2328P - Alternative molecular mechanisms underpinning breast invasive lobular carcinoma identified by genomics-driven artificial intelligence model

Background

A subset of cancers display distinctive phenotypes and harbor pathognomonic genetic alterations. Breast invasive lobular carcinoma (ILC), the most common special histologic subtype of breast cancer (BC), has a distinctive discohesive phenotype, caused by bi-allelic inactivation of CDH1 (E-cadherin), representing a strong genotypic-phenotypic correlation. Here, we sought to apply an artificial intelligence (AI)-based model trained to detect CDH1 bi-allelic mutations (i.e. inactivating mutation and loss of heterozygosity (LOH)) using H&E-stained whole slides images (WSI) as input to identify ILCs with alternative mechanisms of CDH1 inactivation.

Methods

We applied the AI-model trained to detect CDH1 biallelic mutations using WSIs to a cohort of 1,057 BCs with available targeted sequencing data. To determine the molecular underpinning of cases predicted to harbor CDH1 bi-allelic mutations but lacking these genetic alterations by targeted sequencing, we conducted the re-analysis of targeted sequencing data, CDH1 gene promoter methylation assessment and/or whole genome sequencing (WGS) analysis.

Results

We identified 34 cases predicted by the AI-model to harbor a CDH1 bi-allelic mutations that lacked these alterations by targeted sequencing. Reanalysis of targeted sequencing data and/or promoter methylation assessment revealed CDH1 promoter methylation (n=20), CDH1 homozygous deletions (n=3) and CDH1 intragenic deletion associated with LOH (n=1) in these cases. In addition, WGS of an ILC revealed a translocation t(13;16) predicted to result in a novel deleterious fusion gene affecting CDH1 with loss of exons 1-2 and the regulatory regions of this gene, associated with LOH. By applying a genomics-driven AI model we detected alternative mechanisms of CDH1 inactivation in 25/34 (74%) cases.

Conclusions

In the context of genotypic-phenotypic correlations, AI-based methods trained on a genetic ground truth can result in the identification of alternative/convergent molecular mechanisms underpinning histologic entities. This study highlights the potential of combinatorial approaches integrating deep learning methods and genomics in pathology.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Breast Cancer Research Foundation, National Institutes of Health (NIH)/National Cancer Institute.

Disclosure

Y.K. Wang, J. Bernhard, J. Sue, M.C.H. Lee, R. Godrich, A. Casson, J.D. Kunz, B. Rothrock, G.J. Oakley, D. Klimstra, T. Fuchs: Financial Interests, Personal, Stocks/Shares: Paige; Financial Interests, Personal, Full or part-time Employment: Paige. B. Weigelt: Financial Interests, Personal, Research Funding: Repare Therapeutics. C. Kanan: Financial Interests, Personal, Stocks/Shares, Also Consultant: Paige. J.S. Reis-Filho: Financial Interests, Personal, Other, Consultant: Goldman Sachs, Eli Lilly, Saga Diagnostics; Financial Interests, Personal, Other, Member of the Scientific Advisory Board and Consultant: Repare Therapeutics, Paige.AI; Financial Interests, Personal, Advisory Board: Personalis, Roche Tissue Diagnostics; Financial Interests, Personal, Advisory Board, Member of the Scientific Advisory Board: Bain Capital; Financial Interests, Personal, Advisory Board, Ad hoc member of the Pathology Scientific Advisory Board: Daiichi Sankyo, Merck; Financial Interests, Personal, Advisory Board, Ad hoc member of the Oncology Scientific Advisory Board: AstraZeneca; Financial Interests, Personal, Advisory Board, Member of the SAB: MultiplexDX; Financial Interests, Personal, Member of Board of Directors: Odyssey Bio, Grupo Oncoclinicas; Financial Interests, Personal, Stocks/Shares: Repare Therapeutics; Financial Interests, Personal, Other, Stock options: Paige.AI. All other authors have declared no conflicts of interest.