Improved Interpretation of Immunohistochemistry Stained Pharmocodynamic Biomarkers by Histogram Data Display
JD Deeds, Y Gao, D He, RM Mosher. Novartis Institutes for Biomedical Research, Cambridge, MA
Background: Immunohistochemistry (IHC) analysis of pharmacodynamic (PD) markers provides valuable morphologic information but is limited by a lack of standardized methods for quantification. Typically, histopathologists set subjective thresholds for identifying cells that are positive for the presence or absence of a biomarker. Generating continuous data for expression levels across the entire range of expression may assist with interpretation by revealing additional features, improving statistical power, and reducing observer subjectivity.
Design: We examined expression of two biomarkers (pRB and ppRB) using chromagenic IHC stained xenograft tumor samples from compound treated animals. Samples were collected at various time points following compound administration. Staining intensity of nuclei was determined using ImageScope and IHC-Nuclear software (Aperio Technologies). Counts of number of nuclei staining for each of 256 intensity levels was computed for each sample.
Results: Examination of the full spectrum of intensity values revealed many aspects not observed via single threshold cutoffs. In the pRB stained samples, we were able to select any of a wide range of threshold values and observe all treatment-related changes. In addition, at one time point the histogram revealed that there are likely two populations of cells responding differently to treatment. ppRB staining on the other-hand revealed significant overlap between treated and vehicle samples, making intra-group comparisons difficult. Small changes in selected threshold values resulted in large changes in treated versus vehicle ratios.
Conclusions: Manual selection of thresholds is subjective and prone to bias. Varying a single threshold can result in different fold changes in experimental samples relative to control samples. Visualizing the continuous range of expression data has numerous advantages: 1) Histograms can indicate when values are well distributed over the measurable dynamic range. 2) Histogram analysis can reveal multiple subpopulations of cells within a sample which may be behaving differently. 3) Better measures of statistical significance may be obtained by analyzing histograms. Selecting optimal thresholds based on histogram analysis may provide better interpretation of changes of PD markers.
Wednesday, March 11, 2009 9:30 AM
Poster Session V # 219, Wednesday Morning