COMPUTATIONAL INVESTIGATION OF TRANSCRIPTOMIC AND GENETIC UNDERPINNINGs OF AGING AND HGPS

Abstract

Normal aging is a complex process affecting everyone, and also a major risk factor for many complex diseases. Hutchinson Gilford progeria syndrome (HGPS) is a rare genetic disease with symptoms of aging at a very early age. There are some known and other presumed overlaps between HGPS and normal aging process. My goal in this dissertation is to perform computational investigation in both transcriptomic and genomic level to uncover potential underpinnings of these two models using high throughput genomic data.

Firstly in order to detect the common and distinct gene expression patterns between HGPS and normal aging, which might suggest their potential molecular links, I developed a novel approach that leverages co-expressed gene clusters to identify gene clusters whose expression co-varies with age and/or HGPS with limited sample size. Our results recapitulate previously known processes underlying aging as well as suggest numerous unique processes underlying aging and HGPS. Moreover, it isknown that alternative splicing contributes to phenotypic diversity at multiple biological scales, and its dysregulation is implicated in both aging and age-associated diseases in human. We aim to provide more insight into aging and age related diseases by studying splicing regulation. Then secondly we performed the first comparative investigation on splicing predictability of genomic and epigenomic features using a deep neural network model (DNN). We showed genomic features are the primary driver of splicing, and epigenomics is not contributing extra regulatory information independent to genomics. In addition, cross-tissue variability in splicing further complicates its links to age-associated phenotypes and elucidating these links requires a comprehensive map of age-associated splicing changes across multiple

tissues. Thus thirdly we generate such a map by analyzing ~8500 RNA-seq samples across 48 tissues in 544 individuals. Employing a stringent model controlling for multiple confounders, we identify 49,869 tissue-specific age-associated splicing events of 7 distinct types. We find that genome-wide splicing profile is a better predictor of biological age than the gene and transcript expression profiles, and furthermore, age-associated splicing provides an additional independent contribution to age-associated complex diseases. In fact in this specific study we presented the first systematic investigation of age-associated splicing changes across tissues, and further strengthening the links between age-associated splicing and age-associated diseases. Besides aging factor, genetic variations also potentially contribute to age-related disease shown by GWAS studies. However, potential interactions between aging and genomic variations have not been elucidated fully. It is highly likely that phenotypic effect of systemic molecular changes through aging may depend on the genotype ofthe individual. Lastly we approximate the environmental changes by age-associated changes in the levels of regulatory proteins, and exploiting the known mechanisms of transcriptional regulation, explore potential causal interaction between genotype and aging toward explaining age-related transcriptional and ultimately, age-related diseases. We detected numerous interactions across 25 tissues and showed they could potentially be associated with hypertension disease. In summary, our investigations in this dissertation provided predictive hallmarks along with implied molecular basis

insight about normal aging and HGPS in transcriptomic and genetic level.