Transition to invasive breast cancer is associated with progressive changes in the structure and composition of tumor stroma

Abstract

Ductal carcinoma in situ (DCIS) is a pre-invasive lesion that is thought to be a precursor to invasive breast cancer (IBC). To understand how the tumor microenvironment (TME) changes with transition to IBC, we used Multiplexed Ion Beam Imaging by time of flight (MIBI-TOF) and a 37-plex antibody staining panel to analyze 140 clinically annotated surgical resections covering the full spectrum of breast cancer progression. We compared normal, DCIS, and IBC tissues using machine learning tools for multiplexed cell segmentation, pixel-based clustering, and object morphometrics. Transition from DCIS to IBC was found to occur along a trajectory marked by coordinated shifts in location and function of myoepithelium, fibroblasts, and infiltrating immune cells in the surrounding stroma. Taken together, this comprehensive study within the HTAN Breast PreCancer Atlas offers insight into the etiologies of DCIS, its transition to IBC, and emphasizes the importance of the TME stroma in promoting these processes.

Multiplexed Imaging Analysis of the Tumor-Immune Microenvironment Reveals Predictors of Outcome in Triple-Negative Breast Cancer

“…25 Patient data regarding age, tumor grade, stage, cancer site, and clinical outcome -recurrence and overall survival (OS) -were also gathered (Table 1). We additionally gathered MIBI images of breast tissue of 8 healthy patients, a subset of the patients examined by Risom et al 34…”

“…MIBI scans produce images of protein expression from FFPE tissue, where each image has 44 channels; each channel conveys the expression of a certain marker on the tissue sample (Figure 1a). Cellular segmentations for both TNBC and healthy patients' images were provided by Keren et al and Risom et al, who utilized DeepCell, a deep learning technique for identifying individual cells from MIBI data 25,34,35 . Cell type assignment for TNBC patients' images was also performed by Keren et al through a hierarchical methodology (Figure 1b) (Methods).…”

Whole-cell segmentation of tissue images with human-level performance using large-scale data annotation and deep learning

“…This diversity of data allows models trained on TissueNet to handle data from many different experimental setups and biological questions. The images included in TissueNet were acquired from the published and unpublished works of labs who routinely perform tissue imaging [44][45][46][47][48][49][50][51] . Thus, while this first release of TissueNet encompasses the tissue types most commonly analyzed by the community, we expect that subsequent versions of TissueNet will be expanded to include less-studied organs.…”

“…To construct TissueNet, we collected published and unpublished data from numerous tissue imaging labs [44][45][46][47][48][49][50][51] . Each dataset was manually inspected to identify images suitable for model training.…”

“…For example, nuclear translocation of transcription factors like NF-κB and IRF3 is informative for understanding signaling dynamics in immune cells 63,64 , while the degree of membrane staining of HER2 is an important feature in the accurate grading and assessment of breast cancer 65 . To explore the accuracy of subcellular signal prediction, we stained breast cancer samples with a panel of phenotyping markers and imaged them with MIBI-TOF 51 (Methods; Figure 4a ). We created an integrated multiplexed image analysis pipeline, ark-analysis 66 , that integrates Mesmer’s segmentation predictions with downstream analysis.…”