Reliability of CD44, CD24, and ALDH1A1 immunohistochemical staining: Pathologist assessment compared to quantitative image analysis.

Background: The data on the expression of stem cell markers CD44, CD24, and ALDH1A1 in the breast tissue of cancer-free women is very limited and no previous studies have explored the agreement between pathologist and computational assessments of these markers. We compared the immunohistochemical (IHC) expression assessment for CD44, CD24, and ALDH1A1 by an expert pathologist with the automated image analysis results and assessed the homogeneity of the markers across multiple cores pertaining to each woman. Methods: We included 81 cancer-free women (399 cores) with biopsy-confirmed benign breast disease in the Nurses' Health Study (NHS) and NHSII cohorts. IHC was conducted with commercial antibodies [CD44 (Dako, Santa Clara, CA, USA) 1:25 dilution; CD24 (Invitrogen, Waltham, MA, USA) 1:200 dilution and ALDH1A1 (Abcam, Cambridge, United Kingdom) 1:300 dilution]. For each core, the percent positivity was quantified by the pathologist and Definiens Tissue Studio. Correlations between pathologist and computational scores were evaluated with Spearman correlation (for categorical positivity: 0, >0-<1, 1-10, >10-50, and >50%) and sensitivity/specificity (for binary positivity defined with 1 and 10% cut-offs), using the pathologist scores as the gold standard. Expression homogeneity was examined with intra-class correlation (ICC). Analyses were stratified by core [normal terminal duct-lobular units (TDLUs), benign lesions] and tissue type (epithelium, stroma). Results: Spearman correlation between pathologist and Definiens ranged between 0.40-0.64 for stroma and 0.66-0.68 for epithelium in normal TDLUs cores and between 0.24-0.60 for stroma and 0.61-0.64 for epithelium in benign lesions. For stroma, sensitivity and specificity ranged between 0.92-0.95 and 0.24-0.60, respectively, with 1% cut-off and between 0.43-0.88 and 0.73-0.85, respectively, with 10% cut-off. For epithelium, 10% cut-off resulted in better estimates for both sensitivity and specificity. ICC between the cores was strongest for CD44 for both stroma and epithelium in normal TDLUs cores and benign lesions (range 0.74-0.80). ICC for CD24 and ALDH1A ranged between 0.42-0.63 and 0.44-0.55, respectively. Conclusion: Our findings show that computational assessments for CD44, CD24, and ALDH1A1 exhibit variable correlations with manual assessment. These findings support the use of computational platforms for IHC evaluation of stem cell markers in large-scale epidemiologic studies. Pilot studies maybe also needed to determine appropriate cut-offs for defining staining positivity.