FLUORESCENT PROTEINS

Distinguishing GFP from Cellular Autofluorescence

by Andrew W. Knight and Nicholas Billinton

logical processes, monitoring subcellular SUMMARY protein localization, distinguishing suc- cessful transfection or reporting on gene Endogenous autofluorescence is a common nuisance that plagues many expression, and its use impinges on al- a researcher using green fluorescent protein (GFP) for biological in- most every area of biological research.1,2 vestigations in cells and organisms. Yet an arsenal of photonic tech- However, unless GFP is very highly ex- niques is available to tackle the problem and to allow effective dis- pressed or densely localized, its fluores- cence signal will invariably be contami- crimination of GFP from autofluorescence. nated with endogenous cellular or media autofluorescence. Researchers often think they have failed he use of green fluorescent protein sesses such favorable properties as low in using GFP as a marker when they have (GFP) from Aequorea victoria has toxicity and high stability. Also, because succeeded. Their problem is simply de- T evolved into one of the most im- it is inherently fluorescent, simple illu- tecting it amid the sea of other fluores- portant practical advances in cell biol- mination with blue light determines its cent species within biological material. ogy in recent years. GFP genes can now presence without the requirement of ad- This natural fluorescence, commonly be cloned and expressed in a diverse range ditional factors. called autofluorescence, is an ever- of cells and organisms from bacteria and These characteristics make GFP a truly present annoyance for biophotonics re- yeast to plants and animals. GFP pos- versatile marker for visualizing physio- searchers who wish to quantify or visu- alize specific fluorescent markers. Its pres- ence often leads to low signal-to-noise ra- tios and loss of contrast and clarity in fluorescence microscope images. Most researchers tackle the problem by trying to optimize the optical filters used for fluorescence excitation and emission. However, autofluorescence spectra are generally broad, extending over several hundred nanometers. Hence, its interfer- ence is often significant at the same emis- sion wavelengths as GFP, making this ap- proach ineffective.

Figure 1. The excitation spectra of fluorescein isothiocyanate and autofluorescence show the principle of the dual-wavelength differential correction method. An argon-ion laser wavelength is used to excite the fluorescent marker and autofluorescence, while a krypton laser wavelength is used to excite autofluorescence alone.

Reprinted from the September/October 2001 issue of Biophotonics International © Laurin Publishing Co. Inc. FLUORESCENT PROTEINS

Careful selection of optical filters is the most common approach to dis- tinguishing GFP from autofluores- cence. Because of the striking spec- troscopic similarity between fluores- cein and GFP, the commonly available filter set for the fluorescent probe fluorescein isothiocyanate (FITC) is most often employed in vi- sualizing GFP, especially under the microscope. And UV filter sets are useful for viewing wild-type and blue-shifted GFP mutants. However, unless GFP expression is very high, these filters are not specific enough to adequately discriminate between the two.

Optimizing optical filters Suppliers such as Chroma Tech- nology Corp. and Omega Optical Inc. market more than 30 filter sets for GFP, each designed for a particular mutant or specific autofluorescent Figure 2. The fluorescence intensity varies with angle, with respect to the plane of species. This diversity highlights the polarization of the excitation light. reasons why most researchers have chosen to optimize their own filter To tackle the problem of autofluores- cence is flavin, a ubiquitous coenzymatic sets from a combination of suppliers: first, cence, it is first useful to have an idea of oxidation reduction, or redox, carrier in- because autofluorescence arises from a its source. Identification of the species volved in the metabolism of most or- disparate and often unknown range of likely responsible for autofluorescence ganisms and a photoreceptor in plants molecules that can vary enormously can enable the researcher to optimize the and fungi. Derivatives of riboflavin are among different cells and species; and experimental conditions to reduce its con- the most intensely fluorescent examples. second, because new genetically engi- centration or suppress its ability to fluo- Other common species include nicoti- neered GFPs with diverse spectroscopic resce. Many cellular metabolites exhibit namide adenine dinucleotide (NADH) properties are continually being devel- autofluorescence (Table 1). Because cel- and its derivatives, which are crucial to oped. lular extracts are often key components many biochemical reactions in most types In general, filters should be carefully of culture media, such media can also be of cell. Less-well-known sources include selected to pick out areas of the spectrum intensely autofluorescent, compounding lipofuscin, a somewhat enigmatic sub- where GFP can be excited, and its fluo- the problem. stance found to positively stain for lipid, rescence transmitted, with greater effi- The most-cited source of autofluores- carbohydrate and protein. ciency than the autofluorescent species.

TABLE 1. Common Sources of Autofluorescence

Autofluorescent Source Typical Emission Typical Excitation Wavelength (nm) Wavelength (nm)

Flavins 520 to 560 380 to 490 NADH and NADPH 440 to 470 360 to 390 Lipofuscins 430 to 670 360 to 490 Advanced glycation end-products (AGEs) 385 to 450 320 to 370 Elastin and collagen 470 to 520 440 to 480 Lignin 530 488 Chlorophyll 685 (740) 488 FLUORESCENT PROTEINS

The choice of filter bandwidth will de- pend on the excitation source and detec- tor characteristics. Wide bandwidths are A used where color images are captured and the sharp, bright green of GFP can be dis- tinguished from the broad-spectrum au- tofluorescence; narrower bandwidths are used only where quantitative fluorescence intensity measurements are made, such as in fluorimetric or flow cytometric meth- ods. A suitable dichroic filter with a cut- off wavelength between the peak excita- tion and emission wavelengths completes the filter set and minimizes crosstalk in- terference.

Dual-wavelength correction Dual-wavelength methods exploit the fact that autofluorescence excitation and emission spectra are generally broad and have few features, whereas the excitation and emission spectra of GFP are relatively narrow and well-defined. The sample is excited sequentially at two wavelengths: Figures 3 A and B. A fluorescence microscope image of yeast cells expressing GFP one to optimally excite the fluorophore, in an autofluorescent medium (A) is corrected with software to remove such as GFP — inevitably along with the autofluorescence to produce a clearer picture of the cells (B). Courtesy of M.G. autofluorescent components of the cell Barker and J.A. Miyan, Manchester University Institute of Science and Technology. — and the other principally to excite only the autofluorescence. The difference be- tween the two fluorescence measurements obtained from illumination at the re- B spective wavelengths is then equal to the fluorescence of GFP alone. Steinkemp and Stewart used two laser wavelengths (Figure 1) and, although based on an FITC-labeled species, their work would be just as applicable to GFP.3 Conversely, it is possible to excite the sample at a single wavelength and mea- sure at two. For example, GFP fluores- cence intensity measured at 530 nm can be corrected for autofluorescence using the intensity measured at 600 nm, while illuminating only at 488 nm. Both of these methods assume that the autofluores- cence intensity is uniform between the two wavelengths; however, if this is not the case, the ratio of the two fluorescent signals can be adjusted to zero by using unlabeled control cells or media. Because it is a relatively large fluo- rophore (27 kDa) and the actual fluo- rophore element is rigidly encapsulated retains a significant degree of polariza- constant with angle. Autofluorescence in within its cylindrical structure, GFP ro- tion (Figure 2). The intensity of fluores- yeast cells, for example, was recently re- tates in free solution at a slow rate com- cence polarized perpendicular to the ex- ported as being largely unpolarized.4 Thus pared with its fluorescence lifetime. This citation light (Iperp) is approximately half the difference between these respective Ϫ results in a large fluorescence anisotropy that still polarized parallel (Ipara) with the measurements (Ipara Iperp) leaves a large that can be exploited in a simple but ef- excitation light. signal for GFP but a much diminished fective discrimination method. If cells In contrast, the isotropic fluorescence of signal for the autofluorescence. containing GFP are excited with plane- a smaller fluorophore — fluorescein, in The beauty of this method is that, be- polarized light, the resulting fluorescence this example — remains approximately cause it relies on a property of light other FLUORESCENT PROTEINS

As the number of biochemical applications than wavelength, it can be applied to the Autofluorescence is often unevenly dis- discrimination of GFP from autofluores- of GFP continues to tributed in samples; it is localized in spe- cence with substantially the same spec- increase, so the cific subcellular regions or structures in tral properties, which would be very dif- commercial drive exists cells and organelles, such as that arising ficult to achieve using conventional op- to develop ever-brighter from lignin in plants and collagen in an- tical filters alone. The method also can imals. Targeted microscopy methods often fluorescent proteins that be rapidly applied to many assay proto- allow the user to focus on just those areas cols simply by inserting inexpensive po- fluoresce at longer where GFP is localized, while avoiding larizing filters into existing laboratory wavelengths, outshining areas showing high autofluorescence. Such equipment. the autofluorescence. techniques include confocal and two-pho- ton laser-scanning microscopy, established Microscopy image correction produced showing the number of pixels methods of obtaining high-resolution Digitally captured fluorescence micro- of each brightness on a scale from 0 to fluorescence images and three-dimen- scope images of GFP-expressing cells or 255. In Figure 3A, an image of GFP-ex- sional reconstructions of biological spec- organisms can often be manipulated using pressing yeast cells in a medium shows imens. The process involves scanning the software to enhance the visualization of intense autofluorescence. Analysis of the sample with a laser beam focused into a GFP through the background autofluo- pixel brightness distribution pattern pro- small spot through a microscope objec- rescence. Many software packages allow duced from this image indicates two clus- tive. A computer then constructs the image simple adjustments of contrast, color bal- ters: one principally from the uniform by measuring fluorescence as a function ance, color transformation and sharpness autofluorescence, and one from the of laser position, in a technique called to improve the image. In addition, auto- brighter cells (Figure 4). This analysis en- optical sectioning. matic edge detection may allow GFP- ables the selection of a threshold value An attractive property of GFP for mi- labeled cell structures to be specifically between the clusters, and the software can croscopists is its tolerance of fixatives such highlighted. The simplest and crudest remove from the image all pixels that fall as formaldehyde and glutaraldehyde, al- method is to convert the image to RGB below this value. The result is the removal lowing its visualization in preserved tis- (red-green-blue) format and select the of autofluorescence and a much clearer sues. However, formaldehyde enhances green channel signal. picture of the fluorescent cells (Figure 3B). the autofluorescence of flavins and can A more precise method uses a facility With proper calibration, integration of diminish the fluorescence of some GFP available in packages such as Corel the brightness distribution also can allow fusion proteins, while others precipitate PhotoPaint, whereby the image is ana- quantitative analysis of fluorescence cap- out, resulting in misleading bright spots lyzed pixel by pixel, and a histogram is tured in microscope images. throughout the cell. When using these fix-

Figure 4. A distribution map was made of the pixel brightness of a section of the image showing GFP- expressing yeast cells against a background of media autofluorescence. FLUORESCENT PROTEINS

Figure 5. The excitation and emission spectra of enhanced blue, cyan, green, yellow and DsRed fluorescent proteins are widely diverse. Courtesy of Clontech Laboratories Inc., Palo Alto, Calif. atives, it is best to thoroughly wash the nique of fluorescence lifetime imaging like structure with the fluorophore ele- sample prior to viewing to remove all microscopy.5 ment protected in the center, GFP is rel- traces of the chemical. GFP is very sensi- Fluorescence quenching is a simple in atively resistant to photobleaching; it pho- tive to some nail polishes used to seal mi- situ method of reducing nonspecific aut- tobleaches at a reported rate of less than croscope slide coverslips, so a molten ofluorescence without the need for addi- half that of fluorescein. Thus, improve- agarose or rubber cement is recom- tional optics or instrumentation. Tissue ments in contrast may be observed if a mended. GFP fluorescence disappears in sections or cells are stained with a dye microscope image is captured several sec- the presence of absolute ethanol, making that quenches the autofluorescence while onds after the illumination has been this an unsuitable fixative as well. allowing visualization of the narrower- switched on. However, one must be care- Mounting and embedding media used band GFP fluorescence. To be effective, ful if this approach is to be used for for sectioning work can be another source the dye should have an absorbance spec- quantitative work, because GFP will also of autofluorescence. Those of natural ori- trum that significantly overlaps the auto- be photobleached to some extent, and gin, such as Canada balsam or glycerin- fluorescence emission. The dye also the rate of photobleaching may vary, de- albumen, are among the worst culprits. should have a low molecular weight and pending on the presence of modulating Therefore, one of the many commercial- high solubility so that it rapidly and evenly species in the surrounding media. ly available synthetic low-fluorescence diffuses throughout the sample. And fi- mounting media should be used. nally, it should have low cytotoxicity. Photo conversion Finding a dye to satisfy all these crite- Any method that enables excitation of, Time-resolved measurements ria is difficult, and therein lies the limi- or emission from, GFP at an alternative Time-resolved fluorescence spec- tation of this technique. Nevertheless, ex- wavelength may be useful in avoiding troscopy is frequently used to quantify amples of dyes that have successfully areas of the spectrum where autofluores- fluorescently labeled species in the pres- quenched autofluorescence include tolu- cence occurs. Wild-type GFP, for exam- ence of nonspecific autofluorescence. In idine blue, methylene blue and trypan ple, shows an excitation peak at 395 nm its simplest form, the researcher would blue, with peak absorbances of 626, with a shoulder at 475 nm. Prolonged ir- use a pulsed light source and, after a short 609/668 and 607 nm, respectively. Any radiation with UV (280 to 395 nm) or interval during which the autofluores- fluorescence emission from these species blue light (up to 490 nm) causes a pho- cence has decayed, measure the fluores- is in the far-red region of the spectrum, toinduced isomerism reaction that pro- cence of the labeled species. This method which can easily be blocked with appro- duces a progressive decrease in the 395- is best applied to species with a long fluo- priate filters. nm absorption peak, with a correspond- rescence lifetime (>15 ns). GFP has a rel- ing increase in the 475-nm peak. atively short fluorescence lifetime, but Photobleaching Fluorescence resonance energy trans- significant differences exist among mu- When viewing GFP-labeled cells under fer (FRET) is a phenomenon whereby one tants. Lifetimes vary from 1.3 ns for cyan the microscope, several workers have no- fluorescent molecule transfers excitation to 2.6 ns for green S65T and 3.7 ns for yel- ticed that the broad-spectrum yellow- energy in a nonradiative way to a fluo- low fluorescent proteins, although these green cellular or media autofluorescence rophore in proximity. The energy transfer values are known to vary with tempera- tends to fade more rapidly than GFP with requires that the emission spectrum of ture, pH and fusion to other proteins. prolonged illumination. This results in the donor fluorophore overlap the exci- This difference in lifetimes has, however, clearer resolution of the GFP, presumably tation spectrum of the acceptor. Chro- allowed the discrimination of three co-ex- caused by photobleaching of the auto- mophore-mutated GFPs make excellent pressed GFPs using the complex tech- fluorescent species. Because of its barrel- FRET pairs. Blue fluorescent protein emit- FLUORESCENT PROTEINS

ting at 466 nm is a good donor for S65T- chemical characteristics. The availability Techniques for Distinguishing GFP from GFP, which absorbs at 488 nm, while cyan of an expanding range of fluorescent pro- Endogenous Autofluorescence” by N. Billin- fluorescent protein emitting at 505 nm is teins derived from GFP means that the ton and A.W. Knight, published in Analy- an excellent donor for yellow fluorescent researcher has the option of selecting a tical Biochemistry, Vol. 291 (2001), pp. protein, which absorbs at 514 nm. marker with fluorescent properties far re- 175-197. The efficiency of the FRET process is moved from that of the interfering auto- highly dependent on the proximity of the fluorescence, although many demonstrate Meet the authors respective fluorophores; thus, the two significantly lower fluorescence efficien- Andrew Knight and Nicholas Billinton are re- fluorescent proteins should be expressed cies (Figure 5). searchers at Manchester University Institute of as a fusion protein tethered by a short As the number of biochemical appli- Science and Technology in Manchester, UK. spacer of less than 20 residues, or with cations of GFP continues to increase, so Knight is in the department of instrumenta- tion and analytical science, and Billinton in short tags that will strongly interact, bring- the commercial drive exists to develop the department of biomolecular sciences. Their ing the proteins close together. This cre- ever-brighter fluorescent proteins that research is supported by Gentronix Ltd. ates a fusion protein that has widely fluoresce at longer wavelengths, outshin- (www.gentronix.co.uk). The authors may be spaced fluorescence excitation and emis- ing the autofluorescence. contacted by e-mail at andrew.knight@gen- sion bands, making the optimization of Until that happens, the above biopho- tronix.co.uk. optical filters for autofluorescence dis- tonic techniques can be employed to dis- crimination much easier and vastly re- tinguish GFP from endogenous autoflu- References ducing crosstalk. Unfortunately, there is orescence. The most appealing methods 1. (1999). GFPs: Methods in Cell Biology, a need for greater genetic modification, are likely to be those that maintain the Vol. 58. First edition. K.F Sullivan and and any biochemical effects that change exquisite advantages of using GFP in the S.A. Kay, eds. Academic Press. the orientation or distance between the first place; namely, the ability for nonde- 2. (1998). GFPs: Proteins, Properties, fluorophores will significantly diminish structive in vivo expression and mea- Applications and Protocols. First edition. the efficiency of the process. surement. These are noninvasive meth- M. Chalfie and S. Kain, eds. Wiley & ods such as dual-wavelength differential Sons. Spectrally different mutants fluorescence correction, fluorescence po- 3. J.A. Steinkamp and C.C. Stewart (1986). The cloning of the wild-type GFP-en- larization and time-resolved techniques. CYTOMETRY, Vol. 7, pp. 566-574. coding gene from Aequorea victoria in the G 4. A.W. Knight et al (2000). ANALYZT, early 1990s enabled the use of mutagen- Vol. 125, pp. 499-506. esis to create variant proteins with altered This article is abridged from “Seeing the 5. R. Pepperkok et al (1999). CURR. fluorescent properties and other bio- Wood Through the Trees: A Review of BIOL., Vol. 9, pp. 269-272.