This article is Part I of a two-part series regarding viability, resolution improvement, and measurement in fluorescence imaging. Part II will focus on deconvolution.
The number of fluorescent markers available for use in live cell imaging continues to grow. While the vast increase in the number of these probes brings great opportunities for experimental design and an exciting breadth of potential new research targets, there are drawbacks as well.
Because fluorophores excite and emit light within a limited range of the electromagnetic spectrum, researchers often experience emission wavelength overlap—sometimes called bleedthrough—with their fluorescent dyes. When bleedthrough occurs, a marker from one channel appears erroneously in the image data from another channel of the experiment. For instance, there might be bleedthrough between green fluorescent protein (GFP), which has an emission peak of around 509nm, and yellow fluorescent protein (YFP), whose peak is around 527nm. For researchers interested in the co-occurrence or colocalization of probes, this becomes highly problematic: emission overlap can make it seem as though two unrelated markers are highly correlated when they are not.
Figure 1A: HeLa cells double-labeled with green fluorescence protein (GFP) exclusively in the nucleus and yellow fluorescence protein (YFP) exclusively in the cytoplasm. Original image; (bottom) colocalization scatterplot. (Source: Olympus Corporation, Tokyo, Japan) |
Minimizing spectral bleedthrough
Spectral bleedthrough occurs largely because researchers seek fluorescent tags with the sensitivity required to produce sufficient signal-to-noise ratio at multiple wavelengths to provide for image comparison while doing as little damage as possible to their living cells. Sometimes, wider bandpass emission filters are used to enable adequate data collection while also allowing for sample viability. Some researchers use dedicated hardware to enable hyperspectral separation of emission wavelength data, but this solution, although useful, is often quite costly.
Another more common solution to bleedthrough is the use of bandpass filters that are narrow enough to block unwanted fluorescence emission from each imaging channel. Unfortunately, this reduces the overall number of photons reaching the detector and so requires longer exposure times. Researchers can turn up the intensity of excitation light to deliver more stimulation to the system and get more photons out in turn, but most cells do not thrive under strong light, especially the ultraviolet (UV) or near-UV light most often used in fluorescence imaging. In fact, the less light they are exposed to, the better their chances of surviving; phototoxicity can damage or even kill cells. This means that narrow-band emission filters work best with those specimens that inherently emit enough light so that the researcher does not have to increase the excitation light to deadly levels.
Linear unmixing: A more cell-friendly solution
Another approach to maintaining the balance between adequate signal and sample viability is spectral unmixing. Spectral unmixing uses either hardware or software to separate fluorescence emission information. Hardware-based systems often rely on prism, grating, acousto-optic tunable filter, or liquid-crystal tunable filter technologies, which are useful but computationally complex; these systems may also require additional hardware to carry out the unmixing operation.
It is possible, however, to perform linear unmixing in a multiple wavelength system with careful experimental planning and microscope hardware typical to most live-cell imaging laboratories (Figure 1a, 1b). These software-based systems are used widely outside the life sciences to provide bleedthrough-free data with improved sensitivity and better spatial discrimination. They use algorithmic software to separate the color channels and correct for overlap, thus allowing the researcher to use less light to collect the same amount of image data.
Start with reference data
Though this type of unmixing involves no special hardware other than that found in a typical fluorescence laboratory, it does involve doing some preparatory work to establish the level of bleedthrough that appears in reference images. The reference data is used to calculate the likely sources of bleedthrough and correct for them. Reference samples are prepared for each signal source or wavelength expected in the sample. Each of these reference samples must contain only one signal source—for example, if a researcher is imaging a cell line expressing both GFP and YFP, which overlap considerably, samples that contain only one of each of these fusion proteins at a time must be prepared.
In some cases, autofluorescence is a significant problem even with reference samples. When this occurs, a sample containing no experimentally added fluorophores is prepared and reference images acquired using an excitation/emission pair from a spectral range that both matches the autofluorescence signal and is different than the experimental fluorophores. The researcher acquires a full set of reference samples using all of the excitation/emission pairs required for the experimental sample. To help ensure accurate results, researchers collecting reference information try to keep exposure times, gain settings, illumination levels, confocality, and magnification constant with expected experimental parameters.
Figure 1B: HeLa cells double-labeled with green fluorescence protein (GFP) exclusively in the nucleus and yellow fluorescence protein (YFP) exclusively in the cytoplasm. The same image, spectrally unmixed showing differentiation of nuclear/cytoplasmic labeling; (bottom) colocalization scatterplot. (Source: Olympus Corporation, Tokyo, Japan) |
Simple unmixing techniques
There are some basic considerations associated with acquiring images destined for unmixing. First, the researcher needs to sample a clear background region, where signal from the sample does not exist (a confluent field of cells, for instance, is unsuitable for this step in the process). Also, since unmixing operates pixel by pixel, the researcher must also install a zero-pixel-shift excitation/dichroic/emission set on the microscope and adjust Z-dimension (depth) data to assure that there is no pixel shift in the image data in the X, Y, or Z dimensions.
As the researcher proceeds through the specimen to collect Z-axis information, all of the data for each fluorescence channel should be acquired at one time, moving through all the slices in the Z-stack, before moving on to the next channel.
It is tempting to try to speed up acquisition by moving through the specimen layer by layer, collecting all of the imaging data at each depth in turn. While this technique may speed up the process, it has risks. First, slices can be mismatched when returning to the original Z-position. In addition, by repeatedly exposing the sample to each wavelength of excitation light as the researcher moves down in turn through the stack, the sample is repeatedly exposed to the same process of fluorescence stimulation and recovery, and may become photobleached. When imaging, it also is preferable to expose the sample to the longest wavelength first and shortest wavelength last.
When processing image data, the first step is to determine the relative contributions of each fluorophore (and/or intrinsic fluorescence) to each imaging excitation/emission pair. Once this data is collected, pixel-by-pixel unmixing is a relatively simple process, yielding data that is suitable both for morphological measurement and co-occurrence or colocalization analysis.
Some of today’s most advanced commercial image analysis software programs offer spectral unmixing. For instance, the Olympus cellSens Dimension software’s 5D Multidimensional Acquisition module offers a simple, step-by-step interface that enables effortless linear unmixing of a properly collected dataset. Linear unmixing permits the use of fluorochromes where the consequences of emission bleedthrough on sample viability previously hindered these experiments.
Preserving cells
Advances in computer processing power and robust data processing tools for imaging are now providing researchers the opportunity to design experiments that could not have generated satisfactory data in the past. New spectral unmixing software allows the researcher to correct for many signal issues and to conduct experiments with vastly reduced cell mortality as well. The ability to preserve living cells while acquiring usable multichannel fluorescence image data, among the most elusive goals of live cell imaging, is at last accessible to most fluorescence imaging laboratories.