Ultra-Fast Timepoint-Based HCI Unlocks live Cell at Scale

This application note explores the transformative capabilities of the Araceli Endeavor® in advancing high-content live cell imaging. For example, >1,000 cells/well can be imaged for an entire 384-well plate in under two minutes, with submicron resolution across four fluorescent channels and brightfield. We discuss the implications of this speed, enabling kinetic assays that avoid fixation artifacts and parse apart subtle phenotypes. We show data from live cells imaged hourly, as well as days apart, in both brightfield and fluorescence. Machine learning-based phenotyping on these cells illustrates the lessened impact of this fast time point based live imaging. This piece also highlights the advantages of brightfield imaging, offering non-perturbating, non-bleaching imaging ideal for AI-driven analysis. Overall, the Endeavor demonstrates unmatched speed, resolution, and scalability – empowering researchers to develop temporally resolved high-content screening workflows with minimal cellular stress.

Researchers face a critical challenge in high-content screening: traditional methods often overlook dynamic cellular changes, leading to incomplete insights, but live cell imaging does not scale up. The Araceli Endeavor® solves this by enabling rapid live cell imaging without compromising quality or coverage, revealing subtle or transient biological phenomena with unmatched efficiency. Short-lived effects that would otherwise be overlooked can be illuminated: translocation of NFκB would be missed if measured an hour or two later (Simpson and Wafford 2006), and many cytostatic compounds essential to slowing tumor growth are easily found with live cell outgrowth assays but may be missed with fixed point screening (Single et al 2015). Experimental outcomes themselves can change over time, e.g. metastatic cancer cells making a comeback after being seemingly suppressed by NK cells: how do you know that your single fixed endpoint accurately represents real biology? Further, the act of preserving the cell itself may bias the outcome of the assay. Fixation alters the cell, affecting subcellular morphology (Li et al 2017), such as increasing mitochondria granularity and membrane blebbing, causing protein aggregation, losing soluble factors, and changing protein localization (Schnell et al 2012). Valuable assays such as cell killing, proliferation, phagocytosis and viral infectivity all rely on measuring a cellular response over time or rate of change, requiring multiple time point data of live cells (Gelles and Chipuk 2016). By imaging living cells, critical insight can be gained with new assays while avoiding bias.

There is an essential problem with live cell imaging in high content screening: how do you scale up? If it takes 2 hours to image an entire 384- or 1536- well plate with sufficient coverage and resolution, then hourly time points are impossible without significant compromises. Importantly, cell health may be negatively affected by these long acquisition times, with cells spending hours outside of the incubator. Araceli Endeavor® uniquely enables live cell high content imaging due to its uncompromising acquisition speed (Table 1), minimizing time out of the incubator and facilitating the time points your assay requires or finding the optimal endpoint for a given assay. With clear, detailed imaging at a resolution smaller than a single micron (0.27µm/pixel), paired with a large field of view (FOV) and advanced purpose- built optics designed for speed, the Endeavor delivers a majority of the well in 96- 364- and 1536- well plates (wp) in 4 fluorescent channels and brightfield in <6 minutes. Assuming ~1000 cells are needed per well, single FOV imaging in 5 channels happens in ~20 seconds for 96wp and just over a minute for 384 wells, enabling kinetic data to be captured. By imaging up to 30 plates per hour, the Endeavor enables researchers to perform high-throughput kinetic studies in a fraction of the time, accelerating drug discovery and assay development timelines. No longer will a single plate monopolize the imager all day. This combination of coverage, resolution and speed unlocks time point-based live cell imaging without compromise, delivering quality at scale (see first page figure highlight).

Endeavor solves another problem as well: cells change over time, growing and reacting to drug treatments and dyes. Assuming the typical 2 hours to image a 384-well plate, will cells in well A1 behave the same as well P24, imaged 120 minutes later? NFkB localization 1 hour after cytokine treatment shows robust cytosolic to nuclear translocation, but at 3 hours is nearly absent (Simpson and Wafford 2006). Dyes may efflux out of the cell or internalize from the membrane into vesicles (compare timepoints in Figure 1a). By imaging a majority of the well in <6 minutes, or ~1,000 cells/well in <2 minutes (for 96 and 284wp, see table 1), Endeavor ensures experimental consistency across the entire plate for all but the fastest kinetic assays.

To demonstrate the advantages of ultra-fast live cell imaging, we set up a comparison between live cell imaging with Araceli Endeavor® and equivalent imaging on another HCS instrument, asking whether time outside of the incubator (i.e., imaging time) impacts cell morphology. Cells were stained with a minimally toxic green dye and either 1) imaged hourly on Endeavor for 6 hours (Figure 1) and stored in an incubator between runs (“Araceli”), 2) imaged 3 times on Endeavor but left at room temperature for 4 hours, simulating the time required for 2-3 imaging runs on another high content instrument (“competitor mimic”), 3) kept in the incubator for the duration (“control”, 2 plates). All plates were imaged 48 hours and 6 days from the start of the experiment. Qualitative assessment of images revealed that cells imaged at timepoints showed less cellular debris and looked more like controls than plates left out for 4 hours (Figure 1). Additionally, competitor mimic plates displayed an increase in possible stress granules/ autophagic vesicles within distal processes.

To quantitatively assess cellular changes, we turned to machine learning, using ViQi’s AutoHCS, an AI-based phenotyping toolkit, trained on transmitted light and green channels (Figure 2). Importantly, predicting similarity using AI classification permits phenotypic clustering of elusive phenotypes including subtle indicators of cell stress. Here, images are assessed on a per-sample basis searching for similarity between images. As expected, time after plating was the strongest driver of similarity, with plates separating first according to time point. Within primary clusters, plates imaged hourly for 6 hours clustered with the controls at both 48 hours and 6-day timepoints, while the competitor mimic clustered separately, implying a distinct, perturbed phenotype. This was borne out in a confusion matrix (Figure 2b) with fast timepoint-imaged samples scoring closer to controls than competitor mimics. These data indicate that hourly fast imaging results in healthier cells with a lessened impact on cell morphology. By imaging live cells at high speeds, perturbation is minimized, with imaged cells looking more like un-imaged controls over the entire time range.

Traditionally, high content imaging has used fluorescent probes to interrogate cellular state, and with 4 fluorescent channels, the Endeavor is optimized for fast fluorescent imaging. However, in live cell imaging, the methods used for obtaining fluorescence may affect cellular state. Many dyes (including common nuclear stains) are toxic, and the presence of a probe may change the functional state of the very thing you are observing (Lulevich et al 2016). Notably, antibodies can prevent protein binding and localization (Schnell et al 2012), while common cytoskeletal dyes may chelate actin, preventing polymerization and cell division. Extended exposure times cause phototoxicity, especially in ultraviolet wavelengths, and are a particular concern with the repetitive imaging required for kinetic assays. Photobleaching can also be a problem, attenuating fluorescent signal upon repeated exposures. The Endeavor is a widefield system, intrinsically much more light-efficient than confocal systems, with significantly lower light exposure leading to less bleaching and phototoxicity (Khodjakov et al 2006). Binning can even further reduce exposure times, albeit at the expense of resolution. Even with these advantages, however, bleaching and phototoxicity may still be problematic depending on the probes used. Here, we show an example of fluorescence attenuation over time (Figure 3), with dye effluxed out of the cell resulting in a brighter background and less specific cellular staining. These scenarios highlight the significant benefits of using transmitted light imaging. Araceli Endeavor® uses far red-shifted light to reduce phototoxicity with exposure times 5-20ms. In the absence of fluorophores, bleaching is not an issue with stable signal over time. Key subcellular structures are resolved with Endeavor’s transmitted light at submicron resolution: not just nuclei but nucleoli, actin fiber bundles and endfeet, vesicles, endoplasmic reticulum and more (Figures 1 and 3b). This alleviates the need for many probes and dyes, particularly nuclear dyes which are especially toxic due to nucleic acid binding. As such, transmitted light allows cells to be observed unaltered with no toxicity. Another advantage of transmitted light is in detecting unexpected and off-target effects: antibodies and dyes, while specific to their targets, may miss effects not anticipated and specifically stained for. Recent advances in machine learning have uncovered unexpected functional data from transmitted light-based imaging, readily identifying subcellular structures and cellular states (La Greca et al 2021). Neural networks are often trained by correlating fluorescent stains marking known structures and identifying cell states (ground truth) with their corresponding brightfield images. Cell Paint- trained neural nets can predict morphological clustering and toxicity data (Cross-Zamirski et al 2022) and even perform as well as traditional CellProfiler analysis of fluorescent data on inferring mechanism of action (Harrison et al 2023). With the capacity to quickly switch between transmitted and fluorescent light imaging, the Endeavor is an ideal platform to both generate data for training transmitted light data on fluorescence results and screen with combinatorial strategies to alleviate dye toxicity.

This application note introduces a groundbreaking solution for scalable high-content live cell imaging, delivering high-throughput screening with uncompromised sample coverage and resolution. Be it maintaining cell health or assay experimental consistency, other high content imagers cannot deliver true-to-biology high throughput live cell imaging without compromising cell coverage or resolution. With the Endeavor, a screen will no longer miss the critical window of activity or lack the resolution to capture relevant physiological effects. Cellular responses can now be monitored at multiple timepoints to parse apart subtle or transient phenotypes. Time point imaging on the Endeavor makes possible kinetic high content screening in 96- 384- or even 1536- well plates, enabling the next generation of temporally resolved high content assays to be developed.

AI-based analyses power further insight, such as correlating known fluorescent assay outcomes to transmitted light and unbiased phenotyping. Here, machine learning-based phenotypic analysis indicates that time point-based imaging on the Endeavor minimizes cell damage seen with typical imaging over time. Simultaneous fluorescence and brightfield acquisition highlights the Endeavor’s seamless switching between TL and fluorescence, maintaining <6 minute acquisition times while imaging a majority of the well area in both brightfield and FL. In a confluence of innovation, Endeavor propels high content imaging just as more genetic knock- ins and non-perturbing fluorescent probes enable better assays and advances in machine learning unlock new possibilities in transmitted light analysis. Now you can see more, faster, in living cells, enabling you to scale these technologies up into novel screens.

Human bone carcinoma (U2OS) cells were split into 4 groups then plated in glass bottom 96-well high content plates, grown overnight to ~50% confluency, then stained with Biotium viafluor488 at 1:1000 dilution, hydrolyzed for 5 minutes then washed for 15 minutes. All plates were then imaged after staining, 48 hours and 6 days later, with imaging done in 1 plate (‘Araceli”) was imaged hourly for 6 hours, and otherwise kept in an incubator (approximate time outside of incubator 10 minutes/hour). 1 plate was left at room temperature for 3 hours (‘competitor mimic’) and imaged 3 times on day 1, with 2 plates (‘control’) kept in the incubator for the duration of the experiment. At 6 days, cells were washed 4x in PBS 7.4 before and after 15 minutes fixation in 15% formalin. All imaging with Araceli Endeavor in 3 fluorescent channels (exposures: green: 25ms, red: 60ms, far red: 50ms) and brightfield (8ms). Clustering and confusion matrix computed from brightfield and green channels at T0, 48 hour and 6 days, using ViQi Automated High Content Screening (AutoHCS) machine learning platform.

Cross-Zamirski JO, Mouchet E, Williams G, Schönlieb CB, Turkki R, Wang Y. Label-free prediction of cell painting from brightfield images. Sci Rep. 2022 Jun 15;12(1):10001. doi: 10.1038/s41598-022-12914-x.

Gelles JD, Chipuk JE. Robust high-throughput kinetic analysis of apoptosis with real-time high-content live-cell imaging. Cell Death Dis. 2016 Dec 1;7(12):e2493.doi: 10.1038/cddis.2016.332

Harrison PJ, Gupta A, Rietdijk J, Wieslander H, Carreras-Puigvert J, Georgiev P, Wählby C, Spjuth O, Sintorn IM. Evaluating the utility of brightfield image data for mechanism of action prediction. PLoS Comput Biol. 2023 Jul 25;19(7):e1011323. doi: 10.1371/journal.pcbi.1011323.

Khodjakov A, Rieder CL. Imaging the division process in living tissue culture cells. Methods. 2006 Jan;38(1):2-16. doi: 10.1016/j.ymeth.2005.07.007. PMID: 16343936; PMCID: PMC2590767.

La Greca AD, Pérez N, Castañeda S, Milone PM, Scarafía MA, Möbbs AM, Waisman A, Moro LN, Sevlever GE, Luzzani CD, Miriuka SG. celldeath: A tool for detection of cell death in transmitted light microscopy images by deep learning-based visual recognition. PLoS One. 2021 Jun 24;16(6):e0253666. doi: 10.1371/journal.pone.0253666.1087057115605765.

Li Y, Almassalha LM, Chandler JE, Zhou X, Stypula-Cyrus YE, Hujsak KA, Roth EW, Bleher R, Subramanian H, Szleifer I, Dravid VP, Backman V. The effects of chemical fixation on the cellular nanostructure. Exp Cell Res. 2017 Sep 15;358(2):253-259. doi: 10.1016/j.yexcr.2017.06.022.

Lulevich V, Shih YP, Lo SH, Liu GY. Cell tracing dyes significantly change single cell mechanics. J Phys Chem B. 2009 May 7;113(18):6511-9. doi: 10.1021/jp8103358.

Schnell U, Dijk F, Sjollema KA, Giepmans BN. Immunolabeling artifacts and the need for live-cell imaging. Nat Methods. 2012 Jan 30;9(2):152-8. doi: 10.1038/nmeth.1855.

Simpson PB, Wafford KA. New directions in kinetic high information content assays. Drug Discov Today. 2006 Mar;11(5-6):237-44. doi: 10.1016/s1359-6446(05)03696-2.

Single A, Beetham H, Telford BJ, Guilford P, Chen A. A Comparison of Real-Time and Endpoint Cell Viability Assays for Improved Synthetic Lethal Drug Validation. J Biomol Screen. 2015 Dec;20(10):1286-93. doi: 10.1177/