Validation of biomarkers and digital image analysis in breast pathology

Abstract

For women worldwide, the risk of developing breast cancer is second only to that of nonmelanoma skin cancer. Significant improvements have been made in survival over the past decades and today about 80 % of the patients survive 10 years or more after their breast cancer diagnosis. Still, far from all patients enjoy the relatively good survival indicated by statistics on breast cancer patients as one homogenous group. Improving prognostication of aggressive vs. less aggressive disease, and to separate tumors based on genetic differences for optimal treatment strategies, is therefore the focus of intensive research, including this thesis.

In paper I, we compared if tumor characteristics differ depending on what method of sampling the tumor that have been used for analysis. We compared routine immunohistochemistry on surgically resected breast specimens, including stains of the Estrogen receptor alpha (ER), the Progesterone receptor (PR), Human Epidermal growth factor receptor 2 (HER2) and the proliferation-associated protein Ki67, with analysis of the same stains done on material obtained from fine needle aspiration (immunocytochemistry). We found that there were substantial differences in the expression of these biomarkers between the two methods. Thus, the same rules for interpretation of biomarkers cannot be used for immunohistochemistry and immunocytochemistry, and consequently, validation of each method should be performed individually.

In paper II, we explored the scope of digital image analysis in biomarker evaluations. We scored ER, PR, HER2 and Ki67 status in several different regions of breast tumors by both manual methods and digital image analysis. The outcomes of the scoring of these biomarkers were then combined into IHC surrogate subtypes and compared to PAM50 gene expression-based subtypes as well as patient survival. All tested methods of automated digital image analysis of Ki67 outperformed manual scores in terms of sensitivity and specificity for the Luminal B subtype. Comparing digital versus manual testing concordance to all breast cancer subtypes as determined by PAM50 assays, the digital approach was superior to the manual method. The manual and digital image analysis methods matched each other in hazard ratio for all-cause mortality of patients with tumors with a “high” vs “low” Ki67 index. Manual assessments of the biomarkers ER, PR, HER2 and Ki67 were in most aspects less precise than digital image analysis.

In paper III, we evolved the concept of paper I with an evaluation of the concordance of consecutive Ki67 assessments performed on fine needle aspiration cytology versus resected tumor specimens. We investigated how a status of Ki67 “low” and “high” as determined by immunohistochemistry and immunocytochemistry corresponded to overall survival, respectively. Again, Ki67-index varied when the two methods were used on the same tumors, and was prone to switch the classification between low and high proliferation. ER evaluations were discordant in 5.3 % of the tumors, which in the clinical setting would mean that 1 in 20 patients would risk being left out of beneficial endocrine treatment or being given it without benefit. Ki67 “high”, as determined by immunohistochemistry, defined as a proportion of Ki67-positive cells above the 67th percentile of the material, was significantly associated with poor overall survival and a significantly higher probability of axillary lymph node metastasis. This could not be reproduced for immunocytochemistry. In summary, this study adds to the results of paper I, in which we showed discordance between the methods. By including survival data, we now conclude that not merely are the methods discordant, but immunocytochemistry fails to provide prognostic information. Consequently, immunohistochemistry should be regarded as the superior method.

In paper IV, we focused on proliferation comparing the results in the tumors’ hot spot, in the tumor periphery, and as the average proportion of Ki67-positive cells across the whole tumor section. Both manual and digital scores of Ki67 and the recently described marker for mitotic activity, PHH3, were evaluated along with mitotic counts. Their sensitivity and specificity for the gene expression based Luminal B versus A breast cancer subtypes, for the high versus low transcriptomic grade, for axillary lymph node status as well as for their prognostic value for breast cancer specific and overall survival were analyzed. Digital image analysis of Ki67 in hot spots outperformed the other markers in sensitivity and specificity both for gene expression subtypes and transcriptomic grade. In contrast to mitotic counts, tumors with high expression of Ki67, as defined by digital image analysis and high numbers of PHH3-positive cells, had significantly increased HR for all-cause mortality at 10 years from diagnosis. When we replaced the manual mitotic counts with digital image analysis of Ki67 in hot spots as the marker for proliferation when determining histological grade, the differences in estimated mean overall survival between the highest and lowest grades increased. It also added significantly more prognostic information to the classic Nottingham combined histological grade. We conclude that digital image analysis of Ki67 in hot spots might be suggested as the marker of choice for proliferative activity in breast cancer.