UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Epigenetic signatures of prenatal alcohol exposure Lussier, Alexandre André 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_november_lussier_alexandre.pdf [ 34.19MB ]
Metadata
JSON: 24-1.0355868.json
JSON-LD: 24-1.0355868-ld.json
RDF/XML (Pretty): 24-1.0355868-rdf.xml
RDF/JSON: 24-1.0355868-rdf.json
Turtle: 24-1.0355868-turtle.txt
N-Triples: 24-1.0355868-rdf-ntriples.txt
Original Record: 24-1.0355868-source.json
Full Text
24-1.0355868-fulltext.txt
Citation
24-1.0355868.ris

Full Text

EPIGENETIC SIGNATURES OF PRENATAL ALCOHOL EXPOSURE by  Alexandre André Lussier  B.Sc., McGill University, 2012  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES  (Medical Genetics)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  September 2017  © Alexandre André Lussier, 2017   ii Abstract Prenatal alcohol exposure (PAE) can alter the development, function, and regulation of neurobiological and physiological systems, causing lasting cognitive alterations, behavioral deficits, immune dysfunction, and increased vulnerability to mental health problems. In humans, the spectrum of these deficits is known as fetal alcohol spectrum disorder (FASD). Although the molecular underpinnings are not fully elucidated, epigenetic mechanisms are a prime candidate for the programming of physiological systems by PAE, as they may bridge environmental stimuli and neurodevelopmental outcomes. DNA methylation is also emerging as a potential biomarker of early-life events, which may aid in earlier FASD diagnoses. Thus, my overarching aim was to identify epigenetic mechanisms that may contribute to the deficits associated with FASD and act as biosignatures of PAE. Specifically, I used genome-wide approaches to assess underlying gene expression programs and epigenomic profiles in a rat model of PAE and clinical cohorts of individuals with FASD. In the rat model, I identified alterations to gene expression programs in the brain of adult PAE females under steady-state and immune challenge conditions. Building on these long-term alterations to transcriptomic programs, I identified altered DNA methylation patterns persisting from birth to weaning in the hypothalamus PAE animals, suggestive of early reprogramming of neurobiological systems. In parallel, I found concordant alterations to DNA methylation profiles in the hypothalamus and white blood cells of PAE animals, which may reflect systemic effects and potential biomarkers of PAE. To complement the animal model, I also investigated DNA methylation patterns in two clinical cohorts of FASD, where I identified an epigenetic signature of FASD in buccal epithelial cells. As these results raised the possibility of an epigenetic biomarker, I investigated the relevance of DNA methylation as a diagnostic method for PAE, and successfully generated a predictive algorithm that could classify   iii individuals with FASD versus controls. Overall, these findings provide evidence for the biological embedding of PAE’s effects through changes in gene expression and DNA methylation, while setting the stage for the development of novel biomarkers. Ultimately, these may aid in the development of targeted interventions and early screening tools to mitigate the deficits associated with FASD.      iv Lay Summary Prenatal alcohol exposure can result in abnormal brain development, causing Fetal Alcohol Spectrum Disorder (FASD), which is linked to a number of cognitive, behavioural, and immune deficits that last across the lifetime. Although the lasting effects of alcohol on development are well studied, the molecular changes causing these deficits remain relatively unknown. Recent evidence suggests that modifications to DNA structure and regulation, known as epigenetic mechanisms, may play a role in the long-term effects of alcohol on the developing brain and could act as a signature of prenatal alcohol exposure. This thesis presents new evidence for DNA methylation, a small chemical mark added to DNA, as a mechanism in the long-term programming of immune and brain functions. Furthermore, it provides a framework for the use of DNA methylation as a marker of alcohol exposure to diagnose children at-risk for FASD and help lessen some of their long-term problems.    v Preface Please note that all data chapters in this thesis (Chapters 2-5) are presented in manuscript format, as they are currently published (Chapters 2 & 4) or under submission (Chapters 3 & 5).   Portions of Chapter 1 (introduction) have been adapted from previously published manuscripts:  • Lussier AA, Weinberg J, Kobor MS. 2017. Epigenetics studies of fetal alcohol spectrum disorder: where are we now? Epigenomics.  • Lussier AA*, Islam SA*, Kobor MS. Genetics and epigenetics of development. In Gibbs R. & Kolb B. (Eds.). The neurobiology of brain and behavioural development. Elsevier Inc. In press. *Authors contributed equally.  A version of Chapter 2 has been published in the following manuscripts:  • Lussier AA*, Stepien KA*, Neumann SM, Pavlidis P, Kobor MS, Weinberg J. 2015. Prenatal alcohol exposure alters steady-state and activated gene expression in the adult rat brain. Alcoholism: Clinical and Experimental Research. *Authors contributed equally.  • Lussier AA, Stepien KA, Weinberg J, Kobor MS. 2015. Prenatal alcohol exposure alters gene expression in the rat brain: Experimental design and bioinformatic analysis of microarray data. Data in Brief.  The experimental design and animal work for this study was primarily performed by X. Zhang, with assistance from other members of the Weinberg lab. S. Neumann and K. Stepien ran the gene expression microarrays and performed the tissue extractions alongside T. Bodnar and L. Ellis. K. Stepien performed the differential expression analysis, with critical insight from P. Pavlidis. I verified the microarrays findings by RT-qPCR, and performed the bioinformatic   vi analyses related to verification experiments and pathway identification. I also wrote the vast majority of the manuscript. J. Weinberg and M. Kobor provided critical insight at all steps of the process. John Wiley & Sons license number: 4133251106844 (ACER).  Chapter 3 is original and unpublished. I developed the design for this study alongside T. Bodnar, J. Weinberg, and M. Kobor. I performed the animal experiments with the assistance of T. Bodnar, W. Comeau, and the members of the Weinberg lab. I performed the meDIP procedures with aid from M. Mingay from M. Hirst’s lab at the UBC, and next-generation sequencing was performed by the Genome Sciences Centre in Vancouver, BC. I was responsible for all bioinformatic analyses of the meDIP-seq data, and was assisted by A. Morin from the Kobor lab for the pyrosequencing analysis. I was solely responsible for manuscript preparation, with critical feedback from J. Weinberg and M. Kobor.   A version of Chapter 4 has been published as:  • Portales-Casamar E*, Lussier AA*, Jones MJ, MacIsaac JL, Edgar RD, Mah SM, Barhdadi A, Provost S, Lemieux-Perreault LP, Cynader MS, Chudley AE, Dubé MP, Reynolds JN, Pavlidis P, Kobor MS. 2016. DNA methylation signature of human fetal alcohol spectrum disorder. Epigenetics & Chromatin. *Authors contributed equally.  This study was designed by members of NeuroDevNet, a Canadian Network of Centres for Excellence, and collection of samples was performed at multiple FASD clinics across Canada. I was responsible for a portion of the bioinformatic analyses (gene ontology and differentially methylated regions) and wrote the majority of the manuscript. E. Portales-Casamar performed the differential methylation analysis. R. Edgar performed the brain gene expression analysis. J.   vii MacIsaac and S. Mah ran the DNA methylation arrays.  A. Barhddi, S. Provost, LP Lemieux-Perreault and MP Dubé aided in the genetic analyses. M. Jones, P. Pavlidis, A. Chudley, J. Reynolds, and M. Kobor aided in the interpretation of results and manuscript feedback.   Chapter 5 is original and unpublished. This study was designed by members of NeuroDevNet and sample collection was performed in Winnipeg, MB by J. Salmon and A. Chudley. DNA methylation arrays were run by J. MacIsaac and A. Morin, whom also performed the pyrosequencing assay. I was responsible for all bioinformatic analyses and manuscript preparation, with critical feedback from J. Weinberg, M. Kobor, J. Reynolds, and P. Pavlidis.   Chapter 6 (discussion) contains excerpts from a published review: • Lussier AA, Weinberg J, Kobor MS. 2017. Epigenetics studies of fetal alcohol spectrum disorder: where are we now? Epigenomics. The remainder is original and unpublished.    All animal protocols were approved by the University of British Columbia Animal Care Committee and are consistent with the NIH Guide for the Care and Use of Laboratory Animals (certificates: A06-0017, A07-0381, A10-0136, A10-0016, A12-0032).  Ethics for the clinical cohorts of FASD were reviewed and approved by the “Children's and Women's Research Ethics Board – Clinical” (H10-01149). Experimental procedures were reviewed and approved by the Health Research Ethics Boards at Queen's University, University of Alberta, Children's Hospital of Eastern Ontario, University of Manitoba, and the University of British Columbia.   viii Table of Contents Abstract .......................................................................................................................................... ii!Lay Summary ............................................................................................................................... iv!Preface .............................................................................................................................................v!Table of Contents ....................................................................................................................... viii!List of Tables ............................................................................................................................. xvii!List of Figures ........................................................................................................................... xviii!List of Abbreviations ................................................................................................................... xx!Acknowledgements .................................................................................................................. xxiii!Chapter 1: Introduction ................................................................................................................1!1.1! General overview and hypotheses ..................................................................................... 1!1.2! Fetal alcohol spectrum disorder ......................................................................................... 3!1.3! Animal models of FASD ................................................................................................... 5!1.4! Reprogramming of physiological systems by PAE ........................................................... 6!1.4.1! PAE alters hypothalamic-pituitary-adrenal axis function ........................................... 6!1.4.2! PAE induces changes in homeostatic systems ............................................................ 9!1.4.3! PAE causes alterations to immune function and regulation ..................................... 11!1.5! Fetal programming as a framework for the interpretation of PAE-induced deficits ....... 14!1.6! Epigenetic mechanisms link environmental exposures and cellular programs ............... 15!1.6.1! DNA modifications ................................................................................................... 16!1.6.1.1! DNA methylation ............................................................................................... 16!1.6.2! Non-CpG DNA methylation ..................................................................................... 18!1.6.3! DNA hydroxymethylation ........................................................................................ 20!  ix 1.7! Evidence for genetic and epigenetic changes following PAE ......................................... 21!1.7.1! PAE causes both transient and persistent alterations to gene expression programs . 21!1.7.2! PAE alters DNA methylation programs ................................................................... 23!1.8! Current diagnostic tools and biomarkers of FASD .......................................................... 27!1.9! Thesis overview ............................................................................................................... 30!Chapter 2: Prenatal alcohol exposure alters steady-state and activated gene expression in the adult rat brain ........................................................................................................................31!2.1! Background and rationale ................................................................................................ 31!2.2! Materials and Methods ..................................................................................................... 33!2.2.1! Breeding and prenatal ethanol exposure ................................................................... 33!2.2.2! Induction of arthritis and termination of animals ..................................................... 34!2.2.3! Tissue dissection and RNA extraction ...................................................................... 35!2.2.4! Microarray assay of whole genome gene expression and quality control ................ 35!2.2.5! Differential gene expression analysis ....................................................................... 36!2.2.6! Verification of microarray results ............................................................................. 37!2.2.7! Gene Ontology and Pathway analysis ....................................................................... 37!2.3! Results .............................................................................................................................. 38!2.3.1! Developmental Data .................................................................................................. 38!2.3.2! Prenatal ethanol exposure altered steady-state levels of gene expression in the PFC and HPC ................................................................................................................................ 39!2.3.3! Verification of results related to prenatal ethanol exposure with RT-qPCR ............ 43!2.3.4! Gene Ontology and Upstream Regulator Analysis of PAE effects under steady-state conditions .............................................................................................................................. 45!  x 2.3.5! Prenatal treatments resulted in common, graded, and differential effects under steady state conditions ...................................................................................................................... 46!2.3.6! PAE altered neural gene expression in response to an inflammatory challenge ...... 47!2.3.7! Gene Ontology and Upstream Regulator Analysis of PAE effects in response to adjuvant ................................................................................................................................. 51!2.4! Discussion ........................................................................................................................ 53!2.4.1! Prenatal ethanol exposure altered neural gene expression under steady-state conditions .............................................................................................................................. 53!2.4.2! Prenatal ethanol exposure altered the gene expression response to adjuvant ........... 55!2.4.3! Limitations ................................................................................................................ 56!2.4.4! Effects of pair-feeding on neural gene expression: Pair-feeding is a treatment in itself 57!2.4.5! Summary and Conclusions ....................................................................................... 58!Chapter 3: Prenatal alcohol exposure alters DNA methylation patterns during early development ..................................................................................................................................59!3.1! Background and rationale ................................................................................................ 59!3.2! Materials and methods ..................................................................................................... 63!3.2.1! Prenatal treatment ..................................................................................................... 63!3.2.2! Sample collection ...................................................................................................... 64!3.2.3! Blood composition analysis ...................................................................................... 65!3.2.4! Statistical analyses of developmental data ................................................................ 66!3.2.5! DNA extraction ......................................................................................................... 66!3.2.6! Methylated DNA immunoprecipitation and next-generation sequencing ................ 67!  xi 3.2.6.1! Sequencing library preparation .......................................................................... 67!3.2.6.2! Methylated DNA immunoprecipitation ............................................................. 67!3.2.6.3! Sample amplification and indexing ................................................................... 68!3.2.6.4! Next-generation sequencing ............................................................................... 69!3.2.6.5! Sequencing pre-processing and quality control ................................................. 69!3.2.7! Bioinformatic analyses .............................................................................................. 70!3.2.7.1! Peakset generation ............................................................................................. 70!3.2.7.2! Data preprocessing and normalization of the developmental dataset ................ 71!3.2.7.3! Data preprocessing and normalization of the BvB dataset ................................ 72!3.2.7.4! Removing cell-type specific DMRs ................................................................... 73!3.2.7.5! DMR identification ............................................................................................ 73!3.2.7.6! Genomic enrichment .......................................................................................... 74!3.2.7.7! Transcription factor binding site analysis .......................................................... 74!3.2.7.8! Gene ontology analysis ...................................................................................... 75!3.2.8! Bisulfite pyrosequencing .......................................................................................... 75!3.3! Results .............................................................................................................................. 76!3.3.1! Developmental data .................................................................................................. 76!3.3.2! The developmental profile of the rat hypothalamus ................................................. 77!3.3.2.1! PAE caused persistent alterations to DNA methylation patterns in the hypothalamus .................................................................................................................... 78!3.3.2.2! PAE-specific DMRs contained a greater proportion of bioinformatically predicted Bhlhe40 and Srebf1 TFBS ................................................................................ 80!  xii 3.3.2.3! Genes in PAE-specific DMRs were enriched for biological processes associated with hypothalamic functions ............................................................................................. 81!3.3.2.4! The Ddr4 DMR was verified by bisulfite pyrosequencing ................................ 83!3.3.3! Tissue-concordant alterations to DNA methylation patterns .................................... 84!3.3.3.1! White blood cell proportions were not different across groups ......................... 85!3.3.3.2! Tissue-concordant alterations to DNA methylation patterns ............................. 85!3.3.3.3! Several bioinformatically-predicted TFBS were enriched in cross-tissue PAE-specific DMRs .................................................................................................................. 88!3.3.3.4! Genes in cross-tissue PAE-specific DMRs were enriched for various biological processes ........................................................................................................................... 88!3.3.3.5! Verification of DMRs by bisulfite pyrosequencing ........................................... 91!3.4! Discussion ........................................................................................................................ 91!3.5! Summary and conclusions ............................................................................................. 100!Chapter 4: DNA methylation signature of human fetal alcohol spectrum disorder ...........102!4.1! Background and rationale .............................................................................................. 102!4.2! Materials and methods ................................................................................................... 105!4.2.1! Participants and samples ......................................................................................... 105!4.2.2! DNA methylation 450K assay ................................................................................ 106!4.2.3! DNA methylation data quality control and normalization ...................................... 106!4.2.4! Differential methylation analysis ............................................................................ 108!4.2.5! Analysis of effects due to familial and diagnosis status ......................................... 108!4.2.6! Genotyping .............................................................................................................. 109!4.2.7! Sub-sample definition ............................................................................................. 110!  xiii 4.2.8! Ethnic group adjustment ......................................................................................... 112!4.2.9! DNA methylation pyrosequencing assay ................................................................ 112!4.2.10! Brain concordance analysis ................................................................................... 112!4.2.11! CpG island distribution ......................................................................................... 113!4.2.12! Functional enrichment analysis ............................................................................. 113!4.2.13! Co-expression analysis .......................................................................................... 114!4.2.14! Differentially methylated region analysis ............................................................. 115!4.3! Results ............................................................................................................................ 115!4.3.1! The NeuroDevNet FASD epigenetics cohort ......................................................... 115!4.3.2! Children with FASD displayed altered DNA methylation patterns ........................ 116!4.3.3! Ethnic background correction identified FASD-specific DNA methylation patterns 117!4.3.4! Technical verification of FASD DM loci by bisulfite pyrosequencing .................. 121!4.3.5! Overlap of BEC FASD signatures with brain tissue gene expression and DNA methylation ......................................................................................................................... 122!4.3.6! FASD DM loci were enriched in regions of high DNA methylation variability .... 123!4.3.7! Multiple DM sites were associated with imprinted genes and the protocadherin gene cluster 124!4.3.8! Association of FASD differentially methylated loci with neurodevelopmental processes and disorders ....................................................................................................... 127!4.3.9! Differentially methylated regions were identified between FASD cases and controls 130!4.4! Discussion ...................................................................................................................... 135!  xiv 4.4.1! Summary and conclusions ...................................................................................... 141!Chapter 5: DNA methylation as a predictive tool for fetal alcohol spectrum disorder ......143!5.1! Introduction .................................................................................................................... 143!5.2! Materials and methods ................................................................................................... 147!5.2.1! The Kids Brain Health Network cohort of children with FASD ............................ 147!5.2.2! DNA methylation 450K assay ................................................................................ 148!5.2.3! DNA methylation data quality control and normalization ...................................... 148!5.2.4! Differential methylation analysis and validation of NeuroDevNet (NDN) findings 149!5.2.5! DNA methylation pyrosequencing assay ................................................................ 150!5.2.6! The NDN cohort of children with FASD ................................................................ 150!5.2.7! Cohort of individuals with autism spectrum disorder ............................................. 151!5.2.8! DNA methylation as a predictor of FASD status .................................................... 152!5.3! Results ............................................................................................................................ 152!5.3.1! The KBHN cohort of children with FASD ............................................................. 152!5.3.2! Children with FASD and typically developing controls showed differential DNA methylation patterns ............................................................................................................ 153!5.3.3! Bisulfite pyrosequencing verified the differential DNA methylation of CACNA1A 157!5.3.4! DNA methylation patterns classified individuals with FASD versus controls ....... 158!5.3.5! The DNA methylation predictors were not biased by ASD in an independent cohort 162!5.4! Discussion ...................................................................................................................... 163!  xv 5.4.1! Summary and conclusions ...................................................................................... 168!Chapter 6: Conclusion ...............................................................................................................170!6.1! Summary and cross-cutting features .............................................................................. 170!6.2! Limitations ..................................................................................................................... 174!6.2.1! Sexual dimorphisms ................................................................................................ 174!6.2.2! Tissue specificity and cellular heterogeneity .......................................................... 175!6.2.3! Genetic background ................................................................................................ 176!6.2.4! Correlation versus causation ................................................................................... 177!6.2.5! PAE versus FASD biomarkers ................................................................................ 178!6.3! Broader considerations for future epigenome-wide studies of FASD ........................... 178!6.4! Future directions ............................................................................................................ 181!6.5! Conclusions .................................................................................................................... 184!References ...................................................................................................................................186!Appendices ..................................................................................................................................213!Appendix A Supplementary materials for chapter 2 ............................................................... 213!A.1! Supplementary figures .............................................................................................. 213!A.2! Supplementary tables ................................................................................................ 218!Appendix B Supplementary materials for chapter 3 ............................................................... 228!B.1! Supplementary figures ............................................................................................... 228!B.2! Supplementary tables ................................................................................................ 236!Appendix C Supplementary materials for chapter 4 ............................................................... 257!C.1! Supplementary methods ............................................................................................ 257!C.2! Supplementary figures ............................................................................................... 261!  xvi C.3! Supplementary tables ................................................................................................ 270!Appendix D Supplementary materials for chapter 5 ............................................................... 307!D.1! Supplementary figures .............................................................................................. 307!D.2! Supplementary tables ................................................................................................ 309!   xvii List of Tables Table 2.1 Differentially expressed genes in the prefrontal cortex under steady-state conditions 40!Table 2.2 Differentially expressed genes in the hippocampus under steady-state conditions ...... 41!Table 2.3 Upstream Regulator Analysis in the PFC of animals under steady-state conditions .... 46!Table 2.4 Upstream Regulator Analysis in the HPC of animals under steady-state conditions ... 46!Table 2.5 Genes differentially expressed in PFC of Ethanol-exposed animals in response to adjuvant. ........................................................................................................................................ 48!Table 2.6 Genes differentially expressed in HPC of Ethanol-exposed animals in response to adjuvant. ........................................................................................................................................ 48!Table 2.7 Upstream Regulator Analysis of the PFC in adjuvant VS saline animals .................... 52!Table 2.8 Upstream Regulator Analysis of the HPC in adjuvant VS saline animals ................... 52!Table 3.1 Pregnancy outcomes and body weights during gestation and postnatal development . 77!Table 3.2 Biological processes enriched in the developmental profile DMRs ............................. 82!Table 3.3 Biological processes enriched in the tissue-concordant DMRs .................................... 90!Table 4.1 Characteristics of the NeuroDevNet FASD cohort ..................................................... 116!Table 4.2 Characteristics of the more genetically-homogenous sub-sample .............................. 119!Table 4.3 Genes containing 3 or more differentially methylated probes .................................... 126!Table 4.4 Gene ontology function enrichment in genes up-methylated in FASD ...................... 128!Table 4.5 Disease-association enrichment in genes up-methylated in FASD ............................ 129!Table 4.6 Top 30 gene-annotated differentially methylated regions associated with FASD ..... 133!Table 5.1 Characteristics of the NeuroDevNet II FASD cohort ................................................. 153!Table 5.2 Genes containing multiple differentially methylated CpGs in FASD ........................ 156!Table 5.3 Summarized results from the classification algorithms .............................................. 162!  xviii List of Figures Figure 2.1 Overview of the experimental design prior to sample collection and microarray analysis .......................................................................................................................................... 35!Figure 2.2 Prenatal treatment alters gene expression patterns under steady-state conditions. ..... 41!Figure 2.3 Prenatal alcohol exposure alters steady-state gene expression at Day 16 post-saline injection. ........................................................................................................................................ 42!Figure 2.4 RT-qPCR verification of genes altered by prenatal alcohol exposure. ....................... 44!Figure 2.5 Adjuvant exposure alters gene expression at Day 16 post-injection. .......................... 49!Figure 2.6 Ethanol-exposed animals show altered response to adjuvant. ..................................... 50!Figure 3.1 Overview of the experimental design .......................................................................... 65!Figure 3.2 PAE-specific DMRs across pre-weaning development of the hypothalamus ............. 79!Figure 3.3 Enrichment patterns of the developmental DMRs ...................................................... 80!Figure 3.4 Bisulfite pyrosequencing verification of the Drd4 DMR ............................................ 84!Figure 3.5 PAE-specific DMRs concordant across the hypothalamus and white blood cells ...... 86!Figure 3.6 Enrichment patterns of the tissue-concordant DMRs .................................................. 87!Figure 4.1 Flowchart of bioinformatic analyses ......................................................................... 111!Figure 4.2 Visualization and verification of differentially methylated probes ........................... 120!Figure 4.3 Differentially methylated probes are located in regions of variable and intermediate DNA methylation ........................................................................................................................ 124!Figure 4.4  Several CpGs associated with SLC22A18 displays down-methylation in FASD cases..................................................................................................................................................... 126!Figure 4.5  FASD up-methylated genes coexpression network .................................................. 130!Figure 4.6  Differentially methylated regions associated with FASD. ....................................... 134!  xix Figure 5.1 Visualization and verification of the differentially methylated probes ..................... 156!Figure 5.2 Several differentially methylated CpGs were located in the FAM59B gene body .... 157!Figure 5.3 Flowchart of bioinformatic analyses for the DNA methylation predictor of FASD . 159!Figure 5.4 Visualization of the training and test set performance for both DNA methylation predictors ..................................................................................................................................... 161!   xx List of Abbreviations 450K array – Illumina HumanMethylation450 BeadChip array AA – Adjuvant-induced arthritis ADHD – Attention deficit hyperactive disorder ANOVA – Analysis of variance ARBD – Alcohol-related birth defect ARND – Alcohol-related neurodevelopment disoder ASD – Autism spectrum disorder BEC – Buccal epithelial cell BH – Benjamini-Hochberg BWA – Burrows-Wheeler Alignment C – Control CBC/Diff – Complete blood count with differential CFA – Complete Freund's adjuvant CpG – Cytosine-guanine dinucleotide CpH – Cytosine dinucleotide, where H = adenine, cytosine, or thymine CGI – CpG island CI – Confidence interval CNS – Central nervous system Ct – Cycle threshold DM – Differentially methylated DMR – Differentially methylated region DNA – Deoxyribonucleic acid   xxi DNAhm – DNA hydroxymethylation E – Ethanol FAS – Fetal alcohol syndrome FASD – Fetal alcohol spectrum disorder FC – Fold change FDR – False discovery rate GD – Gestational day GEO – Gene expression omnibus GO – Gene ontology GSR – Gene score resampling hmC – hydroxymethylated cytosine HPA – Hypothalamic-pituitary-adrenal HPC – Hippocampus HSD – Honest significant different HYP – Hypothalamus KBHN – Kid’s Brain Health Network MACS – Model-based analysis of chromatin immunoprecipitation and sequencing mCH – methylated CpH dinucleotide MDS – Multidimensional scaling meDIP-seq – Methylated DNA immunoprecipitation and next-generation sequencing NDN – NeuroDevNet  NIH – National Institutes of Health ORA – Over-representation analysis   xxii P – Postnatal day PAE – Prenatal alcohol exposure PCR – Polymerase chain reaction PF – Pair-fed pFAS – Partial fetal alcohol syndrome PFC – Prefrontal cortex PND – Postnatal day RNA – Ribonucleic acid ROC – Receiver operator characteristic RPKM – Reads per kilobase per million RT-qPCR – Reverse-transcriptase quantitative polymerase chain reaction SEP – Socio-economic position SES – Socio-economic status SNP – Single nucleotide polymorphism SV – Surrogate variable SVA – Surrogate variable analysis TF – Transcription factor TFBS – Transcription factor binding site TSS – Transcriptional start site UCSC – University of California, Santa Cruz Genome Browser URA – Upstream regulator analysis UTR – Untranslated region WBC – White blood cell   xxiii Acknowledgements Foremost, I would like to express my sincerest graduate to my two phenomenal supervisors, Dr. Michael S. Kobor and Dr. Joanne Weinberg. Thank you for your encouragement, critical insight and feedback, knowledge, “opportunities”, and support throughout the past 5 years. I could not imagine having any better academic mentors and am honored to have had the privilege to learn from their experience and expertise. I cannot thank you enough. Many thanks to my committee members, Dr. Catharine Rankin, Dr. Daniel Goldowitz, and Dr. Wendy Robinson, whom have provided insightful feedback during my comprehensive exam, committee meetings, and many questions. Thank you for being available on short notice and putting up with sometimes tight deadlines.   Many thanks to past and present members of the Kobor and Weinberg labs – none of this would have been possible without your support. Special thanks to Dr. Tamara Bodnar, whom helped tremendously with the collection of my animal study and was always available to fill me in on the particularities of the immune system. Thank you, wonderful Weinbergers – Charlis, Kasia, Linda, Ni, Parker, Vivian, Wayne, and Wendy - for your friendship and feedback on my work. Thank you, Koborites – Alex, Alice, Alyssa, David, Eric, Evan, Grace, Josh, Julie, Kristy, Lisa, Maria, Meaghan, Olivia, Sachini, Sarah (x3), Sumaiya, Tanya, and Yasmin – for your friendship critical input on my many projects. Special thanks to Dr. David Lin for keeping me company during weekends/late nights and for always being there when I needed to vent. I would also like to thank the members Medical Genetic graduate program, including the many professors, fellow graduate students, and in particular, Cheryl Bishop, whom made navigating my various deadlines very easy.    xxiv This work would not have been possible without the aid of our many esteemed collaborators – Matthew Mingay (Hirst lab), Dr. Martin Hirst, Dr. Ab Chudley, Dr. James Reynolds, and Dr. Kristin Hamre. Particular thanks to Dr. Paul Pavlidis and Dr. Elodie Portales-Casamar, whom were always available to discuss my research and provided valuable insight into my methods and results. I have learned tremendously from their expertise and count myself lucky to have had the opportunity to collaborate with them. To Dr. Osman Ipsiroglu, thank you for your mentorship and advice, which has acted as a focal point for my future career path and lifelong goals.   I would like to acknowledge the funding support I have received over the years, including the Developmental Neurosciences Research Training Fellowship from NeuroDevNet and Brain Canada. I would also like to acknowledge the funding agencies that provided the grant support that made this work possible: US National Institutes of Health/National Institute on Alcohol Abuse and Alcoholism (R37 AA007789 and RO1 AA022460); NeuroDevNet (Canadian NCE); and the Canadian Foundation on Fetal Alcohol Research.  Finally, I would like to acknowledge all my family and friends that have been extremely supportive throughout my degree. I could never have made it this far without such a wonderful supporting cast and I cherish the relationships we have built over the years. Special thanks to my parents, who didn’t mind me calling at odd hours of the night to share a success story or complain about not sleeping enough. Finally, I would like to thank Urshila Sriram for her love, acceptance of my many idiosyncrasies, and support of my work and career. I could not ask for a better partner and am extremely thankful for her companionship.    1 Chapter 1: Introduction 1.1 General overview and hypotheses Adverse early-life conditions have the potential to permanently imprint or program physiological and behavioral systems during development and lead to long-term consequences in offspring (Godfrey & Robinson 1998; Hanson & Gluckman 2008). In particular, prenatal alcohol exposure (PAE) can alter the development, function, and regulation of numerous neurobiological and physiological systems, giving rise to lasting cognitive and behavioral deficits, immune dysfunction, motor impairments, and increased vulnerability to mental health problems in over the life course (Zhang, Sliwowska, & Weinberg 2005; Pei et al. 2011; Mattson, Crocker, & Nguyen 2011). In humans, the broad spectrum of these structural, neurocognitive, physiological, and behavioral abnormalities or deficits is known as fetal alcohol spectrum disorder (FASD) (Hoyme et al. 2016; Stratton, Howe, & Battaglia 1996). Although the exact molecular mechanisms underlying the effects of PAE on neurobiological systems are not yet fully elucidated, epigenetic mechanisms are prime candidates for the programming effects of environmental factors on physiological systems, as they may bridge environmental stimuli and neurodevelopmental outcomes to influence health and behavior well into adulthood (Yuen et al. 2011; Shulha et al. 2013; Kobor & Weinberg 2011). Furthermore, DNA methylation is emerging as a potential biomarker of early-life events and disease, which may prove useful in the early diagnosis of children at risk for FASD.  I hypothesized that PAE alters the transcriptional profiles and DNA methylation patterns of genes that are functionally related to the deficits associated with FASD. As such, my overarching aim was to identify genetic and epigenetic mechanisms that may contribute to the spectrum of physiological and neurobiological alterations associated with FASD and act as   2 potential biosignatures of PAE. To this end, I used genome-wide, discovery-driven approaches to assess underlying gene expression programs and epigenomic profiles in animal models of PAE and clinical cohorts of individuals with FASD.  I built on previous studies that showed physiological alterations following PAE. In particular, previous work demonstrated that female PAE rats display a more severe and prolonged course of adjuvant-induced arthritis (AA), suggestive of underlying alterations in immune regulation (Zhang et al. 2012). Following up on these results, I investigated gene expression profiles under basal and immune challenge conditions in the brain of adult PAE animals to determine whether I could identify long-term alterations to gene expression programs in the brain using an established rat model of PAE (Chapter 2). Based on previous research, I hypothesized that PAE animals would display altered baseline transcriptomic profiles in the hippocampus and prefrontal cortex, which have important regulatory inputs into the stress axis and immune system. I also hypothesized that PAE animals would show a differential gene expression response to AA compared to controls due to their increased vulnerability to this immune challenge. Building on the long-term programming effects of alcohol on the neural transcriptome identified through this investigation, I shifted my focus to the biological embedding of PAE through epigenetic mechanisms (Chapter 3). This particular study focused on the hypothalamus, rather than the hippocampus and prefrontal cortex, as it acts as the central common integrator of several physiological systems in the brain and plays key roles in the stress response, immune function, and homeostatic regulation. Given the persistent effects of PAE on these neurobiological systems, as well as the association between epigenetics and transcriptional regulation, I predicted that PAE would alter genome-wide DNA methylation programs in the hypothalamus across early development, particularly within genes associated with these vital   3 functions. I also predicted that some differential DNA methylation patterns would overlap between central nervous system (CNS) tissue and white blood cells (WBC), potentially reflecting systemic effects of alcohol and biomarkers of PAE. Although the animal model provides important insight into the molecular underpinnings of PAE-induced deficits, it may not fully reflect the epigenetic changes found in individuals with FASD. As such, I assessed DNA methylation profiles in a clinical cohort of FASD, hypothesizing that I could identify genome-wide DNA methylation alterations in the buccal epithelial cells (BEC) of individuals with FASD versus controls (Chapter 4). Given the importance of replication in epigenome-wide association studies, I further assessed the findings from the first clinical cohort in a second, independent sample of individuals with FASD (Chapter 5). I predicted that some of the results from the initial study would validate here, representing a robust signature of PAE in humans. As DNA methylation patterns have previously been used to predict prenatal exposures, I further expected that the DNA methylation signature of FASD could be used to develop an early screening tool that could accurately classify individuals as FASD or controls.  Taken together, the identification of persistent gene expression changes and stable epigenetic alterations in the brain and peripheral tissues may provide insight into the etiology of PAE-induced deficits, while building a foundation for the development of accurate biomarkers of FASD.   1.2 Fetal alcohol spectrum disorder PAE can result in a harmful in utero environment that can cause numerous adverse developmental consequences falling under the umbrella of FASD. At the most severe end of the   4 spectrum is fetal alcohol syndrome (FAS), which can occur with chronic exposure to high doses of alcohol (Jones & Smith 1973). The diagnostic criteria for FAS consist of pre- and post-natal growth retardation, a characteristic set of facial dysmorphologies, and central nervous system alterations, including neurological abnormalities, developmental delays, and intellectual impairment (Stratton, Howe, & Battaglia 1996). Exposure to alcohol at levels that do not produce full FAS can result in either partial FAS (pFAS), where only some of the diagnostic features occur, or in numerous alcohol-related effects that can be primarily physical (alcohol-related birth defects, ARBD) or primarily neurobehavioral (alcohol-related neurodevelopmental disorder, ARND), although ARBD and ARND are not mutually exclusive and both may occur in an individual exposed to alcohol in utero (Stratton, Howe, & Battaglia 1996). Importantly, neurobehavioral/-developmental deficits are consistently seen across the spectrum, and include neurocognitive impairment (cognitive function, learning and memory, executive function), impairment in self-regulation (attention, impulsivity, behavioral regulation, stress responsiveness, mood/affect, sleep abnormalities), and deficits in adaptive function (communication, social behavior, activities of daily living) (Carter et al. 2016; Lynch, Kable, & Coles 2015; Panczakiewicz et al. 2016; Doyle & Mattson 2015; Astley et al. 2009; Streissguth & O’Malley 2000).   Despite the recognition of FAS over four decades ago, PAE remains a leading cause of intellectual disability in North America and worldwide. Although current global estimates place the prevalence of FAS and FASD at 2.9 and 22.8%, respectively, regional incidences vary greatly, with some populations displaying up to an estimated 55% prevalence of FAS (Roozen et al. 2016). By contrast, recent active case ascertainment studies in the USA, Italy, Poland, and Croatia have found that FASD prevalence is approximately 2-5% in the general population (May   5 et al. 2009, 2014, 2015; Petković & Barišić 2013; Okulicz-Kozaryn, Borkowska, & Brzózka 2017; May et al. 2011). Importantly, approximately half of women under 30 years of age in the USA have unplanned pregnancies (May et al. 2004). As such, they may not realize they are pregnant until later in gestation and may continue to consume alcohol during the first trimester of their pregnancy. Indeed, it is estimated that 10-15% of women in Canada and the USA continue to drink throughout pregnancy, with approximately 3% continuing to binge drink, which is particularly deleterious to fetal development (Popova et al. 2017; Bonthius & West 1990).  The degree to which alcohol affects development depends on a variety of factors such as timing, pattern, and level of alcohol exposure, overall maternal health and nutrition, and genetic background, which may influence the disparity between maternal drinking rates and the prevalence of FASD (Pollard 2007). Importantly, the adverse neurodevelopmental outcomes of children with FASD often persist well into adulthood, including metabolic changes, immune dysfunction, altered stress responsitivity, and vulnerability to mental health disorders, such as substance use, depression, anxiety, psychosis, and bipolar disorder (Famy, Streissguth, & Unis 1998; Spohr & Steinhausen 2008; Lemoine et al. 2003; Weyrauch et al. 2017; Popova et al. 2016; Streissguth et al. 2004; Barr et al. 2006; Moore & Riley 2015).  1.3 Animal models of FASD Animal models of PAE were first developed in response to the skepticism that greeted the first description of FAS by Jones and Smith in 1973 (Jones & Smith 1973; Jones et al. 1973). These were particularly important in that effects of alcohol could be investigated with a level of control not possible in the clinical setting, including timing, pattern (acute versus chronic), and dose of alcohol, genetic factors, environment, nutrition, and other drugs. An additional important   6 strength of animal models is the ability to make direct correlations between central and peripheral tissues, as clinical studies do not have ready access to critical tissues such as the brain and other organs, except through biopsy or postmortem specimens, and changes in peripheral tissues do not always reflect alterations in the brain. Furthermore, animal models can provide critical insight into the molecular mechanisms underlying effects of PAE, and can thus pave the way for identification of novel biomarkers. Important recent studies have made significant progress in characterizing the neurodevelopmental, physiological, and behavioral alterations associated with PAE, as well as elucidating molecular mechanisms through which these alterations occur at different doses and patterns of alcohol exposure. In vitro studies have provided further vital insights into the mechanisms by which alcohol affects cellular functions, allowing for the dissection of molecular pathways in highly specific and controlled environments (Liu et al. 2009; Zhou, Zhao, et al. 2011; Hicks, Middleton, & Miller 2010; Veazey et al. 2013, 2015; Balaraman, Winzer-Serhan, & Miranda 2012). These different strategies have provided key insights into the altered neurodevelopmental profiles resulting from PAE and highlight the complex and long-term programming effects of alcohol on numerous developmental processes. Overall, these studies have shown that alcohol is an early life insult that programs developing neurobiological systems and markedly increases risk for adverse outcomes, supporting the hypothesis that the effects of PAE on development may involve the reprogramming of physiological systems (Hellemans, Sliwowska, et al. 2010).  1.4 Reprogramming of physiological systems by PAE 1.4.1 PAE alters hypothalamic-pituitary-adrenal axis function The HPA axis regulates the body’s response to stress, reacting to stimuli threatening   7 homeostasis and/or survival. Briefly, stressors activate the parvocellular neurons of the paraventricular nucleus of the hypothalamus, resulting in secretion of corticotropin-releasing hormone (CRH) (reviewed in Jankord & Herman 2008). In turn, CRH stimulates release of adrenocorticotropic hormone (ACTH) from the anterior pituitary gland. ACTH then acts on the adrenal cortex, causing the secretion of glucocorticoids, cortisol in humans and mainly corticosterone in rodents. These feed back to multiple brain regions, such as the hypothalamus, to inhibit further HPA activation (Herman & Cullinan 1997). The hippocampus and prefrontal cortex are also important regulators of the HPA axis, partially controlling the extent of the stress response. While the prefrontal cortex provides both stimulatory and inhibitory inputs to the HPA axis, the hippocampus contains high levels of glucocorticoid receptors, dampening the stress response (Diorio, Viau, & Meaney 1993; Reul & De Kloet 1985; Jacobson & Sapolsky 1991). Under stressful conditions, glucocorticoids induce rapid physiological changes promoting survival, such as increased gluconeogenesis, reduced reproductive function, and suppressed immune response. However, prolonged exposure to high levels of glucocorticoids can produce deleterious effects, including metabolic, cognitive, and immune dysfunction (McEwen & Stellar 1993).  Furthermore, the HPA axis is highly susceptible to programming during early life (Matthews 2002; Eguchi 1969). Given that the pregnant mother and fetus constitute an interrelated functional unit, maternal exposures and hormone changes may shape developmental trajectories in the fetus (Weinberg 1993). In particular, the fetal HPA axis of an alcohol-consuming mother receives conflicting signals, as ethanol crosses the placenta to directly activate the fetal HPA axis, while activating the maternal HPA axis in parallel, which then exerts negative feedback on the fetal system (Eguchi 1969). However, the influence of the HPA axis on   8 the developing organism may be partially dampened by increased 11β-HSD levels in the placenta of PAE animals, which may reduce the degree to which glucocorticoids can cross the placenta (Lan et al. 2017). Nevertheless, these signals have been shown to reprogram the fetal HPA axis, increasing HPA axis activation and causing deficits in recovery following stress (Weinberg et al. 2008).  Importantly, data from both clinical cohorts and animal models of FASD suggest that PAE itself causes widespread reprogramming of HPA axis function. Infants exposed to alcohol in utero show elevated basal levels of cortisol at 2 and 13 months of age, as well as higher post-stress levels at 13 months (Ramsay, Bendersky, & Lewis 1996; Jacobson, Bihun, & Chiodo 1999). In addition, 5-7 month old children also display increased cortisol reactivity in response to the “still-face” procedure, which is used to assess emotion and stress regulation (Haley, Handmaker, & Lowe 2006). Animal models have identified a similar hyperresponsiveness to stressors following PAE, identifying alterations to central regulation of the HPA axis under basal and stress conditions (Ramsay, Bendersky, & Lewis 1996; Jacobson, Bihun, & Chiodo 1999; Haley, Handmaker, & Lowe 2006; Weinberg et al. 2008). Although basal serum levels of corticosterone and ACTH are not altered in PAE animals, CRH mRNA expression is increased in the hypothalamus of both weanling and adult PAE rats under basal conditions, as are POMC mRNA levels in the anterior pituitary (Lee et al. 1990, 2000; Lee & Rivier 1996; Gabriel et al. 2005, 2017; Redei, Clark, & McGivem 1989; Redei et al. 1993). PAE rats also display deficits in the intermediate range of HPA feedback regulation (2-10h), but not in the fast response (seconds to minutes), suggesting that deficits in feedback regulation may act through cellular programs, rather than direct hormonal signaling (Osborn et al. 1996; Hofmann et al. 1999). Taken together, these findings suggest that PAE reprograms the HPA axis, altering its basal tone and responsivity   9 to stressors, which may be influenced by underlying alterations to cellular programs within regions associated with the HPA axis. 1.4.2 PAE induces changes in homeostatic systems In addition to its role in the stress response, the HPA axis is also intimately connected to the physiological systems regulating homeostasis. In particular, the hypothalamus acts not only as a key regulator of endocrine function but also for autonomic regulation and homeostatic control, regulating growth, sleeping patterns, metabolism, body temperature levels, and other vital functions through its many different nuclei (Squire et al. 2008). As glucocorticoids also play an important role in the regulation of these metabolic and physiological processes, their dysregulation following PAE may be due to both direct effects of ethanol on the function of hypothalamic centers and indirect effects caused by HPA axis dysfunction (Dickmeis 2009). Nevertheless, clinical and animal studies have shown that the hypothalamus is particularly vulnerable to the effects of alcohol during development, displaying broad alterations to homeostatic functions following PAE, including disrupted sleep patterns and circadian rhythms, deficiencies in thermoregulation, and disordered metabolism and feeding behavior (Jones & Smith 1973; Chen et al. 2012; Earnest, Chen, & West 2001; Sei et al. 2003; Zimmerberg, Ballard, & Riley 1987; Werts et al. 2014).  Indeed, children and adolescent with FASD show disrupted sleep patterns related to insomnia and parasomnia, with concomitant alterations to melatonin secretion profiles (Chen et al. 2012; Ipsiroglu et al. 2013; Goril et al. 2016). Rodent models have also identified sleep disturbances following PAE, including shorter circadian sleep-wake cycles and alterations to the structure and function of the suprachiasmatic nucleus (SCN) of the hypothalamus, which   10 synchronizes the circadian rhythm with light (Hilakivi 1986; Earnest, Chen, & West 2001; Spanagel et al. 2005; Chen et al. 2006). Additionally, PAE has lasting effects on the regulation of body temperature in response to circadian rhythms, while delaying the development of thermoregulation in rat pups, suggesting a disruption of the complex interplay between the different homeostatic systems of the hypothalamus (Zimmerberg, Ballard, & Riley 1987; Sei et al. 2003).  Disordered eating patterns are also common in alcohol-exposed children, particularly within feeding behaviors related to a lack of satiety, which suggest that regulatory pathways regulating feeding behavior may be dysregulated (Harper et al. 2014; Werts et al. 2014). Furthermore, children with FASD display higher rates of glucose intolerance and hyperinsulemia, suggesting that they may be more vulnerable to metabolic syndrome (Castells et al. 1981; Lee 2012; Fan et al. 2008). Animal models show similar findings of altered glucose homeostasis, as PAE rats display increased insulin resistance and glucose intolerance following a glucose challenge (Chen & Nyomba 2003). Furthermore, adult PAE animals show elevated serum triglyceride levels and alterations to adiposity, which may increase the risk of cardiovascular disease (Pennington, Shuvaeva, & Pennington 2002; Dobson et al. 2012). Although it is difficult to tease apart the molecular and biological pathways that may influence these phenotypes, they suggest a broad reprogramming of metabolic processes and regulatory mechanisms, which may be mediated through alterations in the hypothalamus.  As a whole, these findings suggest PAE may reprogram the developing homeostatic systems, potentially through indirect effects on HPA axis and direct effects on the hypothalamus, which may ultimately act as a final common integrator of the effects of PAE to mediate some of its widespread and lasting organizational effects on physiological systems.    11  1.4.3 PAE causes alterations to immune function and regulation Beyond the lasting deficits in the stress response and homeostatic regulation, clinical studies and animal models of PAE have also identified broad alterations to immune function and response. Clinical data examining alcohol-induced alterations in immune competence in children and adults with FAS/FASD remain limited. Early investigation found that children with FAS show a higher incidence of a range of major and minor infections, including recurrent otitis media, upper respiratory tract infections, urinary tract infections, sepsis, pneumonia, and acute gastroenteritis (Johnson et al. 1981; Ammann et al. 1982; Church & Gerkin 1988). In addition, alcohol-exposed show decreased eosinophil and neutrophil cell counts in, as well as decreased leukocyte response to mitogens compared to non-exposed children (Johnson et al. 1981; Gottesfeld & Abel 1991). More recent studies have shown that very low birth weight newborns exposed to alcohol in utero have a 15-fold higher incidence of early-onset sepsis as compared to controls matched for race, sex, gestational age, and birth weight (Gauthier, Manar, & Brown 2004). High levels of maternal drinking (binge) during the second trimester have also been shown to increase the risk of infection by approximately 4-fold compared to that in unexposed newborns when controlling for smoking, low maternal income, and size for gestational age (Gauthier et al. 2005).  Animal models have corroborated clinical findings, as fetuses and newborn PAE animals display decreased thymus weight, size, and cell numbers, as well as suppressed B cell development (Ewald & Frost 1987; Ewald & Walden 1988; Clausing et al. 1996; Moscatello et al. 1999). These deficits persist into adulthood, with additional alterations to the immune response being revealed as the animal matures, such as deficits in the response of splenic T cells   12 and lymphoblasts to the mitogen Concanavalin A and/or interleukin-2, as well as increased vulnerability to infections (Weinberg & Jerrells 1991; Norman et al. 1991; Gottesfeld et al. 1990; McGill et al. 2009). PAE animals also display greater susceptibility to immune and inflammatory challenges, showing greater increases in plasma levels of pro-inflammatory cytokines and reduced proliferative responses of B-cells following lipopolysaccharide exposure (Zhang, Sliwowska, & Weinberg 2005). In addition, PAE females display increased severity of joint inflammation and a prolonged course of disease in an adjuvant-induced arthritis (AA) paradigm. This model is used to study the interactions between the neuroendocrine and immune systems, mimicking human rheumatoid arthritis, an auto-immune disorder influenced by early-life experiences and potentially mediated through altered neuroendocrine-immune interactions (Harbuz, Rees, & Lightman 1993; Harbuz, Chover-Gonzalez, & Jessop 2003; Chover-Gonzalez et al. 1999; Bomholt et al. 2004; Colebatch & Edwards 2011; Zhang et al. 2012).  Although the mechanisms underlying the immunoteratogenic effects of PAE on the immune system remain unclear, it has been shown that alcohol consumption increases cytokine levels with chronic alcohol consumption during pregnancy increasing levels of key cytokines in both the fetus and mother (Crews et al. 2006; He & Crews 2008; Ahluwalia et al. 2000). Evidence from other fields suggests that immune stimulation and alterations to the fine balance of pro- and anti-inflammatory cytokines during pregnancy is associated with increased risk for neurodevelopmental disorders, including schizophrenia and autism (Howard 2013; Goines et al. 2011). However, direct links between alcohol-related alterations in the prenatal maternal cytokine balance and immune, neurocognitive, or behavioral outcomes associated with FASD, have yet to be established. Recent work from a range of animal models has shown that alcohol exposure generally increases cytokine production within the brain, a marker of neuroimmune   13 activation. Using third trimester equivalent exposure models, cytokine levels were shown to increase in the cerebellum, cortex, and hippocampus in alcohol-exposed animals (Topper, Baculis, & Valenzuela 2015; Drew et al. 2015). In addition, cytokine levels are altered in the brain following PAE, with increased levels detected in the hippocampus and prefrontal cortex, but decreased levels in the hypothalamus (Bodnar, Hill, & Weinberg 2016). Despite inherent differences between these models, such as method and timing of alcohol administration, species, and cytokine detection method, the concordance of these findings highlights that neuroinflammation may be a cross-cutting feature in both FASD and animal models of PAE.  Importantly, exposure to stressors exacerbates immune deficits in PAE animals, suggesting a potential role for the stress response in these immune deficits (Giberson & Weinberg 1995; Giberson et al. 1997). Of note, the HPA axis and immune response display extensive bidirectional communication, sharing numerous ligands, receptors, and regulatory regions. In addition to its key role in the stress response, the hypothalamus is also an important feedback center for cytokines, while other brain regions regulating the HPA axis, such as hippocampus and prefrontal cortex display high levels of cytokine and immune receptors (Bernardini et al. 1990; Cunningham & De Souza 1993). Furthermore, pro-inflammatory cytokines can stimulate the HPA axis, which, in turn, has the ability to suppress immune function (Haddad, Saadé, & Safieh-Garabedian 2002). As such, PAE may affect the fine-tuned reciprocal interactions between these systems, leading to alterations in both HPA axis and immune function. Overall, the findings of persistent immune deficits and altered responses to immune stressors suggest a reprogramming of immune functions by PAE, potentially acting in concert with the rewiring of brain regions involved in both the stress and immune response.   14 1.5 Fetal programming as a framework for the interpretation of PAE-induced deficits Overall, the persistence of FASD-associated deficits suggests that physiological and neurobiological systems may be reprogrammed by PAE during early life, resulting in a markedly increased risk for adverse outcomes later in life (Hellemans, Sliwowska, et al. 2010). These findings support a role for the interpretation of PAE-induced deficits through the fetal programming hypothesis. This concept suggests that early environmental or non-genetic factors, including maternal undernutrition, stress, and exposure to drugs or other toxic agents, can permanently organize or imprint physiological and neurobiological systems to increase adverse cognitive, adaptive, and behavioral outcomes, as well as vulnerability to diseases or disorders later in life (Godfrey & Robinson 1998; Hanson & Gluckman 2008; Swanson et al. 2009). This concept was first formulated based on epidemiological evidence that low birth weight and other indices of poor fetal growth are associated with increased biological risk for coronary heart disease, hypertension, and type II diabetes/impaired glucose tolerance (i.e., metabolic syndrome) in adult life (Barker et al. 1989, 1993, Barker 1997, 2003, 2004; Barker & Osmond 1986; Barker & Thornburg 2013). Subsequent research revealed that low birth weight per se is unlikely the cause of these risks for disease; rather, low birth weight is a proxy for prenatal environmental adversity, and common factors likely underlie both intrauterine growth retardation and altered physiological/metabolic function (Welberg & Seckl 2001).  Current thinking extends beyond these initial findings, suggesting that signals received during development, such as nutritional and hormonal status, may preemptively lead the organism towards a phenotype best adapted for the anticipated external environment (Hanson, Low, & Gluckman 2011). However, in the event of a mismatch between early and later life environments, this adaptive response may no longer confer a fitness advantage, but instead, lead   15 to deleterious phenotypes (Godfrey et al. 2007). This early-life programming is a manifestation of developmental plasticity, where a single genotype can lead to multiple phenotypic outcomes due to differing environmental conditions (Barker 2007). Importantly, epigenetic mechanisms are emerging as potential mediators for the biological embedding of these early-life environments, as they provide a link between in utero conditions and the genome in the modulation of subsequent developmental trajectories (Meaney 2010; Boyce & Kobor 2015).   1.6 Epigenetic mechanisms link environmental exposures and cellular programs Although genetics may be considered the inscribed “blueprint” underlying the central dogma of molecular biology (i.e. DNA!RNA!protein), epigenetics can be thought of as the regulatory overlay of genetic sequence that fine-tunes gene activity during development and in response to external signals (Boyce & Kobor 2015). From a historical perspective, the term ‘epigenetics’ was first introduced by Conrad Waddington in the early 1940s to describe ‘‘the branch of biology which studies the causal interactions between genes and their products which bring the phenotype into being” (Waddington 1968). Waddington argued that epigenetics play a critical role in the development of multicellular organisms by creating ‘epigenetic landscapes’ that drive cellular differentiation along a programmed trajectory towards a specific cell-type lineages (Waddington 1968). Since the first introduction of this concept, the field of epigenetics has flourished into a highly active area of study aimed at characterizing the molecular mechanisms underlying gene regulation and biological programming. Today, epigenetics is operationally defined as modifications of DNA and its regulatory components, including chromatin and non-coding RNA, to potentially modulate gene transcription without changing the DNA sequence itself (Bird 2007; Meaney 2010; Henikoff & Greally 2016). Notably,   16 Waddington’s initial hypothesis still holds true:  the ontogeny of the ~200 different cell types in the human body is largely shaped by the unique epigenomic profiles and transcriptional activity of each cellular subtype (Domcke et al. 2015; Schuebeler 2015). Accordingly, epigenetic regulation involves both dynamic tissue- and cell type-specific variation during development, as well as the preservation of the cellular memory required for developmental stability. In addition, epigenetic regulation is now becoming increasingly recognized as a potential biological mediator of environmental influences, which can contribute to sculpting the epigenome, although these effects tend to be subtler than those driven by cell type (Feil & Fraga 2012). Importantly, epigenetic mechanisms exist in a seeming paradox between the stability of cellular identity and plasticity of environmental responses, modulating cellular functions through both short- and long-term responses to stimuli (Boyce & Kobor 2015).   1.6.1 DNA modifications Covalent modifications on DNA nucleotides, primarily cytosine, have long been an established form of epigenetic regulation. Specifically, DNA modifications are comprised of DNA methylation (which can occur in the context of cytosine-guanine (CpG) dinucleotides or at non-CpG positions) as well as oxidized derivatives of DNA methylation such as DNA hydroxymethylation.   1.6.1.1 DNA methylation DNA methylation is arguably the most studied epigenetic mark and involves the covalent attachment of a methyl group to the 5’ position of cytosine, typically at CpG dinucleotide sites (Jones & Takai 2001). These CpG dinucleotides occur relatively infrequently in the genome in   17 order to minimize the potential for DNA methylation-induced sequence mutability as methylated cytosines can undergo spontaneous deamination to thymine (Illingworth & Bird 2009; Weber et al. 2007; Gardiner-Garden & Frommer 1987). Areas with comparatively high CpG content in the genome have been termed “CpG islands” (CGIs) and these CGIs are thought to exist as regions that were either never methylated or only transiently methylated in the germline while the rest of the genome experienced a loss of CpGs at methylated sequences (Illingworth & Bird 2009; Weber et al. 2007; Gardiner-Garden & Frommer 1987). Importantly, the DNA methylation status of the ~ 28 million CpG sites in the human genome is often dependent on genomic context (Jones 2012; Ulahannan & Greally 2015). CGIs, which are associated with approximately 50-70% of known promoters, tend to contain low levels of methylation in somatic cells, while non-island CpGs exhibit generally higher methylation levels (Illingworth & Bird 2009; Weber et al. 2007; Saxonov, Berg, & Brutlag 2006). Moreover, DNA methylation is associated with the regulation of gene expression, although its effects on transcription are highly dependent on genomic context (Lam et al. 2012; Jones & Baylin 2007; Edgar et al. 2014). For example, DNA methylation at gene promoters is generally associated with gene expression silencing, although its role may be more variable within gene bodies (Schuebeler 2015; Jones & Baylin 2007). Conversely, in regions of lower CpG density which flank CGIs, known as “island shores”, high DNA methylation levels are typically associated with highly expressed genes, especially if the associated CGI has low methylation (Edgar et al. 2014; Irizarry et al. 2008; Baubec & Schübeler 2014). While the exact mechanisms remain mostly unknown, transcriptional silencing by DNA methylation may potentially occur through the direct blocking transcription factor binding or the recruitment of transcriptional repressors to promoter, enhancers, or insulator regions (Tate & Bird 1993). Although DNA methylation in promoters and enhancer regions tend to negatively   18 correlate with gene expression within an individual, emerging evidence shows that when comparing a single gene across a population, the association between DNA methylation and gene expression can be negative, positive, or non-existent, highlighting the complex relationship between DNA methylation and transcription (Lam et al. 2012; Gutierrez-Arcelus et al. 2013; Jones, Fejes, & Kobor 2013). Moreover, DNA methylation can be both active, by being a likely cause of gene expression variation, or passive, by being a consequence or an independent mark of gene expression levels (Gutierrez-Arcelus et al. 2013; Jones, Fejes, & Kobor 2013). In addition to its role in transcriptional control, DNA methylation within introns has been associated with altered mRNA splicing, and its presence within certain exons potentially regulates alternative transcriptional start sites (Shukla et al. 2011; Maunakea et al. 2010, 2013) . Finally, DNA methylation in repetitive elements, which comprise more than half of the human genome including intergenic sequences, tends to occur at relatively high levels and is associated with maintenance of chromosome structure and genomic integrity (Cordaux & Batzer 2009; Donnelly, Hawkins, & Moss 1999). Perhaps most importantly, in addition to its role in the regulation of developmental programs, DNA methylation is also emerging as a potential biomarker for early-life exposures due to its stability over time and malleability in response to environmental cues (Bock 2009). For instance, DNA methylation signatures can predict an individual’s risk for eczema or prenatal exposure to smoking with good accuracy, suggesting that DNA methylation profiles could potentially be used to screen for various environmental exposures or disorders (Quraishi et al. 2015; Reese et al. 2017).   1.6.2 Non-CpG DNA methylation Although DNA methylation primarily occurs in the context of CpG dinucleotides, it can   19 also occur at CpH (where H = A/C/T) sites. Previous studies have shown that methylated CH dinucleotides (mCH) occur in cultured embryonic stem cells (ESCs) and induced pluripotent stem cells (Ramsahoye et al. 2000; Lister et al. 2009; Laurent et al. 2010; Ziller et al. 2011; Lister et al. 2012). Moreover, analysis of adult human and mouse CNS neurons found that mCH is specifically enriched in neurons compared to other cell types, as non-CpG methylation is nearly absent in non-neuronal adult somatic cells, but can reach up to ~ 25% of all cytosines in neurons of the adult mouse dentate gyrus (Guo et al. 2014; Ziller et al. 2011; Lister et al. 2013). Levels of mCH increase rapidly during early postnatal brain development (mouse, ~2-4 weeks; human 0-2 years), suggesting that it potentially plays an important role in the regulation of postnatal brain development. Genome-wide profiling also showed that in neurons, mCH is present throughout the 5’ upstream, gene-body, and 3’ downstream regions of genes, where it is negatively correlated with gene expression (Guo et al. 2014; Lister et al. 2013). Furthermore, in vitro plasmid reporter gene analyses have shown that CH methylation is associated with transcriptional repression in mouse neurons (Guo et al. 2014). However, mCH is not associated with gene silencing in all cell types, as non-CpG methylation in ESCs positively correlates with gene expression (Lister et al. 2009). It is thought that the distinct distribution and role in gene expression of mCH in different cell types relates to differences in the relative abundance and activity of specific “readers” and “writers” of non-CpG methylation (Kinde et al. 2015). Furthermore, in addition to CH methylation, very recent research has detected the presence of methylated adenosine nucleotides in vertebrates, suggesting that that DNA modification variants may be more diverse than previously thought (Meyer et al. 2012; Dominissini et al. 2013; Koziol et al. 2015; Meyer & Jaffrey 2016).     20 1.6.3 DNA hydroxymethylation In contrast to the well-characterized mechanisms underlying the establishment and maintenance of DNA methylation, the process of DNA demethylation remains unclear. Thought to involve both active and passive pathways, this phenomenon is vital for typical development and genetic regulation, particularly in the brain (Ooi & Bestor 2008; Wu & Zhang 2014; Tognini, Napoli, & Pizzorusso 2015). Active DNA demethylation may potentially occur through the oxidation of 5-methylcytosine, catalyzed by the Ten-Eleven-Translocation (TET) family of enzymes (Tahiliani et al. 2009; Santiago et al. 2014). This process generates a series of oxidized cytosine base variants, including hydroxymethylcytosine (hmC), formylcytosine, and carboxycytosine (Tahiliani et al. 2009; Ito et al. 2010; Ulahannan & Greally 2015). Although the exact details of active DNA demethylation remain unclear, emerging evidence points to a process involving the coordinated activity of a number of key enzymatic players and intermediate modified cytosine species. These cytosine variants may also play a role in modulating chromatin structure or recruiting various factors to key regions of the genome (Sadakierska-Chudy, Kostrzewa, & Filip 2014). For instance, various members of the methyl-CpG-binding domain protein family display different affinities for hmC, and given their role in recruiting different chromatin modifying complexes, hmC could potentially alter chromatin landscapes throughout the genome (Pfeifer, Kadam, & Jin 2013).  DNA hydroxymethylation (DNAhm) is also present at high levels in pluripotent cells and the brain, where it has been implicated in neural stem cell functions, although its exact functional role remains to be uncovered (Ito et al. 2010; Kriaucionis & Heintz 2009; Santiago et al. 2014). Genome-wide mapping of DNAhm in various brain regions, including the frontal cortex, hippocampus, and cerebellum, identified an enrichment of hmC in gene bodies, which was   21 positively associated with gene transcription, particularly at developmentally activated genes (Lister et al. 2013; Wang et al. 2012). Active DNA demethylation and TET activity is also associated with memory formation and addiction in mice, further supporting its functional role in neural activity (Alaghband, Bredy, & Wood 2016).  1.7 Evidence for genetic and epigenetic changes following PAE 1.7.1 PAE causes both transient and persistent alterations to gene expression programs  Epigenetic factors provide an attractive mechanism to mediate the biological embedding of early life events, and their association with transcription makes gene expressions programs an easy target to first assess the molecular underpinning of PAE-induced deficits. Initial evidence of the genome-wide programming effects of alcohol on the genome was identified through changes in transcription. In particular, genome-wide investigations of gene expression programs have identified widespread alterations to gene expression levels in fetal, neonatal, and adult rodent models of PAE, providing important insight into potential mechanisms and pathways involved in PAE-induced deficits (Green et al. 2007; Hard et al. 2005; Zhou, Zhao, et al. 2011; Downing et al. 2012; Kleiber et al. 2012, 2013, 2014; Lussier et al. 2015).  Given the importance of spaciotemporal gene expression during developmental patterning, it is perhaps not surprising that many of the PAE-induced alterations to the transcriptome are closely related to the stage of development that was assessed. For example, differentially expressed genes during early gestation were generally associated with functions in cellular patterning, growth, and development, suggesting that PAE can interfere with typical developmental programs. As gene expression is highly dynamic, quickly responding to environmental and cellular inputs, transcriptional alterations measured soon after alcohol   22 exposure may reflect the intracellular response to the teratogen, rather than stable programming effects of PAE on the genome. By contrast, gene expression profiling in the adult brain, long after the removal of ethanol, may provide additional insight into the long-term effects of PAE on cellular programs. Although these effects are usually subtler, long-lasting changes to the transcriptome have been identified in the whole brain in male adult mice, suggesting that PAE can have lasting effects on the neural transcriptome.  Alterations identified in the entire embryo or brain likely reflect systemic effects of ethanol on the organism or CNS, respectively, and may reflect the broader alterations of PAE on biological functions. In particular, meta-analyses of gene expression patterns across multiple studies of PAE, ranging from whole embryos on embryonic day 9 in mice to the rat hippocampus on postnatal day 100, identified a general inhibition of transcription by PAE, regardless of the model (Rogic, Wong, & Pavlidis 2016). The differentially expressed genes identified in the combined analyses were mainly involved in protein synthesis, mRNA splicing, and chromatin function, suggesting that PAE may broadly influence the regulatory systems of the cell, irrespective of the timing and dosage of alcohol exposure. More recent studies are beginning to focus on specific brain regions, providing functional insight into some of the deficits observed following PAE. For instance, gene expression patterns in the postnatal day 70 mouse hippocampus are altered by a third trimester equivalent exposure to binge levels of alcohol, which may potentially be related to some of the deficits in spatial learning and memory impairment associated with PAE (Chater-Diehl et al. 2016). A recent study also profiled gene expression patterns in human fetal cortical tissue from late first trimester fetuses with PAE (n=2) (Kawasawa et al. 2017). These embryos displayed a shift in the typical balance of splicing   23 isoforms in addition to widespread alterations to transcriptomic programs, suggesting that PAE may influence the fine balance of splice variants in the brain.  Taken together, these findings support that PAE can have both transient and persistent effects on the genome, which may influence the cellular response to ethanol and mediate the vulnerability to adverse long-term health outcomes. Furthermore, PAE-induced deficits may potentially arise through the disruption of epigenetic programs, concurrent with alterations to gene expression patterns.   1.7.2 PAE alters DNA methylation programs A large number of studies have identified changes in DNA modifications in response to prenatal alcohol exposure, and the current thesis will present a snapshot of the different approaches to assess these alterations, which range from “bulk” levels to candidate gene approaches and genome-wide investigations (here, bulk levels are defined as measures of epigenetic patterns that do not delineate specific regions, but rather represent the total levels within a given tissue or cell population). The first evidence of alcohol-induced changes to DNA methylation programs was generated in a mouse model, where embryos were exposed to alcohol during gestational days (GD) 9-11. This study demonstrated that alcohol reduced bulk levels of DNA methylation in the genome, potentially by inhibiting DNA methyltransferase 1 (DNMT1) activity, and opened the door for future studies of epigenetic mechanisms in FASD (Garro et al. 1991). Several studies have extended this line of evidence by studying the effects of alcohol exposure during various stages of development and identifying alterations to bulk levels of DNA methylation in different brain regions under basal and intervention conditions (Otero et al. 2012; Perkins et al. 2013; Chen, Ozturk, & Zhou 2013; Mukhopadhyay et al. 2013; Nagre et al. 2015;   24 Liyanage et al. 2015; Öztürk et al. 2017). For instance, PAE throughout gestation delays the accumulation of DNA methylation in neural stem cells, and increases DNA methylation levels in the mouse hippocampus, a brain region involved in learning and memory (Chen, Ozturk, & Zhou 2013). This same study assessed bulk DNA hydroxymethylation in parallel, identifying a decrease in the neural progenitor cells of the hippocampus, which suggests widespread alterations to DNA methylation programs (Chen, Ozturk, & Zhou 2013). In addition to assessing the impact of PAE on bulk DNA methylation levels, a number of studies have used bulk DNA methylation levels as a measurable outcome for dietary or therapeutic interventions in combination with different behavioral tasks. For example, choline supplementation has been proposed as a potential intervention due to its role as a methyl donor, and has been associated with the partial rescue of behavioral alterations and increased DNA methylation levels in the hippocampus and prefrontal cortex of PAE rats (Thomas et al. 2007; Otero et al. 2012). Similar outcomes are also observed in embryos and neural stem cells treated with alcohol or 5-azacytidine, a potent inhibitor of DNA methylation, suggesting that alcohol-induced deficits are likely related to altered epigenomic profiles and functions (Zhou, Zhao, et al. 2011). These findings demonstrate that developmental alcohol exposure tends to impair the establishment of typical DNA methylation levels, which may reprogram downstream cellular and biological functions.  Proof of principle of alcohol’s programming effects was further exemplified using the agouti viable (Avy) yellow mouse model, which contains a DNA methylation-sensitive element within the Avy locus that regulates coat color (Wolff et al. 1998). In this model, PAE increased the incidence of pseudo-agouti animals, indicating that specific loci are responsive to the effects of alcohol during development and can influence phenotypic outcomes (Kaminen-Ahola et al.   25 2010). As such, more recent studies have sought to identify specific gene targets of PAE-induced epigenetic effects, either through hypothesis- or discovery-driven approaches. An initial study using cultured cells showed that, rather than a global demethylation of the genome, specific regions become more methylated and others less methylated in response to alcohol exposure, suggesting that some regions may be differentially sensitive to alcohol-induced reprogramming effects (Liu et al. 2009). Numerous groups have invested in targeted analyses of epigenetic patterns in genes associated with the deficits observed in individuals with FASD (e.g. immune, stress, cognitive, and otherwise-related) (Vallés et al. 1997; Maier et al. 1999; Downing et al. 2011; Bekdash, Zhang, & Sarkar 2013; Zhang et al. 2015; Ngai et al. 2015; Marjonen et al. 2015; Liyanage et al. 2015). In mice, the expression of Igf2, an imprinted gene involved in growth, is decreased in the embryo and placenta following PAE, concomitant with increased DNA methylation of the differentially methylated region 1 in its promoter and growth deficits in offspring. Choline supplementation during gestation partially rescues the effects of PAE on growth and DNA methylation within this locus, further highlighting a potential role for dietary supplements in the attenuation of alcohol-induced deficits (Downing et al. 2011). PAE also results in increased DNA methylation and decreased expression of proopiomelanocortin (POMC) in the hypothalamus, which is a key regulator of the stress response (Bekdash, Zhang, & Sarkar 2013). Slc6a4, an important serotonin transporter, also displays sex-dependent alterations to DNA methylation and gene expression patterns in the hypothalamus of adult PAE rats (Ngai et al. 2015). While the hypothesis-driven approach has proven fruitful in many regards, it relies heavily on previously identified biological pathways and has not been very successful in identifying novel targets of developmental alcohol exposure.    26 Researchers have also used genome-wide tools to study the effects of alcohol exposure beyond classical candidate pathways (Liu et al. 2009; Hicks, Middleton, & Miller 2010; F.C. Zhou, Chen, & Love 2011; Laufer et al. 2013; Krishnamoorthy et al. 2013; Khalid et al. 2014; Chater-Diehl et al. 2016; Laufer et al. 2015; Portales-Casamar et al. 2016). For example, widespread changes in DNA methylation patterns were identified in the brains of adult male mice, with some alterations overlapping with changes in gene expression profiles (Laufer et al. 2013; Chater-Diehl et al. 2016). These findings provide evidence for the lasting effects of developmental alcohol exposure on the DNA methylome, as well as identifying novel genes associated with alcohol exposure. Moreover, these PAE-related changes in the DNA methylome may alter transcriptional profiles and reprogram physiological systems. The analysis of DNA methylation profiles in buccal epithelial cells (BECs) of children with FASD has revealed widespread alterations to the epigenome, and provided preliminary evidence of a DNA methylation “signature” of FASD (Laufer et al. 2015; Portales-Casamar et al. 2016). While the use of a peripheral tissue, buccal epithelial cells, makes it difficult to readily interpret these findings in the context of FASD-associated deficits, these studies provide important insight into potential biomarkers of PAE in human populations.  These studies highlight the widespread effects of developmental alcohol exposure on DNA methylation patterns, although the direction of change varies depending on the model of alcohol exposure, the tissue analyzed, and the specific genes assessed. While most studies of bulk DNA methylation identify a decrease in methylation levels, potentially due to lower activity of DNA methyltransferases and the inhibition of 1-carbon metabolism by alcohol, results have varied across models due to a number of factors, including differences in levels and timing of alcohol exposure, developmental stage, analyzed tissue, and analysis methods. These findings   27 highlight the importance of using different models to assess the molecular mechanisms underlying the effects of ethanol at different stages, doses, etc. The analysis of DNA hydroxymethylation also remains an elusive topic of research in the context of FASD, only being investigated in a two studies of PAE (Chen, Ozturk, & Zhou 2013; Öztürk et al. 2017). Given its seemingly key role in neurons, it could potentially play an important role in the etiology of FASD. As a whole, multiple lines of evidence support a role for DNA methylation in the fetal programming of biological systems by PAE and represent an important avenue for the discovery of biomarkers of FASD.   1.8 Current diagnostic tools and biomarkers of FASD Early identification and diagnosis of FASD is crucial to mitigate the long-term deficits caused by PAE. While FAS is readily distinguishable due to its well-characterized features (facial dysmorphisms, growth retardation, and CNS alterations), the identification and diagnosis of all individuals under the umbrella of FASD has proven more difficult, as the majority do not present with any physical manifestations of the disorder (Hoyme et al. 2016; Mattson et al. 2013). A diagnosis of ARND requires confirmation of prenatal alcohol, which is not always readily available from medical records or the biological mother (Riley, Infante, & Warren 2011). However, these individuals can still have considerable neurobiological/behavioral impairments, which are often not diagnosed until they reach school age, when their deficits become more apparent in the face of increased social and cognitive pressure (Senturias & Baldonado 2014). Furthermore, behavioral and cognitive interventions may be effective at mitigating some of the deficits caused by PAE and improving long-term health in individuals with FASD (Paley & O’Connor 2011). As earlier diagnosis of FASD is associated with increased positive outcomes,   28 and interventions may have the greatest impact during early development, early screening tools are being developed to aid in the identification and diagnosis of children at-risk of FASD (Streissguth et al. 2004; Fox, Levitt, & Nelson III 2010).  Self-report questionnaires and observations of alcohol-induced physical and neurobehavioral alterations are currently the gold standard for the initial screening of FASD. However, these can often lead to the underestimation of alcohol consumption behavior during pregnancy (Russell et al. 1996; Jones, Bailey, & Sokol 2013; Burns, Gray, & Smith 2010). Several groups have begun to investigate alternate molecular and physiological biomarkers of PAE to supplement these methods. Many of these have focused on the direct or indirect products of ethanol metabolism, which can be measured in a number of biological specimen, including maternal blood, urine, hair, saliva, and sweat; newborn blood, urine, hair, and meconium; and the placenta (Concheiro-Guisan & Concheiro 2014; McQuire et al. 2016). For instance, fatty acid ethyl esters (FAEE) are highly associated with PAE when measured in the meconium of newborns, but their specificity is inconsistent between different cohorts and markers, potentially due to the small number of cases in each study (Bakhireva et al. 2014; Bearer et al. 2003, 1999, 2005; Kwak, Han, Choi, Ahn, Kwak, et al. 2014; Ostrea et al. 2006). A composite measure of 4 FAEEs showed high levels of diagnostic accuracy in a very small cohort, though these results have yet to be fully assessed in large independent studies (Bakhireva et al. 2014). By contrast, FAEE measures in the placenta display high sensitivity and specificity, but 30-56% false positives, while maternal-based assays of ethanol metabolism blood, urine, and hair have not yet been shown to identify PAE at both high sensitivity and specificity (Gutierrez et al. 2015; Kwak, Han, Choi, Ahn, Ryu, et al. 2014; Sarkola et al. 2000).  Importantly, these methods assess in utero exposure to alcohol and their use is restricted to a timeframe shortly after birth, limiting   29 their use in later-life diagnoses (Cabarcos et al. 2015). As such, other measures have been developed in order to identify a persistent biological signature of PAE. Eye tracking measures have also been used in a small cohort of children to distinguish children with FASD, ADHD, or typically developing controls with relatively good accuracy using several features obtained from a short testing session (Tseng et al. 2013). Furthermore, the cardiac orienting response could also potentially be used to assess the effects of PAE on infants, as it performs slightly better than the Bayley Scales of Infant Development-II at classifying children as alcohol-exposure or controls (Mesa et al. 2017). A decision tree has also been developed using neurobehavioral and physical measures to distinguish individuals affected by PAE from typically developing controls (Goh et al. 2016). Epigenetic marks are also emerging as potential biomarkers or signatures of early-life exposures, as they may provide a link between environmental factors and genetic regulation. For example, plasma microRNA (miRNA) in alcohol-exposed pregnant mothers, either alone or in conjunction with other clinical variables, could predict infant outcomes (Balaraman et al. 2016). A combination of high variance miRNAs, smoking history, and socioeconomic status could classify infants affected by PAE versus unexposed controls. These findings suggest that maternal plasma miRNAs may predict infant outcomes, and may be useful to classify difficult-to-diagnose FASD subpopulations.  These findings suggest that molecular screening tools may prove useful in early identification of children with FASD, although they require further optimization and validation. DNA methylation is now a unique position for the development of potentially accurate and stable biomarker of prenatal alcohol exposure given its stability over time and its malleability in response to environmental influences.   30 1.9 Thesis overview The overarching goal of this thesis was to test the hypothesis that that PAE alters the transcriptional profiles and DNA methylation patterns of genes that are functionally related to the deficits associated with FASD. The experimental data will be presented through four separate chapters, which will address the specific aims outlined in section 1.1. Chapter 2, entitled “Prenatal alcohol exposure alters steady-state and activated gene expression in the adult rat brain” is based on the previous identification of PAE-induced alterations to an AA challenge by the Weinberg lab, and seeks to identify long-term changes to gene expression patterns in the rat brain. Chapter 3, entitled “Prenatal alcohol exposure alters DNA methylation patterns during early development”, builds on the findings from the previous chapter, assessing the programming effects of PAE on DNA methylation patterns of the rat hypothalamus and white blood cells during early postnatal development. Chapter 4, entitled “DNA methylation signature of human fetal alcohol spectrum disorder”, takes advantage of a clinical cohort of individuals with FASD, determining whether PAE in humans can influence DNA methylation in peripheral tissues. Chapter 5, entitled “DNA methylation as a predictive tool for fetal alcohol spectrum disorder”, follows up on the findings from the previous chapter, attempting to validate the findings in an independent cohort, while simultaneously developing a predictive algorithm for the screening of individuals with FASD. Finally, the main findings from each data chapter will be integrated alongside a discussion of limitations and future directions for these studies.    31 Chapter 2: Prenatal alcohol exposure alters steady-state and activated gene expression in the adult rat brain  2.1  Background and rationale The prevalence of fetal alcohol spectrum disorders (FASD) in North America is estimated at 2-5% of live births, making prenatal alcohol exposure (PAE) a leading cause of neurodevelopmental disorders (May et al. 2009; Sampson et al. 1997). In addition to lasting neurocognitive deficits, impairments in self-regulation, and deficits in adaptive functioning, children with FASD also display changes in a number of physiological systems, including the immune system, with adverse impacts on both innate and adaptive immunity (Johnson et al. 1981; Streissguth, Clarren, & Jones 1985; Gauthier et al. 2005). Animal models have corroborated clinical findings, with PAE animals displaying behavioural and cognitive deficits, including delays in learning and memory, and altered responsivity to stressors (Hellemans, Sliwowska, et al. 2010). Moreover, PAE animals also exhibit altered development of the thymus, decreased lymphocyte proliferative responses to mitogens, increased susceptibility to infections, and greater vulnerability to immune and inflammatory challenges compared to controls (reviewed in Bodnar & Weinberg 2013). PAE animals also show larger increases in plasma levels of pro-inflammatory cytokines, as well as reduced proliferative responses of B cells to lipopolysaccharide (LPS), and splenic T cells and T lymphoblasts to Concanavalin A and/or interleukin-2 (Zhang, Sliwowska, & Weinberg 2005; Weinberg & Jerrells 1991). Likewise, in an adjuvant-induced arthritis (AA) paradigm, PAE animals show increased severity of joint inflammation and a prolonged course of disease (39 days post-injection, higher incidence of arthritis in PAE compared pair-fed [PF] and control [C]   32 animals) (Zhang et al. 2012). These findings suggest that although PAE causes deficits in adaptive immunity, PAE offspring show increased responses to some immune/inflammatory challenges.  The immune, neuroendocrine and central nervous systems have extensive bidirectional communication, sharing numerous ligands and receptors. Brain regions, such as the prefrontal cortex (PFC) and hippocampus (HPC) not only play a role in the regulation of neuroendocrine function, but also respond to immune/inflammatory molecules, including cytokines and neuropeptides (Crofford et al. 1992). For example, adjuvant injection induces c-Fos expression in the hippocampus for up to 4 months, suggesting a role for this region in AA (Carter et al. 2011). Thus, long-term changes in gene expression may modulate AA manifestation and progression. Indeed, mounting evidence suggests a role for altered gene expression in the etiology of FASD (Kobor & Weinberg 2011). Widespread changes to gene expression levels in fetal and neonatal brains following PAE, as well as long-lasting alterations to the neural transcriptome following alcohol exposure during the neonatal (third-trimester equivalent) period or across all three trimesters have been reported (Green et al. 2007; Hard et al. 2005; Zhou, Zhao, et al. 2011; Kleiber et al. 2012, 2013).  Using saline-injected animals (steady-state) as a baseline, the current study examined brains from adult PAE and control females from the lab’s previous AA study to determine whether long-term alterations in gene expression mediate the altered severity and course of arthritis observed in PAE females (Zhang et al. 2012). Since the PFC and HPC play key roles in both neuroendocrine and neuroimmune processes and show altered function following PAE, PAE-induced alterations in the transcriptome of these regions could result in marked downstream effects, including dysregulation of the immune response and neuroendocrine-neuroimmune   33 interactions (Norman et al. 2009). Whole genome microarrays were utilized to assess gene expression in the PFC and HPC of adult PAE, PF and C females terminated at the peak or during resolution of inflammation (days 16 and 39 post-adjuvant injection, respectively); cohorts of saline-injected PAE, PF and C females were terminated in parallel. Under steady-state condition, we identified changes in gene expression and altered activation states of upstream regulators specific to PAE. Furthermore, at the peak of inflammation, we found not only changes in genes related to PAE, but also, a failure of PAE animals to mount appropriate responses to the immune challenge, showing no change in the activation or inhibition of inflammation-related genes and upstream regulators identified in controls.  2.2 Materials and Methods 2.2.1 Breeding and prenatal ethanol exposure All animal protocols were approved by the University of British Columbia Animal Care Committee and are consistent with the NIH Guide for the Care and Use of Laboratory Animals (National Research Council 2011). Details of the breeding and feeding procedures have been published (Glavas et al. 2007). Briefly, male and female Sprague-Dawley rats (Animal Care Center, University of British Columbia) were paired; presence of a vaginal plug indicated gestation day (GD) 1. Pregnant dams were singly housed and assigned to experimental groups: Prenatal ethanol exposure (PAE; ad libitum access to liquid ethanol diet, 36% ethanol-derived calories); Pair-fed (PF; liquid-control diet, maltose-dextrin isocalorically substituted for ethanol, in the amount consumed by a PAE partner, g/kg body weight/GD); or Ad libitum-fed control (C; laboratory chow, ad libitum). All animals had ad libitum access to water. Experimental diets (Weinberg/Kiever Ethanol Diet #710324, Weinberg/Kiever Control Diet #710109, Dyets Inc.,   34 Bethlehem, PA) were fed from GD 1-21, then replaced with laboratory chow. Litters were weighed and culled at birth to 5 males and 5 females, when possible. Following weaning (postnatal day 22), offspring were group-housed by litter and sex. Female offspring were used in the present study due to their increased susceptibility to arthritis (Whitacre 2001).   2.2.2 Induction of arthritis and termination of animals Details of the adjuvant-induced arthritis (AA) paradigm have been published (Zhang et al. 2012). Female offspring (50-65 days of age) from C, PF, and PAE groups received an intradermal injection of 0.1 ml of a 12 mg/ml suspension of complete Freund’s adjuvant (CFA) or 0.1 ml physiological saline at the base of the tail. Animals were single-housed post-injection, and monitored for clinical signs of arthritis under light anesthesia with isofluorane. Paws were scored individually for redness and swelling on days 7, 10, and every other day thereafter until day 39 following injection (Zhang et al. 2012). Animals were terminated by decapitation, following brief exposure to CO2, in two cohorts: day 16 post-injection or day 39 post-injection (peak or resolution phase of AA, respectively). Each cohort contained 9 adjuvant-injected animals and 5 saline-injected animals for each group (C, PF, and PAE). Brains were rapidly removed, immediately frozen on dry ice, and stored at -70 °C.    35  Figure 2.1 Overview of the experimental design prior to sample collection and microarray analysis  Adult female rats from one of three prenatal treatment groups, control (C), pair-fed (PF), and prenatal alcohol exposure (PAE), were injected with complete Freund’s adjuvant (CFA) to cause adjuvant-induced arthritis (AA). Animals were terminated 16 or 39 days post-injection and microarray analysis of gene expression was performed on the prefrontal cortex (PFC) and hippocampus (HPC).  2.2.3 Tissue dissection and RNA extraction Brains were thawed to 4 °C, and the PFC and HPC were dissected, placed in RNAlater, and stored at -20 °C. Total RNA and DNA were simultaneously extracted from the tissues (Qiagen AllPrep DNA/RNA Mini kit). RNA integrity was determined using the Agilent BioAnalyzer mRNA Nano assay.   2.2.4 Microarray assay of whole genome gene expression and quality control The Ambion Illumina TotalPrep RNA Amplification kit was used to generate cRNA (750 ng) from total RNA (250 ng) for each sample. Expression data were obtained using the Illumina   36 RatRef-12 Expression BeadChip microarray with the Illumina iScan, which provides probe-level data for all expressed genes (~ 1 probe per gene). Datasets were filtered to remove control probes and probes with a detection p-value >0.05 in comparison to negative control probes. After filtering, 20215 and 20069 probes remained in the PFC and HPC, respectively (out of a total 23350 probes). The filtered, log2-transformed gene expression profiles were quantile-normalized within each tissue. 2.2.5 Differential gene expression analysis Gene expression analysis utilized the sva and limma packages in the statistical program R (Smyth 2005). Using sva, surrogate variables representative of heterogeneity from sources other than experimental treatments (e.g. batch effects) were generated. These were included in linear modeling of gene expression with limma, which uses moderated F- and t-statistics to identify significant differences. Gene expression changes were modeled in two ways using separate sample means: effects of prenatal treatment alone on steady state levels of gene expression (saline-injected animals, n=5 per C, PF, PAE group), and interaction of prenatal treatment with an inflammatory challenge (adjuvant- versus saline-injected animals; n=5 for saline, n=9 for adjuvant per C, PF, PAE group). Each probe received a moderated F-statistic, and their p-values were corrected for multiple testing using Benjamini-Hochberg correction. The false-discovery rate (FDR) was controlled at <25% due to the moderate alcohol-exposure paradigm and its relatively subtle effects. Significant changes in PAE compared to controls had a moderated t-statistic p-value <0.05. Sequences for significant probes were queried against the RefSeq database for Rattus norvegicus to identify target transcripts.    37 2.2.6 Verification of microarray results Differentially expressed genes were verified using reverse-transcription quantitative real time PCR (RT-qPCR) on the Corbett Rotorgene 6000 for both PFC and HPC, with the same RNA used for microarray analysis (n=4 in both tissues for each C, PF, and PAE).  Primers were designed using well-established guidelines to obtain gene-level data and multiple reference genes were used to normalize expression data (Nolan, Hands, & Bustin 2006). Three reference genes across a spectrum of expression levels and no evidence for differences across groups (F-statistic p-value >0.05) were selected for each tissue (Supplementary table 2.4). The normalization factor for each sample was calculated using the geometric mean of cycle threshold (Ct) values (Vandesompele et al. 2002). Expression levels relative to the factor were determined, and analysis of variance (ANOVA) was conducted to test for significant differences between groups (Schmittgen & Livak 2008).  2.2.7 Gene Ontology and Pathway analysis Gene ontology (GO) analysis was conducted to identify “Biological Processes” enriched for the effects of prenatal treatment and adjuvant exposure using the gene-score resampling (GSR) method in ermineJ (Lee et al. 2005). The set of candidate FASD genes from the curated Neurocarta database was included in the analysis as a custom GO term (Portales-Casamar et al. 2013) (Supplementary table 2.1). Benjamini-Hochberg correction was used with an FDR of 1% within single brain regions to identify more robust functional enrichment categories. By contrast, a 10% cutoff was used when comparing overlapping effects between brain regions to a broader picture of the effects of PAE on the brain’s transcriptome. Where many GO categories were identified, these were mapped to their parent GO Slim terms using CateGOrizer to determine   38 common categories of altered function (Hu, Bao, & Reecy 2008). Following GO analysis, the Ingenuity© Upstream Regulator Analysis tool (URA, Ingenuity Systems Inc., Redwood City, CA) was used to predict master transcriptional regulators that explain the observed expression changes within the dataset. Genes with a fold-change ≥ 1.2 and p < 0.05 between treatments were analyzed for effects of PAE and adjuvant injection. For steady-state effects of PAE, prenatal groups were compared, while adjuvant effects were assessed by comparing adjuvant- to saline-injected animals in each prenatal group. Significantly activated and inhibited genes were identified through a Z-score > 2 or < -2 respectively, as well as an overlap p-value ≤ 0.1, calculated by Fisher’s Exact test.   2.3 Results 2.3.1 Developmental Data As expected, body weights of PAE dams were lower than those of controls (p<0.001) by the end of pregnancy (GD21) [Group x Day interaction, F(6,99)=17.2, p<0.0001], with PF dams intermediate to PAE and C; dams no longer differed in weight by lactation day 8. At birth, PAE (5.7± 0.17 g) females weighed less than their C (6.5± 0.18 g) counterparts (main effect of group, F(2,66)=7.02, p<0.01), which persisted until weaning (PAE, 51.2±1.4 g; PF, 55.3±1.6 g; C, 55.2±1.5 g) (group x day, F(6,99)=1.96, p=0.079). Blood ethanol levels for dams in this paradigm typically average ~100-150 mg/dl (Uban et al. 2010; Lan et al. 2006).     39 2.3.2 Prenatal ethanol exposure altered steady-state levels of gene expression in the PFC and HPC PAE effects on steady-state levels of gene expression were examined in saline-injected females on Days 16 and 39 post-injection (~ PND 75 and 95, respectively). On Day 16, p-value distributions were skewed towards zero for contrasts of PAE vs C and PAE vs PF, suggesting gene expression differences in PAE compared to C and PF females (Supplementary figure 2.1). Following Benjamini-Hochberg correction, significant effects of prenatal treatment were found for 80 and 30 genes in the PFC and HPC, respectively, at 25% FDR (Figure 2.2). While many genes (43% in PFC, 37% in HPC) showed significant effects of ethanol exposure against both control groups, only a subset (15 in PFC, 4 in HPC; p <0.05) showed changes specific to PAE, in that levels were similar between C and PF animals (Tables 1.1, 1.2; Figure 2.3). These had a number of annotated functions in common, including neurodevelopment, differentiation, neuronal signaling, and regulation of cell death and transcription. By contrast, on day 39 post-injection, no relationship between gene expression and PAE was apparent in either brain region, according to p-value distributions (Supplementary figure 2.1). Moreover, only 2 probes met a 25% FDR, but were not specific to PAE effects (data not shown). Thus, subsequent analyses focused on brains from Day 16 post-injection.         40  Gene Symbol Gene Name Average Expression F p-value q-value Fold change EvC EvPF PFvC H2afv Rattus norvegicus similar to H2A histone family, member V isoform 1 (LOC685909) 10.6 18.7 4.8E-05 0.11 0.65 0.76 0.86 Tcf4 transcription factor 4 11.2 11.4 7.0E-04 0.23 0.67 0.66 1.01 Rnasek ribonuclease, RNase K 13.2 11.1 8.0E-04 0.23 0.68 0.57 1.19 Ppp1r14a protein phosphatase 1, regulatory (inhibitor) subunit 14A 10.0 12.6 4.1E-04 0.23 0.68 0.64 1.05 Rps8 ribosomal protein S8 13.0 11.1 7.9E-04 0.23 0.69 0.74 0.93 ILMN_1372701 na 9.4 11.3 7.3E-04 0.23 0.71 0.79 0.90 ILMN_1374168 na 9.1 10.7 9.4E-04 0.25 0.77 0.73 1.05 Pex11g peroxisomal biogenesis factor 11 gamma 7.0 11.5 6.7E-04 0.23 0.82 0.71 1.16 Ndfip1 Nedd4 family interacting protein 1 11.4 12.1 5.1E-04 0.23 1.32 1.37 0.97 Acsl3 acyl-CoA synthetase long-chain family member 3 10.2 12.2 4.9E-04 0.23 1.36 1.36 1.00 Dusp6 dual specificity phosphatase 6 9.9 12.5 4.4E-04 0.23 1.41 1.21 1.17 Rpl7 ribosomal protein L7 11.6 13.7 2.7E-04 0.22 1.44 1.36 1.05 Med28 mediator complex subunit 28 9.2 11.1 7.9E-04 0.23 1.48 1.29 1.15 Atp6ap1 ATPase, H+ transporting, lysosomal accessory protein 1 11.0 10.6 9.8E-04 0.25 1.50 1.35 1.11 Ap1s2 adaptor-related protein complex 1, sigma 2 subunit 9.7 12.4 4.6E-04 0.23 1.60 1.35 1.19 Table 2.1 Differentially expressed genes in the prefrontal cortex under steady-state conditions   41 Genes with a significantly expression under steady-state conditions in PAE compared to both C and PF animals (p <0.05) in the PFC (a) and HPC (b) at D16 post-saline injection. Bold = p <0.05. na = probe had no specific alignment to RefSeq RNA database.  Gene Symbol Gene Name Average Expression F p-value q-value Fold change EvC EvPF PFvC Cnih2 cornichon homolog 2 (Drosophila) 11.1 16.0 8.1E-05 0.14 0.61 0.60 1.01 Caap1 caspase activity and apoptosis inhibitor 1 9.2 15.2 1.1E-04 0.14 0.68 0.71 0.95 LOC688637 similar to WD repeat domain 36 8.8 15.4 1.0E-04 0.14 1.46 1.36 1.08 Rgs3 regulator of G-protein signaling 3 9.1 14.6 1.4E-04 0.15 1.71 1.83 0.93 Table 2.2 Differentially expressed genes in the hippocampus under steady-state conditions Genes with a significantly expression under steady-state conditions in PAE compared to both C and PF animals (p <0.05) in the PFC (a) and HPC (b) at D16 post-saline injection. Bold = p <0.05. na = probe had no specific alignment to RefSeq RNA database.   Figure 2.2 Prenatal treatment alters gene expression patterns under steady-state conditions.  Venn diagram of the number of the number of probes significantly altered in each contrast at Day 16 post-saline injection, with moderated F-statistic q <0.25 and moderated t-statistic p <0.05 (80 in the PFC, 30 in the HPC). The number of probes with unique effects in PAE versus both PF and C animals are highlighted in grey, and listed in Table 1. The center of each Venn diagram shows the number of probes differentially expressed among all three   42 prenatal treatment groups. The intersection on the left of each diagram shows the number of probes with a common effect of prenatal ethanol exposure and pair-feeding. The intersection on the right of each diagram shows the number of probes with a unique effect of pair-feeding.   Figure 2.3 Prenatal alcohol exposure alters steady-state gene expression at Day 16 post-saline injection.  In the prefrontal cortex (a), 15 genes were differentially expressed in response to ethanol. In the hippocampus (b), 4 genes were differentially expressed in response to ethanol. F-statistic q-value <0.25 for all genes identified. A B C PF PAE C PF PAE   43 2.3.3 Verification of results related to prenatal ethanol exposure with RT-qPCR Of 19 probes showing differential expression due to PAE (Tables 2.1, 2.2), 17 aligned to a sequence in the Rattus norvegicus RefSeq database (ILMN_1372701 and ILMN_1374168 were the exceptions). Specific RT-qPCR primers were successfully designed for 15 of the 17 genes (Supplementary table 2.3; Rps8 and Rpl7 were not analyzed).  Despite differences with microarray technology, RT-qPCR verified the differential expression of 2/11 genes in the PFC (Ap1s2, Dusp6) and 1/4 genes in the HPC (Rgs3), all of which showed increased expression (p <0.1; Figure 2.4a). Moreover, for 7 significantly up-regulated genes in the microarray, changes trended in the same direction by RT-qPCR (Figure 2.4b). No down-regulated genes from microarray analysis showed significantly differences in PAE animals by RT-qPCR, but one gene (Cnih2) also trended downward.  Importantly, positive correlation between microarray and RT-qPCR data was obtained for PAE effects (r2=0.35, p<0.02 Figure 2.4b), and significant genes were corroborated by the small differences between methods shown in the Bland-Altman plot (Figure 2.4c). No correlation was found for PF animals (Supplementary figure 2.4). Collectively, the general agreement between qPCR and microarray data suggested that PAE caused persistent alterations to gene expression in the PFC and HPC.   44  Figure 2.4 RT-qPCR verification of genes altered by prenatal alcohol exposure. (a) Three genes were significantly upregulated in PAE animals (Dusp6 and Ap1s2 in PFC; Rgs3 in HPC). Graphs were plotted as fold change to control animals (where C animals expression = 1) ± SEM. ** = p<0.01, * = p<0.05, #   45 = p<0.1. (b) Fold-changes in expression were positively correlated between microarray and RT-qPCR results for E vs C animals (r2=0.3552, p<0.02). Annotated data points represent genes identified as significant in both methods. (c) Bland-Altman plot of genes identified by microarray analysis. Dotted lines represent the 95% limits of agreement (Bias = 0.06467) and annotated data points represent genes identified as significant in both methods.  2.3.4 Gene Ontology and Upstream Regulator Analysis of PAE effects under steady-state conditions GO analysis was performed to ascertain the broad functional impact of PAE-induced changes in gene expression. Following multiple test correction, 6 processes were altered in the PFC of PAE compared to PF and C animals at a 1% FDR (Supplementary figure 2.2A): positive regulation of cell projection organization, chemical/ion homeostasis, response to virus, and regulation of intracellular transport. In the HPC, gene-score resampling (GSR) identified 79 processes specific to PAE, which were involved in metabolism (24%), cell communication (18%), development (18%), transport (15%), and signal transduction (10%) (Supplementary figure 2.2A). At a 10% FDR, several PAE-specific biological processes overlapped between brain regions: positive regulation of neuron differentiation, dorsal/ventral pattern formation, circadian rhythm, regulation of lymphocyte differentiation, and regulation of lipase activity (Supplementary figure 2.2B). Moreover, GSR also identified the NeuroCarta candidate gene list for FASD in the PFC of PAE females (Portales-Casamar et al. 2013).   As noted, gene sets were then analyzed using Ingenuity’s Upstream Regulator Analysis (URA) to predict master regulators driving the observed expression changes within the dataset. In the PFC, a significant activation of Gast and an activation of Lep that approached statistical   46 significance were identified in PAE compared to PF and C animals (Table 2.3), whereas in the HPC, significant differential activation of Laminin and Ifng was observed (Table 2.4).   Gene Symbol Gene Name Predicted status Z-score Overlap p-value EvC EvPF PFvC EvC EvPF PFvC Gast Gastrin Activated 2.1 2.2 NA 0.05 0.009 1.00 Lep Leptin Activated 2.5 2.6 NA 0.12 0.04 1.00 Table 2.3 Upstream Regulator Analysis in the PFC of animals under steady-state conditions Genes identified using Ingenuity Pathway Analysis Upstream Regulator in the PFC of steady-state animals. Genes with a Z-score ≥2 or ≤-2 and an overlap p-value ≤0.1 are considered significant (bold). Those with no overlap had a p-value of 1 and no Z-score (NA).  Gene Symbol Gene Name Predicted status Z-score Overlap p-value EvC EvPF PFvC EvC EvPF PFvC Ifng Interferon-gamma Activated 3.8 2.5 NA 0.05 0.04 1.00 Laminin Laminin Activated 2.0 2.0 NA 0.01 0.04 1.00 Table 2.4 Upstream Regulator Analysis in the HPC of animals under steady-state conditions Genes identified using Ingenuity Pathway Analysis Upstream Regulator in the HPC of steady-state animals. Genes with a Z-score ≥2 or ≤-2 and an overlap p-value ≤0.1 are considered significant (bold). Those with no overlap had a p-value of 1 and no Z-score (NA).  2.3.5 Prenatal treatments resulted in common, graded, and differential effects under steady state conditions A number of prenatal group effects not specific to PAE were observed in the microarray analysis (Figure 2.2). Of the probes affected by prenatal treatment, many showed the same levels of expression in PAE and PF compared to C animals (Supplementary table 2.5), while a handful   47 were altered in opposite directions by ethanol exposure and pair-feeding (Supplementary table 2.6). Conversely, several genes exhibited graded effects of prenatal treatment, with effects of ethanol greater than those of pair-feeding (PAE>PF>C), or vice versa (PF>PAE>C) (Supplementary table 2.6). Pair-feeding also had some unique effects, particularly in the HPC (Supplementary table 2.7), on genes involved in small molecule metabolism, transport, signal transduction, and stress responses. At a 10% FDR, GSR identified two PF-related processes overlapping between the PFC and HPC: negative regulation of neuron projection development and positive regulation of epithelial cell migration (Supplementary figure 2.2C). Moreover, the curated list of candidate FASD genes from NeuroCarta was also identified in the HPC of the PF group (Portales-Casamar et al. 2013).  2.3.6 PAE altered neural gene expression in response to an inflammatory challenge  Consistent with the findings on steady state gene expression, the greatest effects of immune challenge were observed on Day 16 post-injection (peak of inflammation). The dominant neural response to adjuvant across prenatal treatments was an up-regulation of mRNA levels. However, some genes (8 in PFC, and 4 in HPC) were differentially expressed in PAE compared to PF and C animals (Tables 2.5 and 2.6; Figure 1.5). For all hippocampal genes identified, C and PF animals showed a significant up-regulation of expression, while PAE animals showed no change in expression levels between the saline and adjuvant conditions (Figure 1.6). These genes (Ctgf, Lcn2, Sgk, Vwf) were multifunctional, with roles in growth, proliferation, adhesion, structural organization, and cellular response to immunological or stressful stimuli.    48 Gene Symbol Gene Name Average Expression F p-value q-value Fold change (Adjuvant/Saline) C PF E ILMN_1351665 na 7.0 7.9 3.4E-04 0.17 0.80 0.80 1.12 Ghrhr growth hormone releasing hormone receptor 7.0 8.2 2.5E-04 0.14 0.87 0.78 1.23 ILMN_1354124 na 6.9 7.1 7.0E-04 0.24 0.94 0.99 1.34 ILMN_1364624 na 8.4 7.2 6.3E-04 0.24 1.22 1.06 0.51 ILMN_1372588 na 11.1 8.7 1.7E-04 0.13 1.38 1.10 0.67 ILMN_1351971 na 11.9 9.8 6.5E-05 0.08 1.40 1.23 0.71 Flna filamin A, alpha 8.6 7.1 7.1E-04 0.24 1.33 1.27 0.99 Bhlhe40 basic helix-loop-helix family, member e40 9.5 8.1 2.8E-04 0.15 1.42 1.45 1.02 Table 2.5 Genes differentially expressed in PFC of Ethanol-exposed animals in response to adjuvant.  Genes with a significantly different response to Adjuvant in E compared to both C and PF animals (p <0.05) in the PFC at the peak of inflammation (D16). Bold = p <0.05. na = probe had no specific alignment to current RefSeq RNA database.  Gene Symbol Gene Name Average Expression F p-value q-value Fold change (Adjuvant/Saline) C PF E Sgk1 serum/glucocorticoid regulated kinase 1 11.4 9.1 1.1E-04 0.18 1.63 1.67 1.01 Vwf von Willebrand factor 8.9 15.6 7.3E-07 0.00 1.76 1.70 1.06 Lcn2 lipocalin 2 7.4 18.6 1.1E-07 0.00 1.55 1.92 1.03 Ctgf connective tissue growth factor 10.4 11.4 1.6E-05 0.05 1.77 2.14 0.85 Table 2.6 Genes differentially expressed in HPC of Ethanol-exposed animals in response to adjuvant.  Genes with a significantly different response to Adjuvant in E compared to both C and PF animals (p <0.05) in the HPC at the peak of inflammation (D16). Bold = p <0.05. na = probe had no specific alignment to current RefSeq RNA database.   49  Figure 2.5 Adjuvant exposure alters gene expression at Day 16 post-injection.  8 genes showed significant changes in expression among treatment groups in prefrontal cortex (a). 4 genes demonstrated significant changes among treatment groups in the hippocampus (b). F-statistic q-value <0.25 for all genes identified.  A B C PF PAE LOC501092 LOC690672 LOC501605 LOC680012 Cyp2g1 Ghrhr Flna_predicted Bhlhb40 Vwf Ctgf Lcn2 Sgk   50  Figure 2.6 Ethanol-exposed animals show altered response to adjuvant.  In a subset of genes, Ethanol-exposed animals showed no response to Adjuvant, although pair-fed and control animals responded with an upregulation of the gene (Lcn2, Sgk). In others, gene expression levels in ethanol animals were already elevated compared to pair-feds and controls, but did not change in response to the extent of their control counterparts (Ctgf, Vwf).    51 2.3.7 Gene Ontology and Upstream Regulator Analysis of PAE effects in response to adjuvant  GSR identified numerous biological processes altered in response to adjuvant at a 1% FDR. In both the PFC and HPC, PAE animals had the fewest uniquely altered categories (8% in PFC, and 11% in HPC), while C animals had the most (25% in PFC and 30% in HPC) (Supplementary figure 2.3A). Four PAE-specific processes overlapped between brain regions (Supplementary figure 2.3B): regulation and positive regulation of epithelial cell proliferation, cellular protein complex assembly, and regulation of hormone level. In categories identified only in PF and C (normal response to adjuvant exposure), 6 overlapped between the PFC and HPC: response to organic nitrogen, actin filament-based process, actin cytoskeleton organization, regulation of cell morphogenesis, developmental growth, and mRNA metabolic process (Supplementary figure 2.3C). Moreover, URA of gene sets for both the PFC and HPC predicted several master regulators of PAE-specific response to adjuvant, as well as some present only in PF and C animals. In the PFC, 2 PAE-specific genes (Fn1, Dicer1) and 4 PF/C-specific genes (Agt, Foxo3, P38 Mapk, Osm) were significantly activated, while a single PAE-specific gene, Calmodulin, was significantly inhibited (Table 2.7). In the HPC, 2 PAE-specific genes (Adcyap1, Prl) showed significant inhibition and one, Nr1i3, showed marginally significant activation (Table 2.8). As well, PF/C-specific effects were found for Adamts12 (inhibited) and Foxo4 (activated). Of note, Foxo3 approached significance in the HPC of PF and C animals, representing the only overlap between brain regions.     52 Gene Symbol Gene Name Predicted status Z-score Overlap p-value C PF E C PF E PAE-specific         Calmodulin Calmodulin Inhibited NA 0.4 -2.0 1.00 0.02 0.02 Dicer1 Dicer 1, ribonuclease type III Activated NA NA 2.0 1.00 1.00 0.09 Fn1 Fibronectin 1 Activated 1.3 1.1 2.1 0.03 0.0002 0.03 NON-PAE        Agt Angiotensinogen Activated 2.5 2.2 NA 0.02 0.002 1.00 Foxo3 Forkhead box O3 Activated 2.3 3.1 0.2 0.02 0.002 1.00 Osm Oncostatin M Activated 2.9 2.7 NA 0.1 0.07 1.00 P38 Mapk p38 mitogen-activated protein kinase Activated 2.0 3.2 NA 0.04 0.005 1.00 Table 2.7 Upstream Regulator Analysis of the PFC in adjuvant VS saline animals  Genes identified using Ingenuity Pathway Analysis Upstream Regulator in the PFC of adjuvant versus control animals. Genes with a Z-score ≥2 or ≤-2 and an overlap p-value ≤0.1 are considered significant (bold). Those with no overlap had a p-value of 1 and no Z-score (NA). Gene Symbol Gene Name Predicted status Z-score Overlap p-value C PF E C PF E PAE-specific          Adcyap1 Adenylate cyclase activating polypeptide 1 Inhibited NA NA -2.2 1.00 1.00 0.09 Nr1i3 Nuclear receptor subfamily 1, group I, member 3 Activated NA NA 2.2 1.00 1.00 0.12 Prl Prolactin Inhibited 2.3 NA -2.0 0.24 1.00 0.05 NON-PAE         Adamts12 ADAM metallopeptidase with thrombospondin type 1 motif, 12 Inhibited -2.4 -2.0 NA 0.0003 0.001 1.00 Foxo4 Forkhead box O4 Activated 2.0 2.0 NA 0.07 0.04 1.00 Foxo3 Forkhead box O3 Activated 2.6 2.6 NA 0.13 0.13 1.00 Table 2.8 Upstream Regulator Analysis of the HPC in adjuvant VS saline animals  Genes identified using Ingenuity Pathway Analysis Upstream Regulator in the HPC of adjuvant versus control animals. Genes with a Z-score ≥2 or ≤-2 and an overlap p-value ≤0.1 are considered significant (bold). Those with no overlap had a p-value of 1 and no Z-score (NA).   53 2.4 Discussion Prenatal ethanol exposure altered patterns of neural gene expression under both steady-state and immune challenge conditions. In saline-injected females, we identified PAE-induced changes in the expression of Rgs3, Dusp6, and Ap1s2, as well as activation of upstream regulators involved in metabolism and immune function. At the peak of inflammation, adjuvant injection caused PAE-specific changes in gene expression, and uncovered a failure to mount appropriate responses to inflammatory challenge in PAE animal, as evidenced by the absence of changes in inflammation-related genes and upstream regulators identified in controls.   2.4.1 Prenatal ethanol exposure altered neural gene expression under steady-state conditions  Microarray analysis identified unique effects of PAE on 15 and 4 genes in the PFC and HPC, respectively. These had roles in neurodevelopment, cell death, differentiation, transcriptional regulation, and neuronal signaling. Using RT-qPCR, we successfully verified the significant up-regulation of Dusp6 and Ap1s2 in the PFC, as well as Rgs3 in the HPC. Furthermore, the majority of genes not verified by RT-qPCR trended in the same direction as the microarray. The discrepancy in technical verification may arise from the different methods of measurement between the technologies and the underpowered analysis resulting from a relatively low number of samples. Additional large-scale experiments will be required to fully validate these results at the biological level. It is tempting to speculate that these genes play important roles in the cognitive and behavioural deficits observed in FASD. Ap1s2 is involved in neurodevelopment and associated with intellectual disability and autism spectrum disorder, while Dusp6 promotes apoptosis and is   54 linked to bipolar disorder (Borck et al. 2008; Kim et al. 2012). Activation of Laminin could also be involved in the altered neuronal migration patterns observed in PAE brains (Ozer, Sarioglu, & Gure 2000). Moreover, inappropriate feeding behaviour in children with FASD, as well as altered glucose metabolism and insulin tolerance in PAE animals have been reported (Werts et al. 2014; Harper et al. 2014). As Rgs3 negatively regulates glucose output via cAMP production in hepatic cells, it may also play a role in altered energy metabolism within the brain when combined with the activation of gastrin and leptin in the PFC (Raab et al. 2005). Furthermore, the activation of interferon- γ in the HPC supports a role for this cytokine in the altered immune system activity and response to challenge in PAE offspring.  Previous studies on fetal and neonatal brains have uncovered ethanol-induced alterations in the expression of genes related to energy metabolism, adhesion, cytoskeletal remodeling, cell cycle, proliferation, differentiation, apoptosis, as well as neuronal growth and survival (Green et al. 2007; Hard et al. 2005; Zhou, Zhao, et al. 2011). Long-term PAE studies in brains of adult male mice identified networks related to cellular development, free radical scavenging, and small molecule metabolism, as well as genes involved in cognitive function, anxiety, ADHD, and mood disorders (Kleiber et al. 2012, 2013). Interestingly, none of the genes found here directly overlapped with those previously identified. These disparities are likely due to species- and sex-specific effects, differences between exposure paradigms, and different gene expression patterns in whole brains versus specific regions. As such, these discrepancies highlight the importance of examining both sexes and targeted brain regions to gain deeper insight into PAE effects. It is also possible that immediate changes in gene expression in response to PAE may not persist or that environmental influences cause alterations over the course of development. Moreover, the relatively moderate levels of ethanol exposure (BALs ~120-150 mg/dl) in this paradigm are   55 consistent with those reported for children with FASD who show functional and cognitive deficits (Mattson, Crocker, & Nguyen 2011). Perhaps most importantly, the genes identified here have not previously been examined in gene expression studies, suggesting that we have uncovered novel candidates for the effects of PAE in females. Whether our specific changes are mediated through epigenetic mechanisms remains to be investigated (Kobor & Weinberg 2011).  2.4.2 Prenatal ethanol exposure altered the gene expression response to adjuvant PAE-specific responses to adjuvant were found for 8 and 4 genes in the PFC and HPC, respectively. These had roles in growth, proliferation, adhesion, structural organization, and cellular response to immunological or stressful stimuli. Across all prenatal treatments, adjuvant caused a global increase in gene expression compared to saline-injected animals. Importantly, PAE animals failed to exhibit the up-regulation in expression observed in controls for genes related to immune and cellular responses to stressful stimuli (Ctgf, Lcn2, Sgk, Vwf). Up-regulation of immune-related genes normally occurs in the CNS in response to peripheral inflammatory stimuli or neuroinflammation, which occurs in AA (Ousman & Kubes 2012; X. Liu et al. 2012). PAE animals may fail to detect these immune changes and/or launch the appropriate neuroendocrine/neuroimmune response, which could contribute to the prolonged inflammation observed in our previous AA study (Zhang et al. 2012). Consistent with this finding, most master regulators identified in the Upstream Regulator Analysis were involved in the immune response. For example, P38 Mapk plays a role in signal transduction within the normal inflammatory cascade and is only activated in PF and C animals (Cuadrado & Nebreda 2010). Moreover, Adamts12 modulates neutrophil apoptosis during inflammation, while Osm   56 attenuates the inflammatory response (Dumas et al. 2012; Moncada-Pazos et al. 2012). Thus, inhibition of Adamts12 and activation of Osm in control animals may blunt their responses to adjuvant. Furthermore, Adcyap1 modulates anti-inflammatory responses and is neuroprotective in neurons following inflammation (Waschek 2013). Its inhibition in PAE animals suggests a lower level of protection against inflammation than the one that would occur in controls. In turn, as Prl promotes pro-inflammatory responses, its PAE-specific activation suggests an altered response to adjuvant (Brand et al. 2004). Failure of PAE animals to activate Foxo-related pathways may also play a role in their unique response to adjuvant, as knockdown of Foxo3 or Foxo4 increases inflammatory responses (Hwang et al. 2011; Zhou et al. 2009). The possibility that Foxo3 is already up-regulated in PAE animals, and thus may not change further after adjuvant injection remains to be investigated (Kleiber et al. 2013). The activation of fibronectin in PAE animals is interesting, as it is involved in the development of inflammatory arthritis (Barilla & Carsons 2000). Greater production or sensitivity to this protein could underlie the altered course and severity of AA in PAE animals. Finally, activation of Dicer1 in PAE animals suggests alterations to microRNA processing under stress conditions, previously demonstrated following PAE (Guo et al. 2011).  2.4.3 Limitations Although these results suggest a long-term effect of PAE on the brain’s transcriptome, the interpretability of this study is limited by the small number of animals and variability in transcriptomic profiles both at baseline and in response to AA. These factors could have influenced the identification of differential expressed genes and reproducibility of our results by RT-qPCR. In addition, as the FDR was set at a more relaxed threshold (25%) to capture a greater   57 number of differentially expressed genes, more false-positives may have been identified in the analysis, reflected in the low number of genes verified by RT-qPCR. The animals used in the present study were also obtained from an outbred population, and differences in genetic background could have influenced both the physiological response to AA and gene expression profiles.  An additional limitation of this study is that estrus stages were not determined at the time of termination. We have previously shown that PAE induces changes in basal levels of hippocampal glucocorticoid and serotonin Type 1A (5-HT1A) receptor mRNA as a function of estrous stage, which likely have widespread effects on global expression patterns in the brain (Sliwowska et al. 2008). While most females in the present study were likely in diestrus, estrus cycle variation might partially explain intra-group differences in gene expression (Lan et al. 2009). Taken together, these limitations temper our interpretation of the differential expression results, which require further validation in independent cohorts.   2.4.4 Effects of pair-feeding on neural gene expression: Pair-feeding is a treatment in itself A number of genes were similarly altered, or showed graded and differential effects in PAE and PF compared to C animals (Figure 2.3). These may respond to common effects of ethanol exposure and pair-feeding, such as reduced caloric availability or altered stress system regulation. While both PAE and PF animals receive the same number of calories, PAE dams eat ad libitum whereas PF dams receive a reduced ration, likely resulting in hunger and stress (Harris & Seckl 2011). Moreover, PF dams tend to consume their daily ration within a few hours and are deprived until the next feeding, which may have unique metabolic effects associated with   58 “disordered” eating. Our results suggest that the HPC may be susceptible to fetal programming in response to energy-, and stress-related environmental factors. Interestingly, the curated list of candidate FASD genes from NeuroCarta was identified in the HPC of PF animals, suggesting that these genes are potentially related to common mechanisms underlying prenatal alcohol exposure, nutrition, and stress (Portales-Casamar et al. 2013). Studies such as ours are critical to separate the effects of prenatal stress and prenatal alcohol exposure at the level of gene expression.  2.4.5 Summary and conclusions  Our results support the hypothesis that PAE has long-term effects on gene expression patterns in the brain, as well as on the response to a systemic inflammatory insult. As both the PFC and HPC play important roles in cognitive, neuroendocrine, and immune function, the identified changes in steady-state and activated expression likely contribute to immune-related alterations, as well as cognitive and behavioural deficits arising from PAE. Moreover, an inability to mount appropriate response to immune/inflammatory challenges may contribute to the increased vulnerability of individuals with FASD to infections and immune problems. These findings extend our previous data demonstrating that PAE animals exhibit increased susceptibility to and impaired recovery from an inflammatory challenge, and suggest that the adverse impact of prenatal ethanol exposure on the neural transcriptome may underlie long-term health and developmental outcomes observed in individuals with FASD.    59 Chapter 3: Prenatal alcohol exposure alters DNA methylation patterns during early development 3.1 Background and rationale Early-life environments have the potential to influence the development of biological systems, leading to long-term consequences in offspring (Godfrey & Robinson 1998; Hanson & Gluckman 2008). Of relevance, prenatal alcohol exposure (PAE) can lead to the development of Fetal Alcohol Spectrum Disorders (FASD) in humans, which is associated with a wide variety of adverse effects. Importantly, PAE can alter the development, function, and regulation of numerous neurobiological and physiological systems, giving rise to lasting deficits across the spectrum of FASD, including, but not limited to cognitive and behavioral deficits, impairment to self-regulation and adaptive functioning, immune dysregulation, and increased vulnerability to mental health problems across the lifespan (Zhang, Sliwowska, & Weinberg 2005; Pei et al. 2011; Mattson, Crocker, & Nguyen 2011).  Among the affected neurobiological systems, the hypothalamus is highly susceptible to the programming effects of PAE (Matthews 2002; Eguchi 1969). In addition to its vital role in neuroendocrine regulation, the hypothalamus also acts as the main center for autonomic regulation and homeostatic control, regulating growth, sleep/wake behavior, circadian rhythms, metabolism, body temperature, and other vital functions (Squire et al. 2008). Data from both clinical cohorts and animal models of FASD have identified alterations to physiological functions associated with the hypothalamus. For example, infants exposed to alcohol in utero show both elevated basal and post-stress levels of cortisol, and children with FASD and early life adversity exhibit dysregulation of the cortisol circadian rhythm (McLachlan et al. 2016).   60 Similarly, in animal models of PAE, exposed offspring exhibit hyperresponsiveness to stressors as well as altered central regulation of hypothalamic-pituitary-adrenal (HPA) activity (Ramsay, Bendersky, & Lewis 1996; Jacobson, Bihun, & Chiodo 1999; Haley, Handmaker, & Lowe 2006; Weinberg et al. 2008). Furthermore, PAE also alters sleep patterns and circadian rhythms, leads to deficits in thermoregulation, and is associated with inappropriate feeding behavior (Jones & Smith 1973; Chen et al. 2012; Earnest, Chen, & West 2001; Sei et al. 2003; Zimmerberg, Ballard, & Riley 1987; Werts et al. 2014).  These deficits often persist across the life course of individuals with FASD and PAE animals, suggesting that alcohol may alter developmental trajectories during prenatal life to increase the risk of adverse outcomes (Hellemans, Sliwowska, et al. 2010). Indeed, the fetal programming hypothesis suggests that early environmental or non-genetic factors, including maternal undernutrition, stress, and exposure to drugs or other toxic agents, can permanently organize or imprint physiological and neurobiological systems and increase adverse cognitive, adaptive, and behavioral outcomes, as well as vulnerability to diseases or disorders later in life (Godfrey & Robinson 1998; Hanson & Gluckman 2008; Swanson et al. 2009). As the underlying mechanisms of these effects begin to emerge, it has become apparent that epigenetic mechanisms are prime candidates for the programming effects of PAE on physiological systems, linking environmental factors and neurobiological outcomes while influencing health and behavior well into adulthood (Yuen et al. 2011; Shulha et al. 2013). The term epigenetics broadly refers to the modifications of DNA and its packaging that alter DNA accessibility, which modulate gene expression and cell functions without changes to underlying genomic sequences (Bird 2007). These include direct modifications to DNA, post-translational modification of histones, and non-coding RNAs.    61 DNA methylation is perhaps the most studied epigenetic modification and involves the covalent attachment of a methyl group to the 5’ position of cytosine, typically occurring at cytosine-guanine dinucleotide (CpG) sites (Jones & Takai 2001). Although closely linked to the regulation of gene expression, the association between DNA methylation and transcription depends on genomic context. Whereas DNA methylation typically represses gene expression when located in promoter regions, its effects are more variable for CpGs residing in gene bodies and intergenic regions. DNA methylation can also directly control transcription factor binding to gene regulatory regions, such as enhancers, modulating gene expression patterns (Tate & Bird 1993). In addition to this role in transcriptional control, DNA methylation has been associated altered mRNA splicing when located within introns, and its presence within certain exons may potentially regulate alternative transcriptional start sites (Shukla et al. 2011; Maunakea et al. 2013, 2010). Furthermore, DNA methylation is closely linked to several crucial developmental processes, including genomic imprinting, as well as tissue specification and differentiation, suggesting a crucial role in the regulation of cellular functions and developmental trajectories (Ziller et al. 2013; Smith & Meissner 2013). Perhaps most importantly, DNA methylation is responsive to environmental influences and these changes may be inherited through cell divisions to potentially persist throughout the lifetime (Langevin et al. 2011; Hanson et al. 2011; Yuen et al. 2011). As such, an additional interesting aspect of DNA methylation is its emerging role as a potential biomarker of early-life exposures, as it is easily quantifiable, stable over time, and can be obtained from readily available peripheral tissues, such as buccal epithelial cells and white blood cells (Bock 2009). Given their role in the regulation of gene expression and cell function, as well as their responsiveness to environmental factors, epigenetic alterations provide an attractive mechanism   62 for the biological embedding of the persistent deficits caused by PAE. Mounting evidence suggests a potential role for DNA methylation in the etiology of PAE-induced deficits, as numerous studies have identified alterations to epigenetic programs in the central nervous system of animals exposed to alcohol in utero. These range from differences in bulk levels of DNA methylation to genome-wide changes in DNA methylation patterns, suggesting that PAE can alter the epigenome (Bekdash, Zhang, & Sarkar 2013; Laufer et al. 2013). For example, PAE alters the DNA methylation status of the POMC gene in the hypothalamus (Ngai et al. 2015; Bekdash, Zhang, & Sarkar 2013). As a key regulator of the stress response, alterations to this gene may reflect broader alterations to the regulatory functions of the hypothalamus. Although genome-wide studies have been performed on whole brains in mice, few studies have focused on targeted brain regions. Studies from clinical cohorts of children with FASD have also identified widespread changes to DNA methylation patterns in peripheral tissues (Laufer et al. 2015; Portales-Casamar et al. 2016). However, alterations to central tissue are difficult to directly assess in clinical populations, and while peripheral tissues are more easily accessible, changes in these cells may not fully reflect alterations in the brain (Berko et al. 2014). Furthermore, biological embedding of PAE’s effects earlier in development could potentially lead to more systemic effects on the epigenome, which would be reflected by alterations present across a variety of tissues.  Currently, the genome-wide impact of PAE on DNA methylation within the hypothalamus remains unknown (Ngai et al. 2015; Bekdash, Zhang, & Sarkar 2013). To address this gap, we assessed whether PAE alters DNA methylation profiles in the early postnatal period, and whether altered sites of methylation could serve as biomarkers of gestational alcohol   63 exposure if also identified in peripheral tissues. Using methylated DNA immunoprecipitation and next-generation sequencing (meDIP-seq), we identified statistically significant PAE-specific differentially methylated regions (DMR) that persisted across pre-weaning development of the hypothalamus, in regions that could potentially reflect the neurobiological alterations caused by PAE. In parallel, we identified concordant DNA methylation alterations between white blood cells and the hypothalamus of PAE animals compared to controls on postnatal day (P) 22. Our findings suggest that: 1) PAE causes widespread alterations to DNA methylation patterns in both central and peripheral tissues, potentially reprogramming physiological systems and influencing the deficits observed in FASD; and 2) DNA methylation patterns in peripheral tissue reflect some changes in brain, which could represent systemic effects on the organism and potential biomarkers of PAE.  3.2 Materials and methods 3.2.1 Prenatal treatment Details of the procedures for breeding and handling have been published previously (Bodnar, Hill, & Weinberg 2016). Briefly, nulliparous females (n=39) were pair-housed with a male and vaginal lavage samples were collected daily for estrous cycle staging and to check for the presence of sperm, indicating gestation day 1 (GD1). Pregnant dams were singly housed and assigned to one of three prenatal treatment groups: Prenatal alcohol exposure (PAE) - ad libitum access to liquid ethanol diet, 36% ethanol-derived calories, 6.37% v/v, n =13; Pair-fed (PF) - liquid-control diet, maltose-dextrin isocalorically substituted for ethanol, in the amount consumed by an E partner, g/kg body weight/GD), n =14; or Control (Con) - pelleted version of the liquid control diet, ad libitum, n =12. All animals had ad libitum access to water.   64 Experimental diets (Weinberg/Kiever Liquid Ethanol Diet #710324, Weinberg/Kiever Liquid Control Diet #710109, and Pelleted Control Diet #102698, Dyets Inc., Bethlehem, PA) were fed from gestation days 1-21, and then replaced with laboratory chow. Litters were weighed and culled at birth to 6 males and 6 females, when possible.   3.2.2 Sample collection On P1, 8, 15, and 22, female offspring (max 1/litter) were decapitated, trunk blood collected (at P22 only), and brains removed and weighed; the hypothalamus was then quickly dissected and frozen on dry ice in RNAlater (n=7-11/age/group; Figure 3.1; Qiagen, Hilden, Germany). WBCs were isolated using Ficoll-Paque (GE Healthcare, Uppsala, Sweden), which isolates peripheral blood mononuclear cells (PBMC). All tissue collected was left at 4˚C for 1 day and then frozen at -80˚C until DNA extraction. WBCs were stored in RNAlater at -80˚C until DNA extraction. Due to the large number of animals associated with the experimental design of this study, animals were collected across four different cohorts (breedings), spanning January 2012 – December 2013.     65  Figure 3.1 Overview of the experimental design  We collected the hypothalamus of female offspring from one of three prenatal treatment groups on postnatal days (P) 1, 8, 15, and 22. In parallel, white blood cells were collected on P22 from the same animals as the hypothalamus samples. Each group/age/tissue was composed of four samples for DNA methylation analysis by methylated DNA immunoprecipitated and next-generation sequencing (meDIP-seq).  3.2.3 Blood composition analysis Analysis of blood composition was done on samples from a separate but parallel cohort of animals. Briefly, on P22, trunk blood was collected from female offspring (C: n = 6; PF: n = 5; PAE: n = 5), and analyzed using the Advia120 hematology system, which assesses complete blood counts and differential WBC counts (CBC/Diff function). The reported values include counts for neutrophils, lymphocytes, monocytes, eosinophils, basophils, and large unclassified cells (Supplementary table 3.1).     66 3.2.4 Statistical analyses of developmental data Maternal data during gestation and lactation were analyzed using repeated measures analyses of variance (ANOVA), with prenatal treatment as the between-subjects factor, and postnatal day as the within-subjects factor. As separate cohorts of offspring from each prenatal group were terminated on P1, P8, P15, or P22 (n = 4/group/age/tissue), body weights were analyzed by ANOVAs for the factor of prenatal treatment at each age and in a group*age interaction model. Blood composition data were also analyzed using a two-way ANOVA to identify differences among groups for each WBC subtype.  Significant main effects and interactions were further analyzed by Tukey honest significant difference (HSD) post hoc tests (p<0.05).   3.2.5 DNA extraction Total RNA and DNA were simultaneously extracted from the hypothalamus and white blood cells (n=4/group/age/tissue; Qiagen AllPrep DNA/RNA Mini kit, Hilden, Germany). Frozen tissue was thawed on ice, quickly weighed and placed in lysis buffer for 5 minutes. Homogenization was performed by 5 strokes of an 18G needle, 10 strokes of a 20G needle, and 10 strokes of a 23G needle. The resulting homogenate was centrifuged at 21,000g for 3 minutes and the supernatant was collected for DNA and RNA extraction. White blood cells were thawed on ice and then centrifuged at 10,000g for 10 minutes. RNAlater was carefully removed without disturbing the cell pellet and cells were resuspended in lysis buffer. The cells were then frozen at -80˚C to disrupt cell membranes and then thawed on ice. The resulting homogenate was then used for DNA and RNA extraction. DNA concentration was assessed using Qubit Fluorometric Quantitation (Life Technologies, Carlsbad, USA). Full developmental data on the animals can be   67 found in Supplementary table 3.2.   3.2.6 Methylated DNA immunoprecipitation and next-generation sequencing Our methylated DNA immunoprecipitation following by next-generation sequencing (MeDIP-seq) procedures were adapted from a previously published protocol, and are outlined in detail below (Taiwo et al. 2012).   3.2.6.1 Sequencing library preparation For each sample, 500 nanograms of DNA were diluted in a total volume of 60µL of EB buffer (Qiagen, Hilden, Germany). DNA was then transferred to a 96-well plate and sheared for 1 hour using the Covaris Focused-ultrasonicator. DNA was purified using Ampure XP in 20% polyethylene glycol (PEG) beads to obtain fragments sized from 200-500 basepairs (Beckman-Coulter, Brea, USA). Library preparation was performed on the Bravo Automated Liquid Handling Platform (Agilent, Santa Clara, USA) using the TruSeq DNA PCR-Free Sample Preparation Kit (Illumina, San Diego, USA). Following end-repair and A-tailing, adapters were ligated overnight at room temperature. PCR-free library preparation allowed for the conservation of methylated cytosines for subsequent methylated DNA immunoprecipitation. Finally, DNA was resuspended in 35µL of EB buffer (Qiagen, Hilden, Germany). DNA was quality controlled using Qubit Fluorometric Quantitation and the DNA 1000 Bioanalyzer 2100 kit (Agilent, Santa Clara, USA) to verify concentration and fragment size (250-550bp).   3.2.6.2 Methylated DNA immunoprecipitation For each sample, 400 nanograms of DNA were diluted in a total volume of 50µL of IP   68 Buffer (10mM sodium phosphate buffer, pH7.0, 140mM NaCl, 0.05% triton). DNA was denatured by incubation at 95˚C for 10 minutes, followed by the addition of 48µL ice-cold IP buffer and incubation on ice for 10 minutes. 2µL of anti-5-methylcytosine antibody (Eurogenetec, Liège, Belgium), diluted to 1/50 in IP buffer (1µL of antibody per 1µg of DNA ratio), was added to each sample. Immunoprecipitation reactions were incubated for 16 hours at 4˚C with overhead rotation. Following two 5 minute washes with 150uL of 0.1% BSA/PBS, 50µL of Dynabeads Protein G were incubated with 5µL of secondary antibody (rabbit anti-mouse IgG; Jackson Immunoresearch, West Grove, USA) in 45uL ice-cold IP buffer for 15 minutes at room temperature with overhead rotation. Beads were washed twice with IP buffer to remove unbound secondary antibody and resuspended in 50µL IP buffer. The antibody-bound beads were added to the immunoprecipitation reactions and incubated for 2 hours at 4˚C with overhead rotation. Beads were then washed 6 times with 150µL of ice-cold IP buffer and resuspended in 98.97µL of Proteinase K digestion buffer (TE with 0.5% SDS). Following the addition of 1.25µL Proteinase K (20mg/mL; Qiagen, Hilden, Germany), samples were incubated in a thermomixer for 2 hours at 55˚C with a rotation speed of 1250rpm. The reaction was then allowed to cool at room temperature for 15 minutes. Supernatant was collected and bead cleanup was performed using equal volume SeraMag beads with 30% PEG. DNA was resuspended in 35µL of EB buffer (Qiagen, Hilden, Germany).  3.2.6.3 Sample amplification and indexing Two rounds of PCR amplification per sample were performed in order to reduce PCR amplification bias. The reaction mixes were as follows: 15µL DNA, 27µL H2O, 12µL 5X HF buffer, 1.5µL DMSO, 1.0µL paired-end primer (Illumina), 0.5µL Phusion High-Fidelity DNA   69 polymerase (New England Biolabs), 2µL indexing primer (Illumina – specific to each sample). The amplification cycle was as follows: 98˚C for 1 minute, 12X (98˚C for 15 seconds, 65˚C for 30 seconds, 72˚C for 30 seconds), 72˚C for 5 minutes. Reactions from the same sample were pooled and bead cleanup was performed using SeraMag beads in 20% PEG (102µL of beads per 120µL of reaction). DNA was resuspended in a final volume of 35µL of EB buffer.  3.2.6.4 Next-generation sequencing Indexed meDIP libraries were combined in 3 pools of 20 samples each, distributing samples evenly by tissue, age, and prenatal treatment across all three sets. Next-generation sequencing was performed on the three sample pools by the Genome Sciences Centre in Vancouver, BC, Canada. Each sample pool was run on two HiSeq lanes, which produced approximately 600,000,000 paired-end reads of 125 bases per lane.    3.2.6.5 Sequencing pre-processing and quality control Fastq files were aligned to the most current rat genome (Rn6, July 2014) using the Burrows-Wheeler Transform (BWA) tool to obtain .bam files (Li & Durbin 2009). Bam files were filtered using samtools to remove duplicate reads, unpaired reads, and reads with a minimum quality score below 10. Following alignment and filtering, each the two runs for each sample were merged using samtools to obtain a single .bam file for each sample (Li et al. 2009). Supplementary table 3.3 shows sequencing related information: sample pool, sample index, number of raw reads, number of filtered reads, and total number of reads/sample.      70 3.2.7 Bioinformatic analyses 3.2.7.1 Peakset generation Model-based analysis of ChIP-seq (MACS2; version 2.1.0.20140616) was used to identify enriched regions of DNA methylation across the genome (Zhang et al. 2008). This method models the distance between paired sequencing reads by using a sliding window (twice the bandwidth = 600 bp) to find enriched regions throughout the genome. Without a control, this method calculates a dynamic regional lambda for each peak (10000bp windows) to estimate the local bias to enrichment and compute background levels for fold enrichment. P-values for each peak are calculated through a dynamic Poisson distribution, which incorporates background levels estimated by the local lamba. These are corrected (q-values) using the Benjamini-Hochberg multiple-test correction method. The peak calling to identify peak regions (DNA methylation windows) was performed using the ‘callpeaks’ function on paired end bam files with no control input and the following options: –f BAMPE –m 5 50 –bw 300 –g 2.9e9 –q 0.05. Each sample was modeled individually, generating 60 total peaksets. These were imported into R using the DiffBind package (Stark & Brown 2011; Ross-Innes et al. 2012). As all samples had slightly different predicted peaks, peaksets were combined into common regions using the dba.count function in DiffBind, which removed peaks found in less than 3 samples across the entire dataset and provided the total number of reads within each peak/sample. This created a final dataset of 469,162 peaks and 48 samples from the developmental profile of the hypothalamus, and a final dataset of 350,960 peaks and 24 samples in the P22 hypothalamus and WBC (BvB) peakset.      71  3.2.7.2 Data preprocessing and normalization of the developmental dataset First, the total reads within each peak were adjusted to reads/kilobase by dividing the number of reads within each region by their length. In turn, these were converted to reads per kilobase per million (RPKM) by dividing the reads/kilobase by the total number of reads found in the predicted peaks to account for differences in sequencing depth between samples. The samples in the developmental dataset were highly correlated (r>0.95 for all samples), with samples clustering most closely with animals of the same age (Supplementary figure 3.1). No outliers were detected in this first pass analysis.  Principal component analysis of the normalized RPKM data revealed significant levels of variation associated with batch effects. Notably, MeDIP and DNA extraction rounds were associated with a large proportion of variation within the dataset. However, both these factors were highly confounded with age, as all P22 samples were immunoprecipitated separately and separate ages were extracted together (Supplementary figure 3.2). Nevertheless, to account for these effects, ComBat correction was performed on the RPKM data from the hypothalamic samples to correct the effects of MeDIP round and DNA extraction round in the dataset. Age was also slightly confounded with the breeding from which animals were collected, as not all ages were samples from the different cohorts. Interestingly, some partial effects of breeding remained in the dataset following ComBat correction, suggesting that this covariate was not fully confounded with age. Furthermore, prenatal treatment accounted for a larger proportion of variance within the dataset following ComBat correction, suggesting that the removal of batch effects might allow for the identification of more subtle effects of PAE. The corrected and   72 normalized RPKM values obtained from ComBat were used for plotting purposes, but were converted back to reads/kilobase for downstream statistical analyses.  3.2.7.3 Data preprocessing and normalization of the BvB dataset First, the total reads within each peak were adjusted to reads/kilobase by dividing the number of reads within each region by their length. In turn, these were converted to reads per kilobase per million (RPKM) by dividing the reads/kilobase by the total number of reads found in the predicted peaks to account for differences in sequencing depth between samples. Samples in the BvB peakset were highly correlated within tissue (r>0.96), the main driver of DNA methylation patterns, and well correlated within the same animals (r>0.92). However, one PF WBC sample clustered with the hypothalamus samples, suggesting that it may have been mislabeled during processing. As such, this sample was removed from the dataset, resulting in a dataset of 23 samples (Supplementary figure 3.3).  Principal component analysis of the normalized BvB RPKM data revealed significant levels of variation associated with DNA extraction round batch effects (Supplementary figure 3.4). Tissue type was the covariate most strongly associated with variance in the dataset, although it was slightly confounded with extraction round. While ComBat correction was used to account for the effects of DNA extraction round in the BvB dataset, this approach limited our ability to identify tissue-specific differences, as it removed the majority of tissue-associated variance from the dataset. Again, prenatal treatment was associated with a larger proportion of variance within the dataset following ComBat correction. Interestingly, breeding once again remained a major contributor to variability within the dataset, suggesting that differences between cohorts may have an important influence on epigenetic patterns. The corrected and   73 normalized RPKM values obtained from ComBat were used for plotting purposes, but were converted back to reads/kilobase for downstream statistical analyses.  3.2.7.4 Removing cell-type specific DMRs Using previously characterized transcriptomic profiles from mouse neurons, oligodendrocytes, and astrocytes, we identified DNA methylation peaks within genes that are specifically expressed in each different subtype (1.5 fold expression difference compared to other cell types) (Cahoy et al. 2008). Given the relationship between gene expression and epigenetic patterns, it is possible that alterations to the DNA methylation levels of these genes could reflect changes in the cell-type proportions within this dataset. However, the majority of the peaks in the dataset were located within intergenic regions, with no annotated associations with these genes, reducing our ability to capture cell-type related differences. As such, only regions directed located within neuron-, oligodendrocytes-, or astrocyte-specific genes were removed from further analyses to reduce the potential confounding factor of cell type, resulting in a dataset of 451,112 peaks for downstream analyses of the hypothalamus.   3.2.7.5 DMR identification Linear modeling was performed using edgeR, which is typically used to analyze RNA-seq count data and includes a factor to account for the number of reads in each sample (Nikolayeva & Robinson 2014; Robinson, McCarthy, & Smyth 2010). This method was used to identify differentially methylated regions (DMRs) that were consistently different between PAE animals and both control groups across difference ages and tissues. For both analyses, the model accounted for the effects of collection during different breedings, and p-values were corrected for   74 multiple-testing using the Benjamini-Hochberg method. Statistically significant DMRS at a false discovery rate (FDR) <0.05 were obtained for the following contrasts: PAE versus C, PAE versus PF, and Control versus PF. The final PAE-specific DMRs were statistically significant in both PAEvC and PAEvPF, and were not found in the CvPF contrasts.   3.2.7.6 Genomic enrichment Custom annotations were built for each peakset using the UCSC genome browser gene annotations. Briefly, genomic coordinates of all CpG islands, exons, introns, promoters (TSS -200bp and TSS -1500bp), 3’ untranslated regions (UTR), 5’ UTRs for the rn6 genome were obtained as bed files from the table browser. In parallel, MeDIP-seq peaks were converted to the bed file format and the overlap of genomic features with MeDIP-seq peaks was computed iteratively using the intersectBed function from bedtools, retaining only the peaks that contained the assessed genomic feature (Quinlan & Hall 2010). The overlaps were concatenated into a single annotation set in R, where individual peaks contained information for each potential genomic feature. Of note, regions spanning both introns and exons were deemed intron/exons boundaries. P-values for genomic feature enrichment analyses were calculated by computing background levels of genomic features on 1,000 random subsets of DMRs, using the same number of PAE-specific DMRs.   3.2.7.7 Transcription factor binding site analysis Enrichment of different transcription factors binding sites (TFBS) in PAE-specific DMRs was assessed using the motifEnrichment function of the PWMEnrich package (Stojnic & Diez 2013). DMR DNA sequences were obtained from the UCSC genome browser (Rn6 genome). As   75 no binding motifs were available for the Rattus norvegicus genome, motifs from the Mus musculus genome were obtained from the PWMEnrich.Mmusculus.background. Motifs were summarized using the groupReport function. P-values were calculated by performing enrichment analysis on 1,000 random subsets of DMRs, using the same number of PAE-specific DMRs for each analysis to assess background levels of each TFBS in the different peaksets.   3.2.7.8 Gene ontology analysis The gene-score resampling (GSR) tool of ErmineJ (version 3.0.2) was used to identify gene function enrichment in the differentially methylated genes including the Gene Ontology (GO) annotations molecular function, biological process, and cellular component (Lee et al. 2005). The ermineJ GSR tool was set with the following parameters: max gene set size = 2,000; min gene set size = 2; iterations = 10,000. Once again, statistically significant associations (p<0.05 and multifunctionality score <0.05) were obtained for the following contrasts: PAE versus Control, PAE versus PF, and Control versus PF. The final PAE-specific GO terms were statistically significant in both PAEvC and PAEvPF, and were not found in the CvPF contrasts.  3.2.8 Bisulfite pyrosequencing DNA from the same samples as above subjected to bisulfite conversion using the Zymo EZ DNA Methylation Kit (Zymo Research, Irvine, California), which converts DNA methylation information into sequence base differences by deaminating unmethylated cytosines to uracil while leaving methylated cytosines unchanged. Bisulfite pyrosequencing assays were designed with PyroMark Assay Design 2.0 (Qiagen, Hilden, Germany; Supplementary table 3.4). The regions of interest were amplified by PCR using the HotstarTaq DNA polymerase kit (Qiagen,   76 Hilden, Germany) as follows: 15 minutes at 95°C, 45 cycles of 95°C for 30s, 58°C for 30s, and 72°C for 30s, and a 5 minute 72°C final extension step. For pyrosequencing, single-stranded DNA was prepared from the PCR product with the Pyromark™ Vacuum Prep Workstation (Qiagen, Hilden, Germany) and the sequencing was performed using sequencing primers on a Pyromark™ Q96 MD pyrosequencer (Qiagen, Hilden, Germany). The quantitative levels of methylation for each CpG dinucleotide were calculated with Pyro Q-CpG software (Qiagen, Hilden, Germany). Of note, only PAE and Control animals were assessed by bisulfite pyrosequencing. We selected several DMRs for verification by bisulfite pyrosequencing based on their potential role in PAE-induced deficits, mainly focusing on their associated gene.  3.3 Results 3.3.1 Developmental data  To verify that our alcohol exposure paradigm performed as expected, we assessed whether our prenatal treatments influenced maternal weight gain over pregnancy and pup weights (Bodnar, Hill, & Weinberg 2016; Uban et al. 2010; Hellemans, Verma, et al. 2010). On average, alcohol intake of PAE dams was consistently high across pregnancy, ranging from 0.208 ± 0.014 to 0.268 ± 0.022 mL/kg body weight during week 1, 0.240 ± 0.016 to 0.305 ± 0.017 during week 2, and 0.236 ±0.014 to 0.285 ± 0.019 during week 3 of gestation (Table 3.1). These levels of drinking typically result in blood alcohol levels ~100-150 mg/dL (Uban et al. 2010; Hellemans, Verma, et al. 2010). Separate analyses of maternal body weights during gestation (GD1, 7, 14, 21) and following parturition (P1, 8, 15, 22) showed significant main effects of group (F(2,143)=13.609, p =0.0000039 and F(2,91)=9.559, p =0.00017, respectively) and group X day interactions (F(6,143) = 2.869, p = 0.011 and F(2,91) = 2.566, p = 0.082, respectively)   77 during both gestation and lactation. Both PAE and PF dams weighed significantly less than controls on GD14 and 21, and following parturition (P1). However, catch-up weight gain occurred after birth, when the diets were normalized, and maternal weight differences among groups were no longer significant by P8. We did not observe any significant group differences in the number of live-born pups, or in the average weight of female pups/litter at any of the collection days. These results suggested that our paradigm was performing as expected, with PAE dam showing gaining less weight over the course of pregnancy, though these effects were not reflected in the weight of pups.   C PF PAE N Number of pups 14.7 ± 0.5 14.0 ± 0.8 14.4 ± 0.5 39 (12C; 14PF; 13PAE)  Dam weight   GD 1 295.9 ± 5.1 297.3 ± 5.1 299.4 ± 6.9 39 (12C; 14PF; 13PAE)   GD 7 327.8 ± 3.9 314.4 ± 5.6 313.2 ± 6.6 39 (12C; 14PF; 13PAE)   GD 14 380.4 ± 5.2 356.6 ± 6.1† 353.1 ± 5.4†† 39 (12C; 14PF; 13PAE)   GD 21 478.9 ± 9.5 446.0 ± 6.6†† 434.0 ± 5.8††† 39 (12C; 14PF; 13PAE)   P1 393.9 ± 9.1 365.1 ± 7.8† 348.0 ± 6.3††† 39 (12C; 14PF; 13PAE)   P8 373.3 ± 8.4 356.5 ± 8.4 361.6 ± 5.2 28 (9C; 10PF; 9PAE)  P15 367.9 ± 8.7 356.7 ± 8.6 356.9 ± 1.5 19 (6C; 7PF; 6PAE)  P22 346.8 ± 7.1 333.0 ± 11.4 334.0 ± 1.3 12 (4C; 4PF; 4PAE) Pup weight   P1 6.6 ± 0.2 6.5 ± 0.2 6.4 ± 0.1 39 (12C; 14PF; 13PAE)   P8 17.4 ± 0.4 16.8 ± 0.8 16.3 ± 0.7 28 (9C; 10PF; 9PAE)  P15 36.6 ± 2.4 33.1 ± 0.9 35.3 ± 1.3 19 (6C; 7PF; 6PAE)  P22 59.9 ± 0.9 57.2 ± 2.7 59.1 ± 4.2 12 (4C; 4PF; 4PAE) †PAE = PF < C; †p < 0.05; ††p < 0.01; †††p < 0.001 Table 3.1 Pregnancy outcomes and body weights during gestation and postnatal development  3.3.2 The developmental profile of the rat hypothalamus Our initial analysis of this dataset aimed to identify persistent alterations to DNA methylation patterns in the rat hypothalamus across early development (P1 to P22). More specifically, we analyzed the hypothalamus of female offspring on P1, 8, 15, and 22 using methylated DNA immunoprecipitation (meDIP-seq). These ages were selected as they represent   78 key developmental periods, including birth (P1), the brain growth spurt (P8), eye opening (P15), and weaning (P22) and females were utilized due to their underrepresentation in molecular and genome-wide studies of PAE. (Dobbing & Sands 1979; McCormick & Mathews 2010).   3.3.2.1 PAE caused persistent alterations to DNA methylation patterns in the hypothalamus As cell type proportions are a major driver of DNA methylation patterns, we first removed peaks that were located within genes specifically expressed in neurons, astrocytes, or oligodendrocytes, resulting in a dataset of 48 samples and 451,112 peaks. We assessed the cell-type associated peaks independently by linear modeling (18,050 peaks), identifying few differences between prenatal groups, which suggested that few cell-type associated differences were present in the dataset (Supplementary figure 3.5). To assess persistent alterations to DNA methylation patterns caused by PAE, we performed linear modeling on the hypothalamic samples across all ages with a model that also accounted for differences across breeding cohorts. Using contrast analyses to assess PAE-specific alterations, we successfully identified 118 PAE-specific DMRs at an FDR <0.05 that persisted across all four developmental ages and showed consistently different DNA methylation levels between PAE animals and controls (Figure 3.2; Supplementary table 3.2).  Of these, 47 were up-methylated and 75 were down-methylated in PAE animals versus control groups, and their sizes ranged from 316 to 1027bp (median = 494.5bp). Importantly, meDIP-seq provides relative levels of DNA methylation based on enrichment scores, and thus, the magnitude of change (i.e. % methylation) was not assessed using this method.    79  Figure 3.2 PAE-specific DMRs across pre-weaning development of the hypothalamus A) Contrast analysis revealed 118 PAE-specific differentially methylated regions (DMR), which were significantly different in PAE versus C animals and PAE versus PF animals, but not significantly different between PF versus C. B) The DMRs showed consistent difference between PAE animals and controls across ages. Each row represents a different DMR, while each column shows the mean for all animals within that group/age. Reads per kilobase per million (RPKM) data were scaled and centered to produce a Z-score for each DMR, where those in blue showed less DNA methylation enrichment and those in red showed more enrichment.   Overall, 34 DMRs were located in genes, particularly within those involved in dopamine signaling (Drd4), the immune response (Ifih1, Ccrl2, Il20ra), and blood-brain barrier function (Plvap).  Of note, two overlapping genes, Golga4 and Ctdspl, contained two separate DMRs, and were the only genes with multiple DMRs. Although the entire DMRs set did not show any significant differences in genomic location enrichment compared to the background of the dataset, the up-methylated DMRs displayed significantly less enrichment in CGI and exons   80 (p<0.05), as no up-methylated DMRs were located in these regions (Figure 3.3). Furthermore, the majority of DMRs were located in intergenic regions, and while these were not significantly enriched compared to the entire dataset, these results suggested that intergenic regions may be more responsive to the influence of PAE on the epigenome, and may contain important regulatory regions that are not yet annotated in the rat genome.   Figure 3.3 Enrichment patterns of the developmental DMRs A) Genomic feature enrichment profile of all, up-methylated, and down-methylated DMRs. The probe counts for each feature (blue) were compared to the results from permutation analyses of 118 random regions (orange), which were used to compute the p-value. The majority of DMRs were located in intergenic regions or introns. Up-methylated regions in PAE animals did not contain any CpG islands (CGI) or exons, which is lower than expected by chance (p<0.05). B) Overrepresentation analysis of transcription factor binding sites in the DMRs. Only Bhlhe40 showed higher enrichment in the PAE-specific DMRs (blue) than by random chance (orange) (p<0.05), although Srebf1 and Mlx trended towards significance (p<0.1). *p<0.05, #p<0.1.  3.3.2.2 PAE-specific DMRs contained a greater proportion of bioinformatically predicted Bhlhe40 and Srebf1 TFBS To follow up on the large proportion of intergenic regions in the PAE-specific DMRs, we assessed the enrichment of transcription factor binding sites (TFBS) within these regions using *(*(B A *(#(#(Number of DMRs   81 binding motifs from the mouse genomes. Although the overlap between the rat and mouse genomes is not perfect, the rodent family shares many genomic characteristics and this analysis provides an important first pass analysis of potential regulatory factors within these regions. Following multiple-test correction (FDR<0.05), few TFBS were enriched within these regions compared to background levels. However, the Bhlhe40 binding motif was significantly enriched within the PAE-specific DMRs (p<0.05), while the Srebf1 and Mlx motifs trended towards significance (p<0.10) (Figure 3.3B). These results suggested that certain transcription factors may play a role in the long-term reprogramming of hypothalamic functions by PAE and may act in concert with other factors to sculpt the epigenome and downstream phenotypes.   3.3.2.3 Genes in PAE-specific DMRs were enriched for biological processes associated with hypothalamic functions We performed GO analysis to ascertain the broad functional impact of PAE-induced changes in DNA methylation patterns of the hypothalamus across early development. We identified 20 PAE-specific biological processes (PAEvC and PAEvPF, p<0.05; PFvC, p>0.05; Table 3.2). Of note, the top GO terms were associated with steroid receptor signaling (GO:0042921, GO:0030518, GO:0031958, GO:0030520), a key function of the hypothalamus. Several processes associated with epigenetic regulation (GO:0016577, GO:0006482, GO:0070932) were also enriched in the PAE-specific DMRs, as were processes involved in immune function (GO:0030885, GO:0030886, GO:0002314), and cellular metabolism (GO:0050812).   82 Name ID Number of genes Multi-functionality P-value Multifunctionality p-value PAEvC PAEvPF PFvC PAEvC PAEvPF PFvC Glucocorticoid receptor signaling pathway 0042921 4 0.475 0.00117 0.00432 0.06853 0.0011 0.00456 0.07096 Intracellular steroid hormone receptor signaling pathway 0030518 27 0.681 0.00146 0.00865 0.09115 0.00148 0.00765 0.09009 Corticosteroid receptor signaling pathway 0031958 5 0.442 0.0025 0.01919 0.10193 0.00269 0.01955 0.1019 Regulation of myeloid dendritic cell activation 0030885 2 0.129 0.00816 0.0198 0.14194 0.00843 0.01869 0.14077 Negative regulation of myeloid dendritic cell activation 0030886 2 0.129 0.00816 0.02063 0.1637 0.00843 0.01978 0.163 Histone demethylation 0016577 13 0.397 0.01224 0.02727 0.18051 0.01204 0.02756 0.17928 Protein demethylation 0006482 15 0.365 0.01636 0.0284 0.18496 0.01597 0.02785 0.18571 Protein dealkylation 0008214 15 0.365 0.01636 0.02926 0.2578 0.01597 0.02805 0.26243 Calcium ion export 1901660 3 0.345 0.0166 0.02927 0.32371 0.01739 0.02926 0.32667 Protein sumoylation 0016925 11 0.328 0.01845 0.03449 0.33119 0.01688 0.03389 0.33539 Regulation of protein targeting to membrane 0090313 11 0.631 0.01845 0.03449 0.42205 0.01688 0.03389 0.42409 Intracellular estrogen receptor signaling pathway 0030520 6 0.523 0.01891 0.0354 0.42205 0.01913 0.0354 0.42409 Histone H3 deacetylation 0070932 8 0.419 0.02819 0.04091 0.56227 0.03028 0.04161 0.56796 Relaxation of smooth muscle 0044557 6 0.679 0.03185 0.04117 0.56279 0.03219 0.04121 0.56348 Midbrain-hindbrain boundary development 0030917 3 0.267 0.03285 0.04178 0.72697 0.03382 0.04319 0.72413 GDP-mannose metabolic process 0019673 5 0.252 0.03451 0.04275 0.7401 0.03368 0.04445 0.73884 Protein deacetylation 0006476 20 0.664 0.04114 0.04456 0.76723 0.04009 0.04284 0.76878 Regulation of acyl-CoA biosynthetic process 0050812 4 0.358 0.04459 0.04469 0.89405 0.04633 0.04556 0.89682 Germinal center B cell differentiation 0002314 2 0.073 0.0467 0.04469 0.89405 0.04618 0.04556 0.89682 Negative regulation of nuclear division 0051784 24 0.774 0.04736 0.04683 0.97779 0.04927 0.04737 0.97771 Table 3.2 Biological processes enriched in the developmental profile DMRs   83 3.3.2.4 The Ddr4 DMR was verified by bisulfite pyrosequencing Given that meDIP-seq provides a relative signal of DNA methylation levels, we verified the PAE-specific DMRs using bisulfite pyrosequencing, a quantitative measure of DNA methylation, to ensure that MeDIP-seq could accurately detect alterations in DNA methylation patterns. Importantly, this technique also detects DNA hydroxymethylation, but cannot differentiate between the different cytosine modifications, while MeDIP-seq is specific to DNA methylation. We assessed four different DMRs, based on their potential role in the etiology of PAE-induced deficits.  We first assayed 16 CpGs within the 3’ UTR of the Drd4 DMR (chr1:214,281,174-214,281,640) in the same samples as the meDIP-seq analysis (Figure 3.4). This analysis detected a >5% change in DNA methylation across the DMR on P1 in PAE compared to Control animals. At older ages, several of the CpGs remained significantly different between PAE and controls, with several remaining present on P22. Overall, bisulfite pyrosequencing showed the same direction of change as the meDIP-seq analysis in this DMR.  We also used this method to verify three additional DMRs, located within Ifih1 (chr3:48,561,559-48,561,925), Mycbp (chr5:141,565,784-141,566,172), and Plvap (chr16:19,912,813-19,913,185) (Supplementary figure 3.6). These showed less consistent changes in DNA methylation between the two methods, as some ages appeared to drive DNA methylation patterns more than others and some CpGs showed opposite direction of change between meDIP-seq and pyrosequencing. Nevertheless, small differences were identified between groups, suggesting that meDIP may be sensitive enough to detect small changes in DNA methylation levels. Of note, only a portion of CpGs within each DMR were assessed by bisulfite pyrosequencing due to limitations in read length, suggesting that additional CpGs within   84 the DMR may partially drive some of the differential DNA methylation enrichment identified by meDIP-seq.   Figure 3.4 Bisulfite pyrosequencing verification of the Drd4 DMR 16 CpGs spanning 380 base pairs (bp) of the DMR located in the 3’ UTR of Drd4 were verified by pyrosequencing in the same animals as the meDIP-seq analysis. All CpGs on P1 displayed >5% change in DNA methylation levels between PAE (red) and controls (blue). Of these, several were consistently different across all ages and a number persisted until P22. The total levels of DNA methylation in the DMR also increased with age across all groups.  3.3.3 Tissue-concordant alterations to DNA methylation patterns In parallel to the analysis of developmental DNA methylation in the hypothalamus, we used meDIP-seq to assay DNA methylation in the hypothalamus and WBC of the same P22 females. This analysis aimed to identify tissue-concordant alterations present in both the CNS and peripheral tissue in response to PAE.     85 3.3.3.1 White blood cell proportions were not different across groups As noted, cell type proportions are a major driver of epigenetic variability. However, the volume of blood collected from P22 animals was too small to perform both epigenetic and blood composition analyses on the same animals. As such, we collected samples from P22 animals from an independent but parallel cohort to determine whether PAE alters the proportion of the different WBCs that would be collected using the Ficoll-Paque method. Composition analysis of whole blood indicated the proportions of lymphocytes, neutrophils, monocytes, basophils, eosinophils, and large unclassified cells. Linear modeling revealed no significant differences among prenatal treatment groups, suggesting that PAE does not alter the proportion of the major WBC subtypes (Supplementary figure 3.7). These findings suggested that WBC proportions might not influence differences in DNA methylation patterns between groups in the present dataset.  3.3.3.2 Tissue-concordant alterations to DNA methylation patterns To identify tissue-concordant alterations to DNA methylation patterns associated with PAE, we performed linear modeling on the BvB dataset with a model that also accounted for differences across breeding cohorts: ~Group+tissue+breeding. This method results in the identification of 300 PAE-specific DMRs at an FDR <0.05 that were present in both tissues and showed the same direction of change in PAE animal compared to controls (Figure 3.5; Supplementary table 3.6). Of these, 105 were up-methylated and 195 were down-methylated in PAE animals, and their size ranged from 355 to 2038bp (median = 574). The majority of DMRs also displayed tissue-specific effects in the relative enrichment of DNA methylation, although the magnitude of change was similar between PAE and controls across both tissues (Figure 3.6).   86 Moreover, unsupervised hierarchical clustering of samples using only the PAE-specific DMRs caused the control groups to cluster by tissue, rather than group. By contrast, the PAE were more closely related than those of the PF and C animals, regardless of tissue-type, highlighting that these DMRs likely reflect a mark of alcohol exposure (Figure 3.5).    Figure 3.5 PAE-specific DMRs concordant across the hypothalamus and white blood cells  A) Contrast analysis revealed 300 PAE-specific differentially methylated regions (DMR) between both tissues, which were significantly different in PAE versus C animals and PAE versus PF animals, but not significantly different between PF versus C. B) Heatmap of the DMRs.  Each row represents a different DMR, while each column shows the meDIP-seq data for each animal (n=4, except PF WBC: n =3). Reads per kilobase per million (RPKM) data were scaled and centered to produce a Z-score for each DMR, where those in blue showed less DNA methylation enrichment and those in red showed more enrichment. Samples were grouped using unsupervised hierarchical clustering, causing PAE samples to first cluster together and samples in general to separate by tissue. PAE-specific DMRs showed the same direction of change in both tissues, with some graded effects of tissue type.    87 Again, the majority of DMRs were located in intergenic regions, and were not associated with any gene (Figure 3.6A). However, the DMRs showed decreased enrichment in intergenic regions compared to background levels and more enrichment in intron/exons boundaries, which was driven mainly by the down-methylated regions. These results may reflect the role of DNA methylation in the regulation of splice variants, which could potentially be affected by PAE. Overall, 75 DMRs were located in genes, although the majority of these were once again located in intronic regions. Several DMRs were located in genes involved in immune function (Fgf9, Il18r1) and alcohol metabolism (Adh4). Of note, one DMR spanned 9 different isoforms of the Utg1a family of genes, while Caln1 and Cntnap5c each contained three separate DMRs.    Figure 3.6 Enrichment patterns of the tissue-concordant DMRs A) Genomic feature enrichment profile of all, up-methylated, and down-methylated DMRs. The probe counts for each feature (blue) were compared to the results from permutation analyses of 300 random regions (orange), which were used to compute the p-value. While the majority of DMRs were located in intergenic regions, they showed a lower proportion than expected by random change (p<0.01). By contrast, exon/intron boundaries were overrepresented in the DMRs, particularly within the regions that were down-methylated in PAE animals. B) Overrepresentation analysis of transcription factor binding sites in the DMRs. Several TFBS showed higher enrichment in the tissue-concordant DMRs (blue) than expected by random chance (orange), with GMEB1 showing the highest enrichment at 17% of all DMRs. *p<0.05, **p<0.01.  B A *(**(*(*(*(Number of DMRs   88 3.3.3.3 Several bioinformatically-predicted TFBS were enriched in cross-tissue PAE-specific DMRs  We assessed the enrichment of TFBS within these cross-tissue PAE-specific DMRs to follow up on potential regulatory regions. Following multiple-test correction (FDR<0.05), we identified 16 TFBS enriched within these regions compared to background levels (Figure 3.6B). The most frequent motif belonged to GMEB1, which was found in 16% of all DMRs. Several binding sites for the forkhead box (FOX) family of transcription factors were also enriched in these regions. Of note, the enrichment of Mlx and Srebf1 motifs in the cross-tissue DMRs overlapped with the results from the developmental profile.   3.3.3.4 Genes in cross-tissue PAE-specific DMRs were enriched for various biological processes We performed GO analysis to ascertain the broad functional impact of PAE-induced changes in DNA methylation patterns across the hypothalamus and WBC. We identified 35 PAE-specific biological processes (p<0.05 in PAEvC and PAEvPF, p>0.05 in PFvC; Table 3.3). Of note, the top GO terms were associated with metabolic processes, including aldehyde metabolism (GO:0006081). Several processes were also associated with immune function (GO:0045063, GO:0071351, GO:0032733, GO:0070673, GO:2674), chromatin remodelling (GO:6338, GO:90239), and the stress response (GO:42320).  89 Name ID Number of genes Multi-functionality P-value Multifunctionality p-value PAEvC PAEvPF PFvC PAEvC PAEvPF PFvC Cellular aldehyde metabolic process 6081 29 0.785 0.00089 0.00081 0.0531 0.00094 0.00093 0.05423 T-helper 1 cell differentiation 45063 5 0.483 0.0026 0.00284 0.0531 0.00261 0.00319 0.05423 Amino-acid betaine metabolic process 6577 10 0.484 0.00275 0.00383 0.05739 0.0028 0.0034 0.058 Carnitine metabolic process 9437 7 0.36 0.00284 0.00414 0.06279 0.00321 0.00391 0.06181 Osteoblast fate commitment 2051 2 0.224 0.0044 0.00465 0.09845 0.00413 0.00434 0.09938 Plasma membrane repair 1778 7 0.109 0.00599 0.0051 0.1162 0.00631 0.00474 0.11789 Negative regulation of circadian sleep/wake cycle, REM sleep 42322 2 0.324 0.00788 0.0051 0.14968 0.00829 0.00474 0.15006 Chromatin remodeling 6338 43 0.753 0.01171 0.00597 0.17051 0.01135 0.00569 0.17029 Negative regulation of axon regeneration 48681 3 0.41 0.01139 0.0092 0.17521 0.01204 0.00896 0.17422 Regulation of natural killer cell cytokine production 2727 2 0.293 0.01217 0.01155 0.17521 0.01348 0.01048 0.17422 Positive regulation of natural killer cell cytokine production 2729 2 0.293 0.01217 0.01082 0.22896 0.01348 0.01057 0.2317 Amino-acid betaine biosynthetic process 6578 5 0.219 0.01428 0.01082 0.25627 0.01367 0.01057 0.25577 Glucose 1-phosphate metabolic process 19255 2 0.0827 0.01405 0.0106 0.29844 0.01546 0.01064 0.30042 Cellular response to interleukin-18 71351 2 0.23 0.01587 0.0114 0.31438 0.01748 0.01094 0.317 Protein K63-linked deubiquitination 70536 12 0.126 0.02072 0.01396 0.33657 0.02115 0.01351 0.33588 Carnitine biosynthetic process 45329 3 0.085 0.02242 0.01864 0.34572 0.02301 0.02005 0.34264 Positive regulation of interleukin-10 production 32733 15 0.81 0.02457 0.02429 0.34572 0.02509 0.02327 0.34264 Response to jasmonic acid 9753 3 0.405 0.02597 0.02429 0.37241 0.02675 0.02327 0.36841 Cellular response to jasmonic acid stimulus 71395 3 0.405 0.02597 0.02776 0.38417 0.02675 0.02755 0.38807 Response to interleukin-18 70673 3 0.404 0.02741 0.03021 0.44431 0.02814 0.02905 0.44259 Cofactor catabolic process 51187 13 0.638 0.0291 0.03294 0.44477 0.02836 0.03162 0.44499 Extracellular polysaccharide biosynthetic process 45226 2 0.12 0.02809 0.03374 0.50515 0.02987 0.03253 0.50571   90 Table 3.3 Biological processes enriched in the tissue-concordant DMRs Name ID Number of genes Multi-functionality P-value Multifunctionality p-value PAEvC PAEvPF PFvC PAEvC PAEvPF PFvC Extracellular polysaccharide metabolic process 46379 2 0.12 0.02809 0.03382 0.53397 0.02987 0.03259 0.53274 Acetaldehyde metabolic process 6117 2 0.216 0.03048 0.03547 0.53397 0.03224 0.03553 0.53274 Protein K48-linked deubiquitination 71108 12 0.0357 0.0314 0.03784 0.58202 0.03277 0.03703 0.57955 Cellular response to light stimulus 71482 38 0.821 0.03665 0.03755 0.58809 0.03621 0.03751 0.58519 Podosome assembly 71800 3 0.0518 0.03607 0.03755 0.61294 0.03643 0.03751 0.61412 Micturition 60073 5 0.536 0.04093 0.03865 0.65836 0.03964 0.0376 0.65844 Regulation of histone H4 acetylation 90239 5 0.465 0.04093 0.03969 0.66132 0.03964 0.03965 0.66174 Adenylate cyclase-activating G-protein coupled receptor signaling pathway 7189 26 0.73 0.038 0.04176 0.7519 0.03969 0.04102 0.75163 ER to Golgi ceramide transport 35621 2 0.11 0.03821 0.04171 0.76157 0.03982 0.04149 0.75818 Ceramide transport 35627 2 0.109 0.03821 0.04174 0.81677 0.03982 0.04231 0.81613 Glycolipid transport 46836 2 0.0288 0.03821 0.04318 0.84505 0.03982 0.0426 0.84661 Regulation of circadian sleep/wake cycle, REM sleep 42320 4 0.439 0.04575 0.04318 0.86337 0.04542 0.0426 0.86292 Negative regulation of acute inflammatory response 2674 6 0.674 0.04567 0.04881 0.94037 0.04575 0.04964 0.93962   91 3.3.3.5 Verification of DMRs by bisulfite pyrosequencing We used bisulfite pyrosequencing to compare quantitative levels of DNA methylation between PAE and Control animals in three cross-tissue DMRs. More specifically, we analyzed DNA methylation in the final exon and 3’ UTR of Adh4 (chr2: 243,719,416-243,720,233), the first exon and 5’ UTR of Ctnnbip1 (chr5: 166,485,057-166,485,637), and the first intron of Ffg9 (chr15: 38,377,629-38,378,027) (Supplementary figure 3.8). The main differences in DNA methylation levels were identified between tissues, which sometimes showed different directions of change between PAE and Controls. In particular, a CpG within the Adh4 DMR showed a close to 5% methylation difference in the hypothalamus of PAE animals, but this effect was not present in WBC. Another CpG within the Adh4 locus showed small changes that were consistent between tissues. This pattern was also observed in the Fgf9 locus, which suggested that these may be small, but systemic effects of PAE. By contrast, the Ctnnbip1 locus showed opposite effects between tissues (decreased in the hypothalamus; increased in WBC), suggesting that other factors may come into play. Moreover, as we did not assess quantitative DNA methylation level across the entire DMR due to pyrosequencing limitations, other CpGs may drive the enrichment patterns previously identified by meDIP-seq.   3.4 Discussion Alcohol exposure in utero appears to reprogram physiological and neurobiological systems, increasing the risk of adverse developmental outcomes across the lifespan (Zhang, Sliwowska, & Weinberg 2005; Pei et al. 2011; Mattson, Crocker, & Nguyen 2011). Given the potential role of epigenetic mechanisms in mediating the long-term effects of PAE, the present study aimed to extend previous work on the influence of in utero alcohol exposure on the   92 epigenome, using an animal model of PAE to assess genome-wide DNA methylation patterns during early postnatal development. We identified 118 differentially methylated regions (DMRs) that were altered in the hypothalamus of PAE versus control animals across the pre-weaning period. In parallel, we found 300 DMRs displaying concordant DNA methylation alterations between the hypothalamus and WBC of PAE animals at weaning. Several differentially methylated genes were functionally related to the PAE-induced deficits, including roles in the immune response, neurobiological function, and mental health, while functional enrichment revealed several PAE-specific biological processes, including those related to immune function, the stress response, and epigenetic regulation. In addition, we identified several transcription factor bindings sites that were enriched in the DMRs, which may potentially reflect broader programming effects of PAE on the epigenome. Overall, these findings suggested that PAE causes broad alterations to epigenomic programs in both the CNS and peripheral tissues, suggesting that alterations to DNA methylation patterns could influence broader neurobiological and physiological systems and potential act as biomarkers of PAE.   Our initial analysis of the DMRs revealed several differentially methylated genes that could potentially be relevant to PAE-induced deficits. In particular, the dopamine receptor D4 (Drd4) gene contained a DMR that persisted across the early developmental period. Given its crucial role in dopaminergic function, as well as interactions among dopaminergic, neuroendocrine, and immune systems, alterations to this gene could reflect broader alterations to signaling in the brain. Interestingly, differential DNA methylation patterns of Drd4 are also present in the buccal epithelial cells of individuals with FASD, suggesting that this may be a robust effect of PAE on the epigenome (Portales-Casamar et al. 2016; Fransquet et al. 2016). In   93 addition to this association with FASD, genetic and epigenetic variation in Drd4 has been linked to attention deficit hyperactivity disorder (ADHD), schizophrenia, bipolar disorder, substance-use disorders, and several other neurobiological disorders (Dadds et al. 2016; Ji et al. 2016; Cheng et al. 2014; Kordi-Tamandani, Sahranavard, & Torkamanzehi 2013; Docherty et al. 2012; Ptáček, Kuželová, & Stefano 2011; Bau et al. 2001; Zhang H. et al. 2013; Faraone, Bonvicini, & Scassellati 2014; Chen et al. 2011). Moreover, Golga4 contained 2 PAE-specific DMRs across hypothalamic development, and is known to be overexpressed in the prefrontal cortex of individuals with bipolar disorder (Iwamoto et al. 2004). As a member of the Golgi secretory pathway, it could also potentially influence the secretion of neuropeptides by cells of the hypothalamus, possibly playing a role in altered function or responsivity following PAE (Wong & Munro 2014). Similarly, Plvap expression increases the breakdown and permeability of the blood-brain barrier (BBB) (Shue et al. 2008). As such, slight alterations to its DNA methylation profile could reflect broader effects on the BBB, which, in turn, could affect downstream neurobiological functions.  The tissue-concordant DMRs also contained several genes previously associated with mental health disorders. In particular, Adh4 was differentially methylated across the hypothalamus and WBC of PAE animals, and has been previously associated with alcohol dependence and substance abuse (Luo et al. 2005). Importantly, it is a key component of alcohol metabolism pathways, and could reflect increased susceptibility to the effects of alcohol during development. Furthermore, Caln1 contained 3 separate DMRs, and as it contains a risk allele for schizophrenia in some human populations, it could also play a role in the etiology of FASD (Li et al. 2015).    94 Of note, two genes displayed differential DNA methylation patterns in both the developmental profile and tissue-concordance analysis, Cntnap5c and Ush2a, which may reflect robust alterations to DNA methylation patterns across both age and tissue types. In humans, genetic variation in Cntnap5 is associated with risk for Alzheimer’s disease and bipolar disorder, while its deletion is associated with autism and dyslexia, suggesting that common pathways may come into play between these disorders and FASD (Schott et al. 2016; Xu et al. 2014; Pagnamenta et al. 2010). By contrast, mutations in Ush2a cause Usher syndrome II, which is associated with hearing deficiencies, deficits also commonly found in individuals with FASD (Church & Gerkin 1988). Finally, several DMRs in both datasets were located in genes associated with immune function and response. In particular, Ifih1 was identified across all ages in the hypothalamus; as a receptor for double stranded RNA that responds to viral infections, it could be associated with vulnerability to neuroimmunological deficits (Rice et al. 2014). Fgf9, a key factor in embryonic and glial cell development, was also differentially methylated in both the hypothalamus and WBC (Thisse & Thisse 2005). Furthermore, this growth factor promotes pro-inflammatory environments through Ccl2 and Ccl7 chemokine secretion, consistent with several DMRs that were located in genes associated with pro-inflammatory cytokine and chemokine signaling (Lindner et al. 2015). These included Il20ra and Ccrl2 in the developmental profile, and Il18r1 in the tissue-concordance analysis, suggesting that PAE can influence inflammatory pathways through epigenetic pathways, and ultimately, alter the cellular response to immune challenges.   We also assessed the functional enrichment of genes located within PAE-specific DMRs, identifying a number of biological processes associated with differential DNA methylation   95 patterns in PAE animals compared to controls. In the DMRs identified across hypothalamic development, a large number of GO processes were associated with functions in steroid receptor signaling. The hypothalamus is central to numerous physiological systems that function through steroid hormones, many of which are dysregulated by PAE. As such, this enrichment pattern suggests that DNA methylation may play a role in the reprogramming of hormonal systems during early development, potentially priming physiological systems to new set-points. In addition, several processes in both the developmental and tissue-concordant DMRs were associated with epigenetic regulation, which may reflect the complex interplay between different layers of the epigenetic machinery. Several studies have identified alterations to histone modifications in the brain following developmental alcohol exposure, further highlighting their potential role in FASD (Goldowitz et al. 2014; Chater-Diehl et al. 2016; Veazey et al. 2015; Subbanna et al. 2014, 2013; Zhang et al. 2015; Lussier, Weinberg, & Kobor 2017; Guo et al. 2011; Govorko et al. 2012; Bekdash, Zhang, & Sarkar 2013). A large number of immune-related biological processes were also identified through this analysis, which further highlights the bidirectional communication between the stress response and immune system. Given the close relationship between these systems, altered responsivity of the hypothalamus to immune challenge could potentially alter the organism’s ability to defend against disease or infection. In addition, the top GO term associated with PAE in the tissue-concordant DMRs was “cellular aldehyde metabolic process”, which may reflect lasting effects of PAE on the organism’s ability to metabolize alcohol’s metabolic byproducts and possibly modulate susceptibility to substance abuse later in life. While no overlaps were identified between the specific biological processes identified in the developmental profile and tissue-concordance analyses, both contained a high proportion of processes with immune, endocrine, or epigenetic functions. These findings suggest   96 that PAE may cause systemic effects on the epigenome across multiple tissue types, which may, in turn, influence downstream neurobiological and physiological processes.  Previous studies have identified subtle effects of PAE on gene expression programs and epigenomic patterns, which is consistent with the effects of other prenatal exposures (Laufer et al. 2013; Chater-Diehl et al. 2016; Zhou, Balaraman, et al. 2011; Ladd-Acosta et al. 2014; Berko et al. 2014; Rakyan et al. 2011; Lussier et al. 2015). Regions containing lower CpG density appear to be more responsive to environmental exposures, highlighting the importance of selecting a method that covers a large portion of the epigenome when analyzing an exposure with rather subtle effects (Irizarry et al. 2009a). Thus, we analyzed genome-wide DNA methylation using MeDIP-seq, which is unbiased towards less variable CpG-rich regions and simultaneously reduces the complexity of the dataset by omitting unmethylated regions. As expected, few DMRs across both analyses were identified in CpG-dense regions, such as promoters and CpG islands, while the majority of DMRs were located in intergenic regions and introns, and several were located in intron/exon boundaries, particularly within the down-methylated tissue-concordant DMRs. Given that DNA methylation plays a role in regulating alternative splice variants, these findings may reflect alterations to the balance of different isoforms within the cell, which could influence downstream cellular profiles and phenotypes (Shukla et al. 2011; Maunakea et al. 2013, 2010). Although isoform balance has not been investigated in the context of PAE, studies have shown that alcohol consumption in general can influence the proportions of different splice variants in the brain, supporting a potential role in early-life exposures as well (MacKay et al. 2011; Farris et al. 2015; Lee, Mayfield, & Harris 2014; Mathew et al. 2016). Interestingly, a larger proportion of down-methylated DMRs were   97 identified in both analyses, which is consistent with several studies showing that PAE decreases bulk DNA methylation levels (Otero et al. 2012; Perkins et al. 2013; Chen, Ozturk, & Zhou 2013; Mukhopadhyay et al. 2013; Nagre et al. 2015; Liyanage et al. 2015). These findings provide important insight into different outcomes in different paradigms of alcohol exposure and suggest that similar upstream mechanisms may impact DNA methylation across models, potentially involving changes in one-carbon metabolism or in the activity of DNA methyltransferases.  The large proportion of DMRs located in intergenic regions suggested that these could contain regulatory regions susceptible to the influence of PAE. Given that the rat genome is poorly annotated for regulatory features, we assessed the enrichment profiles of different transcription factor binding sites in the DMRs, which could be influenced by DNA methylation levels within specific loci. While only the binding site for the BHLHE40 transcription factor was significantly enriched in PAE-specific DMRs across early development, we previously identified this gene as differentially expressed in the adult brain of PAE (Lussier et al. 2015). This gene negatively regulates the circadian rhythm, a key function of the hypothalamus that is dysregulated in individuals with FASD (Nakashima et al. 2008). The BHLHE40 transcription factor could potentially play a role in early programming effects of PAE on neurobiological systems, with persistent expression and downstream effect into later life. By contrast, the tissue-concordant DMRs contained a high proportion of significantly enriched TFBS, including SREBF1, which trended towards significance in the developmental profile DMRs. SREBF1 is associated with key metabolic processes for hormonal signaling, as it plays a role in the regulation of cholesterol production (Osborne 2001). It is also associated with Smith–Magenis   98 syndrome, which is characterized by intellectual disability, disordered sleeping, and behavioral problems (Smith et al. 2002). Furthermore, additional TFBS enriched in the BvB dataset included several members of the forkhead box (FOX) family of genes, FOXC1, FOXK1, and FOXO3. In particular, FOXO3 was identified as a hub gene in the brain PAE animals following an immune challenge, suggesting that it may prime biological systems from early in life (Lussier et al. 2015). Finally, the highest represented TFBS in the BvB dataset belonged to GMEB1, which is involved in signal transduction of the glucocorticoid response (Zeng, Kaul, & Simons 2000). Taken together, these findings suggest that the DMRs identified in both the developmental and tissue-concordance analyses may contain key regulatory regions, and that various transcription factors likely act in concert with DNA methylation to mediate the effects of PAE.  Although meDIP-seq allows for the investigation of more variable regions of the epigenome, it presents a particular caveat when assessing DNA methylation levels, as it provides relative levels of DNA methylation across broad regions of the genome, rather than quantitative and granular data. As such, we undertook to verify our findings from the meDIP-seq analysis through bisulfite pyrosequencing, the gold standard for targeted DNA methylation analyses. A limitation of this approach is that bisulfite pyrosequencing detects both methylated and hydroxymethylated cytosines, and there is no way to distinguish the two when analyzing the results from pyrosequencing, resulting in a mixed signal. By contrast, meDIP-seq specifically enriches DNA methylation, as the antibody is highly specific to 5-methylcytosine (Taiwo et al. 2012). In this context, we found that some of the bisulfite pyrosequencing results did not fully confirm the effects observed by meDIP-seq. However, given that neuronal cells contain a high   99 proportion of DNA hydroxymethylation compared to other cell types, it is possible that the observed differences in methodologies are due to the confound of additional epigenetic patterns not assessed in the meDIP-seq analysis. Indeed, a number of studies have shown that developmental alcohol exposure can alter DNA hydroxymethylation programs in neuronal cells in addition to DNA methylation, suggesting that it may also play a role in the etiology of FASD (Chen, Ozturk, & Zhou 2013; Öztürk et al. 2017). In addition, the lack of confirmation could potentially be due to the small number of animals used in the present study, as well as increased variability in the enrichment profiles obtained from meDIP-seq, given the broader regions assessed. Nevertheless, the Drd4 locus identified in the developmental profile of the hypothalamus displayed consistent DNA methylation alterations in both methods, suggesting that meDIP-seq can capture differences in DNA methylation patterns, regardless of the influence of DNA hydroxymethylation. Additional studies are required to fully validate these findings and assess their relationship to the deficits observed following PAE.  One of the main strengths of animal models derives from their ability to directly compare central and peripheral tissue to ascertain potential correlations between the two, which may identify potential biomarkers reflective of brain function in a tissue that is available for study in human populations. In that regard, however, cell type heterogeneity is a major driver of DNA methylation patterns (Farré et al. 2015). Thus, we attempted to partially correct for cellular heterogeneity between groups by removing regions that were associated with the major cell types in the brain, neurons, astrocytes, and oligodendrocytes (Cahoy et al. 2008). However, additional cellular subtypes, such as glia, are also present in the hypothalamus, and could have influenced the results here without our knowledge. We were also limited by the use of regions located   100 within genes, and thus could not correct for intergenic regions that may be associated with cell type.  By contrast, we measured the proportion of different WBC subtypes in an independent cohort of animals. The fact that we did not identify any significant differences in WBC composition of whole blood among groups suggests that this may not have been a factor in driving the DMRs identified in the tissue-concordant analysis. However, as Ficoll-Paque is a highly technical procedure, differences between WBC extractions could have influenced the proportions of cells analyzed in the present study, and we could not correct for such effects. Additionally, as these subtypes can be further subdivided through more sophisticated methods such as fluorescence-activated cell sorting, there is still a possibility that group differences may exist. In contrast to clinical studies of DNA methylation, no bioinformatic tools exist to predict the proportion of different cell types using epigenomic profiles in rats, and future studies should take this into consideration. Nevertheless, we successfully identified several PAE-specific DMRs that showed the same direction of change between the two tissues, suggesting that these regions may be responsive to ethanol across multiple tissues and may represent more stable biomarkers of PAE.   3.5 Summary and conclusions Our results support a role for DNA methylation in the early-life reprogramming of hypothalamic functions by PAE, and suggest that DNA methylation patterns in WBC could potentially be used as a surrogate for alterations in the central nervous system. We identified persistent PAE-induced alterations to the DNA methylome of the hypothalamus, including several DMRs that could, at least in part, underlie some of the deficits observed in FASD.   101 Although PAE-induced alterations to DNA methylation profiles at any of these development ages may not persist into adulthood, changes early in development could alter the developmental trajectory and induce lasting alterations in brain structure and connectivity, or prime physiological systems to different set-points. Of note, we demonstrate for the first time that PAE-specific DMRs can occur across central and peripheral tissues, which potentially represent systemic effects of PAE on the epigenome, and could serve as an epigenetic biomarker or signature of FASD. Taken together, these findings provide insight into the important role of epigenetic alterations in the short and long-term deficits observed in FASD, and provide a foundation for the development of robust biomarkers of PAE.      102 Chapter 4: DNA methylation signature of human fetal alcohol spectrum disorder  4.1 Background and rationale The prenatal environment has the potential to permanently imprint physiological and behavioural systems during development, leading to both short and long-term health consequences. In particular, prenatal alcohol exposure (PAE) can alter the development, function, and regulation of numerous neural and physiological systems, resulting in a variety of deficits falling under the umbrella of Fetal Alcohol Spectrum Disorder (FASD) (Mattson, Crocker, & Nguyen 2011). Over the lifetime, the effects of prenatal alcohol exposure are manifested through cognitive and behavioural deficits, persistent alterations to stress responsivity and immune function, and increased vulnerability to mental health disorders and other comorbidities in individuals with FASD (Zhang, Sliwowska, & Weinberg 2005; Pei et al. 2011; Mattson, Crocker, & Nguyen 2011; Popova et al. 2016). However, the degree to which alcohol exposure causes alterations during development varies, depending on factors such as timing and level of exposure, overall maternal health and nutrition, and genetic background (Pollard 2007). As such, only a small proportion of affected children present with the phenotype of Fetal Alcohol Syndrome (FAS), which is distinguished by growth deficits and facial dysmorphisms in addition to central nervous system dysfunction (Jones & Smith 1973; Astley & Clarren 2000). Nevertheless, the vast majority of children with FASD display physiological and neurobehavioral impairments lasting into adulthood, suggesting persistent programming effects of PAE across the spectrum of FASD (Jacobson et al. 2011).    103 While the etiology of the FASD currently remains unclear, epigenetics is emerging as an attractive candidate for the biological embedding of prenatal and early life experiences in general, and thus is a promising avenue for the study of FASD (Feil & Fraga 2012). Epigenetics refers to modifications of DNA and its packaging that alter the accessibility of DNA, to potentially regulate gene expression and cellular function without changes to the underlying genomic sequences (Bird 2007).  The most studied epigenetic modification in human populations is DNA methylation, which refers to the covalent attachment of a methyl group to the 5’ position of cytosine, typically occurring in the context of cytosine-guanine dinucleotide (CpG) sites (Jones & Takai 2001). CpG sites are relatively rare in the human genome, yet do not occur at random; regions containing higher than expected levels of these dinucleotides have been termed ‘CpG islands’ (CGIs) (Illingworth & Bird 2009). The 2kb regions flanking CGIs are known as CGI ‘shores’, while the areas located beyond shores are known as ‘shelves’ (Doi et al. 2009; Irizarry et al. 2009a; Bibikova et al. 2011). Of note, these regions typically are more variable than CGIs themselves, as they have a greater range of DNA methylation across individuals (Irizarry et al. 2009a). DNA methylation is associated with the regulation of gene expression, although its effects on transcription are highly dependent on genomic context. For example, when located within gene promoters, DNA methylation generally represses gene expression, but this relationship is less well defined for CpGs located within gene bodies and intergenic regions (Jones 2012). Furthermore, DNA methylation is closely associated with several key developmental processes, including genomic imprinting, as well as tissue specification and differentiation (Ziller et al. 2013; Smith & Meissner 2013). DNA methylation patterns are also population-specific, as a number of CpG sites are associated with ethnicity (Fraser et al. 2012; Moen et al. 2013; Heyn et al. 2013). There are a number of possible reasons for this association,   104 including shared environments or associations of epigenetic marks with specific genetic variants (Gutierrez-Arcelus et al. 2013; Wagner et al. 2014; Banovich et al. 2014).  Importantly, DNA methylation is malleable in response to environmental factors and these changes may be inherited through cell divisions, potentially persisting throughout the lifetime (Langevin et al. 2011; Hanson et al. 2011; Yuen et al. 2011). For example, prenatal exposure to cigarette smoke is associated with long-term changes in DNA methylation of the AHRR gene, and maternal under-nutrition during pregnancy leads to altered DNA methylation of IGF2 (Joubert et al. 2012; Heijmans et al. 2008). Several studies have also characterized epigenetic changes following prenatal and postnatal ethanol exposure (Haycock 2009; Haycock & Ramsay 2009; Kobor & Weinberg 2011; Ungerer, Knezovich, & Ramsay 2013; Laufer, Diehl, & Singh 2013; Resendiz et al. 2013; Ramsay 2010). Early work in pregnant mice demonstrated that acute ethanol exposure during mid-gestation (gestational days 9 to 11) causes global genomic loss of DNA methylation in the fetus (Garro et al. 1991). However, recent studies of embryonic cultures exposed to ethanol show that rather than a global demethylation of the genome by ethanol, some regions become more methylated and others less methylated (Liu et al. 2009). Moreover, genome-wide studies in adult mice that were exposed to ethanol prenatally have also identified widespread changes in DNA methylation patterns in the entire brain, further suggesting an important role for epigenetics in the etiology of FASD (Laufer et al. 2013). Finally, a recent study characterized the DNA methylation profile in buccal epithelial cells (BECs) from a small cohort of human FASD samples, identifying alterations in the epigenome of children with FASD, particularly within the protocadherin gene clusters (Laufer et al. 2015).  Collectively, these findings support epigenetic mechanisms as potential contributors to the deficits observed following PAE. However, no large-scale investigations of DNA   105 methylation in individuals with FASD have been performed to date. In order to ascertain the effect of PAE on the human epigenome, the present study investigated the DNA methylation patterns of BECs from 110 children with FASD and 96 age- and sex-matched controls, to our knowledge representing the largest investigation on PAE effects on the human epigenome. Statistically significant alterations between FASD cases and controls were successfully identified following ethnic background correction, with a number of differentially methylated sites and regions located in genes previously associated with alcohol exposure (Liu et al. 2009; Laufer et al. 2015). Taken together, these results support a potential role for DNA methylation in the etiology of the neurobiological deficits observed in children with FASD and represent a potential epigenetic signature of FASD.  4.2 Materials and methods 4.2.1 Participants and samples Children with FASD were recruited from multiple FASD diagnostic clinics across Canada and age- and sex-matched typically developing children were recruited in parallel. Saliva samples and buccal epithelial cells (BECs) were collected for genotyping and DNA methylation analysis respectively (Reynolds et al. 2011). Written informed consent was obtained from a parent or legal guardian and assent was obtained from each child before study participation. The majority of clinics used previously described guidelines for the diagnosis of FASD (Chudley et al. 2005). Briefly, samples were collected from 112 FASD and 102 age- and sex-matched control children aged between 5 and 18. Saliva samples were collected using the Oragene DNA kit (DNA Genotek Inc., Ontario, Canada) according to the manufacturer’s instructions. BECs were collected using the Isohelix buccal swabs and Dri-Capsule (Cell Projects Ltd., Kent, UK). To   106 collect buccal cells, the swab was inserted into the participants’ mouth and rubbed firmly against the inside of the left cheek for 1 minute. The swab was then placed into a sterile tube with a Dri-Capsule and the tube sealed. An identical procedure was followed for the right cheek. Participants did not have any dental work performed 48 hours prior to collection, and no food was consumed less than 60 minutes prior to collection to avoid contamination.  4.2.2 DNA methylation 450K assay DNA was extracted from buccal swabs using the Isohelix DNA isolation kit (Cell Projects, Kent, UK). 750ng of genomic DNA was subjected to bisulfite conversion using the Zymo EZ DNA Methylation Kit (Zymo Research, Irvine, California), which converts DNA methylation information into sequence base differences by deaminating unmethylated cytosines to uracil while leaving methylated cytosines unchanged. 160ng of converted DNA was applied to the HumanMethylation450 BeadChip array from Illumina (450K array), which enables the simultaneous quantitative measurements of 485,512 CpG sites across the human genome, following the manufacturer’s instructions. Chips were scanned on an Illumina HiScan, with the 214 samples run in two batches and each containing an equal number of FASD and control samples, randomly distributed across the chips. Two pairs of technical replicates were included and showed a Pearson correlation coefficient r>0.996 in both cases, highlighting the technology’s reproducibility.  4.2.3 DNA methylation data quality control and normalization The raw DNA methylation data were subjected to a set of rigorous quality controls, first of the samples, and then of the probes. Of the 214 initial samples, 8 were removed from the final   107 dataset due to various quality and concordance issues. Of these, five were removed based on poor quality data, which were identified through skewed internal controls and/or >=5 % of probes with a detection p-value > 0.05. One sample was removed due to a chromosomal abnormality identified in the genotyping and DNA methylation data (XXY; Klinefelter syndrome). The genotypes of the samples, based on the 65 SNP probes contained on the 450K array, were compared to the genotypes from the SNP arrays. The genotypes were highly correlated for all samples (Pearson correlation coefficient r > 0.9), except one, which was excluded from further analyses. Finally, as a pair of monozygotic twins was present in the control group, only one of their samples was chosen at random and retained in the analysis to remove any genetic bias. Next, probes were removed from the dataset according to the following criteria: (1) probes on X and Y chromosomes (N = 11648); (2) SNP probes (N = 65); (3) probes with beadcount <3 in 5 % of samples (N = 3029); (4) probes with 1% of samples with a detection p-value > 0.05 (N = 10163); or (5) probes with a polymorphic CpG and non-specific probes as defined by the Price annotation (N = 20869 SNP-CpG and 41937 non-specific probes; (Price et al. 2013)). A final filtering step was performed to set the methylation values to NA for any remaining probe-sample pair where beadcount <3 or detection p-value > 0.05. Data normalization was performed using the Beta-Mixture Quantile Normalization method on the final dataset, composed of 206 samples (110 FASD and 96 control) and 404,030 probes (Teschendorff et al. 2012). All analyses were performed using M-values, which represent the log2 ratio of methylated/unmethylated, where negative values indicate less than 50% methylation and positive values indicate more than 50% methylation (Du et al. 2010). Percent methylation changes (beta-values) were used in graphical representations of the data and indicate the   108 percentage of methylation calculated by methylated/(methylated + unmethylated), ranging from 0 (fully unmethylated) to 1 (fully methylated).   4.2.4 Differential methylation analysis Given that DNA methylation changes are typically small and that unknown sources of variation, including cellular heterogeneity, may influence the data, surrogate variable analysis (SVA) was performed to identify surrogate variables (SVs) representative of unwanted heterogeneity using the SVA package in R (Leek et al. 2012). Using DNA methylation data from all 206 samples, SVA identified 15 SVs not associated with clinical status (FASD vs control), which, as expected, were only partially correlated with known covariates (Supplemental methods & Supplementary figure 4.2). Linear regression analysis was performed on the dataset with the limma package in R, utilizing a model that included clinical status and all identified SVs as covariates (Smyth 2004). Statistically significant differences between groups were required to show a false-discovery rate (FDR) <0.05 following multiple test correction by the Benjamini-Hochberg method (Benjamini & Hochberg 1995). Further evaluation of potential biological significance was assessed by mean percent DNA methylation differences between FASD and controls.   4.2.5 Analysis of effects due to familial and diagnosis status As the cohort included several sets of siblings and cousins, a sensitivity analysis was performed to identify potential family effects in the dataset. However, little effect of familial origin was observed, indicating that the presence of families in the cohort did not significantly impact the study’s results or require statistical correction (Supplemental methods). Furthermore,   109 this cohort also included children with prenatal alcohol exposure (PAE) that were not formally diagnosed with FASD (27 children). As such, additional differential DNA methylation analyses were performed on the two individual subgroups of FASD cases compared to controls (Supplemental methods & Supplementary figure 4.9). However, as these did not reveal any significant differences between diagnosed FASD cases and PAE children, the PAE cases were included in the FASD group for all analyses.  4.2.6 Genotyping  Genomic DNA was extracted from saliva samples following standard procedures. Briefly, 161 DNA samples were genotyped for 2,443,177 markers using the Infinium HumanOmni2.5-Quad v1.0 BeadChip (Illumina Inc., San Diego, CA, USA) and 54 samples were genotyped for 2,379,855 markers using the Infinium HumanOmni2.5-8 v1.0 BeadChip (Illumina Inc., San Diego, CA, USA) according to the manufacturer’s protocol. For both microarrays, 200ng of DNA (4uL at 50ng/uL) was independently amplified, labeled, and hybridized to BeadChips, then scanned with default settings using the Illumina iScan. Analysis and intra-chip normalization of resulting image files was performed using Illumina’s GenomeStudio Genotyping Module software v.2011 with default parameters. Genotype calls were generated using the Illumina-provided genotype cluster definitions files (HumanOmni2.5-4v1_H.egt and HumanOmni2.5-8v1_C.egt generated using HapMap project DNA samples) with a Gencall cutoff of 0.15. Only the 2 368 900 common SNPs were used for analysis. pyGenClean v1.2.2 and PLINK v1.07 were used for quality control and genetic data cleanup process. SNPs with completion rate <98%, uninformative (MAF=0) and failed for Hardy-Weinberg equilibrium   110 exact test (P value <2.9x10-8) were removed. Samples with completion rate <95% were excluded.   4.2.7 Sub-sample definition Multi-dimensional scaling (MDS) was performed on the participants’ genotype data including 83 founder individuals from the Caucasian population (CEU), 186 from the Japanese and Han Chinese population (JPT-CHB), and 88 from the Yoruba population (YRI) (HapMap; (International HapMap 3 Consortium et al. 2010)). All 195 samples that had both genotyping and DNA methylation data were hierarchically clustered based on the first 4 principal components from the MDS analysis. One individual of African descent was excluded because of their unique ethnicity compared to the rest. All other samples clustered in two major groups: Cluster 1 = 136 samples (49 FASD: 87 Control; mainly Caucasian) and Cluster 2 = 58 samples (53 FASD: 5 Control; mainly First Nations) (Supplementary figure 4.3).  A large imbalance in ethnicities was present between groups, with the majority of controls being Caucasian and most FASD cases being of First Nation descent. Thus, cluster 1, the largest major cluster, was selected as a more balanced sub-sample, both in terms of ethnicity and cases vs. controls, for further analysis (See Figure 4.1 for a summary of the bioinformatic analyses).   111  Figure 4.1 Flowchart of bioinformatic analyses Two analyses were performed in parallel to assess differential DNA methylation between FASD cases and controls. The first analysis, using 206 samples (110 FASD and 96 controls), identified 1661 differentially methylated (DM) sites and 3005 differentially methylated regions (DMR). The second, using a more genetically homogenous subgroup composed of 49 FASD cases and 87 controls, identified 5242 DM sites and 289 DMRs. This second analysis used a p-value threshold of 0.01 to obtain a more conservative list of probes not associated with ethnicity. These were used to filter out the sites identified in the first analysis that might have been confounded by differences in ethnic proportions between the two groups, resulting in a final list of 658 DM CpGs and 101 DMRs free of the confounding effects of ethnicity.  Raw$Data$214$samples$/$485577$CpG$sites$$206$samples$/$404030$CpG$sites$$Removal$of$outlier$samples$&$probe$filtering$$Normalized$Data$206$samples$/$404030$CpG$sites$$BMIQ$normalizaFon$Differen2al$methyla2on$1661$CpG$sites$(FDR$<$5%)$3005$DMRs$(FDR$<$5%)$DifferenFal$methylaFon$(DM)$of$FASD$vs$Control:$Linear$regression$combined$with$Surrogate$Variable$Analysis$$Different$DNA$methyla2on$controlled$for$ethnicity$658$CpG$sites$(356$Up$/$302$Down)$101$DMRs$(61$Up$/$40$Down)$Differen2al$methyla2on$5242$CpG$sites$(pYvalue$<$0.01)$289$DMRs$(FDR$<$5%)$206$samples$136$samples$Cluster$samples$based$on$genotype$$Select$most$homogenous$cluster$$Keep$Overlap$  112 4.2.8 Ethnic group adjustment  Differential DNA methylation analysis was performed as previously described on the more genetically-homogenous sub-sample defined as “Cluster 1” in the MDS analysis above to identify difference between FASD cases and controls. SVA using this sub-sample identified 11 SVs that were added as covariates in linear modeling as described for the full sample. In addition, the inclusion of principal components from the MDS analysis into the regression model to correct for ethnicity was also explored. However, as ethnicity was confounded with the phenotype of interest, direct correction in the model also removed the signal of interest.  4.2.9 DNA methylation pyrosequencing assay Bisulfite pyrosequencing assays were designed with PyroMark Assay Design 2.0 (Qiagen; Supplementary table 4.3). The regions of interest were amplified by PCR using the HotstarTaq DNA polymerase kit (Qiagen) as follows: 15 minutes at 95°C, 45 cycles of 95°C for 30s, 58°C for 30s, and 72°C for 30s, and a 5 minute 72°C final extension step. For pyrosequencing, single-stranded DNA was prepared from the PCR product with the Pyromark™ Vacuum Prep Workstation (Qiagen) and the sequencing was performed using sequencing primers on a Pyromark™ Q96 MD pyrosequencer (Qiagen). The quantitative levels of methylation for each CpG dinucleotide were calculated with Pyro Q-CpG software (Qiagen).  4.2.10 Brain concordance analysis Human brain and blood DNA methylation data from a previously published cohort was used to assess concordance, which was calculated as the Spearman correlation coefficient of DNA methylation at all CpGs between healthy human blood and brain (Farré et al. 2015).   113 Human brain microarray data were obtained from the Allen Brain Atlas (http://human.brain-map.org/static/download; August 1st 2015), which contains normalized expression values for 58,692 probes and 896 brain regions from 6 individuals. Probes were ranked based on their average expression level for each brain region separately and the mean was calculated across all brain regions. All 29,191 genes assayed (which included 389 out of our 404 differentially methylated genes) were sorted based on their highest ranked probe.   4.2.11 CpG island distribution The probes categorization into “North Shelf“, “North Shore”, “Core Island”, “South Shore”, “South Shelf” or “Non-island” was based on the Illumina “RELATION_TO_UCSC_CPG_ISLAND” annotation. The expected counts were calculated with the 404,030 probes remaining after filtering. Statistics were calculated using multinomial goodness of fit chi-square test. As a post-hoc test to evaluate which category is driving the effect, additional chi-square tests were run on each category vs. the sum of all of the other categories.   4.2.12 Functional enrichment analysis The list of imprinted genes was extracted from http://www.geneimprint.com/site/genes-by-species.Homo+sapiens.imprinted-All (June 1st 2014; Supplementary table 4.4), which includes 80 genes with at least one probe among the 404,030 probes remaining after filtering (3035 probes total). The Illumina “UCSC_REFGENE_NAME” annotation was used to map the probes to genes (479 out of 658 DM probes had such annotation and could be mapped). In the event of probes mapping to several genes, the gene with the closest transcription start site (TSS) was selected using the Price annotation (Price et al. 2013). The over-representation analysis   114 (ORA) tool of ErmineJ (version 3.0.2) was used to identify gene function enrichment in the list of up- and down-methylated genes including the Gene Ontology (GO) annotations molecular function, biological process, and cellular component (Lee et al. 2005). The ermineJ ORA tool was set with the following parameters: max gene set size = 1,000; min gene set size = 2; background genes = all genes mapping to the 404,030 probes remaining after filtering.   4.2.13 Co-expression analysis The Gemma tools and database for meta-analysis of functional genomics data were used to perform a co-expression analysis based on existing studies (Zoubarev et al. 2012). The methods used by Gemma have been previously described (Lee et al. 2004). Data sets were obtained from public sources, primarily the Gene Expression Omnibus (Wheeler et al. 2004). For each data set included in the meta-analysis, the Pearson correlation matrix of gene co-expression profiles was computed. Thresholds were applied for statistical significance of correlation, and the resulting sparse co-expression networks were aggregated across data sets. The degree to which a link is replicated across studies is a measure of its reliability; a threshold was set based on a benchmark permutation-based analysis, scaled to the number of data sets aggregated. Using the Gemma on-line tools, a co-expression network was extracted for the 199 up-methylated genes in the master set of microarray experiments for human (282 usable experiments across multiple tissues and experimental conditions) at the stringency recommended by the software, and visualized the results in Cytoscape (Smoot et al. 2011). The resulting network shows the co-expression relationship of the genes in the input list only.   115 4.2.14 Differentially methylated region analysis The identification of differentially methylated regions (DMRs) was performed using previously established guidelines and the DMRcate package in R (Peters et al. 2015; Peters & Buckley, n.d.). Briefly, results from linear modeling with surrogate variables were analyzed using a Gaussian kernel smoother with a bandwith of 1000 base pairs (bp) and scaling factor of 2 to model all CpG sites in the genome in parallel and identify broad regions of differential DNA methylation. P-values were corrected for multiple testing using the BH method, and an FDR cutoff of 0.05 was used to select significant probes between the FASD and control groups. DMRs were then assigned by clustering significant CpGs located within 1000 bp windows that contained two or more CpGs. This analysis was performed on both the full dataset and the more ethnically homogeneous subset of individuals, and the final list of DMRs was obtained through the same process as previously described in the differential methylation analysis. Genomic locations for all DMRs were assigned using the Illumina hg19 annotation.   4.3 Results 4.3.1 The NeuroDevNet FASD epigenetics cohort Participants in the NeuroDevNet Canadian FASD study cohort were recruited from six clinical sites across Canada (Vancouver, BC; Edmonton, AB; Cold Lake, AB; Winnipeg, MB; Ottawa, ON; and Kingston, ON) (Reynolds et al. 2011). More specifically, 110 children with FASD or confirmed PAE and 96 typically developing controls were matched for sex and age, ranging from 5 to 18 years of age, for the analysis of genome-wide DNA methylation patterns (Table 4.1). We note that self-declared ethnicity differed considerably between the FASD and control participants, necessitating stringent statistical corrections, as described below.    116  FASD cases Controls N 110 96 Age 11.55 ± 3.37 11.28 ± 3.38 Sex ! Male ! Female  41% 59%  47% 53% Self-declared ethnicity ! Caucasian ! Other  27% (48%)* 73% (52%)  91% (96%) 9% (4%) *Percentages in brackets include participants with mixed ethnicities including Caucasian Table 4.1 Characteristics of the NeuroDevNet FASD cohort  4.3.2 Children with FASD displayed altered DNA methylation patterns  The DNA methylation profiles of BECs from the complete NeuroDevNet cohort were assessed using the Illumina HumanMethylation450 array, which assays DNA methylation at 485,512 sites across the human genome. Following quality control and normalization to remove probes with bad detection p-values and low bead counts, or those associated with sex chromosomes, SNPs, and polymorphic CpGs, 404,430 sites remained in the final dataset of 206 samples (Price et al. 2013). Although BECs typically represent a relatively homogenous population of cells, they can occasionally be contaminated by white blood cells during collection, thus possibly affecting the results of differential DNA methylation analyses (Jones et al. 2013). To assess whether BEC from the present study had high levels of contamination, principal component analysis of BECs and blood samples obtained from GEO was performed. This analysis did not reveal any considerable blood contamination in our dataset, as evidenced by the distant clustering of samples from both tissue types, though some cell type differences may be present (Supplementary figure 4.1).  Having thus established that cellular heterogeneity was unlikely to confound our results, we next set out to identify alterations in DNA methylation patterns specific to the FASD group.  For this, differential DNA methylation analysis using a   117 two-group design was coupled with surrogate variable analysis (SVA), which corrects for batch effects and any other undesirable variation in the data. This analysis identified 1661 differentially methylated (DM) CpG sites between the FASD group and controls at a false discovery rate (FDR) <0.05, indicating substantial differences in DNA methylation patterns between the two groups. However, self-declared ethnicity in the cohort was strongly confounded with FASD status (Table 4.1). Given that ethnicity has been associated with altered DNA methylation levels, these differences could potentially drive alterations in DNA methylation at these 1661 DM CpG sites (Fraser et al. 2012; Moen et al. 2013; Heyn et al. 2013).  4.3.3 Ethnic background correction identified FASD-specific DNA methylation patterns  To account for ethnicity on a genetic basis, the Illumina HumanOmni2.5 array was used to obtain genotypes at nearly 2.4 million single nucleotide polymorphisms (SNPs) for each child. Participants were clustered by multi-dimensional scaling (MDS) of genotypic data along with publicly accessible data from the HapMap project (Thorisson et al. 2005).  Linear regression of the first four genetic clusters from this analysis with the SVs revealed little correlation with the majority of DNA methylation variation, suggesting that further correction for differences in ethnicity was required to isolate the effect of PAE beyond ethnicity (Supplementary figure 4.2). As such, individuals clustering within the larger and more genetically homogeneous subgroup were selected for further analysis, consisting of 49 FASD cases and 87 controls (Table 4.2; Supplementary figure 4.3 & Supplementary table 4.2). Differential DNA methylation analysis was performed on the more genetically-homogeneous sub-sample to isolate the effects of PAE in the absence of an ethnic confound. In support of less ethnicity-related effects in this subsample, SVA identified fewer SVs compared to the full dataset. Furthermore, the results from DNA   118 methylation analysis in this subgroup displayed only a moderate correlation with those obtained from the full sample (Spearman rank correlation: 0.43), suggesting that ethnicity indeed may have influenced differential DNA methylation patterns in the full cohort, despite our efforts to use SVA to remove the effects of ethnicity. Therefore, the subsample was used to filter out ethnically confounded CpG loci to obtain a subset of DM sites unbiased for ethnicity (Figure 4.1). More specifically, the top 5242 probes (unadjusted p-value <0.01) in the genetically-homogeneous sub-sample were selected as a conservative set of differentially methylated CpG sites between FASD cases and controls that were unaffected by ethnic background. This set was compared to the 1661 DM sites identified in the full sample, and only the probes present in both lists were considered specific effects of FASD, unlikely to be related to effects of ethnicity. Following this strategy, a final list of 658 DM CpG sites significantly altered in FASD cases was obtained at an FDR<0.05  (Supplementary table 4.1), composed of 356 down-methylated and 302 up-methylated sites compared to controls (Figure 4.2AB).  To determine whether this corrective analysis removed some or all effects of ethnicity, differential DNA methylation analysis was performed on FASD cases from the two main ethnic clusters from MDS to tease apart ethnicity and FASD-specific effects between the groups (Supplemental methods). As expected, the ethnicity-corrected CpGs were less, but still partially associated with ethnicity differences in DNA methylation patterns than the uncorrected set of CpGs, as evidenced by the decreased area under the ROC curve (Supplementary figure 4.4). (Supplemental methods). Furthermore, reflecting the economic realities of our study populations, socio-economic status (SES) scores were slightly confounded between groups (p=0.00017; Supplementary figure 4.5), with the FASD group displaying lower overall scores than controls. However, the more ethnically homogeneous subgroup showed less skewing towards low SES in   119 the FASD group (p=0.16; Supplementary figure 4.5), suggesting that the effects of SES might also have been partially accounted for during the correction for ethnic biases between groups.  As such, the ethnicity-corrected set of 658 CpG loci associated with FASD was used in all subsequent analyses. The changes observed in the absolute methylation levels of these DM CpGs were relatively small, consistent with previous human studies of neurological and neurodevelopmental disorders, with percent methylation changes ranging from 0.16% to 13.1% after correction for surrogate variables (Ladd-Acosta et al. 2014). However, 41 DM sites passed an arbitrary threshold for possible biological relevance of greater than 5% difference in DNA methylation levels between groups. Taken together, these results support the hypothesis that FASD is associated with altered DNA methylation patterns, largely free of identified confounding effects due to ethnicity and SES.    FASD cases Controls N 49 87 Age 11.29 ± 3.16 11.29 ± 3.37 Sex ! Male ! Female  43% 57%  41% 59% Self-declared ethnicity ! Caucasian ! Other  51% (76%)* 49% (24%)  93% (97%) 7% (3%) *Percentages in brackets include participants with mixed ethnicities including Caucasian Table 4.2 Characteristics of the more genetically-homogenous sub-sample   120  Figure 4.2 Visualization and verification of differentially methylated probes A. Volcano Plot showing mean methylation differences between FASD and control (x axis) versus log transformed p-values (y axis). 1661 CpG sites with an FDR<0.05 were considered significantly differently methylated between FASD and control, but 1003 of these were ethnically confounded, resulting in the final 658 probes shown in blue. B. Heatmap of top 50 most significant up- (top) and down-methylated (bottom) probes in control (left, grey) versus FASD cases (right, blue). The percent methylation values (ranging from 0 to 1) are adjusted for the covariates from the regression model, then centered, scaled, and trimmed, resulting in a standardized DNA methylation level ranging from -2 to +2 (black to white scale). The mean percent methylation value for each probe (red to blue scale) is the mean methylation value, after adjustment for covariates, for all samples.  C. Verification with pyrosequencing in   121 both FASD (blue) and control (grey) samples. The top panel displays DNA methylation levels measured by the 450k array, the bottom panel, the levels for the same CpG sites measured with pyrosequencing. These CpGs were located in the gene body of SHANK3 (cg10793758), NOS1AP (cg02858267), CACNA1A (cg24800175), and SNED1 (cg19075225), or in the 3’UTR of NOS1AP (cg12486795). Those found in NOS1AP were located in a CpG island, while those in SHANK3 and CACNA1A were located in a north shelf or shore, respectively. The CpG associated with SNED1 was not located near any CpG island. All pyrosequencing data showed significant differences between FASD and controls (p<0.01).  4.3.4 Technical verification of FASD DM loci by bisulfite pyrosequencing To ensure that the results from the differential DNA methylation analysis were not dependent on the method used to measure them, five CpG sites with a difference in percent methylation change greater than 5% in the vicinity of genes with potential biological relevance were selected for verification using bisulfite pyrosequencing on the same samples. Pyrosequencing results confirmed the DNA methylation levels observed on the 450K array, showing similar DNA methylation levels and significant differences between groups (p<0.01) for CpGs located in SHANK3, NOS1AP, CACNA1A, and SNED1 (Figure 4.2C).  Pearson correlations ranged from 0.421 to 0.801 and Bland-Altman plots showed little difference when comparing both methods, suggesting a strong concordance between DNA methylation data from microarray and the different pyrosequencing method (Supplementary figure 4.6). Perhaps more importantly, linear regression analysis of pyrosequencing data confirmed differential DNA methylation between FASD cases and controls in this subset of biologically relevant sites, even in the absence of covariates, as the p-value ranges from 3.7e-04 to 5.5e-03. Collectively, pyrosequencing data verified the findings from the 450K array, suggesting that individuals with FASD had altered DNA methylation patterns compared to typically developing children.    122 4.3.5 Overlap of BEC FASD signatures with brain tissue gene expression and DNA methylation As alterations to DNA methylation patterns in children with FASD were identified in BECs, it is important to note that changes in peripheral tissues do not necessarily reflect alterations in a relevant tissue, such as the brain, even though these two tissues originate from the same germ layer and thus might share some epigenetic concordance (Berko et al. 2014). Therefore, two complimentary approaches were used to obtain an approximation for the relationship of these FASD-associated DM loci to brain biology and possible the etiology of FASD.  First, DM genes were compared to publically available gene expression data from 896 post-mortem brain regions (Allen Institute for Brain Science) to determine whether they were expressed at biologically relevant levels in neural tissue (Hawrylycz et al. 2012). This analysis revealed that 56% of DM genes identified in BECs displayed mRNA expression levels in the brain above the median expression for all genes, with 68% ranked in the top 2/3 of the genes based on mean ranking across ~900 brain regions (Farré et al. 2015). These findings held true whether all DM genes or only the down-/up-methylated genes were considered for analysis. Next, the FASD BEC DNA methylation patterns were compared to DNA methylation patterns from unrelated post-mortem cortical brain specimens previously published by our group (Farré et al. 2015). The overall correlation of mean DNA methylation between BEC and brain samples for all 658 DM CpGs was 0.76 (Supplementary figure 4.7). Taken together, these results indicated that BEC may be a suitable surrogate tissue for brain cells, and that the DM loci presented here could potentially report on biological alterations in neural tissues.     123 4.3.6 FASD DM loci were enriched in regions of high DNA methylation variability Given that genomic location plays an important role in sculpting DNA methylation landscapes and mediating its effects, we ascertained the relative enrichment of FASD DM loci in distinct genomic features. Overall, DM probes had a significantly different distribution than the proportions present on the entire 450K array (Figure 4.3A; down-methylated probes: χ2 = 33.63, p = 2.8e-06; up-methylated probes: χ2 = 13.30, p = 2.1e-02). Compared to all 450K probes, both down- and up-methylated CpGs in FASD cases were significantly under-represented in CpG island cores, which generally show the least amount of variability in DNA methylation levels (down-methylated p= 1.62e-6; up-methylated p= 7.53e-4). By contrast, down-methylated sites were enriched in CpG island shores and shelves (p= 0.04; p= 0.0003), which tend to be more variable than CpG island cores (Irizarry et al. 2009b). Up-methylated sites were over-represented in non-CpG island regions (p=0.009), further supporting a greater effect of PAE on malleable regions of the epigenome. Moreover, the distribution of average methylation levels for DM sites was significantly different than that of all 404,030 sites (Student’s t test; p = 2.5e-09; Supplementary figure 4.8). Further analysis of this phenomenon revealed a significant enrichment for DM CpG sites in the intermediate 20-80% range of methylation levels, while showing a concordant under-representation in the hypo- (<20%) and hyper- (>80%) methylated categories (Figure 4.3B) (Eckhardt et al. 2006). These findings suggested that DM loci in the FASD cases versus controls were mostly located in more variable regions of the epigenome.    124  Figure 4.3 Differentially methylated probes are located in regions of variable and intermediate DNA methylation A. The 658 probes differentially methylated between FASD and control were under-represented in CGI cores (down-methylated p= 1.62e-6; up-methylated p= 7.53e-4), while down-methylated probes were overrepresented in CGI shores/shelves (p= 0.04; p= 0.0003) and up-methylated probes were over-represented in non-CpG island regions (p=0.009). B. The same probes’ average methylation levels are over-represented in the mid-range categories (**p<0.01, ***p<0.0001).  4.3.7 Multiple DM sites were associated with imprinted genes and the protocadherin gene cluster Next, the association of DM loci with different genes was assessed, particularly with regards to whether some of these harbored more than one CpG differentially methylated between FASD and controls. Using genome location annotations from UCSC, the DM sites were mapped to 403 different genes. Of these, 190 were down-methylated, 208 were up-methylated, and 5 displayed inconsistent differences between FASD cases and controls, containing both up- and down-methylated sites, which were likely due to different genomic locations within the genes   125 (Supplementary table 4.2). The Phenocarta resource for gene-disease associations has previously curated a list of susceptibility genes for FASD, identifying 123 potential candidates from both human and animal studies of PAE (Portales-Casamar et al. 2013). However, DNA methylation analysis of the 115 FASD candidate genes assayed on the 450K array did not reveal significant alterations in FASD cases. Nonetheless, twelve genes contained three or more DM loci, including several genes previously involved in studies of alcohol exposure and dependence, but not present in the Phenocarta list, such as SLC6A3 and DRD4 (Table 4.3) (Hillemacher et al. 2009; Zhang H. et al. 2013; Bau et al. 2001; Sánchez-Mora et al. 2011). This short list of DM genes also showed a slight but statistically significant enrichment for imprinted genes (Fisher’s exact test; p=0.02317). The geneimprint website (www.geneimprint.com; June 1st 2014) currently lists 96 human genes as imprinted, 80 of which were assayed by the 404,030 filtered probes on the 450K array. Of these, 5 were differentially methylated in FASD cases versus controls (ATP10A, CPA4, H19, KCNQ1OT1, SLC22A18), with twelve out of fifteen DM CpGs showing lower methylation levels in the FASD group, which resulted in a strong enrichment for imprinted probes in the list of differentially methylated probes (Fisher’s exact test; p = 1.8e-04). In particular, the 6 CpGs located within the SLC22A18 promoter were clustered together, showing a similar pattern between FASD cases and controls, suggesting a robust regional effect of PAE on this gene’s DNA methylation profile (Figure 4.4). Furthermore, fifteen of the 658 DM sites were located within protocadherin genes, including 6 in the PCDHB cluster, 6 in the PCDHGA cluster, 2 in the PCDHA cluster, and 1 in PCDH9. Given the presence of multiple DM CpGs within these genes, these results provide support for imprinted genes and protocadherin clusters as strong candidates for the effects of PAE on the epigenome.     126 Gene # of probes Direction of change Previous reports (PMID) PCDHB gene cluster 6 Up - PCDHGA gene cluster 6 Up - SLC22A18 6 Down 20009564 H19 5 Down 21382472 19519716 19279321 20009564 23580197 HLA-DPB1 5 Up - DES 4 Down - FAM59B (GAREML) 4 Down - SLC38A2 4 Down - CAPN10 3 Up - DRD4 3 Down 20009564* RASSF4 3 Inconsistent - SLC6A3 3 Up 18504048 Table 4.3 Genes containing 3 or more differentially methylated probes   Figure 4.4  Several CpGs associated with SLC22A18 displays down-methylation in FASD cases The covariate-adjusted DNA methylation levels for control (grey) and FASD (blue) samples are shown for SLC22A18AS (top), with the gene structure aligned (bottom). Exons are represented by blocks, and transcriptional   127 direction is indicated by arrows. All CpG sites are noted, those present on the 450K array are black while CpGs not present are grey. The six significantly differentially methylated probes located in the SLC22A18 promoter region are indicated with the horizontal black bar (FDR-adjusted p-value (q) <0.05).  4.3.8 Association of FASD differentially methylated loci with neurodevelopmental processes and disorders  In order to identify broad biological processes associated with altered DNA methylation patterns in FASD children, gene function enrichment analysis was performed on the dataset. As no significant results were obtained from the entire list of DM genes following multiple-test correction, the analysis was performed separately on both the up- and down-methylated gene lists. Given that the up-methylated gene list included several members of the protocadherin beta (PCDHB) and gamma A (PCDHGA) clusters, which are not differentiated by gene function annotations, a single gene from each cluster was conserved for the analysis to avoid any redundancy that may skew the results. As such, only 199 up-methylated genes and 190 down-methylated genes were analyzed for functional annotations using the over-representation analysis (ORA) tool in ermineJ (Lee et al. 2005). While no significant results were obtained using the Gene Ontology (GO) annotation with the list of down-methylated genes, the up-methylated gene list showed enrichment for genes associated with neurodevelopmental processes (Table 4.4), such as neuron parts (20 genes; FDR = 0.051) and projections (19 genes; FDR = 0.082) (Ashburner et al. 2000; Portales-Casamar et al. 2013). Furthermore, using the Phenocarta annotation for associations with diseases, the list of up-methylated genes was enriched for several neurodevelopmental disorders (Table 4.5), including ‘epilepsy syndrome’ (15 genes; FDR = 0.081), ‘autistic disorder’ (12 genes; FDR = 0.092), and ‘anxiety disorder’ (8 genes; FDR   128 = 0.071) (Ashburner et al. 2000; Portales-Casamar et al. 2013). Of note, the up-methylated genes were also marginally enriched for genes associated with substance-related disorder (15 genes; FDR = 0.192). To further examine the regulatory circuitry associated with FASD DM genes, a co-expression analysis of the up-methylated genes across 282 human expression microarray experiments, spanning multiple tissues and experimental conditions, was performed using the Gemma web tools (Zoubarev et al. 2012). Of the up-methylated genes, 86 could be included in the co-expression network (Figure 4.5). The most strongly co-expressed pair was caldesmon 1 (CALD1)-Palladin (PALLD), which are both cytoskeleton-associated proteins (Jin et al. 2009). In addition, a small cluster of the network showed co-expression of several genes (NRXN1, CACNA1A, CDH10, and others) associated with autism and/or epilepsy. Taken together, these findings suggest that altered DNA methylation patterns may potentially relate to the neurobiological deficits of children with FASD.   GO Name GO ID P-value FDR Genes neuron part GO:0097458 1.38E-05 0.051 ATP2B2, CDH13, GABRB1, HEPACAM, KCNAB2, KCND3, KCTD16, NFASC, NMU, NRSN1, NRXN1, P2RX7, PAM, ROBO3, SHANK1, SHANK3, SLC6A1, SLC6A3, SLC8A1, TIAM2, UCN3 vocalization behavior GO:0071625 1.18E-05 0.066 NRXN1, SHANK1, SHANK3 neuron projection GO:0043005 7.31E-06 0.082 CDH13, GABRB1, HEPACAM, KCNAB2, KCND3, NFASC, NMU, NRSN1, NRXN1, P2RX7, PAM, ROBO3, SHANK1, SHANK3, SLC6A1, SLC6A3, SLC8A1, TIAM2, UCN3 Table 4.4 Gene ontology function enrichment in genes up-methylated in FASD       129 Disease Name Disease ID P-value FDR Genes anxiety disorder DOID_2030 1.44E-04 0.071 CRHR2, CYP3A4, GRM8, NOS1AP, P2RX7, PAM, SHANK1, SLC6A3 pervasive developmental disorder DOID_0060040 1.15E-04 0.076 AGAP1, ARID1B, ATP2B2, ATP10A, CDH10, DCUN1D1, DPP6, ESRRB, GABRB1, GRM8, HEPACAM, NOS1AP, NRXN1, PCDHAC2, ROBO3, SDK1, SHANK1, SHANK3, SLC6A3, ST8SIA2 epilepsy syndrome DOID_1826 2.07E-04 0.081 BRD2, CACNA1A, CCR3, CIT, GJD2, GRM1, GRM8, KCNAB2, NRXN1, NTNG2, P2RX7, PAM, SLC6A1, SLC6A3, SLC8A1 autistic disorder DOID_12849 4.70E-05 0.092 AGAP1, ATP10A, CDH10, GABRB1, GRM8, HEPACAM, NOS1AP, NRXN1, ROBO3, SHANK1, SHANK3, ST8SIA2 autism spectrum disorder DOID_0060041 1.01E-04 0.099 AGAP1, ARID1B, ATP2B2, ATP10A, CDH10, DCUN1D1, DPP6, ESRRB, GABRB1, GRM8, HEPACAM, NOS1AP, NRXN1, PCDHAC2, ROBO3, SDK1, SHANK1, SHANK3, SLC6A3, ST8SIA2 substance-related disorder DOID_303 6.85E-04 0.192 ADARB2, ANPEP, CACNA1A, CDH13, CRHR2, FRMD4A, GRM8, KCND3, KISS1R, NMU, NRXN1, SLC6A1, SLC6A3, TIAM2, TRPM4 Table 4.5 Disease-association enrichment in genes up-methylated in FASD       130  Figure 4.5  FASD up-methylated genes coexpression network Nodes represent the up-methylated genes while edges represent their coexpression link. Nodes colored in orange, green, cyan are genes associated with autism spectrum disorder, epilepsy, and anxiety, respectively. The edge width represents the number of experiments in which the coexpression link was identified. The green edges show positive correlations, while the red edges are negative correlations.  4.3.9 Differentially methylated regions were identified between FASD cases and controls To complement the site-specific analysis of differential DNA methylation, which identified several genes with multiple DM CpGs, we next attempted to identify broader patterns of differential DNA methylation using an unbiased approach. Specifically, the identification of region-specific clusters of DM CpGs between children with FASD and controls was performed using DMRcate, an established method that uses a Gaussian kernel smoother to identify broad regions of differential DNA methylation (Peters et al. 2015). In the full dataset, 3005 differentially methylated regions (DMRs) containing two or more CpGs were identified at an   131 FDR<0.05, while in the more homogeneous subset of samples, 289 statistically significant DMRs were identified between groups. Using the same approach to correct for the confounding effects of ethnicity as described in the site-specific analysis, 101 DMRs unbiased by ethnicity were uncovered between individuals with FASD and controls (Supplementary table 4.5).  On average, DMRs spanned 471 nucleotides, with lower and upper limits of 31 and 2,450 base pairs, respectively. DMRs each contained between 2 and 20 CpGs assayed on the 450K array, for a total of 504 unique sites, 75 of which were also identified in the first differential methylation analysis. Of these, 74 overlapped with 95 different genes, and 27 were located in intergenic regions. Of those associated with genes, 25 overlapped with promoter regions (within 1500 bp of the transcriptional start site), 23 with the 5’UTR, 16 with the first exon, 49 with the gene body, and 6 with the 3’UTR, as annotated from the hg19 genome assembly. Moreover, 15 of the top DMRs associated with one or more genes overlapped with those containing multiple DM CpG in the previous analysis, including SLC22A18, SLC38A2, HLA-DBP1, and NOS1AP (Table 4.6; Figure 4.6AB). These showed the same direction of change across the entire DMR, consistent with the individual CpG differential methylation analysis and verification by pyrosequencing, in the case of NOS1AP. Moreover, two DMRs were identified within the protocadherin genes, with 8 CpGs spanning the PCDHGA and PCDHGB clusters and 4 CpGs spanning the promoter of PCDH12, further supporting a potential role for the protocadherin genes in FASD. Importantly, in addition to the genes overlapping with the previous DM analysis, several additional DM genes were identified through this analysis, including UCN3 and ITGAL, key components of the stress and immune response, respectively (Figure 4.6CD). Taken together, these results suggested that the effects of PAE on the DNA methylation went beyond single CpG loci to affect broader chromosomal neighborhoods.   132 Gene symbol(s) DMR location Chr Start position End position # of probes Min    FDR Mean FDR Max beta FC HLA-DPB1 Body 6 33047056 33049505 17 2.59E-50 1.61E-06 0.087 SLC22A18, SLC22A18AS Body, TSS1500, TSS200, 5'UTR 11 2919689 2921176 20 1.21E-29 1.46E-05 -0.049 PPP1R2P1 Body 6 32846924 32847845 18 1.81E-20 9.39E-10 0.026 SLC38A2 TSS1500 12 46767132 46768016 8 1.98E-16 9.78E-09 -0.039 HKR1 TSS1500, TSS200, 1st Exon, 5'UTR 19 37825307 37825679 7 7.51E-16 9.51E-16 0.022 WDR52 5'UTR, 1st Exon, TSS200, TSS1500 3 113160071 113160821 10 1.34E-14 6.02E-13 -0.037 C3orf24 5'UTR, 1st Exon, TSS200, TSS1500 3 10149466 10150487 11 4.41E-13 1.88E-11 0.034 NOS1AP Body, 3'UTR 1 162336877 162337375 5 4.69E-13 8.79E-13 0.039 KCNAB2 5'UTR 1 6093770 6094993 6 9.78E-13 2.86E-07 0.026 F7 TSS1500, TSS200, Body 13 113759771 113760286 6 1.55E-10 1.96E-10 0.029 IFT140, TMEM204 Body 16 1598866 1599150 4 1.81E-10 4.34E-10 -0.036 RGL3 Body 19 11517079 11517436 4 3.06E-10 5.34E-10 0.036 STRA6 5'UTR, 1st Exon, TSS200, TSS1500 15 74494781 74496040 12 4.80E-10 1.06E-04 0.035 TXNRD1, EID3 5'UTR, Body, TSS1500, TSS200, 1st Exon 12 104697193 104697983 11 5.49E-10 3.98E-08 0.024 RNMTL1 Body, 3'UTR 17 695156 695661 3 5.77E-10 3.23E-09 -0.026 C22orf42 Body, TSS200 22 32554848 32555310 5 7.95E-10 7.91E-09 0.022 RADIL Body 7 4869981 4870162 3 2.40E-09 2.48E-09 0.026 ITGAL Body 16 30485383 30485966 6 7.18E-09 5.13E-08 0.022   133 Gene symbol(s) DMR location Chr Start position End position # of probes Min    FDR Mean FDR Max beta FC ZNF710 5'UTR 15 90547692 90548043 3 4.18E-08 5.44E-07 -0.023 PCDHA7, PCDHAC2, PCDHA12, PCDHA6, PCDHA10, PCDHA4, PCDHA11, PCDHA8, PCDHA1, PCDHA2, PCDHA9, PCDHA13, PCDHA5, PCDHAC1, PCDHA3 Body, TSS1500 5 140344290 140344745 4 4.73E-08 1.20E-07 0.019 MAL2 TSS200, 1st Exon, Body 8 120220410 120221797 8 1.26E-07 2.35E-03 -0.022 UCN3 TSS1500, TSS200, 1st Exon, 5'UTR 10 5406543 5407020 8 1.32E-07 3.03E-07 0.016 HKDC1 TSS1500, 5'UTR, 1st Exon 10 70979777 70980067 4 1.37E-07 1.40E-07 0.023 ARHGEF19 Body 1 16533422 16534579 8 1.88E-07 1.11E-04 -0.035 LOC154822 Body 7 158815555 158816392 3 2.36E-07 1.90E-05 -0.043 NDST4 1st Exon, 5'UTR, TSS200, TSS1500 4 116034871 116035232 4 5.96E-07 6.45E-07 0.031 SNED1 Body 2 242009513 242009588 2 6.41E-07 6.48E-07 0.040 PRKDC Body 8 48739161 48739256 2 7.94E-07 8.04E-07 -0.045 CASZ1 5'UTR 1 10847541 10847594 2 2.92E-06 2.92E-06 0.025 HEATR2 Body 7 807596 809109 9 3.11E-06 3.69E-04 0.036 Table 4.6 Top 30 gene-annotated differentially methylated regions associated with FASD  Max fold changes (FC) represented in percent methylation change (beta) in DNA methylation levels of FASD compared to control   134  Figure 4.6  Differentially methylated regions associated with FASD. Percent methylation values adjusted for covariates were plotted across four statistically significant differentially methylated regions (DMRs) between FASD (blue) and controls (grey) identified by DMRcate. A. The HLA-DPB1 DMR spanned 2449 base pairs (bp) of the gene body (red bar) and contained 17 CpGs from the 450K array. B. The NOS1AP DMR contained 5 CpGs over 498 bp, and was located within the body and 3’ UTR (green bar) of the gene. C. The 477 bp UCN3 DMR contained 8 CpGs. One was located within the 5’UTR (dark green dot) and 1st exon (light blue dot), while the remainder were located upstream of the gene’s transcriptional start site (TSS), 1 CpG falling within 1500 bp (black dot) of TSS and 6 located within 200 bp of the TSS (blue bar). D. The ITGAL gene contained 6 unique DMRs over 583 bp of the gene body (red bar).    135 4.4 Discussion This study aimed to assess the effects of PAE on genome-wide DNA methylation patterns and identify an epigenetic signature of FASD, using a large cohort of human subjects. Significant changes to the DNA methylation profiles in BECs of children with FASD compared to age- and sex-matched typically developing controls were identified, with 658 CpGs displaying significantly altered DNA methylation levels, of which 41 had a greater than 5% methylation change. Moreover, 101 DMRs containing two or more sequential DM CpGs were identified throughout the genome, spanning 95 different genes, overlapping with several from the initial differential methylation analysis at single CpG level. The majority of DM genes were highly expressed in postmortem brain samples from the Allen Brain Institute. Moreover, BEC and independent cortical samples showed relatively high concordance of DNA methylation levels. As discussed in more detail below, several lines of evidence converge to support the validity of our data.  First, a number of DM sites and regions were identified within genes and pathways previously associated with PAE.  Second, novel DM sites and regions tended to be involved in pathways implicated in functional deficits of FASD.  Third, broader patterns related to altered neurodevelopmental disorders were identified in sets and networks of genes associated with FASD in our study.  Differential DNA methylation analysis in our case control study comparing children with FASD to children with normal development replicated several associations from previous studies of PAE. One of the most striking similarities is the altered DNA methylation patterns observed in imprinted genes. Several studies have demonstrated the effect of PAE on the H19 imprinted gene in both mice and humans (Stouder, Somm, & Paoloni-Giacobino 2011; Ouko et al. 2009; Haycock & Ramsay 2009). A genome-wide DNA methylation study in mouse embryos exposed   136 to ethanol also identified significant changes within several imprinted genes including both H19 and SLC22A18 (Liu et al. 2009). Results from our study further confirmed these findings, as 5 down-methylated probes in H19, and 6 in SLC22A18 were altered in the FASD cohort, with the latter being identified as a broader DMR as well. Given that imprinting plays a key role in the regulation of normal growth and development, its alteration by alcohol exposure could be a factor in the neurodevelopmental defects observed in children with FASD (Falls et al. 1999). Furthermore, the only other study of genome-wide DNA methylation patterns in individuals with FASD also identified several DM protocadherin genes within the alpha, beta, and gamma clusters, though only one CpG overlapped with the results presented here (Laufer et al. 2015). The differences in specific CpGs within these gene clusters between the two studies might be due to the much larger sample size of our study, as well as our use of multiple test correction to mitigate spurious patterns of differential DNA methylation associated with the FASD group. In addition, the protocadherin clusters are coordinately regulated and are highly susceptible to environmental influences, which may be reflected by their overrepresentation in these studies (Hirayama & Yagi 2017). However, we note that the single CpG site from our study that overlapped with the previous findings (cg21117330) was located in PCDHGA8 and displayed the same direction of change between FASD cases and controls, and thus might represent a reproducible effect of PAE.  In addition to genes previously identified in studies of PAE, DNA methylation changes were also uncovered in a number of additional genes with functional relevance to the deficits observed in FASD. More specifically, analysis of DM probes and regions identified altered DNA methylation patterns within genes related to the immune response, such as HLA-DPB1, a HLA class II histocompatibility antigen, and ITGAL (or CD11A), the integrin alpha L chain. Given   137 that children with FASD often present with numerous deficits in immune function, epigenetic alterations of these genes might reflect functionally relevant underlying biology (Bodnar & Weinberg 2013). A DMR between FASD cases and controls was also identified in UCN3, an antagonist of the CRF type 2 receptor that plays a key role in the stress response. As this gene acts downstream of stress signaling pathways, this alteration might be linked to altered basal levels of corticosterone found in individuals with FASD (Mattson, Crocker, & Nguyen 2011; Ergang et al. 2015). Finally, two members of the dopaminergic system, SLC6A3 and DRD4, each contained three differentially methylated CpGs in FASD cases compared to controls. Both of these genes have also been proposed as modifiers and/or risk factors in alcohol abuse disorders and attention deficit disorder, and thus might potentially play a role in the deficits of attention and executive function in children with FASD (Bau et al. 2001; Sánchez-Mora et al. 2011).  Moving beyond alterations in specific genes related to PAE, broader associations to neurodevelopmental processes and disorders were identified in genes containing differentially methylated CpGs. In particular, the gene co-expression network contained a small sub-network of genes associated with autism and/or epilepsy, and up-methylated genes in FASD cases were enriched for functions related to neurodevelopmental disorders. These results could reflect the pleiotropy of these genes, or perhaps their involvement in developmental functions dysregulated in neurodevelopmental disorders with partially overlapping phenotypes. As many of these genes were also functionally enriched for neuron parts and projections, they could influence processes necessary for typical brain development and partially underlie some deficits observed in children with FASD and other neurodevelopmental disorders.  Comparing epigenetic patterns associated with FASD and autism presented an interesting conundrum. While we identified a small sub-network of genes associated with autism and/or   138 epilepsy in our analysis of the FASD related gene co-expression network, this relationship did not extent to the level of individual CpGs. Comparing the 14 DM genes from BECs recently reported to be associated with autism spectrum disorder, we did not find any overlap with the DM loci identified in our study of FASD children (Berko et al. 2014). The differences between the gene lists may reflect the different origins and phenotypes between the conditions, or that the effects of PAE are more easily identifiable in peripheral tissue than those of autism, or simply false positives and/or false negatives. Regardless, these results imply that at the single CpG level, genes showing differences in DNA methylation between FASD cases and controls are reflective of FASD-specific alterations, rather than broad neurodevelopmental functions. Although it is tempting to speculate that our collective results may be partially related to the functional deficits observed in FASD, it is important to consider that the DNA methylation patterns were derived from BECs. We feel that this concern is partially mitigated by our finding of the majority of DM genes in BECs being consistently expressed across multiple brain regions, and by the DNA methylation patterns in neural tissue displaying high correlation with those in BEC. Moreover, it has been noted by others that BECs might be a good surrogate tissue for human DNA methylation studies, as both buccal and brain cells are derived from the ectoderm (Lowe et al. 2013). Lastly, while our study did not measure DNA methylation in additional tissues, evidence from animal models is emerging to support lasting alterations to both epigenetic and gene expression patterns in neural tissue following PAE (Lussier et al. 2015; Kleiber et al. 2012, 2013; Laufer et al. 2013). Nevertheless, our results must be interpreted with caution in the context of neurodevelopment, as additional studies in postmortem samples from humans are required to rigorously assess the concordance of epigenetic changes associated with FASD between peripheral and central tissues.    139 A further challenge in the interpretation of alterations to DNA methylation patterns in FASD cases versus controls lies in the small effect sizes of environmental exposures on the epigenome. Although the small magnitude of DNA methylation changes observed here are consistent with genome-wide DNA methylation studies in other neurodevelopmental and psychiatric disorders, it is unclear whether such small changes can have a strong effect on cellular functions (Ladd-Acosta et al. 2014; Berko et al. 2014; Rakyan et al. 2011). As a 5% change in DNA methylation levels is typically interpreted as biologically significant, the 41 CpGs displaying >5% differences between FASD cases and controls may reflect more robust PAE-induced alterations to the epigenome. However, slight alterations accumulating in several genes involved in similar processes could combine to have strong effects on biological processes. For instance, as many of the up-methylated genes were co-expressed, small alterations to multiple members of this network could potentially affect the biological functions they regulate.  While our data are very consistent with published work in human epigenome-wide association studies, it is of course possible that the relatively small changes to DNA methylation levels reflect biological biases or even technical noise (Rakyan et al. 2011).  These could originate from a variety of sources, which we attempted to address to the best of our abilities.  For example, while differences in cell type composition can play an important role in driving DNA methylation variation, little to no contamination of the BEC from the present study with white blood cells was identified. These findings suggest that differences in cell type composition may not have affected the observed alterations to DNA methylation patterns in the FASD group, although few blood cell types were covered in our analysis and additional subtypes that were not assessed could potentially have been present in some samples. In addition to differences in cell types, differing postnatal environments between groups might also influence the observed DNA   140 methylation patterns, skewing the results to represent possibly confounding variables other than PAE, such as diet, socio-economic status (SES), and postnatal alcohol exposure. However, the majority of children in the FASD group were living in foster or adoptive homes, rather than the biological family, which hopefully would reduce differences in the rates of alcohol use or food security between groups. By contrast, SES scores were slightly confounded between groups, although this effect was partially mitigated by the focus on the more ethnically homogeneous subgroup, which showed less skewing towards low SES in the FASD cases. Finally, we feel that potential technical issues were reduced through the use of strict quality control and statistical procedures to eliminate unwanted variation in the data. As such, the technical validity of our approach was supported by the verification of 5 DM loci by bisulfite pyrosequencing, the gold standard for targeted DNA methylation analysis.  We note that although most biological and technical issues were addressed by our study design and methods, a particular caveat in the identification of DM loci was manifested by the imbalance in ethnicity across FASD cases and control groups. Given the close relationship between genetic variation and DNA methylation patterns, differences in genetic background between groups may have contributed to the DNA methylation alterations we identified between FASD cases and controls. Other studies have included ethnicity as a covariate during linear modeling to correct for its effects, but no significant DM probes were identified using this approach in our study, as FASD status was confounded with ethnic background (Supplemental methods). Given that self-reports do not always accurately assess ethnicity, SNP genotyping data were used to objectively assign participants to different ethnic groups, based on HapMap samples of known ethnicity. This analysis resulted in the identification of a more homogeneous subgroup of samples, which was used as a comparative control to filter out the influences of   141 ethnicity and related effects, such as SES and cultural confounders, on differential DNA methylation within FASD cases. In turn, this strategy facilitated the removal of ethnically biased probes from the original DM loci, resulting in the successful identification DM CpG sites specific to children with FASD and not confounded for ethnicity. Given the prevalence of ethnically diverse populations in large-scale studies of DNA methylation, this unique approach driven by genetic stratification of subgroups might prove a useful way of dealing with the effects of ethnicity in case control studies beyond the one presented here.   4.4.1 Summary and conclusions Despite the recognition of FAS over 40 years ago, PAE remains the leading cause of developmental disability in the developed world. While several animal studies have investigated the role of epigenetic mechanisms in context of PAE, most human studies have been limited to alcohol consumption and dependence in adults, or a small cohort of children with FASD (Zhang H. et al. 2013; Zhang R. et al. 2013; Philibert et al. 2012; Laufer et al. 2015). As such, this study is the single largest investigation of genome-wide DNA methylation patterns in children with FASD. While one of the greatest challenges with this large cohort was the ethnicity imbalance between the FASD and Control groups, ethnic background correction reduced this confound and allowed the reliable identification of 658 DM CpG sites specific to children with FASD. Although the effect size of changes was small in most cases, 41 sites displayed a greater than 5% change in DNA methylation, which is consistent with previous studies and may reflect the subtle effects of PAE on the epigenome. We also identified 101 DMRs containing two or more DM CpGs, located within 95 different genes and spanning promoter regions, gene bodies, and both 3’ and 5’ UTRs. While these data were collected from BEC, rather than neural tissue, the vast   142 majority of DM genes were highly expressed in the brain, suggesting a potential concordance between peripheral and central tissues. These alterations occurred in several genes previously implicated with PAE and altered neurodevelopment, and displayed functional enrichments for neural process and neurodevelopmental disorders. Although it will be essential to validate these changes in separate cohorts from a different population, these findings provide initial insight into the molecular mechanisms underlying the effects of PAE on children and present a potential role for role for DNA methylation in the etiology of FASD.   143 Chapter 5: DNA methylation as a predictive tool for fetal alcohol spectrum disorder 5.1 Introduction Prenatal alcohol exposure (PAE) can alter the development, function, and regulation of numerous neural and physiological systems, giving rise to lasting cognitive and behavioural deficits, immune dysfunction, motor impairments, and increased vulnerability to mental health problems in adulthood (Zhang, Sliwowska, & Weinberg 2005; Pei et al. 2011; Mattson, Crocker, & Nguyen 2011). In humans, PAE can result in fetal alcohol spectrum disorder (FASD), a leading preventable cause of developmental disability with a North American prevalence currently estimated between 2-5% (May et al. 2009, 2014, 2015). FASD presents through a wide spectrum of phenotypes, ranging from growth deficits and physical abnormalities to cognitive and behavioral deficits.  On the most severe end of the spectrum lies Fetal Alcohol Syndrome (FAS), which is characterized by growth retardation, microcephaly, a distinct set of facial dismorphisms, and central nervous system abnormalities (Jones & Smith 1973; Astley & Clarren 2000). By contrast, Alcohol-Related Birth Defects (ARBD) and Alcohol-Related Neurodevelopmental Disorders (ARND) describe the less severe end of the spectrum, where individuals with confirmed maternal drinking during pregnancy show primarily physical abnormalities or behavioural and/or cognitive abnormalities, respectively (Jacobson et al. 2011).  Although the degree of alcohol’s effects during development varies among individuals, depending on factors such as timing and level of alcohol exposure, overall maternal health and nutrition, and genetic background, individuals across the spectrum show cognitive and behavioral deficits, which can be as serious in those with full FAS as those without any physical   144 features (Pollard 2007). Importantly, FASD has proven difficult to identify at an early age in the absence of overt physical manifestations of the disorder, as ARND requires confirmation of maternal alcohol consumption for diagnosis. As such, many children with FASD are not identified until they reach school age, where they begin to struggle with increased social pressure and cognitive challenges (Senturias & Baldonado 2014). However, early cognitive and behavioral interventions may potentially alleviate some of the deficits caused by PAE and improve the long-term outcomes of individuals with FASD (Paley & O’Connor 2011). As earlier diagnosis is a strong predictor of positive outcomes in individuals with FASD and habilitative care may be have a greater impact during infancy, early screening tools are necessary to help identify at-risk children at a young age and potentially buffer some of the deficits caused by prenatal alcohol exposure (Streissguth et al. 2004; Fox, Levitt, & Nelson III 2010).  Self-report methods are most commonly used to assess PAE and the child’s risk of FASD, these are not always accurate and can lead to underestimation of alcohol consumption behavior during pregnancy (Russell et al. 1996; Jones, Bailey, & Sokol 2013; Burns, Gray, & Smith 2010). Over the past decades, various biomarkers of PAE have been developed to complement self-report questionnaires in the absence of direct alcohol-induced pathologies. More specifically, the latter have focused on the direct or indirect products of ethanol metabolism, which can be measured in biological specimens from both the mother and infant (Concheiro-Guisan & Concheiro 2014). Although these biomarkers are very sensitive to fetal alcohol exposure, they may not be directly related to the biological underpinnings of PAE-induced deficits or the developmental profiles associated with FASD. Furthermore, their use is limited to a short window after birth, which may not be useful in cases where alcohol exposure is   145 not suspected (Cabarcos et al. 2015). As such, objective measures of PAE are needed to aid in the screening and diagnosis of children at risk for FASD.  Importantly, epigenetic marks are now emerging as potential biomarkers or signatures of early-life exposures. Broadly defined, epigenetics refers to modifications of DNA and its regulatory components, including chromatin and non-coding RNA, that potentially modulate gene transcription without changing underlying DNA sequences (Bird 2007; Meaney 2010; Henikoff & Greally 2016). In addition to their role in the regulation of cellular processes, these may also bridge environmental factors and genetic regulation to capture a lasting signature of early environments. In particular, DNA methylation is emerging as a candidate biomarker for environmental exposures and disease. Typically found on the cytosine residues of cytosine-guanine dinucleotides (CpG), this epigenetic mark is both stable over time and dynamic in response to environmental factors (Boyce & Kobor 2015). Several pre- and postnatal environmental influences have been associated with altered DNA methylation patterns, such as maternal nutrition and smoking, supporting their responsiveness to early-life environments and potential use as biomarkers (Joubert et al. 2012; Heijmans et al. 2008). For example, prenatal exposure to cigarette smoke is associated with lasting alterations to DNA methylation patterns, which are now being used as biomarkers of cigarette smoke exposure in infants (Reese et al. 2017). While in its infancy in relation to PAE, this field shows promise for FASD, as the DNA methylome retains a lasting signature of prenatal alcohol exposure in both the central nervous system and peripheral tissues (reviewed in Lussier, Weinberg, & Kobor 2017). Numerous studies have been performed using animal models, and have shown both short term and persistent   146 alterations to DNA methylation patterns in the brain, suggesting that this epigenetic mark may play a role in PAE-induced deficits (Chater-Diehl et al. 2016; Laufer et al. 2013; Liu et al. 2009; Hicks, Middleton, & Miller 2010; Zhou, Chen, & Love 2011; Lussier, Weinberg, & Kobor 2017). By contrast, fewer studies have investigated DNA methylation patterns in children with FASD. More targeted methods identified changes in DNA methylation levels in the promoter region of DRD4 among a large cohort of children exposure to alcohol during breastfeeding in Australia (Fransquet et al. 2016). Others have employed discovery-driven approaches, assessing genome-wide DNA methylation patterns in case-control studies of FASD. The first of these came from a small cohort of children, where the main findings were alterations to DNA methylation patterns in the protocadherin (PCDH) gene clusters (Laufer et al. 2015). Recently, we analyzed DNA methylation profiles in a large cohort of children with FASD, identifying a signature of 658 differentially methylated CpGs (Portales-Casamar et al. 2016). Although few results have been validated across different cohorts, these findings have set the stage for broader applications of DNA methylation in the context of FASD, creating a framework upon which to build future epigenomic studies of PAE.  To validate the findings from our previous DNA methylation signature of FASD, we assessed the genome-wide DNA methylation profiles of buccal epithelial cells (BEC) from an independent cohort of 24 individuals with FASD and 24 sex- and age-matched typically developing controls. Given that our initial study provided a robust framework for genome-wide assessment of DNA methylation patterns in FASD, we used the findings from our initial study as a foundation for the identification of replicable epigenetic alterations following PAE. Notably, nearly 25% of statistically significant associations from the NDN study were validated in this new cohort at a false-discovery rate (FDR) <0.05. In addition to the validation analyses, we also   147 assessed whether DNA methylation profiles could be used to identify individuals with FASD, generating classification algorithms that use DNA methylation levels to predict FASD status with high accuracy. Taken together, these results support a role for DNA methylation in FASD and suggest that it could potentially be used as an early screening tool for at-risk children.  5.2 Materials and methods 5.2.1 The Kids Brain Health Network cohort of children with FASD The present cohort was collected as a replication study by Kids Brain Health Network (KBHN), formerly NeuroDevNet, and is hereby referred to as the KBHN cohort (Reynolds et al. 2011). Written informed consent was obtained from a parent or legal guardian and assent was obtained from each child before study participation. The clinics used previously described guidelines for the diagnosis of FASD (Chudley et al. 2005). Children with FASD and age- and sex-matched typically developing children were recruited from FASD diagnostic clinics in Winnipeg, Manitoba, Canada. Briefly, buccal epithelial cell (BEC) samples were collected for DNA methylation analysis from 25 FASD and 26 age- and sex-matched control children aged between 5 and 18 (Table 1). BECs were collected using the Isohelix buccal swabs and Dri-Capsule (Cell Projects Ltd., Kent, UK). To collect buccal cells, the swab was inserted into the participants’ mouth and rubbed firmly against the inside of the left cheek for 1 minute. The swab was then placed into a sterile tube with a Dri-Capsule and the tube sealed. An identical procedure was followed for the right cheek. Participants did not have any dental work performed 48 hours prior to collection, and no food was consumed less than 60 minutes prior to collection to avoid contamination.    148 5.2.2 DNA methylation 450K assay  DNA was extracted from BECs using the Isohelix DNA isolation kit (Cell Projects, Kent, UK). 750ng of genomic DNA was subjected to bisulfite conversion using the Zymo EZ DNA Methylation Kit (Zymo Research, Irvine, California), which converts DNA methylation information into sequence base differences by deaminating unmethylated cytosines to uracil while leaving methylated cytosines unchanged. 160ng of converted DNA was applied to the HumanMethylation450 BeadChip array from Illumina (450K array), which enables the simultaneous quantitative measurements of 485,512 CpG sites across the human genome, following the manufacturer’s instructions. Chips were scanned on an Illumina HiScan, with the 53 samples run in two batches and each containing a similar number of FASD and control samples, randomly distributed across the chips. Two pairs of technical replicates were included and showed a Pearson correlation coefficient r>0.994 in both cases, highlighting the technology’s reproducibility on our in house-platform. Inter-sample correlations ranged from 0.926-0.99.   5.2.3 DNA methylation data quality control and normalization The raw DNA methylation data was subjected to a rigorous set of quality controls, first of the samples, and then of the probes. Of the 51 initial samples, 3 were removed from the final dataset based on poor quality data, which was identified through skewed internal controls and/or >=5 % of probes with a detection p-value > 0.05 (2 controls and 1 FASD). Next, probes were removed from the dataset according to the following criteria: (1) probes on X and Y chromosomes (n = 11,648); (2) SNP probes (n = 65); (3) probes with beadcount <3 in 10 % of samples (n = 726); (4) probes with 10% of samples with a detection p-value > 0.01 (n = 11,864);   149 or (5) probes with a polymorphic CpG and non-specific probes (N = 19,337 SNP-CpG and 10,484 non-specific probes) (Price et al. 2013). A final filtering step was performed to set the methylation values to NA for any remaining probe-sample pair where beadcount <3 or detection p-value > 0.01. Data normalization was performed using the SWAN method on the final dataset, composed of 48 samples (24 FASD and 24 control) and 431,544 probes (Teschendorff et al. 2012). Finally, batch effects (chip number and chip position) were removed using the ComBat function from the SVA package in R. All analyses were performed using on ComBat-corrected M-values, which represent the log2 ratio of methylated/unmethylated, where negative values indicate less than 50% methylation and positive values indicate more than 50% methylation (Du et al. 2010). Percent methylation changes (beta-values) were used in graphical representations of the data and indicate the percentage of methylation calculated by methylated/(methylated + unmethylated), ranging from 0 (fully unmethylated) to 1 (fully methylated).   5.2.4 Differential methylation analysis and validation of NeuroDevNet (NDN) findings Cell type deconvolution was performed to assess the proportions of CD14, CD34, and buccal epithelial cells in each sample using DNA methylation levels at CpGs highly correlated with these cell types (Smith et al. 2015). Surrogate variable analysis (SVA) was also performed on ComBat-corrected, normalized data using the SVA package in R to identify surrogate variables (SVs) representative of unwanted heterogeneity (Leek et al. 2012). Using DNA methylation data from all 48 samples, SVA identified 6 SVs not associated with clinical status (FASD vs control). As these were partially associated with known covariates, such as cell type proportions and age, the SVs were included in the linear regression analysis to account for their effects. More specifically, linear modeling was performed on the 648 differentially methylated   150 probes identified in the initial NDN study and found in the present dataset using the limma package in R and a model that included clinical status and all identified SVs as covariates(Smyth 2004; Portales-Casamar et al. 2016). Significant differentially methylated probes between groups were identified at a false-discovery rate (FDR) <0.05 following multiple test correction by the Benjamini-Hochberg method and were required to show the same direction of change as the NDN cohort’s findings (Benjamini & Hochberg 1995). Further evaluation of potential biological significance was performed using an arbitrary threshold of >5% mean percent DNA methylation differences between FASD and controls.   5.2.5 DNA methylation pyrosequencing assay Bisulfite pyrosequencing assays were designed with PyroMark Assay Design 2.0 (Qiagen; Supplementary table 5.1). The regions of interest were amplified by PCR using the HotstarTaq DNA polymerase kit (Qiagen) as follows: 15 minutes at 95°C, 45 cycles of 95°C for 30s, 58°C for 30s, and 72°C for 30s, and a 5 minute 72°C final extension step. For pyrosequencing, single-stranded DNA was prepared from the PCR product with the Pyromark™ Vacuum Prep Workstation (Qiagen) and the sequencing was performed using sequencing primers on a Pyromark™ Q96 MD pyrosequencer (Qiagen). The quantitative levels of methylation for each CpG dinucleotide were calculated with Pyro Q-CpG software (Qiagen).  5.2.6 The NDN cohort of children with FASD DNA methylation data from the our previous cohort of children with FASD were obtained from GEO (GSE80261), and normalized as in described in our original publication (Portales-Casamar et al. 2016). This cohort was collected by NeuroDevNet, a Canadian Network   151 of Centers for Excellence, and is hereby referred to as the NDN cohort (Portales-Casamar et al. 2016). Briefly, this dataset was composed of 110 children with FASD or confirmed PAE and 96 age- and sex-matched typically developing controls. The mean age (in years) for individuals with FASD was 11.55 and 11.28 for controls, both ranging from 5-18 years old. A skew in self-declared ethnicity was present between the groups, as the majority of controls identified as Caucasian, while the majority of children in the FASD group identified as First Nations. This skew was addressed in the initial epigenome-wide association study through the use of a more ethnically homogeneous subset of the cohort. DNA methylation data were obtained from buccal epithelial cells using the Illumina 450K array and were normalized using the beta-mixture quantile normalization method.   5.2.7 Cohort of individuals with autism spectrum disorder Normalized DNA methylation data from a publically available dataset of individuals with autism spectrum disorder (ASD) were obtained from GEO (GSE50759). Briefly, this dataset was composed of 48 individuals with ASD and 48 typically developing controls. The samples consisted of 57 males and 39 females, consistent with the skew towards males in ASD. The mean age (8.84) and range (1-28 years old) differed from the NDN and KBHN studies and the genetic ancestry of most individuals was Caucasian (European), though a proportion of the cohort was of Nigerian ancestry. DNA methylation data of these samples were obtained from buccal epithelial cells using the Illumina 450K array.    152 5.2.8 DNA methylation as a predictor of FASD status  A predictive model of FASD status was created using DNA methylation data and the caret package in R. First, a predictive model was created using stochastic gradient boosting on the NDN cohort (110 FASD: 96 control) using both the differentially methylated probes identified in the NDN study (648 probes) and those validated in the KBHN validation cohort (161 probes) (Portales-Casamar et al. 2016). The parameters of the modeling were optimized for area under the receiver operating characteristic (ROC) curve by grid tuning for repeated cross-validation (number of trees 50-1500; 1,5, or 9 interaction depth; 0.1 shrinkage). The optimal model for predicting clinical FASD status using 648 probes was 1500 trees, 5 of interaction depth, and 20 minimum observations per node.  The optimal model for predicting clinical FASD status using 161 probes was 1400 trees, 1 of interaction depth, and 20 minimum observations per node. Next, the KBHN cohort (24 FASD: 24 control) was used as a positive control to verify the predicted sensitivity and specificity of the predictive model. In parallel, 450K data from a cohort of children with autism spectrum disorder (ASD) were tested as a negative control of the model to verify the predicted specificity of the models. Verification of the predictor with these datasets was performed on normalized, uncorrected data to better mimic the potential use of the predictive model by independent groups.  5.3 Results 5.3.1 The KBHN cohort of children with FASD As noted, we analyzed genome-wide DNA methylation patterns from 24 children with FASD or confirmed PAE and 24 typically developing controls, matched for sex and age, ranging from 2 to 18 years of age (Table 5.1). We found that self-declared ethnicity, primary caregiver,   153 and mean age were significantly different between the FASD and control participants (p<0.05). We corrected for the potential effects of age on DNA methylation through the statistical methods outlined below. However, given the heavy confound in self-declared ethnicity and caregiver status, we could not correct for these effects, and relied on the previous correction of ethnic bias in the initial NDN study (see below) (Portales-Casamar et al. 2016).     FASD cases Controls N 24 24 Age      Range 2-18 5-17  Mean 9.1 11.6 Sex       Female 9 13  Male 15 11 Self-declared ethnicity      Caucasian 4 (2)* 22  First Nations 17 (20)* 1  Asian 1 (0)* 1  Not reported 2 0 Caregiver status      Biological parents 7 24  Biological grandparents 3 0  Adopted/legal guardian 8 0   Foster care 6 0 *including individuals with mixed First Nations lineage Table 5.1 Characteristics of the NeuroDevNet II FASD cohort   5.3.2 Children with FASD and typically developing controls showed differential DNA methylation patterns Following quality control and normalization, 431,544 sites of the 485,512 sites remained in the final dataset of 48 samples, which were corrected for batch effects using ComBat. While BECs are mostly homogeneous population of cells, they contain small proportions of CD34- and CD14-positive white blood cells, which can potentially skew DNA methylation analyses. As   154 such, cell type deconvolution was performed to identify any blood contamination in the samples, identifying a trend toward significance in the proportions of different cells types between groups (CD34: p = 0.115; CD14: p = 0.224; BEC: p = 0.068). To account for this factor in addition to other additional potential confounding variables within the dataset, we performed surrogate variable analysis to identify patterns of variation, identifying 6 surrogate variables when protecting the effects of group (FASD vs Control). These were correlated with known sources of variation within the data, including cell type proportions and age (Supplementary figure 5.1).  To identify alterations in DNA methylation patterns specific to the FASD group, we coupled differential DNA methylation analysis using a two-group design with the surrogate variables to correct for undesirable variation in the data. Given that ethnicity-related probes were already accounted for in the NDN study as much as possible, it was concluded that the effects of ethnic background would be lessened by using the final 658 differentially methylated CpGs (Portales-Casamar et al. 2016).  As such, we performed linear modeling on the probes that were differentially methylated in the first study and remained in the dataset after pre-processing (648 CpGs of 658 from NDN). Of these, 161 CpGs displayed differential methylation in the same direction as the initial cohort in the KBHN FASD group compared to the controls at a FDR<0.05  (Figure 5.1A; Supplementary table 5.2). To assess the probability of validating this many probes, random group subsampling and probe subsampling were performed 10,000 times. As none showed more differentially methylated probes than the original replication cohort (maximum = 31 differentially methylated probes), the probability of validating 161/648 probes was < 1e-4 (Supplementary figure 5.2).  Of the 161 validated probes, 82 were up-methylated while 79 were down-methylated in FASD compared to control samples. Several genes contained multiple differentially methylated CpGs across both cohorts, including HLA-DPB1 (5), FAM59B (4),   155 CAPN10 (3), DES (3), SLC6A3 (3), SLC38A2 (3), FAM24A (2), H19 (2), and TGFB1I1 (2) (Table 5.2). Moreover, 53 CpGs showed >5% change in methylation, an arbitrary cutoff often used to gauge potential biological significance. Of note, three genes contained 2 or more DM probes that showed both an FDR<0.05 and change in percent methylation >5%, FAM59B (4 probes), HLA-DPB1 (2 probes), and SLC6A3 (2 probes). In particular, the FAM59B CpGs were located within a CpG island and showed very strong differences in DNA methylation levels between FASD and control groups, with an average 13% methylation change across the array probes in the CpG island (Figure 5.2).  Overall, the percent methylation changes between groups of the 648 analyzed probes were highly correlated between the NDN and KBHN cohorts (r=0.638; figure 5.1B). Across the entire 648 probes analyzed, 462 had the same direction of change, even though the majority did not achieve statistical significance. We also compared the ranking of probes by p-value from linear modeling between the NDN and KBHN cohorts; no significant similarities were identified (p=0.91). Of note, 21 of the significant probes with >5% methylation change in the NDN study were validated in the present analysis (39 of 41 were present in KBHN). This proportion (54%) was much higher than all validated probes (25%), suggesting that these represented potentially more robust effects of alcohol exposure on the epigenome. When using a 5% methylation change as a cutoff, rather than an FDR < 0.05, 62 probes were validated in the KBHN cohort (p<0.1, max FDR = 0.177), of which 25 displayed > 5% change in both cohorts (64% of the NDN probes).       156 Gene # of CpGs Direction of change HLA-DPB1 5 UP FAM59B 4 DOWN DES 3 DOWN SLC6A3 3 UP SLC38A2 3 DOWN CAPN10 3 UP FAM24A 2 UP H19 2 DOWN TGFB1I1 2 DOWN Table 5.2 Genes containing multiple differentially methylated CpGs in FASD     Figure 5.1 Visualization and verification of the differentially methylated probes  A) Heatmap of the 161 validated probes validated in the KBHN cohort at an FDR <0.05 (79 hypermethylated in FASD; 82 hypomethylated in FASD). The percent methylation values (ranging from 0 to 100) were centered, scaled, and trimmed, resulting in a standardized DNA methylation level ranging from −2 to +2 (blue-red scale). B) Scatter plot of the differences in percent methylation between FASD and controls for the 648 differentially probes identified in the NDN cohort. The mean changes between groups were highly correlated between both the NDN and KBHN   157 cohorts (r = 0.638). C) Verification by bisulfite pyrosequencing in FASD (blue) and control (gray) samples verified the difference observed on the 450K array (p=0.0501). The left panel shows the DNA methylation levels from the pyrosequencing assay, while the right panel shows the results from the 450K array. The CpG assayed was located in the CACNA1A gene body (cg24800175).   Figure 5.2 Several differentially methylated CpGs were located in the FAM59B gene body  DNA methylation levels for FASD (blue) and controls (grey) are shown for 10 CpGs within the gene, with the red circles representing the validated hits in KBHN (FDR <0.05). These were located in a CpG island, illustrated by the green bar at the bottom, which showed an average 13% change in DNA methylation levels in individuals with FASD versus controls across all 5 CpGs covered by the 450K array.   5.3.3 Bisulfite pyrosequencing verified the differential DNA methylation of CACNA1A To verify that the differential DNA methylation results did not depend on the method used to measure them, we assessed DNA methylation levels of the cg24800175 probe in CACNA1A. We selected this probes as it was also verified in the initial NDN study, where it similarly showed a >5% change in DNA methylation between individuals with FASD and   158 controls (p=0.0501). Pyrosequencing results confirmed the DNA methylation levels observed on the 450K array, showing similar DNA methylation levels and differences between groups for CpGs located in CACNA1A (Figure 5.1C). The Pearson correlation between these two methods was 0.826 and the Bland–Altman plot showed little difference when comparing the 450K array to pyrosequencing, suggesting good concordance between DNA methylation data from the two methods (Supplementary figure 5.3). Linear regression analysis of pyrosequencing data between FASD cases and controls confirmed differential DNA methylation in this site, even without correcting for covariates (p = 0.04).  5.3.4 DNA methylation patterns classified individuals with FASD versus controls To assess whether DNA methylation data could be used to predict FASD status, we created a predictive algorithm of FASD using machine learning approaches.  First, we selected normalized DNA methylation data from the 206 samples in the NDN cohort (110 FASD: 96 control) in both the 648 initial probes that were also found in the KBHN data. In addition, we also assessed the 161 probes that were validated across both cohorts, though this model may have resulted in over-fitting of the data. Our strategy was to build both predictors (648 probes vs 161 probes) using an initial training cohort (NDN), followed by subsequent testing in the test cohort (KBHN). See Figure 5.3 for an overview of steps used to build the FASD predictor.    159  Figure 5.3 Flowchart of bioinformatic analyses for the DNA methylation predictor of FASD  Briefly, samples from the NDN cohort were used as the training set, and machine learning was performed on either the 648 probes from the initial NDN study, or the 161 probes validated in the present study. The resulting FASD predictor was tested on the KBHN test set, as well as a negative control set composed of individuals with autism spectrum disorder and typically developing controls.  Using a gradient boosting model in the caret package to optimize both sensitivity and specificity (area under the ROC curve), we created two predictive models to assess the probability of FASD based on DNA methylation patterns (Supplementary table 5.3). For the 648 initial probes model, the predicted sensitivity and specificity for the training cohort were 0.922 and 0.978, respectively, for an area under the curve of 0.993 (95% confidence intervals: 0.990-0.995; Figure 5.4A). By contrast, for the 161 probes model, the predicted sensitivity and specificity were 0.887 and 0.892, respectively, forming an area under the curve of 0.955 (95% Training cohort (NDN) 110 FASD: 96 Control 648 initial probes (NDN) OR 161 validated probes (NDN & KBHN) Machine learning FASD predictor Positive control KBHN FASD cohort 24 FASD: 24 Control Testing Negative control ASD cohort 48 ASD: 48 Control   160 confidence intervals: 0.947-0.963; Figure 5.4B). As expected, the 648 model performed much better in the training set, given that the NDN cohort was used to generate these findings.  We next assessed the predictive models using the normalized, batch-corrected DNA methylation data of the KBHN cohort as a test set. Of note, these data were not corrected for any covariates or surrogate variables other than batch correction. In this cohort, the 648 initial probes model performed more poorly, displaying 0.875 sensitivity, 0.542 specificity, and 0.819 area under the ROC curve (Table 5.3; Figure 5.4A). The balanced accuracy of the model in this cohort was 0.708% (95% CI: 0.559-0.830), and the ROC curve was significantly different from the one obtained in the training cohort (p=0.0051). Overall, 11 controls were misclassified as FASD and 3 children with FASD were misclassified as controls, giving a negative predictive value (NPV) of 81.3% and a positive predictive value (PPV) of 65.6%. In contrast to the 648 probes model, the test set confirmed the predictive accuracy of the 161 probes model, though it was potential over-fitting the data. This model displayed 0.917 sensitivity, 0.875 specificity, and 0.944 area under the ROC curve, while the balanced accuracy in this cohort was 0.896% (95% CI: 0.773-0.965), similar to the training dataset (Table 5.3; Figure 5.4B). Overall, 3 controls were misclassified as FASD and 2 children with FASD were misclassified as controls, giving a negative predictive value (NPV) of 88% and a positive predictive value of 91.3%. Moreover, the ROC from the training set and test set were not significantly different (p=0.78), suggesting that the predictor functioned correctly in a similar dataset. Given the discrepancies in ethnic backgrounds between FASD and control groups, the misclassified samples were assessed for differences in self-reported ethnicity, caregiver status, age, or cell-type proportions in the classification. However, no patterns emerged between the correctly and incorrectly classified   161 individuals, suggesting that differences in demographic variables between the groups do not drive their classification.     Figure 5.4 Visualization of the training and test set performance for both DNA methylation predictors  A) The DNA methylation predictor created using the 648 probes identified in NDN showed high accuracy in the training cohort (dark grey; area under the curve = 0.99), but poorer accuracy in the KBHN test set (blue; area under the curve = 0.82; p<0.01). In particular, 11 control samples in the test set were misclassified as FASD, while only 3 individuals with FASD were classified as controls. B) The DNA methylation predictor created using the 161 validated probes also showed high accuracy in the training cohort (dark grey; area under the curve = 0.96), and similar accuracy in the test set (blue; area under the curve = 0.94, p=0.77). Only 3 controls were misclassified as FASD and 2 individuals with FASD were classified as controls.     162  Table 5.3 Summarized results from the classification algorithms  5.3.5 The DNA methylation predictors were not biased by ASD in an independent cohort  BEC samples from an independent ASD cohort served as a negative control to assess the validity of the model in the FASD cohorts. To this end, we used a publically available dataset of 450K array data from the BECs of 48 individuals with autism spectrum disorder (ASD) and 48 typically developing controls from the gene expression omnibus (GSE50759). Using uncorrected, normalized data from this cohort, the two predictors correctly identified the vast majority of individuals in the cohort as non-FASD. The 648 initial probes model misclassified 17 individuals (9 ASD and 8 controls) as FASD, for a specificity of 0.823 (95% CI: 0.732-0.893), slightly lower than the predicted specificity in the training set. By contrast, 12 individuals (7 ASD and 5 controls) were misclassified as individuals with FASD using the 161 probes model, for a specificity of 0.875 (95% CI: 0.792-0.934), which was consistent with predicted values from the model (Table 5.3). The samples did not have any distinguishing features from the     648 probes 161 probes Training set (NDN)       AUC 0.993 0.955   Accuracy 0.943 0.890   Sensitivity 0.922 0.887   Specificity 0.978 0.892 Test set (KBHN)       AUC 0.819 0.944   Accuracy 0.708 0.896   Sensitivity 0.875 0.917   Specificity 0.542 0.875   False positives 11 3   False negatives 3 2   PPV 0.656 0.880   NPV 0.813 0.913 Negative control (ASD)       Accuracy 0.823 0.875   Sensitivity NA NA   Specificity 0.823 0.875   False positives 17 12   163 correctly classified sample, suggesting that the predictive model is not biased for ASD, sex, age, or Nigerian ancestry in independent cohorts.   5.4 Discussion  Epigenetic mechanisms are emerging as potential biomarkers and mediators of environmental exposures, and a growing body of literature suggests that epigenetic factors may be involved in the etiology of FASD. In particular, our recent study using the largest cohort of children with FASD to date identified a signature of 658 differentially methylated CpGs in the BEC of individuals with FASD compared to typically developing controls (Portales-Casamar et al. 2016). Here, we present the first validation of genome-wide DNA methylation data in a small cohort of individuals with FASD, where we successfully validated 161 of the 658 differentially methylated CpGs identified in the initial NDN cohort. Furthermore, we demonstrated that DNA methylation data could be utilized to successfully generate predictive algorithms to classify individuals as FASD or controls with high accuracy. These results indicated that DNA methylation in BEC could potentially be used as a biomarker of PAE to screen children at risk for FASD.  Our present findings represent the first validation of genome-wide DNA methylation alterations in individuals with FASD. Of the 161 validated CpGs at an FDR<0.05, 53 had >5% change in DNA methylation levels, the arbitrary threshold for potential biological relevance. When using a DNA methylation change >5% as a cutoff, rather than a stringent FDR, 62 CpGs were validated, with 25 of those showing this magnitude of change in the NDN cohort as well, suggesting that these regions are more responsive to alcohol’s effects. Importantly, the majority   164 of the CpGs showed the same direction of change between FASD and controls in both cohorts (462/648), and while they did not achieve statistical significance, potentially due to the small size of this cohort, they may reflect consistent alterations of PAE on the epigenome. In addition, we verified the results from the 450K array by bisulfite pyrosequencing, confirming the differential DNA methylation results for a CpG located in CACNA1A and supporting that our findings were not an artifact of array technology. Although the effects of alcohol on the epigenome were relatively subtle, we note that several genes previously associated with PAE or FASD contained multiple differentially CpGs, including FAM59B, H19, HLA-DPB1, and SLC6A3. In particular, DNA methylation alterations in the imprinted gene H19 have been previously associated with PAE in both animal models and clinical cohorts of FASD, and may reflect broader alterations to imprinted genes caused by PAE (Stouder, Somm, & Paoloni-Giacobino 2011; Ouko et al. 2009; Haycock & Ramsay 2009; Portales-Casamar et al. 2016). Moreover, the HLA-DPB1 locus, a member of the major histocompatibility complex proteins, contained several differentially methylated CpGs, which overlapped with a differentially methylated region identified in the NDN study. Given its key function in immune regulation and potential role in rheumatoid arthritis, these alterations could potentially reflect some of the immune changes associated with FASD (Liu et al. 2013). Furthermore, the FAM59B gene contained several CpGs with large changes in DNA methylation levels between individuals with FASD and controls, potentially representing a particularly sensitive locus with regards to PAE. Of note, only one validated CpG was located in a protocadherin gene (PCDHB18), which were considerably enriched in previous genome-wide studies of DNA methylation in individuals with FASD (Laufer et al. 2015; Portales-Casamar et al. 2016). Given that these only showed one overlapping probe, this could indicate higher   165 variability within these gene clusters that may be associated with other variables not present in the current dataset, such as differences in age, BMI, ethnicity, and SES. Of particular interest, we replicated the differential DNA methylation patterns of the two genes involved in dopamine signaling from the NDN cohort, the dopamine transporter SLC6A3 and the dopamine receptor D4 (DRD4). Given the key role of the dopaminergic system in brain development and its interactions with neuroendocrine and immune systems, these alterations could potentially reflect broader changes to signaling pathways in the organism. Of note, the buccal epithelial cells of children exposed to alcohol during prenatal life and breastfeeding also display altered DNA methylation patterns in the promoter region of DRD4 (Fransquet et al. 2016). Furthermore, several disorders previously associated with allelic variation and DNA methylation in this gene show either overlaps or co-morbidities with FASD, including ADHD, bipolar disorder, anxiety disorder, schizophrenia, and substance abuse (Sánchez-Mora et al. 2011; Dadds et al. 2016; Ji et al. 2016; Cheng et al. 2014; Kordi-Tamandani, Sahranavard, & Torkamanzehi 2013; Docherty et al. 2012; Ptáček, Kuželová, & Stefano 2011; Bau et al. 2001; Zhang et al. 2013; Faraone, Bonvicini, & Scassellati 2014; Chen et al. 2011). Given that dopamine signaling plays a key role in brain development and function, it is tempting to interpret these findings in the context of PAE-induced deficits. However, DNA methylation alterations in buccal epithelial cells may not fully reflect alterations in the central nervous system. Nevertheless, it has been suggested that BEC may act as a suitable surrogate tissue in human studies of DNA methylation, as they are also derived from the ectoderm (Lowe et al. 2013). While we did not measure these genes in additional tissues, evidence from animal models suggest that PAE can cause lasting alterations to the epigenome of central nervous system   166 tissues, and as such, these results may represent potentially broader alterations to epigenomic patterns in the brain (Lussier, Weinberg, & Kobor 2017).   Although these findings represent the first validation of genome-wide DNA methylation data in children with FASD, a few particularities of the KBHN cohort limit the interpretability and generalizability of these results. Similar to the initial cohort, the KBHN replication cohort was heavily confounded by ethnicity, as the vast majority of FASD cases were from First Nations communities, while controls were mainly Caucasian. Given that ethnicity influences DNA methylation patterns, differences between groups may have been due to genetic background. Unfortunately, the KBHN cohort was too small to separate the groups into more ethnically homogeneous subsets, a method we had previously used to account for ethnicity-related differences in DNA methylation. As such, we performed linear modeling on the sites that had been previously identified in the NDN study, which were partially filtered for ethnicity-related differences during the analysis of the first cohort. However, some of the top differentially methylated genes could potentially be influenced by ethnicity differences between groups in spite of our best efforts. For instance, three known polymorphisms are located within the FAM59B locus (dbSNP minor allele frequencies: rs774397935: 1.04%; rs4665833: 5.1%; rs181971256: 21.4%). Although none of these are known methylation quantitative trait loci (mQTL), the FAM59B gene body contains several mQTLs in the developing human brain, and genetic variation outside the region could potentially influence DNA methylation levels (Hannon et al. 2015). In addition, nearby genetic variation can also influence DNA methylation patterns in the promoter of DRD4, which may be reflected in this cohort through the skew in ethnicity between groups (Docherty et al. 2012). Although the frequencies of these alleles in First Nation populations have not been assessed, genetic differences between groups could potentially   167 influence DNA methylation levels within this differentially methylated region. Nonetheless, our results suggest that the regulation of these genes might be altered in individuals with FASD, which may potentially occur through direct effects of alcohol on the epigenome or through increased susceptibility to the effects of alcohol due to genetic variation.  In addition to self-declared ethnicity, significant differences in the primary caregiver were present between groups, as all controls lived with their biological families, while the majority of children with FASD were typically in foster care. While the effects of this disparity on the epigenome are unclear, they could influence DNA methylation patterns through a number of factors, including nutrition, early-life adversity, and socio-economic status (SES) (Esposito et al. 2016). However, we also used SVA to account for differences between groups that may have influenced DNA methylation, including cell type proportions, age, and sex. As such, we feel that the potential confounds associated with the cohort design were reduced through our statistical procedures, though future studies with groups balanced for ethnicity and additional variables will be necessary to tease out these differences and further validate our findings.   Finally, we show for the first time that DNA methylation patterns could be used as biomarkers of PAE in clinical populations and can be utilized as predictive variables for FASD. These findings complement and extend previous studies that investigated different molecular and physiological markers to help screen children for potential prenatal alcohol exposure, including alcohol metabolites in mothers and children, circulating miRNA in mothers, and cardiac orienting response in children (Balaraman et al. 2016; Mesa et al. 2017; Goh et al. 2016; McQuire et al. 2016) . In particular, eye tracking measures have been used in a small cohort of children to distinguish children with FASD, ADHD, or typically developing controls with   168 relatively good accuracy (Tseng et al. 2013). In contrast to these studies, the present cohorts were composed of both children diagnosed with FASD and some with confirmed PAE/high-risk of developing FASD. As no follow-up was performed to determine if all children with PAE were ultimately diagnosed with FASD at a later date, the classification models were essentially tuned to screen children at a higher risk for developing FASD with both high sensitivity and specificity. Importantly, our results suggest that DNA methylation predictors can achieve high accuracy in the classification of individuals with FASD versus controls across multiple cohorts. Although the prediction algorithm that used the 161 validated probes showed more consistent results across different cohorts (NDN: 88.9%; KBHN: 89.6%; ASD: 87.5 %) than the 648 probes algorithm (NDN: 94.3%; KBHN: 70.8%; ASD: 82.3%), the use of the validated probes may have caused some over-fitting in the KBHN test set. Nevertheless, it provides an important second validation of the strongest associations with FASD, which likely represent the more robust DNA methylation alterations caused by PAE. Moreover, both predictive algorithms appear to be largely independent of typical confounding factors, such as age, sex, ethnicity, and cell type composition of the samples, as well as ASD. Collectively, these results support the use of DNA methylation as a potential biomarker of PAE and screening tool for FASD.   5.4.1 Summary and conclusions Given the broad spectrum of cognitive, behavioral, and biological deficits caused by PAE, FASD places an important strain on both societal resources and the affected individuals and families. As such, accurate biomarkers are necessary to identify children at risk for FASD at an early age, when interventions are most effective. Our findings provide an important stepping-stone towards epigenetic biomarkers of FASD and set the stage for broader screening tools for   169 neurodevelopmental disorders. Nevertheless, validation of these tools across different cohorts, with varying ages, ethnicities, and environmental exposures will be essential to parse out the strongest associations and create a successful molecular diagnostic tool for FASD.         170 Chapter 6: Conclusion 6.1 Summary and cross-cutting features The work presented in this dissertation highlights the programming effects of PAE on the developing organism, and provides a framework for the use of DNA methylation as a biomarker for FASD. More specifically, I took advantage of an animal model of PAE and two clinical cohorts of children with FASD to identify genome-wide alterations to gene expression programs and DNA methylation patterns.  I profiled genome-wide transcriptomic alterations using gene expression microarrays in the hippocampus and prefrontal cortex of adult female PAE rats under steady-state (basal, saline-injected) and immune challenge (adjuvant-injected) conditions. I identified significant changes in gene expression in PAE compared to controls in response to ethanol exposure alone (saline-injected females), including genes involved in neurodevelopment, apoptosis, and energy metabolism. Moreover, in response to an adjuvant-induced arthritis challenge, PAE animals showed unique gene expression patterns, while failing to exhibit the activation of genes and regulators involved in the immune response observed in control and pair-fed animals. These results support the hypothesis that PAE affects neuroimmune function at the level of gene expression, demonstrating long-term effects of PAE on the CNS response under steady-state conditions and following an inflammatory insult.  Building on these findings of persistent alterations to the brain’s transcriptome, I investigated the early programming effects of PAE on the brain’s epigenome. Specifically, I probed for alterations to DNA methylation programs across early postnatal development in the hypothalamus. As a key regulatory region of the stress response, immune system, and both autonomic and homeostatic regulation, the hypothalamus is a central target for the biological   171 embedding of PAE and I hypothesized that this region would be more responsive to the effects of PAE on the epigenome. I identified numerous differentially methylated regions (DMRs) that showed persistent differences in PAE compared to control animals across pre-weaning development. Importantly, these contained genes enriched for functions in immune regulation, hormonal response, and epigenetic mechanisms, suggesting that epigenetic mechanisms may play a role in PAE-induced alterations to hypothalamic functions. Furthermore, these DMRs also contained a higher proportion of BHLHE40 binding sites, a transcription factor that was also differentially expressed in the prefrontal cortex of PAE animals exposed to adjuvant compared to controls. As BHLHE40 is an important regulator of the circadian rhythm, it could potentially play a role in mediating some of the long-term deficits associated with FASD (Nakashima et al. 2008). Given that central nervous system tissue is not accessible in clinical settings, other than in postmortem specimens, I also assessed the concordance of PAE-induced differential DNA methylation patterns between the hypothalamus and WBC. I identified 300 DMRs that showed the same direction of change in response to PAE in both tissues, which contained genes enriched for functions in immune regulation, the stress response, and chromatin remodeling. These may represent systemic effects of PAE on the developing organism and suggest that WBC could potentially act as a surrogate for CNS alterations in a subset of the epigenome.  Overall, the epigenomic analyses revealed more differential DNA methylation in intergenic regions, suggestive of underlying regulatory regions that could have subtle but broader effects on gene expression profiles and cellular regulation. Furthermore, a number of DMRs were located around intron/exon boundaries, which have been associated with alternative splicing of genes (Shukla et al. 2011; Maunakea et al. 2013, 2010). Although I could not measure the proportions of different isoforms through the gene expression microarray analyses, these   172 findings suggests that PAE could potentially alter the balance of gene isoforms in the developing organism, which could have important ramifications on the developmental trajectories of neurobiological systems.  Finally, across the three different analyses in our rat model of PAE, I identified several alterations to genes involved in immune regulation, suggesting that, even at baseline levels, PAE animals display differential overall cellular responses to immune factors. In addition, the differentially expressed genes in the brain of adult PAE rats were enriched for functions related to lymphocyte differentiation, further highlighting the prevalence of immune-related changes across different studies. Furthermore, all three analyses identified alterations to genes involved in epigenetic regulation, highlighting the complex interplay between the various layers of genetic regulation and stressing the importance of investigating multiple levels of regulatory mechanisms following PAE.   To complement the findings from animal models of PAE, I investigated DNA methylation patterns in a cohort of children and adolescents with FASD. After correcting for the effects of ethnicity, I found 658 significantly differentially methylated CpGs between FASD cases and controls in buccal epithelial cells. Furthermore, over-representation analysis of genes with up-methylated CpGs revealed a significant enrichment for neurodevelopmental processes and diseases, such as anxiety, epilepsy, and autism spectrum disorders. These findings suggest that prenatal alcohol exposure is associated with distinct DNA methylation patterns in children and adolescents. Importantly, I validated these findings in an independent cohort of individuals with FASD, replicating the differential DNA methylation levels at 161 CpGs throughout the   173 genome. These were located in several genes involved in immune function, highlighting the parallels between animal models and clinical cohorts of FASD.  Of particular note, children and adolescents with FASD had altered DNA methylation levels in several genes from the complement system (C1RL), and cytokine/chemokine signaling (CXXC11, IL1R1), as well as alterations to HLA-DPB1, a component of the major histocompatibility complex previously associated with rheumatoid arthritis (Liu et al. 2013; Raychaudhuri et al. 2012). While these findings were identified in a peripheral tissue not directly involved in immune modulation, they may provide insight into changes in global epigenetic patterns associated with altered immune profiles in individuals with FASD. Furthermore, DRD4, a crucial regulator of the dopaminergic system, had altered DNA methylation levels associated with PAE in both the hypothalamus of PAE animals and BEC of individuals with FASD. Given that genetic variation and DNA methylation in this gene has previously been associated with several disorders comorbid with FASD, such as ADHD, depression, schizophrenia, and substance use, this finding suggests that it could potentially play a role in the etiology of several PAE-induced deficits (Ptáček, Kuželová, & Stefano 2011; Bau et al. 2001; Sánchez-Mora et al. 2011; Abdolmaleky et al. 2008; Cheng et al. 2014; Faraone, Bonvicini, & Scassellati 2014; D. Chen et al. 2011; Dadds et al. 2016; Ji et al. 2016; Kordi-Tamandani, Sahranavard, & Torkamanzehi 2013). Taken together, these results highlight the value of animal models in assessing the molecular underpinnings of FASD in the central nervous system, and suggest a potential role for DNA methylation in the etiology of some PAE-induced deficits, including those in self-regulation and immune function. Finally, as these findings raised the possibility of an epigenetic biomarker of FASD, I investigated the potential relevance of DNA methylation in developing a predictive algorithm for   174 PAE. Using these the two clinical cohorts at our disposal, I successfully generated a bioinformatic tool that could classify individuals with FASD versus controls. Importantly, this algorithm could also successfully differentiate between autism spectrum disorder and FASD, suggesting that these epigenetic patterns were likely specific to individuals with PAE.   As a whole, I propose that PAE can leave a lasting impression on the epigenome of central and peripheral tissues, which may potentially influence the deficits observed in individuals with FASD. In turn, these could also be used a biomarkers of PAE to identify individuals with FASD earlier in life, or aid in diagnoses at later ages.   6.2 Limitations These results represent an important step towards understanding the molecular underpinnings of fetal programming by PAE and the deficits associated with FASD. However, their interpretation may be limited by our use of female offspring in the animal model, tissue and cell-type differences in epigenetic patterns, genetic background, and their correlative nature. In addition, while these findings also suggest a potential role for DNA methylation as a biomarker of FASD, our clinical cohorts also contained several individuals with confirmed PAE, but were not yet diagnosed with an FASD.  6.2.1 Sexual dimorphisms In the animal model, I focused our investigation of PAE-induced gene expression and epigenetic alterations on female animals, partially due to their increased vulnerability to autoimmune disorders such as rheumatoid arthritis and the underrepresentation of females in molecular and genome-wide studies of FASD. However, this approach presents an important   175 caveat in the interpretation of our results, as males and females often display sexually dimorphic responses to the effects on alcohol on various neurobiological systems. In particular, males generally show different cognitive and behavioral phenotypes, as well as differential susceptibilities to stressors and mental health disorders compared to their female counterparts (Hellemans et al. 2008; Bale & Epperson 2015; Oldehinkel & Bouma 2011). Given that genetic and epigenetic patterns are highly associated with sex, our findings must also be validated in male animals to fully assess the effects of PAE on the transcriptome and DNA methylome and to understand the sexually dimorphic effects that may exist (Zhang et al. 2011).   6.2.2 Tissue specificity and cellular heterogeneity Given the key role of epigenetic mechanisms in driving cellular identity, cellular heterogeneity is a major drivers of epigenetic variation in large datasets, which may influence the differences identified between groups (Smith & Meissner 2013). As PAE causes neuronal apoptosis, it is possible that slight differences in cell composition are present between prenatal treatment groups (Ikonomidou et al. 2000). As such, DMRs identified in the animal model could potentially be due to underlying differences in cell composition of the hypothalamus or total WBC. Furthermore, as the entire hypothalamus was analyzed, the impact of PAE on its different nuclei, which have widely varying functions, cannot be conclusively assessed (Squire et al. 2008). Nevertheless, differences in genes related to the functions of a particular hypothalamic center could be tentatively assigned and validated in independent studies. Additionally, while cellular composition could also have affected the results obtained from the tissue concordance analysis, I did not identify any difference in the proportions of different WBC subtypes. While these result may suggest few differences between groups, more sensitive methods could   176 potentially subdivide these cell types further to provide a more granular signal of cellular heterogeneity (De Souza et al. 2016).  Furthermore, alterations to the epigenetic patterns of central tissues are not easily measurable in humans, other than in postmortem brain samples. Thus, the vast majority of epigenome-wide association studies are performed in peripheral tissues such as blood and buccal epithelial cells in the hope that they reflect epigenomic variation in the brain. As epigenetic patterns are highly dependent on cell types that may respond differently in the face of the same exposures, these surrogate tissues may not fully portray the true changes driving disease. However, the establishment of common epigenetic profiles between central and peripheral tissues is an ongoing and essential topic of research, and several studies suggest that peripheral tissue could potentially be used as a surrogate for CNS alterations in humans (Walton et al. 2016; Farré et al. 2015; Kaminsky et al. 2012; Davies et al. 2012; Smith et al. 2015; Horvath et al. 2012). These further highlight the power of our animal model study, as it allowed us to make direct correlations between central and peripheral tissue in the same animals and identify concordant alterations to DNA methylation patterns in the hypothalamus and WBC.  6.2.3 Genetic background  Genetic background can also influence DNA methylation patterns throughout the genome, as a large number of CpG sites are associated with genetic variation (Fraser et al. 2012; Moen et al. 2013; Heyn et al. 2013). Of note, our animal model is based on an outbred population of Sprague-Dawley rats, which display a range of genetic diversity. As such, differing genetic backgrounds among prenatal treatment groups could potentially have influenced the DMRs identified in the developmental and tissue-concordant analyses. This issue was further   177 highlighted in the clinical cohorts, where the majority of individuals with FASD were of First Nation descent, while the controls were primarily of Caucasian descent. Although this limitation was at least partially mitigated through the use of statistical methods, they could potentially have influenced the PAE-induced epigenetic alterations identified in these studies. By contrast, I could not account for this potential genetic influence in the animal model, and further studies are required to fully investigate the interaction between genetic variants and epigenomic patterns in the context of PAE.  6.2.4 Correlation versus causation Although the data presented in this thesis lend support to hypothesis that PAE can influence neurobiological systems through epigenetic alterations, they were not designed to examine the exact mechanisms of alcohol’s effects. More specifically, even though gene expression profiles and epigenetic mechanisms are correlated with PAE in cross-sectional clinical cohorts and animal models, it is not yet clear whether their reversal would dampen PAE phenotypes, which would indicate a more causal role. Emerging technologies, such as CRISPR fused to chromatin modifiers, could potentially be used to selectively alter the epigenetic profiles of key genes and provide a more causal link between epigenetic alterations and PAE-induced deficits (Enríquez 2016). Moreover, while epigenetic patterns are associated with gene expression, these relationships are inconsistent across individuals and the functional implications of epigenetic alterations have yet to be fully established (Lam et al. 2012; Gutierrez-Arcelus et al. 2013). In particular, it is not yet clear whether DNA methylation regulates transcription or if it is a result of thereof, making its functional interpretation difficult. As such, prior to making inferences concerning cognitive, behavioral, or physiological outcomes from alcohol-induced   178 epigenetic alterations, a direct line of evidence must first be established between epigenetic patterns, gene expression profiles, and the phenotype in question, either through genetic manipulation or therapeutic interventions in model organisms.  6.2.5 PAE versus FASD biomarkers The present clinical cohorts were composed of both children diagnosed with FASD and some with confirmed PAE/high-risk of developing FASD. Due to constraints of the clinical situation, it was not possible to do follow-up assessments to determine if all children with PAE were ultimately diagnosed with FASD at a later date. As such, the classification models were tuned to screen children at a higher risk for developing FASD, rather than act as a diagnostic tool. Conversely, this method may cast a wider net and help identify children who might benefit from early interventions. Furthermore, it remains unclear whether higher specificity (low false-negative rate) or specificity (low false-positive rate) is preferable when screening for FASD. On the one hand, higher sensitivity would promote early interventions in a greater number of at-risk children, which could mitigate some of their deficits. On the other hand, high specificity would potentially prevent unnecessary interventions with some individuals and reduce the strain on health care resources. As both issues are important for child health and wellbeing, it appears that a balance between the two may be the best compromise, although much higher values and overall accuracy will be necessary before these methods are implemented in the clinic.   6.3 Broader considerations for future epigenome-wide studies of FASD Much headway has been made in characterizing the epigenetic patterns associated with developmental alcohol exposure and their role in fetal programing by PAE. However, a number   179 of key considerations will be crucial for the next wave of genetic and epigenetic studies in FASD. First, most studies of alcohol exposure in animal models focus exclusively on male animals or do not highlight sex-specific differences, an issue found throughout many research fields and recently highlighted by the new funding guidelines from the National Institutes of Health (Clayton & Collins 2014). Since epigenetic patterns are highly associated with sex, this further reduces the generalizability and applicability of findings from animal models of PAE to clinical settings (Zhang et al. 2011). This is particularly relevant to the domain of FASD, as males in general typically display different cognitive and behavioral phenotypes, as well as differential susceptibilities to stressors and mental health disorders compared to their female counterparts (Hellemans et al. 2008; Bale & Epperson 2015; Oldehinkel & Bouma 2011). As such, the paucity of data on females in the FASD research field must be addressed in order to fully assess the role of epigenetics in the etiology of alcohol-induced deficits. Given the wide variety of PAE models, we must also begin to integrate findings from different models of exposure, which vary in terms of dosage (low to high), pattern of exposure (acute or chronic), trimester of exposure, and type of ethanol administration, to identify the most robust epigenetic signatures of PAE.  Additionally, a large portion of whole-genome analyses of genomic and epigenomic patterns have been performed either in cell culture or whole brains, which does not necessarily reflect the downstream functional implications of alcohol-induced alterations. Future studies should begin to assess changes within specific brain regions and primary tissues to further dissect the role of the transcriptome and epigenetics in the various deficits associated with developmental alcohol exposure. Cell type differences must also be taken into account when analyzing these data, as tissue type and cellular heterogeneity are major drivers of epigenetic   180 patterns and may be altered by alcohol exposure. Various strategies can be used to address this issue, including the isolation of single cell types prior to genome-wide analyses, the inclusion of cell type proportions in statistical models, or bioinformatic methods such as cell type deconvolution and surrogate variable analysis. Large-scale network analyses may also provide an alternative method to analyze these types of data, allowing researchers to identify broader patterns of PAE-induced alterations and draw links between the different changes observed (Zoubarev et al. 2012; Zhang & Horvath 2005). In addition, robust statistical methods must be used in the analysis of genome-wide alterations to prevent spurious associations with alcohol exposure. These considerations include the use of multiple-test correction and other methods to correct for discrepancies between groups (age, ethnicity, smoking, etc.), which tend to occur frequently in population studies. Of note, the phenotypes associated with FASD have been rather heterogeneous in human studies. This is perhaps not surprising, given that numerous environmental and genetic influences can modulate the effects of alcohol on the developing organism, including maternal nutrition, dose and timing of alcohol exposure, as well as overall maternal health and genetics (Pollard 2007). Furthermore, these phenotypes are also possibly confounded with genetic ancestry, highlighting our need for large and diverse cohorts to tease apart the subtle influences of PAE on the genome and identify critical periods of vulnerability.  Finally, to fully assess the role of epigenetic mechanisms in PAE-induced associated physiological functions, we must begin to integrate the multiple layers of genetic and epigenetic machinery, from chromatin alterations and DNA methylation to miRNA and lncRNA expression (Lister et al. 2013). Future studies should also assess the concordance of these changes with mRNA expression, as the relationship between epigenetic patterns and transcription is highly   181 complex and has yet to be fully elucidated. Some studies of PAE have already begun to fill this niche, identifying concomitant changes in gene expression, histone modification levels, and DNA methylation patterns of POMC and VGLUT2 (Bekdash, Zhang, & Sarkar 2013; Zhang et al. 2015). However, much work is needed before we can successfully integrate the multiple layers of genome-wide epigenomic regulation in the etiology of FASD.   6.4 Future directions  Although the study of genetic and epigenetic patterns following PAE is progressing at a relatively rapid rate, a number of key issues remain elusive in regards to both mechanisms of fetal programming and biomarkers of FASD. For one, early evidence from some groups suggests that developmental alcohol exposure could potentially have lasting impacts on the epigenome of future generations, suggesting a possible role for inter- or transgenerational epigenetic inheritance (Govorko et al. 2012). While these data are certainly intriguing and raise important ethical considerations in the study and prevention of FASD, they must be interpreted with relative caution due to severe limitations in studying such effects. First, the interpretation of these results must take into consideration the number of generations to determine whether they are considered inter or transgenerational, which are commonly confounded due to the presence of cells for the F2 generation in the pregnant F0 female (van Otterdijk & Michels 2016). Second, these studies were performed in rodent models, which have not yet been shown to display the same inheritance patterns as humans. Third, no cohorts are currently available for the study of transgenerational inheritance in humans, and the current evidence remains tenuous at best. Nevertheless, although much work must be done to fully assess the implications of inter- or   182 transgenerational epigenetic inheritance in FASD, this remains an intriguing and important area of research that certainly warrants further investigation.   Another issue facing the field is that as of yet, and perhaps not surprisingly at this time, the vast majority of epigenetic studies rely on correlation, rather than causation. Given that different environmental factors have been shown to modulate PAE-induced deficits, including stress, immune challenges, nutrition, and early-life adversity, futures studies must also begin to address the differences and similarities between basal and inducible alterations to gene expression and epigenetic patterns. Model organisms, such as mice, rats, zebrafish, M. drosophila, C. elegans, etc. will play a crucial role in addressing this issue, as they allow for finer manipulations of biological systems and tighter control of environmental conditions. Perhaps most importantly, we must begin to position epigenetic mechanisms at the nexus of exposure paradigms and phenotypic outcomes to provide better insight into the etiology of FASD. Furthermore, analysis of both central and peripheral tissues in animal models will be vital before we can begin to make functional inferences in clinical settings, as human epigenetic studies mainly rely on peripheral tissues such as BEC and blood.  Although the degree to which peripheral alterations are linked to the mechanisms underlying FASD remains unknown, they may present a unique opportunity to develop accurate epigenetic biomarkers of PAE. In many cases, the deficits associated with FASD only become evident long after exposure, highlighting the importance of early biomarkers as tools to identify at risk children and mitigate the long-term effects of alcohol. More recent studies in animal models and clinical populations of individuals with FASD are beginning to provide a solid foundation for biomarker discovery with hopes for definitive markers in the relatively near future. Of the utmost importance in this line of research are additional studies to validate current   183 findings and to begin to assess the accuracy and specificity of these types of markers. While a characteristic epigenomic signature appears to occur in the buccal cells of children with FASD, these finding requires additional validation and testing in a clinical setting. Furthermore, strong correlations have been identified between genetic background and epigenetic patterns, particularly in the case of gene by environment (GxE) interactions (Fraser et al. 2012; Heyn et al. 2013; Moen et al. 2013). This work also points to the functional effects of methylation quantitative trait loci (mQTL), defined as an allelic variant that correlates with CpG methylation levels in its vicinity ( Jones, Fejes, & Kobor 2013). A number of studies have explored the occurrence of mQTLs in the human brain, showing that mQTLs tend to occur as cis associations in different brain regions and may underlie risk loci of various neuropsychiatric diseases, such as schizophrenia and bipolar disorder ( Zhang et al. 2010; Gibbs et al. 2010; Gamazon et al. 2013; Hannon et al. 2015; Jaffe et al. 2015). Given the challenges in obtaining cohorts of children with homogenous ethnicities, it will be vital to assess the relevance and implications of methylation quantitative trait loci or allelic variants correlating with nearby CpG methylation levels in the context of FASD. Longitudinal studies will also be integral to the identification of PAE-associated alterations to epigenetic profiles, as cross-sectional studies may not fully reflect the diversity of individuals with FASD across development and aging. Importantly, the field must also begin to move beyond early life outcomes and extend its focus into adolescence and adulthood, as data on adolescents and adults with FASD remain sparse. These studies will further develop a role for altered epigenetic programming in FASD and long-term health outcomes, be they immune, neurological, or stress-related (Moore & Riley 2015). In addition, these may prove crucial to our understanding of the etiology of FASD, particularly given the relationship between aging,   184 disease, and DNA methylation (Jones, Goodman, & Kobor 2015). These longitudinal cohorts will also be necessary to assess the persistence of epigenetic reprogramming by PAE and the potential validity of biomarkers over time. Epigenetic profiles may also serve as better markers of FASD if they are developed in conjunction with different stratification tools, such as magnetic resonance imaging (MRI), eye tracking, physical and mental health diagnostics, and immune markers, to parse out the wide range of deficits associated with FASD and create more accurate diagnostic tools. Finally, we must also begin to assess the overlaps, or lack thereof, in epigenetic patterns among different neurodevelopmental disorders, as they may display similar deficits and share common or overlapping molecular etiologies (Kelleher & Corvin 2015). The integration of these findings will provide important insight into the root causes of these disorders and may provide additional strategies for both diagnostic tools and therapeutic interventions.   6.5 Conclusions Despite the recognition of FAS over 40 years ago, PAE remains the leading cause of developmental disability in the developed world, as recent North American estimates place the incidence between 2-5% ( Jones & Smith 1973; Lemoine et al. 1968; May & Gossage 2001; May et al. 2014, 2015). However, early identification of individuals with FASD remains difficult, limiting the effectiveness of current interventions, which still lack specific molecular or neurobiological targets (Murawski et al. 2015). Although the study of genetic and epigenetic patterns in FASD remains an emerging field, it has provided important contributions to our understanding of the molecular underpinnings of FASD. To date, epigenetic research has identified numerous alterations to gene expression, DNA methylation patterns, chromatin states, and ncRNA expression levels, which provide important neurobiological insight into the deficits   185 associated with FASD, while also potentially uncovering targets for therapeutic intervention. This work has also begun to lay the groundwork for the development of epigenetic biomarkers of PAE, which may be the key to identifying children at risk for FASD. In turn, the identification of valid biomarkers will eventually support the creation of strategies for earlier diagnoses and targeted interventions to improve the lives of children and families affected by FASD.   186 References Abdolmaleky H.M., Smith C.L., Zhou J.-R., & Thiagalingam S. 2008. “Epigenetic Alterations of the Dopaminergic System in Major Psychiatric Disorders.” Methods in Molecular Biology 448: 187–212. Ahluwalia B., Wesley B., Adeyiga O., Smith D.M., Da-Silva A., & Rajguru S. 2000. “Alcohol Modulates Cytokine Secretion and Synthesis in Human Fetus: An in Vivo and in Vitro Study.” Alcohol 21 (3): 207–13. Alaghband Y., Bredy T.W., & Wood M.A. 2016. “The Role of Active DNA Demethylation and Tet Enzyme Function in Memory Formation and Cocaine Action.” Neuroscience Letters 625 (June): 40–46. Ammann A., Wara D., Cowan M., Barrett D., & Stiehm E. 1982. “The Digeorge Syndrome and the Fetal Alcohol Syndrome.” American Journal of Diseases of Children 136 (10): 906–8. Ashburner M., Ball C.A., Blake J.A., Botstein D., Butler H., Cherry J.M., Davis A.P., et al. 2000. “Gene Ontology: Tool for the Unification of Biology. The Gene Ontology Consortium.” Nature Genetics 25 (1): 25–29. Astley S.J., & Clarren S.K. 2000. “Diagnosing the Full Spectrum of Fetal Alcohol-Exposed Individuals: Introducing the 4-Digit Diagnostic Code.” Alcohol and Alcoholism 35 (4): 400–410. Astley S.J., Olson H.C., Kerns K., Brooks A., Aylward E.H., Coggins T.E., Davies J., et al. 2009. “Neuropyschological and Behavioral Outcomes from a Comprehensive Magnetic Resonance Study of Children with Fetal Alcohol Spectrum Disorders.” Canadian Journal of Clinical Pharmacology 16 (1): e178-201. Bakhireva L.N., Leeman L., Savich R.D., Cano S., Gutierrez H., Savage D.D., & Rayburn W.F. 2014. “The Validity of Phosphatidylethanol in Dried Blood Spots of Newborns for the Identification of Prenatal Alcohol Exposure.” Alcoholism: Clinical and Experimental Research 38 (4): 1078–85. Balaraman S., Schafer J.J., Tseng A.M., Wertelecki W., Yevtushok L., Zymak-Zakutnya N., Chambers C.D., & Miranda R.C. 2016. “Plasma miRNA Profiles in Pregnant Women Predict Infant Outcomes Following Prenatal Alcohol Exposure.” PLoS One 11 (11): e0165081. Balaraman S., Winzer-Serhan U.H., & Miranda R.C. 2012. “Opposing Actions of Ethanol and Nicotine on MicroRNAs Are Mediated by Nicotinic Acetylcholine Receptors in Fetal Cerebral Cortical-Derived Neural Progenitor Cells.” Alcoholism: Clinical and Experimental Research 36 (10): 1669–77. Bale T.L., & Epperson C.N. 2015. “Sex Differences and Stress across the Lifespan.” Nature Neuroscience 18 (10): 1413–20. Banovich N.E., Lan X., McVicker G., Geijn B. van de, Degner J.F., Blischak J.D., Roux J., Pritchard J.K., & Gilad Y. 2014. “Methylation QTLs Are Associated with Coordinated Changes in Transcription Factor Binding, Histone Modifications, and Gene Expression Levels.” PLoS Genetics 10 (9): e1004663. Barilla M.L., & Carsons S.E. 2000. “Fibronectin Fragments and Their Role in Inflammatory Arthritis.” Seminars in Arthritis and Rheumatism 29 (4): 252–65. Barker D.J.P. 1997. “Fetal Nutrition and Cardiovascular Disease in Later Life.” British Medical Bulletin 53 (1): 96–108.   187 Barker D.J.P. 2003. “Editorial: The Developmental Origins of Adult Disease.” European Journal of Epidemiology 18 (8): 733–36. ———. 2004. “The Developmental Origins of Adult Disease.” Journal of the American College of Nutrition 23 (sup6). Taylor & Francis: 588S–595S. ———. 2007. “The Origins of the Developmental Origins Theory.” Journal of Internal Medicine 261 (5): 412–17. Barker D.J.P., Godfrey K.M., Gluckman P.D., Harding J.E., Owens J.A., & Robinson J.S. 1993. “Fetal Nutrition and Cardiovascular Disease in Adult Life.” The Lancet 341 (8850): 938–41. Barker D.J.P., & Osmond C. 1986. “Infant Mortality, Childhood Nutrition, and Ischaemic Heart Disease in England and Wales.” The Lancet 327 (8489): 1077–81. Barker D.J.P., Osmond C., Winter P.D., Margetts B., & Simmonds S.J. 1989. “Weight in Infancy and Death from Ischaemic Heart Disease.” The Lancet 334 (8663): 577–80. Barker D.J.P., & Thornburg K.L. 2013. “The Obstetric Origins of Health for a Lifetime.” Clinical Obstetrics and Gynecology 56 (3). Barr H.M., Bookstein F.L., O’Malley K.D., Connor P.D., Huggins J.E., & Streissguth A.P. 2006. “Binge Drinking during Pregnancy as a Predictor of Psychiatric Disorders on the Structured Clinical Interview for DSM-IV in Young Adult Offspring.” American Journal of Psychiatry 163 (6): 1061–65. Bau C.H., Almeida S., Costa F.T., Garcia C.E., Elias E.P., Ponso A.C., Spode A., & Hutz M.H. 2001. “DRD4 and DAT1 as Modifying Genes in Alcoholism: Interaction with Novelty Seeking on Level of Alcohol Consumption.” Molecular Psychiatry 6 (1): 7–9. Baubec T., & Schübeler D. 2014. “Genomic Patterns and Context Specific Interpretation of DNA Methylation.” Current Opinion in Genetics and Development 25 (1): 85–92. Bearer C.F., Jacobson J.L., Jacobson S.W., Barr D., Croxford J., Molteno C.D., Viljoen D.L., Marais A.-S., Chiodo L.M., & Cwik A.S. 2003. “Validation of a New Biomarker of Fetal Exposure to Alcohol.” The Journal of Pediatrics 143 (4): 463–69. Bearer C.F., Lee S., Salvator A.E., Minnes S., Swick A., Yamashita T., & Singer L.T. 1999. “Ethyl Linoleate in Meconium: A Biomarker for Prenatal Ethanol Exposure.” Alcoholism: Clinical and Experimental Research 23 (3): 487–93. Bearer C.F., Santiago L.M., O’Riordan M.A., Buck K., Lee S.C., & Singer L.T. 2005. “Fatty Acid Ethyl Esters: Quantitative Biomarkers for Maternal Alcohol Consumption.” The Journal of Pediatrics 146 (6): 824–30. Bekdash R.A., Zhang C., & Sarkar D.K. 2013. “Gestational Choline Supplementation Normalized Fetal Alcohol-Induced Alterations in Histone Modifications, DNA Methylation, and Proopiomelanocortin (POMC) Gene Expression in Beta-Endorphin-Producing POMC Neurons of the Hypothalamus.” Alcoholism: Clinical and Experimental Research 37 (7): 1133–42. Benjamini Y., & Hochberg Y. 1995. “Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing.” Journal of the Royal Statistical Society. Series B (Methodological) 57 (1): 289–300. Berko E.R., Suzuki M., Beren F., Lemetre C., Alaimo C.M., Calder R.B., Ballaban-Gil K., et al. 2014. “Mosaic Epigenetic Dysregulation of Ectodermal Cells in Autism Spectrum Disorder.” PLoS Genetics 10 (5): e1004402. Bernardini R., Kamilaris T.C., Calogero A.E., Johnson E.O., Gomez M.T., Gold P.W., &   188 Chrousos G.P. 1990. “Interactions between Tumor Necrosis Factor-Alpha, Hypothalamic Corticotropin-Releasing Hormone, and Adrenocorticotropin Secretion in the Rat.” Endocrinology 126 (6): 2876–81. Bibikova M., Barnes B., Tsan C., Ho V., Klotzle B., Le J.M., Delano D., et al. 2011. “High Density DNA Methylation Array with Single CpG Site Resolution.” Genomics 98 (4): 288–95. Bird A. 2007. “Perceptions of Epigenetics.” Nature 447 (7143): 396–98. Bock C. 2009. “Epigenetic Biomarker Development.” Epigenomics 1 (1): 99–110. Bodnar T.S., Hill L.A., & Weinberg J. 2016. “Evidence for an Immune Signature of Prenatal Alcohol Exposure in Female Rats.” Brain, Behavior, and Immunity 58: 130–41. Bodnar T.S., & Weinberg J. 2013. Neural-Immune Interactions in Brain Function and Alcohol Related Disorders. Edited by Changhai Cui, Lindsey Grandison, and Antonio Noronha. Boston, MA: Springer US. Bomholt S.F., Harbuz M.S., Blackburn-Munro G., & Blackburn-Munro R.E. 2004. “Involvement and Role of the Hypothalamo-Pituitary-Adrenal (HPA) Stress Axis in Animal Models of Chronic Pain and Inflammation.” Stress 7 (1): 1–14. Bonthius D.J., & West J.R. 1990. “Alcohol-Induced Neuronal Loss in Developing Rats: Increased Brain Damage with Binge Exposure.” Alcoholism: Clinical and Experimental Research 14 (1): 107–18. Borck G., Mollà-Herman A., Boddaert N., Encha-Razavi F., Philippe A., Robel L., Desguerre I., et al. 2008. “Clinical, Cellular, and Neuropathological Consequences of AP1S2 Mutations: Further Delineation of a Recognizable X-Linked Mental Retardation Syndrome.” Human Mutation 29 (7): 966–74. Boyce W.T., & Kobor M.S. 2015. “Development and the Epigenome: The ‘synapse’ of Gene-Environment Interplay.” Developmental Science 18 (1): 1–23. Brand J.M., Frohn C., Cziupka K., Brockmann C., Kirchner H., & Luhm J. 2004. “Prolactin Triggers pro-Inflammatory Immune Responses in Peripheral Immune Cells.” European Cytokine Network 15 (2): 99–104. Burns E., Gray R., & Smith L.A. 2010. “Brief Screening Questionnaires to Identify Problem Drinking during Pregnancy: A Systematic Review.” Addiction 105 (4): 601–14. Cabarcos P., Álvarez I., Tabernero M.J., & Bermejo A.M. 2015. “Determination of Direct Alcohol Markers: A Review.” Analytical and Bioanalytical Chemistry 407 (17): 4907–25. Cahoy J.D., Emery B., Kaushal A., Foo L.C., Zamanian J.L., Christopherson K.S., Xing Y., et al. 2008. “A Transcriptome Database for Astrocytes, Neurons, and Oligodendrocytes: A New Resource for Understanding Brain Development and Function.” The Journal of Neuroscience 28 (1): 264–78. Carter J.L., Lubahn C., Lorton D., Osredkar T., Der T.C., Schaller J., Evelsizer S., et al. 2011. “Adjuvant-Induced Arthritis Induces c-Fos Chronically in Neurons in the Hippocampus.” Journal of Neuroimmunology 230 (1–2): 85–94. Carter R.C., Jacobson J.L., Molteno C.D., Dodge N.C., Meintjes E.M., Jacobson S.W., May P., et al. 2016. “Fetal Alcohol Growth Restriction and Cognitive Impairment.” Pediatrics 138 (2): 176192. Castells S., Mark E., Abaci F., & Schwartz E. 1981. “Growth Retardation in Fetal Alcohol Syndrome. Unresponsiveness to Growth-Promoting Hormones.” Developmental Pharmacology and Therapeutics 3 (4): 232–41.   189 Chater-Diehl E.J., Laufer B.I., Castellani C.A., Alberry B.L., & Singh S.M. 2016. “Alteration of Gene Expression, DNA Methylation, and Histone Methylation in Free Radical Scavenging Networks in Adult Mouse Hippocampus Following Fetal Alcohol Exposure.” PLoS ONE 11 (5): e0154836. Chen C.P., Kuhn P., Advis J.P., & Sarkar D.K. 2006. “Prenatal Ethanol Exposure Alters the Expression of Period Genes Governing the Circadian Function of β-Endorphin Neurons in the Hypothalamus.” Journal of Neurochemistry 97 (4): 1026–33. Chen D., Liu F., Shang Q., Song X., Miao X., & Wang Z. 2011. “Association between Polymorphisms of DRD2 and DRD4 and Opioid Dependence: Evidence from the Current Studies.” American Journal of Medical Genetics Part B: Neuropsychiatric Genetics 156 (6): 661–70. Chen L., & Nyomba B.L.G. 2003. “Effects of Prenatal Alcohol Exposure on Glucose Tolerance in the Rat Offspring.” Metabolism 52 (4): 454–62. Chen M., Olson H., Picciano J., Starr J., & Owens J. 2012. “Sleep Problems in Children with Fetal Alcohol Spectrum Disorders.” Journal of Clinical Sleep Medicine 8: 421–29. Chen Y., Ozturk N.C., & Zhou F.C. 2013. “DNA Methylation Program in Developing Hippocampus and Its Alteration by Alcohol.” PLoS ONE 8 (3): 1–11. Cheng J., Wang Y., Zhou K., Wang L., Li J., Zhuang Q., Xu X., et al. 2014. “Male-Specific Association between Dopamine Receptor D4 Gene Methylation and Schizophrenia.” PLoS ONE 9 (2): e89128. Chover-Gonzalez A.J., Harbuz M.S., Tejedor-Real P., Gibert-Rahola J., Larsen P.J., & Jessop D.S. 1999. “Effects of Stress on Susceptibility and Severity of Inflammation in Adjuvant-Induced Arthritis.” In Annals of the New York Academy of Sciences, 876:276–86. Chudley A.E., Conry J., Cook J.L., Loock C., Rosales T., & LeBlanc N. 2005. “Fetal Alcohol Spectrum Disorder: Canadian Guidelines for Diagnosis.” Canadian Medical Association Journal 172 (5 Suppl): S1–21. Church M.W., & Gerkin K.P. 1988. “Hearing Disorders in Children with Fetal Alcohol Syndrome: Findings from Case Reports.” Pediatrics 82 (2): 147–54. Clausing P., Ali S.F., Taylor L.D., Newport G.D., Rybak S., & Paule M.G. 1996. “Central and Peripheral Neurochemical Alterations and Immune Effects of Prenatal Ethanol Exposure in Rats.” International Journal of Developmental Neuroscience 14 (4): 461–69. Clayton J. a., & Collins F.S. 2014. “NIH to Balance Sex in Cell and Animal Studies.” Nature 509 (7500): 282–83. Colebatch A.N., & Edwards C.J. 2011. “The Influence of Early Life Factors on the Risk of Developing Rheumatoid Arthritis.” Clinical and Experimental Immunology 163 (1): 11–16. Concheiro-Guisan A., & Concheiro M. 2014. “Bioanalysis during Pregnancy: Recent Advances and Novel Sampling Strategies.” Bioanalysis 6 (23): 3133–53. Cordaux R., & Batzer M.A. 2009. “The Impact of Retrotransposons on Human Genome Evolution.” Nature Reviews Genetics 10 (10): 691–703. Crews F.T., Bechara R., Brown L.A., Guidot D.M., Mandrekar P., Oak S., Qin L., Szabo G., Wheeler M., & Zou J. 2006. “Cytokines and Alcohol.” Alcoholism: Clinical and Experimental Research 30 (4): 720–30. Crofford L.J., Sano H., Karalis K., Webster E.L., Goldmuntz E.A., Chrousos G.P., & Wilder R.L. 1992. “Local Secretion of Corticotropin-Releasing Hormone in the Joints of Lewis Rats with Inflammatory Arthritis.” Journal of Clinical Investigation 90 (6): 2555–64.   190 Cuadrado A., & Nebreda A.R. 2010. “Mechanisms and Functions of p38 MAPK Signalling.” The Biochemical Journal 429 (3): 403–17. Cunningham E.T., & Souza E.B. De. 1993. “Interleukin 1 Receptors in the Brain and Endocrine Tissues.” Immunology Today 14 (4): 171–76. Dadds M.R., Schollar-Root O., Lenroot R., Moul C., & Hawes D.J. 2016. “Epigenetic Regulation of the DRD4 Gene and Dimensions of Attention-Deficit/hyperactivity Disorder in Children.” European Child & Adolescent Psychiatry 25 (10): 1081–89. Davies M.N., Volta M., Pidsley R., Lunnon K., Dixit A., Lovestone S., Coarfa C., et al. 2012. “Functional Annotation of the Human Brain Methylome Identifies Tissue-Specific Epigenetic Variation across Brain and Blood.” Genome Biology 13 (6): R43. De Souza R.A.G., Islam S.A., McEwen L.M., Mathelier A., Hill A., Mah S.M., Wasserman W.W., Kobor M.S., & Leavitt B.R. 2016