Inhibition of cytoplasmic streaming by cytochalasin D is superior to paraformaldehyde fixation for measuring FRET between fluorescent protein-tagged Golgi components


  • Christian Peter Poulsen,

    1. Department of Plant and Environmental Sciences, University of Copenhagen, Copenhagen, Denmark
    Current affiliation:
    1. Department of Plant and Environmental Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
  • György Vereb,

    1. Department of Biophysics and Cell Biology, and MTA-DE Cell Biology and Signaling Research Group, Medical and Health Science Center, University of Debrecen, Debrecen, Hungary
    Search for more papers by this author
  • Naomi Geshi,

    1. Department of Plant and Environmental Sciences, University of Copenhagen, Copenhagen, Denmark
    Current affiliation:
    1. Department of Plant and Environmental Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
  • Alexander Schulz

    Corresponding author
    • Department of Plant and Environmental Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author

Correspondence to: Alexander Schulz, Department of Plant and Environmental Sciences, University of Copenhagen, Thorvaldsensvej 40, DK-1871 Frederiksberg C, Denmark. E-mail:


Protein–protein interaction at the organelle level can be analyzed by using tagged proteins and assessing Förster resonance energy transfer (FRET) between fluorescent donor and acceptor proteins. Such studies are able to uncover partners in the regulation of proteins and enzymes. However, any organelle movement is an issue for live FRET microscopy, as the observed organelle must not change position during measurement. One of the mobile organelles in plants is the Golgi apparatus following cytoplasmic streaming. It is involved in the decoration of proteins and processing of complex glycan structures for the cell wall. Understanding of these processes is still limited, but evidence is emerging that protein–protein interaction plays a key role in the function of this organelle. In the past, mobile organelles were usually immobilized with paraformaldehyde (PFA) for FRET-based interaction studies. Here, we show that the actin inhibitor Cytochalasin D (CytD) is superior to PFA for immobilization of Golgi stacks in plant cells. Two glycosyltransferases known to interact were tagged with cyan fluorescent protein (CFP) and yellow fluorescent protein (YFP), respectively, coexpressed in Nicotiana benthamiana leaves and analyzed using confocal microscopy and spectral imaging. Fixation with PFA leads to reduced emission intensity when compared to CytD treatment. Furthermore, the calculated FRET efficiency was significantly higher with CytD than with PFA. The documented improvements are beneficial for all methods measuring FRET, where immobilization of the investigated molecules is necessary. It can be expected that FRET measurement in organelles of animal cells will also benefit from the use of inhibitors acting on the cytoskeleton. © 2013 International Society for Advancement of Cytometry

For protein–protein interaction studies in live cells, spatial changes over time such as cytoplasmic streaming are a great challenge. Cytoplasmic streaming is characteristic for intracellular transport in large cells and accompanies cell motility. It leads to the displacements of organelles and RNA processing stacks, mixing of the cytosol and reinforcement of cell polarity [1, 2]. As a consequence, proteins of interest rapidly move through the observed cellular region. The velocity of organelle transport can reach 10 μm s−1; for Golgi stacks and vesicles, values between 4.2 and 7 μm s−1 have been documented in plant cells and on average 2.1 μm s−1 in neuronal axons, respectively [3-5]. In plants, cytoplasmic streaming involves the cytoskeletal actin-filaments and myosin motor proteins, as opposed to animal cells, where microtubules and kinesin or dynein are used [6-8].

The Golgi apparatus of plants is central for glycosylation and secretory processes including synthesis of cell wall polysaccharides during cell differentiation. The importance of protein–protein interaction for the function of Golgi-localized enzymes is emerging [9, 10]. Interactions of proteins can be observed in space and time using confocal microscopy and tagged proteins. However, a simple colocalization test would already be positive, if the distance between two proteins in questions were closer than 250 nm. Considering the size of the Golgi apparatus, most proteins in the lumen of its cisternae and vesicles would appear colocalized, also those which are not interacting. Another streaming-insensitive method to study interactions is bimolecular fluorescence complementation (BiFC) [11]. In BiFC, each protein is fused to an incomplete part of a fluorescent protein and can be complemented to a functional fluorescent protein, when the proteins are interacting. However, the two halves of the fluorescent protein bind irreversibly leading to false-positive results [12] and make it impossible to follow transient interactions. Particularly prone to false BiFC positives is the combination of high expression levels and small volume of the target organelle as given in Golgi stacks and vesicles.

Förster resonance energy transfer (FRET) between proteins that are tagged with fluorescent proteins is a more robust method to assess protein–protein interaction: a significant FRET efficiency can only be measured if the two proteins are closer than 10 nm to each other. Thus, FRET allows discriminating interactions from mere colocalization within a diffraction limited spot. The FRET efficiency is inversely proportional to the sixth power of the distance between the two fluorescent proteins in the range of 1–10 nm [13]. To study the interaction of proteins in vivo, tagging with fluorescent proteins is the method of choice [14, 15]. FRET between CFP and YFP can be measured and analyzed in a broad range of microscopic modalities [16, 17], including the simple approach of acceptor photobleaching [18], the more involved intensity-based ratiometric approach [19], or the technically quite challenging lifetime-based approaches [17, 20].

However, both acceptor photobleaching and fluorescence life time measurements are incompatible with imaging of live cells exhibiting cytoplasmic streaming. If bleaching is performed for only 10 s (which is at the lower limits in our hands), Golgi stacks might have moved already more than 40 μm. Accordingly, emission from Golgi stacks measured after acceptor photobleaching can well originate from a population of stacks that have not been in the field of view and/or focus plane during bleaching. Cytoplasmic streaming could, in principal, be only ignored, if the entire cell was bleached evenly (also along the z-axis), and each of the Golgi stacks were imaged in 3D, identified, tracked, and matched in the before- and after-images. This situation is clearly not given in mature plant cells which easily reach a length of 200–500 μm and a thickness of 50–80 μm. When using a confocal microscope, the bleaching efficiency is highest in the focal plane (of 1 μm thickness) and drastically decreases away from this plane in the z-axis.

Fluorescence lifetime measurements are even more sensitive to streaming, as collection of sufficient photon data from a 256 by 256 pixel frame takes minutes rather than seconds. Of the most used FRET evaluation methods, only the ratiometric (intensity-based) quantitative approach [19] would be applicable at speeds sufficient for live cells showing cytoplasmic streaming; however, the strong variation of correction (spillage) factors in space and time owed to the high native fluorescence characteristic for plants renders this approach less useful.

In the past, cytoplasmic streaming was stopped for FRET measurements by using a fixative such as paraformaldehyde (PFA) and, thus, killing the observed cells (see e.g., [21]. PFA works by crosslinking free amino groups and is hence very efficient in stopping any movement, as well as all processes that involve conformational changes of proteins. We have previously used PFA to show interaction of two Golgi-proteins, ARAD1 and ARAD2, forming homo- and hetero-dimers using BiFC and nonreducing gel electrophoresis as well as FRET [22]. Even though FRET clearly allowed discriminating interacting from noninteracting proteins, we found the need to improve the sensitivity of the protocol. Obviously, aldehyde-based fixatives are prone to introduce autofluorescence and quench the fluorescence from reporter proteins [23, 24]. We now explore an alternative method, Cytochalasin D (CytD) for stopping the cytoplasmic streaming of Golgi structures. CytD is a fungal toxin which binds to the barbed end of actin filaments inhibiting association and dissociation at that end [25]. Applied in an appropriate concentration, CytD immobilizes organelles, but is not expected to affect proteins within the Golgi apparatus. A similar approach was undertaken by Hawes and coworkers [26] with the actin inhibitor latrunculin for FLIM measurements. However, this study does not compare the effects of the actin inhibitor and PFA, nor does it analyze the spectral properties of donor and acceptor and their change due to acceptor bleaching.

In the present work, we compare PFA and CytD using a detailed spectral analysis of acceptor photobleaching. Spectral imaging allows compensating for channel bleed-through during FRET imaging of fluorescent proteins and is important in plants as it allows to compensate for background autofluorescence [17, 27]. Moreover, it allows testing whether CytD changes the spectral properties of the fluorescent proteins. We show that CytD is able to immobilize Golgi stacks for FRET measurements. Furthermore, unlike PFA fixation, immobilization of Golgi stacks with CytD preserves the fluorescence intensity of the proteins used and yields higher FRET efficiency values. The detailed spectral comparison clearly showed that PFA is prone to give more artifacts than CytD, which does not change the spectral properties of fluorescent proteins. In conclusion, CytD treatment can be recommended for immobilizing organelles for FRET acceptor photobleaching and can be applied with standard confocal laser scanning microscopy and a simple setup of band-pass filters, i.e., without the need to do spectral analysis.

Materials and Methods


Recombinant fusion proteins with fluorescent proteins (CFP, YFP) were produced using Gateway LR cloning reactions on ARAD1 and ARAD2 cDNA, provided by Dr. Masood Z. Hadi, Joint BioEnergy Institute, California. Destination vectors used were the pEarleyGate 101 vector for full length C-term YFP fusions and the pEarleyGate 102 vector for full length C-term CFP fusions [28, 29]. GnTI-mRFP was kindly provided by Dr. Ricchard Strasser, University of Natural Resources and Life Sciences, Vienna, Austria

Nicotiana benthamiana leaf transformation

Agrobacterium-mediated transient expression of ARAD1-YFP and ARAD2-CFP followed the procedure described in [22]. Briefly, N. benthamiana leaves were infiltrated with a mix of Agrobacterium lines harboring constructs encoding ARAD1-YFP, ARAD2-CFP, and the silencing inhibitor P19, all under 35S promoter. For Figure 2A and the Supporting Information movies, the Golgi markers N-acetylglucosaminyltransferase GnTI-RFP [30, 31] and STmd-YFP [32] were transiently expressed in N. benthamiana leaves.

Image Acquisition

Two adjacent square leaf cutouts of 1 cm2 were incubated in 0.1 M PBS pH 7.4 with either 4% PFA for 10 min at 4°C or 20 μg/ml CytD for 2 h at 20°C and subsequently washed 2 × 10 min in water. Within the range of CytD concentrations published for plant leaves [33, 34], the smallest concentration that arrested organelle movement was 20 μg/ml, while 10 μg/ml did not stop streaming completely. Leaf cutouts were analyzed using a Leica TCS SP5 inverted confocal laser scanning microscope (CLSM) and a 63× water immersion objective, NA = 1.2. Movies documenting Golgi movement (Supporting Information) are presented as time lapse (seven-fold accelerated). Laser settings were the same as for YFP acquisition described below.

For FRET analysis, image acquisition was performed according to the method described by Sieben et al. [9], but with several adjustments. Figure 1 shows a schematic presentation of image acquisition done with the 458 nm argon laser line for CFP and the 514 nm argon laser line for YFP. These laser lines are standard for confocal microscopes; a 435 nm excitation wavelength at the absorption maximum of CFP would be preferable, but is rarely available. First, a complete emission spectrum was taken with 458 nm excitation and detection between 475 nm and 620 nm with a 20 nm band width and a 5 nm step size for 67.92 s, a laser power setting of 1.2 kW/cm2, and pixel dwell time of 0.01 ms. Then, the acceptor (YFP) was excited with the 514 nm laser, its emission was recorded through a 525–600 nm band-pass filter, with pixel dwell time of 0.005 ms. Five scan cycles were recorded with the laser at 3.1 kW/cm2, then photobleaching was done over 15 scan cycles at 97 kW/cm2, and postbleach intensity recorded again in five consecutive scans at 3.1 kW/cm2. The emission spectrum was also measured after acceptor bleaching, using the same settings as in the prebleach acquisition.

Figure 1.

Schematic presentation of the individual steps of image acquisition. Steps 1 and 5: Excitation with 458 nm laser line and acquisition of the emission spectrum (λ-scan) between 475 nm and 620 nm, 30 frames with 20 nm bandwidth and a 5 nm step size. Steps 2 and 4: excitation with 514 nm laser, 5 frames with emission band pass (BP) of 525–600 nm. Step 3. Bleaching with 514 nm laser line at maximum intensity.

Analysis of Spectral Data from Golgi Bodies

As PFA is claimed to alter native sample fluorescence, as well as to quench fluorochromes [23, 24], the effects of PFA and CytD on CFP and YFP spectra were analyzed in the following way. A region of interest (ROI) was drawn around every circular fluorescent signal in each image corresponding to a Golgi body. Average pixel values for each ROI were processed, and presented in Figure 2 as means for all ROIs. A high amount of endogenous fluorescence is often observed in plant material. Data are presented without subtracting the average background signal, which equals approximately YFP alone post bleaching (Figs. 2D and 2E). Spectra were smoothed by calculating a running average of three neighboring measurement points.

Figure 2.

Leaf epidermal cells of N. benthamiana expressing Golgi-specific proteins fused to RFP, CFP or YFP. (A) Confocal image of Golgi stacks in epidermis cells expressing a GnTI-RFP fusion protein. Bar = 10 μm. (B) Cell walls and the nucleus (N) are visible in the corresponding bright field image. (C) The overlay of confocal and bright field image shows that the Golgi stacks are peripherally located in the highly vacuolated cells. (C)–(H) Spectral analysis of CytD and PFA effects on the emission of fluorescent proteins. ARAD2-CFP and/or ARAD1-YFP were analyzed with confocal laser scanning microscopy. Cells were treated either with 4% PFA or 20 μg/ml CytD; graphs are an average of all recordings. (D), (E), and (F) Fluorescence emission spectrum from Golgi stacks. Excitation was at 458 nm, and detection between 475 nm and 620 nm pre and post acceptor photobleaching. (D) PFA treatment, control with donor or acceptor fluorescent protein only. ARAD2-CFP n = 6, two cells and ARAD1-YFP n = 50, two cells. (E) CytD treatment, control with donor or acceptor fluorescent protein only. ARAD2-CFP n = 11, two cells and ARAD1-YFP n = 22, two cells. (F) Comparison of PFA and CytD, FRET from Golgi stacks from CFP and YFP cotransformed cells. CytD and PFA immobilization with n = 112 in five cells and n = 81 in six cells, respectively. (G) The integrated emission between 475 nm and 505 nm was used to calculate the FRET ratio = (DpostDpre)/Dpost. (H) Control of bleaching efficiency. ARAD1-YFP fluorescence (514 nm excitation, 525–600 nm emission) before and after acceptor photobleaching. Bars in (G) and (H) indicate SEM.

Several criteria were set to ascertain that only immobile fluorescent protein-labeled Golgi stacks are evaluated. For cells coexpressing ARAD1-YFP with ARAD2-CFP, or expressing ARAD2-CFP alone, total emission was calculated for each ROI before and after photobleaching YFP as a sum over the spectrum from 475 nm to 620 nm. The change in emission was calculated and normalized to the postbleach intensity. Bodies that exhibited a change over 50%, indicative for movement during image acquisition, were discarded. Eighty-nine percent Golgi stacks in CytD-treated cells and 55% of the PFA-treated ones fulfilled the set requirements, allowing to calculate the FRET ratio (Fig. 2G) as (DpostDpre) / Dpost, where D is the sum of donor emission intensities between 475 nm and 505 nm. This value is proportional to the efficiency in Förster resonance energy transfer.

Bleaching efficiency was determined with cells expressing ARAD1-YFP alone, using the 514 nm laser line for excitation. The ratio of YFP emission intensities at 520 nm and 540 nm is expected to be approximately 1 (see Spectraviewer, This ratio was calculated for each ROI before bleaching; and ROIs outside the range between 0.7 and 1.4 were rejected, as they would indicate movement during the recording. Another condition was that emission through the 525–600 nm band-pass filter upon excitation was reduced by at least 60%. The set requirements were matched by 69% and 50% of Golgi stacks treated with CytD and PFA, respectively (Fig. 2H).

Pixel-by-Pixel Analysis of Absolute FRET Efficiencies

For this analysis, donor fluorescence has been integrated by summing the emission images between 475 and 505 nm, while acceptor fluorescence images were obtained as the average of the five images taken before or after photobleaching the acceptor. Calculation of pixel-by-pixel absolute FRET efficiencies was done according to [18] using the ImageJ [35] plugin published therein. The following corrections were used: background subtraction, correction for donor photobleaching during the acquisition cycle (in the range of 1.2–2.5% coherent with the low laser power used), correction for acceptor crosstalk into the donor channel (2.4% for CytD immobilized samples and 0% for PFA-fixed cells), correction for acceptor photoconversion formed upon bleaching (2.6% for CytD immobilized samples and 1.2% for PFA-fixed cells), and correction for the incomplete photobleaching of the acceptor (in the range of 10–40% unbleached fraction).

In the calculated FRET efficiency maps, pixels below −1 and above 1 were rejected. ROIs corresponding to Golgi stacks were defined in the following algorithm: donor images were thresholded, converted to binary, a close and then an open operation performed followed by the automatic watershed implemented in ImageJ, and then segmented. ROIs were then rejected from further analysis if (a) their size was smaller than 4 square pixels, (b) their circularity was below 0.3, (c) the percentage of pixels above background in the ROI changed by more than 20% in post image, (d) over 20% of their pixels showed out-of-range FRET efficiency and was, therefore, set to NaN, and (e) their averaged FRET efficiency was below 0. For valid ROIs, average FRET efficiency was calculated and its distribution function generated for both CytD immobilized and PFA fixed samples.

Results and Discussion

Confocal microscopy and bright field overlays of epidermis cells show Golgi bodies, labeled with a cis-Golgi-specific protein (see [30]), at the periphery of the highly vacuolated epidermis cells of N. benthamiana (Figs. 2A–2C). The epidermis cells are observed in situ where they cover the complex leaf tissue. This compromises the quality of the bright field image including the contrast of nucleus and organelles (Fig. 2B). Transformation of epidermis cells followed the standard infiltration protocol, where attached leaves were infiltrated with Agrobacterium tumefaciens carrying the genes in question under the constitutive 35S promoter and observed 3–4 days later [36]. Cellular expression levels with this method can be balanced by diluting the A. tumefaciens strains. Transfection efficiency is similar in the cells around the point of infiltration. The observed epidermis cells showed rapid movement of Golgi stacks that take part in cytoplasmic streaming (Supporting Information Movie 1). Although in a dynamic association–dissociation equilibrium between molecules carrying donor and acceptor tags, molar excess of the acceptor would increase FRET efficiency [37], for the Golgi compartment this might lead to an overload of fluorescent proteins and even false positive results. Therefore, we aimed at a balanced expression level of donor and acceptor, which proved appropriate for identifying interacting and noninteracting Golgi proteins in our previous paper [22].

In a previous paper, showing FRET between the two interacting Golgi glycosyltransferases ARAD1 and ARAD2, Golgi movement was stopped by fixation with PFA [22], which quenches fluorescent proteins and may cause autofluorescence. Other aldehyde-based fixatives such as glutaraldehyde can furthermore evoke blue autofluorescence, interfering with CFP, and change the emission spectrum [24, 38, 39]. Quenching by fixation cannot simply be compensated for by increasing the excitation laser power, as only a limited amount of tagged protein is available in the small volume of Golgi stacks. Increasing laser power would increase the photobleaching of the CFP donor, which, when mathematically compensated for, increases the uncertainty of the calculated FRET efficiencies [18]. Autofluorescence from aldehyde fixatives and endogenous sources cannot be reduced by tissue pretreatment with light [23], as this would also bleach the FRET donor and acceptor. As an alternative to aldehyde-based fixatives, we have, therefore, adopted the actin inhibitor CytD for immobilizing Golgi stacks and compare its effect on the interaction data for ARAD1/ARAD2. CytD stops polymerization of actin [25] and thus cytoplasmic streaming (Supporting Information Movie 2). The Supporting Information Figure shows that YFP-tagged ARAD1 and ARAD2 both colocalize with the Golgi marker STmd-CFP.

A crucial point in the comparison of the two immobilizing agents was the question whether CytD shares the disadvantage of PFA in quenching the fluorescent proteins or changing their emission spectra. Therefore, we performed a spectral analysis in situ to be able to quantify contribution of quenching to the known spectra of CFP and YFP in the background of endogenous autofluorescence. All comparative recordings were done with identical excitation and emission settings. The intensity data are not normalized; differences in the height of the emission spectra reflect that PFA fixation quenches both CFP and YFP emission considerably (about 25% at the emission peaks of CFP and YFP) when compared to CytD treatment (Figs. 2D–2F, 2H). It should be noted that the spectra are significantly different from standard in vitro curves of pure CFP and YFP in water at pH 7.3, due to the difference in pH and ion composition in Golgi bodies, the contribution of endogenous fluorescence from, and the unspecific absorption of emission by cell wall and vacuoles. Since upon PFA fixation, the donor emission spectrum is reduced disproportionally more around the acceptor absorption region (Fig. 2D), a decreased overlap integral will cause the underestimation of FRET efficiency. Furthermore, as apparent in Figure 2F, CFP emission is quenched to a greater extent by PFA than YFP emission, so in the acceptor photobleaching protocol dequenching will be underestimated. The origin of spectral changes introduced by PFA are mainly manifested in quenching, where the alteration of the position, mobility, and chemical nature of amino acid residues of the fluorescent protein are most likely involved. PFA's autofluorescence might contribute to these spectral changes.

For a thorough comparison of the two protocols, we have first scrutinized the imaging conditions for acceptor photobleaching in plant leaf cells. A spectral emission analysis allowed us to determine correction factors, estimate FRET, and do a pixel-by-pixel analysis, in order to eventually assess the suitability of PFA and CytD for FRET measurements.

Spectral Analysis to Determine Correction Factors for Acceptor Photobleaching FRET

First, the various possible sources of error in quantitative FRET calculations have been examined using spectral analysis. Figures 2D and 2E show emission scans of Golgi stacks with ARAD2-CFP fusion protein expressed alone as well as ARAD1-YFP expressed alone, either fixed with PFA (Fig. 2D) or immobilized with CytD (Fig. 2E). These controls were done to identify any spectral or photo-physical issue that needs attention for quantitative analysis.

One of the most frequent artifacts in acceptor photobleaching FRET analysis is the photobleaching of the donor during image acquisition and bleaching, which leads to an underestimation of donor dequenching and FRET efficiency. Spectra from CFP-only samples indicate that in spite of reasonably low laser powers used for imaging CFP, there is anyway some photobleaching of this protein (Figs. 2D and 2E). Quantitative determination of the extent of this unwanted process in the AccPbFRET plugin of ImageJ yielded values of 1.2–2.5%. The crosstalk of acceptor into the donor channel can also lead to underestimation of FRET efficiency. While these spectral data do not discernibly show such crosstalk for YFP-only samples, the fact that there is measurable emission of YFP when excited at 458 nm (Figs. 2D and 2E) indicates that there may also be some emission in the 475–505 nm range. In fact, quantitative determination in AccPbFRET showed that for CytD-immobilized Golgi stacks, the average crosstalk to the donor emission range is 2.4% of the signal in the acceptor image with 514 nm excitation. For PFA fixed samples, where fluorescence intensity decreased considerably as compared to CytD immobilization, no crosstalk was observed. Crosstalk could be totally avoided by a shorter excitation wavelength that is not absorbed by YFP. Standard confocal microscopes, however, do not have a suitable laser line at the absorption maximum of CFP. The violet 405 nm line is only absorbed to about 50% by CFP and is, moreover, incompatible with plant material, because it strongly excites endogenous autofluorescence from the cell wall which would bleed through to the donor fluorescence (see e.g., [40]).

It has been reported that CFP can become photo-converted by bleaching at the YFP-excitation wavelength leading to overestimation of FRET [41]. In CFP-only samples, CFP emission was reduced after bleaching, in both cases CytD- and PFA-treated samples, so CFP photoconversion does not appear to be a problem in this experimental setup (Figs. 2D and 2E).

Finally, possible photoconversion of YFP has also been considered [42]. A small increase in emission from 475 to 505 nm post bleaching was observed in CytD-treated cells expressing ARAD1-YFP alone. Quantitative analysis in AccPbFRET showed that the proportion of this signal to the original YFP emission intensity was 2.6% for CytD immobilized samples and 1.2% for PFA-fixed cells. In conclusion, the thorough analysis of the changes in spectra relevant to FRET revealed that CytD does not quench emission of the fluorescent proteins.

Rough Estimation of FRET Based on Spectral Images

Next, we have checked for FRET in the emission spectra of double transformed cells (Fig. 2F). Transfection efficiency might be different for donor and acceptor in cells transformed with a mixture of A. tumefaciens strains. However, in the observed range, the fraction of bleached fluorescent proteins is expected to be proportional only to illumination, but not to fluorophore concentration. Accordingly, moderate differences in the expression rates will only have influence on the bleaching depth which our exact calculations method takes into consideration. Furthermore, the calculation algorithm does correct for the degree of bleaching. Spectral changes owed to FRET were recorded by exciting CFP at 458 nm, and detecting emission between 475 nm and 620 nm at 5 nm intervals pre and postbleaching of the acceptor. FRET occurred in both treatments, as seen in the increase of CFP signal (emission maximum at 480 nm) after photobleaching YFP (emission maximum at 535 nm; Fig. 2F). The integrated donor emission was calculated as a sum of intensities between 475 nm and 505 nm before and after YFP bleaching. Changes in this wavelength range were used to calculate a parameter proportional to FRET efficiency (Fig. 2G). The calculated FRET ratios were 5% and 25% for PFA and CytD, respectively. These values are significantly different from 0 in both cases (P = 0.000002). It can be speculated that one reason for the discrepancy between the two methods of immobilization is the reduced molecular mobility in aldehyde-fixed material, both related to intramolecular conformational changes and to protein–protein interactions. Moreover, one cannot ignore the increased quenching in PFA fixed sample, both before and after bleaching the acceptor (Figs. 2D and 2E). Quantifying YFP emission upon excitation at 514 nm clearly shows that also YFP fluorescence is decreased by PFA (Fig. 2H). Together with a low number of labeled proteins within the limited space of a Golgi stack, quenching alone can lead to measurements just above background and a low signal-to-noise ratio. Improvement of the signal-to-noise ratio indicates that CytD allows analyzing FRET also when the tagged partners occur in small quantities. This is also relevant when FRET is studied using intensity-based, ratiometric approaches [19, 37], a technique that can easily be implemented in wide field and confocal microscopes, as well as for fluorescent lifetime imaging microscopy [20].

Pixel by Pixel Quantitative Analysis of FRET Efficiency

The AccPbFRET plugin [18] for ImageJ was used to calculate FRET efficiency values, applying all the corrections deemed necessary based on spectral analysis. Unwanted bleaching of the donor, acceptor crosstalk, and acceptor photoproduct have all been present, although to a very small extent, hardly altering calculated FRET efficiency values. Incomplete acceptor bleaching was variable, in some acquisition sequences reaching 40%. Exemplary data and the summary of results are presented in Figure 3. Figure 3A is the grey scale image of the donor fluorescence from a double transformed cell. In general, fluorescence values were in a very wide range, in this image up to 170 (with a background level of 14.1). To allow for better visibility, the 8 bit image is contrast stretched on the 14–110 intensity range for presentation purposes only. The area presented corresponds to roughly one cell, showing, in this case, a rather low abundance of Golgi stacks. Figure 3B shows the calculated FRET efficiency map, the 8 bit grey scale representing values between −1 and +1. The distribution histogram in Figure 3C shows how, because of all the various random sources of noise (movement, photon shot noise, and instrument noise) propagating in a Gaussian manner, the calculated values of FRET efficiency show a large dispersion around the mean. Nonetheless, in spite of the technical challenges of such measurements, the pixel by pixel distribution is centered around a median of 0.36 (mean 0.26, mode 0.61).

Figure 3.

Pixel by pixel analysis of FRET. (A) Donor image of a cell showing Golgi stacks. (B) FRET efficiency map of the same cell generated with help of the AccPbFRET plugin for ImageJ (scale bar in (B) 10 μm.) (C) Distribution histogram of FRET efficiency values from ROIs defined as Golgi stacks in (B). (D) Mean FRET efficiencies of pixel averages in CytD immobilized (n = 100) and PFA fixed (n = 41) Golgi stacks. Bars indicate SEM. (E) Distribution of average FRET efficiencies in the Golgi vesicles for the two treatments. Averages below 0 are clipped.

In further analysis, Golgi stacks and their aggregates have been segmented, and mean FRET efficiencies in each stack calculated. Figure 3D presents the average and SEM of these means for all stacks. For CytD immobilized Golgi, FRET efficiency was 0.27 ± 0.01, while in PFA fixed cells this value was 0.17 ± 0.02. Even though the decrease in FRET caused by PFA fixation was not as extensive as that seen for the spectral analysis based on mean raw fluorescence data for each stack (i.e., Fig. 2G), it is abundantly clear that PFA has a negative effect on the measured FRET. This is also supported when plotting the stack by stack distribution of FRET values (Fig. 3E). The most trivial explanation for seeing a higher (still abnormally low) FRET after PFA fixation in this analysis is the very strict rejection criteria applied for each Golgi stack (see Methods section), with rejection of all stacks with mean FRET below 0. While mathematically it is possible to obtain FRET efficiency values below 0, and we do see in the pixel by pixel histograms (Fig. 3C) such values owed to the various sources of noise, biologically this does not make much sense. It is, therefore, reassuring to see that after CytD immobilization—as opposed to PFA fixation—the FRET distribution histogram of stacks (Fig. 3E) flattens out toward 0. Coherent with this, only 25% of Golgi-like ROIs were rejected in total for the CytD immobilized sample, and only 4.5% owed to <0 mean FRET efficiency, while in the PFA fixed samples a total 62% of the stacks were rejected, and of these 35.5% because of <0 mean FRET.

Comparison of the Effects of Cytochalasin D and Paraformaldehyde on Measuring FRET

Fluorescence intensities measured in the presence of CytD were considerably higher than those after PFA fixation, coherent with the assumption that CytD does not interact with the fluorescent proteins themselves, as opposed to PFA, which could change the structure and molecular environment of fluorescent proteins by crosslinking. The higher fluorescence intensities allowed for reducing the excitation laser power to 1.2 kW/cm2, which is 20% of the power needed for reliable fluorescent signals in the previous study [22]. When using this reduced power, the measurement was less affected by unwanted donor photobleaching; however, the inherent noise in the images, as predicted by error propagation, caused a broadening of the pixel-by-pixel FRET efficiency histograms. As a result, some of the segmented Golgi stacks exhibited an average FRET value below 0 (Fig. 3C), and had to be rejected, consequently, from biological interpretation. But while the fraction of stacks rejected for this reason was only 4.5% in CytD immobilized samples, after PFA fixation this proportion was 35.5%. Similarly, when we consider the spectral analysis of these stacks, using PFA as fixative, 45% of the 148 Golgi stacks from 6 different cells were rejected because of anomalous changes in fluorescence intensity during acquisition and bleaching, while out of 125 Golgi stacks from 5 CytD-treated cells only 10% had to be rejected (Fig. 3E). The significant difference in FRET efficiency (Fig. 3D) shows that PFA-fixation leads to a considerable underestimation of protein–protein interaction. It appears that free mobility of the two proteins and the attached fluorophores will increase the chance for FRET. It has previously been shown that the angle between fluorophores influences the energy transfer efficiency in a FRET system [43], which might be arrested in an inefficient conformation when the fluorophores are crosslinked.


The plant Golgi apparatus is the central organelle for biosynthesis and secretory processes of many cell wall polysaccharides and glycoproteins during development. Interactions of proteins are emerging as essential prerequisites for the functionality of the Golgi apparatus. FRET is the most robust method to study protein–protein interaction. However, as the occurrence of FRET is subject to the presence of spectral overlap between donor emission and acceptor excitation, as well as proper orientation of the two fluorophores, the absence or decreased efficiency of FRET is not a necessary indication of the lack of protein interactions. Along these lines, FRET efficiency, which is a numeric characterizing the proportion of donor excited states that relax by transferring the excitation energy to the nearby acceptor in a nonradiative manner, can be taken as an indicator of molecular proximity, but when this number is very low, it may indicate either the lack of proximity, or simply the lack of FRET itself occurring. Consequently, experimental design must be carefully scrutinized and appropriate positive controls added to avoid false conclusion. PFA fixation appears to introduce a design problem, as it seems to alter fluorophore spectra, which directly affect the achievable maximum FRET efficiency, and possibly also influences molecular motions that hinder the rate and multiplicity of donor–acceptor relative orientations appropriate for FRET to occur.

In spite of its widespread use in animal cell studies, FRET has only rarely been used to study plant proteins, and only recently for proteins localized in Golgi bodies due to the mobility of this organelle and challenges of fluorescence acquisition from its small volume [26]. We have utilized an expression system for measuring protein–protein interactions by FRET in plant Golgi stacks [22], and compare here chemical fixation with immobilization of the organelle by a cytoskeleton inhibitor. Given the sensitivity of fluorescent proteins to fixation [24] and mounting media [41], the immobilization of Golgi stacks with Cytochalasin D is superior to chemical fixation for achieving reliable FRET data. This can be applied to interacting proteins in the endomembrane system including ER, Golgi stacks, and secretory stacks or other organelles that are actively moved in the cytoplasm, such as mitochondria, oleosomes and RNA processing bodies [2]. A similar approach can be applied in animal cells where microtubules, and not actin fibers, are involved in organelle movement and streaming [8]. Inhibition of microtubules and motor proteins by inhibitors and antibodies such as in [44, 45] would be promising. However, it is obvious that such an approach cannot be used in those cases where the interaction of cytoskeletal proteins is to be studied, as inhibition itself would interfere with the protein interaction in question and cause artifacts. For the present study, we can exclude the possibility that CytD interference with actin changed the interaction of the ARAD proteins inside the Golgi.

Obviously, not all biological investigations require that the molecular assembly studied be steady; in fact, visible fluorescent proteins find their greatest use in dynamic applications. However, some cellular processes might be fast enough that a proper compromise between shortening exposure time and improving signal-to-noise ratio cannot be reached. This may either happen because of the molecular characteristics of the biological system (as in our case) or the limitations of the available imaging system. In both cases, immobilization is necessary. Our experiments spearhead the notion that blockers of intracellular transport are the better choice, obviating the possibility that fluorescent proteins are compromised by PFA fixation.


Imaging data were collected at the Center for Advanced Bioimaging (CAB) Denmark, University of Copenhagen. We want to thank Johannes Liesche, University of Copenhagen, for critically reading the manuscript.