Calum Gabbutt

Detection of copy number alterations (CNAs) in Barrett’s oesophagus (BO) predicts oesophageal adenocarcinoma (OAC) risk. When CNAs first occur is unknown. Here we show that CNAs typically form in BO many decades prior to first clinical detection of BO or cancer, and often during childhood. Using a novel molecular clock analysis, we analysed spatiotemporal longitudinal whole genome data from 80 individuals within BO surveillance. Chr 9p loss of heterozygosity (LOH) tended to be the first CNA event typically occurring at age 10-30 years. Genome doubling (GD) also typically originated multiple decades prior to cancer develop. Patients who progressed to OAC had an enrichment of CNAs late in life. CNA rate was more predictive of cancer risk than CNA burden. Analysis of OAC whole genomes confirmed the early-life origins of cancer-associated CNAs. The malignant transformation potential of BO may be set during the first decades of life, enabling early cancer interception.

Determining the nature of human tumour growth is challenging given the impracticality of obtaining detailed data across time. A promising solution is to examine DNA regions whose methylation states fluctuate on clinically relevant timescales, permitting their use as high-resolution lineage tracers. However, existing methods developed for analysing the fluctuating methylation loci of normal tissue and lymphoid cancers are inapplicable to large solid tumours. Here we introduce a mechanistic computational model that tracks the evolution of heritable methylation marks as a tumour grows from a single gland to a mass of many cubic centimetres, and a coupled ABC-SMC inference workflow to estimate tumour growth parameters from multi-region bulk methylation arrays. We applied this framework to data from multiple regions of 10 resected colorectal tumours, including 3 adenomas and 7 carcinomas of diverse sizes and clinical stages. By exploring alternative models, we show that intratumour diversity, in terms of methylation errors, stems more from tumour growth via gland fission than from cell turnover within glands. Moreover, the extent of intratumour diversity varies widely between patients, mainly because of eight-fold variation in gland fission rates but also due to differences in methylation and demethylation rates. Inter-gland divergence patterns are consistent with neutral evolution of colorectal tumours and a cancer stem cell fraction of approximately 1%. As well as helping to resolve the nature of colorectal cancer growth and evolution, our results provide proof of principle for a method that may be adapted to infer the biological parameters of other types of solid tumour. Author summary: Estimating how fast, for how long, and in what way a tumour has been growing inside someone’s body is challenging. We sought a solution by looking at chemical marks on DNA that are imperfectly copied every time a cell divides. We built a computer model to track how the patterns of marks diverge among the cells of a colorectal tumour as it grows. By running many simulations and comparing the outcomes to data from 10 human tumours, we were able to reconstruct how each of them had grown. We found that growth rate varies widely and is not linked to the size of the tumour when it is removed. The rates at which the DNA marks change also vary but to a lesser degree. Our approach provides a new, inexpensive way to reconstruct tumour growth histories from tissue samples collected during routine surgery, without having to take multiple measurements over time. This approach could be applied to other human cancer types.

We present PHYFUM, a novel Bayesian phylogenetic method for methylation data that reconstructs the evolutionary history of stem cells and the glandular structures they reside within in normal tissue. Using simulations, we validated this phylogenetic method and confirmed its accuracy. A re-analysis of 22 patients unveiled early gland divergence in the human gut, in contrast to a much later common ancestor in the endometrium, and yielded strong evidence against gland division by segregation of individual stem cells.

Cancer development and response to treatment are evolutionary processes^1,2, but characterizing evolutionary dynamics at a clinically meaningful scale has remained challenging³. Here we develop a new methodology called EVOFLUx, based on natural DNA methylation barcodes fluctuating over time⁴, that quantitatively infers evolutionary dynamics using only a bulk tumour methylation profile as input. We apply EVOFLUx to 1,976 well-characterized lymphoid cancer samples spanning a broad spectrum of diseases and show that initial tumour growth rate, malignancy age and epimutation rates vary by orders of magnitude across disease types. We measure that subclonal selection occurs only infrequently within bulk samples and detect occasional examples of multiple independent primary tumours. Clinically, we observe faster initial tumour growth in more aggressive disease subtypes, and that evolutionary histories are strong independent prognostic factors in two series of chronic lymphocytic leukaemia. Using EVOFLUx for phylogenetic analyses of aggressive Richter-transformed chronic lymphocytic leukaemia samples detected that the seed of the transformed clone existed decades before presentation. Orthogonal verification of EVOFLUx inferences is provided using additional genetic data, including long-read nanopore sequencing, and clinical variables. Collectively, we show how widely available, low-cost bulk DNA methylation data precisely measure cancer evolutionary dynamics, and provides new insights into cancer biology and clinical behaviour.

SUMMARY: The upper airway epithelium in humans is maintained in homeostasis by a resident population of basal stem cells. In the presence of tobacco smoke these gain mutations that significantly increase their risk of transformation to lung squamous cell carcinoma. Previous studies show that a small proportion of stem cells avoid the mutational damage caused by carcinogens in tobacco and are more abundant in the lungs of former smokers than ongoing smokers, indicating unexplained tissue-level genomic recovery. This mirrors epidemiological risk, which falls rapidly after quitting smoking. Somatic evolutionary mechanistic hypotheses have been proposed to explain these observations. Here, we present a computational framework to model each of these hypotheses within the upper airway epithelial stem cell population over the entire patient lifetimes of a cohort with diverse smoking histories. Applying a mechanistic learning approach based on a set of biologically informed metrics to single cell-derived whole-genome sequencing data, we identified subtle differences between epithelia modelled under different combinations of hypotheses. A slow-cycling subpopulation of stem cells, combined with suppression of immune predation of highly mutated stem cells while smoking, best matched observed data, a result converged upon by multiple distinct machine learning methodologies. Our findings, drawing on an evolutionary model of mutagen exposure at a whole-lifetime scale that is not feasible to model in vivo, reveal the mechanisms behind reduction in lung squamous cell carcinoma risk on cessation of smoking and inform future therapeutic interventions to prevent lung cancer initiation.

DNA methylation is emerging as a highly sensitive and specific marker of cancer initiation and progression. How these cancer-specific methylation changes are established in the decades before cancer, however, remains largely unknown. Here, we use a unique collection of longitudinal blood samples collected annually up to 15 years prior to a diagnosis of acute myeloid leukaemia to sensitively track the dynamics of DNA methylation changes at high temporal resolution. We identify thousands of differentially methylated regions (DMRs) that exhibit altered patterns of methylation up to 10 years before acute myeloid leukaemia (AML) diagnosis. Most of these DMRs are strongly associated with expanding clones carrying somatic driver mutations. We identify a subset of ‘epigenetic driver’ DMRs characterised by recurrent CpG alterations that are highly shared across pre-AML cases. These are likely to reflect early epigenetic reprogramming associated with AML development. We also reveal large numbers of stochastic ‘passenger’ CpGs whose differential methylation results from hitch-hiking with clonal expansions driven by somatically acquired genetic events. These passenger CpGs can be exploited for lineage tracing to discover clonal expansions driven by missing driver mutations. Our findings show widespread changes in methylation patterns during the early stages of cancer development which could be utilised for risk prediction and therapeutic intervention.

Cancer treatment frequently fails due to the evolution of drug-resistant cell phenotypes driven by genetic or non-genetic changes. The origin, timing, and rate of spread of these adaptations are critical for understanding drug resistance mechanisms but remain challenging to observe directly. We present a mathematical framework to infer drug resistance dynamics from genetic lineage tracing and population size data without direct measurement of resistance phenotypes. Simulation experiments demonstrate that the framework accurately recovers ground-truth evolutionary dynamics. Experimental evolution to 5-Fu chemotherapy in colorectal cancer cell lines SW620 and HCT116 validates the framework. In SW620 cells, a stable pre-existing resistant subpopulation was inferred, whereas in HCT116 cells, resistance emerged through phenotypic switching into a slow-growing resistant state with stochastic progression to full resistance. Functional assays, including scRNA-seq and scDNA-seq, validate these distinct evolutionary routes. This framework facilitates rapid characterisation of resistance mechanisms across diverse experimental settings.

Cancer development, progression, and response to treatment are evolutionary processes, but characterising the evolutionary dynamics at sufficient scale to be clinically-meaningful has remained challenging. Here, we develop a new methodology called EVOFLUx, based upon natural DNA methylation barcodes fluctuating over time, that quantitatively infers evolutionary dynamics using only a bulk tumour methylation profile as input. We apply EVOFLUx to 1,976 well-characterised lymphoid cancer samples spanning a broad spectrum of diseases and show that tumour growth rates, malignancy age and epimutation rates vary by orders of magnitude across disease types. We measure that subclonal selection occurs only infrequently within bulk samples and detect occasional examples of multiple independent primary tumours. Clinically, we observe that tumour growth rates are higher in more aggressive disease subtypes, and in two series of chronic lymphocytic leukaemia patients, evolutionary histories are independent prognostic factors. Phylogenetic analyses of longitudinal CLL samples using EVOFLUx detect the seeds of future Richter transformation many decades prior to presentation. We provide orthogonal verification of EVOFLUx inferences using additional genetic and clinical data. Collectively, we show how widely- available, low-cost bulk DNA methylation data precisely measures cancer evolutionary dynamics, and provides new insights into cancer biology and clinical behaviour.

Genetic and epigenetic variation, together with transcriptional plasticity, contribute to intratumour heterogeneity¹. The interplay of these biological processes and their respective contributions to tumour evolution remain unknown. Here we show that intratumour genetic ancestry only infrequently affects gene expression traits and subclonal evolution in colorectal cancer (CRC). Using spatially resolved paired whole-genome and transcriptome sequencing, we find that the majority of intratumour variation in gene expression is not strongly heritable but rather ‘plastic’. Somatic expression quantitative trait loci analysis identified a number of putative genetic controls of expression by cis-acting coding and non-coding mutations, the majority of which were clonal within a tumour, alongside frequent structural alterations. Consistently, computational inference on the spatial patterning of tumour phylogenies finds that a considerable proportion of CRCs did not show evidence of subclonal selection, with only a subset of putative genetic drivers associated with subclone expansions. Spatial intermixing of clones is common, with some tumours growing exponentially and others only at the periphery. Together, our data suggest that most genetic intratumour variation in CRC has no major phenotypic consequence and that transcriptional plasticity is, instead, widespread within a tumour.

The dynamical process of cell division that underpins homeostasis in the human body cannot be directly observed in vivo, but instead is measurable from the pattern of somatic genetic or epigenetic mutations that accrue in tissues over an individual's lifetime. Because somatic mutations are heritable, they serve as natural lineage tracing markers that delineate clonal expansions. Mathematical analysis of the distribution of somatic clone sizes gives a quantitative readout of the rates of cell birth, death, and replacement. In this review we explore the broad range of somatic mutation types that have been used for lineage tracing in human tissues, introduce the mathematical concepts used to infer dynamical information from these clone size data, and discuss the insights of this lineage tracing approach for our understanding of homeostasis and cancer development. We use the human colon as a particularly instructive exemplar tissue. There is a rich history of human somatic cell dynamics surreptitiously written into the cell genomes that is being uncovered by advances in sequencing and careful mathematical analysis lineage of tracing data. © 2022 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.