Computational pathology aids derivation of microRNA biomarker signals from Cytosponge samples

Abstract

Background: Non-endoscopic cell collection devices combined with biomarkers can detect Barrett's intestinal metaplasia and early oesophageal cancer. However, assays performed on multi-cellular samples lose information about the cell source of the biomarker signal. This cross-sectional study examines whether a bespoke artificial intelligence-based computational pathology tool could ascertain the cellular origin of microRNA biomarkers, to inform interpretation of the disease pathology, and confirm biomarker validity.

Methods: The microRNA expression profiles of 110 targets were assessed with a custom multiplexed panel in a cohort of 117 individuals with reflux that took a Cytosponge test. A computational pathology tool quantified the amount of columnar epithelium present in pathology slides, and results were correlated with microRNA signals. An independent cohort of 139 Cytosponges, each from an individual patient, was used to validate the findings via qPCR.

Findings: Seventeen microRNAs are upregulated in BE compared to healthy squamous epithelia, of which 13 remain upregulated in dysplasia. A pathway enrichment analysis confirmed association to neoplastic and cell cycle regulation processes. Ten microRNAs positively correlated with columnar epithelium content, with miRNA-192-5p and -194-5p accurately detecting the presence of gastric cells (AUC 0.97 and 0.95). In contrast, miR-196a-5p is confirmed as a specific BE marker.

Interpretation: Computational pathology tools aid accurate cellular attribution of molecular signals. This innovative design with multiplex microRNA coupled with artificial intelligence has led to discovery of a quality control metric suitable for large scale application of the Cytosponge. Similar approaches could aid optimal interpretation of biomarkers for clinical use.

Funding: Funded by the NIHR Cambridge Biomedical Research Centre, the Medical Research Council, the Rosetrees and Stoneygate Trusts, and CRUK core grants.

Keywords: Artificial intelligence; Barrett's oesophagus; Computerized image analysis; Dysplasia; Oesophageal cancer; Screening.