Fig. 1

A modified Cell Painting protocol captures a virus-specific morphological signature. a MRC-5 lung fibroblast cells infected with Human coronavirus 229E (CoV-229E), stained using Hoechst, SYTO 14, Concanavalin A, Wheat Germ Agglutinin and Phalloidin, in combination with an anti-coronavirus nucleoprotein (NP) antibody. Note the presence of non-infected (asterisk) and infected cells. b A representative composite image of infected cells with F-actin in green, nuclei in blue and anti-coronavirus NP antibody in red. Segmentation and classification of individual cells visualized with an outline with infected cells in purple and non-infected cells in yellow. c Morphological profiles of non-infected and infected cells (corresponding to the median profiles of both classes). d Dimensionality reduction using PCA applied to the extracted CellProfiler features per image, coloured according to their infected or non-infected classification based on NP-specific antibody staining. Percentage of variance explained is indicated by %. e With an R² = 0.73 and a Q² = 0.72, the PLS-DA prediction model could accurately predict viral infection on cell painting features as illustrated by the plot for observed vs predicted values, where observed values correspond to classification by NP-specific antibody. f, g Overview of the importance of each of the feature classes, grouped by module, cell compartment and stain if applicable. Absolute means of PLS-DA loadings indicate the importance of different feature classes associated with viral infection. Higher PLS coefficients indicate higher importance of a given feature group in order to separate a given condition (in this case, infected cells) from the controls (non-infected cells)