Molecular epidemiology of clinical tissues with multi-parameter IHC Poster 237 J Ruan 1, T Hope 1, J Rheinhardt 2, D Wang 2, R Levenson 1, T Nielsen 3, H Gardner 2, C Hoyt 1 1 CRi, Woburn, Massachusetts, USA, 2 Oncology Translational Laboratories, Novartis Institutes for Biomedical Research,Cambridge, Massachusetts, USA, 3 British Columbia Cancer Agency, Vancouver B.C www.cri-inc.com Seeing life in a new light.
Abstract Background A common goal in pharmaceutical research is to discover correlations between clinical outcomes and complex protein expression and interaction patterns in tissue sections. Correlations inform target validation, trial design, patient selection, response assessment, and, if trials are successful, the diagnostic component of theranostics. However, to successfully detect multiple, often weakly-expressed targets in clinical tissue sections requires appropriate staining protocols, advanced instrumentation and powerful software. Objective After developing multi-label immunohistochemical staining methods that were quantitative, independent, and specific, the goal was to create and validate an automated, whole-slide scanning imaging system that could be used to capture and distinguish multiple labels. Image analysis algorithms were used to differentiate relevant tissue regions (e.g., malignant and normal epithelia, stroma, necrosis, etc.) and segment cellular compartments (nuclei, cytoplasm, and membrane) to allow for detailed, spatially resolved multiparameter quantitation. Such as system would have an immediate application to signal-transduction research applied to conventional tissue sections. Here we describe this platform, and present results obtained from analysis of breast cancer tissue microarrays (TMAs). A key technological component of this project was multispectral imaging, which enabled the multiplexed detection. Methods and Materials Multispectral imaging, using a spectrally enabled whole-slide scanning system, was performed on two sets of a 712-core TMA (356 patients represented in duplicate). The first set was stained for ER, PR, and Her2, with a counterstain of hematoxylin, and the second stained for PR, Her1, and Her2, with a counterstain of hematoxylin. Single-stain data for comparison was obtained from previously stained and analyzed TMAs. IHC signals were spectrally unmixed and isolated from each other and the hematoxylin counterstain. Machine-learning-based automated image analysis was performed to locate cancer cells, segment subcellular compartments, and extract IHC signals on a per-cell basis. Relative stain intensities on a per-cell basis were analyzed with flowcytometry analysis software. Results and Discussion Multispectral 20x images obtained of each TMA core were acquired and spectrally unmixed at a rate of three cores per minute. Automated image analysis, using algorithms that could be developed by end-users in under 2 hours, took approximately 10 seconds per core, segmenting cancer-containing regions and extracting signals from relevant cell compartments. Concordance (r) between visual scores of single stain sections and the semi-automated scores of triple-stained samples indicate equivalency (r values ranging from 0.65 to 0.85), thus validating that multiplexed IHC faithfully reflects data obtainable with single-stains. Thus, multiplexed staining and detection, coupled with flow-cytometry analysis tools can be used to explore multiple protein expression patterns on a cell-by-cell basis, something that cannot be accomplished with serial single stains. Together, the innovative multispectral platform and software can capture cellular and subcellular expression details in an intact tissue architectural context.
An approach to utilizing multiplexed IHC in targeted drug development Explore pathway patterns with TMAs Correlate patterns with outcomes Field #5 Develop multiplexed IHC protocols 10 3 0.00% 1.69% dab 10 2 10 1 54.00% 43.74% 10 0 10 0 10 1 10 2 10 3 vector red Offer theranostic medical products Perform efficacy studies D x + R x
Multi-Parameter IHC Demonstration with ER, PR, and Her2 Step 1 : Scan TMA, Find Cores, Extract High-Power Fields (20x) RGB Image (20x) of a core
Multi-Parameter IHC Demonstration with ER, PR, and Her2 2 Step : Spectrally unmix stains from each other Spectral Library hematoxylin DAB (ER) Vector Red (Her2)
Multi-Parameter IHC Demonstration with ER, PR, and Her2 Step 3 : Classify tissue with machine learning algorithms (red = tumor, green = stroma, blue = background). Segment cellular compartments, extract spectrally unmixed signals from associated compartments, and export per-cell protein expression. Core (12,6) Core (7,7) Core (12,6) Core (7,7) Core (2,11) Core (10,3) Core (2,11) Core (10,3) Multi-parameter flow-cytometry-like outputs, but retaining tissue architecture and heterogeneity Machine learning-based automated image analysis has simple learn-by-example interface and is fast (10 20 seconds per image)
Multi-Parameter IHC Demonstration with ER, PR, and Her2 Step 4 : correlate clusters with clinical outcomes. Per-cell multi-parameter IHC Core (2,11) Core (10,3) Core (12,6) Core (7,7) 6000 4000 Her2 2000 Her2 6000 5000 4000 3000 2000 0 0 1000 2000 3000 ER 4000 Per-cell multi-parameter IHC 5000 5000 6000 6000 5000 4000 Her2 3000 2000 0 1000 2000 3000 4000 PR Per-cell multi-parameter IHC 3-D scatter plots show multi-parameter data extracted from individual cells. Note distinct clusters. Data points in the plots represent cells, and can be assessed by cluster analysis or quadrant analysis, commonly use in flow cytometry for phenotyping (double negative, single positive, double positive, etc.) 1000 1000 0 0 1000 2000 3000 4000 5000 ER 0 0 1000 2000 3000 4000 5000 6000 PR 0% 0% 11% 75% 1% 0% 16% 69% 2% 98% 1% 13% 99% 0% 3% 12% 98% 0% 0% 0% 2% 84% 0% 0% 2% 0% 8% 92% 0% 14% 0% 100%
Concordance between automated and visual scores Automated, quantitative measurements performed on spectrally unmixed stain signals were compared with visual 0-3+ and % positivity scores performed by pathologists at BCCA on single-stained serial sections of the same tissue micro-array. R-values of.80 to.90, were typical, showing equivalency ER % positivity R=0.80 Her2 0-3+ scores Visual Scores Automated Scores 0 1+ 2+ 3+ 0 554 27 5 4 1+ 13 10 1 1 2+ 3 2 11 3 3+ 1 0 5 35 R=0.85
Spectral Imaging Acquisition Images at different wavelengths Assemble the images into a data cube Spectrum at every (x,y) pixel Spectrum from nucleus with both hematoxylin and DAB Spectra of pure chromogens collected from single- stained sections Nuance TM Multispectral Imaging Systems Spectrum from membrane with just red stain Spectrum from stroma with just hematoxylin Unmixed DAB Component Unmixed Red Component RGB Representation of Spectral Cube Unmixed Hematoxylin Component Once unmixed, stains can be measured accurately.
Conclusions Multispectral imaging coupled with multiplexed IHC labeling is a practical and effective approach to detecting, isolating and quantitating low-expressing signaling pathway proteins in clinical samples CRi automated image analysis and spectral imaging enables practical multi-parameter molecular assessments of cells in intact tissue sections, thus enabling flow-on-a-slide with the added benefit of capturing critical phenotype heterogeneity across tissue sections.