Poster viewing and lunch

34P - Investigating morphological heterogeneity in luminal breast cancer integrating artificial intelligence and spatial transcriptomics (ID 254)

Lecture Time
12:15 - 12:15
Session Name
Poster viewing and lunch
Exhibition area
Fri, 12.05.2023
12:15 - 13:00
  • Nicola Occelli (Brussels, Belgium)
  • Nicola Occelli (Brussels, Belgium)
  • Matteo Serra (Brussels, Belgium)
  • Mattia Rediti (Brussels, Belgium)
  • Laetitia Collet (Brussels, Belgium)
  • Frédéric Lifrange (Liège, Belgium)
  • Xiaoxiao Wang (Brussels, Belgium)
  • Delphine Vincent (Brussels, Belgium)
  • Ghizlane Rouas (Brussels, Belgium)
  • Ligia Craciun (Brussels, Belgium)
  • Denis Larsimont (Brussels, Belgium)
  • David Venet (Brussels, Belgium)
  • Miikka Vikkula (Woluwe-Saint-Lambert, Belgium)
  • Francois P. Duhoux (Brussels, Belgium)
  • Laurence Buisseret (Brussels, Belgium)
  • Françoise Rothé (Brussels, Belgium)
  • Christos Sotiriou (Brussels, Belgium)



Hormone receptor-positive (HR+), HER2-negative (HER2-) breast cancer (BC) accounts for around 65% of all BCs. Invasive lobular and ductal carcinoma (ILC, IDC) show distinct histology and clinical presentation. In this study, our goal is to exploit morphological differences between IDC and ILC by combining artificial intelligence and spatial transcriptomics (ST) to characterize intra-tumor heterogeneity.


We analyzed 131 H&E whole slide images (WSIs) from frozen HR+, HER2- BC samples, of which 43 were ILC and 88 were IDC. Images were morphologically annotated using QuPath. We performed ST (Visium 10X Genomics®) on the ILC samples. A neural network (NN) was trained to perform histology classification from WSIs and detect the most relevant tissue regions for such classification. Gene expression data from ST were used to characterize these regions.


The NN achieved 0.95 ROC AUC in predicting histology (ILC vs IDC). Interestingly, in 36/43 ILC samples, adipose tissue had the highest relative importance in assessing the histological subtype, suggesting crucial morphological differences in adipocytes between ILC and IDC. Of note, we observed intra-sample heterogeneity in the importance levels of tumor areas, with just 13% of the overall tumor cells showing high importance in the classification. We mapped the most relevant tissue regions for histology classification to the ST spots (for ILC). Pathway enrichment analysis on differentially expressed genes (DEG) relative to these spots revealed enrichment in metabolic and adipogenesis-related pathways (padj < 0.05). When limiting the analysis on spots composed by more than 30% of tumor cells, DEG revealed enrichment in metabolic-related pathways (padj < 0.05).


Adipose tissue morphology was revealed to be a key feature in distinguishing histological subtypes in HR+, HER2- BC. Importantly, tumor cells with increased metabolism showed to be crucial in the histological classification, suggesting differences in metabolism between IDC and ILC. Further validation is needed.

Legal entity responsible for the study

The authors.


FNRS, Fondation Jules Bordet, Breast Cancer Research Foundation, Fondation contre le Cancer.


F.P. Duhoux: Financial Interests, Institutional, Advisory Board: Roche, Pfizer, AstraZeneca, Lilly, Novartis, Amgen, Daiichi Sankyo, Pierre Fabre, Gilead, Seagen, MSD; Financial Interests, Institutional, Invited Speaker: Novartis, Pfizer, MSD, Roche, MSD, Boehringer Ingelheim, Pfizer, Novartis, Lilly, AbbVie, Seagen, Gilead, AstraZeneca, Menarini, Immutep; Financial Interests, Institutional, Expert Testimony: Seagen, Novartis, MSD. C. Sotiriou: Financial Interests, Institutional, Advisory Board: Astellas, Vertex, Seattle Genetics, Amgen, INC, Merck & Co; Financial Interests, Personal, Advisory Board: Cepheid, Puma; Financial Interests, Personal, Invited Speaker: Eisai, Prime oncology, Teva; Financial Interests, Institutional, Other, Travel: Roche; Financial Interests, Institutional, Other, Internal speaker: Genentech; Financial Interests, Personal, Other, Regional speaker: Pfizer; Financial Interests, Institutional, Invited Speaker: Exact Sciences. All other authors have declared no conflicts of interest.