May 1, 2018 (Vol. 38, No. 9)

Alexander Schreiner Team Leader, Biological Applications PerkinElmer

Distinguishing Cell Types by Phenotypic Profiling of the Nucleus

The promise of high-content screening is the acceleration of discovery by extracting as much information as possible from cells. However, many high-content screens analyze only a small number of image-based properties and valuable information is not utilized.1 As nearly all screening approaches require a nuclear counterstain to facilitate segmentation, phenotypic profiling of nuclei can offer new and additional perspectives at no extra cost. This study shows how a single Hoechst nuclear stain enables phenotypic profiling and how this can be used to distinguish cell types within co-cultures, or even within seven different cell types, without any further staining or additional phenotypic markers.


Method

To maintain primary cells in vitro, they are often co-cultured with other cells which provide pro-survival signals in the form of trophic factors and cell-cell interactions. In a typical direct co-culture set-up, cell types are mixed within the same well, posing challenges to analyzing them separately. To show how nuclear counterstaining can be used for cell classification, human hepatocytes (HepG2) and mouse fibroblasts (NIH/3T3) were co-cultured and analyzed.

HepG2 liver and NIH/3T3 fibroblast cells were seeded into a CellCarrier-384 Ultra microplate (PerkinElmer) either alone or as co-culture at different ratios (2:1, 1:1, 1:2). Prior to mixing, HepG2 cells were stained with CellTracker Green CMFDA (Thermo Fisher) and NIH/3T3 cells with CellTracker Red CMTPX (Thermo Fisher) to enable validation of the accuracy of the phenotypic classification. For each individual cell type, 96 wells, and for each co-culture condition, 64 wells, were used. Cells were fixed, stained with Hoechst 33342 (Thermo Fisher), and single-plane images were acquired on an Opera Phenix™ high-content screening system (PerkinElmer) using a 20x water immersion objective in confocal mode. Nine fields per well were acquired corresponding to approximately 1,100 cells.


Results of linear classification for HepG2 and NIH/3T3 cells. HepG2 (green bars) and NIH/3T3 (red bars). Cells were either cultured individually or as co-cultures at different ratios. The graph shows the percentage of cells that were classified as either HepG2 or NIH/3T3 by PhenoLOGIC in Harmony software. In a 1:1 co-culture, the percentage of false positives drops to 0.34 – 0.65% of all cells. (Data not shown).2

Image Analysis

To classify individual cells based on Hoechst nuclear staining as either HepG2 or NIH/3T3, images were analyzed using the building blocks approach in Harmony® software (PerkinElmer). In summary, nuclei were segmented, intensity properties calculated, and border objects and mitotic cells removed.2

To determine detailed phenotypic profiles, SER (Spots, Edges and Ridges) texture and advanced STAR (Symmetry, Threshold compactness, Axial, or Radial) morphology parameters were calculated. SER texture quantifies the occurrence of eight characteristic intensity patterns within the image whilst STAR morphology parameters quantify the distribution of either texture features or fluorescence intensities inside a region of interest. A total of 230 parameters were calculated for every nucleus.

Using the PhenoLOGIC™ machine-learning option in Harmony, the parameters best suited to discriminate between two cell types were selected. PhenoLOGIC requires users to supervise training by clicking on about 100 representative objects per class to train the software to distinguish different phenotypes. The software performs a linear discriminant analysis3 to create a linear combination of the most relevant parameters that are then applied to untrained sample wells to classify cells.

To check the accuracy of the classification, the CellTracker intensity in a perinuclear region was calculated. If a cell was classified as one cell type but the respective CellTracker intensity was below a defined threshold, the cell was counted as a “falsely classified” cell. In monocultures, 97.5–97.8% of the cells are classified correctly. The percentage of falsely classified cells decreases in the co-cultures. In a 1:1 co-culture, the percentage of false positives drops to 0.34–0.65% of all cells. This indicates that the advanced texture and STAR morphology properties, together with PhenoLOGIC, allow the phenotypic differentiation of cell types in co-cultures based on Hoechst nuclear staining alone.

The PhenoLOGIC-based classification of HepG2 and NIH/3T3 cells in co-cultures showed that the features used were all SER and STAR morphology properties, prompting an assessment of whether these properties alone would be sufficient to distinguish between even more cell types. Therefore, in a separate experiment, mouse fibroblasts (NIH/3T3), canine kidney epithelial cells (MDCK), human breast adenocarcinoma (MCF7), human lung carcinoma (A549), human hepatocellular carcinoma (HepG2), human fibrosarcoma (HT1080), and human cervical adenocarcinoma (HeLa) cell lines were cultured separately, fixed, and stained with Hoechst 33342 only. Single-plane images were acquired on the Operetta CLS™ high-content analysis system (PerkinElmer) using a 20x water-immersion objective in confocal mode.

Image analysis was performed as for the co-culture experiment, using only advanced SER texture and STAR morphology properties, which were then subjected to unsupervised principle component analysis (PCA) using High-Content Profiler™ (PerkinElmer) secondary data analysis software. PCA is a visualization method especially suited to multiparametric datasets such as phenotypic profiles and reduces the dimensionality, allowing visualization of the similarities or differences among samples. The seven cell lines formed seven different well-separated clusters.2


Conclusion

The staining of cell nuclei contains a plethora of information that can be used for far more than just aiding segmentation during image analysis. As shown here, phenotypic profiling of the nucleus enables distinguishing of cells in co-cultures and up to seven individual cell lines can be separated by leveraging Hoechst nucleus staining. This type of phenotypic analysis can be directly applied to other cell types like primary cells co-cultured with feeder cells. Phenotypic profiling is not limited to the nucleus, thus applying it to other fluorescent labels, or cells labeled by the broader cell painting approach, opens up new horizons for unbiased drug discovery and disease research.4

The prerequisites for this type of phenotypic analysis are high-quality images, software for image segmentation and generation of phenotypic profiles, and a solution for processing complex multiparametric datasets. Imaging on either the Opera Phenix or Operetta CLS high-content screening systems generates high-quality images, whilst Harmony software enables primary image analysis with accurate image segmentation and advanced morphology and texture quantification methods to generate highly descriptive phenotypic profiles. With PhenoLOGIC, Harmony software also delivers an easy-to-use machine learning–based classifier that helps with dimensionality reduction, and further secondary analysis tools for data exploration are available in High-Content Profiler. PerkinElmer’s suite of products for phenotypic profiling enables you to leverage the real “content” of your high-content screening applications.


PerkinElmer

Alexander Schreiner
Team Leader, Biological Applications



























References
1. Singh S, Carpenter A, Genovesio A. Increasing the Content of High-Content Screening. Journal of Biomolecular Screening. 2014;19(5):640-650.
2. Schreiner S, Malle M, Böttcher K. Distinguishing Cell Types by Phenotypic Profiling of the Nucleus. PerkinElmer Application Note, 2017.
3. Fisher R. The use of multiple measurements in taxonomic problems. Annals of Eugenics. 1936;7(2):179-188.
4. Bray M, Singh S, Han H, Davis C, Borgeson B, Hartland C et al. Cell Painting, a high-content image-based assay for morphological profiling using multiplexed fluorescent dyes. Nature Protocols. 2016;11(9):1757-1774.

 

This site uses Akismet to reduce spam. Learn how your comment data is processed.