Improving biomarker assessment in breast pathology

Abstract

The accuracy of prognostic and therapy-predictive biomarker assessment in breast tumours is crucial for management and therapy decision in patients with breast cancer. In this thesis, biomarkers used in clinical practice with emphasise on Ki67 and HER2 were studied using several methods including immunocytochemistry, in situ hybridisation, gene expression assays and digital image analysis, with the overall aim to improve routine biomarker evaluation and clarify the prognostic potential in early breast cancer.

In paper I, we reported discordances in biomarker status from aspiration cytology and paired surgical specimens from breast tumours. The limited prognostic potential of immunocytochemistry-based Ki67 scoring demonstrated that immunohistochemistry on resected specimens is the superior method for Ki67 evaluation. In addition, neither of the methods were sufficient to predict molecular subtype. Following this in paper II, biomarker agreement between core needle biopsies and subsequent specimens was investigated, both in the adjuvant and neoadjuvant setting. Discordances in Ki67 and HER2 status between core biopsies and paired specimens suggested that these biomarkers should be re-tested on all surgical breast cancer specimens. In paper III, digital image analysis using a virtual double staining software was used to compare methods for assessment of proliferative activity, including mitotic counts, Ki67 and the alternative marker PHH3, in different tumour regions (hot spot, invasive edge and whole section). Digital image analysis using virtual double staining of hot spot Ki67 outperformed the alternative markers of proliferation, especially in discriminating luminal B from luminal A tumours. Replacing mitosis in histological grade with hot spot-scored Ki67 added significant prognostic information. Following these findings, the optimal definition of a hot spot for Ki67 scoring using virtual double staining in relation to molecular subtype and outcome was investigated in paper IV. With the growing evidence of global scoring as a superior method to improve reproducibility of Ki67 scoring, a different digital image analysis software (QuPath) was also used for comparison. Altogether, we found that automated global scoring of Ki67 using QuPath had independent prognostic potential compared to even the best virtual double staining hot spot algorithm, and is also a practical method for routine Ki67 scoring in breast pathology. In paper V, the clinical value of HER2 status was investigated in a unique trastuzumab-treated HER2-positive cohort, on the protein, mRNA and DNA levels. The results demonstrated that low levels of ERBB2 mRNA but neither HER2 copy numbers, HER2 ratio nor ER status, was associated with risk of recurrence among anti-HER2 treated breast cancer patients.

In conclusion, we have identified important clinical aspects of Ki67 and HER2 evaluation and provided methods to improve the prognostic potential of Ki67 using digital image analysis. In addition to protein expression of routine biomarkers, mRNA levels by targeted gene expression assays may add further prognostic value in early breast cancer.