AI-assisted discovery of an ethnicity-influenced driver of cell transformation in esophageal and gastroesophageal junction adenocarcinomas
Pradipta Ghosh, … , Kit Curtius, Debashis Sahoo
Pradipta Ghosh, … , Kit Curtius, Debashis Sahoo
Published September 22, 2022
Citation Information: JCI Insight. 2022;7(18):e161334. https://doi.org/10.1172/jci.insight.161334.
View: Text | PDF
Research Article Gastroenterology Immunology

AI-assisted discovery of an ethnicity-influenced driver of cell transformation in esophageal and gastroesophageal junction adenocarcinomas

Abstract

Although Barrett’s metaplasia of the esophagus (BE) is the only known precursor lesion to esophageal adenocarcinomas (EACs), drivers of cellular transformation in BE remain incompletely understood. We use an artificial intelligence–guided network approach to study EAC initiation and progression. Key predictions are subsequently validated in a human organoid model, in patient-derived biopsy specimens of BE, a case-control study of genomics of BE progression, and in a cross-sectional study of 113 patients with BE and EACs. Our model classified healthy esophagus from BE and BE from EACs in several publicly available gene expression data sets (n = 932 samples). The model confirmed that all EACs must originate from BE and pinpointed a CXCL8/IL8↔neutrophil immune microenvironment as a driver of cellular transformation in EACs and gastroesophageal junction adenocarcinomas. This driver is prominent in White individuals but is notably absent in African Americans (AAs). Network-derived gene signatures, independent signatures of neutrophil processes, CXCL8/IL8 expression, and an absolute neutrophil count (ANC) are associated with risk of progression. SNPs associated with changes in ANC by ethnicity (e.g., benign ethnic neutropenia [BEN]) modify that risk. Findings define a racially influenced immunological basis for cell transformation and suggest that BEN in AAs may be a deterrent to BE→EAC progression.

Authors

Pradipta Ghosh, Vinicius J. Campos, Daniella T. Vo, Caitlin Guccione, Vanae Goheen-Holland, Courtney Tindle, Guilherme S. Mazzini, Yudou He, Ludmil B. Alexandrov, Scott M. Lippman, Richard R. Gurski, Soumita Das, Rena Yadlapati, Kit Curtius, Debashis Sahoo

×

Figure 3

Generation and validation of Boolean network map of BE to EAC progression.

Options: View larger image (or click on image) Download as PowerPoint
Generation and validation of Boolean network map of BE to EAC progressio...
(A) Schematic outline the workflow and training data sets used to create a Boolean map of BE to EAC transition using BoNE (4). (B) Graph showing invariant patterns of gene expression changes during BE→EAC transformation. Gene clusters identified by machine learning are indicated in bold. (C) Bar plots show sample classification accuracy across diverse data sets, with corresponding ROC-AUC values. The sample numbers for BE and EAC analyzed in each data set are annotated on the left margin. (D) Violin plots show the composite scores of upregulated gene clusters in NE, NDBE, DBE, and EACs. P values indicate comparison of each sample type against the NE, as determined by Welch’s t test. (E) Violin plots show the composite scores of upregulated gene clusters in NE, normal gastric (NG), normal GEJ (GEJ) and GEJ-ACs. P values indicate comparison of each sample type against the NE (left) or GEJ (right), as determined by Welch’s t test. (F) Pathway analyses of gene clusters derived from the map in B. Red type indicates likely epithelial processes. DBE-LG, dysplastic BE, low-grade dysplasia; LBP, ligand-binding protein; Lig-binding receptor, ligand-binding receptor; MMP, Matrix metalloproteinase; WASP, Wiskott–Aldrich syndrome protein; WAVE, WASP-family verprolin-homologous protein.