1. Introduction

X-ray crystallography is a popular technique for determining the structure of molecules of biological relevance to atomic resolution. However, it cannot be used to determine structure-function relationships in living cells under physiological conditions. Furthermore, in the case of transmembrane proteins, the extra- and intracellular domains can only be crystallized separately, and even then frequently only as incomplete fragments, whose relative alignment cannot be ascertained.

However, following their identification from crystallographic data, it would often be sufficient to determine the spatial relationships of the molecule’s subdomains in order to determine the protein’s structure-function properties. For this reason, various fluorescence microscopy techniques have been developed to study membrane proteins in vivo. One technique is to measure the angles between subdomains and a substrate using fluorescence polarisation microscopy [1]. If a protein label is excited by polarised light, then its emission will also be polarised and its intensity modulated according to the angles between its absorption and emission dipoles and the substrate. If performed at the ensemble level, then the total emission will be unpolarised by asynchronous protein motion and the angle information lost – it must be performed at the single molecule level. This has been used to study the motor action of myosin [2].

More generally, however, internal domain flexibility and segmental motions result in the determination of a protein’s conformation requiring the measurement of intramolecular distances as well as angles. Fluorescence resonance energy transfer (FRET) microscopy is now a standard tool for investigating inter- and intramolecular distances at the nanometre scale [3]. FRET, indeed fluorescence microscopy, is most commonly practiced at the ensemble level in cells. The calculated distance due to FRET is an average across the ensemble of fluorophore-labelled molecules. The consequence is that uncommon or transient distances are hidden. If the number of fluorophore labels is reduced such that the distances can be determined individually, then the distribution of distances can be determined: this is the single molecule version of FRET microscopy [4]. [The term ‘single molecule’ is used loosely; different authors use it to describe individual complexes with both one and multiple protein molecules of interest, each containing either single or multiple fluorescent tags. We use the term here in its most general form.]

In this paper we present a means of simultaneously measuring distances and angles between subdomains of proteins in live cells. In order to measure distributions of these values, it is important to use some method of multiplexing data collection. We thus describe here a wide-field total internal reflection fluorescence (TIRF) microscope for orientation and FRET analysis.

2. Instrumentation and data analysis

We determine angle information using the method reported by Forkey et al [1]. Consider a fluorophore molecule in a microscope. Figure 1 defines the axes x, y, z and the axial and azimuthal angles θ and ϕ. In a standard epifluorescence microscope, the excitation and emission are polarised solely in the x and y axes and it is hence not possible to determine the axial angle because there is no z component. However, by using TIRF illumination, the evanescent wave is predominately polarised in the z-axis when the beam is p-polarised. If the beam’s propagation direction is in the x-z plane, its s-polarised counterpart is polarised in the y-axis alone. TIRF illumination has a further advantage when imaging membrane proteins in that excitation is restricted to a layer a few hundred nm thick, reducing the level of out-of-focus background fluorescence and allowing single molecules to be detected [5].

 

Fig. 1. Reference frame of molecules in the microscope.

Download Full Size | PDF

The instrument is a development of the multidimensional fluorescence microscope previously reported [6]. Briefly, a laser source is coupled to a home-built objective-type TIRF microscope via a single-mode fibre (Fig. 2). The excitation polarisation is purified using a Glan-Taylor polariser and then rotated between orthogonal polarisations using an achromatic liquid crystal polarisation rotator (Displaytech). The fluorescence is split into orthogonal emission polarisations in each of two wavelength bands, giving four channels which are imaged onto the CCD simultaneously. Polarisation-sensitive dichroic mirrors are used to ensure that the fluorescence polarisation is not affected. The switching of the rotator is synchronized to image acquisition such that alternate frames in a series have the same excitation polarisation.

 

Fig. 2. Schematic of single molecule TIRF microscope for simultaneous orientation and FRET microscopy.

Download Full Size | PDF

Custom-written data analysis routines, incorporating Bayesian segmentation methods, were used to identify fluorescent spot locations, measure their intensities in each of the eight sub-image series, and track them through time. Spatial transformations between the eight channels were determined using images of a coverslip sparsely coated with 50 nm microspheres incorporating multiple dyes (Invitrogen). Intensities were scaled according to the relative transmissions of the filters in the four emission paths. They were then adjusted for bleedthrough as described by Ha et al [7]:

The intensity measured in the short wavelength channels comprises donor fluorescence plus a small acceptor bleedthrough contribution: i ^short=i ^Donor+bi ^Acceptor, where b is a constant and 0<b≪1. Similarly, in the long wavelength channel, i ^long=i ^Acceptor+ai ^Donor(0<a≪1). Rearranging gives i ^Donor=(bi ^long-i ^short)/(ab-1) and i ^Acceptor=(i ^long-ai ^short)/(1-ab). The total corrected donor and acceptor intensities are then given by I ^Donor=(1+a)i ^Donor and I ^Acceptor=(1+b)i ^Acceptor, assuming that the proportion of the total fluorescence blocked by the filters is negligible. a, b were calculated by imaging an excess of the two dyes (Cy3 and Atto 647 for the experiments described below) on separate coverslips excited at 532 nm and then measuring the average intensity in each of the eight channels. Their values were found to be the same for both emission polarisations, allowing the calculations to take the simplified forms above, which do not include bleedthrough of orthogonal polarisation components.

The corrected intensities _iI _j - where the excitation and emission polarisation axes, i, j respectively, are the axes x, y or z - are proportional to the product of the emission and donor absorption probabilities P _e and P _a ^Donor [1], moderated by the efficiency of energy transfer E and quantum yields Φ such that:

${}_i{I}_j^{\mathrm{Donor}}=K\left(l-E\right){\Phi }^{\mathrm{Donor}}{P}_a^{i,\mathrm{Donor}}{P}_e^{j,\mathrm{Donor}}$

${}_i{I}_j^{\mathrm{Acceptor}}=\mathrm{KE}{\Phi }^{\mathrm{Acceptor}}{P}_a^{i,\mathrm{Donnor}}{P}_e^{j,\mathrm{Acceptor}}$

where K is a constant of proportionality and

ω=cos² α=0.08 is a constant describing the orientation of the evanescent electric field in the x-z plane, where α=sin^-1(NA/n)=74° is the half-angle of the collection cone of the objective; n is the refractive index of glass. For each fluorophore,

where C _1-3 are constants associated with the use of a high numerical aperture (NA) objective lens [1]:

$C_1=\frac{5-3\cos \alpha -{\cos }^2\alpha -{\cos }^3\alpha }{8\left(1-\cos \alpha \right)}=0.705$

$C_2=\frac{1-3\cos \alpha +3{\cos }^2\alpha -{\cos }^3\alpha }{24\left(1-\cos \alpha \right)}=0.022$ $C_3=\frac{2-3\cos \alpha +{\cos }^3\alpha }{6\left(1-\cos \alpha \right)}=0.274$

It is possible to simplify these equations if the donor’s absorption and emission dipoles are parallel, i.e. if θ ^Donor _a=θ ^Donor _e and ϕ ^Donor _a=ϕ ^Donor _e. By using the measured intensities and these equations, it is possible to determine best fit values for the angles for each fluorophore and the FRET efficiency between them for each time step, with the best fit value for the constant K determined for each complete trace. We forced θ to lie between 0° and 90° and ϕ between 0° and 180° for reasons of molecular symmetry. Two unique values of the azimuthal dipole orientation ϕ are possible: these angles can only be resolved by acquiring additional images with the TIRF illumination in planes other than the x-z plane.

3. Single molecule imaging of EGFR

We tested the microscope using A431 cells exposed to EGF (epidermal growth factor) labelled with the fluorophores Cy3 (donor) and Atto 647N (acceptor). We selected these dyes not only for their suitability as FRET partners, including high quantum yields (Φ^Cy3=0.24 and Φ^{Atto 647N}=0.65) and resistance to photobleaching, but also because their absorption and emission dipoles are close to parallel. This was determined by measuring their anisotropy in 99% glycerol, yielding values of 0.38 and 0.36 respectively.

2 nM EGF^Cy3 and 2 nM EGF^{Atto 647N} were added to A431 cells cultured on glass coverslips and allowed to bind to EGF receptors (EGFR, also known as ErbB1 and HER1) for 30 minutes at 4 °C. The EGFR system is of particular interest because it is involved in signalling mechanisms that regulate cell growth, differentiation, migration and apoptosis [8]. Previous work suggests that many of the features of EGFR signalling can only be studied under physiological conditions at the single molecule level [9].

Although the analysis software is capable of tracking moving objects, it is unreliable when they move more than the typical spot separation between frames because it becomes difficult to disentangle tracks which cross each other. After rinsing, the cells were therefore lightly fixed with 1% paraformaldehyde. This hinders the diffusion of receptors on the plasma membrane (unpublished observations) whilst still allowing fluctuations in their conformation. The cells were excited with 3.5 mW at 532 nm. Images were integrated for 140 ms and acquired at 4 Hz (i.e. the s, p-excitation polarisation cycle repeats at 2 Hz).

Fluorescent spots were identified within the field of view and their intensity v time traces manually examined for all eight channels. Spots whose intensity was comparable to single or small multiples of molecules, identifiable by the magnitude of photobleaching steps, and which show anticorrelated behaviour between any pair of channels were selected for further analysis. Figures 3 and 4 show data from spots selected to demonstrate the principal features of 8 channel single molecule fluorescence. Firstly, Fig. 3 shows a spot with a relatively low signal-to-noise ratio (SNR). Figure 3 (a) is a video showing the intensity of the spot in each of the 8 channels, together with their intensity v time traces prior to bleedthrough correction. The corrected traces are shown in Fig. 3 (b), along with the FRET efficiency and angles determined as described above. In line with standard practice (e. g. [2]), we have displayed only the azimuthal angle value which lies between 0° and 90°: due to degeneracy, the angle may take the values ϕ or 180°-ϕ. There is anticorrelated behaviour between the donor and acceptor intensity traces due to energy transfer, which is particularly clear in the movie. The angle traces below reveal that the change in efficiency is at least partly due to a molecular rotation rather than change in separation: all three largest peaks in the FRET efficiency trace are correlated with an increase in the donor molecule’s axial angle and decrease in its azimuthal angle, consistent with a rotation between two specific conformations. Note that noise in the measured intensities can give parameter combinations inconsistent with the model, for example ratios between two channel intensities which impossibly require sin² θ _e<0 or >1. The best fit is then one in which sin² θ _e=0 or =1 e θ respectively, causing the angles to be fitted to 0° or 90° at several points. The spot recorded in Fig. 4 has a much higher SNR although it emits for a much shorter period. Study of the video and intensity traces reveals strong variation in both spectral- and polarisation-resolved intensities, which shows the acceptor and donor molecules to be rotating – this behaviour is clearly shown in the angle traces.

 

Fig. 3. Single molecule orientation and FRET data for a spot showing low SNR. (a) (Media 1) showing a small region of interest extracted from the images acquired, with all 8 channels shown. The square superimposed on the image highlights a fluorescent spot detected and the intensity v time traces for this spot are shown below. Images with the two excitation polarisations cannot be acquired simultaneously; the square is green for the ‘current’ image (Media 1). (b) Intensity v time traces following correction for bleedthrough (top panel), FRET efficiency and axial and azimuthal angles for Cy3 and Atto 647N - these latter are only calculated for the part of the trace where a fluorescent spot has been detected.

Download Full Size | PDF

 

Fig. 4. Single molecule orientation and FRET data for a fluorescent spot with high SNR (Media 2). Details as for Fig. 3.

Download Full Size | PDF

In this paper, we have presented the FRET data in terms of the efficiency of transfer, rather than converting them into intermolecular distance. We have taken this approach because such a calculation would include the angles measured as well. At this stage, we have no estimate of the uncertainties associated with the angles calculated, which would be compounded upon determination of such distances. Furthermore, the quantum yield at the single molecule level does not have a unique value [10]. We are currently enhancing our Bayesian analysis methods, which will allow us to determine the distances and angles, and hence protein conformations, within specific confidence limits. It will then also be possible to correlate the FRET efficiency with the orientation; their uncertainties make it similarly inappropriate at present.

We are also working on being able to analyse images where the fluorescent spots move more rapidly. This will allow the technique to be applied to cells with rapidly diffusing membrane proteins under physiological conditions, i. e. without light fixation.

We noted earlier that the azimuthal angles cannot be fully resolved without acquiring additional images - with a consequent reduction in temporal resolution. It is not clear that a complete angular determination is always required. One motivation for performing this work is to provide an input into computational molecular dynamics simulations, for which four angle values is already a major reduction in the number of degrees of freedom. Combining the determined angles for the donor and acceptor molecules in a FRET pair may also permit certain values to be discounted as non-physical.

It is not possible to equate changes in axial angle with changes in the protein-membrane angle because the cell membrane is ruffled and hence not necessarily parallel with the cover slip. We are working on combining this multidimensional microscope with methods to measure the membrane profile in order to resolve this issue.

4. Conclusions

In this paper we have described a method for measuring the orientation of single interacting protein molecules simultaneously with the efficiency of energy transfer between them. We have demonstrated that this can be achieved in cells expressing the membrane protein EGFR with fluorophore-tagged EGF ligands. Future work will allow the inter-fluorophore distance to be determined simultaneously with the angles, within confidence limits.