Figure 1.
Overview of the noise genetics approach.
(a) We used a library of endogenously tagged proteins in human H1299 cells. Briefly, a retrovirus introduced YFP as an artificial exon into the introns of genes. Fluorescent clones were selected and sequenced. The clones express full length protein under endogenous control, with an internal YFP tag. (b) We selected 566 unique proteins with high quality movies and correct localization, and performed, or used existing, 24 h time-lapse movies(c) under controlled conditions. (d) Automated image analysis enabled by a mCherry tag in the parental clone enabled automatic tracking of protein level and localization as well as motility of each individual cell over time. (e) To find candidate motility genes, we sought proteins with high absolute correlation between protein level or localization (contrast, texture) and motility parameters (velocity, persistence). We tested a sample of the candidate motility genes by siRNA knockdown.
Figure 2.
Candidate motility genes were found with high correlation between protein and motility features in individual cells.
(a) Cell velocity and angle change (persistence) for a cell in three consecutive frames i−1,i and i+1. (b) Three protein features were compared to two motility features. Contrast and texture sum over pixels in a cell and employs the occurrence matrix p (i, j) equal to the probability that pixels of intensity i and j are adjacent. Texture uses the mean and standard deviation (std) of pixel intensity levels mu and sigma. (c) Pearson correlation coefficients R between protein and motility features show a group of proteins with high absolute correlation. Blue/red denotes low/high R values. (d) Randomized data (gray) showing lower correlations than real data (black) is used to establish a threshold correlation (dashed line) for detecting motility candidate genes with low false discovery rates. (e) Candidate motility genes show positive or negative correlation to protein level or localization measures. Data is shown in density plot along with the regression line of the linear fit. (f) An example for a non-motility candidate that show low absolute correlation.
Figure 3.
Candidate motility genes recover many of the known motility proteins in the screen.
(a) Examples of known motility genes found in the present assay, with their correlation coefficient. (b) New candidate motility genes found in the assay. (c) Known motility genes included in the screen but not found in the assay. Note that three of the genes have marginal R values. (d) Sub-cellular localization of candidate motility genes is enriched for cytoskeleton, Golgi/ER and plasma membrane, similar to all known motility genes (gene-cards).
Figure 4.
Novel motility genes are validated by mild knockdown experiments.
(a) We used anti-YFP siRNA to produce knockdown of the tagged gene in each clone, and tested protein level and motility in 48 time-lapse movies. (b) Examples of known and novel candidate motility genes after knockdown, as well as control genes, that are not known to be involved in the motility process. On the left, the effect of the knockdown on the expression of the protein is shown. In all knockdown experiments, protein fluorescence level was reduced at least by half. On the right, the effect of the knockdown of the gene on cell speed is shown. In the examples shown, candidate motility genes showed a reduction in velocity after knockdown compared to mock treatment, whereas control genes showed no significant reduction in motility. (c) Velocity reduction in knockdown experiments compared to mock treatment shows that 10/11 candidate genes showed a motility defect upon knockdown (blue bars), in contrast to control genes not known to be involved in motility, for which 3 out of 4 showed no significant defect (red bars). Stars denote known motility genes also found in our assay. Error bars stands for standard deviations (SD).