Evolution forecasting for real-time blood-cancer risk prediction

Cancer is a disease of evolution that plays out in our bodies over decades. As our cells divide, occasional errors ("mutations") in DNA replication occur which can disrupt the normal function of the cell. If these "mutant clones" survive long enough they can acquire further cancer-causing mutations that eventually drive the total break-down of controlled cell proliferation. The first steps of this evolutionary process begin years before the development of cancer. This raises the possibility that these early events could be used as a bellwether for predicting who is at risk of cancer. However, because it is difficult to measure the evolution taking place inside our bodies over such long timescales, our understanding of the evolutionary dynamics driving early cancer remains cursory. A key gap in our understanding is our inability to identify which mutant clones will progress to lethal cancers and which will remain benign.

Serial blood samples collected annually from hundreds of thousands of healthy people give us a superpower that can fill in these gaps in understanding. We can "zoom in" on the people who develop cancer, then "rewind" time by analysing blood samples collected years before the cancer was diagnosed. This provides a detailed "fossil record" of the disease, enabling one to determine when the cancer first arose and to watch the entire evolutionary life-history of the tumour unfold, one DNA error at a time. An attractive initial target for this ambitious vision is the aggressive blood cancer Acute Myeloid Leukemia (AML) because it is characterised by a relatively small number of cancer-causing mutations, which are readily detectable in the blood.

This UKRI FLF application sets out an innovative long-term research programme for blood cancer prediction and early detection. To achieve this, first we will combine population genetic theory with the vast amounts of sequencing data from the blood of >50,000 individuals to characterise the evolutionary dynamics that defines "normal". Second, by exploiting an extraordinary set of serial blood samples in hundreds of AML cases and cancer-free controls we will generate a unique dynamic dataset that paints a highly quantitative genetic portrait of how AML evolves from healthy tissue. Third, by mining this rich data resource, we will train a set of mathematical and statistical models that use evolutionary dynamics theory and simulation (which we have previously pioneered) to make probabilistic "forecasts" of leukemia risk from a blood sample. This FLF application will thus establish blood-cancers as a "model system" for early cancer detection by combining unique longitudinal samples, novel sequencing technologies and emerging statistical methods to predict blood cancer risk.

When cancer is detected earlier, it is easier to treat successfully. As such, early detection of cancer is a major strategic focus for the UK. The major impact of this FLF application will be to accelerate the development of early detection tools for blood cancers by addressing three fundamental questions which will help the UK address the major challenge of early detection: predicting which pre-cancerous states will progress to lethal cancers (thus requiring intervention) and which remain benign or indolent. The ultimate impact of this research will be on patient outcomes.

Early detection requires samples collected before symptoms arise. Blood is an ideal system for this because large collections of serial bloods have been collected in the UK, from otherwise healthy people, for a number of long-term studies. These samples can be exploited to "rewind" time in people who develop cancer by analysing samples collected before diagnosis. This insight has been noted by US-based biotech firms with deep pockets (e.g. Grail, Google Life Sciences, Guardant Health and Freenome), all of whom have identified the need to start longitudinal studies of their own. It is crucial the UK regain the initiative by exploiting our unique existing samples resources and combine them with strengths in cancer-evolution, sequencing and quantitative skills. This FLF application will establish just that, by using blood-cancers as a "model system" for early cancer detection and linking unique longitudinal samples, novel sequencing technologies and emerging statistical methods to predict blood cancer risk. To establish this application's impact, we propose to make all (non-identifiable) data and analysis framework from this proposal rapidly available to the wider scientific community via a dedicated study portal. With a team and UK-based community of quantitatively trained clinicians, biologists, engineers and mathematicians dedicated to cancer-prediction though longitudinal profiling of biomarkers, these data will accelerate the development and commercialisation of early cancer detection tests.