Single cell dynamic phenotyping

Live cell imaging has improved our ability to measure phenotypic heterogeneity. However, bottlenecks in imaging and image processing often make it difficult to differentiate interesting biological behavior from technical artifact. Thus there is a need for new methods that improve data quality without sacrificing throughput. Here we present a 3-step workflow to improve dynamic phenotype measurements of heterogeneous cell populations. We provide guidelines for image acquisition, phenotype tracking, and data filtering to remove erroneous cell tracks using the novel Tracking Aberration Measure (TrAM). Our workflow is broadly applicable across imaging platforms and analysis software. By applying this workflow to cancer cell assays, we reduced aberrant cell track prevalence from 17% to 2%. The cost of this improvement was removing 15% of the well-tracked cells. This enabled detection of significant motility differences between cell lines. Similarly, we avoided detecting a false change in translocation kinetics by eliminating the true cause: varied proportions of unresponsive cells. Finally, by systematically seeking heterogeneous behaviors, we detected subpopulations that otherwise could have been missed, including early apoptotic events and pre-mitotic cells. We provide optimized protocols for specific applications and step-by-step guidelines for adapting them to a variety of biological systems.

Randomly selected cells annotated as "yes" or "no" for mitotic state to serve as ground truth.

SUPPLEMENTARY DATA
Supplementary Data S1: Tracking data from Harmony, Imaris and CellProfiler examined by eye to generate the ground truth. To generate ROC curves, randomly chosen well and poorly tracked cells (cell.id) were annotated as "pass" or "fail" in the nuc.by.eye.ROC, columns and associated with respective TrAM values listed in the adjacent columns. To estimate tracking failure prevalence, randomly chosen cell tracks (cell.id) were annotated as "pass" or "fail" in the columns nuc.by.eye.total. For phototoxicity and mitosis data, events were additionally annotated as "TRUE" or "FALSE" in the phototoxicity.by.eye and in the mitosis.by.eye columns, respectively. For the AR translocation assay, cells expressing GFP-AR from a polyclonal population (PC3 GFP-AR) and from a clonal population (K22) were annotated as "TRUE" or "FALSE" for protein translocation in the responder.by.eye column and associated with the calculated GFP nuc/cyto change.

Tracking Cell Morphology and Drug Response:
Changes in cell motility and/or morphology can reflect drug response and are thus key read-outs in drug screening assays. Therefore, we sought to develop an imaging, segmentation, and filtering workflow that would track single cells from the time of drug treatment until detectable response.
Image Acquisition. First, we acquired images for 12 h at 5 min increments including a baseline image before drug treatment.
Phenotype Tracking. In these experiments it was important to consider what type of cell label to use: Nuclear dyes don't provide information on cell morphology and can quickly lead to cytotoxic effects, as demonstrated in Figure 4a.
Fluorophore-tagged protein expression, e.g. tubulin-RFP, provides highly accurate cell morphology information but requires filtering of non-expressing cells, thereby reducing assay throughput. Cell Tracker dyes are non-toxic, membrane-permeant, and generally label cells with uniform intensity and provide robust measurements of the whole cell population. Therefore, we performed experiments to track cellular phenotype based on Cell Tracker Orange. We chose a robust algorithm that was adaptable to cell populations with high variation in size and intensity, a prerequisite for tracking the broad morphology changes in response to drugs. We then used these training data sets to adapt the image analysis pipeline to different cell and drug types (Supplementary Table   S4).
Data Filtering. Finally, we applied downstream filters to obtain more reliable single cell tracks. Details on criteria used to calculate TrAM and applied thresholds are described in the Supplementary Information and summarized in Supplementary Table S7. Below we demonstrate an application of the optimized workflow.

Cell condensation and initiation of apoptosis in response to staurosporine:
Staurosporine has previously been reported to induce cytoplasmic condensation and both caspase dependent and independent apoptosis 1,2 . We therefore applied the above assays to track heterogeneity of the dynamic phenotype in response to staurosporine.
We performed experiments as described above, imaging Cell Tracker Orange stained cells every 5 min for 12 h, to quantify the heterogeneity of treatmentinduced morphological response and the initiation of apoptosis using the CellEvent™ Caspase-3/7 Green Detection Reagent to detect activation of effector caspases in real time. After initial cell tracking based on Cell Tracker Orange, we applied additional image processing to generate cell and background regions that we used to calculate the cell-specific EGFP signal-to-background ratio at all time points (Supplementary Table S5). We flagged cells having ratio > 1.1 at any time point as responders having activated caspases and initiated apoptosis (Supplementary Fig. S10). Finally, we subjected the preliminary tracking data to downstream TrAM filtering described in the Supplementary Information and summarized in Supplementary Table S7.
We first evaluated initiation of apoptosis by caspases on the population level.
Treated cells had a continuous increase in the number of responders during the course of the experiment, reaching 84% of the population at 12 h. In contrast, mock-treated cells only attained a 12% response rate (Supplementary Fig.   11a). We additionally observed morphological responses. Rounding of treated cells occurred, peaking at 6 h followed by cell disintegration (Supplementary   Fig. 10, Supplementary Fig. S11b). Cells responded prominently to staurosporine by condensation, as measured by an average reduction in cell area by 55%, compared to 8% in the untreated population (Supplementary Fig.  11c). Finally, having independently tracked morphological and enzymatic drug response, we asked whether these mechanisms were correlated. In the bulk population, we found a correlation between morphological (defined as > 40% reduction in cell area) and enzymatic response: 85% of cells underwent condensation and apoptosis during the 12 h time course and 4% did not respond either way (Supplementary Fig. 11f). Of the remaining 11%, 3% responded morphologically but never activated caspases, and 8% activated caspases but did not condense. Interestingly, cells with delayed condensation were more likely to undergo apoptosis later in the experiment, suggesting that cell morphology, in most cells, predicts cell death in response to staurosporine treatment. To understand the biological significance of these subpopulations and the interaction of these two response mechanisms will require further investigation. Taken together, results from our dynamic drug response phenotyping assay shows that we can identify temporal relationships between phenotypic response mechanisms and response heterogeneity across cells.

Assay-specific validation points:
Biological Applications: Motility. Fluctuations of cell density have been reported to impact cell speed 3 . In our experiments we detected increased average motility in higher density imaging fields of PC3 cells (Fig. 2c). We also validated Biological Applications: Protein Translocation. Cells overexpressing GFP-labeled androgen receptor are known to translocate the protein to the nucleus within minutes of treatment with the agonist R1881 4,5 . We analyzed multiple PC3 cell lines and measured rapid AR translocation within 30 min, whereas no change in localization was detected in mock-treated cells or cells expressing GFP (Fig. 3a).
We also validated that responding cells from the clonal and polyclonal cell lines translocated AR at comparable rates by manual assessment of 97 well-tracked responding cells (44 clonal and 53 polyclonal PC3 GFP-AR cells,

Supplementary Data S1).
Biological Applications: Phototoxicity. The nuclear stain DRAQ5 is a compound known to induce cytotoxic effects in a time-dependent manner 6 . Nuclear condensation is a well-established indicator for phototoxic imaging conditions and an early marker for apoptosis 7,8 . Using DRAQ5 for extended time-lapse experiments and population clustering based on nuclear morphology, we detected condensing nuclei reflective of early phototoxic events (Fig. 4b). We validated the fraction of condensing cells detected through our assay (19%) by evaluating 100 cells by eye (17 % identified as phototoxic events, Supplementary Data S1).
Biological Applications: Mitosis. Nuclear area has been shown to stably increase during interphase, followed by an episode of nuclear swelling an hour before cell division 9,10 . Analysis of G2 phase duration in HeLa cells (1 h, estimated via nuclear area change, Fig. 4 d+e) were comparable to previous reports 11 .
Biological Applications: Drug Response -staurosporine. Staurosprine is a potent drug known to initiate both caspase dependent and independent apoptosis within hours of treatment 1,2 . In our experiments, we verified both cell condensation and caspase-activation after 2 h of treatment with staurosporine ( Supplementary   Fig. 11f).