Introduction

Oesophageal adenocarcinoma (OAC) remains a poor prognosis cancer type. The pre-malignant lesion for this type of cancer is called Barrett’s Oesophagus (BO), which has two subtypes, Gastric Metaplasia (GM) and Intestinal Metaplasia with GM (IM). Both BO and OAC are genetically heterogeneous lesions and little is known about the genetic relationship between OAC and pre-cancerous BO, as well as the clinical significance of GM which is not considered to be BO in the US guidelines. Furthermore, the considerable intra-tumour heterogeneity (ITH) may underlie the observed chemo-resistance. There have not been any large-scale studies to assess ITH in OAC and associated BO. Aside from multi-region sequencing, another way of overcoming sampling bias of the heterogeneous tissue is through circulating tumour DNA (ctDNA). Liquid biopsy sampling also has been proven to be highly prognostic in multiple cancer types and recent advances in ctDNA detection have now provided a potential avenue for this sampling approach to utilised in OAC.

Hypothesis

Deciphering the tissue heterogeneity of OAC and associated BO will help determine the evolutionary relationship between these tissue types with clinical relevance. Blood biopsy may provide a method to overcome tissue heterogeneity and inform clinical management in OAC.

Aims

• Characterise the intra tumour heterogeneity in OAC • Better understand the evolution of OAC, by using phylogenetic analysis of BO and OAC • Compare the genetic architecture of Barrett’s IM and GM • Evaluate the clinical potential of ctDNA detection for minimal residual disease using two different strategies, namely cancer panel versus personalised assay.

Materials and Methods

Multi-region whole exome sequencing was performed on chemotherapy-naïve 398 samples from 79 patients with OAC. The samples were manually micro-dissected and then sequenced to a mean depth of 150x. Clustering was performed using PyClone and tree building by CloneEvol. Mutational signatures were identified using deconstructSigs.

The gene panel ctDNA analysis included 245 double-spun plasma samples and 78 peripheral blood samples from 97 patients. The samples were prepared using the Roche expanded Avenio ctDNA panel. The personalised panel included 53 double-spun plasma samples collected from 20 patients and processed using the Signatera assay.

Results

Multi-region sequencing revealed considerable ITH in OAC, with the majority of mutations being subclonal. Moreover, SMAD4 which has been previously shown to be a biomarker for poor prognosis, was more often subclonal. In addition, the majority of the BO mutations are also subclonal, indicating substantial ITH in this premalignant lesion. As expected, BO had less driver mutations than OAC, but the driver mutations in BO were more likely to be subclonal. Comparison between IM and GM, demonstrated they had distinct mutational signature profiles and GM was significantly less mutated than IM. Interestingly, the variant allele frequency (VAF) of the tumour and IM mutations was similar, however the VAF of the GM mutations was significantly lower than the tumour and IM mutations, which was surprising given that GM samples generally had a higher cellularity than IM samples. Also, driver gene mutations in BO were not always shared with the tumour, and in some cases there were no shared clones between the BO and tumour samples.

Both the gene panel and personalised panel demonstrated ctDNA to be prognostic for disease free and cancer-specific survival in the post-surgery setting despite the ITH. Using the Roche gene panel assay that is comprised of 77 cancer genes, 14 of which are OAC driver genes, TP53 was the most frequently mutated gene (15%), followed by APC (8%), then KRAS (6%). The sequencing of peripheral blood cell samples (available for 80%) revealed that 23% of cases had at least one variant derived from Clonal Haematopoiesis of Indeterminate Potential (CHIP) that could confound the analysis. Once CHIP was removed, median cancer-specific survival in the post-surgical setting for ctDNA positive patients was 10 months compared to 30 months for ctDNA negative patients (hazard ratio 5.55, 95% CI: 2.42-12.71, P = 0.0003). In order to overcome the diverse genetic landscape and the low ctDNA shedding in OAC, a bespoke panel based on the WGS of the primary may be advantageous. The pilot evaluation (n=20) of a personalised panel was weakly prognostic for disease free survival (p= 0.042) and could detect minimal-residual disease down to 0.01%. Moreover, this method gave an average lead time of nearly one year.

Discussion

Multi-region analysis revealed that OAC is a highly heterogenous cancer type, which could explain why OAC is frequently resistant to chemotherapy. In addition, the results from this study suggest that BO is also heterogeneous. Furthermore, GM and IM are quite distinct in terms of their evolution, with differences in their mutational signatures, mutation burden, and mutation VAF providing biological rationale for the lower cancer conversion rate that has been observed for GM. This finding may substantiate the clinical decision not to include GM cases in Barrett’s diagnosis and surveillance programmes. Based on these results, an alternative approach to take could be to design surveillance programmes that are further stratified to distinguish between patients with GM and IM when the BO segment exceeds 3 cm.

The results from the plasma analysis demonstrates that ctDNA is prognostic post-operatively in OAC. Moreover, the technology available is highly sensitive, indicating the potential clinical application of liquid biopsy sampling. However, if a gene panel approach is used, a peripheral blood cell sample should also be sequenced to eliminate variants derived from clonal haematopoiesis and this is suboptimal given the heterogeneous nature of the driver gene landscape. A personalised approach has promise for detection of minimal residual disease that should be further evaluated in larger prospective trials.