Multispectral live-cell imaging

Abstract

Fluorescent proteins and vital dyes are invaluable tools for studying dynamic processes within living cells. However, the ability to distinguish more than a few different fluorescent reporters in a single sample is limited by the spectral overlap of available fluorophores. Here, we present a protocol for imaging live cells labeled with six fluorophores simultaneously. A confocal microscope with a spectral detector is used to acquire images, and linear unmixing algorithms are applied to identify the fluorophores present in each pixel of the image. We describe the application of this method to visualize the dynamics of six different organelles, and to quantify the contacts between organelles. However, this method can be used to image any molecule amenable to tagging with a fluorescent probe. Thus, multispectral live-cell imaging is a powerful tool for systems-level analysis of cellular organization and dynamics.

INTRODUCTION

The discovery and development of genetically encoded fluorescent reporters has revolutionized live cell imaging. However, though the number of spectrally variant fluorescent proteins and vital dyes is large and ever increasing, the ability to distinguish more than a few different fluorescent reporters in a single sample is severely hampered by the considerable spectral overlap in their excitation and emission (R. Neher & Neher, 2004). Fluorescence spectral imaging is a technology that allows many different spectrally variant fluorescent markers to be distinguished in a single sample (Garini, Young, & McNamara, 2006). The most widely used approach for spectral deconvolution of images, called linear unmixing (LU), involves a matrix inverse operation to find the best fit of known fluorophore spectra to that of the recorded spectrum at every pixel in a digital image (R. Neher & Neher, 2004). This and other multispectral acquisition and analysis approaches have been used in commercial instruments to distinguish multiple combinations of organic dyes in fixed microbes (Valm et al., 2011; Valm, Oldenbourg, & Borisy, 2016) and fixed neuronal tissue (Tsuriel, Gudes, Draft, Binshtok, & Lichtman, 2015). Nevertheless, the application of spectral imaging remains underdeveloped in live cell experiments.

There are two general approaches for spectral imaging, emission-based and excitation-based (Figure 1). In emission-based spectral imaging, a spectral detector is used to collect an entire emission spectrum at every pixel in an image (Figure 1a). A LU algorithm based on the emission spectra of the fluorophores is then applied, allowing fluorophores with highly overlapping spectra to be distinguished, even if they are present in the same pixel (Figure 2). Commercially available microscopes generally use this approach. Alternatively, an excitation-based approach can be used (Figure 1b). In this approach, the sample is sequentially excited at different wavelengths. A single image is captured at each excitation wavelength, so little information about the emission spectra of the fluorophores is obtained. Nevertheless, a LU algorithm based on the excitation spectra of the fluorophores can be used to distinguish the fluorophores (Valm et al., 2017). Because most commercially available instruments use an emission-based approach, this unit will provide a detailed protocol for imaging using a confocal microscope with a spectral detector. We will also describe sample preparation, LU, and image analysis. Many of these steps are the same whether an emission- or excitation- based approach is used, and we will point out some special considerations to keep in mind for excitation-based approaches (see strategic planning). The protocol described here is for multispectral imaging of six different organelles: endoplasmic reticulum (ER), mitochondria, Golgi, lysosomes, peroxisomes, and lipid droplets (LDs). However, this method can be modified to visualize any cellular molecules amenable to tagging with a fluorescent protein or organic dye.

Figure 1: Emission- vs. excitation-based spectral imaging approaches.

(a) Schematic of the hardware used for 6-color emission-based confocal microscopy. The specimen is excited using three lasers simultaneously, by point-scanning illumination. Emitted light is collected by a linear array of detector elements after being dispersed by a reflective dispersion grating. (b) Schematic of the hardware used for 6-color excitation-based microscopy. The specimen is excited using six light sources sequentially, for example by light sheet illumination (as in Valm et al. 2017). Emitted light passes through a series of interference filters and is collected using a camera. Image adapted from Valm et al. 2017.

Figure 2: Spectral imaging and LU strategy.

(a) To derive the values for the known fluorophore matrix, images of singly labelled cells were acquired at each wavelength and under the same acquisition conditions used to acquire images of 6-label cells. Intensity values centered at 512 nm and 591 nm were zero for all cells because these detector elements were blocked to prevent scattered laser excitation light from reaching the detector. (b) Graphical representation of the unmixing matrix. The normalized intensity values at each wavelength range from 0 to 1. (c) Published emission spectra for the following fluorophores: CFP, EGFP, YFP, mOrange2 (Shaner, Steinbach, & Tsien, 2005); mApple (Rodriguez et al., 2016); and BODIPY 665/676 (Life Technologies, 2010). (d) Observed emission spectra for the same fluorophores, generated using the images shown in (a). Observed spectra differ slightly from theoretical spectra in (c) due to the influence of detector and objective spectral sensitivity, the discrete sampling of the spectra by the detector, and the influence of physical pins in front of detector elements used to block scattered excitation light. (a-c) Images adapted from Valm et al. 2017.

STRATEGIC PLANNING

Selection of appropriate fluorophores and instrumentation are crucial for successful multispectral imaging. As with any fluorescence-based approach, fluorophores should be chosen for maximum brightness and photostability. In addition, for multispectral imaging, fluorophores must be chosen for optimal separation. The imaging strategy presented here distinguishes fluorophores based on their emission spectra. The error in distinguishing two or more fluorophores with highly overlapping emission spectra using LU depends upon the noise in the recorded spectra as well as the degree to which the fluorophore spectra overlap. Therefore, fluorophores should be chosen so that their emission peaks are as well separated as possible and the brightness of the different fluorophores at their peak emission wavelength should be as similar as possible to each other. The ThermoFisher SpectraViewer (https://www.thermofisher.com/us/en/home/life-science/cell-analysis/labeling-chemistry/fluorescence-spectraviewer.html) and Semrock Searchlight (https://searchlight.semrock.com) are great resources for exploring the spectra of various fluorophores. In addition, Nikon’s MicroscopyU website (https://www.microscopyu.com/techniques/fluorescence/introduction-to-fluorescent-proteins) provides molar extinction coefficients and quantum yields for many fluorescent proteins.

We describe in detail a protocol using five fluorescent proteins (cyan fluorescent protein, CFP; enhanced green fluorescent protein, EGFP; yellow fluorescent protein, YFP; mOrange2; and mApple), and one vital dye (BODIPY 665/676). However, various combinations of fluorophores are possible. We have avoided using fluorophores that are excited by a 405 laser for live cell imaging, due to the high phototoxicity of short wavelength light. If fixed cells are used then this is not a limitation, and an additional excitation wavelength can be added. In addition to preparing multiply-labeled cells for multispectral imaging, note that it is necessary to prepare singly-labeled cells for each of the fluorophores used, in order to acquire reference spectra.

Alternatively to using an emission-based approach, it is possible to distinguish fluorophores based on their excitation spectra. In this case, rather than exciting all fluorophores simultaneously and using a spectral detector, the fluorophores are excited sequentially using a different laser for each fluorophore. Ideally, the absorbance in different parts of the specimen could be measured; however, in practice it is much more practical to measure the emitted light through a filter using available detectors, e.g., cameras. Therefore, the measured quality is a convolution of the excitation and emission curves. When using an excitation-based multispectral approach, it is necessary to choose fluorophores with excitation maxima that are as well separated as possible and whose quantum efficiencies at peak absorbance wavelength are as close as possible.

The LU algorithm requires at least as many channels in the recorded image data as there are different fluorophores to be distinguished in the sample. In this protocol we describe multispectral live cell imaging of organelles. Because many of the organelles move very rapidly, we found that it was critical to excite and image all fluorophores simultaneously when using a laser scanning confocal microscope (Figure 1a). To do this, it is necessary to use a detector with six or more channels. The Zeiss LSM780/880 and Nikon A1 are examples of confocal microscopes capable of 32-channel imaging. For imaging structures that move more slowly, it may be possible to use sequential excitation. In this case, sequential acquisition of two 3-channel images could be used to acquire the six channels necessary for LU.

For live cell, spectral imaging using an excitation-based approach, one must be able to modulate the excitation light source very quickly. To this end, acousto-optical beam splitters (AOBS)s are recommended for fast switching of excitation light and scientific complementary metal-oxide semiconductor (sCMOS) detectors are useful for fast image acquisition (Figure 1b).

After image acquisition and LU, images can be analyzed in a variety of ways. In this protocol, we provide instructions for using organelle analysis code developed in Mathematica (available at: https://sourceforge.net/projects/organelle-interactome/). This code is used to segment the multispectral images, and to measure many organelle properties, including the number and size of organelles. It also calculates the number of contacts between different pairs of organelles (e.g., ER and mitochondria), where a contact is defined as two objects being present in overlapping or adjacent pixels. For the globular organelles (lysosomes, peroxisomes, and LDs), it calculates the fraction of these organelles that contact each of the other labelled organelles. Finally, “model” cells in which the globular organelles are placed at random positions are generated, and the number of organelle contacts in these model cells is calculated. This allows one to compare the frequency of observed organelle contacts with the frequency expected based on a random distribution. Thus, taken together, this protocol outlines a multispectral sample preparation, image acquisition, and image analysis pipeline that can be used to address many biological questions.

BASIC PROTOCOL 1: SAMPLE PREPARATION FOR MULTISPECTRAL LIVE-CELL IMAGING

Materials

4- or 8-well chamber slides (eg. LabTek, ibidi)

Fibronectin from human plasma (Millipore)

Phosphate-buffered saline (PBS, Life Technologies)

Imaging medium (see recipe)

Opti-MEM (Gibco)

Mammalian cells (e.g., Cos-7, ATCC)

Lipofectamine 2000 (Life Technologies)

Purified plasmid DNA:

• -CFP-LAMP1 (CFP fused to lysosomal-associated membrane protein 1)
• -Mito-EGFP (mitochondrial targeting sequence of cytochrome c oxidase subunit VII fused to EGFP)
• -ss-YFP-KDEL (fusion of the bovine prolactin signal sequence with YFP and a KDEL ER retention sequence)
• -mOrange2-SKL (mOrange2 fused to the peroxisome targeting sequence SKL)
• -mApple-SiT (mApple fused to the 15 N-terminal amino acids of sialyltransferase)

NOTE: Many of these plasmids are available from Addgene (https://www.addgene.org). Cerulean and venus are brighter versions of CFP and YFP respectively, and also work well with the other fluorophores in this suite.

Vital dyes:

• -BODIPY 665/676 (Molecular Probes)

Transfecting/labeling cells for multispectral imaging

1. Select fluorophores (see strategic planning above). In this protocol we will provide an example using five fluorescent proteins targeted to different organelles, and the vital dye BODIPY 665/676.

2. Dilute fibronectin (from human plasma) in PBS to a final concentration of 10 μg/ml.

3. Coat chamber slides by aliquoting a sufficient volume of fibronectin (10 μg/ml) into each well to cover the whole surface, e.g., 200 μl for each well of an 8-well slide. Incubate at room temperature for 15-20 min.

4. Aspirate fibronectin. It is possible to save the fibronectin and reuse several times to conserve reagent.

5. Plate cells at the manufacturer’s recommended seeding density. For Cos7 cells, we use 1.5×10⁴ cells/well of an 8-well chamber slide. Incubate overnight at 37°C with 5% CO₂.

6. Transfect cells with 0.05 – 0.2 μg of each plasmid/well of an 8-well chamber slide. It is often necessary to adjust the ratio of the different plasmids so that all the proteins are expressed at roughly the same level. For example, we use:

   • -
     0.20 μg CFP-LAMP1 (lysosomes)
   • -
     0.05 μg mito-EGFP (mitochondria)
   • -
     0.10 μg ss-YFP-KDEL (ER)
   • -
     0.05 μg mOrange2-SKL (peroxisomes)
   • -
     0.10 μg mApple-SiT (Golgi)
   • -
     0.50 μl Lipofectamine 2000 (=1 μl/μg DNA)

7. Combine 100 μl Opti-MEM with Lipofectamine 2000, and 100 μl Opti-MEM with DNA. Incubate at room temperature for 5 min. Combine Lipofectamine and DNA mixtures, and further incubate at room temperature for 15-20 min. Aspirate growth medium from cells in chamber slides, and add transfection mixture. Incubate at 37°C with 5% CO₂ for 4-6 hours.

8. In addition to preparing multiply-transfected cells, prepare a singly-labelled sample for each fluorophore.

   In the case of fluorescent proteins, transfect cells with the same amount of plasmid DNA as is used for multiply-labelled cells, but reduce the amount of Lipofectamine 2000 to maintain a 1:1 ratio of Lipofectamine: DNA.

9. Replace transfection medium with imaging medium.

   It is possible to use different media depending on the cell type, but it is very important to use antibiotic-free medium following transfection. It is also preferable to use phenol red-free medium from this point on.

   For labelling lipid droplets, add BODIPY 665/676 to the medium at a final concentration of 50 ng/ml. Add BODIPY 665/676 to the multiply-labeled samples, and to the BODIPY 665/676 only samples, but not to the cells transfected to express a single fluorescent protein.

10. After replacing the medium, incubate overnight at 37°C with 5% CO₂.

BASIC PROTOCOL 2: MULTISPECTRAL IMAGE ACQUISITION

Materials

Laser scanning confocal microscope equipped with spectral detector and appropriate lasers/dichroics

For live cell imaging: stage-top incubation chamber with CO2

Software for linear unmixing (e.g., Zen, Elements)

Software for spatial deconvolution (e.g., Huygens)

Optimizing multispectral image acquisition

1. Use the eyepiece to find a cell that expresses all five fluorescent proteins (FPs) and is labeled with BODIPY.

   All cells should label with BODIPY. Transfection rates vary greatly depending on the cell line used. Most cells that are transfected express all five FPs, but in some cells one or more of the FPs may be expressed at a significantly higher or lower level than the others. To find cells that express all five FPs at roughly equivalent levels, it is helpful to have filter cubes that will discriminate all five fluorophores.

2. Begin setting the image acquisition parameters in scan mode. Start with the shortest wavelength laser only (458 nm). This excitation wavelength will maximally excite CFP and GFP. Set the laser power such that the dynamic range in the image is maximized but no pixels are saturated in any spectral channel.

   Saturated pixels will cause errors and artefacts in the unmixing result.

3. In scan mode, turn on the next longest wavelength laser (514 nm). This laser will maximally excite YFP but will also excite other FPs in the repertoire. With the 458 nm laser on simultaneously, adjust the 514 nm laser power such that the dynamic range in the image is again maximized with no saturation.

4. Finally, while keeping the 454 nm and 514 nm lasers on, turn on the longest wavelength laser (594 nm) and set its power such that the dynamic range in the image is maximized without any saturation.

5. Set the number of channels to record. When using a 32-element detector, the maximum number of channels is 32. However, some of these channels may be out of the useful range for the specific experiment. For example, the shortest excitation wavelength in this protocol is 454 nm, so it is not useful to collect wavelengths below 454 nm. This is why Figure 2a-b shows 26 rather than 32 channels.

6. Record a single still image of a cell with the empirically derived laser settings, then apply LU to the image (see below) before acquiring time lapse images to determine if all fluorophores are present in the cell and at sufficient expression levels.

Linear unmixing

1. Acquire spectral images of each of the singly labelled-samples (as in Figure 2a).

   Once the laser power settings have been optimized for the 6-labelled cells, images of cells labelled with only one of the six fluorophores should be acquired using exactly the same image acquisition settings as for the 6-labelled cells.

2. Use the microscope unmixing software to extract reference spectra from the singly-labelled cells for each of the six fluorophores used in the experiment.

   The extracted reference spectra may differ from published spectra due to the influence of detector and objective spectral sensitivity, the discrete sampling of the spectra by the detector, and the influence of physical pins in front of detector elements used to block scattered excitation light (see Figure 2c-d).

3. Use the microscope unmixing software to apply LU to the images of six-labeled cells.

4. After unmixing an image of a cell, determine the quality of the image visually. Adjust image acquisition settings if necessary.

   If any of the six fluorophore labels is very weak in intensity or absent from the cell, some back and forth adjustments should be made. It could be that the selected cell did not express a particular FP sufficiently. In this case, try a different cell. Alternatively, the acquisition settings can be adjusted. Namely, the power for the laser that maximally excites the fluorophore that is weak or missing should be increased. If the settings are adjusted, then the acquisition of reference spectra from all six singly-labelled cells must be repeated, and these new reference spectra used for unmixing all images acquired with the adjusted settings.

Spatial deconvolution

1. For computational efficiency, spatial deconvolution should be applied to images after applying LU. After unmixing, images are reduced from 26 channel spectral images to 6 channel unmixed images, greatly reducing the time needed to perform spatial deconvolution.

2. Ideally, spatial deconvolution should be performed using experimentally derived point spread functions (PSF)s from subresolution beads that are labeled with the same fluorophores used in the cellular experiments that have been imaged using the spectral detector and unmixed as for cells. In practice, because it is impractical to make subresolution beads with fluorescent proteins, theoretical PSFs may be used.

BASIC PROTOCOL 3: ANALYSIS OF ORGANELLE INTERACTIONS

Materials

ImageJ software

Mathematica software license

Organelle interactome code, available at: https://sourceforge.net/projects/organelle-interactome/

NOTE: For detailed instructions on using Mathematica code see the readme file and example images available at https://sourceforge.net/projects/organelle-interactome/

Creation of an image mask

1. Open the first frame of the YFP (ER) image of your cell in ImageJ.

2. Use Image→Adjust→Brightness/Contrast… to adjust the brightness so that you can clearly see all of the ER signal. The ER gives a proxy of the boundaries of the cell. DO NOT hit [Apply].

3. Click on the free-form drawing tool and draw a line around the edge of the cell.

4. Hit [Reset] on the B&C dialog window to reset the brightness in the image.

5. Hit Control+F to fill in the cell outline.

6. Close the B&C window. Open the threshold window by clicking Image→Adjust→Threshold… Move the threshold slider so that only the filled in area of the cell mask is brought to foreground.

7. Hit [Apply]. Save the mask image with any name you like, but make sure the last characters in the name are “mask.tif”.

Analysis of organelle interactions

Technical Note: To use the Mathematica organelle analysis code, the unmixed images must be named in the following format:

CFP images: “*ch00.tif”

GFP images: “*ch01.tif”

YFP images: “*ch02.tif”

mOrange images: “*ch03.tif”

mApple images: “*ch04.tif”

BODIPY images: “*ch05.tif”

where the * character represents a wildcard string of any characters.

1. To begin, open the notebook file: “AnnotatedCodeInteractomeCombined.nb” in Mathematica by double clicking on it.

2. A dialog window will open up. Hit [Browse…] to choose the directory where your image files are located, then hit the next [Browse…] choose a directory to save the results file to, and finally enter a name for the result file in the box provided. Then hit [OK].

3. In the next dialog window, use the dropdown menu to select which frame of the images you would like to analyze. Then, enter the number of pixels to dilate the organelle of interest. We suggest using 1 pixel dilation. Finally, enter the number of times to run model images (note, generating model images is computationally intensive and may take a long time and will vary depending upon your computer. Try running the model only 1 time first to see how long the process will take, then try running more than one model to generate large model data sets. When you have entered all the requested information, hit [OK].

4. The next window that opens up is titled, “mOrange Image.” In the center of the dialog window will be an image of the mOrange image channel (peroxisome channel) after applying a segmentation procedure to identify the objects in the image. At the top of the window is a menu bar with five different buttons. Pressing each button runs a different segmentation algorithm on the image. Choose the segmentation algorithm that gives the best result. We suggest using the “Cluster” algorithm for mOrange labeled peroxisomes. When you are satisfied with the result, scroll down to the bottom of the window and hit [OK].

5. The next window will be that for the GFP image and so on for all 6 organelle channels. Repeat step 4 for the GFP, CFP, mApple and BODIPY channels. For the YFP channel (ER), a local adaptive threshold is run. Instead of choosing the segmentation algorithm, you can adjust the neighbourhood size of the local threshold. We suggest starting with a value of ‘50’.

   Examples of ideal segmentation for the six organelles are shown in Figure 4.

6. When all calculations are complete, the results are saved as an excel file in the directory that you selected in step one, and a final dialog window opens up to tell you that the analysis is complete and the results have been saved. Hit [OK].

7. The excel file with all the results contains two columns. The Column ‘A’ is a description of the measurement made, and Column ‘B’ is the computed measurement. The spread sheet contains 335 rows with some breaks in between different categories of measurements.

Figure 4: Segmentation.

Examples of segmentation based on algorithmically-determined, optimal intensity threshold values. Scale bar, 10 μm. Image adapted from Valm et al. (2017).

REAGENTS AND SOLUTIONS

Imaging Medium

DMEM without glutamine and phenol red (Corning, 17-205-CV)

2 mM L-glutamine (Corning, 25-005-CI)

10% fetal bovine serum (Corning, 35-011-CV)

BODIPY 665/676

Prepare a stock solution at a concentration of 5 mg/ml in DMSO. Aliquot and store at −80°C. Perform a serial dilution to obtain a working concentration of 50 ng/ml in imaging medium.

COMMENTARY

Background information

The protocols as written here describe the application of spectral imaging with live cells using a laser scanning confocal microscope equipped with a 32-element spectral detector. However, many different spectral microscope systems suitable for biological imaging are commercially available. A complete discussion of these different systems is beyond the scope of this article; however, parameters that should be considered when choosing a spectral imaging system for live cell imaging include: speed, dynamic range and sensitivity, number of simultaneously collected wavelength channels, ability to acquire optical sections, price, and included spectral analysis software.

As well, the protocols described here used LU to assign fluorophore identity in recorded spectral images. The analysis of spectral image data is an active area of research, and novel algorithms including blind approaches to fluorophore identification are being developed. For example, non-negative matrix factorization may be particularly useful for assigning identity when one or more of the fluoropphores present in a sample is unknown, as is often the case with autofluorescence (R. A. Neher et al., 2009).

Critical Parameters

Several parameters affect transfection efficiency. It may be necessary to optimize cell density, as well as the amount of plasmid DNA and transfection reagent. In addition, it is critical that all the fluorophores in the sample are of comparable brightness (i.e., do not differ by more than a factor of 10). To achieve this, it may be necessary to adjust the ratio of the different plasmids, and/or the concentration and incubation duration of organic dyes. Various media can be used depending on the cell type, but after transfection the medium should be antibiotic-free, and for imaging the medium should not contain phenol red.

For imaging organelles or other structures that move rapidly, it is critical to excite and image all the fluorophores simultaneously. In this protocol, we use fluorophores that can be excited simultaneously using 458, 514, and 594 lasers. Other combinations are also possible, but for live cell imaging it is preferable to avoid using a 405 laser, which exhibits high phototoxicity.

Be sure to prepare samples labeled with each fluorophore individually, in addition to multiply-labeled samples. These are used to acquire reference spectra using the same acquisition parameters (e.g., laser power, gain). During imaging, it may be necessary to go back-and-forth between multiply- and singly-labelled samples until optimal image acquisition parameters are determined. We recommend empirically determining imaging parameters that work for the majority of multiply-labelled cells, and then using these settings to acquire images of singly-labelled samples to generate reference spectra. Subsequently it may be necessary to exclude some multiply-labelled cells because one or more of their fluorophores is too dim or bright when imaged using the “compromise” settings. However, this is more practical than acquiring many sets of reference spectra at different settings.

Anticipated Results

Raw data will be in the form of 6- to 32- channel images. Ideally, for the 6-labelled cells you should observe signal in most of the channels; for the singly-labelled cells, you should observe signal in the channels corresponding to the emission spectra of the fluorophores (Figure 2a). No pixels should be saturated. If the fluorescent proteins are all expressed at roughly equivalent levels and LU is performed correctly, the anticipated result after unmixing is a 6-channel image in which the fluorophores are clearly distinguished (i.e., they localize to separate compartments; see Figure 3). After segmentation using the Mathematica organelle analysis code, the anticipated result is a binary image with the expected morphology for each organelle (see Figure 4).

Figure 3: Anticipated result after LU.

(a) Micrographs of a COS-7 cell expressing LAMP1-CFP, mito-EGFP, ss-YFP-KDEL, mOrange2-SKL, and mApple-SiT, and labelled with BODIPY 665/676. After LU, the six fluorophores are clearly separated. Scale bars, 10 μm. (b) Zoom-up of a region of the cell shown in (a). Scale bars, 5 μm. Image adapted from Valm et al. (2017).

An important step in any quantitative image analysis procedure is to calculate the uncertainty in the result. In the case of spectral imaging, this is critical because the LU operation can introduce artefacts in the computed image in terms of what fluorophores are present in every pixel. However, propagation of uncertainty through the LU operation is difficult and impractical. Therefore, we recommend an empirical approach to estimate the uncertainty in assigning fluorophore identity to every pixel in the image after linear unmixing. With this approach, the singly labeled cells that were imaged exactly as the 6-labeled cells, and from which reference spectra were extracted, are themselves subjected to linear unmixing using all six fluorophore reference spectra. The resulting unmixed image will contain six fluorophore channels for every singly labeled cell, even though the specimens themselves were transfected or labeled with only one FP or dye. The ratio of correctly to incorrectly assigned fluorophores gives an estimate of the error in the unmixing result, and should be similar for the six-labeled cells as long as images of those cells have equivalent signal to noise ratios as the singly-labeled cell images.

Time Considerations

Optimization of cell labeling can take some time, since adjusting plasmid ratios and dye concentration/incubation duration requires some trial and error. Once the labeling protocol is optimized and reference spectra have been acquired, the amount of time required to image a sample for multispectral imaging is similar to other live cell imaging experiments. However, additional time must be allocated for LU and image analysis. It is helpful to have the necessary software installed on a workstation computer, so that the computer used to operate the microscope is not required for image processing.

Table 1.

Troubleshooting guide for live cell multispectral imaging and analysis

Problem	Possible Cause	Solution
Many cells die after transfection	Antibiotics are present in the medium	Use antibiotic-free medium after transfection
Many cells die after transfection	Cell density is too low	Increase the cell density or lower the amount of DNA/lipofectamine
Transfection efficiency is low	Cell density is too high	Decrease cell density or increase the amount of DNA/lipofectamine
Cells die during imaging	pH is not optimal	Use a stage-top incubator with CO2, or use CO2-independent medium
Cells die during imaging	Phototoxicity	Avoid using 405 laser; decrease laser power and increase gain or pixel dwell time to compensate
Reference spectrum is not as expected	Signal-to-noise ratio (SNR) is low	Use averaging to increase SNR
Reference spectrum is not as expected	Fluorophore has different spectra in different environments (e.g., lysosome)	Acquire more than one reference spectrum for the same fluorophore; this will result in additional channels after LU
A fluorophore is not detected in 6-labelled cells	Laser settings are not optimal	Increase power for the laser that maximally excites the missing fluorophore
LU result is not as expected	Fluorophore brightness varies greatly (>10-fold)	Adjust the ratio of plasmid DNA; for organic dyes, adjust either the concentration or incubation time
LU result is not as expected	Objects have moved during sequential excitation	Excite and image all fluorophores simultaneously
Mathematica does not recognize image files	Files are not in the appropriate format	Ensure files are named in the correct format (e.g., *ch00.tif), and that image extension is .tif (not .tiff)

SIGNIFICANCE STATEMENT.

We describe a method for live-cell imaging of cells labeled with up to six fluorophores. We apply this method to study the dynamics and contacts among six organelles: the endoplasmic reticulum, Golgi, mitochondria, peroxisomes, lysosomes, and lipid droplets. Although organelles perform distinct biochemical functions, their activities must be closely coordinated. Thus, multispectral live-cell imaging allows for the systems-level analysis of the organization of organelles in space and time. This method can further be applied to visualize any cellular molecules amenable to tagging with a fluorescent protein or organic dye.