Automated recognition of cell phenotypes in histology images based on membrane- and nuclei-targeting biomarkers

Abstract

Background: Three-dimensional in vitro culture of cancer cells are used to predict the effects of prospective anti-cancer drugs in vivo. In this study, we present an automated image analysis protocol for detailed morphological protein marker profiling of tumoroid cross section images.

Methods: Histologic cross sections of breast tumoroids developed in co-culture suspensions of breast cancer cell lines, stained for E-cadherin and progesterone receptor, were digitized and pixels in these images were classified into five categories using k-means clustering. Automated segmentation was used to identify image regions composed of cells expressing a given biomarker. Synthesized images were created to check the accuracy of the image processing system.

Results: Accuracy of automated segmentation was over 95% in identifying regions of interest in synthesized images. Image analysis of adjacent histology slides stained, respectively, for Ecad and PR, accurately predicted regions of different cell phenotypes. Image analysis of tumoroid cross sections from different tumoroids obtained under the same co-culture conditions indicated the variation of cellular composition from one tumoroid to another. Variations in the compositions of cross sections obtained from the same tumoroid were established by parallel analysis of Ecad and PR-stained cross section images.

Conclusion: Proposed image analysis methods offer standardized high throughput profiling of molecular anatomy of tumoroids based on both membrane and nuclei markers that is suitable to rapid large scale investigations of anti-cancer compounds for drug development.

Figure 1

Image processing pipeline used to analyze immunohistochemically stained tumoroid cross section images. After the tumoroid cross sections were digitized at 40 ×, different tissue components were identified using unsupervised clustering and classification. Using the spatial distribution of pixels associated with positively and negatively stained cells, the staining patterns were identified. At the same time, the spatial distributions of DNA-rich pixels were analyzed and individual DNA spots were identified.

Figure 2

Sample synthetic images used to validate the computational image analysis algorithms used in the manuscript. First, a three-dimensional tissue block was generated randomly within a 128 μm × 128 μm × 64 m volume. Random tissue maps were obtained as 9 horizontal cross sections of this tissue block at 4 μm intervals near the vertical center, where the membranes, cytoplasms, and the nuclei of Ecad+/PR+ and Ecad-/PR- cells were marked with different labels. The associated Ecad-stained and PR-stained cross section images were generated by assigning colors from among those observed in the reference tumoroid cross section images followed by smoothing (bottom row). Overall, 5 such tissue blocks were generated, producing a total of 45 random tissue maps and respectively stained synthetic images. Image acquisition was modeled at 40×.

Figure 3

Processing of membrane-targeting Ecad stained tumoroid cross section images. The original image at 40× (top), the segmentation (middle), and the deduced staining patterns are shown (left column) along with marked high magnification boxes of width 156 μm (right column).

Figure 4

Processing of nuclei-targeting PR stained tumoroid cross section images. The original image at 40× (top), the segmentation (middle), and the deduced staining patterns are shown (left column) along with marked high magnification boxes of width 156 μm (right column).

Figure 5

Aerial and DNA spot percentages of positively and negatively stained regions. The percentages based on areas and the number of DNA spots in positively and negatively stained regions were very similar between different cross sections of tumoroids. The positive staining percentages of PR and Ecad staining followed each other in general to within a difference of 12% even though differences as large as 30% were also observed. In the charts above, positive staining is shown in red and negative staining is shown in blue.