A toolbox for the promotion of healthy ageing: Phenotypic prediction from genes and environment

Many different factors influence the health of individuals, be they domestic animals or humans. These factors can broadly be categorised as either genetic or environmental. Thus the genes inherited from parents and the environments encountered during life are paramount in determining health status as one ages. These factors may also interact, such that individuals with one genetic make-up may react well to a particular environment, whereas a different genetic make-up may react badly. Where a substantial proportion of the genetic and environmental factors can be identified it is possible to provide accurate predictions of individuals' health as they age. Using such genetic information in prediction has great potential as it can be measured early in life and is unchanging throughout life. So there is the potential to be aware in advance of the environmental conditions that will optimise the future health of individuals. Such prediction is potentially a powerful tool to promote healthy ageing and wellbeing in both humans and companion animals, as it allows increasing efficiency of interventions, such as recommended diets or even drug treatments, and the targeting interventions towards those individuals who will most benefit. Combining genetic and environmental information is therefore the natural way to proceed when predicting how animals or humans will age and this project is concerned with developing accurate mathematical and statistical models to do this. Research in animals and humans has started the process of identifying genes affecting the traits associated with healthy ageing such as obesity or bone strength. However it has become clear that traits associated with healthy ageing are generally controlled by large numbers of genes with small effects. To unequivocally find such genes and accurately estimate their effects requires very large studies and relatively few genes have as yet been identified. Thus the amount of variation explained jointly by all the genes found in studies so far is usually much less than 10%, even though genetic variation in total may explain as much as 80% of the overall variation. Alongside genetic information, factors such as age, gender, diet and other lifestyle characteristics are often major contributors to how individuals develop. In addition, it is often known that metabolic or predisposing traits like glucose or lipid concentration in blood may correlate with health. Such traits may be more amenable to measurement or may be measured earlier than overall health status and may be used as indicators or predictors of future health. Thus information can also be combined across traits to improve the accuracy of prediction, and to allow prediction of (unmeasured) correlated traits. With this background we propose to develop mathematical methods which make best use of available genomic information and to combine this information with environmental data and across multiple traits. We will use several different approaches and compare them in their ability to accurately predict performance and how they may be extended to account for data from many traits and environments. We plan to apply and extend methods currently used in animal breeding for the related task of identifying genetically superior animals for breeding. These will be compared with machine learning methods from computer science. We plan to demonstrate the effectiveness of these methods applied to the analysis of data from human populations on body mass index - a proxy for obesity - and blood glucose levels, and will also include in the analyses environmental variables like smoking, diet and exercise. The data are currently available from human studies and methods and results will be relevant to this species. In due course, the methods developed will be directly applicable to companion animals as data become available.

The most prevalent non-infectious causes of poor health in companion animals and humans are related to aging and are multifactorial in origin: partly genetic and partly environmental, with these factors acting independently or through complex interactions. Where a substantial proportion of the systematic factors can be identified and modelled it is possible to provide accurate predictions of future life events including specific risks of developing these age-related syndromes or diseases in as yet unaffected individuals or beneficial or adverse drug response in those already affected. Such prediction is potentially a powerful tool to promote healthy ageing in both humans and animals, as it allows increasing effectiveness of interventions, and cost-effectiveness, by stratifying the population into subgroups, e.g. at risk/not at risk, responders and non-responders to particular environmental interventions and then targeting interventions. Key to accurate prediction is our ability to explain as much as possible of the trait variance, both genetic and non-genetic and we propose to develop methods that allow us to make best use of the information recorded in a clinical context or in population studies. This information includes records of environmental exposures, phenotypes of interest and genomic information. We propose improving predictive performance by better using information through: 1) Taking appropriate account of all genetic marker information, not just that from few highly selected markers 2) Using models that account for environment and genomic factors and their interactions 3) Utilising information from correlated traits in a multitrait framework 4) Identifying optimum prediction methods from those specific for problems where the set of predictive variables is much larger than the number of observations We will combine and refine these different methods and demonstrate their efficacy using simulated data and crossvalidation on large-scale GWAS data.

Impact to the academic community Our findings will be disseminated to the academic community by publishing in peer reviewed journals and by presenting them at national and international meetings. Publication in high-impact, peer-reviewed journals is necessary to give our finding authority. These journals will include general scientific journals where possible, and also leading specialist journals in the areas of genetics, animal breeding, veterinary medicine and medicine, and applications of machine learning methods (see Track Record and CVs). The range of stakeholders who will benefit from our research is very broad, and includes academics in the area of veterinary medicine, both for companion animals and livestock and medicine. Impact to the general public The MRC press office is the link between MRC scientists and the media, and is in charge of publicising research activities from MRC scientists. It also provides corporate information to the media and reports on recent research findings on the MRC news page. The press office represents our first step towards the communication of the results of our investigations to the lay public. In addition, the MRC participates annually in events like the Edinburgh Science Festival, and these are very excellent platforms for the dissemination of science, in which we would be very happy to participate. A specific and important issue to take into account when dealing with obtaining/using genetic information from humans is the fact that a proportion of the population is concerned about 'genetic discrimination'. An important task would be to emphasise the benefits that will come from using genetic information, and that, with a strong NHS, the risks of discrimination are somehow lower. Impact to potential users We will provide excellent tools for the prediction of phenotypic risk and hope to provide the foundation its application in veterinary and clinical medicine, as a means to promote lifelong health and wellbeing. We will engage directly with potential users of the approach and with policy makers to discuss aspects of the implementation. Potential users and groups involved in these discussions should include veterinarians, clinicians, clinical geneticists, epidemiologists, and the animal breeding and pharmaceutical industries. We have links with these user groups through ongoing collaborations (see Track Record section for examples). Careful dissemination of our findings to these groups and policy makers is of utmost importance and our work could be the basis for a larger collaborative project that could be more focussed on implementation and cost-effectiveness of the methods we will develop in the different fields of application. In addition to other dissemination already proposed, we propose to set up a workshop in the last semester of our project, to disseminate and discuss our findings amongst potential user groups and policy makers, and discuss potential applications. Impact to Health We believe the methodology proposed could significantly increase the accuracy of prediction of phenotypic value of health-related traits both in managed animals and in humans compared to currently used methods. This would represent an important step towards the possibility of using high-throughput genotyping at a population-wide level as a powerful tool to prevent poor health in old age. With escalating treatment costs, good screening strategies become more cost effective. To allow the full implementation of the methodology SNP array genotyping on a population basis is needed. This will become possible in the near future as genotyping costs decrease. Economic Impact Once each individual is typed, its genetic information could be used to inform on his or her risk to suffer a wide range of disorders later in life, and that will make the strategy very cost effective and will have an enormous impact on the way health services work, and on health.