medicine: ...... 19 1.7 Tools for Modelling Genetic ...... 20 1.7.1 Heterologous Expression Assays...... 21 1.7.2 Animal Models ...... 22 1.7.3 Neuronal Networks ...... 24 1.8 Multielectrode Arrays ...... 25 1.9 Dissertation Aims ...... 28 CHAPTER 2 | DEVELOPMENT OF A RODENT CORTICAL CULTURE MULTIELECTRODE ARRAY ASSAY ...... 29 2.1 Abstract ...... 29 2.2 Introduction...... 31 2.3 Methods ...... 33 2.3.1 Animals ...... 33 2.3.2 Media Preparation ...... 34 2.3.3 MEA Preparation ...... 34 2.3.4 Primary Culture Procedure ...... 35 2.3.5 Network Activity Recordings ...... 36 2.3.6 Raw Data Processing...... 37 2.4 Results ...... 40 Page 7

2.5 Discussion ...... 48 2.6 Conclusion ...... 52 CHAPTER 3 | DEVELOPING A NETWORK MODEL OF DRAVET SYNDROME ON MULTIELECTRODE ARRAYS ...... 54 3.1 Abstract ...... 54 3.2 Introduction ...... 56 3.3 Method ...... 63 3.3.1. Animals ...... 63 3.3.2 Media Preparation ...... 63 3.3.3 MEA Preparation ...... 63 3.3.4 Primary Culture Procedure ...... 64 3.3.5 Raw Data Processing...... 64 3.4 Results ...... 64 3.5 Discussion ...... 79 3.6 Conclusion ...... 82 CHAPTER 4 | EXPLORING PREDICTIVE VALIDITY OF GENETIC DRAVET SYNDROME MODEL ON MULTIELECTRODE ARRAYS...... 84 4.1 Abstract ...... 84 4.2 Introduction...... 86 4.3 METHOD ...... 92 4.3.1 Animals ...... 92 4.3.2 Media Preparation ...... 92 4.3.3 MEA Preparation ...... 92 4.3.4 Primary Culture Procedure ...... 92 4.3.5 Compound Testing...... 92 4.3.6 Raw Data Processing...... 93 4.4 Results ...... 94 4.5 Discussion ...... 118 4.6 Conclusion ...... 125 CHAPTER 5| DISCUSSION ...... 126 REFERENCES ...... 132 SUPPLEMENTARY MATERIAL ...... 145

Page 8

LIST OF FIGURES

Figure 1.1. Two common MEA designs...... 26 Figure 2.1. Representation of network activity on a raster plot...... 38 Figure 2.2. Raster plot of network activity at DIV 14...... 41 Figure 2.3. Variability within and between cultured networks...... 43 Figure 2.4. Parameters from networks at 3 time-points...... 44 Figure 2.5. Parameters of culture at different time-points and densities...... 46 Figure 3.1. Raster plots representing 30sec of activity from SCN1A+/-C57BL/6 networks...... 65 Figure 3.2. Spiking parameters in SCN1A+/-/C57BL/6 networks ...... 66 Figure 3.3. Bursting parameters in SCN1A+/-/C57BL/6 networks...... 67 Figure 3.4. Connectivity parameters in SCN1A+/-/C57BL/6 networks...... 68 Figure 3.5. Raster plots representing 30sec of activity from SCN1A+/-/SV129 networks...... 70 Figure 3.6. Spiking parameters in SCN1A+/-/SV129 networks...... 71 Figure 3.7. Bursting parameters in SCN1A+/-/SV129 networks...... 72 Figure 3.8. Connectivity parameters in SCN1A+/-/SV129 networks ...... 73 Figure 3.9. Spiking parameter ratios between SCN1A+/- networks in C57BL/6 and SV129 backgrounds...... 75 Figure 3.10. Bursting parameter ratios between SCN1A+/- networks in C57BL/6 and SV129 backgrounds...... 76 Figure 3.11. Connectivity parameter ratios between SCN1A+/- networks in C57BL/6 and SV129 backgrounds...... 77 Figure 4.1. Spiking parameters in C57BL/6 networks in response to LTG...... 95 Figure 4.4. Spiking parameters in SCN1A+/- and WT networks in response to LTG...... 99 Figure 4.5. Bursting parameters in SCN1A+/- and WT networks in response to LTG.... 100 Figure 4.6. Connectivity parameters in SCN1A+/- and WT networks in response to LTG...... 101 Figure 4.7. Spiking parameters in C57BL/6 networks in response to STP...... 103 Figure 4.8 Bursting parameters in C57BL/6 networks in response to STP...... 104 Figure 4.9. Connectivity parameters in C57BL/6 networks in response to STP...... 105 Figure 4.10. Spiking parameters in SCN1A+/- and WT networks in response to STP. ... 107 Figure 4.11. Bursting parameters in SCN1A+/- and WT networks in response to STP. . 108 Figure 4.12. Connectivity parameters in SCN1A+/- and WT networks in response to STP...... 109

Page 9

Figure 4.13. Spiking parameters in C57BL/6 networks in response to CBD...... 111 Figure 4.14. Bursting parameters in C57BL/6 networks in response to CBD...... 112 Figure 4.15. Connectivity parameters in C57BL/6 networks in response to CBD...... 113 Figure 4.16. Spiking parameters in SCN1A+/- and WT networks in response to CBD. .. 115 Figure 4.17. Bursting parameters in SCN1A+/- and WT networks in response to CBD. 116 Figure 4.18. Connectivity parameters in SCN1A+/- and WT networks in response to CBD...... 117 Supplementary Figure (SF) 1. Raster plots representing 30sec of activity from SCN1A+/- C57BL/6 networks before and after application of 100uM lamotrigine...... 146 Supplementary Figure (SF) 2. Raster plots representing 30sec of activity from SCN1A+/- C57BL/6 networks before and after application of 40uM stiripentol...... 148 Supplementary Figure (SF) 3. Raster plots representing 30sec of activity from SCN1A+/- C57BL/6 networks before and after application of 10uM cannabidiol...... 150

Page 10

ACKNOWLEDGEMENTS

I would like to first and foremost thank my supervisors Steven Petrou and Christopher

Alan Reid. Thank you for your expert guidance and support throughout my degree.

There was no question I felt I could not ask. Snezana Maljevic, thank you for your advice and encouragement, especially throughout my writing process. I will always appreciate the time you made for me. Dulini Mendis, thank you for sharing this MEA journey with me through all the ups and downs. Elena Gazina, thank you for imparting me with your infinite wisdom of molecular biology. Know that at least some of it got through to me. Lynley Cordeiro and Kay Richards, thank you for always being up for a laugh and for making the lab a warm and welcoming place. To Sasha Zaman and

Gabriel Jones, thank you for your many years of friendship and support. Here’s to many years ahead. To the whole Epilepsy and Channels laboratory group, thank you all for supporting me over the past 6 years. I enjoyed working with every single one of you.

To Mum, Dad and Liam, thank you for your undying support along my journey and for always having my back. To Nicole Gooey, thank you for being a constant shoulder to lean on. To the H Pylori Groupies, my favourite fish, thank you for being an unyielding source of delight. To David Edwards, my hardest times were made easier because of you. To all my friends and family, thank you for sticking by me.

I would also like to thank the Florey Animal House staff, the Multichannel Systems

Staff, my examiners and chair, and the funding sources that allowed me to pursue my degree (The Australian Postgraduate Award; The Caitlin’s Fund Scholarship)

Page 11

CHAPTER 1| LITERATURE REVIEW

1.1 Epilepsy

Epilepsy is a serious neurological disease estimated to affect more than 65 million people, with a lifetime prevalence of 1%. It is most broadly characterised by a person having at least one epileptic seizure, following which their “demonstrates a pathologic and enduring tendency to have recurrent seizures” (Fisher et al. 2014). An epileptic seizure is further defined as “a transient occurrence of signs and/or symptoms due to abnormal excessive or synchronous neuronal activity in the brain” (Fisher et al.

2005). Epileptic seizures manifest in a variety of ways such as loss of consciousness, impaired cognitive functioning, involuntary movements or the inability to move, and various sensory disturbances (Kumar et al. 2012; Fisher and Kaplan 2004). While most patients reach seizure freedom through therapeutic intervention, one third of people diagnosed with epilepsy continue to live with poorly managed seizures due to either drug resistance or extreme pharmacological side effects (Ngugi et al. 2010; Hesdorffer et al. 2011; Engel 2013; Oyrer et al. 2018).

1.2 Classifications:

1.2.1 Seizures

Classification of epilepsy is the first step towards patient diagnosis and treatment. It informs the understanding of the type of seizures a patient has, potential triggers of these seizures, likely prognosis for the patient, potential comorbidities and most importantly, effective therapies. The classification of epilepsies remained unchanged from 1989 until 2017, when the system was updated to reflect a deeper understanding of

Page 12 underlying mechanisms, diagnosis, and treatment attained through scientific research

(Scheffer et al. 2017; Falco-Walter et al. 2018). Seizures are first classified according to type of onset: generalised, focal, or unknown. Generalised epileptic seizures almost instantly engage both hemispheres of the brain and result in loss of consciousness or awareness. They are further categorised as motor (eg. tonic-clonic) or non-motor (eg. absence) based on involuntary movement. Focal seizures originate unilaterally as opposed to bilaterally, and can evolve to a bilateral tonic-clonic seizure or remain localised. Focal seizures can be further classified as ‘aware’ or ‘impaired awareness’

(formerly simple and complex) seizures, and finally as motor or non-motor. Unknown onset seizures refer to seizures of unknown origin, and can also be classified as motor and non-motor (Falco-Walter et al. 2018; Scheffer et al. 2017).

1.2.2 Epilepsy Type & Syndromes

After seizure classification progresses into an epilepsy diagnosis, the epilepsy type and syndrome type (where appropriate) can then be classified. Epilepsy types can be classified as focal, generalised, combined generalised and focal, and unknown. Epilepsy classifications are defined by the combined total classification of the patients’ seizures.

A combined general and focal category has been added to include syndromes that express both seizure types, such as Lennox-Gastaut or Dravet Syndrome (DS). Finally, where applicable, the epilepsy syndrome can be classified. Syndrome classification considers epilepsy type in combination with other features such as EEG, imaging, age- dependent features, and comorbidities. This level of classification helps dictate therapeutic interventions and informs outcome predictions. (Oyrer et al. 2018; Scheffer et al. 2017; Falco-Walter et al. 2018).

Page 13

1.2.3 Etiology

Throughout this classification process, there is a heavy focus on seizure etiology, which can be divided into six categories: structural, genetic, infectious, metabolic, immune, and unknown. An epilepsy disorder is said to have genetic etiology when “it directly results from a known or presumed genetic mutation in which seizures are a core symptom of the disorder” (Scheffer et al. 2017). Genetic etiology in epilepsy disorders are common, to the point that Idiopathic Generalised Epilepsy (IGE), a common subgroup of generalised epilepsies including childhood and juvenile absence epilepsies, juvenile myoclonic epilepsy, and generalised tonic-clonic seizures alone, have been more accurately classified as Genetic Generalised Epilepsy (GGE) to highlight their underlying etiology (Scheffer et al. 2017).

1.3 Genetic Epilepsy

While genetic mutations are being proposed as the underlying cause of an increasing number of epilepsies, in most cases the underlying gene(s) are not known (Scheffer et al. 2017; Falco-Walter et al. 2018). In 1995, Berkovic and colleagues described the first epilepsy gene (Steinlein et al. 1995). In 2017, this number had increased to 84 known epilepsy genes, and 609 epilepsy-related genes (Wang et al. 2017). Historically, the first efficient method of determining a genetic cause of epilepsy was mapping pedigrees of the families of effected individuals, and performing linkage analyses and positional cloning to identify the pathogenic mutation or mutations. While this was an exciting proof of principle, it was costly, time-consuming, and unable to fully capture the depth of the genetic influence in epilepsy (Poduri and Lowenstein 2011). Most notably, it was unable to detect spontaneous de novo mutations which are not represented in pedigrees.

Page 14

As many severe epilepsies arise through de novo mutations, this is a significant shortcoming (Allen et al. 2013). Genome-wide association and copy number variant investigations also resulted in limited findings. In recent years, the application of next generation sequencing (NGS) has accelerated the process of identifying pathogenic mutations to an unprecedented degree. In 2010 a large, collaborative study was formulated with the aim of using NGS to sequence at least 4000 genomes in 4000 epilepsy patients across the world (Consortium 2012). Through exome sequencing, this and other similar collaborations have been instrumental in discovering novel genes associated with severe epilepsies, most notably epileptic encephalopathies (Epi 2015;

Consortium 2017; Allen et al. 2013; Suls et al. 2013; Appenzeller et al. 2014).

1.4 Epileptic Encephalopathy

Epileptic encephalopathies (EEs) are a heterogeneous group of severe epilepsies “where the epileptic activity itself contributes to severe cognitive and behavioural impairments above and beyond what might be expected from the underlying pathology alone” (Berg et al. 2010). Most EEs are broadly defined by frequent seizures coupled with abnormal

EEG patterns, and progressive developmental delay. EEs fall along a spectrum of severity and can manifest at any age; however, the term is most commonly used to describe seizure disorders with onset in infancy or childhood. This is unsurprising since as many as 40% of epilepsies with onset between the ages of 1-3 years can be attributed to EE (Eltze et al. 2013; Scheffer et al. 2017; McTague et al. 2016). The etiology of EEs is commonly genetic, and is often caused by a mutation in a single gene. One gene can be implicated in many different EEs, as well as several genes can cause in a single EE syndrome (McTague et al. 2016). One further commonality of EEs is that they are often

Page 15 refractory to standard antiepileptic drugs (AEDs) making treatment a series of trial and errors that may never result in long term seizure freedom (Khan and Al Baradie 2012;

Jain et al. 2013).

1.6 Dravet Syndrome

Dravet Syndrome (DS) is a severe form of infantile epilepsy and was the first EE with a known genetic cause (Claes et al. 2001). DS presents in previously healthy and normally developing infants at around 6 months of age. These infants develop febrile status epilepticus and over the next six months, exhibit convulsive episodes that are often triggered by fever. When the children are between 1-4 years of age, other seizure types emerge. At 6 years of age, while the frequency of seizures reduces, they persist into adulthood. Seizures are also accompanied by severe cognitive impairment, ataxia, and social interaction deficits (Scheffer et al. 2009; Genton et al. 2011; Li et al. 2011).

Patients with DS report significantly lower quality of life, significantly more behavioural problems, and a mortality rate between 3.7-17.5% (Sinoo et al. 2019;

Cooper et al. 2016). Seizures in DS patients are often refractory and in some cases, traditional AEDs can further exacerbate the symptoms (Catterall 2018).

1.5 Sodium Channels

The first gene to be associated with an EE was SCN1A. This remains the most common epilepsy gene to this day, with over one thousand mutations described (Salgueiro-

Pereira et al. 2019). Additionally, mutations in the SCN1A gene have been identified as the underlying cause of DS in 80% of patients (Bender et al. 2012). SCN1A codes for

Page 16 the -subunit of a gated sodium channel (Nav1.1), which regulates neuronal excitability and is responsible for the initiation and propagation of action potentials.

Sodium (Nav) channels are expressed at a higher concentration at initial segments

(AIS), the site of (AP) initiation (Kole et al. 2008). Each Nav channel consists of one pore-forming  subunit, coupled with one or two non-pore forming  subunits. It is this  subunit that allows Na+ to pass through the membrane, depolarizing the cell and resulting in the rising phase of an action potential. Reduced or loss of function of this channel due to epilepsy-causing mutations lead to fewer action potentials and consequentially less excitability of the network. While it seems contradictory that a decrease in network excitability would lead to seizures, it has been widely accepted that the SCN1A mutation specifically effects the excitability of parvalbumin-positive interneurons. These are GABAergic inhibitory neurons and their role is to regulate firing of excitatory pyramidal neurons. The loss of function in inhibitory neurons causes disinhibition of excitatory neurons, resulting in increased and deregulated firing of excitatory neurons, generating seizure activity (Yu et al. 2006;

Ogiwara et al. 2007; Dutton et al. 2013).

SCN1A mutations are so heavily implicated in DS that genetic screening is recommended in infants and children that experience repeated febrile seizures, as it is believed that these seizures are the initial manifestation of the syndrome. However, detection of an SCN1A mutation is not enough to confirm a DS diagnosis, as individuals with detected mutations have variable progression regarding phenotypic severity. A study analyzing seizure phenotype in 200 individuals with an SCN1A mutation found that while 45.5% were diagnosed with DS, the remaining participants were diagnosed

Page 17 with milder epilepsies such as GEFS+ (23.8%), febrile seizures (14.8%), and focal seizures (3.5%). Furthermore, 8.9% of participants with SCN1A mutations were healthy individuals showing no signs of epilepsy. The severity of the mutation is also not entirely predictive of seizure outcomes. Traditionally, truncation mutations are associated with DS, as these mutations typically result in total loss of function of the protein as they completely ablate the pore-forming region. Likewise, missense mutations are associated with milder epilepsy types as these mutations, while changing the structure of the SCN1A protein, often do not affect the pore forming region.

However, of the 127 mutations identified in this cohort, 13 truncation mutations were identified in individuals with mild phenotypes, and 37 missense mutations outside the pore-forming region were seen in DS patients (Cetica et al. 2017). The genetic background of different individuals, as well as mosaic mutations and modifier genes are thought to contribute to phenotypic severity, highlighting that a one-size-fits-all approach to epilepsy diagnosis, prognosis and treatment avenues is not suitable for this disease (Marini et al. 2011; Scheffer et al. 2009).

Although much is known about the etiology of DS, medical interventions and drug therapy have so far been largely ineffective. Complete seizure control is typically not attainable, with the primary goal to reduce the most severe seizure types while minimizing the negative effects of AEDs (Knupp and Wirrell 2018). The lack of response to standard treatment coupled with the diverse underlying genetic component makes DS, and epilepsy syndromes as a whole, a promising candidate for the successful implementation of precision medicine (Striano et al. 2016).

Page 18

1.6 Precision Medicine:

Precision medicine is the practice of designing targeted therapies for individual patients based on underlying gene-specific mechanisms, ideally in combination with genetic background. This is in opposition to administering therapy based on treating symptoms alone (Maljevic et al. 2017). Precision medicine is not a novel idea; for instance, blood transfusions dependent on blood type have been in place for more than a century.

However, with the recent development of rapid and large scale genome sequencing, the promise of widespread personalized medicine is starting to be realised (Collins and

Varmus 2015). One example of precision medicine in action is the development of the cystic fibrosis (CF) drug Ivacaftor. CF is caused by reduced or complete loss of function of the CFTR channel, resulting in improperly regulated cellular salt and water balance.

In 3-4% of patients, this is due to specific gene variant G551D in the CFTR gene.

Ivacaftor is a small-molecule drug introduced in 2012 and is the first CF drug with a mechanisms of action of counteracting the underlying genetic defect rather than treat surface symptoms. Patients on Ivacaftor showed a slower decline in lung function compared to standard treatment, as well as lower rates of hospitalization amongst users.

Since its release, Ivacaftor has been found to be effective against CF caused by 37 different mutations, meaning 14% of the CF population in the US were eligible for this treatment in 2017 (Feng et al. 2018; Condren and Bradshaw 2013).

Oncology is another field in which precision medicine advises patient care (Nör and

Gutkind 2018; Nakagawa and Fujita 2018; Low et al. 2018). A recent study performed whole exome and RNA sequencing on a subset of patients with advanced pancreatic cancer. Therapeutically relevant mutations were identified in 48% of the cohort, and

Page 19 pathogenic mutations identified in 18%. Furthermore, 30% of patients had their clinical management altered once their genomic data was known to more appropriately treat the disease (Aguirre et al. 2018). A platform for integrating genomic data into the workflow of patient care was proposed and implemented in a cohort of 769 cancer patients in a

2017 clinical trial. In cases where genome analysis alone was not sufficient to identify avenues for therapy, patient-derived animal and cellular models such as PDX mice

(patient-derived xenograph mice, the development of which involves implanting patient tumor cells into immunocompromised animals) and organoids were used as a platform for high throughput drug screening to identify novel therapies. Two patients displayed similar mutations however the drug screening process clearly distinguished them as having disparate therapeutic needs. Another patient displayed pharmacoresistance to all but one drug screened (Pauli et al. 2017). These findings highlight the importance of developing predictive and specific genetic models of disease for further classification and drug screening at a patient level.

1.7 Tools for Modelling Genetic Epilepsies

In order to translate the genetic findings underpinning severe EEs into actionable therapies for patients using precision medicine, the pathomechanisms and drug interactions must be understood. These are investigated by expressing the mutation in an experimental model. These can range from simple single-cell models to complex in vivo animal models. Ideal models of genetic epilepsy will a) express the genetic risk factors identified in patients; b) recapitulate disease mechanisms and pharmacosensitivity seen in patients; c) be suitably scalable for high-throughput

Page 20 screening; and d) have the ability to represent the scope of pathological mutations including secondary effects (Epi 2015).

1.7.1 Heterologous Expression Assays

The primary assay used to investigate the effects of a genetic mutation on protein function are heterologous expression assays. These assays involve incorporating the mutated gene into a cell line that otherwise would not express this gene. This allows the gene product to be assessed in isolation, and will not be confounded by expression of a similar native gene (Epi 2015). Common cell types for such models are Xenopus laevis oocytes, human kidney (HEK) cells or Chinese hamster ovary (CHO) cells. These cells are ideal for epilepsy modelling as they express very few endogenous voltage gated ion channels. The expression of the desired protein in these cells can then be analysed using electrophysiological, biochemical, or antibody staining methods (Maljevic et al. 2017).

Most commonly, the effects of mutations on sodium channel function are measured using the patch clamping technique, where the current and voltage of single channels can be measured.

Many SCN1A mutations have been assessed using heterologous expression assays, and these assays help determine whether the mutation leads to loss of function, gain of function, or has mixed effect (Schutte et al. 2016). In a recent study, Berecki et al

(2019) investigated the effects of the T226M Nav1.1 channel mutation, known to cause profound developmental and epileptic encephalopathy that starts earlier and seems to be more severe than DS. Through expression of this mutation in CHO cells and utilizing the patch clamping technique, a unique epileptogenic mechanism that corresponds to a

Page 21 gain of function of SCN1A was identified. It was determined that at low levels of synaptic input, the mutation enhanced excitability of PV interneurons, increasing net inhibition. However once synaptic input reached ‘typical’ levels, a block resulted in overexcitation and disinhibition. This mechanism was found to be

“functional dominant negative” meaning the T226M channel can reduce the activity of wild-type channel, supporting the observations that the phenotype expressed as a result of a T226M is more severe than seen in Dravet Syndrome. Also recently, three de novo missense mutations identified in DS patients were tested using this system, and severity in vitro was found to correlate with clinical phenotype severity, illustrating the utility of this technique in screening mutations to inform disease progression and prognosis

(Nissenkorn et al. 2019). This system has also proven useful for testing potential therapeutic compounds as it has the ability to detect modulated channel function as a result of drug application (Thompson et al. 2011). While the heterologous expression system has been validated against neuronal tissue and is suitably scalable for high throughput screening, it does not consider background genetics or interactions with other endogenously expressed proteins or structures in neurons. It also does not successfully model cell compartment trafficking deficits, as heterologous expression systems lack the complexity of the structural composition of neurons. Therefore, more complex models are required when unravelling the true impact of a mutation on a biological system (Maljevic et al. 2017).

1.7.2 Animal Models

Epileptogenetic mechanisms are most commonly studied in rodent models, as rodents display clear seizure phenotypes that correlate with human patients harboring an

Page 22 identical mutation. Mouse models of SCN1A mutations have been instrumental in determining the underlying cause of DS. Many mouse models of DS have been generated, including targeted deletion of SCN1A exon 1 and exon 26 (Miller et al. 2014;

Yu et al. 2006), point-mutation knock-ins (Ogiwara et al. 2007; Martin et al. 2010; Tsai et al. 2015) and a transgenic model with a human GEFS+ mutation (Tang et al. 2009).

A knockout SCN1A model, intended to recapitulate the truncation mutations seen in patients, revealed reduced excitability in parvalbumin positive interneurons, while activity in excitatory neurons remained unaffected (Yu et al. 2006; Kalume et al. 2007).

Furthermore, it was found that happloinsufficiency of Nav1.1 resulted in disinhibition of cortical excitatory neurons through impaired function of GABAergic interneurons

(Tai et al. 2014). It has also been found in mouse models that selective deletion of

Nav1.1 in inhibitory neurons results in seizures and high mortality, whereas the same deletion in only excitatory neurons has the opposite effect (Ogiwara et al. 2013).

Rodents are particularly useful as models of DS as they produce spontaneous seizures, mirroring what is seen in human patients. This is not a feature able to be replicated in more simple DS models. Rodents also exhibit many of the comorbidities seen in DS.

Mice with a loss-of-function allele demonstrate hyperactivity, anxiety-like behaviours, lower sociability, a decrease in spatial learning, and memory impairment (Ito et al.

2013). Finally, rodent models display pharmacological validity, responding similarly to human patients when given AEDs. Hawkins et al. (2017) evaluated a battery of AEDs in a mouse model of DS and found that lamotrigine worsened the seizure phenotype, while clobazam, valproic acid, stiripentol, levetiracetam, and phenobarbital improved the seizure phenotype. This is consistent with what is known in human patients, with

Page 23 clobazam, valproic acid and stiripentol often prescribed in combination, and lamotrigine commonly avoided due to seizure exacerbation. While there are many positives to rodent models in genetic epilepsy, lack of scalability is a major issue regarding the pursuit of precision medicine. Animal models are expensive and time-consuming to produce, and phenotype expression of a mutation is heavily influenced by the genetic background of the model used (Maljevic et al. 2017; Epi 2015). This does not align with the ultimate goal, to find a treatment for each individual based on the expression of their specific mutation.

1.7.3 Neuronal Networks

Artificially cultured neuronal networks are a promising avenue for high throughput genetic screening. It is suspected that many forms of epilepsy arise through one mutation affecting multiple neural cells by impacting cell interactions. Additionally, therapeutic compounds that act through indirect compensation of mutation effects can only be evaluated or identified if many neuronal processes are apparent in the model

(Epi 2015).

The electrical nature of neural activity facilitates capturing the behaviour of neurons through measurements of the internal or external environment. At the macroscale, this has led to the development of EEG, fMRI, and PET scanning technologies which capture broad features of the overall activity of a brain in vivo (Obien et al. 2015). At the microscale, neuronal networks have been extensively studied using slice and artificially generated network models. An example of this is multi-site patch clamping.

Multi-site patch clamping involves patch-clamping a small number of neurons

Page 24 simultaneously, allowing the relationship between these neurons to be determined.

While multi-site patch clamping allows the degree of network connectivity between neurons to be precisely measured, only a very small portion of the entire network can be monitored this way and it is a very time consuming practice (Spira and Hai 2013).

While both these technologies are useful for developing an understanding of the mechanisms behind genetic epilepsies, neither provide a suitable platform for high- throughput screening.

1.8 Multielectrode Arrays

Multielectrode array (MEA) technology was first introduced in 1972, however due to cost and technological restrictions, it has only recently become accessible for widespread laboratory use. In vitro MEAs are small chips with embedded microelectrodes flanked by a well chamber, allowing disassociated neurons to be cultured on the surface of the electrodes (Figure 1.1). The electrodes are tightly packed and detect neuronal activity by transducing the change in voltage from the extracellular environment. The spatial resolution of the chip varies, with chips ranging from three electrodes to capture broad network changes, to thousands of electrodes able to detect changes along a single axon. (Cotterill et al. 2016). Electrodes are generally made with metallic conductors such as , or platinum. The purpose of this is to achieve low impedance, generating the highest possible to noise ratio (SNR) so that the activity of the networks is clearly identifiable over background noise (Obien et al.

2015). Each electrode can detect the voltage changes from the nearest and other supporting cells in the extracellular field. The detected magnitude of this activity depends on the proximity of the neuron to the electrode (Illes et al. 2007; Arnold et al.

Page 25

2005). Single action potentials (spikes) are detected when a neuron is active over one electrode. Bursts of action potentials are detected when isolated pockets of neurons generate action potentials in synchrony. Network bursts are detected once the network reaches maturity and many neurons across all electrodes generate an action potential in synchrony.

a) b)

Figure 1.1. Two common MEA designs. The swMEA (a) consists of 64 titanium electrodes embedded in glass for higher density, lower throughput investigation. The mwMEA (b) consists of 12 gold electrodes in each well of a 24-well plate, embedded in

FR4 to aid cellular adhesion to the base of the well. This is for lower density, higher throughput investigation such as compound screening.

These recent developments in MEA technology have made the platform more desirable for high-throughput and toxicology studies, most specifically with the introduction of the multiwell MEA (mwMEA) plate. Multiwell plates are set up in the style of cell culture plates with electrodes placed in the bottom of each well. Multiwell

MEAs are far more efficient than traditional single-well MEAs (swMEAs) as swMEAs require more manual labor in the way of media changes, and can only be recorded from

Page 26 one at a time, resulting in potential error. While previously it was labour intensive to study more than 10 networks at a time using the single-well system, the introduction of the mwMEA plate allows for simultaneous growth of 12, 24, 72, or 96 cultures on a single cell culture plate. By extension, experiments can be performed in parallel on one plate of cultures, reducing the time necessary for performing these experiments by up to

96 fold (Cotterill et al. 2016). The efficiency of the mwMEA system expands its capabilities, allowing for high-throughput compound screening. Recordings from MEAs have been used to describe and further characterize disease-state networks by using genetically manipulated animal neurons or induced pluripotent stem-cell neurons. By combining pharmacological effects with disease-state networks, it has been possible to further demonstrate what effects specific neuroactive drugs have on disease-state networks in vitro (Yi et al. 2015).

MEAs have been employed to enhance the understanding of neuronal communication, as well as a platform for disease modelling. Ito et al. (2013) utilized the MEA system, in combination with immunofluorescent staining, to evaluate the relationship between synaptic development and neuronal activity. It was found that the increase in glutamatergic and GABAergic synapses during development correlated with increased neuronal activity, which plateaued at the same time as synapse proliferation ceased.

MEAs have become popular in cardiac research as a model of Long QT syndrome

(Delisle and January 2017), detecting drug-induced arrhythmia (Navarrete et al. 2013), and in screening for cardiotoxicity of novel compounds (Li et al. 2016). They have also been utilised in pain research (Yang et al. 2018), and most recently have been used in screening novel compounds for seizure liability (Bradley et al. 2018).

Page 27

1.9 Dissertation Aims

These studies highlight the MEA system as a useful tool in neuronal network modelling and drug screening. In this thesis, I have aimed to further utilize the MEA system as a platform for disease modelling and drug screening in genetic epilepsy. I first established a workflow for developing stable cortical mouse neuron cultures on mwMEAs, as described in chapter 2. In the literature, extensive variability in the baseline activity of cultured networks between different laboratories culturing under different conditions was reported. It was for this reason that I established the parameters and optimal conditions for generating stable and consistent networks for use in future experiments.

In chapter 3 I utilized the optimized protocol developed in chapter 2 to examine the effectiveness of the MEA system in recapitulating a genetic mouse model of epilepsy. I aimed to determine whether the MEA system was sensitive enough to detect a seizure phenotype in a validated model of DS that expresses spontaneous seizures in vivo. I also investigated whether the system would detect a phenotype in networks derived from a mouse model that contained a SCN1A mutation but did not express spontaneous seizures. In chapter 4, I further investigated the utility of the MEA system for drug screening in genetic epilepsy. I applied three AEDs with known effects in DS onto the primary cultured mouse neurons and measured their effects on the activity profile of the network, investigating whether these drug effects were specific to the DS networks over normal networks. This was to understand whether the MEA system could be used as a platform for drug discovery and screening by detecting a drug-gene interaction effect.

These three studies explore the potential for modelling of genetic epilepsies on the

MEA platform, and assess the utility of this platform for screening promising compounds in a sufficiently complex yet high throughput system.

Page 28

CHAPTER 2 | DEVELOPMENT OF A RODENT CORTICAL CULTURE MULTIELECTRODE ARRAY ASSAY

2.1 Abstract

Introduction: Multielectrode arrays (MEAs) have proven to be a useful tool for studying the activity and connectivity of neuronal networks in vitro. However, there has not been a consensus on the optimal parameters to measure electrophysiological development, as the ideal parameters vary based on intended use and between laboratory groups. This has contributed to a large variation in baseline network activity seen between different laboratory groups using different protocols. This extensive variability suggests that before use as a platform for genetic modelling and drug screening, the ideal parameters for generating stable networks must be established. In this chapter, I aim to investigate the optimal parameters of the MEA system as a platform for future network modelling.

Aims: To generate a workflow to produce stable neuronal networks that can be used as the basis for further genetic modelling and drug screening studies.

Methods: Primary cortical mouse neurons were dissociated and plated on mwMEAs at varying concentrations and the activity was measured at three separate time-points to assess network maturation and stability. The activity profile of the network was established by extracting spiking and bursting features using custom MATLAB scripts described in Mendis et al. (2019); Mendis et al. (2014).

Page 29

Results: Spiking and bursting parameters were significantly different between DIV14 and DIV21, but not between DIV21 and DIV28. Additionally, cells at seeding density

11,646 cells/mm2 produced networks with the highest synchrony value.

Conclusion: After testing three seeding densities at three time-points, it was determined that a seeding density of 11,646 cells/mm2 at DIV21 produced stable networks that displayed adequate synchrony as a baseline for further exploration of this platform utilised in all chapters of this thesis.

Key Words

MEA, Multielectrode Array, Neuronal Networks, Cortical Neurons

Abbreviations

MEA – Multielectrode Array swMEA – Single Well Multielectrode Array mwMEA – Multiwell Multielectrode Array

MFR – Mean Firing Rate

NB – Network Burst

MFRIn – Mean Firing Rate Inside Bursts

MFROut – Mean Firing Rate Outside Bursts

NBSI – Network Burst Start Interval

DIV – Days in vitro

Page 30

2.2 Introduction

Multielectrode arrays (MEAs) have proven to be a useful tool for studying the activity and connectivity of neuronal networks in vitro. Neuronal networks grown on MEA chips demonstrate many of the characteristics observed in intact neuronal networks, such as spikes (single extracellular action potentials), bursts (groups of extracellular action potentials), and network bursts (bursts originating simultaneously at different spatial areas of the culture). The spontaneous activity recorded in these networks change in response to stimulation with various pharmacological compounds, mirroring what is seen in in vivo networks. Because of this, primary neuronal cultures on MEAs have emerged as a platform to study neurophysiology, neuropharmacology and neurotoxicology (Cotterill et al. 2016).

Many studies have been performed using low-throughput MEA systems to classify the physiological maturation of neuronal networks grown on these chips over time. These studies have used cultures derived from different brain regions, source animal age, cell densities, MEA types, and time-points to assess network development to maturation.

However, there has not been a consensus on the optimal parameters to measure electrophysiological development, as the ideal parameters vary based on intended use and between laboratory groups. This has contributed to a large variation in baseline network activity seen between different laboratory groups using different protocols

(Vassallo et al. 2017).

It has been previously reported in low-throughput MEA systems that neuronal network development retains low levels of activity (single spikes, no bursting activity) through

Page 31 the first 7 days in vitro. From 7 days on, cultures develop coordinated spiking across channels, coordinated bursting across channels, and finally network synchrony across the culture. However, this network synchrony and maturation varies between labs with some reporting it after two weeks and others stating it takes as long as five weeks or beyond to truly establish network stability (van Pelt et al. 2005).

Culture age is not the only factor in the development of networks on MEAs. It has been reported that cell density influences the development of the networks. Studies have shown that cell cultures at lower densities take longer to develop network synchrony and temporal stability (a lack of significant changes in parameters once networks reach a certain age) than high density cultures (Biffi et al. 2013). The reason for this is not known. Although it was hypothesized that lower density cultures contained fewer neurons to make contact with the electrode making network connectivity more difficult to detect. However, in these cases it has been seen that the number of neurons in contact with the electrodes does not differ between low and high density cultures (Wagenaar et al. 2006). Although the biological cause of this phenomenon is unknown, cell density is still a factor that must be considered when developing cultured networks.

While many parameters have been examined using higher density (swMEA) systems, the multiwell system is only recently being explored in this way. A study by Cotterill et al. (2016) examined the development patterns of cortical cultures on the mwMEA system and compared this to the swMEA development over time. Using primary rat cortical neurons on 48-well MEA plates containing 16 electrodes per well, development was measured at DIV 5, 7, 9, and 12, and the culture activity was described using 12

Page 32 features that described firing rate, bursting synchrony, and network spike properties.

They reported rapid development of spikes, bursts and coordinated activity, as well as the development of network synchrony over the 12 days. They concluded that this activity was similar to the development of swMEAs previously reported in the literature. Therefore, it can be concluded that fewer electrodes on the surface of the

MEA does not impact the ability for network development and connectivity to be characterized.

As there is no consensus for the optimal culture conditions for mouse primary cortical neurons on MEAs, as well as a general lack of publications regarding the mwMEA system, this study aims to both establish a workflow for the use of primary cortical mouse neurons on mwMEAs and to classify the development of these cultures on a mwMEA system. We also aimed to observe through the course of this study the network stability and the variability between identically seeded cultures. For the purposes of this study network stability is defined here as temporal stability of spontaneous electrophysiological activity from week to week during development, measured in 11-min recordings. The resulting workflow can then be used to continue further experiments using genetically modified animal neurons as well as to classify neuroactive drugs.

2.3 Methods

2.3.1 Animals

Mice were sourced from the ARC in Perth, Western Australia and were of genotype

C57BL/6J background strain. Animals were maintained and housed in the Florey

Page 33

Institute animal breeding facility. Litters were taken from the mother from 0-2 days after birth. All animal experiments were approved by the Howard Florey Institute

Animal Ethics Committee (AEC#14-075) and performed in accordance with the

Prevention of Cruelty to Animals Act and the NHMRC Australian Code of Practice for the Care and Use of Animals for Scientific Purposes.

2.3.2 Media Preparation

Prior to culturing, culture media was made using 89.3% Minimum Essential Medium

(MEM, Merck KGaA), 0.9% 1 M HEPES (Sigma-Aldrich), 6 mg glucose (Merck

KGaA) per 1 ml of MEM, 8.9% fetal bovine serum (Thermo Fisher Scientific) and

0.9% Penicillin/Streptomycin (P/S) solution (Sigma-Aldrich).

Neuronal media for culture maintenance was also made using Neurobasal-A medium

(Thermo Fisher Scientific) supplemented with 1.9% B27 (Thermo Fisher Scientific),

0.95% GlutaMax (Thermo Fisher Scientific), 0.95% HEPES (Thermo Fisher Scientific) and 0.95% P/S (Sigma-Aldrich).

2.3.3 MEA Preparation

24 well MEAs were sourced from Multichannel Systems (product code 24W700/100F-

288). Each well contained 12 gold electrodes (diameter = 100uM spaced 700uM apart) embedded in a base of FR4. Four days before culture preparation, each MEA was plasma cleaned for 3.5 minutes to increase the hydrophilic nature of the surface of the chip. One day before culture preparation, each well of the MEA chip was coated with

0.1% Polyethylineimine (PEI) (Sigma-Aldrich) for 1hr at room temperature then rinsed

Page 34 four times with distilled water and left under a UV light for 1 hour. The plate was then left to dry overnight. Prior to beginning the primary culture procedure, each well was coated with 2% (Sigma-Aldrich) for 2-3 hours and stored in the cell culture incubator at 37oC. Laminin was removed immediately prior to plating. Cell culture media was also prepared and stored at 4oC.

2.3.4 Primary Culture Procedure

C57BL/6 animals were taken at postnatal day 0-2 and decapitated with small surgical scissors without the use of anaesthetic. The brain was exposed by making a sagittal cut into the skull across the corpus callosum and peeling it back to expose the brain. The whole brain was removed and placed in a petri dish containing ice cold dissection solution (HBSS/HEPES) (Sigma-Aldrich). The cortex was isolated and the rest of the brain discarded. The tip of the animals tail was cut off at this point for genotyping. The cortex was chopped into 1-2mm pieces, transferred to a 50mL falcon tube and left to sit on ice in the dissection solution. The cortex was transferred to another 50mL falcon tube and dissociated in 0.25% Trypsin (Sigma-Aldrich), warmed to 37oC for 6 minutes, after which 500uL 0.0032% DNase (Sigma-Aldrich) diluted in culture media (described above) was added and the tube was placed again at 37oC for another four minutes.

The tube was then removed from the incubator and 5ml of culture media was added.

The cortical pieces were then centrifuged at 1500 rpm for 5 minutes at room temperature. Most the supernatant was aspirated, leaving 500uL of culture media remaining in the bottom of the tube. 250uL of DNase diluted in culture media was added and left to rest for 1 minute. 10mL of culture media was added and the cortices

Page 35 were mechanically triturated with a 1000uL Eppendorf pipette. The tube was centrifuged at 1500rpm for 5 minutes. The supernatant was aspirated and 6ml of culture media was added. Cortices were mechanically triturated for a second time and passed through a 40uM cell strainer (Corning), which had previously been wet with 1ml culture media into a clean 50mL falcon tube. The old tube was washed out with 3mL culture media to collect any cells left behind and this was then passed through the strainer.

10uL of the cells were reserved for counting while the remaining 10mL was centrifuged at 1500rpm for 5 minutes at room temperature. In the interim, cells were counted using an automated cell counter (Countess; ThermoFisher Scientific) and the laminin solution was aspirated off the surface of the MEAs. After cell counting, the supernatant was aspirated and cells were resuspended to a density of 3.125 x106cells/ml and 120uL,

90uL or 60uL of this suspension was plated into each MEA well, resulting in a final

2 2 2 density of 11,646 cells/mm , 8,734 cells/mm and 5,823 cells/mm respectively. Wells were topped up with culture media to make a total well volume of 120uL. Where possible, 12 wells of a 24-well MEA plate were plated using the cells from a single animal. MEA plates were placed back in the incubator at 37oC and after two hours, the media was removed and placed with 500uL Neurobasal-A medium (described above)

At DIV3, 5uM cytosine arabinoside (Merck KGaA) (5uM) was added to the media to prevent glial proliferation. This was removed at DIV5 and a complete media change was performed three times a week.

2.3.5 Network Activity Recordings

At DIV 14, 21, and 28, an 11 minute recording was taken using the software Multiwell

Screen from Multichannel systems. Each MEA plate was left to sit on the amplifier

Page 36 headstage for 5 minutes to adjust, and the plate was covered by a lid that allows carbogen (5% CO2, 95% O2) to pass over the plate to maintain the CO2 environment established in the incubator. Recordings were performed with a sampling rate of 20kHz and a digital voltage resolution of 24bit. After 11 minutes of recording, the plate was transferred back to the incubator. Full media changes were performed 24 hours prior to recordings so as not to artificially alter the baseline activity of the networks. At the completion of the experiment, cells were washed off the bottom of the plate with Terg-

A-Zyme and rinsed with water. They were then left to dry.

2.3.6 Raw Data Processing

Raw data was processed using methods previously described (Mendis et al. 2019;

Mendis et al. 2014). After the data had been recorded, the program Multichannel Data

Manager converted the files to H5 format for use in MATLAB. Electrode voltage were high-pass filtered at 300Hz following which spikes were detected based on precision timing. Spikes were detected if the highest phase of the spike was greater than

6 times the standard deviation of noise. The average noise level was calculated for each channel and channels with outlying noise levels were discarded. Noise was measured intermittently every two milliseconds. Noisy channels where the average noise levels were outliers compared to the noise of other channels and channels where the noise levels varied abnormally (Noisemax > 2Noisemin) were removed from analysis. Though spikes recorded from a single channel can originate from several neurons, spikes were not sorted into individual units.

Page 37

Bursts and Network Bursts (NBs) were detected using an adaptive algorithm based on firing rate (Mendis et al. 2016b; Mendis et al. 2019). Bursts on single channels were defined as rapid successions of 3 or more spikes with inter-spike intervals lower than a custom threshold that changes between networks and is based on the firing rate outside of bursting periods. NBs were detected where more than 20% of the channel-wise bursts overlapped in time. Network characteristics were extracted in terms of firing and bursting parameters (Fig 2.1).

Figure 2.1. Representation of network activity on a raster plot. Graphic that represents 20 seconds of activity of one network on a raster plot. a) Shows the spikes

(represented by one vertical line) and bursts (represented by multiple vertical lines), with the x axis representing time in seconds and the y axis representing 30 individual channels within one well. Each horizontal row represents one channel, or electrode, on the MEA. Each downward dash represents a spike, and a group of dashes together represents a burst. Network bursts (NBs) are depicted as a series of bursts lining up in the majority of channels and are labelled here, encapsulated by the red rectangles. The final feature represented is the network burst start interval (NBSI), which represents the

Page 38 time interval between the start of two consecutive network bursts. b) shows an expanded view of a burst, and highlights the burst core, a tightly clustered series of spikes, and the extended burst. In this example, the burst core lasts for 0.5 sec while the total burst lasts for 0.75 sec. c) shows an expanded network burst which is represented here as lasting for 1.5 sec. The time interval between the starts of bursts of the first and last bursting channels that participated in the NB is called Jitter.

Average amplitude of spikes and mean firing rates were calculated for each channel and were averaged across channels. Mean firing rates (MFR) were also calculated for bursting periods (MFRIn) and for non-bursting periods (MFROut). Burst features were calculated for both single-channel bursts as well as network bursts. Single channel burst parameters include the number of spikes inside bursts and burst durations. Network burst (NB) parameter features consist of NB durations, network burst start interval

(NBSI-time interval between the start of two consecutive NBs), and the jitter between channels participating in the NB (time interval between the starts of bursts of the first and last bursting channels that participated in the NB). The NB rate was also calculated, as well as Cohen’s Kappa, a measure of temporal synchronisation that spans from -1 (no synchronisation) to 1 (complete synchronisation) with a value of 0 indicating the amount of synchronisation expected by random chance (McHugh 2012). Kappa was obtained by dividing the time interval recorded for each channel into 10ms bins. A bin was marked as 1 if one or more spikes were detected in this duration. If no spikes were detected, a bin was marked as 0. Cohen’s Kappa was then calculated between all pairs of active electrodes within each well of the MEA to give a final value of synchrony between active electrodes. For each parameter the percentage change from baseline was

Page 39 calculated. Analysis of extracted parameters was done using Mann-Whitney U tests and

Kruskal-Wallis followed by a post-hoc Dunn’s test for multiple comparisons in

Graphpad Prism Version 8.

2.4 Results

Using the culturing and recording protocols and the analysis tools described above, the maturation and stability of primary cortical mouse neurons on mwMEA plates was assessed, as well as potential causes of variability. Primary cortical neurons derived from C57BL/6 animals at age P0-P2 were extracted and cultured on 24-well MEA plates for four weeks. Recordings were taken at DIV 14, 21 & 28.

The activity of the cortical cultures while developing was assessed to determine whether known characteristics of cortical networks were generated using the above protocol. An example of a culture at developmental day 14 can be observed via a raster plot (Fig

2.2). From the raster plot, spikes, bursts and network bursts were all detected in these cultures. This confirms that these cultures express the anticipated features of neuronal networks on MEAs previously reported in the literature (Cotterill et al. 2016).

Page 40

Figure 2.2. Raster plot of network activity at DIV 14. Representative raster plot of 60 seconds of activity from a C57BL/6 cortical culture at DIV 14. Each red bar on the x axis represents detection of a network burst. Spikes, bursts and network bursts can all be detected in this plot.

After establishing the viability of the cultures, potential sources of variability between each cultured network were investigated. Variability may arise from the preparation and seeding process. Two anticipated sources of variability were a) the discrepancy that occurs between recordings due to using different recording plates, and b) variability between different preparations. Preparations from four animals provided 9-12 samples each across two plates (Fig 2.3a). The variability between plates was observed for both

MFR and NB Rate using a Mann-Whitney U test. There was no significant difference in

MFR and NB Rate between the plate A and plate B samples for any of the four animals tested, confirming that plate-to-plate variability is not a factor to be concerned with when going forward (Fig 2.3d, 2.3e).

Page 41

There was a significant decrease in mean firing rate (Fig 2.3b) between networks derived from Mouse 1 compared to Mouse 2 (18.62 CI [15.57, 23.61], 7.33 CI[6.87,

8.47], p<.001) and a significant increase between Mouse 2 and Mouse 3 (7.33 CI[6.87,

8.47], 18.24 CI[16.19, 24.15], p<.001). There was also a significant decrease in NB rate

(Fig 2.3c) between Mouse 1 and Mouse 2 (1.41 CI[0.99, 1.77], 0.36 CI[0.29, 0.52], p<.001) and Mouse 1 and Mouse 3 (1.41 CI[0.99, 1.77], 0.65 CI[0.51, 0.82], p<.05), and a significant increase between Mouse 2 and Mouse 4 (0.36 CI[0.29, 0.52], 0.89

CI[0.70, 1.06], p<.01). A two-way ANOVA shows no significant interaction between plate and preparation variables for firing rate or network burst rate. This demonstrates the sensitivity of the assay to variability in culture preparation or individual animal variability. A power analysis determined that the animal numbers required were between n = 1 and n = 8 to account for the animal-to-animal variability. As each parameter investigated has a different degree of internal variability, each requires greater or fewer animals. This was addressed by ensuring samples were taken from multiple preparations to minimise noise.

Page 42

a) Plate A Plate B

Mouse 1 Mouse 2 Mouse 3 Mouse 4

Firing Rate Network Burst Rate b) c) ** * 40 **** 3 **** ****

30 c

c 2

e

e

s

s

/

/

s s

20 t

e

s

k

r

i

u

p b

s 1 10

0 0 1 2 3 4 1 2 3 4 Mouse Number Mouse Number

d) e) Network Burst Rate Firing Rate 3 40

30 c

c 2

e

e

s

/

s

/

s s

20 t

e

s r

k

i

u p

b 1 s 10

0 0 1 2 3 4 1 2 3 4 Mouse Number Mouse Number

Figure 2.3. Variability within and between cultured networks. Demonstration of

animal (b, c) and plate (d, e) based sources of variability within C57BL/6 primary

cultures. Recordings were taken at DIV 14.

Page 43

After determining potential sources of variability, the development of the networks was investigated. To classify the development of the C57BL/6 cultures, their growth was monitored over time. We observed the activity of the networks at DIV 14, 21 and 28.

The characteristics chosen as benchmarks to examine maturation of the networks were

MFR, NB Rate, and Cohen’s Kappa. The MFR indicates overall activity of the cultures, the NB Rate represents the degree to which the cultures are forming functional networks, and the kappa value is used to measure whether synchrony is occurring within each network. A Kappa value of 1 indicates total synchrony, and a value of 0 indicates the degree of synchrony expected through random chance. We performed a

Kruskal-Wallis test followed by a post-hoc Dunn’s test to compare the differences between cultures at DIV 14 (n = 513), 21 (n = 509) and 28 (n = 290).

a) Mean Firing Rate b) Network Burst Rate **** **** ****

15 * 1.0 c

c e

10 e

s

s

/

/

s

s t

e 0.5

s

k

r i

u

p s 5 B

0 0.0 14 21 28 14 21 28 Days in Culture Days in Culture

c) Kappa 0.5 0.4

a 0.3

p

p a

K 0.2

0.1

0.0 14 21 28 Days in Culture

Figure 2.4. Parameters from networks at 3 time-points. Comparisons of cultures at

Page 44

DIV 14, 21 and 28 for parameters describing a) mean firing rate, b) network bursting rate and c) Kappa.

The MFR of C57BL/6 networks (Fig 2.4a) are significantly higher at DIV14 than at

DIV 21 (12.65 CI[13.86, 15.37], 11.22 CI[12.01, 13.06], p<.05) and at DIV 28 (12.65

CI[13.86, 15.37], 10.64 CI[11.20, 12.87], p<.001). There is no significant difference between DIV21 and DIV28. The NB Rate (Fig 2.4b) is also significantly higher at DIV

14 than at DIV 21 (0.90 CI[0.89, 0.98], 0.41 CI[0.47, 0.54], p<.001) and at DIV 28

(0.90 CI[0.89, 0.98], 0.37 CI[0.42, 0.54], p<.001). There is no significant difference between DIV21 and DIV28. Finally, we found no significant differences in Kappa (Fig

2.4c) between DIV14, DIV21, and DIV28.

Overall, the networks appear to change the most significantly between DIV14 and

DIV21, with a significant decrease seen in both spiking and bursting activity. There is no significant difference between DIV21 and DIV28 for any of the parameters observed, which implies that the characteristics seen at DIV21 are retained to DIV28 and at this point the cultures are less likely to vary due to further growth and development. There is no significant difference between the Kappa values between

DIV 14, 21 or 28, indicating that synchronicity as measured by Kappa is not altered by maturation of the networks. It also indicates that the synchronicity measure is not influenced by the firing or bursting rate. The decrease in variability and apparent stabilisation at DIV21 provides the framework for a stable baseline with which to conduct further experiments.

Page 45

To further determine the effects of seeding density on the developed networks, preparations were seeded at three different densities: 50% of the original seeding density (187500 cells/well, 5,823 cells/mm2) (DIV 14 n = 87, DIV 21 n= 70, DIV 28 n

= 43), 75% of the original seeding density (281250 cells/well, 8,734 cells/mm2) (DIV

14 n = 90, DIV 12 n = 91, DIV 28 n = 79), and 100% of the original seeding density

(375000 cells/well, 11,646 cells/mm2) (DIV 14 n = 513, DIV 21 n = 509, DIV 28 n =

290). Comparisons were made using a Kruskal-Wallis followed by a post-hoc Dunn’s test with recordings taken at DIV 14, DIV 21 and DIV 28. To reduce preparation-based variability, samples from each preparation were seeded at each of the three densities.

a) Mean Firing Rate b) Network Burst Rate 25 **** 1.5 **** **** **** DIV14 20 DIV21 **** **** DIV28

c * ****

e 1.0 *

15 c

s

e

/

s

s

/

e

s

t

k

s

i r

p 10

u s B 0.5

5

0 0.0 50% 75% 100% 50% 75% 100% Density (% of original) Density (% of original) c) Kappa 0.5

0.4

0.3

a

p

p a

K 0.2

0.1

0.0 50% 75% 100%

Density (% of original)

Figure 2.5. Parameters of culture at different time-points and densities. Comparisons of cultures at DIV 14, 21 and 28 at 50% density, 75% density, and 100% density The

Page 46 parameters of interest describe a) mean firing rate, b) network bursting rate and c)

Kappa.

The MFR of C57BL/6 networks at 100% density (Fig 2.5a) are significantly higher at

DIV14 than at DIV 21 and DIV28. There were no significant differences between any of the time-points for the 50% density or 75% density networks. The NB Rate (Fig

2.5b) is significantly higher at DIV 14 than at DIV 21 for the 50% density networks

(1.07 CI[0.98, 1.27], 0.52 CI[0.51, 0.70], p<.001), the 75% density networks (1.33

CI[1.08, 1.32], 0.73 CI[0.69, 0.93], p<.001), and the 100% density networks (as previously described). The NB rate is significantly higher at DIV 14 than at DIV 28 for the 50% density networks (1.07 CI[0.98, 1.27], 0.43 CI[0.38, 0.56], p<.001), the 75% density networks (1.33 CI[1.08, 1.32], 0.55 CI[0.47, 0.62], p<.001), and the 100% density networks (described above). The NB rate at DIV 21 is also significantly higher than at DIV 28 in the 75% density networks (0.73 CI[0.69, 0.93], 0.55 CI[0.47, 0.62], p<.05). There is no significant difference between DIV 21 and DIV 28 in the 50% density and 100% density networks. There are also no significant differences in Kappa values between any time-points within any cell density (Fig 2.5c).

Indicative criteria for detecting a stable baseline is a lack of significant change between each parameter at each time-point. Additionally, a moderate Kappa level is preferred as this indicates a well-connected network that does not succumb to floor or ceiling effects when investigating drug effects. If the Kappa level is too low, inhibitory drug effects will not be properly detected due to the floor effect of having non-synchronous baseline activity. Alternatively, if the Kappa level is too high, excitatory drug effects will be

Page 47 difficult to detect as there will be a ceiling effect resulting from the baseline having achieved maximum synchrony before compound application. Overall, there was no significant difference in MFR between any of the three time-points at 50% and 75% seeding density. At 100% seeding density, there is a significant difference between

MFR between DIV 14 and DIV 21, but no difference between DIV 21 and DIV 28, indicating that the network has not yet stabilised at DIV 14 but has stabilised by DIV

21. There are significant differences between DIV 14 and both DIV 21 and DIV 28 at all densities, implying that the networks are not yet stabilised at DIV 14. At 75% density, there is also a significant difference between DIV 21 and DIV 28, indicating that the network has not yet stabilised at DIV 21 and it is unknown whether it has stabilised by DIV 28. For 50% and 100% density, there is no significant difference between DIV 21 and DIV 28, indicating again that the network has stabilised at DIV 21 for this parameter. There are no significant differences in Kappa value between each time-point at any density. However, when further compared between densities, there is a significant increase between Kappa values at all time-points from 50% density to 75% density, and then again from 75% density to 100% density. This implies that an increase in density leads to a significant increase in synchronicity within the cultures.

2.5 Discussion

The aims of this study were to a) develop a workflow that produced active, viable primary cultured networks onto mwMEA plates, b) address potential sources of variability that arise when developing a large-scale screening system, and c) classify the baseline characteristics of these cultures to determine which conditions lead to networks most optimised to be the basis for further development of a high-throughput assay for

Page 48 diagnosis and drug discovery. It was first determined that the cultures were viable through observation of the network activity represented by the spikes, bursts, and network bursts detected in the raster plots. Potential sources of variability that may arise between different plates and different preparations/animals used were investigated, and their impact on the activity of the cultured networks was addressed. It was found that there was no variability originating from the use of different plates when using neurons from the same preparation. This is demonstrated by the lack of significant differences when comparing activity generated from networks sourced from the same preparation across plates A and B for both MFR and NB rate (Fig 2.3b, 2.3c). However, there are significant differences between MFR and NB rate when comparing samples from different preparations (Fig 2.3d, 2.3e). This signifies that while between-plate variability is not an issue that needs to be addressed, between-preparation variability is a significant confounder we need to factor into further analysis. This has been reported also by Biffi et al. (2013) in swMEAs and Cotterill et al. (2016) in mwMEAs. Cotterill also noted that culture-to-culture variability was much greater than plate-to-plate variability, which mirrors what we detected. This variability can be mitigated by sampling from wells from a variety of different preparations, rather than conducting all experiments with the samples from one preparation.

The development of cultures over time was observed by recording at DIV 14, DIV 21 and DIV 28. Both MFR and NB Rate were significantly higher at DIV 14 than at DIV

21 or DIV 28, and that there was no significant difference for both parameters between

DIV 21 and DIV 28. This signifies a stability of the networks at DIV 21 as this is the first time-point at which no further change is detected. No significant change in Kappa

Page 49 value between the three time-points was detected, which may indicate that the Kappa value, and therefore overall synchrony of the networks, stabilise prior to DIV 14. For the purposes of establishing which timepoint provides a stable baseline for further experimentation, DIV 21 was chosen as the most ideal time-point, as it is the earliest recorded time-point at which the networks stabilise, meaning no further change is detected. Stabilization of cultured mouse neuronal networks on MEAs at DIV 21 has also been reported by Charlesworth et al (2016) who reported an increase in MFR and burst rate over time until stabilisation at 21 days, at which point the networks show highly synchronised activity. This was also seen in embryonic rat hippocampal neurons on MEAs, which demonstrate an increase in burst activity during the first three weeks before stabilising (Habibey et al. 2017). While DIV 28 demonstrates the same activity level, there is a greater risk of the cultures aging and detaching from the chip. Thus, it is best to choose the point at which the cultures first stabilise. Finally, while there was no difference in Kappa values in in any seeding density at any time-point, the Kappa values at density 11,646 cells/mm2 were the highest of all three densities, signifying the greatest degree of synchrony within the networks. However this value was not so high that an increase in synchrony would be undetectable. As synchrony is an important signifier of mature networks, the ideal culturing and recording parameters are those that produce networks with the most synchrony to allow for more detailed observations about changes in synchrony in future studies (Jewett et al. 2016). Based on these observations, the parameters that produce networks that stabilize the earliest and have the highest recorded synchrony measure are networks seeded at 11,646 cells/mm2, developed to DIV 21. Further experiments will use these parameters as the baseline for network development.

Page 50

When comparing the baseline activity profile of these networks to the activity profiles obtained by other groups, both similarities and differences can be seen. A study by Lu et al. (2015) measured baseline activity of cortical cultures derived from mice at P0 and seeded at a density of 3x103 cells/mm2, identifying DIV 21 as the ideal time-point to use as a baseline for further analysis. They reported a MFR of approximately 2 spikes/second and NB Rate of approximately 0.06 bursts/second at DIV 14, and a MFR of approximately 4.2 spikes/second and NB rate of 0.14 bursts/second at DIV 21. This is a much lower MFR and NB rate than reported here. The reason for this difference is most likely seeding density, as the seeding densities used in our study are approximately

4x greater than those used by Lu and colleagues. Another study utilizing mouse cortical cultures on MEAs conducted by McSweeney et al. (2016) reported a seeding density of

3.33 x104 cells/mm2. This density is approximately 3x greater than the densities used to determine our baseline activity. However, McSweeney reported a MFR of approximately 1.8 spikes/second and an NB rate of 4 busts/minute at DIV 14, and a

MFR of approximately 3.5 spikes/second and NB rate of 6 bursts/minute at DIV 19

(DIV 21 was not reported). This is far less than we have reported, despite the seeding density being greater. This may be due to viability factors (for example, more successful adhesion of cells in our cultures), or differences in recording and spike detection.

McSweeney and colleagues reports that a burst must have a minimum of 8 spikes, whereas our threshold is 3 spikes. Further studies describe comparisons between laboratories and report major discrepancies in MFR and NB Rate across cell culturing techniques, recording, and analysis parameters (Vassallo et al. 2017). Another difference seen across laboratory groups is the parameters reported. For instance, there are many measures of network connectivity other than Cohen’s kappa that can be useful

Page 51 in MEA network analysis. Open source software package SpyCoDyn utilises multiple measures of synchrony such as cross-correlation, partial correlation, joint entropy, and transfer entropy (Pastore et al. 2018). Each of these approaches measure a different aspect of network connectivity. Cross-correlation can be utilised to determine a link between two electrodes as a function of time, while partial-correlation expands on this by differentiating between direct and indirect connections (Poli et al. 2016). Joint entropy is a linear measure using inter-spike intervals to determine the likelihood that a pair of neurons are connected and may be used to infer causality (Garofalo et al. 2009).

Transfer entropy allows for the identification of causal relationships based on a time- series by taking into account and estimating the part of neuron activity that relies on the activity of a connected neuron (Garofalo et al. 2009; Pastore et al. 2016). The use of these parameters will depend on the software available and the complexity of the required analysis. This demonstrates that each laboratory utilizing MEA techniques will have varying baselines and may use different parameters in their reporting, which stresses the importance of establishing an internal baseline within each laboratory group, as we have done here. Further experimentation will involve utilizing this protocol with the optimised seeding density and recording DIV to generate disease models and conduct high throughput drug screening at a network level.

2.6 Conclusion

In this study a workflow for producing stable and replicable networks on mwMEAs has successfully been developed. Neurons taken from postnatal mouse pups plated at a high density will stabilize for use as the baseline for further experimentation at DIV21. It was confirmed that variability between preparations significantly impacts on network

Page 52 development, whereas variability between plates has no effect. The ability to establish a mature neural network model in 21 days for long- and short-term experimentation highlights the high-throughput nature of this platform. Further experiments will utilize this workflow and culturing parameters to generate genetic disease models and perform high throughput drug screening.

Page 53

CHAPTER 3 | DEVELOPING A NETWORK MODEL OF DRAVET SYNDROME ON MULTIELECTRODE ARRAYS

3.1 Abstract

Introduction: DS expression is heavily affected by genetic background, as patients with complete loss of function mutations express symptoms along the spectrum of mild to severe, and a varied combination of specific symptoms. Therefore, when modelling DS, the platform used must be able to detect differences in phenotype that may occur as a result of genetic background. In two mouse models with different genetic backgrounds,

SCN1A haploinsufficiency results in two different phenotypes. In a C57BL/6 background, SCN1A haploinsufficiency generates a model with spontaneous seizures and early mortality, as seen in human DS. In the SV129 background, SCN1A haploinsufficiency does not result in spontaneous seizures, generating a model that expresses no spontaneous phenotype. In this chapter, I aim to explore the ability to capture a spontaneous seizure phenotype in networks derived from a mouse model of

DS, and compare this activity to that of networks derived from a mouse model with no expected DS phenotype.

Aim: To model DS on MEAs using cortical neurons from two mouse models with

Nav1.1 haploinsufficiency. These models have a C57BL/6 background or SV129 background. This will determine MEA sensitivity to detecting differences in phenotype, replicating what is seen in human patients with variable penetrance.

Page 54

Method: Cortical neurons from SCN1A+/-/C57BL/6 and SCN1A+/-/SV129 animals were cultured on MEAs and spiking, bursting and network connectivity features were extracted. Parameters were compared using Mann-Whitney U tests.

Results: SCN1A+/-/C57BL/6 networks showed significantly increased spiking, decreased burst rate and increased burst size, and increased time between network bursts. The SCN1A+/-/SV129 networks showed significantly decreased bursting and few other significant differences. These networks were significantly different from each other when change from baseline was compared.

Conclusion: MEAs successfully detected an epileptic phenotype in SCN1A+/-/C57BL/6 networks through feature analysis and comparison to known epilepsy networks. This phenotype was not observed in the SCN1A+/-/SV129 networks. Additionally, these networks were significantly different from each other when change from baseline was compared, indicating that it would be possible to significantly separate a mild from severe DS phenotype using this system.

Key Words

MEA, Dravet Syndrome, Genetic Modelling, SCN1A, Nav1.1, Sodium Channel

Abbreviations

DS – Dravet Syndrome

ADNFLE - Autosomal dominant nocturnal frontal lobe epilepsy

MEA – Multielectrode Array

Page 55

MFR – Mean Firing Rate

NB – Network Burst

MFRIn – Mean Firing Rate Inside Bursts

MFROut – Mean Firing Rate Outside Bursts

NBSI – Network Burst Start Interval

DIV – Days in vitro

3.2 Introduction

Modelling genetic epilepsies has traditionally been achieved using animal models. The use of genetically altered rodents as models of human disease has been utilized both for determining disease pathogenesis as well as for studying therapeutic interventions.

Mouse models of DS with Nav1.1 mutations have been shown to accurately model the human condition, showing temperature-dependent and age-dependent onset, along with progression to myoclonic and generalized tonic-clonic seizures (Oakley et al. 2009).

These rodents are sourced from different genetic lines or strains that have been inbred to express certain characteristics. These different characteristics can modulate the phenotype of a single-gene mutation, resulting in a severe or mild disease expression.

This effect of genetic background on gene expression varies depending on the mutation observed. DS expression is heavily affected by genetic background, as patients with complete loss of function mutations express symptoms along the spectrum of mild to severe, and a varied combination of specific symptoms (Dravet et al. 2005). As the mutation in each of these cases has completely neutralized the function of the protein, genetic background is the only variable that can influence the expression of the

Page 56 disorder. Therefore, when studying DS in rodent models, the effect of genetic background must be considered (Frankel 2009; Schauwecker 2011).

One common genetic background used to model epilepsy is the C57BL/6 mouse strain.

This strain is traditionally used as a vehicle for epilepsy modeling when studying effects of genetic background as it is reported to be highly resistant to seizures (Ferraro et al.

2002; Tan et al. 2007). However, other studies have reported that the C57BL/6 strain is highly susceptible to febrile seizures, demonstrating the variability of seizure models

(Van Gassen et al. 2008). Another strain traditionally utilized in epilepsy modelling is

129S1/SvImJ (referred to as SV129 in this thesis). Embryonic stem cells from SV129 strains are the most commonly used for gene targeting, resulting in this strain being highly prevalent in epilepsy research. Less is known about the seizure susceptibility of the SV129 strain as it is commonly crossed with C57BL/6 (Rivera and Tessarollo

2008).

Yu et al. (2006) studied the effect of genetic background on seizure susceptibility caused by the DS mutation by ablating the SCN1A gene in an embryonic mouse stem cell and generating two different mouse lines from the stem cell clones. One line was derived from the 129/SvJ background and generated Nav1.1-129sv mice and the other from the C57BL/6 background to generate Nav1.1-B6 mice. These mice were then monitored to determine the severity of their phenotype. At P9, it was determined by immunoblots that there was a decrease in expression of Nav1.1 protein expression compared to wild-type by 50% in heterozygous animals and 100% in homozygous animals in both strains. Additionally, homozygous knockout Nav1.1 animals exhibited

Page 57 ataxia and seizures beginning at P9, and none survived beyond P15 regardless of genetic background. Heterozygous animals however, were found to have different phenotypic expression, seemingly as a result of genetic background. SCN1A heterozygous animals maintained in the 129/SvJ background had a survival rate nearly the same as WT at 15 weeks (90%), while those maintained in a C57BL/6 background had a survival rate of 20% at 13 weeks. Additionally, no behavioral seizures were recorded in mice maintained in the SV129 background despite being highly prevalent in those maintained in the C57BL/6 background. It was concluded that haploinsufficiency of the Nav 1.1 protein alone was not sufficient to cause a notable phenotype, and that genetic background could influence whether an individual was more or less resistant to seizures caused by DS (Yu et al. 2006; Miller et al. 2014).

The effects of the DS mutation can also be seen at the intracellular level in animal models. It is widely supported that DS is caused by the loss of function of the Nav1.1 subunit in, specifically, parvalbumin-positive interneurons. Selective deletion of SCN1A in mouse interneurons result in reliable expression of DS symptoms and related comorbidities at the neuronal level (Cheah et al. 2012; Han et al. 2012; Ogiwara et al.

2013). However, the same SCN1A deletion in excitatory neurons causes no such effects

(Dutton et al. 2013). In addition, it has been shown that not only does the ablation of

SCN1A result in the expression of DS symptoms at a neuronal level, the degree of expression is modulated by genetic background. This mirrors what is seen in the whole animal model. Rubinstein et al. (2015) performed a series of experiments comparing the activity of DS and WT neurons in SV129 and C57BL/6 genetic backgrounds. It was reported that the SCN1A+/- interneurons in the C57BL/6 background demonstrated a

Page 58 significant reduction in action potential firing capability compared to WT neurons.

SCN1A+/- interneurons in the SV129 background also demonstrated a reduction in AP firing capability compared to WT, however the reduction was more significantly pronounced in the C57BL/6 background. This suggests that the degree of severity seen in DS animals of C57BL/6 and SV129 backgrounds can be recapitulated at a neuronal level. It also suggests that the expression of DS in humans can be replicated in animal models and then further at the cellular level.

While these differences are observable at a neuronal level, they have not yet been observed at a neural network level. However, other epileptic networks have been studied at this level, giving some insight into the phenotype of genetic epilepsies on

MEAs. Gullo et al. (2014) reported increased burst duration in cultured primary mouse neurons modelling ADNFLE on MEAs. They also reported a decrease in network burst rate, an increase in the number of spikes within bursts in ADNFLE cultures compared to

WT, and an increase of the inter-burst intervals by more than 50% in ADNFLE cultures from WT. A febrile seizure model has also been tested on the MEA platform using an

SCN1A mouse carrying the NaV1.1-R1648H mutation. It was reported that the disease networks showed significantly increased burst duration, significantly fewer network bursts, and a significantly longer inter-network burst interval. While the NaV1.1-

R1648H is not a DS mutation, this confirms that it is possible to observe the effect of

SCN1A mutations on the MEA platform (Hedrich et al. 2014).

Additionally, while not an epilepsy mutation, the development of GABAergic parvalbumin-positive interneurons (the site of Nav1.1 disfunction caused by the SCN1A

Page 59 mutation) on MEAs has been investigated to determine the effect of these neurons on developing and mature neuronal networks. It is widely accepted that GABAergic interneurons contribute significantly to the generation of rhythmic network oscillations in the brain during development, as well as precise regulation of the timing of action potentials. As cortical neurons mature their activity shifts from highly synchronous networks with little influence from GABA, to a pattern driven by GABA which exhibits higher frequency and lower synchrony than in their immature state. This is referred to as the GABAergic shift. (Ben-Ari et al. 2007; Moody and Bosma 2005; Garaschuk et al.

2000; Allene et al. 2008).

The impact of different GABAergic interneurons and the degree to which the

GABAergic shift regulates cultured network activity on MEAs was investigated by

Baltz et al. (2010). Neocortical primary rat neurons from the dorsal and lateral cortical regions were grown on MEAs and monitored until DIV21. Lateral cortical derivates demonstrated an increase in spike and burst frequency as the networks aged, with the bursts becoming less synchronous and shortening over time. After adding a GABAA receptor blocker, the spike frequency, burst duration and numbers of electrodes involved in burst detection was increased, while the burst frequency and variation between the inter-burst intervals was decreased. This indicates that GABAergic interneurons have a significant effect on the firing patterns of cultured cortical neurons.

In these networks, two distinct GABAergic neuron populations were detected, with 50% being large parvalbumin positive interneurons. In the dorsal cortical derivative networks that did not contain parvalbumin-positive interneurons, or in cultures where GABAA receptors were blocked by antagonists, this pattern was not observed. This

Page 60 demonstrates that parvalbumin-positive neurons have a detectable impact on dissociated cortical networks derived from the lateral cortex, and that this impact is to generate a pattern of activity more diverse than the highly synchronous immature developing network, and that their inhibition has a similar effect to GABAA blockers in vitro.

Other means of observing the effects of GABAergic inhibition on networks reveal a similar pattern. Computational modelling of developing neural networks reveals that chronic blocking of GABAA receptors generates highly synchronous networks characterized by homogenous bursts with increased burst duration (Baltz et al. 2011).

Electrical stimulation of networks with and without GABAA further reveal that

GABAergic interneurons regulate gradual recovery of the networks within milliseconds of a stimulus, whereas networks with blocked GABA recover abruptly after approximately 5 seconds. These observations correspond to what is seen in neural networks during the GABAergic shift, highlighting the importance of GABAergic interneurons in developing mature networks. As parvalbumin-positive interneurons are heavily implicated in the GABAergic shift that drives spiking and bursting patterns observed in mature cultures, their inhibition may result in cultured networks expressing an immature pattern of high frequency, highly synchronous activity. (Baltz and Voigt

2015; Reimer et al. 2012). As the SCN1A mutation that causes DS inhibits parvalbumin positive interneuron functioning, the effects of the SCN1A mutation on cortical cultures on MEAs may be similar to what is observed in these prior studies utilizing GABA inhibition. The overall aim of this study is to see if the effects of SCN1A haploinsufficiency induced by the R1407X mutation (Ogiwara et al. 2007) can be detected at a neuronal network level, using MEA technology.

Page 61

Aim: To determine whether differences from WT can be detected in the heterozygous

SCN1A+/- networks, and whether any differences observed are significantly different or similar to each other. This is so that the degree of difference between the heterozygous networks and the WT counterparts for each genetic background can be measured to determine the relative severity.

Hypothesis 1: A significantly different activity profile between SV129 and C67/B6

SCN1A+/- networks will be detected as these backgrounds express the DS phenotype at different intensities in vivo and in vitro, with the C57BL/6 strain showing a significant phenotype at both a whole animal and cellular level, and the SV129 strain showing no phenotype at the whole animal level and a vastly reduced phenotype at the cellular level.

Hypothesis 2: A significant increase in firing rate in heterozygous networks will be detected due to increased inhibition of inhibitory interneurons.

Hypothesis 3: A similar pattern to the ADNFLE and febrile seizure cultures will be detected as each are models of epileptic networks, as well as high firing rate due to the reduced function of parvalbumin positive interneurons. Therefore, we anticipate longer bursts, increased spikes within bursts, a decreased network burst rate, and an increase in inter-network burst start interval in SCN1A+/- cultures compared to WT. It was expected that this increase will be significantly greater in the C57BL/6 networks as opposed to the SV129 networks to correspond with previous findings reported in the literature.

Page 62

3.3 Method

3.3.1. Animals

SCN1A+/-/SV129 and SCN1A+/-/C57BL/6 mice were maintained in house at the Florey

Institute animal breeding facility. Litters were taken from the mother from 0-2 days after birth. All animal experiments were approved by the Howard Florey Institute

Animal Ethics Committee (AEC#14-075) and performed in accordance with the

Prevention of Cruelty to Animals Act and the NHMRC Australian Code of Practice for the Care and Use of Animals for Scientific Purposes. SCN1A+/- animals were sourced from the Yamakawa group and contain the truncation mutation R1407X (RX). RX

(heterozygote) mice were maintained on 129S1/SvImJ (inbred strain from Jackson

Laboratories Stock No: 002448); RX heterozygote mice on this background do not have an overt phenotype and normal survival. Male RX heterozygote (129S1/SvImJ) mice were crossed with female C57Bl/6 (inbred strain obtained from Animal Resources

Centre, W.A.) to generate RX heterozygote F1(129S1/SvImJ/ C57Bl/6J) mice; F1 heterozygote RX mice display spontaneous seizures (from approximately P18) and premature death.” (Ogiwara et al. 2007).

3.3.2 Media Preparation

See 2.3.2

3.3.3 MEA Preparation

See 2.3.3

Page 63

3.3.4 Primary Culture Procedure

See 2.4.3

3.3.5 Raw Data Processing

See 2.3.5

3.4 Results

Using the culturing and recording protocols and the analysis tools described above we observed the characteristics of SCN1A+/- cultures derived from SV129 and B6 genetic backgrounds. Recordings were taken at DIV 21. The first background tested was

C57BL/6 as we anticipated we would see a clear phenotype based on previous descriptions in the literature. Primary cortical neurons from SCN1A+/-/C57BL/6 animals at ages P0-P2 were extracted and cultured on 24-well MEA plates for three weeks.

Recordings were taken at DIV 21. A Mann-Whitney U test was used to determine the differences between SCN1A+/- (n = 129 (13 animals from 5 litters)) and WT (n = 107

(11 animals from 5 litters)) networks. A power analysis was performed to detect effect size and it was found that D = 0.38. We analysed three categories of parameters to assess total network activity. There are four parameters that represent single spike activity in the cultures (Fig 3.2), three parameters that represent bursting activity (Fig

3.3) and two parameters that represent network connectivity (Fig 3.4).

Page 64

Figure 3.1. Raster plots representing 30sec of activity from SCN1A+/-C57BL/6 networks. Two raster plots showing a representative 30sec of activity from a) WT and b) SCN1A+/- networks on a C57BL/6 background.

From the raster plots, it can be seen that the SCN1A+/- networks (Fig 3.1a) show fewer network bursts and an increased number of spikes within bursts compared to the WT networks (Fig 3.1b). Parameter extraction and analysis was required to further investigate these properties.

Page 65

a) Mean Firing Rate b) Mean Firing Rate Within Burst

20 150 ** ***

15 c

c 100

e

e

s

s

/

/

s

s e

10 e

k

k

i

i

p

p s s 50 5

0 0 WT SCN1A+/- WT SCN1A+/- C57BL/6 C57BL/6

c) Mean Firing Rate Outside Burst d) Average Spike Amplitude

5 40

4

d 30

n o

c 3

e

V s

/ 20

u s

e 2

k

i p

s 10 1

0 0 WT SCN1A+/- WT SCN1A+/- C57BL/6 C57BL/6

Figure 3.2. Spiking parameters in SCN1A+/-/C57BL/6 networks Four parameters describing firing activity from WT and SCN1A+/- cell networks in the C57BL/6 genetic background. These parameters are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average amplitude of spikes. The recordings were taken at DIV 21.

Page 66

In the C57BL/6 background, SCN1A+/- networks have a higher mean firing rate (Fig

3.2a) than WT (13.96Hz [CI 13.11, 15.65], 11.78 Hz [CI 10.62, 12.86], p<.01).

SCN1A+/- networks also have a higher mean firing rate within bursts (Fig 3.2b) than WT networks (112.1 Hz [CI 107.7, 116.1], 96.97 Hz [CI 86.15, 109], p<.01). The mean firing rate outside of bursts and average amplitudes of spikes (Fig 3.2c, 3.2d) are not significantly different between SCN1A+/- and WT.

a) Network Burst Rate b) Mean Single Channel Burst Duration

0.8 150 **

d 0.6 n

o 100

c

e

s s

/ 0.4

m

s

t s

r 50 u

b 0.2

0.0 0 WT SCN1A+/- WT SCN1A+/- C57BL/6 C57BL/6

c) Mean Single Channel Burst Size

30 *

s 20

e

k

i

p

s

. o

n 10

0 WT SCN1A+/- C57BL/6

Figure 3.3. Bursting parameters in SCN1A+/-/C57BL/6 networks. Three parameters describing bursting activity from WT and SCN1A+/- cell networks in the C57BL/6 genetic background. These parameters are a) network burst rate b) mean single

Page 67 channel burst durations and c) mean single channel burst size. The recordings were taken at DIV 21.

The network burst rate (Fig 3.3a) of SCN1A+/- networks in the C57BL/6 background is significantly lower than WT (0.34 Hz [CI 0.31, 0.39], 0.49 Hz [CI 0.39, 0.57], p < .01).

The mean single channel burst duration (Fig 3.3b) shows no significant difference between WT and SCN1A+/- networks. The mean single channel burst size (Fig 3.3c) is significantly less in WT networks than in SCN1A+/- networks (17.11 [CI 16.26, 19.67],

21.80 [CI 18.69, 23.48], p<.05). a) Mean Network Burst Start Interval b) Mean Jitter 4 ** 150

3 ****

100

c s

e 2

s m

50 1

0 0 WT SCN1A+/- WT SCN1A+/- C57BL/6 C57BL/6

Figure 3.4. Connectivity parameters in SCN1A+/-/C57BL/6 networks. Two parameters describing network synchrony from WT and SCN1A+/- cell networks in the C57BL/6 genetic background. These parameters are a) mean network burst start interval and b) mean jitter. The recordings were taken at DIV 21.

The mean network burst start interval (Fig 3.4a) of SCN1A+/- networks in the C57BL/6 background is significantly more than WT (2.92sec [CI 2.49, 3.20], 2.03sec [CI 1.74,

Page 68

2.57], p<.01). The mean jitter (Fig 3.4b) in SCN1A+/- networks was also significantly higher than WT (90.87ms [CI 85.96, 97.42], 69.98ms [CI 64.13, 83.46], p<.0001).

It appears that SCN1A+/- networks in a C57BL/6 background demonstrate increased firing compared to WT networks, with significantly more of this firing occurring within bursts. SCN1A+/- networks have a significantly lower burst rate and larger burst size than seen in WT networks. Finally, at the network level there is a significantly longer interval between bursts and the next spiking activity, and significantly increased jitter in

SCN1A+/- networks compared to WT cultures.

To determine whether a phenotype caused by the SCN1A+/- can be observed in an

SV129 background, primary cortical neurons from SCN1A+/-/ SV129 animals at ages

P0-P2 were extracted and cultured on 24-well MEA plates for three weeks. Recordings were taken at DIV 21. A Mann-Whitney U test was used to determine the differences at

DIV 21 (n WT = 86 (12 animals from 6 litters); n SCN1A +/- SV129 = 110 (13 animals from 6 litters). A power analysis was performed to detect effect size and it was found that D = 0.41. These parameters assessed are the same as described for the C57BL/6 background.

Page 69

Figure 3.5. Raster plots representing 30sec of activity from SCN1A+/-/SV129 networks. Two raster plots showing a representative 30sec of activity from a) WT and b) SCN1A+/- networks on a SV129 background.

From the raster plots, it can be seen that the SCN1A+/- networks (Fig 3.5a) show fewer network bursts and a reduced overall firing rate compared to the WT networks (Fig

3.5b). Parameter extraction and analysis was required to further investigate these properties.

Page 70

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts

20 **** 150 * d

d 15

n

n o

o 100

c

c

e

e

s

s /

/ 10

s

s

e

e

k

k i

i 50

p

p s s 5

0 0 WT SCN1A+/- WT SCN1A+/- SV129 SV129 c) Mean Firing Rate Outside Bursts d) Average Spike Amplitude

3 40

d 30 n

o 2

c

e

V s

/ 20

u

s

e k

i 1 p

s 10

0 0 WT SCN1A+/- WT SCN1A+/- SV129 SV129

Figure 3.6. Spiking parameters in SCN1A+/-/SV129 networks. Four parameters describing firing activity from WT and SCN1A+/- cell networks in the SV129 genetic background. These parameters are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average amplitude of spikes. The recordings were taken at DIV 21.

The mean firing rate (Fig 3.6a) of SCN1A networks in SV129 background is significantly lower than WT (12.77 Hz [CI 10.93, 13.45], 16.02 Hz [CI 14.53, 18.11], p<.001). The mean firing rate inside bursts (Fig 3.6b) is also significantly lower in

SCN1A+/- networks than in WT (117.1 Hz [CI 108.8, 124.2], 109.1 Hz [CI 101.9, 116.8],

Page 71 p < .05). There is no difference in mean firing rate outside bursts or average spike amplitude between SCN1A+/- and WT cultures (Fig 3.6c, 3.6d).

a) Network Burst Rate b) Mean Single Channel Burst Duration

1.0 150

0.8 * d

n 100 o

c 0.6

e

s

s

/

m s

t 0.4 s

r 50 u b 0.2

0.0 0 WT WT SCN1A+/- SCN1A+/- SV129 SV129

c) Mean Single Channel Burst Size

30

. 20

o

n

e

k

i p

s 10

0 WT SCN1A+/- SV129

Figure 3.7. Bursting parameters in SCN1A+/-/SV129 networks. Three parameters describing bursting activity from WT and SCN1A+/- cell networks in the SV129 genetic background. These parameters are a) network burst rate b) mean single channel burst durations and c) mean single channel burst size. The recordings were taken at DIV 21.

The network burst rate (Fig 3.7a) of SCN1A networks in SV129 background is significantly lower than WT (0.29 Hz [CI 0.26, 0.50], 0.41 Hz [CI 0.37, 0.50], p <

Page 72

0.05). The mean single channel burst duration and mean single channel burst size are not significantly different between WT and SCN1A+/- networks (Fig 3.7b. 3.7c).

b) Mean Jitter a) Mean Network Burst Start Interval 150 5 * 4 100

3

s

c

e

m s 2 50

1

0 0 WT SCN1A+/- WT SCN1A+/- SV129 SV129

Figure 3.8. Connectivity parameters in SCN1A+/-/SV129 networks. Two parameters describing network connectivity from WT and SCN1A+/- cell networks in SV129 genetic background. These parameters are a) mean network burst start interval and b) mean jitter. These recordings were taken at DIV 21.

The mean network burst start interval (Fig 3.8a) of SCN1A+/- networks in SV129 background is significantly higher than WT (3.37sec [CI 1.98, 3.83], 2.43sec [CI 2.00,

2.68], p<.05). The mean jitter (Fig 3.8b) shows no significant difference between

SCN1A+/- and WT networks.

Overall, it appears that SCN1A+/- networks in the SV129 background have a significantly lower mean firing rate and less spikes within bursts than in the WT networks. There is also significantly less bursting activity, and a greater length of time between bursting and spiking activity than seen in WT cultures. When comparing this to the activity of SCN1A+/- and WT cultures in the C57BL/6 backgrounds, there are some Page 73 clear differences, such as an opposite effect of the SCN1A+/- mutation on spiking activity. Knowing this, we compared the ways in which the SCN1A+/- genotype interacted with the genetic background in both SV129 and C57BL/6 networks. This was to determine whether the effect of the SCN1A+/- mutation on genetic backgrounds is quantifiable between backgrounds, as well as to determine whether the MEA technique is a viable method for detecting different severity of epilepsy types. We have done this by obtaining the average of all WT values and dividing each SCN1A+/- value by this number. Standard deviation (SD) was calculated by pooling the variance from both WT and SCN1A+/- samples. This was done separately for each background. The parameters used are the same ones used for the previous comparisons. The sample sizes are the same as stated previously for SV129 and C57BL/6 backgrounds. Welch’s t-test was used to identify significant differences and the mean and SD are reported below.

Page 74

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts 2.0 1.5 ****

****

e

e

n

n i

1.5 i

l

l e

e 1.0

s

s

a

a

B

B

1.0

o

o

t

t

o

o i

i 0.5

t

t a

0.5 a

R R

0.0 0.0 C57BL/6 SV129 C57BL/6 SV129

c) Mean Firing Rate Outside Bursts d) Average Spike Amplitude

2.5 2.0 e

2.0 e

n

n i

i 1.5

l

l

e

e s

1.5 s

a

a

b

B

1.0

o

o

t

t

1.0

o

o

i

i

t

t a

a 0.5 R R 0.5

0.0 0.0 C57BL/6 SV129 C57BL/6 SV129

Figure 3.9. Spiking parameter ratios between SCN1A+/- networks in C57BL/6 and

SV129 backgrounds. Four parameters describing firing activity through the ratios to baseline in C57BL/6 and SV129 networks. These parameters are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average amplitude of spikes. The recordings were taken at DIV 21.

The ratio between SCN1A+/- and WT cultures are significantly different for mean firing rate (Fig 3.9a), with the ratio in the SV129 background being significantly lower than what is seen in the C57BL/6 background (0.77±0.47, 1.12±0.48, p<.0001). The ratio between SCN1A+/- and WT cultures is also significantly different for mean firing rate inside bursts (Fig 3.9b), with the ratios in the SV129 background being significantly less than in the C57BL/6 background (0.92±0.26, 1.12±0.30, p<.0001). There are no

Page 75 significant differences between mean firing rate outside bursts (Fig 3.9c) and average spike amplitude (Fig 3.9c).

a) Network Burst Rate b) Mean Single Channel Burst Duration

2.5 2.0 e

e 2.0

n

n i

i 1.5

l

l

e

e s

s 1.5

a

a

B

B

1.0

o

o

t

t

1.0

o

o

i

i

t

t a

a 0.5 R R 0.5

0.0 0.0 C57BL/6 SV129 C57BL/6 SV129 c) Mean Single Channel Burst Size

2.0

e n

i 1.5

l

e

s

a B

1.0

o

t

o

i t

a 0.5 R

0.0 C57BL/6 SV129

Figure 3.10. Bursting parameter ratios between SCN1A+/- networks in C57BL/6 and

SV129 backgrounds. Three parameters describing bursting activity through the ratio to baseline in C57BL/6 and SV129 networks. These parameters are a) network burst rate b) mean single channel burst durations and c) mean single channel burst size. The recordings were taken at DIV 21.

There are no significant differences between C57BL/6 and SV129 cultures for network burst rate ratios, single channel burst duration, and single channel burst size (Fig 3.10).

Page 76

a) Mean Network Burst Start Interval b) Mean Jitter

3 2.0 *

e

e

n

n i

i 1.5

l

l e

e 2

s

s

a

a

B

B

1.0

o

o

t

t

o

o i

i 1

t

t a

a 0.5

R R

0 0.0 C57BL/6 SV129 C57BL/6 SV129

Figure 3.11. Connectivity parameter ratios between SCN1A+/- networks in C57BL/6 and SV129 backgrounds. Two parameters describing network synchrony through the differences in change from SCN1A+/- and WT in C57BL/6 and SV129 networks. These parameters are a) network burst start interval and b) jitter. The recordings were taken at DIV 21.

There is no significant difference in network burst start interval (Fig 3.11a). For mean jitter, (Fig 3.11b) the ratio in the SV129 background is significantly lower than in the

C57BL/6 background (1.04±0.60, 1.23±0.57, p<.05).

To compare backgrounds overall, SCN1A-/+ cultures on a C57BL/6 background have a significantly higher mean firing rate (Fig 3.2a), and mean firing rate inside bursts (Fig

3.2b) when compared to WT networks. In the SV129 background the opposite is seen, with SCN1A+/- networks having a significantly lower mean firing rate (Fig 3.6a) and decreased mean firing rate inside bursts (Fig 3.6b) compared to WT networks. These differences are significant from each other, as can be seen when comparing the ratios for these parameters for each genetic background (Fig 3.9a, 3.9b). For mean firing rate outside bursts and spike amplitude, there are no significant differences seen at any stage

Page 77 when comparing SCN1A+/- networks with WT (Fig 3.2d, 3.6d), or when comparing them between backgrounds (Fig 3.9d).

When looking at bursting parameters, SCN1A+/- networks show significantly lower bursting rate than WT networks across both genetic backgrounds (Fig 3.3a, 3.7a).

However, when comparing these differences to each other, they are not significantly different (Fig 3.10a). Burst sizes are significantly bigger in SCN1A+/- networks compared to WT in the C57BL/6 background (Fig 3.3b), however there is no significant difference in the SV129 background (Fig 3.7b), and the degree of difference is also not significantly different between the two genetic backgrounds (Fig 3.10b). For network burst duration, there is no significant difference in either background (Fig 3.3c, 3.7c), nor when comparing the degrees of difference (Fig 3.10c).

In the network connectivity parameters, there is a significant increase in network burst start intervals in SCN1A+/- cultures compared to WT in both genetic backgrounds (Fig

3.4a, 3.8a). The differences from WT are not significantly different between genotypes

(Fig 3.11a). There is also significant increase in mean jitter in SCN1A+/- networks from

WT in the C57BL/6 background (Fig 3.4b), but not in the SV129 background (Fig

3.8b). The changes from WT were significantly different between genotypes (Fig

3.11b).

Page 78

3.5 Discussion

The purpose of this study was to determine a) whether the MEA system is appropriate to model DS and b) whether it is sensitive enough to discriminate between the presence and absence of phenotype expression seen in different genetic backgrounds. The ability to quantify the differences between phenotypic expression in order to generate a network activity ‘fingerprint’ was also investigated for potential diagnosis and drug discovery purposes. This was achieved by culturing primary neurons sourced from mouse models of DS on MEA plates. These two mouse models contained the same

SCN1A+/- mutation in two different genetic backgrounds, C57BL/6 and SV129. The activity of the primary neurons were measured at DIV21 using the MEA system and the activity profile was generated for each background.

Initially we investigated the effects of the SCN1A+/- mutation in the C57BL/6 genetic background, as this was reported in the literature to result in strong expression of the DS phenotype. At DIV21, the networks exhibited a significantly higher firing rate over WT networks, as well as significantly increased firing inside bursts when compared to WT networks. SCN1A+/- networks also demonstrated a significantly reduced network burst rate and significantly increased burst size, as well as significantly increased synchrony as demonstrated by a significant increase in the network burst start interval. This is a profile of epileptic networks and supports our hypothesis that we will see a decreased burst rate, increased number of spikes within bursts and a greater length of time between network bursts, as is seen in the ADNFLE and febrile seizure models of epilepsy on the MEA system (Gullo et al. 2014; Hedrich et al. 2014). Additionally, the increase in firing rate and decrease in network burst rate reflects the blocking of

Page 79

GABAergic interneurons and reflects what is seen in networks on MEAs containing a high number of parvalbumin positive interneurons when a GABAA blocker is applied

(Baltz et al. 2010; Baltz et al. 2011; Baltz and Voigt 2015). It is possible that the haploinsufficiency of Nav1.1 on the parvalbumin-positive GABAergic interneurons in

SCN1A+/-/C57BL/6 networks result in the reduction of the GABAergic shift. This would prevent the networks from transitioning into a mature state brought on by this shift, and instead remain in a highly synchronous immature state. One additional parameter in which we found significant differences was increased jitter, which was unexpected as an increase in jitter correlates with decreased synchrony and is not a characteristic of epileptic networks. However, the synchrony that is measured by jitter is relative to synchrony within individual network bursts and not synchrony between the network bursts. Therefore, while jitter is a useful measurement of the ‘tightness’ of network bursts, it may not represent overall synchrony throughout the network. Even knowing this however, a decrease in synchrony within bursts is a characteristic of networks that have matured along the GABAergic shift and therefore would not be expected in a network with haploinsufficiency in the parvalbumin-positive interneurons.

These findings were then compared to the expression of the SCN1A+/- mutation in the

SV129 genetic background. The SV129 background is reported to resist expression of the DS phenotype. The SCN1A+/- networks in the SV129 background showed a significantly reduced mean firing and bursting rate, as well as decreased spike rate inside bursts, increased network burst start intervals and no change in jitter. This is not the profile expected for epileptic networks, as it was anticipated that we would see a significantly increased spike rate, as well as signifiers for other ‘synchronicity’

Page 80 parameters, such as a significant decrease in jitter as an addition to the increase in network burst start intervals. This supports our hypothesis that the DS phenotype would not be expressed in SCN1A+/- networks in a SV129 background on the MEA system.

It was then determined whether the MEA assay can statistically differentiate between different phenotypic expressions of the same genetic mutation for diagnostic and drug discovery purposes. This involved comparing the degree of change between SCN1A+/- and WT networks for each parameter in each genetic background to assess whether the differences seen were significant between the different genetic backgrounds. When comparing firing activity, there were significant differences between the SV129 and

C57BL/6 background for two of the four parameters. This is most likely due to the previous observation that the SCN1A+/- networks in the SV129 background do not demonstrate a typical epileptic phenotype whereas this phenotype is present in the

C57BL/6 background. There are no significant differences in bursting parameters, suggesting that the SCN1A+/- mutation effects the bursting parameters in SV129 and

C57BL/6 networks in the same way, despite different significant differences seen when comparing SCN1A+/- to WT in either background. This is most likely due to the differences in SCN1A+/- from WT trending in the same direction, regardless of significance. Finally when looking at synchronicity parameters, there is no difference seen in network burst start intervals, and significant difference seen in jitter. Once again the significant differences may be due to the fact that the typical epileptic phenotype was not present in the SV129 background and is present in the C57BL/6 background.

Page 81

While this technique aims to detect the differences between networks, it is still unknown as to what causes these differences. Compensatory mechanisms that mitigate the effects of seizure-causing mutations have been proposed to explain other cases of variable degrees of penetrance based on genetic background or sample age (Bando et al.

2013; Sun et al. 2013; LaSarge and Danzer 2014; Lin et al. 2017). This implies that a gene in the SV129 background is activated or silenced in order to minimize the increased firing rate that would otherwise be caused by a SCN1A+/- mutation. Theories behind the differences in seizure susceptibility have been proposed, such as endogenous

L1 retrotransposition, and loci responsible for strain differences in seizure susceptibility have been mapped to chromosomes 1, 2, 4, 5, 7, 10, 11, 12, 15 and 18 (Schauwecker

2011; Akagi et al. 2008). In a study comparing gene expression between C57BL/6 and

SV129 animals, high density oligonucleotide arrays were used to identify gene expression in multiple brain regions in each genotype, and compare expression. Several genes in chromosomes 1 and 4 relating to seizure susceptibility were found to be different between genotypes (Sandberg 2000). Therefore, there is a precedent set for genetic differences effecting seizure penetrance at a chromosomal level that translates to the network level.

3.6 Conclusion

From this study, we believe there is enough evidence to determine that the

SCN1A+/-/C57BL/6 networks demonstrate features of an epileptic network, and that this is most likely due to the SCN1A+/- mutation reducing the active parvalbumin-positive

GABAergic interneurons. Therefore, we can suggest that the MEA system is appropriate for modelling severe epilepsy mutations. The MEA system was also

Page 82 successful in not falsely capturing a phenotype that we did not believe to be present in the SV129 background. Finally, we believe that the MEA system has been successful in finding an ‘activity fingerprint’ for DS that is distinguishable from networks that do not have this phenotype, due to the number of significant differences between the C57BL/6 and SV129 backgrounds. This adds to the strength of MEA technology as a diagnostic tool for further genetic and drug testing in a high-throughput system.

Page 83

CHAPTER 4 | EXPLORING PREDICTIVE VALIDITY OF GENETIC DRAVET SYNDROME MODEL ON MULTIELECTRODE ARRAYS

4.1 Abstract

Introduction: While 70-80% of patients with new-onset epilepsy eventually enter remission through the use of antiepileptic drugs, 20-30% of patients continue to experience refractory epilepsy. Despite the ongoing development of new AEDs, overall outcomes for patients with newly diagnosed epilepsy remain unchanged (Tang et al.

2017). It is for this reason that research into the development of new antiepileptic drugs and improvement of drug discovery platforms is ongoing. MEAs have proven to be successful in screening drugs for toxicity and for therapeutic evaluation. Their high throughput capabilities and ability to utilize human iPSC neurons makes them ideal tools for drug screening in precision medicine. In this chapter, I explore the potential ulitity of the activity of cultured neuronal networks to predict pharmcosensitivity in DS.

Aim: To determine whether activity of cultures neuronal networks from a DS mouse model will show predictive validity for antiepileptic drugs.

Method: SCN1A+/-/C57BL/6 primary cortical neurons were cultured on MEAs and three drugs with known effects in DS (lamotrigine, stiripentol and cannabidiol) were applied.

Spiking, bursting and connectivity parameters were extracted and analysed to determine the drugs effect on the MEA networks, and to discern whether there was a significant effect in the haploinsufficient networks compared to the normal networks.

Page 84

Results: Significant changes across all parameters were observed in networks after application of lamotrigine, stiripentol, and cannabidiol. However very few significant differences were specific to the DS networks.

Conclusion: While the MEAs were able to detect and differentiate between the different compounds, these differences were not specific to the DS networks. We are therefore not able to confirm that this in vitro model is predictive for pharmacosensitivity in DS.

Further studies that include the testing of different concentrations and a larger variety of drugs are required to further examine the utility of this in vitro model in DS.

Investigating the utility of the neuronal culture-MEA assay in other neurodevelopmental disorders based on different genetic mutations is also warranted to investigate the potential scope of this model as a drug screening platform.

Key Words

Abbreviations

AChE – Acetylcholine

MOA – Mechanism of Action

AED – Antiepileptic Drug

STP = Stiripentol

LTG = Lamotrigine

CBD = Cannabidiol

MEA – Multielectrode Array

MFR – Mean Firing Rate

Page 85

NB – Network Burst

MFRIn – Mean Firing Rate Inside Bursts

MFROut – Mean Firing Rate Outside Bursts

NBSI – Network Burst Start Interval

DIV – Days in vitro

4.2 Introduction

Drug treatment is an integral part of seizure management in people with epilepsy. While

70-80% of patients with new-onset epilepsy eventually enter remission through the use of antiepileptic drugs (AEDs), 20-30% of patients continue to experience refractory epilepsy (Löscher et al. 2013). Drugs developed in recent years improve patient outcomes, however this improvement is primarily due to improved tolerability and minimization of harmful drug interactions. It remains that 85-90% of patients that are pharmacoresistant to first generation antiepileptic drugs still do not achieve seizure freedom using the more recently developed second and third generation drugs (Franco et al. 2016; Regner et al. 2018). It is for this reason that research into the development of new epileptic drugs is ongoing.

MEA technology is novel in its ability to capture the activity of neuronal networks over time in a high-throughput manner. This makes them an interesting candidate for quickly evaluating the effects of neuroactive compounds on complex neuronal network models of disease (Dhindsa and Goldstein 2015; Johnstone et al. 2010; Novellino et al. 2011;

McConnell et al. 2012; Wallace et al. 2015; Eder et al. 2016; Odawara et al. 2016).

Page 86

Before developing the MEA system as a disease model testing AEDs, the effectiveness of the MEA in detecting neuroactive compounds must be assessed. Shafer and colleagues have tested the multiwell MEA system in the context of the ToxCast program, a screening battery of ~500 in vitro assays for testing the effects of environmental compounds (Valdivia et al. 2014). Neuroactive compounds were identified by testing positive in one or more of 20 ToxCast Novascreen assays relating to ion channels (NSV_IC), testing negative in the NSV_IC assay but positively in other

ToxCast assays, or by known neuroactivity reported in the literature. After screening 92 compounds, MEAs were found to have a sensitivity of >70% to neuroactive compounds, identifying compounds that had been undetected with prior screening techniques. Notably, it was reported that MEAs were highly sensitive to compounds affecting voltage gated sodium channels, further supporting the evidence that this system is useful as an epileptic model. Shafer and colleagues have since increased the number of compounds screened on MEAs to 1055, highlighting the incredibly high- throughput nature of the system (Strickland et al. 2018). A great deal of validation has been done regarding toxicity screening in MEAs and the platform has been confirmed to be robust, and reproducible within and between laboratory groups (Novellino et al.

2011; McConnell et al. 2012; Vassallo et al. 2017).

MEAs can also be used to quickly and easily eliminate or identify potential therapeutic compounds to pursue based on their effect on neuronal network activity (however other tests will be required to examine the non-neuronal side effects of such compounds).

Gross and colleagues screened a battery of novel AChE blockers, comparing them to

Page 87 eserine, a known AChE inhibitor and drug clinically approved as effective against

Alzheimer’s Disease (Keefer et al. 2001). Of the seven novel compounds tested, two were determined functionally toxic, meaning while they do not cause cytotoxicity, they resulted in an irreversible cessation of all neural activity. Based on this knowledge it would be unwise to pursue these compounds for therapeutic application. From the remaining five compounds, potential mechanisms of action, effectiveness, and therapeutic window were evaluated. It was determined that two test compounds had a broader therapeutic window than eserine, and the excitatory or inhibitory nature of each compound was revealed. Petrou and colleagues have taken this one step further and utilized the MEA system to reveal the mechanism of action (MOA) of two previously unclassified compounds. Two naturally occurring antinociceptive compounds with unknown mechanisms of action (cannabidiol and conolidine) were added to cortical primary mouse neurons grown on MEAs and their activity profile was compared to that of a battery of compounds with known MOAs. Both experimental compounds showed the greatest similarity to a known Cav2.2 channel blocker. Further testing using whole- cell patch clamping confirmed that conolidine and cannabidiol both inhibit Cav2.2 channels, strongly suggesting this to be the MOA of both drugs (Mendis et al. 2019).

These studies highlight the effectiveness of the MEA system in not only screening for toxic compounds, but in detecting subtle differences between drug activity and identifying promising novel compounds for pursuit in drug development.

Finally, it has been reported in the literature that MEA technology has the sensitivity to detect a phenotype in neural network models of genetic disorders, and revert or diminish this phenotype through the application of therapeutic compounds. Using

Page 88 primary neurons sourced from rodent models of genetic epilepsy, Gullo et al. (2014) successfully established an in vitro model of ADNFLE. This model showed spontaneous hyperexcitability which was significantly reduced after application of carbamazepine, an AED currently used therapeutically in epilepsy patients. Human patient iPS cell-derived neural networks have also proven to be useful disease models on MEAs. Motor neurons containing a SOD1A4V/+ mutation were derived from iPS cells of patients with motor neuron disease (ALS). In combination with patch-clamping, it was found that SOD1A4V/+ produced networks with increased spontaneous firing compared to controls, and that the AED Retigabine significantly suppressed this firing

(Wainger et al. 2014). Most recently, small fiber neuropathy (SFN) patient iPS cells have allowed the concept of precision medicine to be fully realized. Neurons derived from patient iPS cells demonstrated significantly increased excitability compared to control networks in both MEAs and patch-clamp models. Lacosamide, an AED currently being trialed for use in SFN patients, significantly reduced the excitability of these networks in both the single cell and network level. After treatment with

Lacosamide, this patient reported a significant decrease in pain they had not experienced with any drugs trialed so far (Namer et al. 2019). This case highlights the potential for MEAs to be an integral part of precision medicine in genetic disease modelling as it allows for robust, rapid screening of many different therapeutic compounds for further study, with the findings potentially directly translatable to human patients.

As described in the previous chapter, Dravet Syndrome (DS), an epileptic channelopathy, is primarily caused by a variety of mutations in SCN1A resulting in

Page 89 happloinsufficiency of the Nav1.1 ion channel (Brunklaus and Zuberi 2014). It is hypothesized that this reduction of functional Nav1.1 channels is a major contributor to the DS phenotype (Bechi et al. 2012). To determine whether a drug will be effective against DS, disease models that integrate Nav1.1 happloinsufficiency may result in the discovery of effective compounds that address the pathophysiological mechanisms of the disorder in a way that pure seizure models would not detect (Franco et al. 2016). DS is a promising candidate for drug screening on the MEA platform as it has been robustly modelled in other systems and demonstrates a clear phenotype, while remaining largely resistant to drug treatment.

Lamotrigine, Stiripentol and Cannabidiol are three AEDs that have specific effects in

DS patients. It is reported that lamotrigine, a sodium channel inhibitor, is largely ineffective and often exacerbates the seizure phenotype of patients with DS, a finding that translates to both whole animal and neuronal in vivo models (Guerrini et al. 1998;

Anderson et al. 2017; Hawkins et al. 2017). Stiripentol is a GABA modulator and is the only drug specifically indicated for use in DS patients. Both animal models and patient data report stiripentol to reduce the frequency and duration of seizures (Cho et al. 2018;

Griffin et al. 2018; Chiron and Dulac 2011). Stiripentol has multiple mechanisms of action and is often used in combination with clobazam as it increases the levels of the metabolite norclobazam, enhancing clobazams’ therapeutic effects (Giraud et al. 2006).

In isolation, Stiripentol is a positive allosteric modulator on GABAA receptors and is particularly effective if that receptor contains an α3 subunit (Fisher 2009). Stiripentol acts by increasing the opening time of GABA-activated channels pre and post- synaptically, potentiating brain GABA levels resulting in an anticonvulsant effect

Page 90

(Quilichini et al. 2006; Fisher 2011). Cannabidiol has been gaining popularity as a therapeutic agent for DS. In SCN1A+/- animal models of DS, cannabidiol reduced the duration and frequency of induced seizures, as well as the frequency of spontaneous seizures (Kaplan et al. 2017). In patient trials involving children and young adults with

DS, cannabidiol was added to their current treatment repertoire. A significant decrease in monthly convulsive seizure frequency, as well as a decrease in overall seizures was seen in the cannabidiol group as opposed to the placebo group was found (Laux et al.

2019; Devinsky et al. 2017). Cannabidiol is proposed to have a multimodal mechanism of action, with identified targets including inhibition of GPR55 and T-type voltage gated calcium channels, and stimulation of serotonin receptors (Chen et al. 2018;

Kaplan et al. 2017; Gaston and Friedman 2017; Devinsky et al. 2014). As of 2019, several countries including Australia, Canada, Germany, and the United Kingdom, have legalized the medical use of cannabis (Samanta 2019). Cannabidiol has also been approved by the FDA under the name ‘Epidiolex’ for the treatment of DS in patients two years and older (Abu-Sawwa and Stehling 2020).

This aim of this study is to investigate the use of MEAs as a screening method for evaluating the effects of antiepileptic drugs on a genetic model of DS. To determine whether the MEA system is robust enough to discriminate between the behavior of

AEDs on wildtype and mutant networks, we have applied three AEDs with known effects in DS to the cultured SCN1A+/- networks and have analysed their effects on spiking, bursting, and network connectivity.

Hypothesis 1: The compounds will influence activity profile of the cultured networks.

Page 91

Hypothesis 2: These effects will be significantly different in the wildtype phenotype compared to the mutant phenotype.

Hypothesis 3: The drug effects will correspond with what is reported in the literature, ie lamotrigine will exacerbate the DS phenotype, while stiripentol and cannabidiol will reduce it.

4.3 METHOD

4.3.1 Animals

SCN1A+/-/C57BL/6 mice were maintained in house at the Florey Institute animal breeding facility. Litters were taken from the mother from 0-2 days after birth. All animal experiments were approved by the Howard Florey Institute Animal Ethics

Committee (AEC#14-075), and performed in accordance with the Prevention of Cruelty to Animals Act and the NHMRC Australian Code of Practice for the Care and Use of

Animals for Scientific Purposes.

4.3.2 Media Preparation

See 2.2.2

4.3.3 MEA Preparation

See 2.2.3

4.3.4 Primary Culture Procedure

See 2.3.4

4.3.5 Compound Testing

Page 92

Each well of a mwMEA plate received a full media change 24hr prior to compound

testing. Three compounds tested are listed in Table 4.1 and stock solutions were further

diluted in media to reach the final concentration in the well. Recordings were taken for

11 minutes prior to compound addition to establish a baseline. Each compound was then

added to one well of a 24-well MEA plate in a 5uL aliquot, along with a comparable

DMSO vehicle control in a corresponding well. The plates were left to sit for 1 minute

after drug addition to allow the cultures to stabilize following perturbation and then 11

minute recordings were again taken. For drugs with multiple concentrations, another

5uL of compound and a corresponding DMSO control was added after this initial

recording and once again an 11 minute recording was taken. Concentrations of drugs

were selected on the basis of therapeutic concentration reported in the literature (Chiron

et al. 2000) and through prior experimentation at effective concentrations on neurons on

MEAs (Mendis et al. 2016a; Mendis et al. 2019).

Compound Name Source Stock Final Concentration Concentration

Cannabidiol Under the Tree 75mM 10uM, 20uM BioPharmaceuticals Pty Ltd, Forte dei Marmi, Italy

Lamotrigine L3791, Sigma-Aldrich 85mM 100uM, 200uM Stiripentol S6826, Sigma-Aldrich 85mM 20uM, 40uM, 60uM

Table 4.1. Description of compounds. Drug names, source, stock concentration, and

final concentration in the MEA well.

4.3.6 Raw Data Processing

See 2.3.6 Page 93

4.4 Results

Using the culturing and recording protocols and the analysis tools described above we observed the characteristics of C57BL/6 SCN1A+/- cultures before and after the application of different antiepileptic compounds. The first compound tested was

Lamotrigine (LTG). Primary cortical neurons from C57BL/6 SCN1A+/- animals at ages

P0-P2 were extracted and cultured on 24-well MEA plates until DIV21-DIV28. We added LGT to the MEA wells via a series of 5uL aliquots of 10mM drug diluted in cell culture media, made from the initial 85mM stock solution. A corresponding DMSO control was also added at each concentration. A Mann-Whitney U test was used to determine whether the drug had a significant effect on the activity of the networks compared to the DMSO control, to determine that the MEA system could differentiate between drug and control. The percentage change from baseline activity for DMSO and

LTG conditions at 100uM LTG was compared (DMSO n = 33 (from 8 animals), LTG n

= 32 (from 8 animals)). At 200uM LTG, many of the wells were silenced completely by the drug, making it difficult to determine more nuanced changes in the activity profile of the networks. Therefore, only the 100uM condition is shown below. The raster plots representing 30 seconds of network activity can be found in the supplementary figures

(SF1).

Page 94

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts 20 **** 60 **

0 40

-20 %

% 20 -40

0 -60

-80 -20

DMSO LTG DMSO LTG

c) Mean Firing Rate Outside Bursts d) Average Spike Amplitude 50 8 ** **** 6

0 % % 4

-50 2

-100 0

DMSO LTG DMSO LTG

Figure 4.1. Spiking parameters in C57BL/6 networks in response to LTG. Percentage change from baseline in cultures after addition of 100uM compared to their DMSO controls. The parameters used represent spiking activity and are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average spike amplitude.

In the WT networks, the change in mean firing rate (Fig 4.1a) from baseline after addition of 100uM LTG is significant in comparison to the DMSO control, with LTG resulting in a decrease from baseline by 53%, and DMSO only resulting in a 7.5% decrease (-53.09 [CI -59.01, -38.12], -7.51 [CI -14.02, -4.85], p<.001). The change in mean firing rate inside bursts (Fig 4.1b) after addition of 100uM LTG is also significant

Page 95 compared to the DMSO control, with an increase from baseline by 22.2% compared to a decrease from baseline by 5.5% (22.18 [CI 10.49, 40.27], -5.48 [CI -6.91, 14.04], p<.01). The mean firing rate outside bursts (Fig 4.1c) decreases significantly by 85.3% compared to a 15.8% decrease with DMSO (-85.32 [CI -87.43, -65.30], -15.81 [CI -

18.54, 37.27], p<.001). Finally, average spike amplitude (Fig 4.1d) significantly increased by 3.7% in response to 100uM LTG compared to a 1.4% increase with

DMSO (3.73 [CI 2.93, 6.98], 1.36 [CI 0.21, 3.04], p<.01).

a) Network Burst Rate b) Single Channel Burst Duration 20 **** 150 *** 0 100

-20 % -40 % 50

-60 0 -80

-100 -50

DMSO LTG DMSO LTG c) Single Channel Burst Size 150 ****

100

% 50

0

-50

DMSO LTG Figure 4.2. Bursting parameters in C57BL/6 networks in response to LTG.

Percentage change from baseline in cultures after addition of 100uM LTG compared to their DMSO controls. The parameters used represent bursting activity and are a) network burst rate b) single channel burst duration c) single channel burst size.

Page 96

In the WT networks for bursting parameters, the change in the network burst rate (Fig

4.2a) from baseline after addition of 100uM LTG is significant in comparison to the

DMSO control, with LTG resulting in a decrease from baseline by 54.2%, and DMSO resulting in a 6.5% decrease (-54.18 [CI -65.62, -40.60], -6.47 [CI -10.77, -0.06], p<.001). The change in single channel burst duration (Fig 4.2b) after addition of 100uM

LTG is significant compared to the DMSO control, with an increase from baseline by

63.8% compared to a decrease from baseline by 0.78% (63.78 [CI 43.47, 132.2], -0.78

[CI -6.97, 9.79], p<.001). The single channel burst size (Fig 4.2c) increases significantly at 100uM LTG by 67.7% compared to a 1% increase in the DMSO control

(67.68 [CI 51.78, 115.7], 0.95 [CI -5.57, 6.57], p<.001).

a) Network Burst Start Interval b) Jitter 250 10 **** 200 0 -10

150 % % -20 100 -30 50 -40

0 -50

DMSO LTG DMSO LTG c) Kappa 100 **** 80

60 % 40

20

0

DMSO LTG Figure 4.3. Connectivity parameters in C57BL/6 networks in response to LTG.

Page 97

Percentage change from baseline in cultures after addition of 100uM LTG compared to their DMSO controls. The parameters used represent network connectivity and are a)

NBSI b) jitter c) kappa value.

In the WT networks for the network connectivity parameters, the change in the NBSI

(Fig 4.3a) from baseline after addition of 100uM LTG is significantly increased in comparison to the DMSO control, with LTG resulting in an increase from baseline by

90.5%, and DMSO resulting in a 7% increase (90.45 [CI 93.11, 215.2], 6.94 [CI 2.33,

14.89], p<.001). The change in jitter (Fig 4.3b) after addition of LTG is not significant between drug and DMSO controls. The Kappa value (Fig 4.3c) increases significantly at

100uM LTG by 57.5% compared to a 7.2% increase in the DMSO control (57.53 [CI

49.68, 82.63], 7.18 [CI 2.68, 11.21], p<.001).

Lamotrigine at 100uM significantly influences the activity profile of the C57BL/6 WT networks. Compared to the DMSO control, lamotrigine appears to reduce the overall firing rate and the firing rate outside bursts, as well as overall network bursts. Burst length and size are both significantly increased. Finally, the NBSI and Kappa is significantly increased. The overall pattern of activity appears to be that in WT cultures, lamotrigine decreases spiking and bursting rate, while increasing the size and length of the bursts. It also increases synchrony within the culture in the form of increased NBSI and Kappa value. After establishing that lamotrigine had a noticeable effect on primary cultured cells in MEAs, it was of interest to determine whether differences in response to lamotrigine would be significantly detected in SCN1A+/- networks (n = 16 (from 4 animals)) compared to WT (n = 32 (from 8 animals)).

Page 98

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts 50 100 ** 50 0

0 % % -50 -50 -100 -100

-150 -150 WT SCN1A+/- WT SCN1A+/-

c) Mean Firing Rate Outside Bursts d) Average Spike Amplitude 50 10 *

0 5

** % % -50

0 -100

-150 -5

WT SCN1A+/- WT SCN1A+/- Figure 4.4. Spiking parameters in SCN1A+/- and WT networks in response to LTG.

Percentage change from baseline in WT an SCN1A+/- cultures after addition of 100uM

LTG. The parameters used represent spiking activity and are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average spike amplitude.

The change in mean firing rate (Fig 4.4a) from baseline after addition of 100uM LTG is not significantly different between the SCN1A+/- and WT networks. The mean firing rate inside bursts (Fig 4.4b) is significantly reduced in SCN1A+/- networks compared to WT, with a decrease from baseline by 9.1% compared to a 22.2% increase in WT networks (-

9.07 [CI -61.39, 10.63], 22.18 [CI 10.49, 40.27], p<.01). The mean firing rate outside

Page 99 bursts (Fig 4.4c) decreases significantly by 95.4% compared to a 85.3% decrease in WT networks (-95.37 [CI -97.45, -84.53], -85.32 [CI -87.43, -65.30], p<.01), Average spike amplitude (Fig 4.4d) is significantly different in SCN1A+/- networks compared to WT, with an increase from baseline by 0.9% compared to a 3.7% increase in WT networks

(0.87 [CI -21.91, 6.53], 3.73 [CI 2.93, 6.98], p<.05).

a) Network Burst Rate b) Single Channel Burst Duration 50 150

100 0

50 % -50 % 0

-50 -100 -100

-150 -150

WT SCN1A+/- WT SCN1A+/-

c) Single Channel Burst Size 150

100

50

% 0

-50

-100

-150

WT SCN1A+/- Figure 4.5. Bursting parameters in SCN1A+/- and WT networks in response to LTG.

Percentage change from baseline in SCN1A+/- cultures after addition of 200uM LTG compared to WT controls. The parameters used represent bursting activity and are a) network burst rate b) single channel burst duration c) single channel burst size.

Page 100

The change in the network burst rate (Fig 4.5a) from baseline after addition of 100uM

LTG is not significantly different in the SCN1A+/- networks compared to WT. There is

also no significant difference in single channel burst duration or single channel burst

size between SCN1A+/- and WT networks after application of LTG (Fig 4.5b, 4.5c).

a) Network Burst Start Interval b) Jitter 300 50

200 0

100 % % -50 0 -100 -100

-200 -150

WT SCN1A+/- WT SCN1A+/-

c) Kappa 100

50

0 % -50

-100

-150

WT SCN1A+/- Figure 4.6. Connectivity parameters in SCN1A+/- and WT networks in response to

LTG. Percentage change from baseline in SCN1A+/- cultures after addition of 100uM

LTG compared to WT controls. The parameters used represent network connectivity

and are a) NBSI b) jitter c) kappa value.

The change in the NBSI, jitter, and Kappa value between SCN1A+/- and WT after

addition of 100uM LTG are not significantly different (Fig 4.6).

Page 101

It appears that lamotrigine significantly decreases the mean firing rate inside bursts, mean firing rate outside bursts, and spike amplitude to a greater degree in SCN1A+/- networks than in WT networks. It has no significant on overall mean firing rate, or any of the bursting and network connectivity parameters. In the WT networks compared to

DMSO control, we also see that lamotrigine significantly decreases spike rate, as well as reducing the number of network bursts while increasing their length and size. It was also observed that lamotrigine increases network synchrony.

Stiripentol (STP) was then tested on the MEA network. Stiripentol is the only drug prescribed specifically for Dravet syndrome. While a variety of concentrations was tested, we chose to make comparisons at the 40uM concentration as this is what has been reported as the therapeutic concentration in vivo (Chiron et al. 2000). Once again we first wanted to determine whether the effects of stiripentol on neuronal networks could be detected on the MEA system by adding 40uM stiripentol to WT networks (n =

32 (from 11 animals)) and a comparable DMSO control (n = 34 (from 12 animals)), and taking 11 minute recordings before and after application. The raster plots representing

30 seconds of network activity can be found in the supplementary figures (SF2).

Page 102

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts **** ** 10 10

0 0

-10 -10

% % -20 -20

-30 -30

-40 -40 DMSO STP DMSO STP c) Mean Firing Rate Outside Bursts d) Average Spike Amplitude * 40 4 **

20 2

0 0

% % -20 -2

-40 -4

-60 -6

DMSO STP DMSO STP

Figure 4.7. Spiking parameters in C57BL/6 networks in response to STP. Percentage change from baseline in cultures after addition of 40uM STP compared to their DMSO controls. The parameters used represent spiking activity and are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average spike amplitude.

In the WT networks, the change in mean firing rate (Fig 4.7a) from baseline after addition of 40uM STP is significant in comparison to the DMSO control, with STP resulting in a decrease from baseline by 26.2%, and DMSO only resulting in a 3.5% decrease (-26.21 [CI -35.75, -16.84], -3.53 [CI -9.79, 11.82], p<.001). The change in mean firing rate inside bursts (Fig 4.7b) after addition of 50uM STP is also significant

Page 103 compared to the DMSO control, with a decrease from baseline by 23.9% compared to a decrease from baseline by 3.7% (-23.85 [CI -33.38, -5.52], -3.75 [CI -10.51, 20.37], p<.01). The mean firing rate outside bursts (Fig 4.7c) decreases significantly by 7.2% compared to a 6.3% increase with DMSO (-7.15 [CI -27.64, 14.71], 6.35 [CI -33.62,

140.6], p<.05). Finally, average spike amplitude (Fig 4.7d) significantly decreased by

2.2% in response to 40uM STP compared to a 0.7% increase with DMSO (-2.17 [CI -

8.60, 0.13], 0.67 [CI -1.06, 5.05], p<.01). a) Network Burst Rate b) Single Channel Burst Duration 40 20 **** 30 0

20 % % -20 10 -40 0

-10 -60

DMSO STP DMSO STP c) Single Channel Burst Size 10 **** 0

-10

% -20

-30

-40

-50

DMSO STP Figure 4.8 Bursting parameters in C57BL/6 networks in response to STP. Percentage change from baseline in cultures after addition of 40uM STP compared to their DMSO controls. The parameters used represent bursting activity and are a) network burst rate b) single channel burst duration c) single channel burst size.

Page 104

In the WT networks for bursting parameters, the change in the network burst rate (Fig

4.8a) from baseline after addition of 40uM STP is not significant in comparison to the

DMSO control. The change in single channel burst duration (Fig 4.8b) after addition of

40uM STP is significant compared to the DMSO control, with a decrease from baseline by 36.2% compared to a decrease from baseline by 5% (36.22 [CI -38.99, -19.24], -4.99

[CI -14.60, 21.78], p<.001). The single channel burst size (Fig 4.8c) decreases significantly at 40uM STP by 31.9% compared to a 4.6% decrease in the DMSO control

(-31.90 [CI -40.88, -23.47], -4.62 [CI -11.89, 8.63], p<.001).

Network Burst Start Interval a) b) Jitter * *** 10 20

0 0

% -20

% -10

-20 -40

-30 -60

DMSO STP DMSO STP

c) Kappa 10

0

-10

% -20

-30

-40

-50

DMSO STP Figure 4.9. Connectivity parameters in C57BL/6 networks in response to STP.

Percentage change from baseline in cultures after addition of 40uM STP compared to

Page 105 their DMSO controls. The parameters used represent network connectivity and are a)

NBSI b) jitter c) kappa value.

The NBSI (Fig 4.9a) from baseline after addition of 40uM STP is significantly decreased in comparison to the DMSO control, with STP resulting in a decrease from baseline by 21.7%, and DMSO resulting in a 4.2% decrease (-21.68 [CI -26.08, 2.58], -

4.22 [CI -9.11, 20.62], p<.05). Jitter (Fig 4.9b) significantly decreases by 27.8% in the

STP condition compared to a 3.5% increase in the control. The Kappa value (Fig 4.9c) is not significantly different between the STP and DMSO conditions.

Overall, it appears that 40uM stiripentol significantly changes the activity profile of the

WT C57BL/6 cortical networks. In comparison to DMSO controls, all four spiking parameters are significantly reduced. While the network bursts are not significantly different between drug condition and control, single channel burst duration and single channel burst size are also significantly decreased. The connectivity paramters are also significantly affected, with NBSI and jitter significantly decreased, but no change is seen in Kappa value. It appears that stiripentol reduces overall spiking activity and bursting activity, as burst size and length also significantly decreases. Finally the network connectivity appears to significantly decrease, with networks becoming less synchronized as demonstrated by NBSI and kappa value. Once a significant effect had been detected in the WT networks, we then wanted to determine whether stiripentol had any discriminate effects on SCN1A+/- networks (n = 19 (from 7 animals) over WT (n =

32 (from 11 animals)).

Page 106

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts 10 10

0 0

-10 -10

% -20 % -20

-30 -30

-40 -40

-50 -50 WT SCN1A+/- WT SCN1A+/- c) Mean Firing Rate Outside Bursts d) Average Spike Amplitude 20 50 ****

0 0 % % -20 -50

-40 -100

-60 -150 WT SCN1A+/- WT SCN1A+/- Figure 4.10. Spiking parameters in SCN1A+/- and WT networks in response to STP.

Percentage change from baseline in WT an SCN1A+/- cultures after addition of 40uM

STP. The parameters used represent spiking activity and are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average spike amplitude

The change in mean firing rate, mean firing rate inside bursts, and mean firing rate outside bursts (Fig 4.10a, 4.10b, 4.10c) is not significantly different between the WT and SCN1A+/- networks following the application of 40uM STP. Average spike amplitude (Fig 4.10d) is significantly different in SCN1A+/- networks compared to WT,

Page 107 with a decrease from baseline by 95% compared to a 2.2% decrease in WT networks (-

95.02 [CI -95.59, -81.11], -2.17 [CI -8.60, 0.13], p<.001).

a) Network Burst Rate b) Single Channel Burst Duration 40 20 **

20 0

0 %

% -20 -20 -40 -40

-60 -60 WT SCN1A+/- WT SCN1A+/- c) Single Channel Burst Size 10

0

-10

% -20

-30

-40

-50 WT SCN1A+/- Figure 4.11. Bursting parameters in SCN1A+/- and WT networks in response to STP.

Percentage change from baseline in SCN1A+/- cultures after addition of 40uM STP compared to WT controls. The parameters used represent bursting activity and are a) network burst rate b) single channel burst duration c) single channel burst size.

The change in the network burst rate or single channel burst size (Fig 4.11a, 4.11c) from baseline after addition of 40uM STP is not significantly different between WT and

SCN1A+/- networks. The change in single channel burst duration (Fig 4.11b) is significant in comparison to the WT control, with STP resulting in a decrease from

Page 108 baseline by 14.2%, in SCN1A+/- networks and 36.2% in WT networks (-14.23 [CI -

20.04, 4.44], -36.22 [CI -38.99, -19.24], p<.01).

a) Network Burst Start Interval b) Jitter 150 10

0 100

-10 % 50 % -20

-30 0 -40

-50 -50 WT SCN1A+/- WT SCN1A+/- c) Kappa 20

0

% -20

-40

-60 WT SCN1A+/- Figure 4.12. Connectivity parameters in SCN1A+/- and WT networks in response to

STP. Percentage change from baseline in SCN1A+/- cultures after addition of 40uM

STP compared to WT controls. The parameters used represent network connectivity and are a) NBSI b) jitter c) kappa value.

The change in the NBSI, jitter, and Kappa value between SCN1A+/- and WT after addition of 40uM STP are not significantly different (Fig 4.12).

It appears that stiripentol has very little discriminate effect on SCN1A+/- networks compared to WT for total spike rate and spike rate inside or outside bursts. The spike

Page 109 amplitude of SCN1A+/- networks decreases significantly compared to the WT networks.

There were no discriminate effects on network burst rate or burst size between SCN1A+/- and WT networks, however there was a significant different in network burst duration, with the WT networks showing a significant decrease compared to what was seen in the

SCN1A+/- networks. Stiripentol does not affect network synchrony in SCN1A+/- networks in comparison to WT. Once again, it was seen that while the effects of STP on cortical neural cultures were detected by the MEA system, these changes were largely indiscriminate between SCN1A+/- and WT.

Finally, cannabidiol (CBD) was applied to the cultured networks to determine whether the MEA system was able to detect an effect on the cultured networks. Additionally,

CBD was investigated to investigate whether CBD showed a similar pattern of activity to the other antiepileptic drugs tested. 10uM CBD was added to WT networks (n = 37

(from 14 animals)) and a comparable DMSO control (n = 36 (from 11 animals)) and 11 minute recordings were taken before and after addition. The raster plots representing 30 seconds of network activity can be found in the supplementary figures (SF3).

Page 110

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts 5 **** 15

0 10 -5

% 5 % -10

-15 0 -20

-25 -5 DMSO CBD DMSO CBD

c) Mean Firing Rate Outside Bursts d) Spike Amplitude 10 **** 4 0 2

-10 %

% 0 -20

-2 -30

-40 -4 DMSO CBD DMSO CBD

Figure 4.13. Spiking parameters in C57BL/6 networks in response to CBD.

Percentage change from baseline in cultures after addition of 10uM CBD compared to their DMSO controls. The parameters used represent spiking activity and are a) mean firing rate b) mean firing rate inside bursts c) mean firing rate outside bursts and d) average spike amplitude.

In the WT networks, the change in mean firing rate (Fig 4.13a) from baseline after addition of 10uM CBD is significant in comparison to the DMSO control, with DMSO resulting in a decrease from baseline by 19.4%, and DMSO only resulting in a 4.8% decrease (-19.39 [CI -26.38, -12.78], -4.83 [CI -5.41, 2.39], p<.001). The change in mean firing rate inside bursts (Fig 4.13b) after addition of 10uM CBD is not significant

Page 111 compared to the DMSO control. The mean firing rate outside bursts (Fig 4.13c) decreases significantly by 25.4% compared to a 4.4% decrease with DMSO (-25.36 [CI

-40.00, 22.93], -4.41 [CI -9.23, -1.35], p<.001). Finally, average spike amplitude (Fig

4.13d) was not significant between DMSO and drug conditions.

a) Network Burst Rate b) Single Channel Burst Duration 10 40

0 30

-10 % % 20 -20

10 -30

-40 0 DMSO CBD DMSO CBD

c) Single Channel Burst Size 30

20

% 10

0

-10 DMSO CBD

Figure 4.14. Bursting parameters in C57BL/6 networks in response to CBD.

Percentage change from baseline in cultures after addition of 10uM CBD compared to their DMSO controls. The parameters used represent bursting activity and are a) network burst rate b) single channel burst duration c) single channel burst size.

Page 112

Change in the network burst rate, single channel burst duration, and single channel burst size (Fig 3.14) from baseline after addition of 10uM CBD is not significant in comparison to the DMSO control.

a) Network Burst Start Interval b) Jitter 50 20

40 10 30

0 % 20 % -10 10

0 -20

-10 -30 DMSO CBD DMSO CBD c) Kappa 200 *

150

100 % 50

0

-50 DMSO CBD

Figure 4.15. Connectivity parameters in C57BL/6 networks in response to CBD.

Percentage change from baseline in cultures after addition of 10uM CBD compared to their DMSO controls. The parameters used represent network connectivity and are a)

NBSI b) jitter c) kappa value.

Change in the NBSI and jitter (Fig 4.15a, 4.15b) from baseline after addition of 10uM

CBD was not significant when compared to the DMSO control. The Kappa value (Fig

4.15c) after CBD addition was significantly increased by 30.7% when compared to an

Page 113 increase of 5.3% in the DMSO control (30.71 [CI 14.84, 113.4], 5.26 [CI -26.72,

20.01], p<.05).

Overall, it appears that 10uM CBD significantly reduces the overall mean firing rate and mean firing rate outside bursts in WT networks compared to a vehicle control. CBD does not appear to have a significant impact on any of the bursting parameters, nor on

NBSI and jitter. However, it does appear to significantly increase network connectivity as measured by the Kappa value. CBD then was applied to WT and SCN1A+/- networks to determine whether CBD had a significant effect on epileptic networks over non- epileptic networks, as was done for STP and LTG. 10uM CBD was added as a 5uL aliquot to both WT networks (n = 37 (from 14 animals)) and SCN1A+/- networks (n = 30

(from 8 animals) and 11-minute recordings were taken before and after drug application.

Page 114

a) Mean Firing Rate b) Mean Firing Rate Inside Bursts 10 30

0 20 % -10 % 10

-20 0

-30 -10 WT SCN1A+/- WT SCN1A+/-

d) Spike Amplitude c) Mean Firing Rate Outside Bursts 4 20

0 2

% 0

% -20

-40 -2

-60 -4 WT SCN1A+/- WT SCN1A+/-

Figure 4.16. Spiking parameters in SCN1A+/- and WT networks in response to CBD.

Percentage change from baseline in WT an SCN1A+/- cultures after addition of 10uM

CBD. The parameters used represent spiking activity and are a) mean firing rate b)

mean firing rate inside bursts c) mean firing rate outside bursts and d) average spike

amplitude

Page 115

a) Network Burst Rate b) Single Channel Burst Duration 10 40

0 30

-10 % % 20 -20 10 -30

-40 0 WT SCN1A+/- WT SCN1A+/-

c) Single Channel Burst Size 40

30

20 % 10

0

-10 WT SCN1A+/-

Figure 4.17. Bursting parameters in SCN1A+/- and WT networks in response to CBD.

Percentage change from baseline in SCN1A+/- cultures after addition of 10uM CBD compared to WT controls. The parameters used represent bursting activity and are a) network burst rate b) single channel burst duration c) single channel burst size.

There was no significant difference between the effect of 10uM CBD on the SCN1A+/- and WT cortical cultures for any of the spiking or bursting parameters (Fig 4.16, 4.17).

Page 116

a) Network Burst Start Interval b) Jitter 80 20

10 60

0 % % 40 -10 20 -20

0 -30 WT SCN1A+/- WT SCN1A+/- c) Kappa 200 *

150

100 % 50

0

-50 WT SCN1A+/-

Figure 4.18. Connectivity parameters in SCN1A+/- and WT networks in response to

CBD. Percentage change from baseline in SCN1A+/- cultures after addition of 10uM

CBD compared to WT controls. The parameters used represent network connectivity and are a) NBSI b) jitter c) kappa value.

Change in the NBSI and jitter (Fig 4.18a, 4.18b) after addition of 10uM CBD was not significant in the SCN1A+/- networks when compared to the WT control. The Kappa value (Fig 4.18c) after CBD addition was significantly different in SCN1A+/- networks compared to WT, with an increase of 14.6% when compared to an increase of 30.7% in the WT networks (14.59 [CI 3.64, 20.57], 30.71 [CI 14.84, 113.4], p<.05).

Page 117

Overall, it appears that CBD does not notably change the SCN1A+/- networks in comparison to WT. None of the spiking or bursting parameters are significantly different between genotypes. The kappa value is significantly decreased compared to

WT, while NBSI and jitter show no change.

4.5 Discussion

The purpose of this study was to determine a) whether antiepileptic compounds have a detectable effect on neuronal networks cultured on MEAs, b) whether the drugs affect the SCN1A+/- networks differently to the WT networks and c) whether these drug effects correspond with what has previously been reported in the literature. This was achieved by culturing primary C57BL/6 SCN1A+/- neurons on 24-well MEA plates and allowing them to mature to DIV 21-28. AEDs with known effects on a DS phenotype were applied, and the activity profile was measured.

Initially, we looked at the effects of lamotrigine on the WT networks to establish the extent to which this drug modulated the cultures on MEAs. Of the spiking parameters,

100uM lamotrigine significantly reduced the mean firing rate, as well as significantly decreasing the firing rate outside bursts. Firing rate inside bursts however was significantly increased, along with spike amplitude. It should also be noted that there was a significant difference in spike amplitude from baseline in WT networks after applying lamotrigine and no difference to the SCN1A+/- networks. This may imply that spike amplitude is not an ideal parameter for identifying the effects of drugs acting via voltage-dependent Na+ channels on DS networks as this mutation causes a reduction in these channels, so the effects may not be as visible or pronounced from baseline as

Page 118 those in the WT networks. However, lamotrigine has been shown to reduce spike amplitude even in WT networks, so this result was unexpected (Greenhill and Jones

2010; Albertson et al. 2011). Another explanation for this significant difference in amplitude between the WT and SCN1A+/- networks may be due to the inability of each electrode to discern between individual neurons grown at high densities. Therefore, if multiple neurons near the electrode fire instantaneously, this may be detected as a single large spike amplitude rather than an amalgamation of smaller ones. Disruptions in the network such as adding a compound could result in a change in amplitude as more or less neurons around the electrode may be firing at that time, increasing or reducing the total recorded amplitude. This may have occurred in the WT networks, inflating the perceived spike amplitude compared to the SCN1A+/- networks. This may be rectified by increasing the n values to account for cultures that are more sensitive to perturbation.

Lamotrigine also had a significant effect on bursting parameters, resulting in significantly reduced network burst rate, and significantly increased burst size and burst duration. Finally, there was a significant effect on network connectivity, with significantly more time between network bursts, significantly less jitter, and a significantly higher kappa value. Overall, lamotrigine appears to synchronize the network, as less spikes are occurring outside bursts, bursts are fewer but longer and larger with significantly more time between their occurrence, and the kappa value is significantly higher. This is not an expected characteristic of an antiepileptic drug, as epileptic seizures are theorized to be a result of increased neuronal network synchrony.

Therefore we anticipated that this drug would reduce network synchrony as part of its function as an antiepileptic compound.

Page 119

When comparing the effects of lamotrigine on disease model to wildtype, lamotrigine results in significantly less spiking both inside and outside bursts, and a significantly reduced spike amplitude in the SCN1A+/- networks compared to WT. This was an unexpected finding, as lamotrigine promotes seizure activity in human DS patients.

Therefore, it was expected that lamotrigine would increase spiking activity in SCN1A+/- networks in comparison to WT. There are no significant differences in the effects of lamotrigine for bursting or network connectivity parameters. While this suggests that the MEA system has a level of sensitivity to discriminate between a drug response in

SCN1A+/- compared to WT networks, the reduced spiking parameters do not appear to be an indicator of an overall increase in seizure activity, as was expected in this model.

Stiripentol, a reasonably effective drug in DS patients, was also tested on these networks. At 40uM, stiripentol significantly decreased all four spiking parameters in

WT networks. While the network burst rate was not significantly different, burst duration and burst size were also significantly decreased after stiripentol application.

NBSI and jitter were significantly decreased, with no significant change seen in Kappa.

Decrease in spike rate, burst size, and burst duration are all indicators of anti-seizure effects, as an increase in these parameters have been identified as a marker of an epileptic networks on MEAs (Gullo et al. 2014; Hedrich et al. 2014). It is notable that lamotrigine significantly increased burst size and duration, where stiripentol has significantly decreased these parameters. Although this effect does not appear to be specific to the SCN1A+/- networks, as the same changes occur in WT networks, it indicates that the 24-well MEA system can discriminate between the effects of different

AEDs, a finding that may be helpful in future screening of novel compounds.

Page 120

When comparing the effects of stiripentol on SCN1A+/- and wildtype networks,

SCN1A+/- networks show significantly decreased spike amplitude and significantly increased burst duration. As with lamotrigine, it was unexpected to see this drastic change in the spike amplitude after application of stiripentol. While it is expected that stiripentol will reduce network activity due to GABA potentiation, this drastic decrease warrants further investigation. Similar to lamotrigine, this decrease may be due to the reduction of activity in neurons surrounding the recording electrode, meaning that fewer action potentials will be detected and those that are will be further away so will be recorded as a low amplitude spike. It is also a possibility that the networks used had an unusually low amplitude at baseline, therefore any change in amplitude that may not be detected in a more robust network will be exacerbated here. This may be rectified by increasing the n values to determine whether this change is real or a result of networks sensitive to mechanical perturbation. No other spiking, bursting, or network connectivity parameters show significant differences. While this suggests that the MEA system has can discriminate between a drug response in SCN1A+/- compared to WT networks, the changes in these two parameters do not appear to be an indicator of decreased seizure activity, as was expected in this model.

Cannabidiol at 10uM significantly decreased overall firing rate and firing rate outside bursts, while having no significant effect on firing rate inside bursts and spike amplitude. Cannabidiol also had no significant effect on the bursting parameters, and no effect on NBSI and jitter, while significantly increasing the Kappa value. Overall, cannabidiol does not appear to have as significant an effect on the activity profile of the

Page 121 network compared to lamotrigine and stiripentol. This may be due to the multifaceted mechanism of action of cannabidiol, resulting in a more nuanced activity profile in neuronal networks (Laux et al. 2019). Cannabidiol also does not appear to significantly alter the activity profile of the network significantly in SCN1A+/- cultures compared to

WT, with the only significant change being a significant reduction in the Kappa value.

There may be a few reasons a significant ‘seizure rescuing’ phenotype was not detected in the SCN1A+/- networks after drug application. Future directions would testing compounds at different concentrations, as there is varying information published regarding therapeutic concentration, and a relevant dose may have been missed in this study (Hawkins et al. 2017; Chiron et al. 2000). There is also value in testing compounds at varying concentrations, as it has been previously reported that compounds can have drastically different effects on MEA networks depending on the applied concentration (Keefer et al. 2001). A wider range of compounds should also be tested, including valproate and clobazam which are first line AEDs used in the treatment of DS. Finally, it would be beneficial to test drugs in combination with each other as this is how they are therapeutically delivered. For example, combining stiripentol with valproate and clobazam to determine whether it may be the drug interactions that result in an overall seizure decrease (Wirrell 2016).

Furthermore, it is not uncommon for genetic models of disease to recapitulate a disease phenotype that does not translate to drug interactions. A recent study evaluated a mouse model of DS for predictive validity in drug discovery by administering a battery of known AEDs to SCN1A+/- mice. Seizures in these models were either spontaneous or

Page 122 hypothermia-induced. While the hypothermia-induced model responded to all drugs in the expected manner, the majority of compounds known to be effective in DS patients did not significantly impact seizure frequency in the spontaneous seizure group

(Hawkins et al. 2017). This is not an isolated case, as predictive validity is an ongoing issue in drug discovery in epilepsy. There is currently no model of DS presenting with spontaneous seizure activity that provides predictive validity (Griffin et al. 2018). As the DS model described in this study displayed an epileptic activity profile without being induced by proconvulsant drugs or outside stimulation, it may be the case that providing this kindling stimulus will strengthen the seizure phenotype and a clearer drug response in SCN1A+/- network discernible from WT may be observed.

A more immediate avenue for further investigation may involve a more complex approach to analysis of the network activity profile. In this instance, parameters have been extracted that are useful for observing individual characteristics of network behavior to generate a total ‘network fingerprint’. These parameters however may not be as useful when combining to observe a total ‘network fingerprint’ as some of the parameters are related and therefore will change in response to the same network activity (for example, NBSI is inversely proportional to NB rate so will always change together). Additionally, in order to statistically compare the activity of entire networks using every extracted parameter, correction for multiple comparisons (such as the

Bonferroni correction) will mask significant differences due to the vast number of compared parameters. This problem will be encountered even after eliminating related parameters to reduce the number of comparisons. This may be resolved by utilizing advanced multiparametric data analysis techniques. Methods for multiparametric data

Page 123 analysis are gaining traction in the field of MEA research as data acquisition and feature extraction methods advance, allowing for increasingly nuanced measurement of network activity. Dimensionality reduction is a useful tool in multiparametric analysis and involves reducing a large number of features into a small number of useful features without losing information from the parameters extracted. This reduces the complexity of the dataset for ease of analysis without having to sacrifice the richness of extracted information. This would allow the transformation of all 10 spiking, bursting and connectivity features described in this study into a single data point for each sample.

This results in a complex comparison of a combination of parameters and also takes into account the interactions between parameters, rather than analyzing each parameter separately. This technique is more sensitive to network changes than individual parameter analysis, and has been utilized in detecting drug effects on MEA networks

(Mendis et al. 2019; Mendis et al. 2016a; Mendis et al. 2014). Support for parameter reduction comes from a study by Giugliano et al. (2004) who conducted a mean field analysis of networks of integrate-and-fire neurons combined with experimental MEA data. It was observed that as few as three generic mechanisms were required to artificially reproduce typical spike and burst patterns observed in complex networks.

Therefore, extracting many more parameters from MEA data may contain redundant information rather than providing further depth, and not aid in revealing insights into different molecular mechanisms. However, analysis of multiple parameters without using dimension reduction is an avenue that has been explored extensively by

NeuroProof, a commercial enterprise that extracts over 200 parameters from raw MEA data in order to identify and capture the unique activity of these networks (Bader et al.

2016). Carrillo-Reid et al. (2015) also reported an alternative to dimension reduction in

Page 124 the form of locally linear embedding and has successfully recapitulated network burst activity from raw spike time stamp data. Utilizing these alternate feature extraction methods may highlight differences in the SCN1A+/- drug response as compared to normal controls not detected in the individual feature comparisons described here.

4.6 Conclusion

From this study, there is enough evidence to conclude the MEA system has successfully detected the effects of different AEDs on the neural networks. These effects were seen across spiking, bursting and network connectivity parameters, generating a detailed activity profile of the effects of these drugs on cultured networks. When comparing the effects of these compounds between DS and wildtype networks, some significant differences were detected, however not to the extent that the conclusion can be drawn that these drugs reduced or increased convulsant activity. We believe that the MEA system has proven to be a promising platform for this kind of testing and that this preliminary study provides direction for more thorough investigation. Future directions will involve pursuing a more detailed analysis procedure, as well as increasing the sample size and testing a wider variety of compounds and drug interactions using this assay.

Page 125

CHAPTER 5| DISCUSSION

The aim of these studies was to a) establish a workflow for utilizing the mwMEA system in high-throughput screening and drug discovery b) utilize this workflow to determine the sensitivity of the system in modelling an epileptic phenotype, and c) assess the efficiency of the system as a platform for drug screening by measuring the reaction to AEDs applied to the epileptic disease networks. I first tested different culturing and maturation parameters in C57BL/6 animals and developed a workflow by which primary cortical mouse neurons generated a stable baseline in 24-well mwMEAs.

Compared to other publications utilising primary cortical mouse neurons, these networks demonstrated a higher spiking and bursting rate despite a similar plating density, highlighting the importance of establishing an internal baseline for further study, as opposed to relying on available literature (McSweeney et al. 2016; Vassallo et al. 2017).

I then tested the utility of the primary neuronal culture-based MEA assay as a potential disease model for genetic epilepsies. Cortical networks were cultured from mice with

SCN1A haploinsufficiency with two distinct genetic backgrounds. On the C57BL/6 genetic background, SCN1A haploinsufficiency results in mice exhibiting spontaneous seizures and suffering sudden death. This is a validated model of DS. In contrast. on the

SV129 genetic background, SCN1A haploinsufficiency results in mice that do not exhibit spontaneous seizures. Spiking, bursting, and connectivity features of the network were compared My results demonstrated remarkable parallels with neuronal cultures generated from DS mice: the seizure-prone C57BL/6 background revealed significantly greater excitability compared to cultures from DS mice on the seizure- Page 126 resistant SV129 background. The DS phenotype was therefore found to be significantly different and demonstrated similar behaviour to other known genetic epilepsy models generated on MEAs (Gullo et al. 2014).

Finally, drugs known to have therapeutic or harmful effects in human DS patients

(stiripentol, lamotrigine and cannabidiol) were applied to the network models of DS on

MEAs. While it was confirmed that these compounds significantly affected the activity profile of the networks, there was not enough significant discrimination of disease compared to control networks to conclude that these drugs have an impact on DS networks specifically. Further analysis of the available data should be undertaken to more robustly examine drug interactions, as well as screening of a wider variety of drugs at different concentrations. This will more thoroughly investigate the promise of this system in drug screening for therapeutic compounds against DS and, by extension, other genetic epilepsy disorders.

While the MEA system appears to be a successful platform for investigating the network properties and drug responses of cultured neurons, there are some caveats. The biggest problem facing the use of MEAs as a platform for investigating phenotypes of genetic diseases is the innate variability of the system. It is evident through the findings

I have reported here and others in the literature that the MEA system must go through a rigorous optimisation process before the parameters for a stable baseline can be generated. These methods appear to be different for each laboratory and cell type used.

The number of samples and degree of preliminary testing required to establish this baseline reduces the promise of MEAs as a high throughput system, as a significant

Page 127 degree of work must be undertaken before any meaningful phenotypic differences can be seen (Vassallo et al. 2017). The sample size required in this study for observing a phenotypic difference between two genetic models is not feasible when attempting high throughput drug screening for individual mutations in precision medicine. A technique that could be employed to reduce the required sample size is to utilize matched pairs within litters to reduce the variability seen between preparations. However, this requires entire preparations to be cultured and potentially discarded based on detected genotype.

Once again this is not a very high throughput process. For this reason, MEAs may be more valuable in measuring responses to drug application, as each sample becomes its own control, and less emphasis is placed on generating consistency between the baseline activity of networks. This repeated measures design also means the sensitivity of the assay is improved, and a lower sample size required, further establishing MEAs as high throughput system for this application. However, the lack of differential effect I have reported between drug application on SCN1A+/- and normal networks may indicate that the same number of samples is required in cases when we want to assess if the drug changes the disease activity signatures towards the control, which is what we expect would provide therapeutic benefits in patients. This again reduces the potential for high throughput drug screening in individual patient mutations.

This limitation may be overcome by adopting a data extraction method that provides a rich categorization of the networks. Improvements in feature extraction have seen a development in more complex methods of analysing a network beyond simple spike and burst reporting. A study by Mendis et al. (2019) utilized a complex feature extraction and analysis method to identify the MOA of two previously unclassified compounds.

Page 128

This employs a method known as dimensionality reduction, which adopts numerous analysis tools historically used in the field of bioinformatics such as principle components analysis and multi-dimensional scaling to reduce multiple characteristics and feature interactions in the networks down to a single data point, as described in section 4.5. This is superior to comparisons made between individual features only as it is more representative of the dynamics of the networks. Another aspect of the networks that may expose further differences is analysing local field potentials, a low frequency component of extracellular signals that represent accumulated network activity surrounding electrodes that critically differ from high frequency components such as action potentials. Investigation of drug effects on LFPs has demonstrated that using a combination of low and high frequency features to sort compounds based on the MOA results is significantly more accurate classification than using LFP or high frequency features alone (Mendis et al. 2016a). The development of more sensitive feature extraction and analysis techniques may, therefore, result in a reduction of necessary sample size when extracting the desired information.

On a larger scale, modelling genetic epilepsy still presents a challenge. While phenotypically validated genetic animal models have existed for decades and have been instrumental in evolving our understanding of the pathomechanisms underlying many genetic epilepsies, there has been little improvement in the treatment for individuals with refractory seizures (Löscher et al. 2013). Five hypotheses have been put forward in an attempt to explain refractory epilepsies. One of these is the gene variant hypothesis, which postulates that pharmacoresistance arises due to genetic variants of proteins involved in the effectiveness of AED mechanisms of action. Another is the target

Page 129 hypothesis, which suggests that epilepsy-related changes in the AED targets (such as ion channels) render these AEDs ineffective (Tang et al. 2017). As these will differ for every individual, they likely will not be incorporated into animal models of EEs.

Additionally, significant differences in gene expression and protein function exist between humans and mice, with 20% of genes showing distinct expression patterns in the cortex alone (Zeng et al. 2012). In order to take the genotypic variation of each patient into account and minimise the interference of non-human genetic expression patterns, precision medicine and personalised therapy using human stem cell derived neurons may be the best way forward (McNeish et al. 2015).

Human iPSC derived neurons have proven to be a successful avenue for modelling genetic epilepsies, including DS. Human iPSC neurons demonstrating an EE phenotype have been engineered through gene editing or obtained using DS patient cells. analysis has shown that iPSC-derived neurons with SCN1A mutations demonstrate significant impairment in the generation of action potentials in GABAergic neurons, consistent with what is seen in other genetic models of DS. These neurons are also pharmacologically validated and respond to AEDs (Jiao et al. 2013; Higurashi et al.

2013). Human iPSC-derived neurons have also been examined on the MEA system and have exhibited the spiking, bursting, and network connectivity profile expected of mature networks (Odawara et al. 2016; Sasaki et al. 2019). Human iPSC models using

MEAs have been generated for neurodevelopmental disorders such as Alzheimer’s

Disease and Autism Spectrum Disorder (Mangan et al. 2018; Marchetto et al. 2016).

The effects of AEDs on induced epileptiform activity in iPSC neuronal networks on

MEAs have also been investigated, demonstrating the potential for this assay to be

Page 130 utilised in drug screening (Odawara et al. 2016). While this is a very promising avenue for discovery it is not yet the perfect system. One limitation is that the timeline for human iPSC maturation can reach 10-13 weeks in vitro compared to the 3 weeks reported in cortical mouse cultures. However, recent advances in neural differentiation have seen the timeline for human iPSC maturation drastically improving, with Izsak et al (2019) reporting a synchronously active neuronal network in aggregate cultures after two weeks of cultivation. A further limitation is that human iPSC networks demonstrate a great degree of variability between networks (Odawara et al. 2016; Hempel et al.

2017). However this variability may be overcome using the analysis techniques described above, as well as utilizing advances in technology as this problem is investigated further.

Overall, this thesis has successfully investigated a promising assay as a platform for modelling genetic epilepsies in the context of precision medicine. A workflow has been generated and a phenotype of a refractory EE successfully identified. This is proof of principle that the DS phenotype is visible on the MEAs in preparation for further studies utilizing patient stem cells. This brings the field of epilepsy research once step closer to realising the ultimate goal of precision medicine.

Page 131

REFERENCES

Abu-Sawwa R, and Stehling C. 2020. 'Epidiolex (Cannabidiol) Primer: Frequently Asked Questions for Patients and Caregivers', The Journal of Pediatric Pharmacology and Therapeutics : JPPT, 25: 75-77. Aguirre AJ, Nowak JA, Camarda ND, Moffitt RA, Ghazani AA, Hazar-Rethinam M, Raghavan S, Kim J, Brais LK, Ragon D, Welch MW, Reilly E, McCabe D, Marini L, Anderka K, Helvie K, Oliver N, Babic A, Da Silva A, Nadres B, Van Seventer EE, Shahzade HA, St. Pierre JP, Burke KP, Clancy T, Cleary JM, Doyle LA, Jajoo K, McCleary NJ, Meyerhardt JA, Murphy JE, Ng K, Patel AK, Perez K, Rosenthal MH, Rubinson DA, Ryou M, Shapiro GI, Sicinska E, Silverman SG, Nagy RJ, Lanman RB, Knoerzer D, Welsch DJ, Yurgelun MB, Fuchs CS, Garraway LA, Getz G, Hornick JL, Johnson BE, Kulke MH, Mayer RJ, Miller JW, Shyn PB, Tuveson DA, Wagle N, Yeh JJ, Hahn WC, Corcoran RB, Carter SL, and Wolpin BM. 2018. 'Real- time Genomic Characterization of Advanced Pancreatic Cancer to Enable Precision Medicine', Cancer Discovery, 8: 1096. Akagi K, Li J, Stephens RM, Volfovsky N, and Symer DE. 2008. 'Extensive variation between inbred mouse strains due to endogenous L1 retrotransposition', Genome Research, 18: 869-80. Albertson AJ, Yang J, and Hablitz JJ. 2011. 'Decreased hyperpolarization-activated currents in layer 5 pyramidal neurons enhances excitability in focal cortical dysplasia', Journal of neurophysiology, 106: 2189-200. Allen AS, Berkovic SF, Cossette P, Delanty N, Dlugos D, Eichler EE, Epstein MP, Glauser T, Goldstein DB, Han Y, Heinzen EL, Hitomi Y, Howell KB, Johnson MR, Kuzniecky R, Lowenstein DH, Lu Y-F, Madou MRZ, Marson AG, Mefford HC, Esmaeeli Nieh S, O'Brien TJ, Ottman R, Petrovski S, Poduri A, Ruzzo EK, Scheffer IE, Sherr EH, Yuskaitis CJ, Abou-Khalil B, Alldredge BK, Bautista JF, Berkovic SF, Boro A, Cascino GD, Consalvo D, Crumrine P, Devinsky O, Dlugos D, Epstein MP, Fiol M, Fountain NB, French J, Friedman D, Geller EB, Glauser T, Glynn S, Haut SR, Hayward J, Helmers SL, Joshi S, Kanner A, Kirsch HE, Knowlton RC, Kossoff EH, Kuperman R, Kuzniecky R, Lowenstein DH, McGuire SM, Motika PV, Novotny EJ, Ottman R, Paolicchi JM, Parent JM, Park K, Poduri A, Scheffer IE, Shellhaas RA, Sherr EH, Shih JJ, Singh R, Sirven J, Smith MC, Sullivan J, Lin Thio L, Venkat A, Vining EPG, Von Allmen GK, Weisenberg JL, Widdess-Walsh P, and Winawer MR. 2013. 'De novo mutations in epileptic encephalopathies', Nature, 501: 217-21. Allene C, Cattani A, Ackman JB, Bonifazi P, Aniksztejn L, Ben-Ari Y, and Cossart R. 2008. 'Sequential generation of two distinct synapse-driven network patterns in developing neocortex', Journal of Neuroscience, 28: 12851-63. Anderson LL, Hawkins NA, Thompson CH, Kearney JA, and George AL. 2017. 'Unexpected Efficacy of a Novel Sodium Channel Modulator in Dravet Syndrome', Scientific Reports, 7: 1682. Appenzeller S, Balling R, Barisic N, Baulac S, Caglayan H, Craiu D, De Jonghe P, Depienne C, Dimova P, and Djémié T. 2014. 'De novo mutations in synaptic

Page 132

transmission genes including DNM1 cause epileptic encephalopathies', The American Journal of Human Genetics, 95: 360-70. Arnold FJ, Hofmann F, Bengtson CP, Wittmann M, Vanhoutte P, and Bading H. 2005. 'Microelectrode array recordings of cultured hippocampal networks reveal a simple model for transcription and protein synthesis‐dependent plasticity', The Journal of Physiology, 564: 3-19. Bader B, Ehnert C, Jügelt K, and Schröder U. 2016. Functional phenotypic comparison of activity patterns from different mouse brain region-specific neuron cultures with developing human iPSC-derived neuronal networks. Baltz T, De Lima A, and Voigt T. 2010. 'Contribution of GABAergic interneurons to the development of spontaneous activity patterns in cultured neocortical networks', Frontiers in Cellular Neuroscience, 4. Baltz T, Herzog A, and Voigt T. 2011. 'Slow oscillating population activity in developing cortical networks: models and experimental results', Journal of neurophysiology, 106: 1500-14. Baltz T, and Voigt T. 2015. 'Interaction of electrically evoked activity with intrinsic dynamics of cultured cortical networks with and without functional fast GABAergic synaptic transmission', Frontiers in Cellular Neuroscience, 9. Bando SY, Silva FN, Costa LdF, Silva AV, Pimentel-Silva LR, Castro LHM, Wen H-T, Amaro E, Jr., and Moreira-Filho CA. 2013. 'Complex Network Analysis of CA3 Transcriptome Reveals Pathogenic and Compensatory Pathways in Refractory Temporal Lobe Epilepsy', PLoS ONE, 8: e79913. Bechi G, Scalmani P, Schiavon E, Rusconi R, Franceschetti S, and Mantegazza M. 2012. 'Pure haploinsufficiency for Dravet syndrome Na(V)1.1 (SCN1A) sodium channel truncating mutations', Epilepsia, 53: 87-100. Ben-Ari Y, Gaiarsa J-L, Tyzio R, and Khazipov R. 2007. 'GABA: a pioneer transmitter that excites immature neurons and generates primitive oscillations', Physiological reviews, 87: 1215-84. Bender AC, Morse RP, Scott RC, Holmes GL, and Lenck-Santini P-P. 2012. 'SCN1A mutations in Dravet syndrome: impact of interneuron dysfunction on neural networks and cognitive outcome', Epilepsy Behav, 23: 177-86. Berg AT, Berkovic SF, Brodie MJ, Buchhalter J, Cross JH, van Emde Boas W, Engel J, French J, Glauser TA, Mathern GW, Moshe SL, Nordli D, Plouin P, and Scheffer IE. 2010. 'Revised terminology and concepts for organization of seizures and epilepsies: report of the ILAE Commission on Classification and Terminology, 2005-2009', Epilepsia, 51: 676-85. Biffi E, Regalia G, Menegon A, Ferrigno G, and Pedrocchi A. 2013. 'The Influence of Neuronal Density and Maturation on Network Activity of Hippocampal Cell Cultures: A Methodological Study', PLoS ONE, 8: 1-16. Bradley JA, Luithardt HH, Metea MR, and Strock CJ. 2018. 'In Vitro Screening for Seizure Liability Using Microelectrode Array Technology', Toxicological Sciences, 163: 240-53. Brunklaus A, and Zuberi SM. 2014. Dravet syndrome—From epileptic encephalopathy to channelopathy.

Page 133

Carrillo-Reid L, Lopez-Huerta VG, Garcia-Munoz M, Theiss S, and Arbuthnott GW. 2015. 'Cell assembly signatures defined by short-term synaptic plasticity in cortical networks', International journal of neural systems, 25: 1550026. Catterall WA. 2018. 'Dravet Syndrome: A Sodium Channel Interneuronopathy', Current opinion in physiology, 2: 42-50. Cetica V, Chiari S, Mei D, Parrini E, Grisotto L, Marini C, Pucatti D, Ferrari A, Sicca F, Specchio N, Trivisano M, Battaglia D, Contaldo I, Zamponi N, Petrelli C, Granata T, Ragona F, Avanzini G, and Guerrini R. 2017. 'Clinical and genetic factors predicting Dravet syndrome in infants with SCN1A mutations', Neurology, 88: 1037-44. Charlesworth P, Morton A, Eglen SJ, Komiyama NH, and Grant SGN. 2016. 'Canalization of genetic and pharmacological perturbations in developing primary neuronal activity patterns', Neuropharmacology, 100: 47-55. Cheah CS, Yu FH, Westenbroek RE, Kalume FK, Oakley JC, Potter GB, Rubenstein JL, and Catterall WA. 2012. 'Specific deletion of NaV1.1 sodium channels in inhibitory interneurons causes seizures and premature death in a mouse model of Dravet syndrome', Proceedings of the National Academy of Sciences, 109: 14646-51. Chen KA, Farrar M, Cardamone M, Gill D, Smith R, Cowell CT, Truong L, and Lawson JA. 2018. 'Cannabidiol for treating drug‐resistant epilepsy in children: the New South Wales experience', Medical Journal of Australia, 209: 217-21. Chiron C, and Dulac O. 2011. 'The pharmacologic treatment of Dravet syndrome', Epilepsia, 52: 72-75. Chiron C, Marchand MC, Tran A, Rey E, d'Athis P, Vincent J, Dulac O, and Pons G. 2000. 'Stiripentol in severe myoclonic epilepsy in infancy: a randomised placebo- controlled syndrome-dedicated trial', The Lancet, 356: 1638-42. Cho MJ, Kwon SS, Ko A, Lee ST, Lee YM, Kim HD, Chung HJ, Kim SH, Lee JS, Kim DS, and Kang HC. 2018. 'Efficacy of Stiripentol in Dravet Syndrome with or without SCN1A Mutations', Journal of clinical neurology (Seoul, Korea), 14: 22-28. Claes L, Del-Favero J, Ceulemans B, Lagae L, Van Broeckhoven C, and De Jonghe P. 2001. 'De Novo Mutations in the Sodium-Channel Gene SCN1A Cause Severe Myoclonic Epilepsy of Infancy', The American Journal of Human Genetics, 68: 1327-32. Collins F, and Varmus H. 2015. 'A New Initiative on Precision Medicine', The New England Journal of Medicine, 372: 793-95. Condren ME, and Bradshaw MD. 2013. 'Ivacaftor: a novel gene-based therapeutic approach for cystic fibrosis', The journal of pediatric pharmacology and therapeutics : JPPT : the official journal of PPAG, 18: 8-13. Consortium E. 2012. 'Epi4K: Gene discovery in 4,000 genomes', Epilepsia, 53: 1457-67. Consortium TEK. 2017. 'Phenotypic analysis of 303 multiplex families with common epilepsies', Brain, 140: 2144-56. Cooper MS, Mcintosh A, Crompton DE, McMahon JM, Schneider A, Farrell K, Ganesan V, Gill D, Kivity S, and Lerman-Sagie T. 2016. 'Mortality in Dravet syndrome', Epilepsy Res, 128: 43-47.

Page 134

Cotterill E, Hall D, Wallace K, Mundy WR, Eglen SJ, and Shafer TJ. 2016. 'Characterization of Early Cortical Neural Network Development in Multiwell Microelectrode Array Plates', Journal of Biomolecular Screening, 21: 510-19. Delisle BP, and January CT. 2017. 'Advancing precision medicine for the treatment of long-QT syndrome type 2: shedding light on lumacaftor', European Heart Journal, 39: 1456-58. Devinsky O, Cilio MR, Cross H, Fernandez‐Ruiz J, French J, Hill C, Katz R, Di Marzo V, Jutras‐Aswad D, and Notcutt WG. 2014. 'Cannabidiol: pharmacology and potential therapeutic role in epilepsy and other neuropsychiatric disorders', Epilepsia, 55: 791-802. Devinsky O, Cross JH, Laux L, Marsh E, Miller I, Nabbout R, Scheffer IE, Thiele EA, and Wright S. 2017. 'Trial of cannabidiol for drug-resistant seizures in the Dravet syndrome', New England Journal of Medicine, 376: 2011-20. Dhindsa RS, and Goldstein DB. 2015. 'Genetic Discoveries Drive Molecular Analyses and Targeted Therapeutic Options in the Epilepsies', Current Neurology and Neuroscience Reports, 15: 70. Dravet C, Bureau M, Oguni H, Fukuyama Y, and Cokar O. 2005. 'Severe myoclonic epilepsy in infancy: Dravet syndrome', Adv Neurol, 95: 71-102. Dutton SB, Makinson CD, Papale LA, Shankar A, Balakrishnan B, Nakazawa K, and Escayg A. 2013. 'Preferential inactivation of Scn1a in parvalbumin interneurons increases seizure susceptibility', Neurobiol Dis, 49: 211-20. Eder A, Vollert I, Hansen A, and Eschenhagen T. 2016. 'Human engineered heart tissue as a model system for drug testing', Advanced Drug Delivery Reviews, 96: 214- 24. Eltze CM, Chong WK, Cox T, Whitney A, Cortina-Borja M, Chin RFM, Scott RC, and Cross JH. 2013. 'A population-based study of newly diagnosed epilepsy in infants', Epilepsia, 54: 437-45. Engel J. 2013. Seizures and epilepsy (Oxford University Press). Epi PMC. 2015. 'A roadmap for precision medicine in the epilepsies', The Lancet. Neurology, 14: 1219-28. Falco-Walter JJ, Scheffer IE, and Fisher RS. 2018. 'The new definition and classification of seizures and epilepsy', Epilepsy Res, 139: 73-79. Feng L, Grosse S, Green R, Fink A, and Sawicki GS. 2018. 'Precision Medicine In Action: The Impact Of Ivacaftor On Cystic Fibrosis–Related Hospitalizations', Health Affairs, 37: 773-79. Ferraro TN, Golden GT, Smith GG, DeMuth D, Buono RJ, and Berrettini WH. 2002. 'Mouse strain variation in maximal electroshock seizure threshold', Brain Research, 936: 82-86. Fisher JL. 2009. 'The anti-convulsant stiripentol acts directly on the GABAA receptor as a positive allosteric modulator', Neuropharmacology, 56: 190-97. Fisher JL. 2011. 'The effects of stiripentol on GABAA receptors', Epilepsia, 52: 76-78. Fisher R, Acevedo C, Arzimanoglou A, Bogacz A, Cross H, Elger C, Engel Jr J, Forsgren L, French J, and Glynn M. 2014. 'ILAE official report: a practical clinical definition of epilepsy', Epilepsia, 55: 475-82.

Page 135

Fisher RS, Boas WvE, Blume W, Elger C, Genton P, Lee P, and Engel Jr. J. 2005. 'Epileptic Seizures and Epilepsy: Definitions Proposed by the International League Against Epilepsy (ILAE) and the International Bureau for Epilepsy (IBE)', Epilepsia, 46: 470-72. Fisher RS, and Kaplan P. 2004. Imitators of epilepsy (Demos Medical Publishing). Franco V, French JA, and Perucca E. 2016. 'Challenges in the clinical development of new antiepileptic drugs', Pharmacological Research, 103: 95-104. Frankel WN. 2009. 'Genetics of complex neurological disease: challenges and opportunities for modeling epilepsy in mice and rats', Trends in genetics : TIG, 25: 361-67. Garaschuk O, Linn J, Eilers J, and Konnerth A. 2000. 'Large-scale oscillatory calcium waves in the immature cortex', Nat Neurosci, 3: 452-59. Garofalo M, Nieus T, Massobrio P, and Martinoia S. 2009. 'Evaluation of the Performance of Information Theory-Based Methods and Cross-Correlation to Estimate the Functional Connectivity in Cortical Networks', PLoS ONE, 4: e6482. Gaston TE, and Friedman D. 2017. 'Pharmacology of cannabinoids in the treatment of epilepsy', Epilepsy & Behavior, 70: 313-18. Genton P, Velizarova R, and Dravet C. 2011. 'Dravet syndrome: The long-term outcome', Epilepsia, 52: 44-49. Giraud C, Treluyer J-M, Rey E, Chiron C, Vincent J, Pons G, and Tran A. 2006. 'In vitro and in vivo inhibitory effect of stiripentol on clobazam metabolism', Drug metabolism and disposition, 34: 608-11. Giugliano M, Darbon P, Arsiero M, Lüscher H-R, and Streit J. 2004. 'Single-Neuron Discharge Properties and Network Activity in Dissociated Cultures of Neocortex', Journal of neurophysiology, 92: 977-96. Greenhill S, and Jones RS. 2010. 'Diverse antiepileptic drugs increase the ratio of background synaptic inhibition to excitation and decrease neuronal excitability in neurones of the rat entorhinal cortex in vitro', Neuroscience, 167: 456-74. Griffin A, Hamling KR, Hong S, Anvar M, Lee LP, and Baraban SC. 2018. 'Preclinical Animal Models for Dravet Syndrome: Seizure Phenotypes, Comorbidities and Drug Screening', Frontiers in pharmacology, 9: 573. Guerrini R, Dravet C, Genton P, Belmonte A, Kaminska A, and Dulac O. 1998. 'Lamotrigine and seizure aggravation in severe myoclonic epilepsy', Epilepsia, 39: 508-12. Gullo F, Manfredi I, Lecchi M, Casari G, Wanke E, and Becchetti A. 2014. 'Multi- electrode array study of neuronal cultures expressing nicotinic β2-V287L subunits, linked to autosomal dominant nocturnal frontal lobe epilepsy. An in vitro model of spontaneous epilepsy', Frontiers in Neural Circuits, 8. Habibey R, Latifi S, Mousavi H, Pesce M, Arab-Tehrany E, and Blau A. 2017. 'A multielectrode array microchannel platform reveals both transient and slow changes in axonal conduction velocity', Scientific Reports, 7: 8558. Han S, Tai C, E Westenbroek R, H Yu F, Cheah C, Potter G, L Rubenstein J, Scheuer T, de la Iglesia H, and Catterall W. 2012. Autistic-like behaviour in Scn1a(+/-) mice and rescue by enhanced GABA-mediated neurotransmission.

Page 136

Hawkins NA, Anderson LL, Gertler TS, Laux L, George Jr AL, and Kearney JA. 2017. 'Screening of conventional anticonvulsants in a genetic mouse model of epilepsy', Annals of clinical and translational neurology, 4: 326-39. Hedrich UB, Liautard C, Kirschenbaum D, Pofahl M, Lavigne J, Liu Y, Theiss S, Slotta J, Escayg A, Dihne M, Beck H, Mantegazza M, and Lerche H. 2014. 'Impaired action potential initiation in GABAergic interneurons causes hyperexcitable networks in an epileptic mouse model carrying a human Na(V)1.1 mutation', J Neurosci, 34: 14874-89. Hempel CM, Werley CA, Dempsey GT, and Gerber DJ. 2017. 'Targeting neuronal function for CNS drug discovery', Drug Discovery Today: Technologies, 23: 17- 25. Hesdorffer DC, Logroscino G, Benn EK, Katri N, Cascino G, and Hauser WA. 2011. 'Estimating risk for developing epilepsy: a population-based study in Rochester, Minnesota', Neurology, 76: 23-7. Higurashi N, Uchida T, Lossin C, Misumi Y, Okada Y, Akamatsu W, Imaizumi Y, Zhang B, Nabeshima K, Mori MX, Katsurabayashi S, Shirasaka Y, Okano H, and Hirose S. 2013. 'A human Dravet syndrome model from patient induced pluripotent stem cells', Molecular Brain, 6: 19. Illes S, Fleischer W, Siebler M, Hartung H-P, and Dihné M. 2007. 'Development and pharmacological modulation of embryonic stem cell-derived neuronal network activity', Experimental Neurology, 207: 171-76. Ito S, Ogiwara I, Yamada K, Miyamoto H, Hensch TK, Osawa M, and Yamakawa K. 2013. 'Mouse with Nav1.1 haploinsufficiency, a model for Dravet syndrome, exhibits lowered sociability and learning impairment', Neurobiol Dis, 49: 29-40. Izsak J, Seth H, Andersson M, Vizlin-Hodzic D, Theiss S, Hanse E, Ågren H, Funa K, and Illes S. 2019. 'Robust Generation of Person-Specific, Synchronously Active Neuronal Networks Using Purely Isogenic Human iPSC-3D Neural Aggregate Cultures', Frontiers in Neuroscience, 13. Jain P, Sharma S, and Tripathi M. 2013. 'Diagnosis and Management of Epileptic Encephalopathies in Children', Epilepsy Research and Treatment, 2013: 9. Jewett KA, Christian CA, Bacos JT, Lee KY, Zhu J, and Tsai N-P. 2016. 'Feedback modulation of neural network synchrony and seizure susceptibility by Mdm2- p53-Nedd4-2 signaling', Molecular Brain, 9: 32. Jiao J, Yang Y, Shi Y, Chen J, Gao R, Fan Y, Yao H, Liao W, Sun X-F, and Gao S. 2013. 'Modeling Dravet syndrome using induced pluripotent stem cells (iPSCs) and directly converted neurons', Human Molecular Genetics, 22: 4241-52. Johnstone AFM, Gross GW, Weiss DG, Schroeder OHU, Gramowski A, and Shafer TJ. 2010. 'Microelectrode arrays: A physiologically based neurotoxicity testing platform for the 21st century', NeuroToxicology, 31: 331-50. Kalume F, Frank HY, Westenbroek RE, Scheuer T, and Catterall WA. 2007. 'Reduced sodium current in Purkinje neurons from Nav1. 1 mutant mice: implications for ataxia in severe myoclonic epilepsy in infancy', Journal of Neuroscience, 27: 11065-74.

Page 137

Kaplan JS, Stella N, Catterall WA, and Westenbroek RE. 2017. 'Cannabidiol attenuates seizures and social deficits in a mouse model of Dravet syndrome', Proceedings of the National Academy of Sciences, 114: 11229-34. Keefer EW, Norton SJ, Boyle NA, Talesa V, and Gross GW. 2001. 'Acute toxicity screening of novel AChE inhibitors using neuronal networks on microelectrode arrays', NeuroToxicology, 22: 3-12. Khan S, and Al Baradie R. 2012. 'Epileptic Encephalopathies: An Overview', Epilepsy Research and Treatment, 2012: 8. Knupp KG, and Wirrell EC. 2018. 'Treatment Strategies for Dravet Syndrome', CNS Drugs, 32: 335-50. Kole MH, Ilschner SU, Kampa BM, Williams SR, Ruben PC, and Stuart GJ. 2008. 'Action potential generation requires a high sodium channel density in the axon initial segment', Nat Neurosci, 11: 178. Kumar HKMCC, Das S, Ghosh LK, and Gupta B. 2012. Epilepsy and its Management: A Review. LaSarge C, and Danzer S. 2014. 'Mechanisms regulating neuronal excitability and seizure development following mTOR pathway hyperactivation', Frontiers in Molecular Neuroscience, 7. Laux LC, Bebin EM, Checketts D, Chez M, Flamini R, Marsh ED, Miller I, Nichol K, Park Y, Segal E, Seltzer L, Szaflarski JP, Thiele EA, and Weinstock A. 2019. 'Long-term safety and efficacy of cannabidiol in children and adults with treatment resistant Lennox-Gastaut syndrome or Dravet syndrome: Expanded access program results', Epilepsy Res, 154: 13-20. Li BM, Liu XR, Yi YH, Deng YH, Su T, Zou X, and Liao WP. 2011. 'Autism in Dravet syndrome: prevalence, features, and relationship to the clinical characteristics of epilepsy and mental retardation', Epilepsy Behav, 21: 291-5. Li X, Zhang R, Zhao B, Lossin C, and Cao Z. 2016. 'Cardiotoxicity screening: a review of rapid-throughput in vitro approaches', Archives of toxicology, 90: 1803-16. Lin W-H, Giachello CNG, and Baines RA. 2017. 'Seizure control through genetic and pharmacological manipulation of Pumilio in Drosophila: a key component of neuronal homeostasis', Disease Models & Mechanisms, 10: 141-50. Löscher W, Klitgaard H, Twyman RE, and Schmidt D. 2013. 'New avenues for anti- epileptic drug discovery and development', Nature reviews drug discovery, 12: 757. Low SK, Zembutsu H, and Nakamura Y. 2018. 'Breast cancer: The translation of big genomic data to cancer precision medicine', Cancer science, 109: 497-506. Lu C, Chen Q, Zhou T, Bozic D, Fu Z, Pan JQ, and Feng G. 2015. 'Micro-electrode array recordings reveal reductions in both excitation and inhibition in cultured cortical neuron networks lacking Shank3', Molecular Psychiatry, 21: 159. Maljevic S, Reid CA, and Petrou S. 2017. 'Models for discovery of targeted therapy in genetic epileptic encephalopathies', Journal of neurochemistry, 143: 30-48. Mangan K, Harms L, Kannemeier C, Kim K-H, Bader BM, Jügelt K, Schröder O, Freitas B, and Jones E. 2018. 'IDENTIFICATION OF MEASURABLE PHENOTYPES RELEVANT

Page 138

TO ALZHEIMER’S DISEASE USING HUMAN IPSC-DERIVED NEURONS', Alzheimer's & Dementia, 14: P1130. Marchetto MC, Belinson H, Tian Y, Freitas BC, Fu C, Vadodaria KC, Beltrao-Braga PC, Trujillo CA, Mendes APD, Padmanabhan K, Nunez Y, Ou J, Ghosh H, Wright R, Brennand KJ, Pierce K, Eichenfield L, Pramparo T, Eyler LT, Barnes CC, Courchesne E, Geschwind DH, Gage FH, Wynshaw-Boris A, and Muotri AR. 2016. 'Altered proliferation and networks in neural cells derived from idiopathic autistic individuals', Molecular Psychiatry, 22: 820. Marini C, Scheffer IE, Nabbout R, Suls A, De Jonghe P, Zara F, and Guerrini R. 2011. 'The genetics of Dravet syndrome', Epilepsia, 52: 24-29. Martin MS, Dutt K, Papale LA, Dube CM, Dutton SB, de Haan G, Shankar A, Tufik S, Meisler MH, Baram TZ, Goldin AL, and Escayg A. 2010. 'Altered function of the SCN1A voltage-gated sodium channel leads to gamma-aminobutyric acid-ergic (GABAergic) interneuron abnormalities', J Biol Chem, 285: 9823-34. McConnell ER, McClain MA, Ross J, Lefew WR, and Shafer TJ. 2012. 'Evaluation of multi-well microelectrode arrays for neurotoxicity screening using a chemical training set', NeuroToxicology, 33: 1048-57. McHugh ML. 2012. 'Interrater reliability: the kappa statistic', Biochemia medica, 22: 276-82. McNeish J, Gardner Jason P, Wainger Brian J, Woolf Clifford J, and Eggan K. 2015. 'From Dish to Bedside: Lessons Learned While Translating Findings from a Stem Cell Model of Disease to a Clinical Trial', Cell Stem Cell, 17: 8-10. McSweeney KM, Gussow AB, Bradrick SS, Dugger SA, Gelfman S, Wang Q, Petrovski S, Frankel WN, Boland MJ, and Goldstein DB. 2016. 'Inhibition of microRNA 128 promotes excitability of cultured cortical neuronal networks', Genome Research, 26: 1411-16. McTague A, Howell KB, Cross JH, Kurian MA, and Scheffer IE. 2016. 'The genetic landscape of the epileptic encephalopathies of infancy and childhood', The Lancet Neurology, 15: 304-16. Mendis DC, Halgamuge SK, Morrisroe E, and Petrou S. 2014. "Changes in propagation delays for quantifying pharmacological effects on cortical cultures." In 7th International Conference on Information and Automation for Sustainability, 1-6. Mendis GD, Morrisroe E, Reid CA, Halgamuge SK, and Petrou S. 2016a. 'Use of local field potentials of dissociated cultures grown on multi-electrode arrays for pharmacological assays', Conf Proc IEEE Eng Med Biol Soc, 2016: 952-56. Mendis GDC, Berecki G, Morrisroe E, Pachernegg S, Li M, Varney M, Osborne PB, Reid CA, Halgamuge S, and Petrou S. 2019. 'Discovering the pharmacodynamics of conolidine and cannabidiol using a cultured neuronal network based workflow', Scientific Reports, 9: 121. Mendis GDC, Morrisroe E, Petrou S, and Halgamuge SK. 2016b. 'Use of adaptive network burst detection methods for multielectrode array data and the generation of artificial spike patterns for method evaluation', Journal of , 13: 026009.

Page 139

Miller AR, Hawkins NA, McCollom CE, and Kearney JA. 2014. 'Mapping genetic modifiers of survival in a mouse model of Dravet syndrome', Genes Brain Behav, 13: 163-72. Moody WJ, and Bosma MM. 2005. 'Ion channel development, spontaneous activity, and activity-dependent development in nerve and muscle cells', Physiological reviews, 85: 883-941. Nakagawa H, and Fujita M. 2018. 'Whole genome sequencing analysis for cancer genomics and precision medicine', Cancer science, 109: 513-22. Namer B, Schmidt D, Eberhardt E, Maroni M, Dorfmeister E, Kleggetveit IP, Kaluza L, Meents J, Gerlach A, Lin Z, Winterpacht A, Dragicevic E, Kohl Z, Schüttler J, Kurth I, Warncke T, Jorum E, Winner B, and Lampert A. 2019. 'Pain relief in a neuropathy patient by lacosamide: Proof of principle of clinical translation from patient-specific iPS cell-derived nociceptors', EBioMedicine, 39: 401-08. Navarrete EG, Liang P, Lan F, Sanchez-Freire V, Simmons C, Gong T, Sharma A, Burridge PW, Patlolla B, Lee AS, Wu H, Beygui RE, Wu SM, Robbins RC, Bers DM, and Wu JC. 2013. 'Screening drug-induced arrhythmia [corrected] using human induced pluripotent stem cell-derived cardiomyocytes and low-impedance microelectrode arrays', Circulation, 128: S3-13. Ngugi AK, Bottomley C, Kleinschmidt I, Sander JW, and Newton CR. 2010. 'Estimation of the burden of active and life-time epilepsy: a meta-analytic approach', Epilepsia, 51: 883-90. Nissenkorn A, Almog Y, Adler I, Safrin M, Brusel M, Marom M, Bercovich S, Yakubovich D, Tzadok M, Ben-Zeev B, and Rubinstein M. 2019. 'In vivo, in vitro and in silico correlations of four de novo SCN1A missense mutations', PLoS ONE, 14: e0211901. Nör J, and Gutkind J. 2018. "Head and neck cancer in the new era of precision medicine." In.: SAGE Publications Sage CA: Los Angeles, CA. Novellino A, Scelfo B, Palosaari T, Price A, Sobanski T, Shafer T, Johnstone A, Gross G, Gramowski A, Schroeder O, Jügelt K, Chiappalone M, Benfenati F, Martinoia S, Tedesco M, Defranchi E, D'Angelo P, and Whelan M. 2011. 'Development of Micro-Electrode Array Based Tests for Neurotoxicity: Assessment of Interlaboratory Reproducibility with Neuroactive Chemicals', Frontiers in Neuroengineering, 4. Oakley JC, Kalume F, Yu FH, Scheuer T, and Catterall WA. 2009. 'Temperature- and age- dependent seizures in a mouse model of severe myoclonic epilepsy in infancy', Proc Natl Acad Sci U S A, 106: 3994-9. Obien MEJ, Deligkaris K, Bullmann T, Bakkum DJ, and Frey U. 2015. 'Revealing neuronal function through microelectrode array recordings', Frontiers in Neuroscience, 8. Odawara A, Katoh H, Matsuda N, and Suzuki I. 2016. 'Physiological maturation and drug responses of human induced pluripotent stem cell-derived cortical neuronal networks in long-term culture', Scientific Reports, 6: 26181. Ogiwara I, Iwasato T, Miyamoto H, Iwata R, Yamagata T, Mazaki E, Yanagawa Y, Tamamaki N, Hensch TK, Itohara S, and Yamakawa K. 2013. 'Nav1.1 haploinsufficiency in excitatory neurons ameliorates seizure-associated sudden

Page 140

death in a mouse model of Dravet syndrome', Human Molecular Genetics, 22: 4784-804. Ogiwara I, Miyamoto H, Morita N, Atapour N, Mazaki E, Inoue I, Takeuchi T, Itohara S, Yanagawa Y, Obata K, Furuichi T, Hensch TK, and Yamakawa K. 2007. 'Nav1.1 localizes to of parvalbumin-positive inhibitory interneurons: a circuit basis for epileptic seizures in mice carrying an Scn1a gene mutation', The Journal of neuroscience : the official journal of the Society for Neuroscience, 27: 5903-14. Oyrer J, Maljevic S, Scheffer IE, Berkovic SF, Petrou S, and Reid CA. 2018. 'Ion channels in genetic epilepsy: from genes and mechanisms to disease-targeted therapies', Pharmacological reviews, 70: 142-73. Pastore VP, Godjoski A, Martinoia S, and Massobrio P. 2018. 'SpiCoDyn: A Toolbox for the Analysis of Neuronal Network Dynamics and Connectivity from Multi-Site Spike Signal Recordings', Neuroinformatics, 16: 15-30. Pastore VP, Poli D, Godjoski A, Martinoia S, and Massobrio P. 2016. 'ToolConnect: A Functional Connectivity Toolbox for In vitro Networks', Frontiers in neuroinformatics, 10: 13-13. Pauli C, Hopkins BD, Prandi D, Shaw R, Fedrizzi T, Sboner A, Sailer V, Augello M, Puca L, and Rosati R. 2017. 'Personalized in vitro and in vivo cancer models to guide precision medicine', Cancer Discovery, 7: 462-77. Poduri A, and Lowenstein D. 2011. 'Epilepsy genetics--past, present, and future', Current opinion in genetics & development, 21: 325-32. Poli D, Pastore VP, Martinoia S, and Massobrio P. 2016. 'From functional to structural connectivity using partial correlation in neuronal assemblies', Journal of neural engineering, 13: 026023. Quilichini PP, Chiron C, Ben-Ari Y, and Gozlan H. 2006. 'Stiripentol, a putative antiepileptic drug, enhances the duration of opening of GABAA-receptor channels', Epilepsia, 47: 704-16. Regner GG, Pereira P, Leffa DT, de Oliveira C, Vercelino R, Fregni F, and Torres ILS. 2018. 'Preclinical to Clinical Translation of Studies of Transcranial Direct-Current Stimulation in the Treatment of Epilepsy: A Systematic Review', Frontiers in Neuroscience, 12. Reimer T, Baumann W, and Gimsa J. 2012. 'Population burst of parvalbumin-positive interneurons inhibit spiking pyramidal cells in spontaneously active cortical in vitro networks', J. Chem, 6: 1033-42. Rivera J, and Tessarollo L. 2008. 'Genetic Background and the Dilemma of Translating Mouse Studies to Humans', Immunity, 28: 1-4. Rubinstein M, Westenbroek RE, Yu FH, Jones CJ, Scheuer T, and Catterall WA. 2015. 'Genetic background modulates impaired excitability of inhibitory neurons in a mouse model of Dravet syndrome', Neurobiology of Disease, 73: 106-17. Salgueiro-Pereira AR, Duprat F, Pousinha PA, Loucif A, Douchamps V, Regondi C, Ayrault M, Eugie M, Stunault MI, Escayg A, Goutagny R, Gnatkovsky V, Frassoni C, Marie H, Bethus I, and Mantegazza M. 2019. 'A two-hit story: Seizures and genetic mutation interaction sets phenotype severity in SCN1A epilepsies', Neurobiology of Disease, 125: 31-44.

Page 141

Samanta D. 2019. 'Cannabidiol: A Review of Clinical Efficacy and Safety in Epilepsy', Pediatric Neurology, 96: 24-29. Sandberg R. 2000. 'Gene expression profiling of brain regions in inbred mouse strains reveals candidate genes for phenotypic variation', Masters Thesis in Biomedicine. Stockholm: Karolinska Institute. Sasaki T, Suzuki I, Yokoi R, Sato K, and Ikegaya Y. 2019. 'Synchronous spike patterns in differently mixed cultures of human iPSC-derived glutamatergic and GABAergic neurons', Biochemical and Biophysical Research Communications, 513: 300-05. Schauwecker PE. 2011. 'The relevance of individual genetic background and its role in animal models of epilepsy', Epilepsy Res, 97: 1-11. Scheffer IE, Berkovic S, Capovilla G, Connolly MB, French J, Guilhoto L, Hirsch E, Jain S, Mathern GW, Moshé SL, Nordli DR, Perucca E, Tomson T, Wiebe S, Zhang Y-H, and Zuberi SM. 2017. 'ILAE classification of the epilepsies: Position paper of the ILAE Commission for Classification and Terminology', Epilepsia, 58: 512-21. Scheffer IE, Zhang Y-H, Jansen FE, and Dibbens L. 2009. 'Dravet syndrome or genetic (generalized) epilepsy with febrile seizures plus?', Brain and Development, 31: 394-400. Schutte SS, Schutte RJ, Barragan EV, and O'Dowd DK. 2016. 'Model systems for studying cellular mechanisms of SCN1A-related epilepsy', Journal of neurophysiology, 115: 1755-66. Sinoo C, de Lange IM-L, Westers P, Gunning WB, Jongmans MJ, and Brilstra EH. 2019. 'Behavior problems and health-related quality of life in Dravet syndrome', Epilepsy & Behavior, 90: 217-27. Spira ME, and Hai A. 2013. 'Multi-electrode array technologies for neuroscience and cardiology', Nat Nanotechnol, 8: 83-94. Steinlein OK, Mulley JC, Propping P, Wallace RH, Phillips HA, Sutherland GR, Scheffer IE, and Berkovic SF. 1995. 'A missense mutation in the neuronal nicotinic acetylcholine receptor alpha 4 subunit is associated with autosomal dominant nocturnal frontal lobe epilepsy', Nat Genet, 11: 201-3. Striano P, Vari MS, Mazzocchetti C, Verrotti A, and Zara F. 2016. 'Management of genetic epilepsies: From empirical treatment to precision medicine', Pharmacological Research, 107: 426-29. Strickland JD, Martin MT, Richard AM, Houck KA, and Shafer TJ. 2018. 'Screening the ToxCast phase II libraries for alterations in network function using cortical neurons grown on multi-well microelectrode array (mwMEA) plates', Arch Toxicol, 92: 487-500. Suls A, Jaehn JA, Kecskes A, Weber Y, Weckhuysen S, Craiu DC, Siekierska A, Djemie T, Afrikanova T, Gormley P, von Spiczak S, Kluger G, Iliescu CM, Talvik T, Talvik I, Meral C, Caglayan HS, Giraldez BG, Serratosa J, Lemke JR, Hoffman-Zacharska D, Szczepanik E, Barisic N, Komarek V, Hjalgrim H, Moller RS, Linnankivi T, Dimova P, Striano P, Zara F, Marini C, Guerrini R, Depienne C, Baulac S, Kuhlenbaumer G, Crawford AD, Lehesjoki AE, de Witte PA, Palotie A, Lerche H, Esguerra CV, De Jonghe P, and Helbig I. 2013. 'De novo loss-of-function mutations in CHD2 cause a fever-sensitive myoclonic epileptic encephalopathy sharing features with Dravet syndrome', Am J Hum Genet, 93: 967-75.

Page 142

Sun H, Kosaras B, Klein PM, and Jensen FE. 2013. 'Mammalian target of rapamycin complex 1 activation negatively regulates Polo-like kinase 2-mediated homeostatic compensation following neonatal seizures', Proc Natl Acad Sci U S A, 110: 5199-204. Tai C, Abe Y, Westenbroek RE, Scheuer T, and Catterall WA. 2014. 'Impaired excitability of somatostatin- and parvalbumin-expressing cortical interneurons in a mouse model of Dravet syndrome', Proc Natl Acad Sci U S A, 111: E3139-48. Tan HO, Reid CA, Single FN, Davies PJ, Chiu C, Murphy S, Clarke AL, Dibbens L, Krestel H, Mulley JC, Jones MV, Seeburg PH, Sakmann B, Berkovic SF, Sprengel R, and Petrou S. 2007. 'Reduced cortical inhibition in a mouse model of familial childhood absence epilepsy', Proceedings of the National Academy of Sciences, 104: 17536-41. Tang B, Dutt K, Papale L, Rusconi R, Shankar A, Hunter J, Tufik S, Yu FH, Catterall WA, Mantegazza M, Goldin AL, and Escayg A. 2009. 'A BAC transgenic mouse model reveals neuron subtype-specific effects of a Generalized Epilepsy with Febrile Seizures Plus (GEFS+) mutation', Neurobiol Dis, 35: 91-102. Tang F, Hartz AMS, and Bauer B. 2017. 'Drug-Resistant Epilepsy: Multiple Hypotheses, Few Answers', Frontiers in neurology, 8: 301-01. Thompson CH, Kahlig KM, and George AL, Jr. 2011. 'SCN1A splice variants exhibit divergent sensitivity to commonly used antiepileptic drugs', Epilepsia, 52: 1000- 9. Tsai M-S, Lee M-L, Chang C-Y, Fan H-H, Yu IS, Chen Y-T, You J-Y, Chen C-Y, Chang F-C, Hsiao JH, Khorkova O, Liou H-H, Yanagawa Y, Lee L-J, and Lin S-W. 2015. 'Functional and structural deficits of the dentate gyrus network coincide with emerging spontaneous seizures in an Scn1a mutant Dravet Syndrome model during development', Neurobiology of Disease, 77: 35-48. Valdivia P, Martin M, LeFew WR, Ross J, Houck KA, and Shafer TJ. 2014. 'Multi-well microelectrode array recordings detect neuroactivity of ToxCast compounds', NeuroToxicology, 44: 204-17. Van Gassen KLI, Hessel EVS, Ramakers GMJ, Notenboom RGE, Wolterink-Donselaar IG, Brakkee JH, Godschalk TC, Qiao X, Spruijt BM, Van Nieuwenhuizen O, and De Graan PNE. 2008. 'Characterization of febrile seizures and febrile seizure susceptibility in mouse inbred strains', Genes, Brain and Behavior, 7: 578-86. van Pelt J, Vajda I, Wolters PS, Corner MA, and Ramakers GJA. 2005. 'Dynamics and plasticity in developing neuronal networks in vitro.' in, Progress in Brain Research (Elsevier). Vassallo A, Chiappalone M, Lopes RDC, Scelfo B, Novellino A, Defranchi E, Palosaari T, Weisschu T, Ramirez T, and Martinoia S. 2017. 'A multi-laboratory evaluation of microelectrode array-based measurements of neural network activity for acute neurotoxicity testing', NeuroToxicology, 60: 280-92. Wagenaar DA, Pine J, and Potter SM. 2006. 'An extremely rich repertoire of bursting patterns during the development of cortical cultures', BMC Neuroscience, 7: 11. Wainger BJ, Kiskinis E, Mellin C, Wiskow O, Han Steve SW, Sandoe J, Perez Numa P, Williams Luis A, Lee S, Boulting G, Berry James D, Brown Robert H, Cudkowicz Merit E, Bean Bruce P, Eggan K, and Woolf Clifford J. 2014. 'Intrinsic Membrane

Page 143

Hyperexcitability of Amyotrophic Lateral Sclerosis Patient-Derived Motor Neurons', Cell Reports, 7: 1-11. Wallace K, Strickland JD, Valdivia P, Mundy WR, and Shafer TJ. 2015. 'A multiplexed assay for determination of neurotoxicant effects on spontaneous network activity and viability from microelectrode arrays', NeuroToxicology, 49: 79-85. Wang J, Lin Z-J, Liu L, Xu H-Q, Shi Y-W, Yi Y-H, He N, and Liao W-P. 2017. 'Epilepsy- associated genes', Seizure, 44: 11-20. Wirrell EC. 2016. 'Treatment of Dravet Syndrome', Can J Neurol Sci, 43 Suppl 3: S13-8. Yang Y, Mis MA, Estacion M, Dib-Hajj SD, and Waxman SG. 2018. 'NaV1.7 as a Pharmacogenomic Target for Pain: Moving Toward Precision Medicine', Trends in Pharmacological Sciences, 39: 258-75. Yi Y, Park J, Lim J, Lee CJ, and Lee S-H. 2015. 'Central Nervous System and its Disease Models on a Chip', Trends in Biotechnology, 33: 762-76. Yu FH, Mantegazza M, Westenbroek RE, Robbins CA, Kalume F, Burton KA, Spain WJ, McKnight GS, Scheuer T, and Catterall WA. 2006. 'Reduced sodium current in GABAergic interneurons in a mouse model of severe myoclonic epilepsy in infancy', Nat Neurosci, 9: 1142-9. Zeng H, Shen EH, Hohmann JG, Oh SW, Bernard A, Royall JJ, Glattfelder KJ, Sunkin SM, Morris JA, and Guillozet-Bongaarts AL. 2012. 'Large-scale cellular-resolution gene profiling in human neocortex reveals species-specific molecular signatures', Cell, 149: 483-96.

Page 144

SUPPLEMENTARY MATERIAL

Page 145

Supplementary Figure (SF) 1. Raster plots representing 30sec of activity from

SCN1A+/-C57BL/6 networks before and after application of 100uM lamotrigine. Six raster plots showing a representative 30sec of activity from: WT network a) before and b) after the application of DMSO control, WT network c) before and d) after the application of 100uM lamotrigine, SCN1A+/- networks e) before and f) after the application of 100uM lamotrigine. Seconds are represented on the x axis and channels number for each recording is represented on the y axis.

Page 146

Page 147

Supplementary Figure (SF) 2. Raster plots representing 30sec of activity from

SCN1A+/-C57BL/6 networks before and after application of 40uM stiripentol. Six raster plots showing a representative 30sec of activity from: WT network a) before and b) after the application of DMSO control, WT network c) before and d) after the application of 40uM stiripentol, SCN1A+/- networks e) before and f) after the application of 40uM stiripentol. Seconds are represented on the x axis and channels number for each recording is represented on the y axis.

Page 148

Page 149

Supplementary Figure (SF) 3. Raster plots representing 30sec of activity from

SCN1A+/-C57BL/6 networks before and after application of 10uM cannabidiol. Six raster plots showing a representative 30sec of activity from: WT network a) before and b) after the application of DMSO control, WT network c) before and d) after the application of 10uM cannabidiol, SCN1A+/- networks e) before and f) after the application of 10uM cannabidiol. Seconds are represented on the x axis and channels number for each recording is represented on the y axis.

Page 150

Minerva Access is the Institutional Repository of The University of Melbourne

Author/s: Morrisroe, Emma

Title: Multielectrode Arrays: A tool for modelling genetic epilepsy

Date: 2019

Persistent Link: http://hdl.handle.net/11343/234064

File Description: Final thesis file

Terms and Conditions: Terms and Conditions: Copyright in works deposited in Minerva Access is retained by the copyright owner. The work may not be altered without permission from the copyright owner. Readers may only download, print and save electronic copies of whole works for their own personal non-commercial use. Any use that exceeds these limits requires permission from the copyright owner. Attribution is essential when quoting or paraphrasing from these works.