The functional impact of copy number variation in the human genome

Copy number variation (CNV) is a class of genetic variation where large segments of the genome vary in copy number among different individuals. It has become clear in the past decade that CNV affects a significant proportion of the human genome and can play an important role in human disease. With array-based copy number detection and the current generation of sequencing technologies, our ability to discover genetic variants is running far ahead of our ability to interpret their functional impact. One approach to close this gap is to explore statistical association between genetic variants and phenotypes. In contrast to the successes of genome-wide association studies for common disease using common single nucleotide polymorphism (SNP) as markers, the majority of disease CNVs discovered so far have low population frequencies and are mainly involved in rare developmental disorders. Another strategy to improve interpretation of genomic variants is to establish a predictive understanding of their functional impact. Large heterozygous deletions are of particular interest, since (i) loss-of-function (LOF) of coding sequences encompassed by large deletions can be relatively unambiguously ascribed and (ii) haploinsufficiency (HI), wherein only one functional copy of a gene is not sufficient to maintain normal phenotype, is a major cause of dominant diseases. This thesis explored both approaches. Initially, I developed an informatics pipeline for robust discovery of CNVs from large numbers of samples genotyped using the Affymetrix whole-genome SNP array 6.0, to support both the association-based and prediction-based study. For the disease association strategy, I studied the role of both common and rare CNVs in severe early-onset obesity using a case-control design, from which a rare 220kb heterozygous deletion at 16p11.2 that encompasses SH2B1 was found causal for the phenotype and an 8kb common deletion upstream of NEGR1 was found to be significantly associated with the disease, particularly in females. Using the prediction-based approach, I characterized the properties of HI genes by comparing with genes observed to be deleted in apparently healthy individuals and I developed a prediction model to distinguish HI and haplosufficient (HS) genes using the most informative properties identified from these comparisons. An HI-based pathogenicity score was devised to distinguish pathogenic genic CNVs from benign genic CNVs. Finally, I proposed a probabilistic diagnostic framework to incorporate population variation, and integrate other sources of evidence, to enable an improved, and quantitative, identification of causal variants