Quantitative imaging of histological samples

(See a list of the products featured in this article.)

One of the benefits of fluorescence imaging is the ability to obtain quantitative data about the presence, location, and intensity of fluorescent signals in a fluorescently stained sample. Unfortunately, it can also be challenging due to the inherent loss of architectural details when there is no fluorescent marker associated with specific structures in the cells or tissue. Often a matched, chromogenically stained histological sample is created and viewed with a brightfield microscope to provide contextual information for use in interpreting the fluorescently stained sample. However, differences between the fluorescence and brightfield acquisition systems—for example, differences in spatial resolution between a monochrome and a color camera—can complicate the comparison of the stained samples.

To address the need for quantitative colorimetric imaging, we have developed a method for analyzing histological samples on a high-content imaging instrument using a specialized brightfield unit that allows for acquisition of the target chromophore’s absorption on a monochrome CCD camera. A histological sample is illuminated individually with differently colored light-emitting diodes (LEDs) in the transmitted light path, and the image from each channel is captured using a monochrome camera. These individual images can then be used to reconstruct a composite image of the chromogenically stained tissue by applying Maxwell’s theory of color composition [1], which states that you can synthesize all colors of light from the three primary colors (blue, yellow, and red) (Figure 1). Moreover, the contribution of each color component can be quantified using software that measures optical density, stained pixels, and other spatial features. Using this methodology, we can automatically, repeatedly, and quantitatively analyze colorimetric images without user intervention. In addition, because the transmitted light path remains compatible with the fluorescent light path, we are able to directly compare matched histological and fluorescently stained samples.

Figure 1. Application of Maxwell’s theory of color composition to histological staining. Tissue microarrays (TMA) of normal human tonsil tissue were stained with the blue nuclear counterstain hematoxylin, the red cytoplasmic counterstain eosin Y, and human anti–Ki-67 antibody, which targets a nonhistone nuclear protein. The antibody was detected with a horseradish peroxidase (HRP)–conjugated secondary antibody followed by staining with the HRP substrate diaminobenzidine (DAB), which forms an insoluble brown product upon oxidation by HRP. For acquisition, the samples were illuminated with blue, green, and red brightfield light-emitting diodes (LEDs, chosen to match the absorption spectra of the stains) in the transmitted light path. Samples were automatically acquired using 10x magnification on the Thermo Scientific™ CellInsight™ CX7 High-Content Analysis Platform (Cat. No. CX7A1110).

CellInsight CX7 HCA Platform for colorimetric imaging

To demonstrate quantitative colorimetric imaging, we used tissue microarrays (TMA) that were stained with the blue nuclear counterstain hematoxylin, the red cytoplasmic counterstain eosin Y, and a horseradish peroxidase (HRP) conjugate of human anti–Ki-67 antibody in conjunction with the HRP substrate diaminobenzidine (DAB). All data were acquired using the Thermo Scientific™ CellInsight™ CX7 High-Content Analysis (HCA) Platform, which is equipped with a five-color brightfield LED system that illuminates the samples with discrete wavelengths of light. Quantitative analyses were then performed using the Thermo Scientific™ HCS Studio™ Cell Analysis Software’s histology algorithm, which was developed to individually quantify the image obtained from each LED and also to optimize the typical parameters for histological analysis, including optical density staining measures and user-defined grading systems.

Figure 2 shows a stained TMA core sample that was acquired using blue, green, and red LED illumination; these images were then used to create the composite image either without or with color coding (the left panel shows a brightfield image without staining, the middle panel shows the colored image with staining, the right panel shows the algorithm overlays for the objects of interest). To generate an object count for analysis, we used the LED channel image corresponding to the blue hematoxylin staining to set object boundaries on the cell nucleus. To quantify the Ki-67 protein levels, we measured the brown DAB staining, which required analysis of both the red and green LED channel images. Once the cell mask and channel combinations are defined, the algorithm can detect the assigned criteria (such as pixels above a staining threshold) in each of the images and composites.

Figure 3 shows the application of this quantitative analysis to many different tissue samples and cancer types. Comparisons to control tissue demonstrated that significantly greater Ki-67 protein staining (p < 0.1) occurred in over half of the cancer samples. There were some exceptions, however. For example, sample group 6 (diffuse small noncleaved cell lymphoma of colon) exhibited low levels of Ki-67 protein staining (Figure 3A) but a highly variable optical density (OD) reading (Figure 3B). When the images for this group were manually reviewed, it was determined that folding of the TMA sample was responsible for this phenomena (data not shown), indicating that variability of OD within samples could be a valuable quality control measure.

Figure 2. Histological data acquisition and analysis on the CellInsight CX7 HCA Platform. Tissue microarrays of a diffuse large B cell lymphoma of neck were stained with the blue nuclear counterstain hematoxylin, the red cytoplasmic counterstain eosin Y, and human anti–Ki-67 in conjunction with a horseradish peroxidase (HRP)–conjugated secondary antibody and diaminobenzidine (DAB), which forms an insoluble brown product. (A) A stained tissue microarray core sample was acquired as separate fields using blue, green, and red illumination, a 40x objective, and tiling to create the entire sample. (B) Images were color-encoded to represent the typical visual acquired with a color camera. (C) Images were then individually analyzed: objects detected (blue outlines) and Ki-67 staining (red spots) were measured, analyzed, and compared to other groups within the array.

Figure 3. Quantitative analysis of tissue microarray samples. A series of tissue microarray samples were stained and analyzed using conditions described in Figure 2. All were analyzed using the Thermo Scientific™ HCS Studio™ 2.0 Cell Analysis Software’s histology algorithm for (A) total percent staining of Ki-67 (calculated as the number of DAB-stained pixels/total number of pixels × 100% in the acquired field) and (B) average optical density of each sample. Samples included: 1) mucosa-associated B cell lymphoma of thyroid, 2) diffuse plasmacytic lymphoma of small intestines, 3) diffuse large B cell lymphoma of colon, 4) diffuse large B cell lymphoma of groin, 5) diffuse small noncleaved cell lymphoma of left groin, 6) diffuse small noncleaved cell lymphoma of colon, 7) diffuse large B cell lymphoma of neck, 8) diffuse lymphocytic plasmacytoid lymphoma of lower jaw, 9) diffuse large B cell lymphoma of right oxter, 10) diffuse T cell lymphoma of right knee joint, and 11) cancer-adjacent normal tonsil tissue (control). * Indicates the statistical significance (p < 0.1) of the Ki-67 staining compared to the control (11).

Learn more about CellInsight CX7

The ability to automatically acquire, analyze, and store colorimetric images using the CellInsight CX7 HCA Platform enables the quantitative analysis of histological samples and provides researchers with a means of defining and sharing staining criteria.

Reference

  1. Maxwell JC (1890) Chapter XXIII: On the theory of three primary colours. In The Scientific Papers of James Clerk Maxwell. Vol. 1. Edited by WD Niven. Cambridge: Cambridge University Press. pp. 445–450.

Article download

Download a hyperlink-enabled, printer-friendly version of this article.
Download now