Image acquisition and processing
Slides were scanned with a ScanScope CS (Aperio Technologies, Vista, CA) using brightfield imaging at 20× magnification. Specimen areas were selected and individual images were saved in a 24-bit RGB TIFF file format with a resolution of 1 μm/pixel using the ImageScope software (Aperio Technologies). The automated analysis of the TIFF image files was performed using the programming language IDL 6.3 (ITT Visual Information Solutions, Boulder, Colorado) [14, 15] for IHC quantification using RGB [12], normalized RGB [6], HSL [8, 13], and CMYK as described below.
CMYK image analysis method
We adapted a CMYK model with maximum grey component replacement, a criterion in which the lowest brightness level is subtracted from all channels. This criterion effectively discriminates color differences by subtracting out the grey level, a level of equal intensity among the three channels (C, M, Y). Consequently, this subtraction sets one of the color channels to zero in each image. The selection of this display criterion for an image using a popular software Adobe® Photoshop®7.0, involves the following sequence of menu selections: Edit, Color settings, Custom CMYK and Black generation set at Maximum. In this study, CMYK values were derived from the CMYK_Convert library procedure in IDL 6.3 using the equations:
(* indicates multiplication)
K (black) = 255 - maximum(R, G, B)
C = 255* [1 - R/(255 - K)] (if K = 255 then C = 0)
M = 255* [1 - G/(255 - K)] (if K = 255 then M = 0)
Y = 255* [1 - B/(255 - K)] (if K = 255 then Y = 0)
Mean intensity measurements in regions of interest were computed without the use of positive color intensity thresholds. However, proportions of labeled area in tissue were determined with a positive range intensity of 21+ and 93+ for the Yellow and nRed channels, respectively.
Xenograft models
Spectral characteristics of AEC, DAB and NovaRed were displayed with the different color models using color bars that were generated from a range of representative IHC staining derived from the xenograft studies described below, as well as additional IHC archived slides that showed a greater range of staining intensity.
Changes in STAT3 phosphorylation (p-STAT3) levels were associated with different conditions, including growth factor stimulation and sampling techniques as previously described [16]. Formalin-fixed paraffin embedded specimens were obtained to compare IHC staining using different chromogens AEC, DAB or NovaRed, and hematoxylin counterstain. Briefly, a tumor xenograft was excised and divided into two equal parts. One part was further divided and subjected to immediate formalin fixation, or protein extraction for Western blot analysis [17]. The second part was incubated with 50 ng/ml epidermal growth factor (EGF, Sigma-Aldrich, Oakville, ON) in PBS for 20 minutes prior to specimen analysis. The antibody S727p-STAT3 (BD Biosciences, Mississauga, ON) was used for Western blotting at a dilution of 1:100, and anti-GAPDH used at a dilution of 1:5000 (Ambion, Foster City, CA). Protein bands were detected with ECL plus fluorescence, imaged with a Typhoon system and their intensities were quantified (mean intensity × band area) and normalized with GAPDH using ImageQuant5.2 software (GE Healthcare, Little Chalfont, Buckinghamshire, UK).
Second, p-STAT3 levels were also quantified in formalin-fixed paraffin embedded sections of fine-needle samples obtained from a tumor xenograft. These samples were subjected to either immediate or delayed fixation after 5, 20 or 60 minutes, which caused an increase in p-STAT3.
Clinical test sets for comparison with CMYK model
Image analysis was performed on two clinical image data sets collected for other histopathological studies. The first study comprised of a total of 414 sections obtained from three adjacent sections of 138 frozen uterine cervical carcinoma biopsies. Carbonic anhydrase-IX-labeled DAB was previously evaluated with a positive color selection method [18]. In brief, a visual selection of a range of positive brown color pixels was performed using Adobe®Photoshop® 7.0 and batched analyzed using the IDL 6.3 programming language. Image analysis was performed in areas designated by tumor masks.
The second study comprised of a total of 282 formalin-fixed and paraffin embedded NSCLC biopsies arranged in 8 tissue microarrays which were IHC-stained for EGFR-labeled NovaRed [19]. Three cores of 0.6 mm diameter represented each case with evaluation completed on cases represented by at least 2 cores (n = 256). Two evaluators visually scored with a scale 0–3 (0 for absence of membrane staining in cancer cells) and their scores were strongly associated (r = 0.97). Thus, the mean of core scores was used as the final score for individual cases. Image analysis was performed in areas designated by whole tissue core masks since anti-EGFR demonstrated high specificity for cancer cells.
Immunohistochemistry
Specimens were processed with routine protocols using anti-S727p-STAT3 (BD Biosciences), anti-CA-IX (a generous gift of Dr. Adrian Harris, University of Oxford, UK) [15] and anti-EGFR (clone 31G7, Zymed Technologies, Invitrogen, Burlington, ON). The Idetect Ultra HRP Detection System (ID Labs Inc, London, Ontario) was used to visualize either DAB (Dako Corp, Carpinteria, CA), NovaRED or AEC (Vector Laboratories). Counterstaining was performed with Gill modified hematoxylin (Harleco®).