Quantcast
Skip to content

Spectral Unmixing in Fluorescence Microscopy

Spectral Unmixing

In an ideal world, when using multiple fluorescent molecules as tags in microscopy, these molecules should have distinct and non-overlapping emission spectra. But the world, as you may have noticed, is not ideal in most things—and microscopy is no exception.

In reality, we have to make a series of compromises to avoid getting false positive signals. The first of these—hardly a compromise, in fact—is to be careful about fluorophore choice. When we are doing a double labelling, it makes good sense to choose two fluorophores with spectra widely spaced apart to avoid mixing. Why combine GFP with Rhodamine, when you can combine it with Texas Red?

Things are not always what they seem

But sometimes it is not as simple as that. There is a multitude of fluorophores available in the market today, and you may have to use fluorophores with partially overlapping emission spectra either because a triple (or more) labelling is needed or simply because that’s what is available in your lab. Using very narrow bandpass filters is an obvious solution to the problem—not a perfect one, as you end up discarding a large proportion of the signal but at least it gets rid of the crosstalk. Using a confocal microscope and doing sequential scanning is obviously the best solution: you excite and record one fluorophore at a time. Given the fact that excitation efficiency of a fluorophore decreases rapidly as we move away from the peak, one should be trouble-free with this approach. If you have a freely tunable excitation source (i.e., a white light laser) and fully adjustable detection windows, then you’re luckier than most. You’d have to try really hard to get into trouble but you still can. If your fluorophores are close together, you could get false positive signal even with sequential scanning, by extending the detection window for the one fluorophore into the emission range of the second fluorophore. And there is also sample autofluorescence to worry about.

This cautionary tale should not confuse anyone: using a confocal microscope with enough spectral versatility is the ideal way to record multi-channel fluorescence. To avoid this scenario, just choose detection windows of the appropriate width at your confocal—and then maybe get a little help from the software, which indispensable when using widefield systems.

True colors: spectral unmixing

In some cases, ratiometric techniques—including spectral unmixing—can be used to quantify the fluorescence of each of the fluorophores in a specimen. So it is possible to estimate the crosstalk that occurs when two fluorophores are imaged together, by individually measuring the crosstalk when specimens containing only a single label are imaged. Two excitation wavelengths and two emission detection channels are used, and, after the signal ratio is determined for each probe, the raw data can be recalibrated using this ratio. This approach, however, becomes less useful if the emission spectra of the probes are too close to one another, and also for a larger number of fluorophores.

There is, happily, another, more efficient way to get rid of crosstalk: spectral imaging with linear unmixing or spectral unnmixing.

To be able to implement linear unmixing, you first have to obtain a “lambda stack” (x,y,λ). A region of interest in the x-y dimension must be scanned at various intervals along the wavelength axis to establish the pattern of intensity changes at different emission bands. This way, the emission spectrum of the fluorophores can be determined by plotting the pixel intensity versus the center wavelength of each emission band.

Each fluorophore is a unique snowflake

Each fluorophore has a unique spectral signature that’s independent of the overlap with other fluorophores. This spectral signature can be used to assign the real contribution of the particular fluorophore to each pixel in a lambda stack. The algorithm compares the summed spectra measured in an image against a library of predicted spectra according to the best-fit parameters given by the software. This way the emission light can be unmixed. The individual fluorescent protein emission fingerprints can be used to clearly separate the contribution from each probe and reassign color to regions that would otherwise appear mixed. After the contribution of each spectral component is determined, the lambda stack can then be segregated into individual images for each fluorophore.

Algorithms for linear unmixing are based on the assumption that, in pixels where colocalization occurs, the intensity of the measured emission signal is linearly proportional to the sum of the intensities for each probe. Obviously, this is true as long as there are no saturation phenomena where linearity is compromised. Another case of compromised linearity are images collected from fluorescence resonance energy transfer experiments (FRET), as this phenomenon leads to a reduction of fluorescence intensity of the donor and an increase in the fluorescence intensity of the acceptor.

Obviously, the denser the sampling along the λ axis, the more accurate the profile. The ability to have a freely tunable excitation source and fully adjustable detection windows, as mentioned above, not only makes spectral separation easier in general imaging, but it also gives the capability of generating more precise lambda stacks than what could be achieved with fixed line excitation and fixed bandpass detection. The production of useful fluorescence emission spectral profiles can fail if the instrument is not equipped with the optimum filters. Being able to freely configure the excitation wavelength and the detection wavelength range enables the user to design custom bandpass settings for virtually any fluorophore that exists or will exist in the future.

Spectral Unmixing

Leave a Comment





This site uses Akismet to reduce spam. Learn how your comment data is processed.

Scroll To Top
Share via
Copy link