Acute myeloid leukaemia (AML) is an aggressive blood cancer which claims the lives of 70-80% of patients within 5 years of diagnosis. Like many other cancers, AML usually develops as a consequence of serial acquisition of somatic driver mutations; a process that starts many years, or even decades, before diagnosis. This raises the prospect that early detection of ‘pre-leukaemic’ mutations could be used to identify individuals at high risk of developing AML, in whom early intervention could halt the disease before it fully develops. One of the difficulties with early detection of AML is that clonally expanded leukaemia-associated mutations are also found in the blood of healthy individuals, a phenomenon termed ‘clonal haematopoiesis’. However, most individuals with clonal haematopoiesis will never progress to AML and so a key challenge is the identification of individuals most at risk. To do this, we need a better understanding of the evolutionary dynamics of clonal haematopoiesis in the years, or decades, before AML occurs and how this differs from the dynamics of clonal haematopoiesis in individuals that remain cancer-free.

We sought to understand this process by first studying the acquisition and expansion of the initial clonal haematopoiesis driver mutation. Using blood sequencing data amassed from ~50,000 individuals, combined with insights from evolutionary theory, we developed a framework to quantify the mutation rates and fitness effects of clonal haematopoiesis variants down to single nucleotide resolution. This enabled us to build a league table of the fittest and potentially most pathogenic variants in blood. We also quantified the distribution of fitness across key clonal haematopoiesis genes and found the distribution to be highly skewed, with most mutations in these genes conferring either a weak or no fitness effect. Our framework also reveals that whilst cell-extrinsic effects are likely crucial in some situations, the combined effects of chance (when a mutation arises) and cell-intrinsic fitness differences are the major forces shaping clonal haematopoiesis.

Mosaic chromosomal alterations (mCAs) can also be important drivers in AML and ~3% of individuals aged ~40-70 have a clonally expanded mCA detectable in >1% of their blood cells. We therefore adapted our framework to quantify the mutation rates and fitness effects of mCAs in blood and applied this to data generated from ~500,000 individuals in UK Biobank. We find most mCAs confer growth rates of ~10-20% per year and find correlation between mCA fitness and blood cancer risk. In contrast to the strong age dependence observed in single nucleotide variant prevalence in blood, we find mCA age dependence to be more variable, particularly in women, suggesting the risk of acquisition and/ or expansion of certain mCAs is non-uniform throughout life and is influenced by gender-specific factors.

To determine how the dynamics of clonal haematopoiesis differs in individuals who progress to AML, we identified longitudinal blood samples that had been collected annually at multiple timepoints from individuals who subsequently developed AML, as well as age-matched controls who remained cancer free. We developed a custom error-corrected duplex sequencing platform to detect mutations in 34 clonal haematopoiesis/AML-associated genes, genome-wide mCAs and AML-associated translocations and used this to perform an integrative assessment of the genetic changes in these samples. We find there are four main evolutionary patterns in the years preceding AML diagnosis: linear evolution, evolution with clonal interference, static evolution and late evolution. We calculate the age at acquisition of the first and second mutations and, whilst the initial driver mutation is often acquired early in life, there are some very fit ‘uber drivers' which appear to occur as the initial event just ~4 years pre-diagnosis. We find that the variants we identified as ‘highly fit' in clonal haematopoiesis are significantly enriched pre-AML and we were able to determine how fitness effects changed with the acquisition of subsequent mutations. NPM1 mutations, which characteristically occur late in AML development and have never been seen in individuals who do not progress to AML, can be detected as early as 2 years pre-AML diagnosis, highlighting the benefit afforded by low VAF variant calling, particularly in high-risk individuals.

This quantitative analysis of clonal haematopoiesis, combined with an integrated assessment of genetic changes in longitudinal blood samples from individuals who progress to AML, reveals important insights into the evolutionary dynamics of mutations in the years preceding AML. Understanding which features distinguish pre-malignant from benign clonal evolution is key for risk stratification of clonal haematopoiesis and will aid in the development of rational monitoring approaches and identification of those who may benefit from early intervention studies.