Blood- and saliva-based DNA methylation biomarkers of disease

Thousands of proteins can be measured from blood samples, offering key insights into disease risk prediction. We were one of the first groups to show how individual proteins and multi-protein signatures associated with disease outcomes in UK Biobank. Others have shown that weighted combinations of blood-based proteins can track the health of 11 organs (organ ageing). A limitation of this work is that many proteins fluctuate over the short-term; a single time-point measure may not reflect important longer-term effects. We have shown that blood-based DNA methylation (DNAm) proxies for proteins (protein epigenetic scores or EpiScores) can yield stronger associations with disease outcomes than measured proteins. EpiScores provide a more stable estimate of chronic exposures than single-time-point protein measurements. We will develop EpiScores for individual proteins and protein-based organ age estimates and their changes over time in minimally and non-invasively collected biosamples (blood and saliva). The EpiScores will then be tested in relation to incident disease outcomes. A novel multivariate GWAS framework and time-aware graphical models will be developed to infer causal pathways between proteins/their EpiScores and disease outcomes. Our data include the world’s largest blood- and saliva-based DNAm cohort (Generation Scotland – GS, n=18,869 and n=10,491, respectively) and a longitudinal study with >10,000 proteins and genome-wide DNAm measured at up to 4 and 6 time-points per volunteer, respectively (Lothian Birth Cohort 1936 – LBC1936, n=795). These data have already been or are currently being generated, independently of this application. In the proposed work, we will conduct a suite of multi-omic analyses to: 1. build EpiScores for 11 organ age measures and >10,000 proteins (LBC1936). These will be projected into blood-based DNAm data from 18,869 GS volunteers and tested as predictors of 10-year onset of 175 diverse disease outcomes (minimum 30 incident cases). 2. extract longitudinal slopes for the >10,000 proteins, 11 organ ages and an overall multi-organ ageing measure and derive EpiScores for their rates of change (LBC1936). The EpiScores will then be considered as predictors of incident disease outcomes as per (1). 3. determine which DNAm organ age and protein EpiScores correlate highly across 600 paired GS blood/saliva samples prior to being projected into 10,491 GS individuals with salivary DNAm and up to 8-years of incident disease data. 4. apply the disease-associated EpiScores to blood- and saliva-based cohorts from two international epigenetics consortia (N>80 cohorts, N~90,000 samples) in external testing and replication of our findings. 5. develop and apply a novel multivariate genome-wide association study framework to identify the shared and unique SNP correlates of a measured protein and its EpiScore analogue. This work will be done for up to 4,000 protein/protein EpiScore pairs in 774 Generation Scotland individuals. 6. use summary-level bivariate associations between the omics layers (SNPs, CpGs, EpiScores and proteins) and disease outcomes to build graph models that identify age-specific causal pathways. All of our findings will be made publicly available and we will create easy-to-use tools and user interfaces to allow the wider scientific community to efficiently and securely generate DNAm biomarkers within their own cohorts. By building signatures for disease risk from an accessible biosample (saliva), cross-referencing these to findings from blood, and developing and applying novel causal inference frameworks, our work has the potential to transform how we assess biomarkers of ageing.