The authors' goal was to develop an experimental system in which undergraduates could perform quantitative inquiry investigations with transgenic plants expressing GFP, thus supporting the educational goals of the National Research Council (1996
). Instruments designed to measure surface GFP expression are expensive and thus inaccessible for many research and most teaching laboratories. To overcome this barrier, the authors wanted to utilize inexpensive techniques to detect and quantify GFP expression. The initial plan was to use commercially available hand-held LED illumination systems designed for GFP detection. However, these devices were not suitable because of non-uniform illumination and filters that produced photographic artefacts (for example, see Fig. E). Consumer camera filters were also not suitable because of numerous photographic artefacts, some of which are presented in Fig. C and D. Dichroic filters were an improvement over consumer photographic filters (Fig. ), but were not effective in blocking blue excitation light (Fig. F and G).
The epifluorescent device illustrated in Fig. works like the filter cubes found in epifluorescent compound microscopes (Ruzin 1999
; Billinton and Knight 2001
), and is designed to be inexpensive while still being effective in imaging red-shifted GFP fluorescence. The device can be attached to most commercial SLR cameras and many dissecting microscopes because it is mounted with a Cokin (Piktus, France) lens filter holder which includes a number of size adapters. The fact that GFP fluorescence was measured in pot-grown plants (Figs and ) in addition to plants growing on agar medium (Fig. ) shows that the device is effective and flexible. Additionally, because of the flexibility of the apparatus design, this technique can be used in many research applications. The spectral properties of the apparatus (Fig. ) should allow it to work with non-plant organisms such as GFP-expressing Xenopus laevis
and Caenorhabditis elegans
. The apparatus has been successfully used with GFP-expressing Danio rerio
along with fluorescent minerals like willemite and calcite [see Additional Information
The device is inexpensive because it was constructed mostly from mass-produced materials. Blue LEDs have been shown to be effective light sources for fluorescent microscopy (Chin-Sang, Website
; Martin et al. 2005
). The light source (Fig. ) contained a Luxeon V royal blue LED. The spectrum is reported to have a range of ~420–500 nm with a single emission peak of 490 nm (Technical Data DS34, Lumileds Lighting), falling within the reported excitation wavelengths of red-shifted GFPs (Chiu et al. 1996
; Zhang et al. 1996
; Patterson et al. 2001
). This lamp has been successfully used to view C. elegans
expressing high levels of GFP without the addition of an excitation filter (Chin-Sang, Website
); however, when the spectrum was observed with a hand-held spectroscope, detectable emission of green light was observed. Thus, an excitation filter was included with the apparatus.
Dichroic filters designed to enhance machine vision devices were used because they are much less expensive than epifluorescent microscope filter sets. The excitation filter (Fig. ) was made of magenta dichroic glass because it does not emit light longer than ~500 nm (Fig. ). The 45° reflective filter (50 mm × 50 mm) was large enough to accommodate the field of view of a 100-mm macro-lens, reflect light with wavelengths <500 nm, and provide additional improvement of the excitation light quality by allowing wavelengths >500 nm to pass through the filter cube. Reflected light with wavelengths <500 nm was blocked by the 45° reflective filter and the long-pass yellow dichroic filter, resulting in images with little or no reflected light and allowing fluorescence caused by GFP and chlorophyll to be detected (Figs H). This combination of filters produced the greatest signal-to-noise ratio in GFP-expressing transgenic arabidopsis (Fig. ). When the green channel pixel intensity of the raw files was measured, the counts observed in GFP-expressing transgenic arabidopsis were 16-fold greater than those observed in the equivalent exposure of wild-type plants. If the red fluorescence caused by chlorophyll is not desired in the photograph, a short-pass cyan dichroic filter can be added (Fig. ) to block wavelengths longer than ~590 nm (Fig. ), thus obscuring the light emitted by chlorophyll (Figs I, and B). Unfortunately, adding the cyan filter reduces the signal-to-noise ratio to 12.5-fold above background (Fig. ).
The performance of the device is comparable to fluorescent dissecting microscopes specially designed to detect GFP (Tritech Research model SMT1-FL). To compare the two instruments, a mixed population of GFP-expressing and wild-type seeds was sown at a high enough density so that two or more germinating seedlings could be viewed in the microscope's ×6 field of view. After photographing using the Canon EOS 30D camera body, the same plants were photographed using the filter cube. The observed signal-to-noise ratios were 4.99 and 5.81, respectively. The differences in the ratios were not statistically significant.
The machine vision filters are not optimized for GFP fluorescence. Thus, the sensitivity of the apparatus can be improved by using dichroic filters specifically made for GFP detection. However, research-grade 45° beam-splitting filters are ~8 times more expensive than the filter used in this study. The barrier filters are ~18 times more expensive. The manufacturing tolerance of the machine vision filters is somewhat broader (±3 % at R50 %) than that of epifluorescence microscope dichroic filters (±2 % at R50 %) (per Technical Support, Edmund Optics Inc., Barrington, NJ, USA). However, the device can be reliably produced as long as spectrophotometric scans of the filters are used to confirm the filters' optical properties. The comparable signal-to-noise ratio of the apparatus and a research-grade fluorescent dissecting microscope show that the use of mass production dichroic filters is a pragmatic compromise between costs and performance.
Both the CMOS and CCD image detectors responded quantitatively to red-shifted GFP fluorescence. When a titration of purified EGFP was photographed with the filter cube (Fig. ), the linear regression of the green channel pixel intensity produced a statistically significant correlation coefficient in response to GFP surface density (ng mm−2) with the regression line passing through a y-intercept of zero. The cameras responded linearly to exposure time as well (Fig. ). When green channel pixel intensity was plotted against GFP surface density by exposure times (ng s mm−2), the resulting regression line was statistically significant. Thus, pixel intensity is directly proportional to both GFP surface density and exposure time (Fig. ) when the protein is contained in a transparent solution.
The implication of these results is that both CMOS and CCD digital SRL cameras might be used to quantify surface GFP fluorescence. However, comparisons between surface fluorescence and in planta
GFP protein concentrations have not been made in this investigation. Other investigators have found linear relationships between surface fluorescence and GFP protein concentrations in plant tissues (Blumenthal et al. 1999
; Harper et al. 1999
; Richards et al. 2003
; Halfhill et al. 2004
). Because both GFP fluorescence and background fluorescence are affected by physiology and development in situ
, standard curves will need to be performed for each experimental system investigated.
Colour CMOS and CCD detectors contain Bayer colour filter arrays that give each pixel some degree of colour specificity by preferentially transmitting green, blue or red photons (Turchetta et al., Website
). The transmittance spectrum of each green pixel filter overlaps the transmittance spectrum of both the red and blue pixel filters in the Canon EOS 30D/40D (Buil, Website
) and the Nikon D80 (Schmitt, Website
). To confirm that EGFP fluorescence is detected primarily in the green channel of digital images, exposure time series of photographs were taken of purified EGFP. When the filtered cube's barrier filter contained a combination of yellow and cyan dichroic glass, GFP fluorescence was only observed in the green channel of the Canon camera (Fig. A). When the cyan dichroic glass was removed, a steeper linear response was observed in the green channel. Additionally, a 6.6-fold smaller linear response was also detected in the red channel (Fig. B). The latter result indicates that some of EGFP's fluorescence was detected by the red channel when the barrier filter only contained yellow dichroic glass. The Nikon camera, in contrast, did not produce a red-channel response when the filter cube only contained a yellow barrier filter (Fig. C), indicating this camera has better colour specificity for EGFP fluorescence.
To ascertain whether chlorophyll fluorescence affects the green channel of the camera, leaf pigments were extracted with methanol and quantified spectrophotometrically (Meeks and Castenholz 1971
). When the green leaf extracts in the wells of microtitre plates were photographed with the filter cube containing a yellow dichroic barrier filter, a positive linear response with exposure time was observed in the red channel of both the Canon and Nikon digital cameras (Fig. ). A slightly negative slope was observed in the green channel of the Canon image files (Fig. A). The negative slope was probably due to chlorophyll light absorption blocking the green background fluorescence caused by the microtitre plates. In subsequent experiments with the Nikon camera (Fig. B), a different style of blackened microtitre plate was used that produced 4-fold less background counts. The sum of these results suggests that fluorescent measurements of red-shifted GFPs can be made from counts observed in the green channel when using SLR digital cameras.
Long exposures should not be taken when making quantitative measurements with digital cameras. Both CMOS and CCD detectors lose colour fidelity when their photodiodes become saturated (Fellers and Davidson, Website
; Tian et al. 2005
). Once saturated, contiguous photodiodes start to accept charges, a phenomenon known as blooming (Fellers and Davidson, Website
) or crosstalk (Tian et al. 2005
). The effect of blooming became evident in Fig. when the green channel mean pixel intensity was >60 000 cpp. Sixteen-bit cameras have a theoretical saturation level of 216
= 65 536 cpp. In the photographs taken with the filter cube that blocked red light with a cyan filter (Fig. B), counts in the red channel were detected in exposures >10 s, which corresponds to the exposure times that the green channel pixel intensities approached 60 000 cpp in the CMOS-containing Canon camera. Additionally, the blue channel started to produce counts in photographs of plants expressing GFP after 30 s of exposure (Fig. A and B) but not in photographs of wild-type plants. This result indicates that the blue photodiodes started to accept charges when the green pixels became saturated. Similar results were observed with a CCD-containing Nikon camera [see Additional Information
]. Therefore, pixel intensity measurements should not be taken when any of the colour channels approach saturation.
Autofluorescence can interfere with the detection of GFP. For example, Zhou et al. (2005)
showed that chlorophyll can completely obscure GFP fluorescence in Medicago truncatula
(alfalfa) and Oryza sativa
(rice). The chlorophyll concentration varies with the age and species of leaf. They proposed that the chlorophylls absorbed the excitation photon, thus reducing GFP fluorescence. Fortunately, mature arabidopsis leaves are not as affected by its chlorophyll as alfalfa (Zhou et al. 2005
). However, materials other than red-fluorescing chlorophyll appear to produce interfering fluorescence in arabidopsis.
Green background fluorescence was detected in the leaves of wild-type arabidopsis (Figs , and ). Since methanol leaf pigment extracts did not produce detectable green fluorescence (Fig. ), substances other than chlorophyll are the likely source of this background. Several substances have been identified as potential sources of unwanted green autofluorescence, including lignin, flavins, nicotinamide-adenine dinucleotide phosphate and aromatic amino acids (Billinton and Knight 2001
). Experiments with variegated varieties of plants (Fig. A) showed increased green fluorescence in the albino regions of the leaves. In arabidopsis, there was 3.06-fold greater green autofluorescence. Similar results were observed in variegated garden sage [see Additional Information
]. Zhou et al. (2005)
showed that ethanol extraction of whole leaves can remove pigments that obscured GFP fluorescence. Green autofluorescence increased by 2.4-fold when ethanol was used to extract leaf pigments from wild-type arabidopsis (Fig. D–F). The increase in autofluorescence could be caused by reducing the interference caused by chlorophyll (Zhou et al. 2005
) or by releasing fluorescent soluble materials (Billinton and Knight 2001
) when cells were ruptured. In either case, these results indicate that materials other than chlorophylls are responsible for endogenous green background fluorescence.
Blumenthal et al. (1999)
have shown that subtracting the in situ
fluorescence spectrum of wild-type tobacco from the fluorescent spectrum of GFP-expressing tobacco produced ‘differential emission spectra’ that were nearly identical to the fluorescent spectrum of purified GFP. This observation indicates that the fluorescence spectrum of GFP-expressing plants is the sum of the GFP fluorescence and its autofluorescence. Thus, the green channel counts observed in negative controls should be subtracted from the counts observed in GFP-expressing plants. Developmental age (Harper and Stewart 2000
; Halfhill et al. 2003
; Hraška et al. 2006, 2008
), environmental growth conditions (Halfhill et al. 2004
) and plant species (Zhou et al. 2005
) can affect the efficiency of GFP fluorescence detection. Thus, controls should be physiologically identical to GFP-expressing plants in terms of genetic background, developmental ages and growth conditions. Care should be taken when conducting experiments involving stress physiology. When leaves become necrotic or chlorotic (Fig. ), green fluorescence becomes evident in wild-type plants. A detailed discussion of the factors to consider when developing controls is presented by Halfhill et al. (2004)
The experiments presented in Figs , and were conducted in a manner that reduced the variability between treatment and control observations. The wild-type plants were the same ecotype as the transgenic lines. The wild-type plants and the GFP-expressing plants were sown in the same pot on the same day and grown at a density where they were not competing for light. To ensure developmental uniformity, all intensity measurements were taken from expanded cotyledons or the first true leaves. As a result, when background counts were subtracted from GFP measurements, the background counts were observed from plant tissues that were nearly identical both physiologically and developmentally.
Non-green leaf tissue can have high levels of green fluorescence (Figs A–C and ). However, some non-green plant tissues can be used in GFP expression experiments. For example, etiolated cotyledons, hypocotyls and roots expressing GFP produce statistically greater green pixel intensities than the corresponding organs of wild-type plants (Fig. ).