Introduction
Endoscopic surveillance of Barrett’s esophagus (BE) is recommended to identify dysplasia (low grade [LGD], high grade [HGD]) or adenocarcinoma [EAC]) for endoscopic intervention. However, this paradigm is limited by sampling error and subtle endoscopic changes, leading to missed HGD and EAC. We aimed to validate promising markers previously discovered by whole genome methylation sequencing for discriminating nondysplastic BE (NDBE) from HGD and EAC by simultaneously querying copy number aberration (CNA) and methylated DNA markers (MDMs).
Methods
DNA was extracted from prospectively collected esophageal brushing samples from 169 patients, from a multicenter study, independent from the prior discovery experiment. Patients had NDBE (42), LGD (23), HGD (26), and EAC (41), along with 37 non-BE squamous epithelia samples (NE). Endoscopic brushings were obtained from distal 5cm of the squamous epithelium and cardia in NE patients and from the visible BE mucosa in those with BE, with one brush used for each 5cm. Clinical and endoscopic data were abstracted, and histology was confirmed by a gastrointestinal pathologist. Custom DNA capture probes were prepared by IDT for 200 MDMs (selected from the prior discovery experiment). Targeted (MDM) and shallow whole genome (CNA) sequencing on enzyme converted DNA libraries was performed using the Illumina NextSeq platform. The number of MDMs was reduced by removing MDMs with low coverage and MDMs with high variability in the control group (NDBE). Further variable selection was performed using the VSURF package in R and the final model was generated using random forests comparing NDBE vs HGD/EAC. CNA burden was estimated with the trimmed median absolute deviation (tMAD) score from the ichorCNA package.
Results
Patient clinical data are summarized in Table 1. One LGD sample was excluded due to poor sequencing. Based on a cross-validated MDM selection, a model of four MDMs (Fig 1) achieved a cross-validated AUC of 0.87 (0.81-0.94, 95% CI) for the identification of HGD/EAC. Incorporating the tMAD score for CNA achieved a cross-validated AUC of 0.92 (0.87-0.98) for the identification of HGD/EAC. At a specificity threshold of 80%, the 4-MDM model identified 35 (85% [71-94%]) EAC, 23 (88% [70-98%]) HGD, 14 (63% [41-83%]) LGD, and 8 (19% [9-34%]) NDBE patients (observed specificity of 81%). The 4-MDM + tMAD model identified 38 (93% [80-98%]) EAC, 23 (88% [70-98%]) HGD, 11 (50% [28-72%]) LGD, and 8 (19% [9-34%]) NDBE patients (observed specificity of 81%).
Conclusion
In this validation experiment, CNA and methylated DNA markers (individually and in combination) show promising discrimination for the differentiation of BE-HGD/EAC from NDBE. Since both data types can be assayed from endoscopic brush specimens, they could be utilized for the molecular augmentation of histologic analysis in endoscopic surveillance.

Table: Baseline demographic characteristics of patients whose endoscopic brushing samples were included in the experiment. NE: normal esophagus, NDBE: non-dysplastic BE, LGD: Low grade dysplasia, HGD: High grade dysplasia, EAC: esophageal adenocarcinoma
Figure: Plot of smoothed curves for four regions included in model