Phenotypic Plasticity Of The Actin Cytoskeleton

Context

Although F-actin has a large number of binding partners and regulators, the number of phenotypic states available to the actin cytoskeleton is unknown. Here, we quantified 74 features defining F-actin and cellular morphology in >25 million cells after treatment with a library of 114,400 structurally diverse compounds. After reducing the dimensionality of these data, we found that only ~25 recurrent F-actin phenotypes emerged, each defined by distinct quantitative features that could be machine learned. We identified 2003 unknown compounds as inducers of actin-related phenotypes, including two that directly bind talin, a core regulator of integrin activity. Moreover, we observed that compounds with distinct molecular mechanisms could induce equivalent phenotypes and that initially divergent cellular responses could converge over time. These findings suggest a conceptual parallel between the actin cytoskeleton and gene regulatory networks; where the theoretical plasticity of interactions is nearly infinite, yet phenotypes observed in vivo are constrained into a limited subset of practicable configurations.

Content

This file describes key quantitative datasets from the research paper:
High content imaging of unbiased chemical perturbations reveals that the phenotypic
plasticity of the actin cytoskeleton is constrained
Nicole S. Bryce, Tim W. Failes, Justine R. Stehn, Karen Baker, Stefan Zahler, Yulia Arzhaeva, Leanne Bischof, Ciaran
Lyons, Irina Dedova, Greg M. Arndt, Katharina Gaus, Benjamin T. Goult, Edna C. Hardeman, Peter W. Gunning and
John G. Lock.
Lead contact: john.lock@unsw.edu.au
Published in Cell Systems (2019).
File name: Bryce_et_al_Quantitative_Dataset.csv
This file contains raw post-image analysis quantitative data, excluding only observations (i.e. rows) that
contained missing vales (124,767 observations included). Each observation pertains to a single experimental
condition capturing the effects of a single chemical compound on a population of SK-N-SH neuroblastoma
cells. Imaging and image analysis methods are described in the manuscript.
This file contains 81 columns. Columns 1-6 contain key metadata including: a unique Row ID per condition;
WellName (unique value per experimental condition); Experiment_Date (experimental batch number and
experimental data; year-month-day); Plate_Well (a second unique identifier linking the condition to its source
384-well plate and well); Plate (an identifier linking each condition to its source 384-well plate); Well (an
identifier linking each condition to its source well); Control Types (categorical defining unknown compound
‘Treament’, DMSO negative control, and Jasplakinolide, Latrunculin or TR100 positive controls); Plate
(unique ID per experimental 384 well plate). Columns 7-81 reflect raw image analysis parameters that define
the actin cytoskeleton and cellular phenotype. These parameters are defined in Supplementary Information
Table S1.
File name: Bryce_et_al_Quantitative_Dataset_Normalised.csv
This file contains data as defined above (Bryce et al Quantitative Dataset.csv) with the following alterations.
First, observations have been excluded based on the number of cells analysed per condition, such that only
conditions with >10 & <700 cells analysed are retained (124,546 observations included). Second, columns
82-155 now include z-normalised versions of raw image analysis parameters present in columns 7-81. Znormalisation was performed relative to the population of unknown compounds applied within each
experimental replicate (i.e. each experimental batch/date) as described in methods using robust statistics
(median and median absolute deviation; MAD).
File name: Bryce_et_al_Quantitative_Dataset_Normalised_with_TSNE.csv
This file contains data as defined above (Bryce et al Quantitative Dataset Normalised.csv) with the following
alterations. First, observations containing a proportion of duplicated values were excluded as this is not
compatible with subsequent dimension reduction methods such as t-SNE (t-distributed stochastic neighbour
embedding; 124,343 observations included). Second, the X (column 156, TSNE_X) and Y (column 157,
TSNE_Y) output coordinates for each observation are included following optimised t-SNE dimension
reduction. The final column ‘clusters’ (column 158) details the cluster membership (1-25) of each observation,
with the exception of observations that were ‘non-clustered’ – as defined by OPTICS unsupervised cluster
detection in the t-SNE space.
File name: Bryce_et_al_Quantitative_Dataset_Normalised_TSNE_non-clustered-removed.csv
This file contains data as defined above (Bryce et al Quantitative Dataset_Normalised_with_TSNE.csv) with
the ‘non-clustered’ observations (as defined by OPTICS unsupervised cluster detection in the t-SNE space)
excluded (123,108 observations included).
File name: Bryce_et_al_PCA_pre_Dimension_Reduction_Optimisation_output.csv
This file contains Row ID and 25 columns of PCA data based on normalised image parameters as in
Bryce_et_al_Quantitative_Dataset_Normalised.csv, but with duplicate value exclusion applied as in
Bryce_et_al_Quantitative_Dataset_Normalised_with_TSNE.csv. This PCA data forms the basis for
dimension reduction approaches using t-SNE and UMAP.
File names: Dynamic_Drug_Treatment_Data_output_for_R_plotting_TA_compounds.csv and
Dynamic_Drug_Treatment_Data_output_for_R_plotting_TA_compounds_outliers_included.csv
These files contain phenotypic (image analysis features) and in particular the t-SNE dimension reduction
coordinates that are used for the plotting of dynamic converging phenotype trajectories in Figure 4J, as well
as the outlier analysis in Figure S5B-D.

Acknowledgements

Bryce, Nicole S. et al. (2019), Data from: High-content imaging of unbiased chemical perturbations reveals that the phenotypic plasticity of the actin cytoskeleton is constrained, Dryad, Dataset, https://doi.org/10.5061/dryad.1cg2dq2