Sir: During apoptosis, cells break up into membrane bound bodies that contain condensed chromatin fragments and remnants of cell organelles (apoptotic bodies). 1 Apoptosis often occurs at a low frequency in breast cancer (typically< 5% 2), making accurate counting difficult. In addition, the counting of apoptotic bodies, following staining with haematoxylin and eosin (H&E), is usually assessed visually, and this is a laborious and time-consuming process. Caspase-3 is synthesized as an inactive proenzyme which is activated by cleavage in cells undergoing apoptosis (reviewed by Porter and Janicke3). Apoptotic cells can be detected more easily using immunohistochemistry (IHC) with an antibody for activated (cleaved) caspase-3 (Figure 1A, B) than counting H&E-stained cells alone (Figure 1C, D). In addition, activated caspase-3 can be observed in the nuclei of cells without morphological features of apoptosis, suggesting earlier detection of the apoptotic cascade, as well as in apoptotic bodies (Figure 1E). Therefore we explored the potential of activated caspase-3, detected by immunohistochemistry, to detect apoptotic cells in breast cancer accurately, by measuring (a) all caspase-3 positive cells or (b) morphologically apoptotic cells with caspase-3 staining. Furthermore, it was envisaged that as improvements have been made to digital imaging and analysis platforms over the past few years, the efficiency, reproducibility and accuracy with which percentages of apoptosis are determined could also all be improved upon using digital pathology. Hence, we also investigated if a semi-automated image analysis approach using the Ariol® platform (Leica Biosystems, Newcastle, UK) could be used to determine the level of apoptosis in breast cancer. A total of 285 breast tissue microarray (TMA) cores (subset of 359 previously described breast carcinomas4 were stained by immunohistochemistry for active caspase-3 (Abcam ab2302; Abcam, Cambridge, UK), also as described previously. 4 Cores were scored manually (median of 170 cells per core; 95% confidence interval= 23.6) using two separate algorithms:(i) percentage of all objects staining with caspase-3 and (ii) percentage of caspase-3 positive apoptotic bodies (ie objects staining positive with caspase-3 and containing condensed DNA). The results from each method were compared. Each method detected a range of frequencies of apoptosis that were closely matched (0.36–40%). The percentage of caspase-3 positive cells was correlated significantly using Spearman’s rho nonparametric comparison with the percentage of apoptotic bodies (R2= 0.9602, P= 0.01; y= 0.9715x+ 0.0346). The interclass correlation coefficient for the two methods was 0.981.

A total of 214 breast cancer tissue microarray (TMA) cores were scanned at high resolution (× 20) using Ariol® SL-50 image analysis system (Genetix; as described previously5), and defined areas for scoring were set. Tumour cells in these areas were identified manually. Tumour cells were counted manually for those containing no brown staining (‘negative cells’) and those with ‘brown objects’(median of 220 cells per core; 95% confidence interval= 31.2). A classifier based on cell shape and colour was defined on Ariol® platform for the ‘negative’cells and another for cells with ‘brown objects’. Both the number of negative cells and brown objects were correlated significantly between the manual and Ariol® counts (Figure 2A, B, respectively). The interclass correlation coefficient between the percentage of apoptosis determined using manual counts and those provided by Ariol® was 0.815.