Three-level data structures are common in modern epidemiological research. The Childhood to Adolescence Transition Study (CATS), where repeated measures from individuals are clustered within schools, is one such example. While almost all research suffer from missing data, it can be particularly problematic in longitudinal cohorts. Although MI is a popular approach for handling missing data, its validity depends on appropriate tailoring of the imputation model to the substantive analysis, so as to ensure congeniality between the imputation and analysis models. In the context of a three-level substantive analysis with incomplete data, this means the imputation model should account for the multilevel structure, in addition to any other features in the substantive analysis such as non-linear relationships and/or interaction effects. A number of MI approaches including (i) pragmatic adaptations of the widely accessible single-level MI approaches using dummy indicators (DIs) for the higher-level clusters and imputing repeated measures in wide format, (ii) pragmatic adaptations of two-level MI approaches with DIs for the higher-level clusters or imputing repeated measures in wide format, and (iii) MI approaches using three-level imputation models, can be used to handle incomplete three-level data. However, there are limited evaluations of these approaches in the literature. In particular, with complex substantive analyses involving interactions, non-linear terms, and time-varying higher-level cluster memberships conducting MI appropriately ensuring congeniality remains challenging. The current research investigated the performance of MI approaches for handling incomplete three-level data, to provide guidance in the contexts of: a three-level random intercept model, a three-level random intercept model involving interactions or non-linear terms, and a static-effects cross-classified random effects model (CCREM) accounting for time-varying higher-level cluster memberships. In each context, the MI approaches were implemented using joint modelling (JM) and fully conditional specification (FCS). Substantive model compatible (SMC) MI that automatically tailors the imputation model to the substantive analysis, was also considered where relevant. The approaches were evaluated using simulation studies based on CATS, conducted under a number of scenarios by varying factors such as the number of clusters, cluster sizes, missing data proportions, missing data mechanisms and the intra-cluster correlations (ICCs). The methods were also illustrated using CATS data. Results indicated the pragmatic adaptations of the standard single-level MI approaches are adequate for handling incomplete three-level data in most substantive analyses, namely when the substantive analysis is a random intercept only model, a random intercept model with an interaction between the time-varying exposure and time, or a static-effects CCREM. However, in each substantive setting if the data are sparse due to large amounts of missing data and/or small cluster sizes, or the ICCs are small, the two-level MI approaches imputing repeated measures in wide format or three-level MI approaches should be used, with the latter being the only option if the repeated measures are also irregularly measured. If the analysis model contains interactions between the time-varying exposure and baseline variables or non-linear terms, the three-level SMC-MI approach (available in software Blimp) is recommended. The findings of this PhD provide initial guidance on how best to impute incomplete three-level data in three different substantive contexts. Further evaluation of these approaches across a range of contexts and studies are needed to develop guidance in other settings.

Birth defects are the leading cause of infant death and the most common are congenital heart defects (CHD). In the majority of cases, the underlying genetic cause is unknown. There is increasing evidence that variants in non-coding DNA within the functional cis-regulatory elements (CREs) controlling gene expression can cause CHDs. However, little is known about the human CREs that are crucial for cardiovascular development. To address this knowledge gap, we bioinformatically identified cardiovascular CREs based on the occupancy of the CRE by the homeodomain protein NKX2-5 and cardiac chromatin histone modifications. This search defined several putative CREs within the loci of genes specifically implicated in heart development and disease. We show that human regulatory elements are functionally conserved and capable of driving reporter transgene expression in vivo throughout the developing cardiovascular system of medaka (Oryzias latipes). Using heart muscle cells and endothelial cells derived from human embryonic stem cells in which these CREs were deleted, we identified molecular perturbations in the gene regulatory network controlling vascular development. Among the CREs identified, we show that the deletion of the human enFLT1 enhancer in vitro, a highly conserved enhancer within the FLT1 locus, using endothelial tube formation assays did not disrupt vessel formation. However, deletion of the endogenous enFLT1 enhancer in medaka results in blood vessel malformations, indicating an important role for this regulatory element in vivo. These findings highlight the crucial role of human regulatory elements and their function as a regulator and buffer of transcriptional regulation in cardiovascular development. Our studies further support the hypothesis that perturbations in human CREs are important contributors of developmental defects.

Background Children with traumatic brain injury (TBI) often present with deficits in working memory and decision-making. As these are considered core executive function skills, children with impairments in working memory and decision-making are more likely to exhibit secondary deficits in academic performance, adaptive behaviour, social skills, and overall quality of life. However, there is a lack of research assessing decision-making skills, or the rehabilitation of working memory and decision-making in this clinical population. The current study aimed to develop a novel measure of decision-making, using a randomised controlled trial to implement and evaluate the efficacy of a computerised working memory intervention (Cogmed), and investigate the contribution of decision-making as a mediator in childhood TBI and dysregulation profile. Method The study was conducted in Victoria, Australia. Participants in the decision-making tool development, and the mediation study comprised two groups, the TBI group (n=49), and the typically developing healthy control group (n=22). Children were assessed directly on standardised measures of executive function such as working memory, and parent reports were obtained via questionnaires of adaptive behaviour and dysregulation. Participants in the clinical trial were children with TBI (n=31) who presented with difficulties in executive function and/or working memory on screening. Child assessments and parent reports were obtained at three time points: baseline, immediately post intervention, and at 6 months follow-up. Results and Conclusion This study was the first study to develop and validate a novel measure of decision-making in children following TBI, thereby providing preliminary evidence of the clinical utility of this measure. This study was the second study to investigate Cogmed in this clinical population, with the results of the clinical trial suggesting minimal evidence for Cogmed’s efficacy in improving working memory, decision-making, and functional outcomes. Finally, this study was the first to report on the role of decision-making as a mediator between childhood TBI and dysregulation, with results suggesting decision-making underpins problems of dysregulation in children following TBI.

Streptococcus pyogenes (the group A Streptococcus, GAS) causes an immense communicable and noncommunicable global burden of disease across a diverse clinical spectrum. The potential for immunisation to reduce this burden was recognised centuries ago, however vaccine development has been frustrated by scientific, regulatory, and commercial obstacles. A renewed global effort has prioritised advancement of new human infection models to accelerate vaccine development. Controlled human infection (‘human challenge’) trials, where healthy volunteers are deliberately infected with a pathogen under tightly controlled conditions, are a powerful tool for studying infections and advancing development of vaccines and drugs. The overall aims of this thesis were to detail the context, rationale, and steps taken to successfully establish a new GAS pharyngitis controlled human infection model. Each project is a milestone on that pathway, namely: understanding the impeded status of vaccine development as the starting point for new initiatives, including the controlled human infection model; judicious selection of a challenge strain; designing the protocol for an initial challenge study; undertaking the landmark trial to establish the new model; and exploiting this unique opportunity to study host-pathogen interactions in the organism’s only natural (human) host. This thesis spans a remarkable period in the modern history of GAS vaccine development. Involvement of the World Health Organization (WHO) has led to a realistic vaccine development roadmap aiming to overcome the obstacles that have prevented an explosion of GAS research and a backlog of pre-clinical candidates from translating into meaningful progress. A strong rationale for challenge strain selection is central to maximising the impact of controlled human infection research and mitigating risk to participants. A multifaceted approach was undertaken that incorporated molecular epidemiology, in vitro assays, whole-genome sequencing, and animal model studies, to ultimately select an emm75 GAS strain for use in the new model. Human experiments in the 19th and early 20th century established GAS as the cause of scarlet fever and 1970s human infection studies demonstrated efficacy of prototype monovalent M-protein vaccines. Drawing on these contributions as proof of concept, a protocol for a new GAS pharyngitis controlled human infection model was developed in accordance with modern Good Clinical Practice and Good Manufacturing Practice standards. To establish the pharyngitis controlled human infection model, a trial was done to identify a dose of emm75 GAS that causes acute pharyngitis in at least 60% of healthy adult volunteers. The starting dose level was confirmed with pharyngitis diagnosed in 17 of 20 healthy adult participants (85%; 95% CI: 62-97). When the dose was de-escalated, only one of five participants was diagnosed with pharyngitis. There were no serious adverse events, and a convincing clinical picture of acute symptomatic pharyngitis was supported by microbiological, biochemical, and immunological results. To explore antibody responses during experimental human pharyngitis, serum immunoglobulin G (IgG) and salivary immunoglobulin A (IgA) responses to an array of 17 candidate vaccine antigens were studied at multiple timepoints. Responses were highly variable in serum and saliva across the panel of antigens, and no obvious natural correlate of protection was evident. Overall, serum IgG responses targeting multiple antigens increased for participants with pharyngitis across the post-challenge timepoints and did not increase in those without pharyngitis. Although the low number of participants without pharyngitis precluded definitive conclusions, an opposite pattern was observed for salivary IgA, with increased responses in most who did not develop pharyngitis, and flat or decreased responses in those who did have pharyngitis. The model presented in this thesis responds directly to a key priority outlined in the WHO GAS vaccine development roadmap. It sets the stage for future human challenge trials to evaluate vaccines, accelerate drug development, and study host-pathogen interactions.

Epidermolysis bullosa (EB) is a group of genetic blistering disorders caused by mutations in the structural proteins of the skin. EB is characterised anatomically by skin fragility, blistering and scarring, and frequently results in pain, itch and – in severe cases – death. The dystrophic subtype of EB is caused by mutations in the human COL7A1 gene which encodes for collagen VII, and can be inherited in either a dominant or recessive manner. Dominant dystrophic epidermolysis bullosa (DDEB) is commonly caused by glycine substitutions within the collagenous domain of collagen VII. Improving understanding of the pathophysiology of DDEB should enable the development of more effective treatments. To this end, having suitable mouse models for DDEB would be helpful, but none have yet been reported. This thesis describes the generation of two novel mouse lines, each of which carries a modified Col7a1 allele (p.G2037R and p.G2028R) that was introduced using CRISPR and corresponds to the most common COL7A1 mutations found in DDEB patients (p.G2043R and p.G2034R). Dominant inheritance of either of these alleles results in a phenotype that closely resembles that seen in DDEB patients. Specifically, mice carrying these alleles show recurrent blistering that is first observed transiently around the mouth and paws in the early neonatal period and then again around the digits from 5-10 weeks of age. Histologically, the mice show micro-blistering and reduced collagen VII immunostaining. Biochemically, collagen VII from these mice displays reduced thermal stability, which we also observed to be the case for DDEB patients carrying the analogous mutations. Unlike previous rodent models of EB which frequently show severe disease and early lethality, these mouse models – which to our knowledge are the first for DDEB – show no impairment in survival and have a relatively mild phenotype, and thus represent a practically and ethically tractable means for better understanding and treating EB. As proof-of-concept of the utility of these DDEB mice, animals carrying the p.G2037R allele were assessed in standardised pain behavioural tests and observed to be hypersensitive to both mechanical and thermal stimuli using the von Frey and Hargreaves’ assays respectively, mimicking what has been observed previously in EB patients. A range of different pharmacological agents – some of which are already used clinically to treat pain in EB patients and others of which are more experimental – were then objectively compared across a range of relevant doses and their analgesic effectiveness in DDEB mice was found to vary. While most drugs provided only a partial reduction in pain hypersensitivity, the non-steroidal anti-inflammatory agent, meloxicam, as well as a novel analgesic agent in early pre-clinical development were both found restore sensitivity to that of wild type mice, thus demonstrating that the DDEB mice may be useful preclinical tools for future drug development in EB.

Behavioural emergencies due to high-risk behaviours are increasing globally in paediatric health care settings and place the child, their family, other inpatients and staff at high-risk of harm. High-risk behaviours have been reported in up to two thirds of children with autism spectrum disorder (ASD) and intellectual disability (ID). Admission to hospital can trigger high-risk behaviours in children, especially those with ASD or ID, and increase their intensity or frequency. Patient and staff safety can be compromised if staff are not confident or competent in managing aggressive outbursts. As well as the safety risks, these behaviours result in delayed treatment, prolonged procedure times, increased health care costs and poorer health outcomes. Having a critical mass of well-trained staff who are confident in preventing and managing behavioural emergencies is important. However, these skills may be beyond the usual scope of practice of the acute paediatric health care provider. Paediatric acute health care is a complex specialty encompassing children with a wide range of developmental levels who operate within an array of unique family-centred care systems. Staff need many and varied skills and strategies that can be adapted to effectively interact with the spectrum of ages, developmental stages, neurodiversity and parental/carer involvement experienced in the acute paediatric setting. Simulation-based education (SBE) has been found to be superior to traditional clinical medical education for clinical skill acquisition and is gaining acceptance as a training method for teaching skills in managing clinical aggression. Simulation training allows practice of skills without patient risk and may be more effective than traditional teaching formats for aggression management. The aim of this thesis is to better understand the cause of behavioural emergencies caused by children in hospital and to use this knowledge to develop, implement and evaluate a simulation-based training program for acute care paediatric staff on managing aggression in children with ASD or ID. The program of research was established with four objectives linked to four independent studies, each informing the next. First, we conducted a systematic review to assess the effectiveness of simulation-based training for increasing de-escalation knowledge, skills and behaviour of staff working in the acute care setting. Second, we conducted a one-year retrospective study of behavioural emergencies that triggered an emergency response team attendance in a quaternary paediatric hospital to understand their frequency, type, context and management. Third, we designed a simulation-based education session about prevention and management of aggression in a neurotypical adolescent. We used a proof-of-concept study design to evaluate the acceptability of this training and to gain an understanding of the impact on participants’ self-perceived confidence levels in managing clinical aggression. Fourth, we designed a more specific simulation training program about management of aggression and high-risk behaviours in adolescents with ASD and ID in the hospital setting. Our final study was a pilot and feasibility cluster randomised controlled trial (RCT) of this simulation-based education, with mixed method design. Understanding aggression and high-risk behaviours in paediatric health care settings is an emerging area of research. The findings of these four studies have increased our understanding of the prevention and management of aggression and high-risk behaviours in acute care paediatric settings. Simulation-based education could become an important part of a multi-faceted strategy that is now needed to reduce the prevalence of high-risk behaviours in acute care settings and paediatric hospitals, and to improve outcomes if behavioural emergencies occur. Further evaluation of this training format for managing high-risk behaviours (including aggression), in children with ASD and ID is required. As well as simulation education, features of provider and patient systems need to be considered when formulating solutions to this complex issue in the paediatric acute care setting.

Cystic fibrosis (CF) is a multisystem disease caused by mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene. Lung disease which is the major source of morbidity and mortality in CF, begins in infancy and persists and progresses in the first years of life. The most sensitive measure of CF lung disease is computed tomography (CT) and the majority of children show evidence of bronchiectasis, which represents irreversible structural lung disease, by five years of age. The rate with which lung disease develops and progresses varies greatly between people with CF. Accurate prediction of future lung disease severity in people with CF would facilitate significant improvements in clinical care, as well as provide helpful information to people with CF and their families, health authorities and CF researchers. It has been demonstrated that CF lung disease severity is determined by a combination of non-CFTR genetic and environmental factors. However, no robust predictive biomarkers have been developed. Epigenetic mechanisms are those which regulate gene expression but do not alter the DNA sequence itself. DNA methylation is the most widely studied epigenetic mechanism, and an individual’s DNA methylation profile is determined by their underlying genotype and environmental exposures (including in utero). DNA methylation has been studied as a predictive biomarker in other childhood diseases, and given it is determined by the same factors which determine CF lung disease severity, is well positioned to act as a biomarker in CF. Despite this, DNA methylation based biomarkers of future lung disease severity have not been explored in CF. The thesis presents a series of projects which aimed to identify DNA methylation based biomarkers of future CF lung disease severity. The projects in this thesis utilised biospecimens and clinical data from the world leading Australian Respiratory Early Surveillance Team for Cystic Fibrosis (AREST CF) cohort. In order to better understand previous attempts at identifying predictive biomarkers, Chapter Two and Three describe reviews of the evidence regarding non-CFTR genetic modifiers of CF lung disease and DNA methylation based biomarkers of future health outcomes in children. The reviews highlight the importance of appropriate study design when trying to identify predictive biomarkers, which then informed design of the subsequent epigenome wide association study (EWAS). Chapters Four and Five describe the development of methods that were necessary for subsequent EWAS. In Chapter Four protocols for cryopreservation, and flow cytometry based phenotyping and sorting of specific cell populations were developed and validated for use with paediatric bronchoalveolar lavage (BAL). These methods were then used in Chapter Five to obtain purified populations of the common cell types in BAL, which were then used to develop reference epigenomes for these cell types. These reference epigenomes allowed adjustment for cell composition in the subsequent EWAS, which is a crucial step in DNA methylation studies involving samples with multiple cell types. Chapters Six and Seven describe the EWAS studies that attempted to identify predictive biomarkers. Chapter Six involved genome wide DNA methylation profiling of BAL collected at six years of age, and assessed the ability to predict the presence or absence of bronchiectasis at nine years of age. No predictive biomarkers were identified. Chapter Seven involved genome wide DNA methylation profiling of BAL collected at one year of age, and assessed the ability to predict the presence or absence of bronchiectasis at five years of age. Seven predictive biomarkers were identified, which when assessed using area under the receiver operator curve analysis had extremely good performance as predictive biomarkers. Several of the predictive biomarkers were related to genes that are relevant to CF lung disease pathophysiology and hence may represent therapeutic targets. By following best practice for EWAS, and in particular using tissue specific samples, adjusting for cell composition and assessing outcome using the gold standard measure, this thesis succeeded in identifying DNA methylation based biomarkers of future lung disease severity in CF. If these biomarkers are validated in external cohorts they could dramatically improve the care of children with CF around the world. N.B. There are two files that have been submitted separate to this thesis document as supplementary files. They have been highlighted as appropriate in the thesis. In addition, the code used for statistical analysis is available at: https://atlassian.petermac.org.au/bitbucket/pages/OS/paed-cf- methylation/master/browse/docs/index.html

Abstract Introduction Kidney transplantation is the treatment of choice for eligible individuals with end-stage kidney disease, conferring superior quality and quantity of life compared to remaining on dialysis. However, transplant outcome for many patients remains suboptimal, with graft life years falling substantially short of the recipient’s life expectancy due to graft rejection and failure. At the same time, typical immunosuppressant doses to control rejection lead to cumulative morbidity and premature mortality from side-effects such as infections, malignancy, and cardiovascular disease. Objectives The overall objective of the research in this dissertation was to describe the clinical practice of immunosuppression in kidney transplantation and explore an approach to improve dosing to optimise safety and effectiveness. The thesis used time-to-event and pharmacoepidemiologic analysis of a large volume registry data set; systematic review and critical re-appraisal of pharmacokinetic-pharmacodynamic (PKPD) and randomised controlled trials (RCT) of immunosuppressants; and an investigator initiated, Australian multi-centre pharmacokinetic study with population pharmacokinetic modelling. 1. Optimal dose and dosing strategy of immunosuppressant agents over the long term have not been conclusively established. Using over 5 decades of immunosuppressant dosing and kidney transplant outcome data from the Australia and New Zealand Dialysis and Transplant Registry (ANZDATA) a time-to-event analysis and pharmacoepidemiologic data exploration was performed. This was used to describe immunosuppressant regimen and dose size evolution over time, and association with outcome (Chapter 3). 2. Mycophenolate mofetil (MMF) is used in the majority of kidney transplant recipients, yet there is conflicting opinion on whether dose should be individualised, along with substantial practice variability and ongoing prevalence of inefficacy and immunosuppressant toxicities. Through a comprehensive systematic review of both PKPD and RCT evidence the case for concentration-controlled dosing (CCD) of MMF was critically re-appraised, to refine understanding and guide best practice (Chapter 4). 3. Initial underexposure to mycophenolic acid (MPA, active metabolite of MMF) is seen in up to a third of kidney transplant recipients thus increasing risk of acute rejection. An investigator-initiated, Australian multi-centre pharmacokinetic (PK) study (ADOPT) was undertaken to describe the change in PK of total and unbound MPA in the pre-transplant and early post-transplant period. This was used to assess the ability of pre-transplant drug concentrations to predict post-transplant individual MMF dose requirement (Chapters 5 and 6), and to understand the change in total and unbound MPA drug concentrations in the early post-transplant weeks (Chapter 5). 4. Optimising MPA exposure is a practical burden in the initial phase post-transplant. Population PK (PopPK) modelling of the ADOPT data was performed to provide a deeper understanding of changing MPA PK peri-transplant and in the initial post-transplant weeks, in order to (a) re-assess ability to predict post-transplant MMF dose requirement after controlling for changing covariate factors and (b) understand the pattern of change in unbound MPA PK over the initial post-transplant weeks and (c) evaluate the potential value of Bayesian dosing to achieve an unbound MPA concentration target (Chapter 6). Key findings - There have been 4 key immunosuppressant regimen eras, with successive improvement in patient and graft outcomes by era. Multivariate analysis revealed more modest improvements in the maintenance phase in the period 2007 - 2015 compared with 1998 - 2006. Dose size for prednisolone and tacrolimus has decreased over the entirety of these 2 periods (1998 – 2015), in both initial and maintenance phases, with possible over-reduction of tacrolimus in the maintenance phase. The initial dose of MMF (combined with tacrolimus) has increased then decreased over this period. - The pharmacokinetic-pharmacodynamic (PKPD) and RCT evidence clearly favours CCD of MMF using Target Concentration Intervention. The contradictory RCT data can be objectively explained by accounting for the use of Therapeutic Drug Monitoring compared with Target Concentration Intervention CCD. - Pre-transplant concentrations were unable to improve prediction of post-transplant individual MMF dose requirement using the trapezoidal exposure calculation technique. Using a PopPK model and accounting for changing renal function, however, pre-transplant concentrations provided some information on post-transplant dose requirement. - There is a substantial increase in clearance of unbound mycophenolic acid (MPA, active metabolite of MMF) from pre- to post-transplant. Unbound MPA exposure stabilises by 2 weeks post-transplant, making it a more practical measure for CCD. A PopPK model has been developed that could be used for Bayesian dosing to an unbound MPA target.

Infection and inflammation are critical in the pathogenesis of cystic fibrosis (CF) lung disease. Detection of pathogenic bacteria and neutrophilic inflammation in the lower airways of infants with CF is associated with the earlier development of bronchiectasis and lower lung function at six years of age. The role of the airway microbiota, its bacterial community network, in the evolution of CF lung disease is not known, nor are the long-term sequelae of early life lower airway infection and inflammation. This thesis addresses these questions through analysis of a birth cohort of 100 infants, newly diagnosed with CF, predominantly by newborn screening, in Victoria, Australia, in the early 1990s, and managed in a single specialist CF centre where anti-staphylococcal prophylaxis was not prescribed. Participants in this birth cohort underwent serial bronchoscopy and bronchoalveolar lavage (BAL) from recruitment in infancy and semi-annually thereafter during early childhood. Quantitative microbiological culture was performed and inflammatory markers measured contemporaneously. 16S ribosomal RNA gene sequencing was subsequently performed on stored samples to describe the lower airway microbiota. Comparison of the lower airway microbiota of 13 newly diagnosed infants with CF to that of disease-controls, infants undergoing clinically indicated bronchoscopy but without evidence of respiratory infection, revealed profound differences in the CF airway microbiota within the first six months of life, driven by increased Staphylococcus and more pronounced in unwell infants. Longitudinal analysis of 95 BAL samples from 48 infants with CF, aged 6 weeks to 6.5 years, demonstrated a diverse and dynamic lower airway microbiota, in which reduced diversity was associated with dominance of proinflammatory pathogens and increased airway inflammation. Twenty-five year longitudinal follow-up of 79 of the 100 original birth cohort members (45.6% male, 67.1% P.Phe508del homozygous) was conducted to explore the associations between early lower airway infection with Staphylococcus aureus and Pseudomonas aeruginosa, as well as neutrophilic inflammation, and survival, rate of decline of pulmonary function and severity of structural lung disease in adulthood. Survival analyses were conducted using Kaplan Meier curves and Cox proportional hazard models. Annual rate of decline of forced expiratory volume in one second, percent predicted, calculated according to the Global Lung Initiative reference equations, was estimated using a linear mixed-effects model with random intercept and slope, and severity of structural lung disease was quantified using the CF-CT scoring system. There was no association between lower airway infection with S. aureus in the first two years of life and 25-year survival or development of functional or structural lung disease. Infection with the AUST-01 P. aeruginosa strain in the first five years of life was associated with increased mortality, while infection with other P. aeruginosa strains was associated with a trend towards reduced survival and poorer longitudinal lung function. Neutrophilic inflammation in the first two years was not associated with poorer long-term outcomes. Together, these findings highlight the need for careful evaluation of current approaches to anti-staphylococcal prophylaxis in newly diagnosed infants with CF, while suggesting a role for optimisation of the early lower airway microbiota.

Focal Cortical Dysplasia (FCD) is one of the most common brain malformations and a frequent cause of intractable epilepsy. Affected individuals often require surgical removal of the affected cortex for seizure control. The European Epilepsy Brain Bank estimates that different subtypes of FCD are the reason for up to 70.6% of epilepsy surgery related to brain malformations. Given the clinical importance of this disorder, there is an urgent need to further understand its biology and disease mechanism. Although FCD was first described 50 years ago, the genetic and pathogenic mechanisms of the disorder remain incompletely understood. We hypothesise that the delineation of disease mechanism underlying FCD will not only improve diagnosis and clinical management, but will also provide novel insights into the mechanisms underpinning human brain development and function. This project applied modern genomic technologies to comprehensively study a unique archive of surgical brain tissues derived from individuals affected by FCD. The first part of this study focussed on understanding the genetic basis of a subtype of FCD, known as bottom-of-sulcus dysplasia (BOSD). BOSD is a highly localised subtype of FCD with the dysplastic features being maximal at the bottom of sulcus. Using deep sequencing, we analysed blood and brain derived gDNA from 20 BOSD brain specimens. We found that brain-specific, pathogenic somatic MTOR variants accounted for 50% of the cases, with an additional two cases caused by heterozygous germline variant in DEPDC5 or NPRL3. These genes are all involved in the mTOR signalling pathway. Further characterisation of four somatic MTOR cases found that the somatic variants were more abundant in the bottom of sulcus compared to the nearby cortex, with a positive correlation with the severity of histopathology. Our study shows that BOSD belongs to an expanding group of “mTORopathies” and highlights that BOSD lesion is highly localised, with well-defined genetic and pathologic substrates. In the second part of this study, we tested the germline + somatic “two-hit” hypothesis in FCD. The “two-hit” hypothesis suggests that individuals with heterozygous germline variant acquired FCD lesion as a result of a somatic second “hit”. In a case study of an individual with FCD and a heterozygous germline DEPDC5 variant, we identified a somatic second “hit” in the same gene. By combining experimental and clinical findings, we showed in this individual that the brain region with the highest somatic variant allele frequency was also the region of seizure onset, suggesting a relationship between the somatic variant and epileptogenicity. We further characterised the brain specimens and found that somatic variant allele frequency positively correlated to the density of dysmorphic neurons. Furthermore, laser capture microdissection showed that the somatic variant was only detected in the dysmorphic neurons and not in the normal neurons. Collectively, we provided supporting evidence for the “two-hit” mechanism and demonstrated the molecular and cellular characteristics of FCD lesion. The third part of this study involved the application of single nucleus RNA-sequencing (snRNA-seq) to study 26 surgical brain specimens resected due to a range of brain malformations including FCD. These brain malformations are closely related and commonly characterised by the presence of abnormal cell types i.e. dysmorphic neurons and balloon cells. We generated 284,891 single nuclei transcriptomes, the largest snRNA-seq dataset for epilepsy brain specimens to date. Using a combination of bioinformatic analysis and molecular characterisation, we identified the probable transcriptional profiles of dysmorphic neurons and balloon cells. Furthermore, we identified commonly perturbed pathways in both cell types, namely calcium signalling pathway and glutamatergic synapse regulation, which may help explain the pathogenic mechanism underlying these different types of closely related brain malformations. In conclusion, this study has provided fundamental insights into the genetic and pathogenic mechanism underlying FCD and related brain malformations. In addition to improving genetic diagnosis and clinical management, these findings have the potential to pave the way for the identification of novel druggable targets by defining the genetic and transcriptomic profiles of FCD.

This thesis represents a compilation of peer-reviewed papers addressing the care for children presenting to emergency departments with bronchiolitis. After an introductory chapter, papers related to one of several themes are briefly introduced and placed in context with each other. The papers, and their compilation in this thesis, aim to provide guidance for paediatricians and emergency physicians caring for children on how to bronchiolitis in hospital, with specific attention paid to hydration methods, reducing low value care and high flow nasal cannula use. Many aspects of this work, including certain research findings, are being used by, or have been implemented in, the Emergency Department at Royal Children’s Hospital and other emergency departments across Victoria, Australia, as well as interstate and overseas. I take primary responsibility for the initial concept, design, protocol development and successful execution of all projects used for this thesis. To my knowledge, no other co-author has presented papers from this compilation for submission as part of a degree. Exceptions are the use of parts of the projects in the fulfilment of Advanced Trainee requirements for the Royal Australasian College of Physicians or the Australasian College for Emergency Medicine, or for advanced medical students in the fulfilment of a Bachelor of Science through the University of Melbourne.

Introduction: Reference intervals (RI) are an important clinical assessment tool used by physicians and laboratory professionals to interpret test results. A RI is commonly defined as a range between the 2.5th and 97.5th percentile and represents the middle 95% of the reference population. The estimation of RIs is a complex process which involves defining a reference population, selecting reference individuals, collecting, and testing reference samples, and applying appropriate statistical methods. Estimation of RIs in children is challenging for a number of reasons. There is a need for adequate representation of the general population in the reference population. Analytes are influenced by the nutrition, development, and growth of a child. Hence, RIs should reflect these age-dependent changes in the analytes as children grow older. Most laboratory tests are instrument dependent and variation between different analysers has been documented in several studies. In addition, the best statistical methods for estimating and reporting RIs varying with continuous age are still not widely agreed upon. Finally, there are no available guidelines for laboratories to validate continuous age-specific RIs for their use. Hence, this project addressed the following four gaps: 1) the lack of appropriate and reliable continuous age-specific paediatric RIs, 2) an absence of head-to-head comparison of laboratory results on different analysers, 3) the lack of clear evidence and accompanying guidelines regarding the best statistical methods for constructing continuous age-specific paediatric RIs, and 4) the lack of guidelines for validating continuous age-specific RIs. Methods: This research project consists of four studies. Data collected as part of the HAPPI Kids study, from children 30 days to < 18 years, were used to compare laboratory test results on different analysers and to establish and validate continuous age-specific RIs. Differences in mean test results of analytes by analyser types were investigated using mixed-effects regression analysis and by comparing maximum variation between analysers with analyte-specific allowable total error reported in the Westgard QC database. Continuous age-specific paediatric RIs were estimated using quantile regression where power variables in age were selected based on fractional polynomial regression for the mean, with modification by sex when appropriate. The continuous age-specific paediatric RIs were considered valid if more than 90% of test results from a routine laboratory were within the intervals. A systematic review was conducted to examine the range of statistical methods used over the past 25 years for the estimation of age-specific RIs and to identify trends in usage and reporting. A simulation study was conducted to evaluate and compare statistical methods for constructing continuous age-specific RIs in children under different scenarios and for different sample sizes. Results: The variation in the mean test results across five analysers was not clinically significant for 24 out of 30 biochemistry analytes examined and common continuous age-specific RIs were established for children 30 days to 18 years. The continuous age-specific paediatric RIs were validated in a routine laboratory after initial analysis for most analytes and after secondary analysis for a few analytes. According to the results of the systematic literature review, a wide variety of statistical methods have been used for estimating age-specific RIs in children over the past 25 years. However, there has been insufficient uptake of modern statistical methods in estimating continuous age-specific paediatric RIs for analytes that change with age. The results of the simulation study show that in comparison to the commonly used methods for estimating continuous age-specific paediatric RIs, the applied novel method that we applied performed better in most scenarios for sample sizes of 400 and over. Discussion: The comparison of test results for common biochemistry analytes should provide evidence of lack of differences between analysers and in turn improve clinical interpretation. It is recommended that continuous age-specific RIs are used as a gold standard for interpreting test results. Laboratories should estimate or validate continuous age-specific paediatric RIs for their use by applying appropriate statistical methods based on the evidence provided in this research.