We present a method for the acquisition of fluorescence reporter time courses from single cells using micropatterned arrays. The protocol describes the preparation of single-cell arrays, the setup and operation of live-cell scanning time-lapse microscopy and an open-source image analysis tool for automated preselection, visual control and tracking of cell-integrated fluorescence time courses per adhesion site.
Live-cell Imaging of Single-Cell Arrays (LISCA) is a versatile method to collect time courses of fluorescence signals from individual cells in high throughput. In general, the acquisition of single-cell time courses from cultured cells is hampered by cell motility and diversity of cell shapes. Adhesive micro-arrays standardize single-cell conditions and facilitate image analysis. LISCA combines single-cell microarrays with scanning time-lapse microscopy and automated image processing. Here, we describe the experimental steps of taking single-cell fluorescence time courses in a LISCA format. We transfect cells adherent to a micropatterned array using mRNA encoding for enhanced green fluorescent protein (eGFP) and monitor the eGFP expression kinetics of hundreds of cells in parallel via scanning time-lapse microscopy. The image data stacks are automatically processed by newly developed software that integrates fluorescence intensity over selected cell contours to generate single-cell fluorescence time courses. We demonstrate that eGFP expression time courses after mRNA transfection are well described by a simple kinetic translation model that reveals expression and degradation rates of mRNA. Further applications of LISCA for event time correlations of multiple markers in the context of signaling apoptosis are discussed.
In recent years, the importance of single-cell experiments has become apparent. Data from single cells allow the investigation of cell-to-cell variability, the resolution of intracellular parameter correlations and the detection of cellular kinetics that remain hidden in ensemble measurements1,2,3. In order to investigate cellular kinetics of thousands of single cells in parallel, new approaches are needed that enable monitoring the cells under standardized conditions over a time period of several hours up to several days followed by a quantitative data analysis 4. Here, we present Live-cell Imaging of Single-Cell Arrays (LISCA), which combines the use of microstructured arrays with time-lapse microscopy and automated image analysis.
Several methods for generating microstructured single-cell arrays have been established and published in literature5,6. Here, we briefly describe Microscale Plasma-Initiated Protein Patterning (µPIPP). A detailed protocol of the single-cell array fabrication using µPIPP is also found in reference7. The use of single-cell arrays enables alignment of thousands of cells on standardized adhesion spots presenting defined microenvironments for each cell and thus reduces a source of experimental variability (Figure 1A). Single-cell arrays are used to monitor the time courses of fluorescent markers purposed to indicate a variety of cellular processes. Long-term microscopy in scanning time-lapse mode allows for monitoring a large area of the single-cell arrays and hence sampling single-cell data in high-throughput over an observation time of several hours or even days. This generates time-line stacks of images from each position of the array (Figure 1B). In order to reduce the large amount of image data and to extract the desired single-cell fluorescence time courses in an efficient way, automated image processing is required that takes advantage of the positioning of cells (Figure 1C).
The challenge of LISCA is to adapt the experimental protocols and computational tools to form a high-throughput assay that generates quantitative and reproducible data of cellular kinetics. In this article we provide a step-by-step description of the individual methods and how they are combined in a LISCA assay. As an example, we discuss the time course of enhanced green fluorescent protein (eGFP) expression after artificial mRNA delivery. eGFP expression following mRNA delivery is described by reaction rate equations modeling translation and degradation of mRNA. Fitting the model function for the time course of eGFP concentration to the LISCA readout of the fluorescence intensity for each individual cell over time yields best estimates of model parameters such as the mRNA degradation rate. As a representative result we discuss the mRNA delivery efficiency of two different lipid-based transfection agents and how their parameter distributions differ.
Figure 1: Representation of the LISCA workflow combining (A) micro-patterned single-cell arrays (B) scanning time-lapse microscopy and (C) automated image analysis of recorded image series. The single-cell arrays consist of a two-dimensional pattern of cell-adhesive squares with a cell-repellent interspace leading to an arrangement of the cells on the micropattern, as can be seen in the phase-contrast image as well as the fluorescence image of eGFP expressing cells (A). The entire microstructured area is imaged in a scanning time-lapse mode repeatedly taking images at a sequence of positions (B). Recorded image series are processed to read out the fluorescence intensity per cell over time (C). Scale bars: 500 µm (A), 200 µm (C). Please click here to view a larger version of this figure.
Figure 2: Data acquisition combining single-cell microarrays (A) with scanning time-lapse microscopy (B). As preparation of the time-lapse experiment, a single-cell array with a 2D micropattern of adhesion squares is prepared (1), followed by cell seeding and the alignment of the cells on the micropattern (2) as well as the connection of a perfusion system to the six-channel slide, which enables liquid handling during the time-lapse measurement (3). A scanning time-lapse experiment is set up (4) and the cells are transfected on the microscope by injecting an mRNA lipoplex solution through the perfusion system during the time-lapse experiment (5). Scale bars: 200 µm. Please click here to view a larger version of this figure.
1. Microstructured single-cell array fabrication (Figure 2A)
Figure 3: Single-cell microarray fabrication by µPIPP. (1) PDMS stamps with a three-dimensional micropattern structure on the surface are arranged on a coverslip of a six-channel slide. (2) The coverslip with the PDMS stamps on it is treated with oxygen plasma to make the surfaces hydrophilic. (3) PLL-PEG is added. It is absorbed into the microstructure by capillary forces and makes the surfaces not covered by the PDMS stamp cell-repellent. (4) The coverslip is rinsed with water to remove the remaining PLL-PEG. Then, the PDMS stamps are removed and a six-channel sticky slide is stuck unto the coverslip. (5) Fibronectin, a protein of the extracellular matrix, is added to make the areas without PLL-PEG cell-adhesive. (6) The six-channel slide is washed with phosphate-buffered saline. Please click here to view a larger version of this figure.
2. Cell seeding (Figure 2A)
NOTE: For the following washing steps, add the respective liquid to one reservoir and then remove an equal volume of liquid from the opposite reservoir of a channel.
Figure 4: Cellular self-organization and quality control of µPIPP array. (A) The microstructured surface consists of squared FN-coated adhesion spots shown in red surrounded by a cell-repellent polymer. (B) After cell seeding, the HuH7 cells are randomly distributed and (C) adhere mainly on the adhesion spots over a time period of 4 h. Reprinted with permission 7. Scale bars: 200 µm. Please click here to view a larger version of this figure.
3. Perfusion system (Figure 2A)
NOTE: The use of a perfusion system is only required if reagents or fluorescent markers need to be added during the course of the time-lapse measurement. Depending on your needs, you can connect each channel to a separate perfusion system or connect several channels in series to the same perfusion system. The number of perfusion systems corresponds to the number of independent experimental conditions. Connect the tubes under sterile conditions in a biosafety cabinet and avoid the inclusion of air bubbles in the perfusion system. If no perfusion system is used, add the reagents/markers in a biosafety cabinet before the time-lapse measurement. The perfusion system is in-house fabricated, the used material is listed in the Table of Materials. The assembly of the perfusion system has been described previously9.
4. Time-lapse microscopy (Figure 2B)
NOTE: For long-term measurements, maintain a stable temperature of 37 °C and a stable CO2 level. As an alternative to CO2-dependent cell growth medium, use L15 medium for which no gas incubation system is required.
NOTE: For quantitative imaging, use cell growth medium without phenol red during the time-lapse measurement to reduce background fluorescence and use the same settings of the time-lapse protocol as well as the same microscope for technical replicates.
5. Fluorescent marker – mRNA transfection (Figure 2B)
NOTE: For a transfection in two channels connected by a tubing system, a total volume of 600 µL transfection mix is needed (300 µL for one channel). The indicated volumes refer to a transfection in two connected channels.
6. Image analysis and fluorescence readout
Figure 5: Automated image processing of time-lapse image series using PyAMA. (1) Phase-contrast and fluorescence image series for each imaging position are imported. (2) Cell contours are determined by segmentation on the phase-contrast image stack. (3) A background correction is applied to the fluorescence images. (4) The cell contours are tracked over time and pre-selected for export. (5) The fluorescence intensity is integrated based on the tracked cell contours. (6) Single-cell cell areas and integrated fluorescence intensities are evaluated and time courses for each cell are exported. Scale bars: 100 µm. Please click here to view a larger version of this figure.
The LISCA approach enables to efficiently collect fluorescence time courses from single cells. As a representative example we outline how the LISCA method is applied to measure single-cell eGFP expression after transfection. The data of the LISCA experiment is used to assess mRNA delivery kinetics, which is important for the development of efficient mRNA drugs.
In particular we demonstrate the different impact of two lipid-based mRNA delivery systems with respect to the time point of translation onset and the expression rate at the single-cell level. We cultured cells and divided the batch into two populations. One subpopulation was transfected with lipoplexes as described in the protocol section. The other subpopulation was transfected using the same mRNA with the same final mRNA concentration, but with lipid nanoparticles (LNP) as delivery system, which were produced using microfluidic mixing12. Due to a different lipid composition and the different fabrication of the mRNA delivery systems of the lipoplexes and the LNPs we expect an impact on the translation kinetics as the uptake kinetics should be influenced. Using the LISCA method we quantify the time point t0 of translation onset after the transfection and how strong the cells express eGFP, which depends on the product of transfected mRNA molecules m0 and the translation rate kTL. To obtain these two parameters we fit a three-stage translation model as sketched in Figure 6A. After successful release of the mRNA molecules m at time point t0 in the cytosol, the mRNA is translated with rate kTL into unmaturated eGFP G*, which is non-fluorescent. The unmaturated G* matures with rate kM to eGFP G, the fluorescence intensity of which is measured during the time-lapse measurement. The mRNA as well as the (unmaturated and maturated) eGFP degrade over time with respective degradation rates δ and γ. The model is described by ordinary differential equations and the analytical solution for G is used as a model function for parameter estimation. The model function is fitted to each of the single-cell time courses as shown in Figure 6B with example time courses (grey) and the respective fits (green). In Figure 6C we show the histograms of the time point t0 of translation onset and the expression rate m0kTL of lipoplex transfected cells. As both parameters are estimated for each cell, the correlation of these parameters can be analyzed as shown in the scatterplot (Figure 6D, blue data) and can be compared to cells transfected with LNPs (red). As shown in Figure 6D, cells transfected with LNPs show less cell-to-cell variability compared to cells transfected with lipoplexes and the population average shows a faster onset of translation as well as a higher expression rate (thick dots with black outline).
These two data sets are just one example how LISCA can be used to study translation after mRNA transfection. Further investigations can be made for example with regard to mRNA stability dependent on mRNA sequence modifications 14, varied reporter protein stabilities 15, or siRNA mediated mRNA degradation 16.
Figure 6: Data analysis of single-cell eGFP translation kinetics. (A) The translation kinetics of the reporter protein eGFP after mRNA delivery can be described by a three-stage reaction rate equation with the respective parameters. (B) The model is fitted (green traces) to each single-cell eGFP expression time course (gray traces) to estimate the model parameters such as the time point (t0) of transfection onset and the expression rate (m0kTL), two parameters to quantify mRNA delivery efficacy. (C) Histograms of the parameter distribution for the transfection onset time and the expression rate. (D) As the parameters are estimated for each cell, a scatter plot of these parameters shows parameter correlation. The small dots represent the parameters of a single cell. The plot shows cells transfected with mRNA LNPs (red) and mRNA lipoplexes (blue). The thick dots with black outline correspond to the respective population average. Please click here to view a larger version of this figure.
Here we described LISCA as a versatile technique to follow cellular kinetics of intracellular fluorescent labels at the single-cell level. In order to perform a successful LISCA experiment, each of the described steps of the protocol section must be established individually and then all steps must be combined. Each of the three major aspects of LISCA feature crucial steps.
Single-cell microarray fabrication
The quality of the microarray is crucial as the cellular alignment on the microarray is not only important for all further experimental steps but also has influence on the data quality. For this reason, the geometry of the pattern and the fabrication method must be adapted with respect to the used cells. The representative results discussed in this article have been generated with the liver carcinoma cell line HuH7 aligned on a (30 µm)2 square pattern. This pattern geometry is also suitable for other cell lines such as A549 or HEK293. For larger cells such as the BEAS-2B, a larger square pattern with 35 µm edge length and an inter-square distance of 80 µm can be used. The described seeding procedure is optimized for the HuH7 cells but it can be adapted to many other adherent cells17. For example, some cell lines need different sizes of the adhesion spots or need larger spacing between the adhesion spots to avoid cell-cell contacts through elongated cells.
The micropattern should be adjusted such that the adhesion site area meets roughly the average area of a cell in standard culture dishes. The geometry can be round or squared and has no measurable effect of cell viability. Cell motions seem to be more restricted on a squared pattern compared to round pattern, where cells are often seen to rotate. Separate viability testing is recommended when new cell lines are used for the first time. Seeding density needs to be adjusted for specific cell lines to reach a high occupancy of the adhesion spots and to minimize double occupancies of adhesion spots at the same time. Typically, the resulting occupancy of number of cells per adhesion site follows Poisson statistics and is either zero, one or two. Hence total occupancy between 60% and 80% should be aimed at to avoid double occupancies17. The seeding protocol may need to be adapted regarding the number of seeded cells and the time duration between seeding and the first washing step. For example, a washing step 30 min after seeding instead of 1 h (see steps 5 and 6 of protocol section 2 Cell seeding) will reduce the number of double-occupied adhesion spots but will also lower the total number of occupied adhesion spots for HuH7 cells.
As the connection of a tubing system to the channel slides is not mandatory, it is easier to establish the use of a reagent or fluorescent marker without using a tubing system to check for the best suitable concentration and incubation time. After a suitable protocol is established for the reagent/marker of interest, the tubing system can be included in the workflow.
Time-lapse microscopy
Another crucial step are the settings of the time-lapse measurement. For example, the exposure time for the fluorescence marker must be chosen carefully to avoid photobleaching of the fluorophore and phototoxicity effects but still ensure a good fluorescence signal. Another important parameter of the time-lapse setup is the best suitable combination of spatial and temporal resolution, which heavily depends on the observed cellular kinetics. If a high temporal resolution is needed, only a smaller number of positions in the microarray can be scanned between two time points, which reduces the scanned observation area and, thus, also reduces statistics. The observation area is furthermore not only dependent on the number of scanned positions but also on the objective and the size of one field of view of the camera. In the given example, a time resolution of 10 min is sufficient using a 10x objective to observe the translation kinetics. This combination allows to scan 70-100 positions per time point depending on the size of the camera chip, the speed of the microscope stage, and the number of imaging channels (e.g., phase-contrast and eGFP fluorescence).
Image processing
The analysis of the total fluorescence per cell is facilitated as the cells are positioned in an array. Yet, the quality of the single-cell fluorescence time courses, in particular the signal-to-noise ratio (SNR), depends on the algorithm used for the integration of cell fluorescence intensities from the image series. The image processing steps of the software tool PyAMA are shown in Figure 5. The determination of the integration areas for the single-cell fluorescence readout is particularly important because the cell contour of living cells varies with time. PyAMA integrates the fluorescence intensity over a cell contour, which can be determined by a built-in segmentation algorithm. An alternative would be to integrate over fixed boundaries based on the micropattern geometry12,16. The current version of PyAMA performs image segmentation based on a threshold on the standard deviation of pixel neighborhoods of the phase-contrast image channel. Segmentation results can be imported from external software as well. For future versions, native support for segmentation based on machine learning and for integration over fixed boundaries is planned.
PyAMA offers to filter out outlier cells or anomalous time courses using an interface that allows for visual inspection of both the fluorescence image as well as the fluorescence time courses of selected cells (Figure 1C). Examples of cells that may be excluded from analysis are cells that undergo division or apoptosis or that are located outside of an adhesion site. Aggregations of multiple cells erroneously recognized by the tracking algorithm as a single cell also need to be filtered out to ensure that the time courses originate from single cells. PyAMA performs a pre-selection of cells to reduce the amount of manual interaction required for filtering out anomalous cells. The pre-selection can be inspected and corrected by hand before exporting the time courses of the selected cells to compensate for inaccuracies of the pre-selection. The pre-selection of the current version of PyAMA is based on a threshold for the cell size to deselect cell aggregations. For future versions, an additional machine-learning based pre-selection is planned, which allows to account for further criteria including the examples for anomalous cells described above.
In summary, the LISCA approach employs single-cell arrays to efficiently collect single-cell fluorescence time courses. Confining cells to microfabricated adhesion sites facilitates tracking and image analysis. Furthermore, cells are cultured in standardized local microenvironments and, hence, are presented with uniform surface area when exposed to agents like lipid nanoparticles for transfection. This aspect is particularly beneficial when cellular heterogeneity within a population is investigated. The µPIPP technique described here is one of many microfabrication techniques that result in regular microarrays of protein patterns. The reader is referred to literature reviewing microcontact printing, photolithographic approaches and soft lithography as alternative fabrication processes6. Depending on the cell line, the one or the other patterning technique may be preferable. In our experiments, µPIPP technique showed well-positioned cells after seeding of cells due to cellular self-sorting, which relies on residual cell adhesiveness on the PEGylated area so that cells are able to search in random migration and spread on the protein target arrays with differentially larger adhesiveness17.
LISCA allows for the acquisition of large numbers of single-cell time courses of arbitrary fluorescence markers. Analysis of fluorescence signals at the single-cell level, in contrast to bulk experiments, yields pristine single-cell time courses and reveals cell-to-cell variability. Cellular heterogeneity plays an eminent role in cell fate decisions such as apoptosis18,19,20. In this context, we recently extended the LISCA approach using two fluorescence channels allowing for analysis of temporal correlations of two cellular events at any one time indicating particular stages within the cell-death signaling cascade19. In this field of research, LISCA presents itself as an alternative to flow cytometry measurements. While flow cytometry undeniably provides a faster workflow and typically yields larger statistics, the data are acquired at one particular point in time. Full time dependencies yield estimates of kinetics parameters and reveal temporal correlations between different fluorescence signals, which otherwise are difficult to access. In order to use multiple fluorescence channels, the setup requires automated filter wheels or multiple cameras. In this case also single-cell FRET analysis should be feasible and enable time-resolved studies of proximity between fluorescently labeled molecules. One drawback of image-based analysis like LISCA is the labor linked to visual controls and outlier detection. Here machine learning tools could ease data processing and allow fully automated data analysis. In the future, using automated microscopy platforms, rapid array microfabrication and artificial intelligence, the throughput and applicability of LISCA could be substantially increased. Furthermore, subsequent analysis of cells that exhibit unusual fluorescence responses using microfluidic extraction 21and single cell genomics is a frequent requirement in pharmaceutical industry. The protocol presented in this article suits the demand for studying the kinetics of cellular processes with single-cell resolution and adequate statistics.
The authors have nothing to disclose.
This work was supported by grants from the German Science Foundation (DFG) to Collaborative Research Center (SFB) 1032. Support by the German Federal Ministry of Education, Research and Technology (BMBF) under the cooperative project 05K2018-2017-06716 Medisoft as well as a grant from the Bayerische Forschungsstiftung are gratefully acknowledged. Anita Reiser was supported by a DFG Fellowship through the Graduate School of Quantitative Biosciences Munich (QBM).
Adtech Polymer Engineering PTFE Microtubing | Fisher Scientific | 10178071 | |
baking oven | Binder | 9010-0190 | |
CFI Plan Fluor DL 10x | Nikon | MRH20100 | |
Desiccator | Roth | NX07.1 | |
Eclipse Ti-E | Nikon | ||
eGFP mRNA | Trilink | L-7601 | |
Female Luer to Tube Connector | MEDNET | FTL210-6005 | |
Fetal bovine serum | Thermo Fisher | 10270106 | |
Fibronectin | Yo Proteins | 663 | |
Filter set eGFP | AHF | F46-002 | |
Fisherbrand Translucent Platinum-Cured Silicone Tubing | Fisher Scientific | 11768088 | |
HEPES (1 M) | Thermo Fisher | 15630080 | |
Incubation Box | Okolab | OKO-H201 | |
incubator | Binder | 9040-0012 | |
L-15 without phenol red | Thermo Fisher | 21083027 | |
Lipofectamine 2000 | Thermo Fisher | 11668027 | |
Male Luer | in-house fabricated consisting of teflon | ||
Male Luer to Tube Connector | MEDNET | MTLS210-6005 | alternative to in-house fabricated male luers |
NaCl (5 M) | Thermo Fisher | AM9760G | |
Needleless Valve to Male Luer Connector | MEDNET | NVFMLLPC | |
NIS Elements | Nikon | Imaging software Version 5.02.00 | |
NOA81 | Thorlabs | NOA81 | Fast Curing Optical Adhesive for tube system assembly |
Opti-MEM | Thermo Fisher | 31985062 | |
PCO edge 4.2 M-USB-HQ-PCO | pco | ||
Phosphate buffered saline (PBS) | in-house prepared | ||
Plasma Cleaner | Diener Femto | Pico-BRS | |
PLL(20 kDa)-g[3.5]-PEG(2 kDa) | SuSoS AG | ||
silicon wafer mit mircorstructures | in-house fabricated | ||
Sola Light Engine | Lumencor | ||
sticky slide VI 0.4 | ibidi | 80608 | |
Sylgard 184 Silicone Elastomer Kit | Dow Corning | 1673921 | |
Tango 2 | Märzhäuser | 00-24-626-0000 | |
Ultrapure water | in-house prepared | ||
uncoated coverslips | ibidi | 10813 | |
Injekt-F Solo, 1 mL | Omilab | 9166017V | with replacement sporn |