Increased risk of hormone therapy failure in breast cancers expressing low phospho-Stat5: Validation of quantitative immunofluorescence assay parameters
M. A. Girondo, A. R. Peck, B. Freydin, I. Chervoneva, T. Hyslop, A. J. Kovatich, J. A. Hooke, C. D. Shriver, E. P. Mitchell, H. Rui , Cancer Research , DOI: 10.1158/0008-5472.SABCS13-P1-08-20
Previous analyses of three breast cancer cohorts revealed that loss of phospho-Stat5 in breast cancer is associated with significantly elevated risk of hormone therapy failure (1, 2). Nuclear localized tyrosine phosphorylated Stat5 (Nuc-pYStat5) may therefore have clinical value as a predictive marker.
Analysis of two of the three previously reported anti-estrogen treated patient cohorts used pathologist scoring of diaminobenzidine (DAB) chromogen-stained Stat5. However the third cohort, analyzed by quantitative immunofluorescence analysis (QIF) on the Genoptix/HistoRx AQUA platform, revealed a greater hazard ratio than the cohorts analyzed by pathologist DAB-scoring. To extend and validate these observations, we applied the Nuc-pYStat5
cutpoint derived in our previous study (2) to an independent cohort of anti-estrogen-treated breast cancer patients using two distinct QIF software platforms, AQUA and Definiens Tissue Studio. Tissue Studio relies on supervised machine learning and multiparametric features of a high-resolution whole slide image to identify cancer cell regions, while AQUA software relies on costaining of a tumor marker to identify cancer cell regions. The two QIF platforms produced highly concordant Nuc-pYStat5 levels (R2linear = 0.96, p<0.001, N = 344) and confirmed a significant elevated risk of failing antiestrogen therapy in patients whose tumors had lost Nuc-pYStat5 (Hazard ratio 3.6; 95% CI 1.8-7.4; P<0.02; N = 98). On both QIF platforms, Nuc-pYStat5 remained an independent marker after multivariate adjustment for standard pathology parameters, including ER/PR, HER2, age, node status and grade, with a hazard ratio of 5.8 (95% CI 1.3-22.2; P = 0.02; N = 52). High concordance between Nuc-pYStat5 levels produced by the two QIF platforms held up in a second independent dataset of more than 300 breast cancer specimens (R2 linear = 0.97, p<0.001, N = 382). Nuc-pYStat5 levels by the two QIF methods remained highly concordant across the entire dynamic range in both patient cohorts. Furthermore, high concordance was also observed between replicate QIF analyses of Nuc-pYStat5 on serial tumor microarray sections stained in the same run on an automated immunostainer (Concordance Correlation Coefficient (CCC) = 0.96; 95% CI 0.96-0.97). Modest inter-assay staining variation (CCC = 0.84; 95% CI 0.82-0.87) for Nuc-pYStat5 when serial tumor microarrays were stained on different runs several days apart could be corrected for by normalization procedures (CCC = 0.94; 95% CI 0.92-0.95). This progress supports the utility of QIF analysis of Nuc-pYStat5 levels in human breast cancer and further documents the potential value of Nuc-pYStat5 as a predictive marker of response to antiestrogen therapy. The study confirms that further retrospective and prospective validation studies are warranted.