Abstract P1-05-03: Integrating spatial transcriptomics and high-resolution morphological annotation to investigate tumor heterogeneity and PAM50 molecular subtyping in lobular breast cancer

Abstract

Abstract Background: Invasive lobular breast carcinoma (ILC) represents 5 to 15% of all invasive breast cancers (BCs). Here, we aim to investigate inter- and intra-tumor heterogeneity in terms of microenvironment composition, PAM50 molecular classification and proliferation (genomic grade index [GGI]) by combining spatial transcriptomics (ST) and accurate morphological annotation. Methods: Spatial RNA sequencing (Visium - 10X Genomics) was performed on frozen tumor samples from 15 primary estrogen receptor positive, HER2-negative ILC patients with long-term follow up. Hematoxylin/eosin slides were morphologically annotated integrating manual and machine learning-based approaches reaching single-cell resolution (QuPath software). The relative histomorphological categories (HC) composition of each spot across the ST slide was computed as percentage of pixels, while the level of proximity of different HC was evaluated computing the proportion of co-occurring HC at each spot. The PAM50 subtypes (AIMS R package) and GGI were computed on spots containing at least 40% of tumor (merging all the spots belonging to each sample [pseudo-bulk] for PAM50, while calculating mean and standard deviation [SD] across spots for GGI). PAM50 was also computed on the pseudo-bulk of the whole set of spots per sample. Wilcoxon and Spearman rank tests were used to compare continuous variables and assess correlations. Results: Out of 15 tumors, 7 were T2 or T3, 6 were node positive at diagnosis and 14 were grade 2. Four patients experienced disease relapse. Morphological annotation revealed that an average (per patient) of 20.3% (5.6-46.7%), 65.9% (45.5-83.5%) and 6.5% (0.0-27.1%) of the spots corresponded to tumor, stroma and fat tissue respectively. Larger tumors (T2-3 vs T1) presented a higher proportion of fat tissue and tumor cells, although these differences did not reach statistical significance. The levels and spatial variability of proliferation, measured using GGI, were higher in T2-3 compared to T1 tumors (p = 0.072 and 0.040, respectively). Of note, higher spatial variability of proliferation was also associated with node-positive tumors (p = 0.066). By computing the PAM50 classification using the pseudo-bulk, 9 samples were classified as luminal A, 1 as luminal B and 5 as normal-like. Of interest, when focusing on the tumor enriched spots, 60% of the samples previously classified as normal-like were re-classified as luminal A. Samples from patients who relapsed showed a higher fraction of fat tissue at the level of the whole slide (14.4% vs 3.5%; p = 0.018), with an increased co-localization of fat tissue and tumor cells at the spot level as well as higher proliferation values, although not significant. Conclusions: High proportion of fat tissue together with higher co-localization of fat tissue with tumor cells are associated with poor outcome in ILC. Higher spatial variability of proliferation is associated with larger tumors, lymph node positivity and recurrence. The proportion of stroma and fat tissue affected substantially the PAM50 classification of the tumors. Further validation is needed. Citation Format: Matteo Serra, Laetitia Collet, Mattia Rediti, Frédéric Lifrange, David Venet, Xiaoxiao Wang, Delphine Vincent, Ghizlane Rouas, Danai Fimereli, David Gacquer, Andrea Joaquin Garcia, Isabelle Veys, Ligia Craciun, Denis Larsimont, Miikka Vikkula, François Duhoux, Françoise Rothé, Christos Sotiriou. Integrating spatial transcriptomics and high-resolution morphological annotation to investigate tumor heterogeneity and PAM50 molecular subtyping in lobular breast cancer [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P1-05-03.