Autism spectrum disorder (ASD) is a group of etiologically and phenotypically heterogeneous neurodevelopmental disorders defined by deficits in social communications and mental flexibility. It is established that variation in hundreds of genetic loci contributes substantially to ASD risk. This has raised the question of whether these mutations, which are found across disparate genes, affect similar biological pathways, perturb brain development at a particular time point, or disrupt a specific brain system. Addressing this is critical to develop a molecular and neurobiological understanding of ASD. However, searching for such convergence is made challenging by the genetic complexity of ASD and the molecular, cellular, and circuit-level complexity of the brain. Here, I apply comprehensive profiling of RNA levels (the transcriptome) by RNA sequencing (RNA-seq) to characterize the role of genes in normal human brain development and ASD. My overarching hypothesis is that there exist molecular regulators of transcription which are particularly susceptible to mutations in ASD, and that their regulatory targets are affected in ASD brain. My general approach is to organize the hundreds to thousands of disparate changes in transcript levels into more tractable and biologically meaningful gene sets or modules that are differentially expressed, co-expressed, or co-regulated. I then integrate these gene sets or modules with molecular and phenotypic information from whole genome studies and targeted experimental studies in order to systematically reveal new insights about ASD neurobiology and highlight specific genes and pathways worth investigating further. I first assess the role of ASD risk genes in normal brain development and then apply RNA-seq to measure transcriptomic changes in postmortem ASD brain to evaluate whether convergent neurobiological pathways are affected. I find robust evidence that developmentally co-expressed, co-regulated, and physically interacting genes are affected in ASD during normal brain development. Rare, highly deleterious variants predominantly exert their effect by disrupting major transcriptional and chromatin regulators in early fetal development, while less deleterious inherited variants affect late prenatal and early postnatal cellular and circuit maturation through alterations in synaptic function. Additionally, across most individuals with ASD, I find strongly shared changes in synaptic and neuronal genes at both a gene expression and transcript splicing level in cortex. Moreover, shared cortical changes are also seen in a genetically defined subtype of autism, duplication 15q syndrome (dup15q). In contrast, ASD-associated changes in cerebellum are weaker. Co-expression network analysis identifies specific cell types and circuits that are affected, and highlights specific transcriptional regulators likely to play a role in ASD pathology. Taken together, these results identify roles for transcriptional regulators in ASD and define the potential consequences of their dysregulation.

Autism Spectrum Disorders (ASDs) are a group of heritable neruodevlopmental disorders. Both common and rare genetic variants are known to play a role in ASDs. However the functional impact of genetic variants remains largely unexplored. In this study, we conducted transcriptome profiling analysis to uncover the expression alterations that are associated with autism. The transcriptome profiling also aids us exploring the regulatory patterns of genetic variants, and better understanding the genetic models of autism. Since brain tissue is not accessible on a large scale, we profiled mRNAs of lymphoblast cell lines (LCLs) from three independent cohorts to determine whether we could detect a reproducible blood gene expression pattern associated with ASD. RNA from a total of 978 patients, and 651 controls, including 607 unaffected siblings analyzed for differential expression. Although few genes were consistently differentially expressed between ASD and controls, we did find five (CMKOR1, DKFZP564O0823, PITPNC1, PRKCB1 and VIM) that were differentially expressed in at least two cohorts LCLs and previously published brain samples. Similarly, using LCL gene expression to classify subjects by disease status performed only slightly above chance. Using weighted gene co-expression network analysis (WGCNA), we were able to identify a module correlated with ASD in both AGRE and NIMH cohorts that overlapped with genes previously found to be mis-expressed in post mortem brain from ASD cases. eQTL analysis identified SNPs that were associated with LCL gene expression, including several in AHI1, a Joubert Syndrome gene dysregulated in ASD brain and lymphoblasts. Four of the 23 SNPs that were significantly correlated with the expression level of AHI1 reside in the same haplotype block previously associated with ASD, suggesting that risk for ASD is mediated via AHI1 transcript levels. Overall, we found a weak, but consistent signal in LCLs further suggesting that peripheral lymphoblast gene expression may be useful for studying ASD.

Rare variants including Copy Number Variants (CNVs) and Single Nucleotide Variants (SNVs) are found to play an important role to the etiology of ASD together with common variants. We next interrogated gene expression in lymphoblasts from 244 families with discordant siblings in the Simons Simplex Collection in order to identify potentially pathogenic variation. Our results reveal that the overall frequency of significantly mis-expressed genes (which we refer to here as outliers) identified in probands and unaffected siblings do not differ. However, in probands, but not their unaffected siblings, the group of outlier genes is significantly enriched in neural-related pathways including neuropeptide signaling, synaptogenesis and cell adhesion. We demonstrate that outlier genes cluster within the most pathogenic CNVs (rare de novo CNVs) and can be used to prioritize rare CNVs of potentially unknown significance. Several non-recurrent CNVs with significant gene expression alterations are identified, including deletions on chromosome 3q27, 3p13 and 3p26, and duplications at 2p15, suggesting these as potential novel ASDs loci. In addition, we identify distinct pathways disrupted in 16p11.2 microdeletions, microduplications and 7q11.23 duplications, and show that specific genes within the 16p CNV interval correlate with differences in head circumference, an ASDs relevant phenotype. This study provides evidence that pathogenic structural variants have functional impact on transcriptome alterations in ASDs at a genome-wide level, and demonstrates the utility of this approach for prioritization of genes for subsequent functional analysis.

Genetic studies have identified dozens of ASDs susceptibility genes, yet the interaction between ASD risk genes are pooly understood. In the aim of identify the molecular mechanisms and potential convergening pathways of ASD risk genes, the last chapter of my research utilizes transcriptome profiling to answer two questions: 1) do these genetic loci converge on specific laminar expression patterns, and 2) where does the phenotypic specificity of ASDs arise, given its genetic overlap with intellectual disability (ID)? To answer these, we mapped ASDs and ID risk genes to non-human primate and human brain transcriptome. We found ASDs genes are enriched in superficial cortical layers and glutamatergic projection neurons at the circuit level. Furthermore, we show that the patterns of ASDs and ID risk genes are distinct, providing a novel biological framework for investigating the pathophysiology of ASDs. In this chapter, we demonstrated the importance of understanding ASD gene interaction with systems biology method.

Neuropsychiatric disorders are behavioral conditions marked by intellectual, social, or emotional deficits that can be linked to diseases of the nervous system. Autism spectrum disorder (ASD), schizophrenia (SCZ), bipolar disorder (BP), major depressive disorder (MDD), and attention deficit and hyperactivity disorder (ADHD) are common, heritable diseases each with a prevalence exceeding 1% of the population, none of which can be characterized by discernable anatomical or neurological pathologies. Genetic association studies have identified mutations in hundreds of genes that contribute to risk for at least one of these disorders, and have shown that a substantial fraction of the genetic liability is shared between many of these neuropsychiatric diseases. It has long been hoped that with enough genetic evidence we will identify the biological pathways, developmental time points, and brain regions that, when disrupted, give rise to neuropsychiatric disorders. However, the cellular and functional complexity of the human brain, as well as the genetic complexity of neuropsychiatric disease, make it difficult to search for such convergence.

In this thesis, I investigate global and local transcriptional regulation within and across 12 regions of the human brain in order to investigate the regional specificity of neuropsychiatric disorders. I develop novel bioinformatics methods – ranging from data processing to network construction – to identify whether the transcriptional regulation of a set of genes is shared or specific. I hypothesize that local, region-specific transcriptional regulation corresponds directly to cell types and processes that are specific to, or far more prevalent in, a given region; that cross-regional transcriptional regulation corresponds to cell types that show little heterogeneity across brain regions; and that genetic disruption of region-specific transcriptional programs results in regional susceptibility. I use a systems-biology approach to summarize transcriptional regulation into reproducibly co-expressed gene sets (“co-expression modules”), which can be analyzed statistically to identify common functions, pathways, and cell types. I then integrate data from genetic association studies to ascertain gene sets conferring outsized risk for neuropsychiatric disorders, thereby implicating the corresponding pathways for further investigation in disease etiology. Finally, I use the network structure itself to investigate the genetic architecture of ASD and SCZ in terms of omnigenics and network polygenics.

Chapter 1 presents the biological background for the studies and summarizes some of the major studies of neuropsychiatric disorders along with their principal methods and conclusions. In chapter 2, utilizing my multi-regional co-expression approach, I identify 12 brain-wide, 114 region-specific, and 50 cross-regional co-expression modules. Nearly 40% of expressed genes fall into brain-wide modules and correspond to major cell classes and conserved biological processes, while region-specific modules comprise 25% of expressed genes and correspond to region-specific cell types. The detailed study in chapter 3 demonstrates that neuropsychiatric risk concentrates in both brain wide and multi-regional modules, implicating major core cell types in disease etiology but not region-specific susceptibility. Chapter 4 presents a new and more general framework for defining genetic networks. Using this framework, I show that the network pattern of ASD-associated rare loss-of-function mutations, as well as the large number of significant targets for trans master regulators in BP and SCZ, support a classical polygenic architecture with thousands of directly causal genes. These results suggest that a nontrivial component of risk for neuropsychiatric disease comes from the global polygenic disruption of neuronal function and neuronal maturation.