Abstract P6-04-03: Computational pathology based HER2 expression quantification in HER2-low breast cancer

Abstract

Abstract Background HER2 directed therapies for breast cancer (BC) rely on accurate estimation of HER2 expression by pathologist scoring of immunohistochemically (IHC) stained tissue according to ASCO/CAP guidelines Emerging HER2-targeted antibody drug conjugates (ADCs) like trastuzumab deruxtecan (T-DXd), have demonstrated efficacy in the HER2-low (IHC 1+ or IHC 2+/ISH-) population (Modi, NEJM 2022). A deeper understanding of the spectrum of HER2 expression and its spatial distribution could provide insights about the mode of action of ADCs, including potential bystander activity. Computational pathology-based methods like Quantitative Continuous Scoring (QCS) can help here by objectively quantifying HER2 expression levels on a per cell basis from digitized HER2 IHC slides (Gustavson, SABCS 2020). We applied QCS to a cohort of HER2-negative (HER2-neg) patients (pts) from a retrospective study (NCT04807595) to quantify the prevalence of HER2 expression in this population and investigate the relationship with manual scoring. Methods To analyze the prevalence of HER2 expression in the HER2-neg population, we used available digital images (N=207 pts) from retrospectively rescored HER2 slides from tumors categorized as HER2-low or IHC 0 (IHC 0 or >0< 1+). QCS algorithm was applied to perform an instance segmentation of each tumor cell into the membrane, cytoplasmic and nuclear sub-compartments. HER2 expression levels on the membrane were estimated from a Hue-Saturation-Density model (Van der Laak, JQCS 2000) in terms of optical density (OD). Descriptive statistics and spatial modelling were used to aggregate cell-level information to a slide level score using the membrane OD values and tumor cell locations. A novel Spatial Proximity Score (SPS) was used to mathematically model the proportion of tumor cells that could potentially be targeted either directly or via bystander activity of ADCs. The analysis is ongoing, complete results with additional patient data to be presented. Analytical validation of the QCS algorithm demonstrated high correlation between OD values as measured on the automatically detected membranes from QCS and those measured on consolidated manual membrane annotations (N=2157 cells) from three pathologists (R = 0.993). This is very similar to the correlation observed between individual pathologists (R = 0.995). Results In the analyzed cohort (N=207), median OD of HER2-low tumors was significantly higher compared to IHC 0 tumors (one-sided Wilcoxon p-value < 0.001). A significant increase of OD values was observed for increasing IHC categories from 0 through >0 < 1+ and 1+ to 2+/ISH- (one-sided Jonckheere-Terpstra p-value < 0.001). OD values within each IHC category showed considerable variability, particularly in IHC 1+ and IHC 2+. In 49% of pts (N=101), greater than 88% of tumor cells expressed HER2 at any intensity (OD≥10). Among the remaining 106 pts, the number of potentially ADC-susceptible cells (within 25μm radius of HER2 expressing cells) as estimated by SPS was at least double the amount estimated by single cell-based scores alone in 45 cases (42%) and increased by at least 50% in another 12 cases (11%). Conclusions Computational approaches such as QCS can help us to objectively characterize the spectrum and spatial distribution of HER2 expression. These mathematical models contribute to our understanding of potential mechanisms of action of ADCs. While this study confirmed a general association of QCS-based scores with manual IHC categories, we also saw considerable variation, as some IHC 1+ or 2+ samples had low OD. Building on these and other promising initial results (Gustavson et al, SABCS 2020), we will further explore clinical relevance of QCS-based scoring. Eventually, digital scoring may be able to define data-driven signatures to select HER2-low pts that might benefit from HER2 targeted therapies. Citation Format: Andreas Spitzmüller, Ansh Kapil, Anatoliy Shumilov, Jessica Chan, Lemonia Konstantinidou, Zonera Hassan, Mark Gustavson, Danielle Carroll, Della Varghese, Gareth D. James, Akira Moh, Andrew Livingston, Victoria de Giorgio-Miller. Computational pathology based HER2 expression quantification in HER2-low breast cancer [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 P6-04-03.