Cell Painting High-Content Image Assay

Cell painting involves a set of fluorescent reagents that are used to visualize and analyze the spatial organization of cellular structures and components. The assay is often used during drug screening to profile small molecules or other compounds on stained proteins in living cells. The pattern of cellular staining may revel morphological differences in samples treated by small molecule compounds.

1. Plate cells—cells are plated into 96- or 384-well plates at the desired confluency
2. Treatment/perturbation—causes desired phenotype of interest, either by chemical or genetic means
3. Fixation and staining—after treatment, cells are fixed, permeabilized and stained for the desired markers using individual reagents or a cell painting kit.
4. Image acquisition—the plate is sealed and loaded into an HCS-imager and images are acquired from every well. Image acquisition time will vary based on the number of images per well sampled, sample brightness, and the extent of sampling in the z-dimension.
5. Analysis—using automated software or previously designed scripts, features are extracted from the six-channel data to indicate a diverging phenotype, and these features are analyzed by cluster analysis (or a similar technique) to create several phenotypic profiles.

For an example of the cell painting workflow and protocol, see this cell painting application note.

Data for cell painting are typically acquired using a high-content screening (HCS) imaging system. HCS systems employ fluorescent imaging not unlike a traditional fluorescence microscope but are designed specifically to image multi-well (typically 96- or 384-well) plates at maximum speed for highest data throughput. A combination of traditional widefield and confocal fluorescence capabilities are often equipped, the latter being necessary for thick samples like organoids/spheroids or where maximum brightness and sensitivity is paramount.

Data acquisition is only the starting point. Processing those large data sets is often the rate limiting step during an experiment. Data processing can be performed using a variety of computing methods, but the goal is to elucidate significantly different phenotypic features between cellular populations to identify populations that are unique. This level of data processing has only recently become possible as data sets can now reach into the tens of terabytes.

Figure 1. Representative image and data readout from an HCS data set acquired on a Cellinsight CX7 LZR Pro system.

Figure 2. Phenotype comparisons of untreated vs. pharmacological control exposure in U2OS cells. Cells were treated with compounds of interest at 1–100 μM final concentrations for 48 hours in 96-well imaging plates. After treatment with the compounds, cells were immediately labeled using the Image-iT Cell Painting Kit and analyzed using the CellInsight CX7 LZR Pro instrument.
 

Figure 3. Example of a cell painting experiment with modified reagents. Reagents can be modified for specific experimental requirements, as is the case in the above showing the evaluation of cardiomyocyte mechanisms of action. H9c2 cells were seeded at 2,000 cells per well in 96-well plates. Two hours after seeding, the cells were treated with the listed compounds for 4, 24, or 96 hours before immunofluorescent labeling and analysis using the Cellomics Compartmental Analysis bioapplication.

Bray et al. (below) is the publication on which the cell painting technique is based. This where the term “cell painting” was first used in literature and is considered the standard protocol.

1. Bray, MA., Singh, S., Han, H. et al. Cell Painting, a high-content image-based assay for morphological profiling using multiplexed fluorescent dyes. Nat Protoc 11, 1757–1774 (2016). PMID: 27560178 

Schiff et al. (below) describes how machine-learning using a neural network was employed to discover Parkinson’s Disease phenotypes in cultured patient fibroblasts compared against healthy controls. The authors report this this method is accurate enough to reliably differentiate between healthy, sporadic, and LRRK2/genetic–driven disease states.

2. Schiff, L., Migliori, B., Chen, Y. et al. Integrating deep learning and unbiased automated high-content screening to identify complex disease signatures in human fibroblasts. Nat Commun 13, 1590 (2022). PMID: 35338121