Using genetics of biomarkers and Bayesian shrinkage prediction to identify genetic factors of giant cell arteritis and its complications

Over the last decade, significant progress has been made in characterising the genetic basis of many common diseases. Genetic discovery has been mainly driven by large consortia combining study cohorts to perform genome wide association studies with increased statistical power. For rare diseases and for useful clinical measures such as disease progression, large sample sizes are difficult or impossible to collect, slowing the rate of genetic discoveries. One of these diseases is giant cell arteritis (GCA). Accumulating evidence points to a genetic predisposition for GCA but lacks robust characterization of the genetic variants contributing to disease risk [1, 2].
GCA is the most common adulthood vasculitis. It occurs exclusively in people over 50 years of age and is characterised by inflammation of large blood vessels. Vasculitis leads to ischaemic complications, including irreversible vision loss. These occur in 19% of UK patients, despite prompt treatment. Polymyalgia rheumatica (PMR), which is characterised by muscle pain and stiffness, is present in 50% of GCA patients. Most patients commence glucocorticoid monotherapy, which is gradually reduced once clinical symptoms have abated. However, both GCA and PMR have a high relapse rate, with 50% remaining glucocorticoid dependent 2-3 years later, leading to significant toxicity and adverse events.
We will study the genetic component of GCA using one of the largest collections of GCA cohorts with genome-wide genotypic data and clinical phenotype data, including PMR and ischemic complications. To increase statistical power, we propose to reduce the size of the hypothesis space in two novel ways
1. Feature engineering: Collapse the genome-wide genetic data into a smaller number of genetic proxy measures (locus-specific genetic risk scores) for relevant intermediate traits, including immune and vascular biomarkers and clinical traits.
2. Advanced statistical methods: Apply Bayesian sparse prediction models to learn which genetic scores are most informative. Evaluate possible extensions to the hierarchical shrinkage prior distribution that group genetic scores (e.g. based on biological pathways) or GCA patients (e.g. based on PMR occurrence).
Two recent publications demonstrate the strength of the proposed methodology over conventional GWAS in important clinical areas, where sample sizes are relatively small [3, 4].
Aims
The aim of this project is to use genotypic data to elucidate the pathogenesis of GCA and related clinical phenotypes (PMR/ischaemic complications) and to identify new treatment targets for subsequent translational research. The key objectives are:
1. Identify relevant intermediate traits (immune and vascular traits, cytokines, protein levels, gene expressions) and compute genetic risk scores. A curated database with results from over 20,000 GWAS, as well as functionality for computing genetic scores, is available through our GENOSCORES platform.
2. Apply and extend Bayesian sparse prediction models to evaluate the effect of candidate genetic scores on GCA susceptibility (case/control analysis) and ischaemic complications (survival analysis). 3. Use bioinformatics databases, such as Reactome, to map identified scores to biological pathways, and assess if scores for other pathway molecules are also associated with GCA.
4. Use cheminformatics databases, such as DrugBank, to establish if identified scores correspond to known drug targets and evaluate the potential for drug repositioning.
5. Evaluate the possibility of following up novel targets by generating and analysing proteomic or transcriptomic data in available biosamples or using additional cohorts, such as the UK Biobank.