ACT-Discover: identifying karyotype heterogeneity in pancreatic cancer evolution using ctDNA

Abstract

Background: Liquid biopsies and the dynamic tracking of somatic mutations within circulating tumour DNA (ctDNA) can provide insight into the dynamics of cancer evolution and the intra-tumour heterogeneity that fuels treatment resistance. However, identifying and tracking dynamic changes in somatic copy number alterations (SCNAs), which have been associated with poor outcome and metastasis, using ctDNA is challenging. Pancreatic adenocarcinoma is a disease which has been considered to harbour early punctuated events in its evolution, leading to an early fitness peak, with minimal further subclonal evolution.

Methods: To interrogate the role of SCNAs in pancreatic adenocarcinoma cancer evolution, we applied whole-exome sequencing of 55 longitudinal cell-free DNA (cfDNA) samples taken from 24 patients (including 8 from whom a patient-derived xenograft (PDX) was derived) with metastatic disease prospectively recruited into a clinical trial. We developed a method, Aneuploidy in Circulating Tumour DNA (ACT-Discover), that leverages haplotype phasing of paired tumour biopsies or PDXs to identify SCNAs in cfDNA with greater sensitivity.

Results: SCNAs were observed within 28 of 47 evaluable cfDNA samples. Of these events, 30% could only be identified by harnessing the haplotype-aware approach leveraged in ACT-Discover. The exceptional purity of PDX tumours enabled near-complete phasing of genomic regions in allelic imbalance, highlighting an important auxiliary function of PDXs. Finally, although the classical model of pancreatic cancer evolution emphasises the importance of early, homogenous somatic events as a key requirement for cancer development, ACT-Discover identified substantial heterogeneity of SCNAs, including parallel focal and arm-level events, affecting different parental alleles within individual tumours. Indeed, ongoing acquisition of SCNAs was identified within tumours throughout the disease course, including within an untreated metastatic tumour.

Conclusions: This work demonstrates the power of haplotype phasing to study genomic variation in cfDNA samples and reveals undiscovered intra-tumour heterogeneity with important scientific and clinical implications. Implementation of ACT-Discover could lead to important insights from existing cohorts or underpin future prospective studies seeking to characterise the landscape of tumour evolution through liquid biopsy.

Trial registration: ClinicalTrials.gov NCT02795650.

Keywords: Cell-free DNA; Circulating tumour DNA; Copy number; Intra-tumour heterogeneity; Pancreatic cancer; Tumour evolution; cfDNA; ctDNA.

Fig. 1

Cohort overview. Cohort of patients with metastatic pancreatic cancer. The date of metastatic biopsy relative to the time of diagnosis, systemic anti-cancer therapy, liquid biopsies and death are annotated. ctDNA - circulating tumour DNA; VAF - variant allele frequency

Fig. 2

Genomic characteristics. The top panel shows the mutation load for each tumour, split into clonal mutations that are found in all samples, and subclonal mutations found only in a subset. The second panel shows the fraction of each admixed sample estimated to be derived from tumour (purity), split by whether the sample was derived from a tissue sample of primary tumour (n = 1), metastasis (n = 16), patient-derived xenograft (n = 8), or cfDNA (n = 6). Only samples with an estimated tumour purity > 5% are shown. The third panel shows the different genomic events for a set of cancer genes of interest. Mutations are represented as tiles, and somatic copy number aberrations (SCNAs) are overlaid as triangles. The colour of the tiles and triangles represents the timing of the mutations and copy number events, respectively. The final panel shows the proportion of mutations attributed to different mutational signatures for each patient. The five patients on the right of the figure, which are greyed out, do not have sufficient mutations (less than 50) to perform reliable signature deconvolution

Fig. 3

Tumour-informed analysis of somatic mutations and copy number alterations. A Simulated data testing the sensitivity of ACT-Discover at tumour content between 0 and 10% with differing numbers of simulated SNPs per segment. At a simulated purity of 5–10%, segments containing only 50 SNPs could be rescued using ACT-Discover. At a simulated purity of 0.1%, only segments containing 50,000 SNPs or more could be rescued. B Downsampling approach testing the sensitivity of ACT-Discover in the context of reduced effective tumour content. At simulated purity of ~ 10% (i.e. 10% PDX content), 97% of this allelic imbalance remained detectable. Even at 1% effective purity, 20% of allelic imbalance was detectable, and at 0.1% effective purity, it was still possible to detect allelic imbalance, albeit only 5%. C Number of mutations identified within each sample, coloured by whether they were identified using de novo somatic mutation calling or whether they required a tumour-informed approach, and the number of copy number alterations identified within each sample, coloured by whether they were identified using de-novo copy number calling or whether they required a haplotype-informed approach. D Copy number profiles of metastasis and three cfDNA samples of patient PANVH2. For each segment, a sample was selected that had the greatest absolute difference of BAF values to 0.5. The SNPs were then coloured based on whether they were greater than 0.5 (orange) or less or equal than 0.5 (purple). This colouring was then used for the SNPs in the segment in all other samples. It was not possible to phase SNPs coloured in grey. Segments where in at least one sample SNPs coloured in purple have BAF values greater than 0.5 can be classified as having mirrored subclonal allelic imbalance. Examples of this can be seen on chromosome 2q and 7p

Fig. 4

Heterogeneity of somatic copy number alterations in an untreated cancer. A Timeline of patient PAN113 showing sample collection and treatment. B Phylogenetic tree showing mutation clusters arising prior to, and after, systemic anti-cancer treatment. A putative driver mutation in CREBBP arises during treatment. C Heterogeneity of somatic copy number alterations (SCNAs). Events occur in parallel, potentially indicative of convergent evolution, or are heterogeneous between a metastatic sample and cfDNA extracted at the same time point. Ongoing heterogeneity is detected over time in subsequent cfDNA samples. D Copy number profiles of metastasis and three cfDNA samples of patient PAN113