Abstract P2-09-03: Quantification of HER2 expression and spatial biology using computational pathology: A cross-assay validation study in breast cancer

Abstract Background Conventional pathologic scoring of HER2 by IHC is proven to distinguish potential responders to trastuzumab but has not been effective for next generation antibody drug conjugates (ADCs) such as trastuzumab deruxtecan (T-DXd), which is capable of bystander killing. Several alternative approaches have been deployed to measure HER2, including immunofluorescence and mRNA sequencing. We have developed a novel and fully automated computational pathology technique, Quantitative Continuous Scoring (QCS), to quantify the level and distribution of HER2 from digitized HER2 IHC slides in an objective, quantifiable, and reproducible manner on a per-cell basis [Gustavson et al., SABCS 2020]. To further validate this approach, we performed a systematic multi-omic comparison of QCS to orthogonal methods of HER2 quantitation on a cohort of primary and metastatic breast cancer cases (N=30). Methods HER2 was evaluated using three independent methods on serial tissue sections obtained from 30 archival FFPE breast cancer samples distributed over the full range of HER2 expression, from 0 to 3+. HER2-IHC staining (clone 4B5, Roche Tissue Diagnostics) was performed using standard methods and cases were scored by two pathologists using CAP/ASCO guidelines and H-scores were assigned. We performed FISH (HER2 IQFISH pharmDx [Dako]; PathVysion HER-2 DNA Probe Kit [Vysis]), mRNA quantification of ERBB2 transcript levels (Nano String), and immunofluorescence (IF; HER2 clone 29D8, CST). Imaged with Vectra (Akoya) and analyzed with Halo (Indica). QCS readouts were generated from the above-mentioned digital images of IHC slides by using a fully automated image analysis pipeline; readouts included per-cell staining intensity measurements of membranes and cytoplasmic sub-compartments in terms of optical density (OD) [Van der Laak, JQCS 2000], which were aggregated to a single slide-level score. Additionally, using the OD measurements and the cell locations, a Spatial Proximity Score (SPS) was computed, summing the percentage of cells with OD≥10 (corresponding to the limit of visual detection of IHC staining) as well as the percentage of cells with OD< 10 within a prespecified radius (25µm) of a neighboring cell with OD≥10. Results Our analysis demonstrated that QCS-based scoring correlates with orthogonal measurements used in this study. Comparing protein-based assays, the observed Pearson correlation was R=0.88 between QCS median membrane OD and IHC H-scores, R=0.86 with IF-based HER2 mean cell expression intensity, and R=0.85 with IF-based H-scores. Correlation with transcriptomic profiling was R=0.81 for OD vs. mRNA, however ERBB2 transcript levels did not distinguish between HER2 0, 1+, and 2+ FISH negative cases, while QCS was able to do so. Correlation between protein-based and nucleic-acid based assays were numerically worse, with R=0.64 for OD vs. FISH. All samples (including those with HER2 IHC scores of 0 and H-Scores < 10) had at least ~20% of cells with quantifiable HER2 expression by OD, the presence of which was confirmed using IF. For cases in the lowest quartile of HER2 expression by OD, SPS identified 20-50% additional HER2-null cells that were in close proximity to HER2-expressing cells that may be vulnerable to bystander killing. Conclusion QCS-based scoring is consistent with orthogonal protein-based measurements across the range of HER2 expression. Most importantly, QCS derived-spatial analysis features identify additional patients in the lower end of HER2 expression that might be highly relevant for ADC response prediction, particularly if a drug exerts bystander activity. Further clinical verification and validation on large cohorts is needed. Footnote: This study was approved by the IRB at MSKCC. Citation Format: Joshua Drago, Zonera Hassan, Jan Zaucha, Ansh Kapil, Fatemeh Derakhshan, Fresia Pareja, Shimulov Anatoliy, Fanni Ratzon, Travis J Hollman, Claire Myers, Jessica Chan, Andrea Spitzmuller, Mark Gustavson, Danielle Carroll, Dara Ross, Jorge Reis-Filho, Carl barrett, Sihem Khalifa, Schmidt Guenter, Hadassah Sade, Sarat Chandarlapaty. Quantification of HER2 expression and spatial biology using computational pathology: A cross-assay validation study in 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 P2-09-03.