We present a simple protocol to obtain fluorescence microscopy movies of growing yeast cells, and a GUI-based software package to extract single-cell time series data. The analysis includes automated lineage and division time assignment integrated with visual inspection and manual curation of tracked data.
Fluorescence time-lapse microscopy has become a powerful tool in the study of many biological processes at the single-cell level. In particular, movies depicting the temporal dependence of gene expression provide insight into the dynamics of its regulation; however, there are many technical challenges to obtaining and analyzing fluorescence movies of single cells. We describe here a simple protocol using a commercially available microfluidic culture device to generate such data, and a MATLAB-based, graphical user interface (GUI) -based software package to quantify the fluorescence images. The software segments and tracks cells, enables the user to visually curate errors in the data, and automatically assigns lineage and division times. The GUI further analyzes the time series to produce whole cell traces as well as their first and second time derivatives. While the software was designed for S. cerevisiae, its modularity and versatility should allow it to serve as a platform for studying other cell types with few modifications.
Single-cell analysis of gene expression has furthered our understanding of many aspects of gene regulation. Static snapshots of fluorescent reporter expression using flow cytometry or microscopy provide useful information on the distribution of single-cell expression, but lack the history and evolution of time series data required to directly inform gene expression dynamics. Fluorescence time-lapse microscopy presents a means to obtain both single-cell measurements and their history. Various experimental and analytical techniques have been developed to obtain and quantify movies of fluorescent reporter expression, thus imparting insights into gene regulation features (see 1 for a review) such as cell-to-cell variation 2,3, bacterial persister formation 4, transcription initiation and elongation 5, transcriptional bursting 6,7, cell-cycle dependence 8,9, and heritability 10. However, obtaining quality single-cell fluorescence time series involves significant technical challenges in culturing a monolayer of cells in a controllable environment and in high-throughput quantification of the acquired fluorescence movies. Here, we describe a procedure to obtain and analyze fluorescence movies of S. cerevisiae with no required experience in cell culture device manufacture or in software development (Figure 1).
First, we detail an example protocol to generate fluorescence time series movies for budding yeast expressing one or more fluorescent reporters. Though customized microfluidic culture chambers have been built and successfully employed previously 11-13, we use a commercially available microfluidic device from CellAsic (Hayward, CA). The system confines cells to monolayer growth and allows continual control of the perfusion environment. The microscopy protocol we present is a simple means to obtain fluorescence movies of budding yeast, but any modified experimental protocol (a customized culture device, alternative media conditions, etc.) yielding similar fluorescence movie data of single yeast cells may be substituted.
Next, we outline the analysis of the movies using a graphical user interface (GUI) -based software package in MATLAB (Mathworks, Natick, MA), dubbed the GUI for Rapid Analysis of Fluorescence Time Series (GRAFTS), to extract time series data for single cells. GRAFTS has similar features to the versatile, open-source software package Cell-ID 14 in segmenting and tracking cells and in extracting fluorescence intensity and geometric information. However, GRAFTS provides important additional features. First, it offers easy interactive editing of segmentation and tracking results to verify data accuracy, rather than just statistical gating of outlier region traces after analysis. Moreover, it extends the analysis to automatically designate lineage and cell-cycle points of interest of budding yeast. Determining when mother and daughter divide to form two independent cell regions is crucial to determining whole cell (mother including any connected bud) measurements throughout the cell cycle 8. The suite consists of three modules to accomplish these tasks. The first segments cell regions based on the contrast between focused and unfocused bright field images, and allows the user to define and visually test segmentation parameters. The second tracks (using Blair and Dufresne’s MATLAB implementation of the Crocker et al. IDL routine, available at: http://physics.georgetown.edu/MATLAB/) and measures cell regions through time; automatically assigns lineages; and enables visual inspection and error correction. A simple plotting GUI is included to query single-cell properties. The third module ascribes bud emergence and division times, and outputs whole cell time series data as well as their first and second time derivatives (as discussed in 9). The analysis module outputs the data as a space-delimited text file for subsequent study in the statistical software of choice. Thus, the package enables the user to extract high quality time series data through a graphical interface. We have used this method to estimate real-time transcription rates in single budding yeast cells as a function of the cell cycle 9. While the modules have been optimized for budding yeast, the parameters or, if necessary, the freely available code may be adapted for other organisms and image types. Segmentation, tracking, and lineage assignment algorithms may be specific to types of imaging assigned and the organism in question. The existing algorithms could be replaced, but still retain the GUI interface that allows user-friendly visual inspection and correction of segmentation and tracking errors that invariably occur with any algorithm.
1. Obtain Fluorescent Microscopy Movies of Single Yeast Cells Growing in a Microfluidic Chamber
2. Format and Segment Data for Tracking Using the FormatData GUI in MATLAB (Figure 2)
3. Track Cells and Lineages through Time with ProcessTimeSeries GUI, and Curate ID and Lineage Assignments (Figure 5)
4. Exporting Data for Analysis
A successfully performed and analyzed experiment will yield mostly continuous time series for single whole cells with realistically assigned bud emergence and division times. As an example, we performed the above protocol with a haploid yeast strain expressing an integrated copy of Cerulean fluorescent protein (CFP) driven by the constitutive ADH1 promoter to observe how growth and global expression may vary over time in single cells (Table 4, Y47). We ran the time series analysis to obtain whole cell (Figure 7A) measurements over time, and a dependence on the cell cycle emerges for both volume and expression as found in 9. After bud formation, the total combined volume (mother + bud) increases more quickly than before bud emergence (consistent with 8,16) while protein concentration, or average intensity, decreases slightly (Figures 7B and 7C). The rise in combined integrated protein (volume x concentration in this case) also accelerates after budding (Figure 7D). Comparing to the mother region alone (Figure 7B&D), these results demonstrate the importance of properly incorporating bud contributions to the whole cell measurement and highlight the need for proper bud formation and division time assignments. As we have reported 9, we can use the differentiated time series to calculate an instantaneous relative mRNA level M(t) and transcription rate A(t), which both also increase after budding (Figure 7E and 7F):
where P(t) is the total protein at time t and γM is the mRNA degradation rate 0.04 min-1 17. These are not absolute quantities as our protein time series are measured in an arbitrary fluorescent unit.
The ability to generate stable time derivatives of measurement time series also benefits kinetic studies. We use a yeast strain expressing an observable, chimerical transcription factor with switchable transcription activity (Table 4, Y962). The master transcription factor of the phosphate starvation pathway, Pho4p, is controlled through nucleo-cytoplasmic localization in response to extracellular phosphate concentratio n18. We replaced the DNA binding domain of Pho4p with tetR, which binds the tet operon (tetO), and C-terminally fused the new transcription factor to a yellow fluorescent protein to create Pho4-tetR-YFP 9. By switching the phosphate level in the perfusion media, we could then toggle expression of an integrated CFP driven by a synthetic promoter composed of 7 tetO binding sites and the CYC1 minimal promoter (7xtetOpr-CFP). A time series analysis shows when the transcription factor is localized to the nucleus (using a fusion of RFP and the nuclear protein Nhp2p to create an RFP-based nuclear submask as in step 2.9-10), and when the target gene starts and stops transcription (Figure 8). By analyzing the derivative series over time, activation and deactivation times of the target promoter in each single cell can be inferred even when the concentration of the stable fluorescent reporter is high.
Figure 1. Schematic of protocol. The protocol describes steps for 1) microfluidic culture, 2) fluorescence, time-lapse microscopy, 3) data analysis, and 4) output of single-cell time series of cell measurements. Text in red indicates information required for data analysis.
Figure 2. FormatData GUI. The interface is used to import, format, and segment image data for later calculations and visual inspection. Numbers indicate the protocol step corresponding to each component. Click here to view larger figure.
Figure 3. SegTest GUI. The interface is used to test cell segmentation parameters and visually confirm segmentation accuracy as in step 2.7.
Figure 4. ColorThreshTest GUI. The interface is used to test the intensity threshold needed to create a subcellular mask from the chosen fluorescence channel as in step 2.10.
Figure 5. ProcessTimeSeries GUI. The interface is used to measure, track, visually inspect, and edit cell regions and lineage assignments. Numbers indicate the protocol step corresponding to each component. Click here to view larger figure.
Figure 6. AnalysisParams GUI. The interface is used to enter parameters required for the time series analysis as in steps 4.3 and 4.4.
Figure 7. Single-cell time series of a haploid yeast expressing ADH1pr-CFP in microfluidic culture over several generations. (A) Bright field (top row) and CFP (bottom row) micrographs are shown at the indicated time points for a cell throughout the experiment. For the segmented mother cell (blue outline) and its buds (green outline), the contiguous whole cell is outlined in red. Raw mother and bud (B) volume and (C) protein concentration time series were smoothed to remove measurement noise, and (D) integrated CFP fluorescence was calculated as the product of volume and concentration. The whole cell (red) trace is extended past division to keep a running total that is easy to fit to a differentiable smoothing spline across divisions (B and D). The (E) relative mRNA level and (F) instantaneous transcription rate are calculated using equations (1-2) and the spline fit in (D).
Figure 8. Promoter transcription rate response to nucleo-cytoplasmic shuttling of a Pho4-YFP fusion in single yeast cells. (A) Micrographs from the four indicated acquisition channels are shown for a cell with a switchable YFP-fused Pho4-tetR transactivator that (B) localizes to the RFP-marked nucleus in response to low phosphate and (C) drives expression of a 7xtetOpr-CFP reporter. (D) The inferred transcription rate changes with localization on average (black line) and at the single-cell level (red and magenta lines, where red represents the contiguous cell region outlined in red in A). Sharp drops in CFP expression in B coincide with cell division when some protein is lost to the daughter cell.
Measurement Name | Description |
Time | Image frame number |
Area | Total number of pixels comprising the region |
Volume | (4/3)πabc, where a = ½ major axis length of region, b = ½ minor axis length, and c = b or ½ “Chamber height” (whichever is smaller) |
(X)_(Y)mean | Mean pixel intensity for the region mask (Y) in color channel (X) |
(X)_(Y)median | Median pixel intensity for the region mask (Y) in color channel (X) |
(X)_(Y)std | Standard deviation of pixel intensities for the region mask (Y) in color channel (X) |
(X)_(Y)iqr | Interquartile range of pixel intensities for the region mask (Y) in color channel (X), 75th percentile – 25th percentile value |
(X)_(Y)T20pct | Mean intensity for the top 20% of pixel intensities for region mask (Y) in color channel (X). Useful in comparison with the next measurement to determine subcellular localization without a submask. |
(X)_(Y)B80pct | Mean intensity for the bottom 80% of pixel intensities for region mask (Y) in color channel (X). Useful in comparison with the previous measurement to determine subcellular localization without a submask. |
(X)_(Y)M50pct | Mean intensity for the middle 50% of pixel intensities for region mask (Y) in color channel (X) |
(X)_cellint | Integrated fluorescence of color channel (X) (mean pixel intensity x volume) for the whole cell (mother and any connected bud) |
(Z)Raw | Raw time series data output by “Time Series Analysis” for variable (Z) |
(Z)Smooth | Spline-smoothed raw time series data output by “Time Series Analysis” for variable (Z) |
(Z)Whole | Whole cell (mother + attached bud (Z)Rawsmooth) time series data output by “Time Series Analysis” for variable (Z) |
(Z)Cont | Whole cell time series data made continuous at divisions (“running total”) output by “Time Series Analysis” for variable (Z) |
(Z)WhSmooth | Spline-smoothed (Z)Whole cell time series data output by “Time Series Analysis” for variable (Z) |
(Z)ddt | First time-derivative of spline-smoothed (Z)Whole cell time series data output by “Time Series Analysis” for variable (Z) |
(Z)d2dt2 | Second time-derivative of spline-smoothed (Z)Whole cell time series data output by “Time Series Analysis” for variable (Z) |
Table 1. Measurement variable nomenclature in ProcessTimeSeries GUI.
Cell ID color | Lineage status |
yellow | mother assigned |
pink | new bud (red line drawn to assigned mother) |
green | no mother assigned |
Table 2. Cell ID color code for lineage assignment status.
File Name | Description |
FormatData.m | Data input GUI that formats movies and segments BF images for processing by ProcessTimeSeries.m |
FormatData.fig | FormatData GUI window layout |
BFsegment.m | BF segmentation module, with extractregions2.m and segmentregions.m |
SegTest.m | Segmentation quality testing GUI |
SegTest.fig | SegTest GUI window layout |
extractregions2.m | First component of BF segmentation module to identify regions |
segmentregions.m | Second component of BF segmentation module to separate and clean cell regions |
color2submask.m | Fluorescence-based submask segmentation module |
ColorThreshTest.m | Fluorescence-based threshold mask testing GUI |
ColorThreshTest.fig | ColorThreshTest GUI window layout |
tiffread2.m | Extracts data *.TIF or *.STK files for use in MATLAB |
imalign.m | Image registration using 2D cross correlation |
ProcessTimeSeries.m | Data processing GUI that tracks regions, assigns lineages, and enables visual correction of errors. Also provides GUI-based plotting capability and outputs whole-cell time series data |
ProcessTimeSeries.fig | ProcessTimeSeries GUI window layout |
cleanMask.m | Eliminates mask regions based on parameters in ProcessTimeSeries GUI |
track.m | Tracks cell regions based on region centroids |
OptimizeLineages.m | Lineage assignment module determines optimal mother-bud relationships based on physical distance and past and future budding events |
getClosestObjects.m | Finds neighbor cells for each new cell ID (bud) |
SelectVariables.m | Measurement selection GUI for choosing variables of interest |
SelectVariables.fig | SelectVariables GUI window layout |
GeneratePlots.m | Plotting GUI for data in ProcessTimeSeries window |
GeneratePlots.fig | GeneratePlots GUI window layout |
AnalyzeCellSeries.m | Time series analysis module determines cell division time and outputs whole cell measurements as described in the text |
assignDivisionsInf2.m | Assigns bud sprout and cell division times |
AnalysisParams.m | Cell time series analysis parameter input |
AnalysisParams.fig | AnalysisParams GUI window layout |
Table 3. GRAFTS file descriptions.
Strain ID | Genotype (W303-based) | Reference |
Y47 | MATa leu2::ADH1pr-CFP-hisG::URA3::kanR::hisG | 19 |
Y962 | MATa ADE+ NHP2::RFP-NAT spl2Δ::LEU2 pho4::TRP1 pho84Δ::klura3 phm4::HIS3MX6 leu2Δ::TEF1m7pr-PHO4ΔP2-tetR-cYFP URA3::7xtetOpr-CFP | 9 |
Table 4. Yeast strains used in this study.
The above protocol describes a simple method to obtain and analyze fluorescence time series data with limited experience in microfluidics or in software development. It allows one to obtain time-lapse fluorescence movies of single yeast cells; extract relevant cell size and expression measurements; curate tracking and lineage assignments; and analyze the behavior of whole cells over time using a commercially available microfluidic culture device and a versatile graphical user interface (GUI). While the experimental, segmentation, and tracking steps have been approached in various ways previously 11,13,14, the above procedure is designed to make these techniques more accessible to a wider subset of the biology community. While the above procedure is optimized for a particular microfluidic culture device, the overall analysis can be adapted to similar devices as off-the-shelf microfluidic culture technologies become cheaper and more customizable. Fundamentally, the segmentation and region measurement algorithms we employ are similar to existing methods 13,14, but the GRAFTS software adds the ability to visually curate region tracking and lineage assignments and to assign cell-cycle phases accurately. Both of these features are crucial to accurately calculating the time-derivatives of data series.
There are several steps in the protocol which significantly impact the quality of the resulting time series data. Initially overloading the culture chamber with cells will decrease perfusion in the chamber (especially noticeable in kinetic experiments where chamber refresh time is important), and overgrowth will occur earlier in the experiment time course. This can best be avoided by starting with lower cell loading pressure for shorter times and gradually increasing one or both until an appropriate cell density is achieved. Stage position selection for imaging is a related and equally important consideration. Knowing the doubling time of the strain, plan how many cells will be in the image frame over time and anticipate how crowding will affect media diffusion. Ten dispersed cells will grow into ten microcolonies, but ten cells initially grouped together will grow into a single, large colony (we have noticed nutrient diffusion limitations at 6 psi to the center of a colony when it is greater than ~14 cells in diameter). Extra effort in the upstream protocol steps also facilitates manual data curation later, and improves the output time series. Ensure the plate is firmly mounted in the stage and use the image registration option in the FormatData GUI to vastly improve the accuracy in automatically tracking single cells through time. Spend time in choosing the best possible segmentation parameters also to reduce the number of errors which require attention. The tracking software works best with slight oversegmentation of cell regions rather than undersegmentation because remerging a split cell is a simpler task than drawing lines to divide regions; however, increased oversegmentation increases errors in lineage assignments due to the presence of false, new regions. Furthermore, be sure that all mother-bud relationships are properly assigned so that measurements for the whole cell will be accurate. If a bud is missed by the segmentation or grows outside of the image boundary, consider ending the mother’s data series to prevent spurious results. This is easily done by changing the mother region to a unique ID (enter ID of “0”) at the erroneous frame, and using the “Delete ALL” feature to remove all future instances of the new ID. Close attention to these steps will greatly increase the accuracy of the raw time series data.
To establish valid interpretation of the data output by pressing “Time Series Analysis”, we recommend a few additional inquiries. Verify bud emergence and division times are accurately determined based on the user-input parameters. Manually record budding and division times for a test movie set and compare to the algorithm results. To visually estimate the time cytokinesis completes between a mother and daughter, look for their boundary to narrow and darken. (NB: A test strain with a fluorescently-marked nucleus aids in manual observation of division time. Another method would be to fluorescently-mark the plasma membrane and look for the gap between mother and daughter cells to close.). Also, check whether total fluorescence for a cell is better calculated as the average region intensity multiplied by the volume (when captured light originates from a thin cross-section of the cell) or as the integrated fluorescence over the region (when captured light originates from the entire cell). If the fluorescence profile across a cell matches the curve of a semi-ellipse described by the major and minor axes of the region 7, captured light originates from the entire cell, and total fluorescence should be calculated as (mean intensity) x (area). In our case at 63X with a depth of field much less than the cell height, the fluorescence profile is relatively flat and total fluorescence is calculated as (mean intensity) x (volume). Another consideration is the photobleaching of the fluorophore. The software does not consider photobleaching in the analysis, and thus the fluorescence time series output represents only the observable reporter. Photobleaching processes depend heavily on the acquisition settings and time interval used to obtain fluorescence images, the nature of the fluorophore, and various other experiment-specific parameters. Photobleaching also may not be well-described by a simple, first-order rate expression, which prevents a general algorithm for its treatment. We therefore do not account for photobleaching in the time series output, but the analytical method may be used to measure the kinetics of this process to aid the user’s interpretation. Lastly, choose smoothing parameters suitable for the observed time series. Ideally, the smoothing splines will eliminate high-frequency measurement noise while preserving real features in the data. The parameters required to achieve this depend on the characteristics of the particular time series and may vary between experiments.
The software was intended to be flexible for analysis of various time-lapse fluorescence microscopy data. Any number of color channels can be included and any may be used to create a sub-region mask. While the algorithms work well for movies of budding yeast cells, many of the functionalities should extend to movies of other organisms as well. Region measurement (excepting volume), tracking, and curation depend only on the cell mask and are thus adaptable. The segmentation, lineage assignment, cell division determination, and time series analysis modules can be independently modified to suit specific needs. In addition, rather than using the included segmentation module, a pre-segmented mask movie can be input, and phase-contrast or differential interference contrast images can be substituted for bright field. The GUIs can then be used primarily to track and curate ID and lineage assignments.
The authors have nothing to disclose.
We thank Emily Jackson, Joshua Zeidman, and Nicholas Wren for comments on the software. This work was funded by GM95733 (to N.M.), BBBE 103316 and MIT startup funds (to N.M.).
Name of Reagent/Material | Company | Catalog Number | Comments |
Y04C Yeast Perfusion Plate | CellAsic | Y04C-02 | |
ONIX Microfluidic Perfusion Platform | CellAsic | EV-262 | |
Axio Observer.Z1 Microscope | Zeiss | ||
Plan-Apochromat 63X/1.40 Oil DIC objective | Zeiss | 440762-9904-000 | |
Cascade II EMCCD camera | Photometrics | ||
Lumen 200 metal-halide arc lamp | PRIOR Scientific | ||
Triple-bandpass dichroic filter cube and excitation and emission filter set | Chroma Technology Corp | set #89006 | Used for YFP (Venus/Citrine), CFP (Cerulean), RFP (mCherry/tdTomato) |
MAC 5000 controller and filter wheels | Ludl Electronic Products | ||
MATLAB R2011a | Mathworks | 64-bit version handles large data files better than 32-bit |