Optimal tumor sampling for immunostaining of biomarkers in breast carcinoma

Abstract

Introduction: Biomarkers, such as Estrogen Receptor, are used to determine therapy and prognosis in breast carcinoma. Immunostaining assays of biomarker expression have a high rate of inaccuracy; for example, estimates are as high as 20% for Estrogen Receptor. Biomarkers have been shown to be heterogeneously expressed in breast tumors and this heterogeneity may contribute to the inaccuracy of immunostaining assays. Currently, no evidence-based standards exist for the amount of tumor that must be sampled in order to correct for biomarker heterogeneity. The aim of this study was to determine the optimal number of 20X fields that are necessary to estimate a representative measurement of expression in a whole tissue section for selected biomarkers: ER, HER-2, AKT, ERK, S6K1, GAPDH, Cytokeratin, and MAP-Tau.

Methods: Two collections of whole tissue sections of breast carcinoma were immunostained for biomarkers. Expression was quantified using the Automated Quantitative Analysis (AQUA) method of quantitative immunofluorescence. Simulated sampling of various numbers of fields (ranging from one to thirty five) was performed for each marker. The optimal number was selected for each marker via resampling techniques and minimization of prediction error over an independent test set.

Results: The optimal number of 20X fields varied by biomarker, ranging between three to fourteen fields. More heterogeneous markers, such as MAP-Tau protein, required a larger sample of 20X fields to produce representative measurement.

Conclusions: The optimal number of 20X fields that must be sampled to produce a representative measurement of biomarker expression varies by marker with more heterogeneous markers requiring a larger number. The clinical implication of these findings is that breast biopsies consisting of a small number of fields may be inadequate to represent whole tumor biomarker expression for many markers. Additionally, for biomarkers newly introduced into clinical use, especially if therapeutic response is dictated by level of expression, the optimal size of tissue sample must be determined on a marker-by-marker basis.

Figure 1

Heterogeneity of MAP-Tau expression in a whole tissue section of breast carcinoma. (a) H&E stain. (b) Immunofluorescence. Nuclei are labeled with DAPI. Cytokeratin is labeled with Cy3. MAP-Tau is labeled with Cy5.

Figure 2

Cross validation design. (1) Division of cohort into test set and training set. Repeated 10 times. (2) Division of training set into learning set and evaluation set. Repeated 1,000 times. (3) Fitting of linear regression over learning set. Performed for sample sizes of one to thirty five field of views (FOVs). Calculation of average prediction error over evaluation set. Red arrow indicates first local minimum. (4) Calculation of average prediction error over the test set. Gray arrow indicates over local minimum over 10 training sets. Black arrow indicates smallest value within one standard error of average first local minimum.

Figure 3

Coefficient of variation (%) by epitope with 95% confidence intervals.