Independent Prognostic Value of Intratumoral Heterogeneity and Immune Response Features by Automated Digital Immunohistochemistry Analysis in Early Hormone Receptor-Positive Breast Carcinoma

Immunohistochemistry (IHC) for ER, PR, HER2, and Ki67 is used to predict outcome and therapy response in breast cancer patients. The current IHC assessment, visual or digital, is based mostly on global biomarker expression levels in the tissue sample. In our study, we explored the prognostic value of digital image analysis of conventional breast cancer IHC biomarkers supplemented with their intratumoral heterogeneity and tissue immune response indicators. Surgically excised tumor samples from 101 female patients with hormone receptor-positive breast cancer (HRBC) were stained for ER, PR, HER2, Ki67, SATB1, CD8, and scanned at 20x. Digital image analysis was performed using the HALO™ platform. Subsequently, hexagonal tiling was used to compute intratumoral heterogeneity indicators for ER, PR and Ki67 expression. Multiple Cox regression analysis revealed three independent predictors of the patient's overall survival: Haralick's texture entropy of PR (HR = 0.19, p = 0.0005), Ki67 Ashman's D bimodality (HR = 3.0, p = 0.01), and CD8+SATB1+ cell density in tumor tissue (HR = 0.32, p = 0.02). Remarkably, the PR and Ki67 intratumoral heterogeneity indicators were prognostically more informative than the rates of their expression. In particular, a distinct non-linear relationship between the rate of PR expression and its intratumoral heterogeneity was observed and revealed a non-linear prognostic effect of PR expression. The independent prognostic significance of CD8+SATB1+ cells infiltrating the tumor could indicate their role in anti-tumor immunity. In conclusion, we suggest that prognostic modeling, based entirely on the computational image-based IHC biomarkers, is possible in HRBC patients. The intratumoral heterogeneity and immune response indicators outperformed both conventional breast cancer IHC and clinicopathological variables while markedly increasing the power of the model.