Aim To compare SP6 and MIB1 antibodies for Ki67 staining in breast cancer.
Background Immunohistochemical detection of Ki67 has been widely used to assess the proliferative fraction in breast cancer. Ki67 is used prognostically and is the primary end-point for some presurgical trials. MIB1 has been the preferred antibody, but SP6 has become available, with apparently improved performance. The importance of Ki67 led us to systematically compare SP6 with MIB1.
Methods Two sets of tissue microarrays were used. These were constructed from formalin-fixed paraffin-embedded breast cancers: (i) 177 cancers with data on response to an aromatase inhibitor for advanced disease (cohort 1); (ii) 200 mainly oestrogen-receptor-positive cancers without response data (cohort 2). Twenty-eight pairs of core-cut biopsies taken before and after aromatase inhibitor treatment were also assessed (cohort 3). Stained sections were examined either visually or by using an image analysis system (Ariol).
Results There was a strong correlation between the two antibodies in all cohorts of samples scored visually (cohort 1: n=161, r=0.93, p<0.0001; cohort 2: n=194, r=0.84, p<0.0001; cohort 3: n=54, r=0.89, p<0.0001). Correlation between visual and Ariol scores was markedly better with the SP6 antibody (r=0.71 and r=0.88 for MIB1 and SP6, respectively). Ki67 related similarly with time-to-treatment failure with the two antibodies (cohort 1). Changes in Ki67 values with the two antibodies after 2 weeks of aromatase inhibitor treatment also correlated strongly.
Conclusions SP6 and MIB1 provide highly comparable measures of Ki67 that predict progression of advanced disease similarly. SP6 is substantially better suited than MIB1 to image analysis.
- breast cancer
- image analysis
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Ki67 is a nuclear non-histone protein that was first identified by Gerdes et al in 1991.1 It is universally expressed by proliferating cells in the late G1, S and G2/M phases of the cell cycle. Several studies have shown that its expression correlates with other well-known markers of proliferation, such as S-phase fraction,2 3 mitotic index,2 4 tyrosine kinase2 and bromodeoxyuridine incorporation.5–7
Immunohistochemical detection of Ki67 antigen has been widely used to assess the growth fraction in human malignancies, including breast cancer. Several studies have reported that baseline Ki67 has prognostic value in breast cancer.8 9 A meta-analysis of 46 studies involving more than 12 000 patients showed that high Ki67 levels were associated with increased probability of relapse and worse survival.10 Ki67 measurement is infrequently included in the routine assessment of primary breast cancer,11 but an important role as an intermediate marker of response has emerged in the neoadjuvant setting, and its incorporation as one of four immunohistochemical markers as a prognostic panel for breast cancer has recently been advocated.12
Changes in Ki67 levels in neoadjuvant studies of endocrine treatment have predicted the clinical outcome of much larger adjuvant trials, and on-treatment Ki67 measurements have been found to be better predictors of long-term outcome than pretreatment levels.13–17 Measurement of Ki67 was included in calculation of a new predictive score: PEPI (preoperative endocrine prognostic index), which estimates risk of relapse and survival rates after presurgical endocrine therapy.16 A role in neoadjuvant chemotherapy is less well established, but tumours with high Ki67 expression are more likely to respond to neoadjuvant chemotherapy,18 19 and post-treatment measurement of Ki67 is also a strong indicator of recurrence-free and overall survival.20
The original monoclonal antibody against Ki67 was used for immunostaining of proliferating cells in unfixed tissues. Development of antibodies such as MIB1 and MIB3 allowed the staining to be performed in formalin-fixed, paraffin-embedded samples after heat retrieval.21 Several antibodies against Ki67 have since become commercially available, but a mouse monoclonal antibody, MIB1, has been widely adopted for over 15 years.9 More recently, a rabbit monoclonal antibody, SP6, has become available and has been shown to enhance sensitivity and intensity of Ki67 staining when compared with the mouse monoclonal Ki67 antibody.22 SP6 has been also successfully applied for Ki67 measurements by other groups.23 24 Given the burgeoning importance of Ki67 as a biomarker and the well-established position of MIB1 antibody, any new antibody requires formal evaluation prior to its widespread use. We have therefore conducted a systematic series of studies comparing SP6 with MIB1, with visual scoring or image analysis, and an evaluation of their association with clinical outcome in advanced breast cancer.
Material and methods
Archival formalin-fixed paraffin-embedded tissue blocks from three patient cohorts were used in the present study. Cohort 1: primary or locally recurrent breast tumour samples from women diagnosed between 1974 and 2003, and subsequently treated with third-generation aromatase inhibitors (AIs) for advanced disease (n=177); samples were analysed in tissue microarrays (TMAs). Cohort 2: mainly oestrogen-receptor (ER)-positive primary breast carcinomas from the TransATAC program25 (n=200); samples were analysed in TMAs. Cohort 3: pairs of formalin-fixed paraffin-embedded core-cut tumour biopsies taken before and after 2 weeks of treatment of 28 postmenopausal patients with anastrozole.26
TMAs were constructed with a manual tissue arrayer (MTA-1; Beecher Instruments, Sun Prairie, Wisconsin, USA) using 600 μm cores. H&E-stained slides were reviewed by a pathologist and/or an experienced technician. Three areas of invasive tumour, away from in-situ or benign tissue components, were marked on the slides and the corresponding paraffin blocks for TMA construction. Three cores were extracted from each donor block and assembled into three recipient blocks (TMAs A, B and C).
Sections (4 μm thick) were cut onto X-tra Adhesive glass slides (Surgipath, Richmond, Illinois, USA), dewaxed in xylene, and then rehydrated by immersing in decreasing grades of industrial methylated spirits baths and rinsed in water. Antigen retrieval was by microwaving at 800 W in citrate buffer pH 6.0 for 10 min. Staining was performed using a Dako Autostainer (Dako, Glostrup, Denmark) and the protocol for the Dako REAL Detection System. MIB1 primary antibody (Dako) was used at a dilution of 1:40, whereas a 1:100 dilution was used for SP6 antibody (Abcam, Cambridge, UK). All dilutions and washes were performed with Dako antibody diluent and Dako wash buffer, respectively. On completion of the autostainer run, sections were removed from the instrument, counterstained with Mayer's haematoxylin, dehydrated, and mounted.
Ki67 visual scoring
Stained sections were examined visually with a ×40 objective and a 10×10 eye-piece graticule. The percentage of positive nuclei as a proportion of total number of tumour cells was recorded. In cohorts 1 and 2, the entire invasive tumour area for each core was scored by a single analyst (LZ), with confirmation of regions for scoring with a second analyst (JS) as necessary. The observers were blinded to patient outcome. In cohort 3, 10 high-power fields were scored in each core-cut biopsy.
Slides were analysed using the Applied Imaging Ariol image analysis system (Genetix, New Milton, Hampshire, UK). Slides were scanned using the TMAsight assay designed to capture images of individual cores using a ×20 objective. Images acquired through three filters (red, green and blue) were converted by Ariol software into colour reconstructions. The settings of MultiStainHighRes script were adjusted in a process called ‘training of classifiers’ to analyse objects of specific colour, size and shape, and to distinguish positive from negative cells. Only invasive breast cancer areas were included. Images of all TMAs were analysed using the settings established during the training.
The associations between Ki67 expression with the two antibodies were determined using Spearman rank and Pearson correlations in the GraphPad Prism software (http://www.graphpad.com/, version 5.0a).
Cox regression was employed to examine the relationship between MIB1 and SP6 and time-to-treatment failure (TTF) defined as the time from starting to being withdrawn from AI treatment. Values from MIB1 and SP6 were log-transformed and their relationship with hazard of treatment failure determined. The HR expresses the increase in hazard that occurs for every one unit increase on the natural log scale of MIB1 or SP6. Logistic regression was employed to examine the relationship between MIB1 and SP6 and the odds of clinical benefit (complete or partial response or stable disease for at least 6 months). Both factors were log-transformed; the OR expresses the increase in the odds of clinical benefit that occurs for every one unit increase on the natural log scale of MIB1 or SP6.
Overall, the performance of MIB1 and SP6 antibodies was similar in terms of staining intensity and the proportion of positive cells. SP6 stained sections showed lower non-specific general background staining in comparison with MIB1 (figure 1, top panels). The membranous/cytoplasmic reactivity was present in about 1% cases stained with MIB1, but not with SP6 (figure 1, bottom panels).
The main goal of analyses using cohort 1 was to compare the association of Ki67 using the two antibodies with a clinical end-point. A total of 161 out of 177 cases were suitable for visual Ki67 measurement. The total number of Ki67-positive and Ki67-negative cells in all available replicate cores (A, B and C) was assessed, and the percentage of Ki67 was calculated (pooled data). Only cases with ≥20 positive cells or ≥300 cells in total were included in the analysis. There was a highly significant positive correlation between MIB1 and SP6 antibodies (Pearson: r=0.93, p<0.0001; Spearman: r=0.95, p<0.0001; figure 2). The results of univariate analyses showed similar borderline significant association of high Ki67 with short TTF with the two antibodies, the association being marginally stronger with SP6 (MIB1: HR 1.15, 95% CI 1.00 to 1.33, p=0.052; SP6: HR 1.19, 95% CI 1.00 to 1.43, p=0.045).
The variability in visual Ki67 measurements between replicate cores was calculated as the coefficient of variation (CV) in all cases with two or more available cores. MIB1 showed higher median CV (%) compared with SP6 (43.7 versus 36.7, respectively).
The aim of analyses with cohort 2 was to determine the correlation between visual and image analysis scores for the two antibodies. Visual scores showed a highly significant positive correlation between MIB1 and SP6 antibodies (Pearson: r=0.84, p<0.0001; Spearman: r=0.89, p<0.0001; figure 2). Correlation between visual and image analysis scores was markedly better with the SP6 antibody (r=0.88) than with MIB1 (r=0.71); (figure 3).
A CV was calculated in all cases with two or more available cores for both antibodies and the two scoring methods. In contrast to the data obtained from cohort 1, MIB1 showed a lower median CV (%) compared with SP6 (42.0 versus 51.4, respectively) when Ki67 was scored visually. When Ariol was used to measure Ki67, MIB1 showed higher median CV (%) compared with SP6 (65.5 versus 60.2, respectively).
The aim of analyses with cohort 3 was to determine the comparability between the two antibodies in assessing changes in Ki67 in core-cut biopsies taken before and after endocrine therapy; in this case an AI: anastrozole. Data were available from 27 pairs of samples. Fold changes in Ki67 were assessed by calculation of the ratio of post-treatment to pretreatment Ki67 before natural log-transformation. There was a strong correlation between data from the two antibodies (Pearson: r=0.85, p<0.0001; Spearman: r=0.84, p<0.0001; figure 4). Calculation of mean fold change for both antibodies showed an average 72% suppression of Ki67 in post-treatment samples in both cases (MIB1 mean post:pre Ki67: 0.28 (95% CI 0.18 to 0.44); SP6 mean post:pre Ki67: 0.28 (95% CI 0.19 to 0.39)). There were no significant differences between suppression of Ki67 seen with the MIB1 and SP6 antibodies (p=0.79).
The measurement of Ki67 by immunohistochemistry, as well as being used as a marker of prognosis by some, is also being used in some neoadjuvant trials as an eligibility criterion (as a surrogate for luminal B status),24 a primary end-point26 and a means of triaging patients away from endocrine treatment alone.17 Validity of the methods used and the comparability of data both within and between centres are critical for the valid interpretation of the data and application of Ki67 for the purposes described above.
The data presented here demonstrate the highly similar performance of MIB1 and SP6 antibodies in visual analysis, and the better performance of SP6 in image analysis where the r value compared with visual scoring was 0.88 with SP6, and only 0.71 with MIB1. This better correlation appears to be explained by the reduced background with SP6, allowing the software classifiers to identify positively staining nuclei with greater confidence. This is a substantial advantage for quantitative markers where removal of subjectivity in scoring between observers is important, and the acquisition of images and analytical details allows accurate audit and comprehensive record keeping.
The presence of membranous/cytoplasmic staining of the MIB1 antibody has been well documented, but the mechanism of this immunoreactivity remains unknown.27–29 Leonardo et al have shown that this aspecific MIB1 staining can be reduced by increasing the antibody incubation temperature to 37°C,30 but others have not observed reduction of membranous/cytoplasmic staining with such changes.31 There is one report showing that, in breast carcinomas, membranous and cytoplasmic Ki67(MIB1) expression was present in 8% of cases and was significantly associated with tumours that were high grade, HER2 amplified and ER negative.31 In our study, the aspecific membranous/cytoplasmic Ki67 staining with MIB1 antibody was observed in about 1% cases. As the number of cases was low, we were unable to associate it with any other tumour feature. The membranous/cytoplasmic reactivity was not observed when SP6 antibody was applied.
While analytical comparability is important to demonstrate, it is also essential that the SP6 antibody should also be shown to perform at least as well as MIB1 in its clinical use. We chose two settings in which to test this. The two antibodies both revealed a relationship with TTF with AI use in the advanced breast cancer setting. Although in both cases the relationship was of borderline significance, it was notable that the HR with SP6 was higher than that for MIB1. We also tested the changes in Ki67 found in 27 pairs of core-cut biopsies from ER-positive primary breast carcinomas taken before and after 2 weeks of use of AIs that are known to suppress Ki67 levels by between 70% and 80%.14 32 This type of treatment setting is often referred to as a ‘window of opportunity’ study and has gained favour over recent years for the evaluation of the antiproliferative effect of new drugs in clinical development. The two antibodies demonstrated a similar mean suppression of Ki67, and there was a strong correlation between them for individual changes within tumours. The apparently slightly poorer correlations in this circumstance reflect the much lower levels of Ki67 in the AI-treated samples and the inevitably lower precision at low levels.
In summary and conclusion, the SP6 antibody has a similar performance to MIB1 for visual analysis, and improved performance for image analysis. Additional evidence of its valid application for prognostic evaluation and as a dynamic marker in studies of pharmacological efficacy provides strong support for its use in clinicopathological assessment of proliferation in breast cancer.
Rabbit monoclonal SP6 antibody showed a similar performance to mouse monoclonal MIB1 antibody for visual analysis of Ki67.
SP6 was substantially better suited than MIB1 to image analysis.
SP6 and MIB1 provided highly comparable measures of Ki67 that predicted progression of advanced disease similarly.
Funding The work was funded by The Breakthrough Breast Cancer Research Centre, The Mary-Jean Mitchell Green Foundation and the NIHR Royal Marsden Hospital Biomedical Research Centre.
Competing interests None to declare.
Provenance and peer review Not commissioned; externally peer reviewed.