Abstract P5-14-08: Machine learning based histopathology images analysis reveals cancer stemness in TNBC patient with 17p loss

Abstract

Abstract Background: The area of computational pathology has made huge progresses due to advances in artificial intelligence (AI) and machine learning technologies. It has been applied to many research and translational tasks which provide great improvement on medical diagnosis and treatment. Cancer stem-like cells (CSC) have been consistently reported for its key role in Triple negative breast cancer (TNBC) . Given the large amount of existing H&E stained histological slides of TNBC, digital identification of CSC could benefit the evaluation of tumor status and prediction of patients’ response to chemotherapy. Here we proposed an AI framework based on Convolutional Neural Network (CNN) to predict CSC from the histological images of TNBC patient. And our preliminary work suggested that chromosome 17p loss, a common genetic variations in breast cancer, is linked to cancer stemness. Methods: A modified GoogleNet model was adopted as our CNN classifier. Consecutive breast cancer tissue microarrays (TMA), which stained with H&E ,SOX2, OCT4 and NANOG antibodies respectively by IHC, were used as training dataset for the CNN model. Gene expression data from the TCGA and METABRIC datasets were used to identify gene signatures associated with CSC. The connectivity map (CMAP) and Cancer Cell Line Encyclopedia (CCLE) were used for screening compounds that target stemness in cancer cell lines with chromosomal 17p loss. HS578T and EO771 cells with or without heterozygous 17p loss (11b in EO771) were used for in vitro experiments. Female immunodeficient nude (Nu/J) mice were used for animal studies. Results: The well trained GoogleNet model was applied to TNBC patient diagnosis images in TCGA BRCA dataset. By analyzing patient genomic alteration on chromosomal level, we found that loss of chromosome 17p associate with high cancer stemness in TNBC. Flow cytometry assays also demonstrated higher ALDH1 activity and higher CD44+/CD24−/low cell population in HS578T cells with 17p loss. RNA-seq of HS578T cells revealed that most CSC marker genes were located in the unregulated differentially expressed genes (DEGs) of 17p loss cells. We next compared the cytotoxicity of chemotherapy drugs including doxorubicin, paclitaxel, docetaxel and cisplatin on 17p loss and 17p intact HS578T cells, 11b loss and 11b intact EO771 cells, in terms of IC50 value. The IC50 value of indicated drug on 17p loss HS578T cells with were 3-6 fold higher than their IC50 on 17p intact HS578T cells. Similar result was observed in EO771 cells. Next, 17p loss and 17p intact HS578T cells were orthotopically implanted into the Nu/J mice. Under the doxorubicin treatment, mice bearing 17p loss HS578T derived tumors had larger and heavier tumors in compare to mice bearing 17p intact tumors. Next, we did a computational drug screening to identify compounds that can target the cancer stemness in 17p loss cells. Screened out compounds were tested for their cytotoxicity on 17p loss and 17p intact HS578T cells and FK866 showed the most pronounced efficacy on inhibiting the viability of 17p loss cells, compared to 17p intact cells. Followup experiments demonstrated that FK866 can decrease the CSC features induced by doxorubicin, both in vitro and in vivo. FK866 also potentiates the effect of doxorubicin on treating TNBC cells with 17p loss, which provide a drug combination potential for TNBC patient with 17p loss. Conclusions: A CNN based model was developed to identify CSC from TNBC histopathology images. The images analysis combined with patient genomic data revealed that chromosome 17p loss associated with cancer stemness in TNBC. This result was confirmed using assays on TNBC cells with or without 17p loss. A computational drug screening was performed to identify candidates that targeting stemness in 17p loss cells. FK866 was identified and it potentiates the anti-tumor effect of doxorubicin on treating TNBC cells with 17p loss. Our study provides a novel strategy on applying AI to precision treatment for cancers. Citation Format: tianhan dong, jiannan liu, yuanzhang Fang, Ziyu Liu, Xiongbin Lu, kun huang. Machine learning based histopathology images analysis reveals cancer stemness in TNBC patient with 17p loss [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P5-14-08.