FoxO3a nuclear-cytoplasmic pulsing dynamics in a mammary epithelial cell line following growth-factor treatment with varying ERK/AKT activation loads. Dataset 2 of 3: single-cell reporter pulsing metrics. - Dataset (ID:20288)

HMS Dataset ID: 20288
Dataset Title: FoxO3a nuclear-cytoplasmic pulsing dynamics in a mammary epithelial cell line following growth-factor treatment with varying ERK/AKT activation loads. Dataset 2 of 3: single-cell reporter pulsing metrics.
Screening Lab Investigator: Somponnat Sampattavanich, Bernhard Steiert, Bernhard Kramer
Screening Principal Investigator: Peter K. Sorger
Assay Description: The nuclear-cytoplasmic pulsing behavior of a FoxO3a fluorescent reporter (FoxO3aN400-Venus) expressed in the 184A1 mammary epithelial cell line was assessed at the single-cell level using live imaging of cells that were untreated or treated with one of six growth factors in the absence or presence of an AKT or MEK inhibitor.
Assay Protocol: 1. Cells stably expressing FoxO3aN400-Venus and NLS-mCherry reporters were plated in 96-well plates at ~6 x 105 cells/cm2 and cultured at 37°C with 5% CO2 in DMEM/F12 (Invitrogen) supplemented with 5% horse serum, 20 ng/mL EGF, 10 µg/mL insulin, 0.5 µg/mL hydrocortisone, 100 ng/mL cholera toxin, 50 U/mL penicillin, and 50 µg/mL streptomycin (PMID: 12372298).
2. Prior to drug and growth factor addition, the cells were washed twice with PBS, placed in serum-free medium (DMEM/F12 with penicillin/streptomycin and no phenol red) for 5 hr, washed again, and placed in fresh serum-free medium.
3. DMSO or drug was added.
4. The cells were imaged using a 10x objective on a Nikon Eclipse inverted fluorescence microscope fitted with an environmental chamber maintained at 37°C with 5% CO2. Images were collected at 5 min intervals using a Hamamatsu ORCA-ER cooled CCD camera and Spectra-X light engine (Lumencor).
5. At 75 min after drug addition, imaging was briefly paused. Growth factor diluted in serum-free medium was added within 1 minute, resulting in an ~5% increase in culture volume, and imaging was resumed and continued for 24 hr.
6. Cell tracking was performed using a custom MATLAB script for cross-correlation between adjacent frames with manual validation. Image segmentation was performed using NLS-mCherry nuclear reporter signal.
7. Quantification of FoxO3a level in each cell is reported as the mean fluorescent intensity per pixel in the cytosolic region, the mean in the nuclear region, and the log10 ratio of mean cytosolic to nuclear fluorescent intensity (HMS LINCS Dataset #20287).
8. To evaluate differences in single-cell FoxO3a reporter translocation dynamics, functional principal component analysis (fPCA) was employed. Similar to classical principal component analysis, this technique generates an empirical set of orthonormal basis functions ψi (t) with


explaining maximal variance. The analysis was first restricted to the synchronous early response: t = -70 to 80 min. Equidistantly spaced cubic b-splines were used to convert the input signals to continuous time courses. One and a half data-points per basis function were chosen to smooth the signals and prevent overfitting. Artifacts at the boundaries, i.e. beginning and end of the traces, arising from the lack of measurements before and after the observed time interval did not occur due to the high number of traces. Three orthonormal basis functions (harmonics) were calculated, explaining more than 95% of the variance. Inside the resulting three-dimensional subspace, a new basis was chosen by rotation


with Euler angles θ1 = -30°, θ2 = 5°, and θ3 = -5°, resulting in harmonics that were more interpretable. The first describes the steady-state cell-to-cell variability before stimulation, the second the sustained response, and the third the transient part after ligand addition. As ligand identity and dose do not influence the pre-stimulation behavior, it is sufficient to analyze the principal component scores of harmonics 2 and 3. For the late phase, the harmonics only contain large time scales, reflecting that there is no non-trivial synchronous behavior for times t > 80 min.
9. To calculate pulsatile score for times t > 80 min, preprocessing was utilized to bowdlerize artifacts from traces such as spikes resulting from e.g. cell division or loss of tracking. Missing values were added by interpolation. The first three fPCA harmonics were employed to remove the large time scales. Then, the detrended signal was smoothed using N/3 b-splines for N data points. Peaks were detected on the smooth data representation. Due to overfitting, the pulsatile traces are mixed with small artificial peaks. These were dropped if the peaks were smaller than a certain threshold. The final detrended, interpolated, and peak-adjusted signals were then lumped into the following features fi:
(1) number of edges (approximately 2x the number of peaks)
(2) range (min-to-max)
(3) signal-to-noise ratio = range/(1/(N-1)*sqrt(RSS)) with RSS = distance of spline to data
(4) peak duration = mean(edge_start to neighboring_edge_end)
(5) peak distance = mean(min to neighboring min; max to neighboring max)

Whenever no peaks were detected, peak duration and peak distance were set to 300 min. A reference value ri and a weight wi were defined for all features fi. A positive value for wi was chosen for features where a larger number represents more pulsing, i.e. nEdges, range, and signal-to-noise ratio. A negative number for wi was assigned to features where a larger number indicates less pulsing, i.e. peak duration and distance. Using the features fi, the reference values ri, and the weights wi, pulsatile score


was calculated and compared to a threshold for assigning each trace into either the pulsing or the non-pulsing group (HMS LINCS Dataset #20288).
10. Finally, mean fPCA values and features were calculated across all single-cell traces for a given condition, and the fraction of pulsing cells per condition

was calculated (HMS LINCS Dataset #20289).
11. The complete set of images for this dataset is viewable online through our OMERO server at omero.hms.harvard.edu/webclient/?show=plate-913, and a lookup table with data file data column definitions is contained within the downloadable data package for this dataset.
HMS Dataset Type: Analysis
Date Publicly Available: 2017-05-02
Most Recent Update: 2017-05-02