Document resource
Introduction Early detection of oesophageal adenocarcinoma(EAC) will facilitate preventive treatment for a major public health priority. There are efforts to devise better non-invasive tests for the precursor lesion of EAC defined by metaplastic transformation from squamous oesophageal epithelium to intestinal metaplasia known as Barrett’s oesophagus(BE). This study aimed to devise a quantitative, higher-throughput diagnostic using the acellular fluid residue from the capsule sponge.Methods We selected biomarker candidates from existing bulk and single cell transcriptomic data based on their differential expression in non-dysplastic BE compared to normal squamous and gastric tissue, and subsequent validation at protein level with immunohistochemical staining. Shortlisted proteins were quantified from capsule sponge supernatants using ELISAs. We trained a machine learning algorithm(lightGBM) on secretory protein data from a retrospective(N=161) and prospective(N=130) training cohort and independently tested it in an external prospective cohort (N=132). We then performed an external validation using ultra-sensitive single molecule arrays(SIMOA) for the strongest two markers.Results Transcriptomic data identified a goblet cell-of-origin for BE markers TFF3, SPINK4, FABP2, and REG4, while MEP1A was expressed by foveolar cells and CXCL8 by lymphocytes in BE epithelium. We established a diagnostic signature using sponge supernatant circulating expression of six-proteins which achieved an AUC of 0.87 when tested in an independent prospective cohort. This rapid assay could identify BE among symptomatic reflux patients with a positive predictive value of 95.6%, 71% sensitivity and 97% specificity. Employment of ultra-sensitive SIMOA for REG4 and FABP2 increased AUC to 0.925, 84% sensitivity and 92% specificity ( figure 1).Conclusion We have developed a novel quantitative approach to make the capsule sponge a higher-throughput diagnostic test for BE.Abstract P132 Figure 1ROC curves for predictive BE model obtained with traditional ELISA compared to ultra-sensitive SiMoA