WO2005064319A1 - Identification of single molecules - Google Patents

Abstract

A computer-based method identifies single molecule light sources, such as fluorescent single molecules, in a series of pixellated images of an optical section of a specimen. The method comprising the steps of: (a) determining a pixel group size which corresponds to the diffraction-limited image area of a single molecule light source in the optical section; (b) over the series of images, tracking temporal changes in intensity of pixel groups of said size; and (c) identifying single molecule light sources in the images by selecting those pixel groups from step (b) which exhibit a change in tracked intensity corresponding to an expected intensity behaviour of a single molecule light source in the optical section.

Description

IDENTIFICATION OF SINGLE. MOLECULES

FIELD OF THE INVENTION

The present invention relates to a method and system for identifying single molecules in a series of pixellated images of an optical section of a specimen.

BACKGROUND OF THE INVENTION

Single molecule optical studies of fluorescent probes in vitro (Funatsu et al . , 1995) and in living cells (S^'a o et al . , 2000) have provided new insights on the events underlying cell signalling, trafficking and cytoskeleton-membrane arrangements (Seisenberger at al., 2001; lino et al . , 2001; Harms et al . , 2001b) . For example, protein properties and functions can be estimated directly at a single molecule level (Watanabe and Mitchison, 2002) .

Many such single molecule optical studies have been performed using Total Internal Reflection (TIR) Fluorescence Microscopy (TIRFM) (Axelrod, 1992) . When light, such as a laser beam, obliquely illuminates the interface of two media of different refractive index from the high to the low refractive index direction at an incident angle that is greater than the critical angle for total internal reflection, a shallow electromagnetic field called the "evanescent field" is produced in the low refractive index medium. In TIRFM single molecule studies, the high refractive index medium is typically a glass slide and the low refractive index medium is the sample of interest. The evanescent -field extends from the interface of the two media and exponentially decays in aqueous solution or cytoplasm with a decay length of about 100 nm, leaving most of the sample volume un-illuminated . Laser epi-illumination can also be used to detect fluorescent single molecules in vitro or at cell membranes (Sase at al . , 1995; Harms et al . , 2001b), but in this case the fluorophores must be positioned solely in the plane of focus and not be present in bulk solution.

In the last decade, in vivo single molecule studies have been boosted by the introduction of Green Fluorescent Protein (GFP) technology (Kendal and Badminton, 1998). A protein's cDNA sequence can be inserted into commercially available plasmids to produce a fusion protein construct with GFP. This can then be used for cell transfection (Harms et al . , 2001a). If the GFP- fusion protein is present in the cells at nanomolar concentrations then fluorescence of single GFP-protein molecules can be visualized as discrete spots of light; (Pierce et al . , 1997, Sako et al . , 2003)

One of the problems associated with imaging single GFP molecules in living cells is the high autofluorescence levels (Harms et al . , 2001a). Also, unless the GFP molecules are firmly attached to a cellular structure (membrane or cytoskeleton) they diffuse too rapidly to be visualised as discrete objects. When studying membrane-binding proteins, these problems are ameliorated by using TIRFM. However, unbound (freely diffusing) fluorophores still present high background signals.

Detection and tracking of individual fluorophores is a critical first step in analyzing fluorescent single molecule^' data sets and presents a significant challenge in terms of data handling and analysis. Manual selection of individual spots observed on sequences of video data takes significant time and is subject to observer bias.

SUMMARY OF THE INVENTION In general terms, the present invention provides an automated method of identifying single molecules with an aim of providing improved single molecule detection and thereby facilitating temporal and spatial analysis of single molecule trajectories.

The invention is suitable for detection of single molecules tagged with a variety of different types of fluorescent probe, e.g. quantum dot, single fluorophores, multiple fluorophores, fluorescent beads and so forth.- However, in some circumstances it may also be suitable for ^' detection of naturally-occurring fluorescent single molecules. . As used herein, the term

"fluorescent single molecule" encompasses single molecules artificially tagged with fluorophores, and also naturally- occurring fluorescent single molecules. More broadly, the invention can also be applied to single molecules attached to other types of tag, e.g. nanoparticle or nanogold tagged molecules in which the tags strongly reflect incident light. Thus, as used herein, the term "single molecule light source" encompasses single molecules attached to tags which are intended to be viewed optically, such as fluorescent single molecules and molecules attached to tags which reflect incident light.

In a first aspect, the present invention provides a computer- based method of identifying single molecule light sources in a series of pixellated images of an optical section of a specimen, the method comprising the steps of: (a) determining a pixel group size which corresponds to the diffraction-limited image area of a single molecule light source in the optical section; (b) over the series of images, tracking temporal changes in intensity of pixel groups of said size; and (c) identifying single molecule light sources in the images by selecting those pixel groups from step (b) which exhibit a change in tracked intensity corresponding to an expected ^' intensity behaviour of a single molecule light source in the optical section.

Preferably, the single molecule light sources are fluorescent single molecules.

The pixel group size, average intensity, temporal properties etc. of the single molecule light source in the specimen may be predetermined using one or more control specimens containing known dilutions of the single molecule light source to be identified.

The diffraction-limited image area of each single molecule light source is generally much greater than its physical size, i.e. typically the image will be determined by the limit of optical resolution (which is about -250 nm) .

The method may be performed on a stored series of images, or performed on-line when single molecule light source identification can be computed faster than the rate of data collection. Thus the method may comprise an initial step of obtaining the series of pixellated images of the optical section of the specimen.

By determining an appropriate pixel group size and then selecting pixel groups whose intensity behaviour matches what would be^' expected from a single molecule light^" source, it is possible to automate the identification of single -molecule light sources. Thus observer bias and error can be avoided, and consistent identification criteria can be applied to all data. This improves subsequent statistical analysis of the results.

In order to reduce systematic, experimental and photon noise, the tracked intensity of each pixel group can be based on a temporal running averaged intensity, e.g. using the computed mean or median intensity from 4 to 10 sequential image frames.

To compensate for low frequency changes in temporal and/or spatial background intensity level, the tracked intensity of each pixel group can be determined relative to a background intensity. For example, the -appropriate background intensity . for each group can be determined from the intensity of the pixel ring surrounding the group.

Preferably, the change in tracked intensity is a step change in intensity. Fluorescent single molecules should exhibit a stepwise decrease in intensity due to photobleaching or fluorophore removal from the optical section by diffusion. On the other hand, .sudden fluorophore appearance in the optical section will tend to produce a stepwise increase in intensity in the corresponding pixel group. Thus when such stepwise changes are seen in pixel groups of the appropriate size, they can be associated with underlying fluorescent single molecule behaviour.

The rate of photobleaching is directly proportional to illumination intensity, whereas fluorophore 'disappearance from the optical section should be independent of illumination . intensity. Thus by conducting experiments at different illumination intensities (e.g. by changing the duty-cycle ratio of the laser beam illumination or by attenuating the laser source) , it is possible to distinguish between the rates of fluorophore photobleaching and fluorophore diffusion out of the focal plane.

A related aspect of the invention provides a method of analysing a series of pixellated images of an optical section of a specimen. For example, having identified single molecule light sources in the images according to the method of the previous aspect, such a method may further comprise: determining the lateral position of the identified single molecules in the optical section; and/or tracking lateral movement of the identified single molecules in the optical section; and/or collecting statistical data on the identified. single molecules .

The methods of the invention discussed above may conveniently be " implemented in software, for execution on any appropriate digital computer.

Thus- further aspects of the invention respectively provide a computer system operatively configured to implement any of the methods of the previous aspects of the invention; computer programming product or products (such as ROM, RAM, floppy discs, hard drives, optical compact discs, magnetic tapes, and other computer-readable media) carrying computer code for implementing any of the methods of the previous aspects of the invention; and a computer program per se for implementing any of the methods of. the previous aspects of the invention.

For example, a computer-based system for identifying single molecule light sources in a series of pixellated images of an optical section of a specimen, may comprise a processor which is operatively configured for: (a) over the series of images, tracking temporal changes in intensity of pixel groups of a size corresponding to the diffraction-limited image area of a single molecule light source in the optical section; and (b) ^' identifying single molecule light sources in the images by selecting those pixel groups from step (a) which exhibit a change in tracked intensity corresponding to an expected intensity behaviour of a single molecule light source in the optical section.

Such a system may be used to identify single molecule light sources on-line. However, usually it is convenient to obtain and store the series of images for subsequent analysis, in which case, the system may further comprise an image storage device for storing the series of images and communicating the series of images to the processor. The device may be e.g. ROM or RAM or a device for reading media on which the images have been previously stored. Such media may include floppy^' discs, hard drives, optical compact discs, magnetic tapes etc.

A further aspect of the -invention provides a microscope system comprising: a microscope for imaging an optical section of a specimen; an image registering device, such as a camera, arranged for registering a series of pixellated images obtained by the microscope; and a computer system according to a previous aspect for identifying single molecule light sources in the series of images registered by the image registering device.

Preferably, the microscope is adapted to perform total internal reflection fluorescence microscopy.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention will be described by way of example with reference to the accompanying drawings.

Fig. 1 shows schematically^' a fluorescent single molecule imaging apparatus;

Fig. 2 shows (a) a schematic sequence of pixellated images, (b) a "drop" image calculated from the sequence of images, (c) the thresholded "drop" image, and (d) the "step" image calculated from^' the thresholded "drop" image and the .sequence of images, and (e) the XY positions of the fluorophores causing the spots of the "step" image; Fig. 3 shows (a) an example intensity track used to calculate a step index, and (b) another example intensity track used to calculate a pedestal index;

Fig. 4 shows (a) the "rise" image calculated from a further sequence of images, (b) the thresholded combined image representing spots present in both the "rise" image and a "drop" image (not shown) , and (c) the "pedestal" image calculated from the thresholded combined image and the sequence of images;

' Fig. 5 shows (a) a representative frame from an image sequence recording GFP on a glass surface at a density of 0.2-0.3 molecule μm^-2, (b) the "drop" image formed from the image seguence, (c) the "step" image formed from the image sequence (d) the calculated centroid of each identified fluorescent single molecule from the image sequence, (e) -representative intensity tracks of individual fluorophores, and (f) the intensity drop distribution of the single GFP molecules bound to the glass surface;

Fig. 6 shows (a) individual TIRFM images taken at 0, 6 and 12 seconds after the start of recording of the lamella of a living mouse myoblast under continuous laser illumination, (b) representative fluorescent intensity tracks exhibiting single step photobleaching, (c) representative fluorescent intensity tracks for landing and subsequently disappearing spots, (d) the intensity drop distribution for fluorescent single molecules bound to the cell membrane, and (e) the variation in the average fluorescence of the total area shown (570μm²) during the recording;

Fig. 7 shows (a) individual TIRFM images taken at 0, 8 and 16 minutes during a time-lapse recording of one cell that was illuminated for^' 350ms in every 5s interval, giving an illumination duty_, ratio of 0.07, (b) representative fluorescence intensity tracks, (c) a plot of average cell fluorescence against time, (d) a histogram showing the lifetime distribution of 775 individual spots, and (e) an image sequence of one fluorescent single molecule that landed on the cell membrane and stayed attached for over 140s, together with the fluorescence intensity track of this spot measured at ^'5s intervals; and

Fig. 8 shows plots of rate constant against illumination duty ratio for GFP-PH domain on plasma membrane and GFP-PH domain on anti-GFP antibody.

DETAILED DESCRIPTION

The following section considers the invention as applied to the ^■ identification of fluorescent single molecules. However, the skilled reader would appreciate that similar considerations can be applied -to the identification of other types of single molecule light source.

Theoretical Considerations

If the bulk concentration of a fluorophore is in the nanomolar (-nM) range then there will be approximately one molecule per cubic micron. If an optical sectioning technique is used then individual fluorophores should be visible as separate spots of light. For this to form a viable experimental system, the source of exciting light must be sufficiently intense and the optical detector system suitably sensitive to enable images to be -captured at video rate. Measurements with "best" precision, often require compromises between certain physical parameters e.g. between time-resolution and signal-to-noise ratio.

In the experiments described below, we observed single proteins moving inside a living cell. To achieve this, we expressed the proteins as fusion constructs with green fluorescent protein (GFP) . An advantage of this approach is that any protein (or protein domain) can be specifically tagged and its localisation Observed within a live cell with minimal intervention (Ludin et al . , 1998). However, some disadvantages should be considered: GFP is a fairly large fluorescent tag (27kDa mol . ' wt . ) (Prasher et al . , 1992) and may therefore interfere with normal functioning of the protein under study; expression of a fluorescently tagged protein usually results in overexpression of the protein in the cell, which may affect cell behaviour and/or function; GFP fluorescence is excited by blue light ^■ (around 488nm) which causes other substances in a living cell to fluoresce; finally, along with normal photobleaching processes, GFP also exhibits temporary "blinking" and "switching" behaviour's (Dickson et al., 1997) which complicate interpretation of kinetic data.

We have used a mutant isoform of GFP (eGFP; GFP-65T) (Heim et al . , 1994) that has superior excitation and emission properties compared to wild type GFP. This isoform is excited at 488nm and emits at 510nm. Its fluorescence lifetime is about 3.2 ns

(Volkmer et al . , 2000) and so emission saturation should occur at 3^χl0⁸ photons per second, ^' however, measured values are closer to 8.5^χl0⁶ s^-1, probably because of a slower, competing, de- excitation mechanism (Harms et al., 2001). Using 488nm light, excitation saturation occurs at an incident intensity of 29kW.cιrf ² (i.e. 29^χl0³/hv = 7.2^χl0²² photons . s^-1. cm^-2, and if quantum efficiency = 60% and absorption cross-section = 1.9 A² this gives the observed output of 8.5^χl0⁶ photons . s^_1. fluorophore^" ). At saturating excitation the average time to photobleaching is about 4ms (Harms et al., 2001) and shows mono-exponential behaviour, giving 30,000 photons output per fluorophore before photobleaching. At lower excitation intensities both average emission rate and time to photobleaching increase linearly with excitation power.

From a practical standpoint, excitation intensity (e.g. laser power) should be adjusted to give an emission rate and fluorophore lifetime that is appropriate to the type of information required. Photon counting statistics, noise level and mean time to photobleaching all interact in a somewhat competing manner. To observe slow processes, low powers should be used in order to increase the average time until photobleaching. Fast processes require high laser power to increase the photon emission rate so that fast data collection " is possible.

Next, we need to. consider how a fluorescent single molecule can be imaged in three-dimensional specimens like living cell's. Standard epi-fluorescence microscopy is usually unsuitable because fluorophores throughout the thickness of the specimen produce out-of-focus background noise due to poor z-axis discrimination (Paige et al., 2001). Instead, a form of optical sectioning is required; either confocal microscopy or total internal reflection fluorescence microscopy (TIRFM) (Axelrod, 2001) . For our studies, TIRFM'is the method of choice because it enables a very thin section (<100nm) to be viewed continuously as a wide field (i.e. non-scanned) image. This means that individual fluorophores that enter the evanescent field can be seen as separate, diffraction-limited, spots and that background fluorescence from out-of-focus fluorophores and cell autofluorescence (mainly from flavonoids) is minimised.

Experimental Procedures

Single Molecule Imaging Apparatus

A detailed description of the apparatus is also described by

Mashanov et al . , 2003. The apparatus is shown schematically in Fig. 1.

Microscope

We used an objective lens-coupled TIRFM based on used a Zeiss inverted microscope (Axiovert 100TV, Zeiss, Germany) . A custom built holder with two small (3mm diameter) round mirrors mounted at 45° to the microscope axis was positioned in the slot in the objective- turret, underneath the objective lens, normally used for insertion of the DIC Wollaston prism carrier. We used a Zeiss, Fluar 100_* 1.45NA objective lens and Zeiss, low fluorescence, 1.51 refractive index, immersion oil.

Laser

We used an argon ion, all lines, TEM 0,0, single mode laser (532-AP-A01, Melles Griot, CA, USA), which had an output of up to 5θmW output at 488nm, less than 2 mrad divergence (e.g. diffraction limited output) and reasonable (<100 μrad) beam pointing stability. Leakage of other laser lines (i.e. 515nm) was excluded using an externally mounted, 488nm laser line, band pass filter (e.g. XL06, Omega Optical, Brattleboro, VT, USA). The cooling fan was mounted away from the optical table to prevent mechanical noise causing fluctuations in the beam angle (and hence TIR angle) . The laser illumination levels were controlled via RS232 interface, custom-built, mechanical shutter system, that opened or closed within 10ms, and by addition or subtraction of neutral density filters.

To vary the illumination, we controlled the laser power by a time-lapse approach, in which the specimen was illuminated for 0.25, 0.5, 1, or 1.5 seconds in every 5-second interval. This ensured that the imaging _. conditions (e.g. angle of the input laser beam, hence evanescent field strength, camera gain and signal-to-noise ratio)- remained constant during data collection. Samples were illuminated with an attenuated beam (7% of normal levels) , during the first second of each illumination cycle and the angle of the reflected beam was used to control the focus.

Optical Table

All optical components, - including laser and microscope, were mounted on a vibration-isolated workstation "VH workstation", Newport Corp., Irvine, CA, USA). A rail-mounted optical component system was used to construct the optical circuit.

Optical Circuit

A Newtonian beam expander (f=10 and f=160) was constructed to produce a collimated, expanded laser beam (of Ca.l2mm diameter). A 50μm diameter pinhole accurately positioned at the intermediate focus of the beam expander produced spatial filtering by removing higher-order scattered light. A laser- line band-pass filter, inserted after the beam expander improved spectral purity. Finally, a field diaphragm placed immediately after the beam expander enabled the size of ' the final TIR spot (at the specimen plane) to be controlled. Translation of the final beam expander lens (f=160mm) in the (x,y) -plane, enabled the angle of incidence of the TIR spot to be changed without affecting spot position. Translation of this lens in the z- plane (in the direction of light propagation) focussed the final beam. This was adjusted so that the beam exiting the microscope objective lens was "best collimated".

After passing through the beam-expansion and filtering optics, the laser beam was reflected from a plane mirror mounted on a kinematic mount. This was 200mm from a third lens (f=200mm) which itself was a further 200mm from the back focal plane of the objective lens. Tilting of the mirror again allowed the TIR angle to be adjusted without affecting spot position. Translation of the lens in the (x,y) -plane enabled spot position to be adjusted without greatly affecting TIR angle.

To align the beam the bright-field lamp is switched on so that light entered the objective lens and is then reflected from the custom built TIR mirrors. Optical components are then inserted, one by one, so that the light coming from the objective follows an undeviated path. This procedure ensures that all the components are correctly aligned with the microscope axis. When the laser is then switched on it should retrace the same path and be approximately in the correct alignment. Fine adjustments are then made so that the laser beam is emitted from the front lens of the microscope objective. At this point the laser beam will be approximately collimated and will be emitted at an unspecified angle, presenting possible danger to the , experimenter. Hence, it is advisable to wear suitable goggles (Glendale Inc. Lakeland, FL, USA) at all times.

To optimise beam collimation, the exiting beam is projected onto the ceiling (if an inverted microscope is used) or a piece of card if an upright microscope is used. Then by z-translation of the second beam expansion lens, the beam can be brought to "best collimation". x-y-translation of the same lens or tilting of the mirror causes the exiting beam angle to change and hence TIR angle when the specimen is in place. This should be adjusted until the beam exits the lens at about 10° to the horizontal (80° incident angle). Now, when the specimen (e.g. glass slide + water medium) is mounted, the beam angle will reach the critical angle for the glass-water interface (61.5-62°). and should no longer exit the lens, but, instead will reflect back from the coverslip-water interface and be coupled out of the microscope on a roughly parallel path to the input beam by the second mirror positioned at the back focal plane. As a control specimen we suggest use of a dilute solution (0.05%W/V) of 500rtm diameter, yellow-green fluorescent, latex beads (Bang's Laboratories, Fishers, IN, USA) .

Camera

We used a digital CCD camera coupled to a two-stage, multichannel plate, image-intensifier (GemStar, Photonic Science, Millham, East Sussex, UK) . The entire system (camera + intensifier) was computer-controlled and data transfer digital at 10 or 12-bit resolution. The ability to change parameters such as exposure time, pixel binning, region of interest (ROI) , intensifier gain and CCD camera gain has the drawback that there are many different overall gain and signal-to-noise ratios. After systematically measuring the signal-to-noise ratios at different gain combinations on control specimens in our system, we optimised settings- for best signal-to-noise at gain appropriate for identification fluorescent single molecules at a. given frame rate.

• Computer Hardware and Software

A proprietary, digital frame-grabber { Snapper DataCell, UK) was used to control the camera and capture data at acquisition rates of 20 megapixel . s^"1. We found modern IBM-PC type computers (Pentium III, 1 GHz or better) to be fast enough for simultaneous real-time image display, hard drive recording (at a maximum speed of 15 Mpixel.s^-1) and offline image analysis.

We wrote our own software to record and analyse data at. sufficient speed or capacity (30 Mbytes, s^""1 and up to 500 Mbytes per record) . The software controlled all of the microscope, camera, laser and oveable optical components and stored^' all settings together with other experimental parameters with each data sequence.

GFP Molecules In Vitro

Glass coverslips were glued to microscope slides by UV-curing adhesive (RS - components, UK) to construct rigid flowcells, 0.15 mm thick glass spacers forming a central channel. Buffer solution (pH 7.4) with anti GFP antibodies (Abeam Ltd, UK) 5 μg ml^-1 was left in flowcell for 5 min and after a few washings was replaced with buffer containing GFP 20 ng ml^-1 for 5 min.

Unbound GFP was washed with excess of buffer solution and the flowcells were placed on the microscope for observations.^'

Autofluorescent Protein Construction and Transfection

Many signaling and motor proteins have an amino acid sequence called the Pleckstrin Homology (PH) domain (Lemmon et al.,

1995) , known for its membrane-binding property. PH domains can selectively bind phosphoinositols incorporated in cell membrane (Dowler et al . , 2000). We chose PH domains of myosin X for this study (Yonezawa et al . , 2000; Berg and Cheney, 2002). Myosin X has three PH domains but its membrane binding properties are not known . •

Cloning

Full-length myosin X cDNA was amplified from mouse myoblasts by long rt-PCR using primers based on the mouse sequence (AJ249706) (Yonezawa et al . , 2000) and sequenced in entirety to check for sequence fidelity. Primers flanking the PH domains, or the FERM domain, with BamHI site on the forward and a Sail site on the reverse primer, were used with this cDNA in an rt-PCR reaction.

' 5 The resulting DNA fragments (nt3615-4500 for PH123, nt 4740- nt5175 for M-yTH4, and nt5385-6270 for FERM) were directly cloned into TOPO (Invitrogen, UK) , and sequenced to confirm sequence fidelity. The BamHI-Sall fragments were excised from TOPO and cloned into PeGFP-Nl and peGFP-Cl (BD Biosciences Clontech,

10 USA) , to create PH123-GFP, MyTH4 or FERM-GFP, with the GFP fused either to the N- or to the C-terminus . These constructs were transiently expressed in COS-7 cells, and the cell lysates were immunqblotted, using an anti-GFP antibody (AbCam Ltd, .UK) , to confirm that the expressed proteins were of the predicted size.

15 In the experiments described here, we used FERM-GFP, MyTH4 and PH123-GFP in which GFP was fused to the N-terminus, but the C- terminal fusions gave similar results.

Cell Cul ture and Transfection

COS-7 cells were cultured in DMEM (Gibco) supplemented with 10% 20. fetal bovine serum (Gibco) . Cells were plated on glass coverslips coated with 0.01% gelatin (Sigma) 24 hours prior to transfection using DEAE dextran. Cells were fixed using 4% paraformaldehyde, mounted in Pro-long antifade, and confocal sections were taken using a Zeiss Pascal confocal microscope.

25 Mouse myoblasts were isolated from the "immortal mouse" (h2k^b- tsA58) and cultured as described by Peckham et al . , 2001. They were transfected with PH123-GFP or peGFP-Nl (Clontech) using FuGene-6 (Roche Diagnostics, UK) following the recommended protocol. After 24 hours, the cells were plated onto glass

30 coverslips, and allowed to settle. 8 hours later, the medium was replaced with Hank's balanced salt solution containing 20mM HEPES (pH 7.4), without serum, and the coverslip was assembled into a small chamber for viewing the cells on an inverted microscope. Approximately 10% of the cells had a suitable level of expression for TIRFM measurements following this transfection procedure, such that single molecules could be visualised as single spots. We estimate that the final concentration of GFP- fusion protein in the cell was in the nanomolar range (i.e. approximately 1 molecule per μm³, 5000 molecules/cell) . This was confirmed by independent measurements using confocal microscopy, where we compared the fluorescence of a range of concentrations of GFP in solution, to that of live transfected cells and found the concentration _. to be lOnM. Under equivalent conditions in neutrophils, local concentration of Pi(3,4,5)P₃ was 50nM (local concentration of 5μM, Insall, R.H., and Weiner O.D'. , 2001).

Fluorescent Single Molecule Detection Algorithm (SFDA)

The SFDA exploits three properties of fluorescent single molecules: diffraction-limited size of fluorophores on high- resolution images; steady level of fluorescence under steady illumination; and instant photobleaching to background fluorescence level.

The SFDA analysed each sequence of pixellated images (shown schematically in Fig. 2a) recorded by the CCD camera in up to three passes. In the first pass a 4-10 frames temporal running average of intensity in the diffraction-limited area (which in our set-up we typically determined to be a 5χ5 pixel group of

0.16 μm²) centred on each pixel on the images was tracked to find the biggest drop in intensity in this area during the record. The background intensity obtained from the intensity of pixels in the ring surrounding each area was subtracted from the intensity drop to eliminate or reduce the contribution of objects with sizes bigger than the diffraction-limited area.

Each pixel was then associated with^' the intensity drop of the pixel group of which it was the centre to produce the "drop"image of Fig. 2b. This image records spots which exhibited a significant sudden reduction of intensity at some time over the record. The use of the running average intensity greatly reduced noise fluctuations, which could mask photobleaching events .

We applied an intensity threshold to the "drop" image to limit our analysis to selected pixels on the image, where changes in intensity of the corresponding diffraction-limited area, were significantly higher than the surrounding background noise. The thresholded image is shown in Fig. 2c. In the case of single GFP molecules in vi tro, most of the selected spots had a diffraction-limited size., a constant level of fluorescence, a narrow intensity distribution (Fig. 5f) and exhibited a one step fluorescence drop to background level, strongly suggesting that we had detected single GFP molecules. In such good- conditions we could then go on to collect statistical information about the detected spots. However, living cell samples had higher levels of temporal and spatial noise, so in this case further filtering (a second pass) was employed to reduce noise and eliminate some blinking or moving spots .

In the second pass, temporal intensity tracks for each selected pixel were collected and tested for "step" behaviour. We assumed that fluorescent single molecules attached to the surface would have a steady level of fluorescence, and, after instantaneous photobleaching or fluorophore removal from the . optical section, would have a steady background intensity noise. The intensity track for each selected pixel was formed by calculating, at each image frame, the average intensity of the diffraction-limited area centred on that pixel the intensity ^' tracks being formed by calculating, at each image frame (i.e. time instant) , the average intensity of the diffraction-limited area centred on that pixel and subtracting from this the average^' intensity in the pixel ring surrounding the diffraction-limited area. Fig. 3a shows an example of such an intensity track. The aim of the test was to create a "step" image of spots in which pixels with a high "step" index, I_s, would be brightest. ϊ_s was determined from the expression: ΔΓ Is = ° ^beforeDrop "afterDrop where ΔI is the time-averaged intensity drop from the respective track when a spot disappears, SD_befo_reDrop is the standard deviation of intensity from the track before spot disappearance, and SDafte_rDrop is the standard deviation of intensity from the track after spot disappearance. The time-averaging used to determine ΔI was based on fixed time intervals before and after spot disappearance, as shown on Fig. 3a. For computational convenience, these time intervals could be the same as that previously used for the intensity running average of the first pass.. SDbeoreDrop and SDafterDro were calculated for the whole track record respectively before and after the intensity drop in order to eliminate landing, blinking, moving, and multi-fluorophore spots, except that if the background fluorescence changed during long records it was sometimes beneficial to limit the periods for the SD_beforeDro and SD_afterDop calculations . Thick straight lines on Fig. 3a represent idealised single molecule fluorescence behaviour (i.e. the fluorophore emits a constant number of photons per time unit) . The "step" index helped to distinguish fluorescent single molecule events from various noise events occurring on the record. Each selected pixel was 'associated with its I_s value to produce the "step" image of Fig. 2d. Pixels with high I_s were considered to be caused by single non-blinking fluorophores which had been photobleached or removed from the -optical section during a given record. A further intensity threshold could be applied at this stage.

In the third pass, we calculated approximate fluorophore XY positions in the optical section. We were also able to further discriminate whether individual spots were caused by fluorescent single molecules. Because each diffraction-limited area occupied a significant number of pixels on the CCD camera (typically 25-30 pixels) , we used spatial intensity extremes to calculate fluorophore XY-coordinates (in pixel units) .

The photons emitted by a fluorescent single molecule strike the surface of the detector according to a two dimensional Gaussian distribution, whereby the pixel at the centre of the distribution corresponds to the fluorophore XY position. This pixel records a higher intensity than any of the pixels around it. Furthermore, the average intensity of all the pixels in the first ring around the centre pixel is higher than the average intensity of the pixels in the second ring around the centre pixel. We used these properties of the distribution to find the centre pixel, if any, of each spot formed on the "drop" or "step" image.

For example, for a diffraction-limited area corresponding to a 5x5 pixel group, one of the pixels of each spot was determined to be at the XY position of a fluorophore causing the spot if the following criteria were met: the intensity of that pixel was 10-20% higher than the average intensity of the first ring of surrounding pixels, and the average intensity of the pixels of the second surrounding ring was 10-20% lower than the average intensity of the first ring. For diffraction-limited areas corresponding to different pixel group sizes the Gaussian distribution will broaden or narrow and the criteria should be adjusted accordingly. Fig. 2e shows the XY positions of the fluorophores causing the spots of Fig. 2d.

A similar multi-pass automated procedure was used for the identification of fluorescent single molecules appearing' in the optical section (e.g. fluorophores dynamically binding to an imaged surface such as a cell membrane) . In this case manual detection is especially difficult because fluorophores are not present on the surface at the beginning of a record, and the record needs to be scanned many times for spots coming and staying on the surface for short time. Fluorescent single molecules attached to free floating proteins move fast, even in viscous cytoplasm media. They can arrive in and leave evanescent field in a fraction of millisecond, which means that fluorescent spots appear suddenly in optical section images due to attachment to substrata and suddenly disappear due to detachment or photobleaching.

To detect such "landing" spots we used in a first pass a temporal running average, of intensity in the diffraction-limited area around each pixel to find the biggest sudden increase in intensity during the record. The resulting "rise" image (Fig. 4a) represented pixel groups with a significant sudden increase of intensity over the record. The "rise" value associated with each pixel could be added to the previously determined "drop" value for the pixel, and an appropriate threshold then applied, to limit the analysis to selected pixels forming spots which suddenly appeared and later disappeared during the^' record. Alternatively, the "rise" value of each pixel could be subtracted from the corresponding "drop" value to eliminate from the analysis spots which suddenly appeared during the record.

An example thresholded combined image is shown in Fig. 4b. The image shows spots present in both the "rise" image of Fig. 4a and a "drop" image (not shown) and represents fluorophores which "landed" on the surface during a sequence of images and photobleached or disappeared from the surface before the end of the sequence.

In a second (optional) pass, an index similar to I_s was used to better identify those selected pixels whose intensity behaviour matched that expected of "landing" and subsequently disappearing fluorescent single molecules. Temporal intensity tracks were again collected for each selected pixel (the intensity tracks being formed by calculating, at each image frame, the average intensity of the diff action-limited area centred on that pixel and subtracting from this the average intensity in the pixel ring surrounding the diffraction-limited area) . We then used a "pedestal" index (I_p) to create a "pedestal" image where the -intensity value of each selected pixel was given by:

AΓ^* I„ = S D_high S D low where ΔI is the time-averaged intensity drop from the respective track at the time of spot disappearance, SD_hih is the standard deviation of intensity from the track during spot presence on the surface, and SDι_ow is the standard deviation of intensity from the track after the spot disappeared.

Fig. 3b shows an example of a pixel intensity track caused by a "landing" and subsequently disappearing fluorescent single molecule. Thin straight lines on Fig. 3b represent idealised single molecule fluorescence behaviour. The time-averaging used to determine ΔI was based on fixed time intervals before and after spot disappearance, as shown on Fig. 3b (except that the "high interval" was shortened if the spot disappeared soon after appearance) . Rather than calculating SDι_0W from the intensity track before spot appearance, a fixed time interval after spot disappearance was used to calculate the background area fluorescence for SDι_ow- This method- gives a better match with fluorescent single molecule intensity estimates in conditions of decreasing background fluorescence in living cells. Pixels with high I_p were considered to be caused by fluorescent single molecules appearing and then photobleaching or being removed from the opt-ical section during a given record. Each selected pixel was associated with its I_p value to^' produce the "pedestal" image of Fig. 4c. A further intensity threshold could be applied at this stage. The "pedestal" index, like the "step" index, helped to distinguish fluorescent single molecule events from various noise events occurring on the record.

In a third pass, using the method of spatial intensity extremes described earlier, we calculated fluorophore XY positions in the optical section.

After fluorescent single molecules identification, ^'we collected information about fluorophore intensities, lifetimes, ^• landing times and fluorophore coordinates. This information was used to calculate fluorescent single molecule intensity distributions, lifetime distributions, landing rates, and inter-fluorophore ^■distances.

Tracking Lateral Positions of Fluorescent Single Molecules

Because, following the third pass of the SFDA, we knew the approximate position (to within about 200nm) of each fluorophore_. on the "drop", "step", "rise" or "pedestal" image, we could use a conventional fluorescent single molecule tracking algorithms (see e.g. Cheezum et al . , 2001) to calculate the exact position of each fluorophore during its presence on the surface, and subsequently automatically calculate its lateral diffusion coefficient.

Direct fit of a Gaussian curve to the fluorophore intensity profile was found to be a suitable algorithm for tracking fluorescent single molecule lateral position. We used images of GFP molecules on a glass surface to calculate the fluorescent single molecule point spread function (PSF) for our optical system. The PSF had a "Gaussian" shape and the same width at half-maximum for fluorescent single molecules on glass as on cell membranes. ^' We used the least-squares method to find the best fit of normalized PSF to fluorophore images, the x and y coordinates having sub-pixel resolution. The mean squared displacement (MSD) and coefficient of lateral diffusion,

(4ti_mage) , was calculated for each fluorophore during itspresence on the record.

Results

Characterization of Single GFP Molecules on Glass Substrates

We recorded GFP on glass surfaces at a density of 0.2-0.3 molecule μm^"2. Fig. 5a shows a representative frame from the record. Single GFP molecules were clearly seen as isolated spots of a diffraction-limited size in the optical section. We applied the SFDA to detect spots which disappeared from the surface during the image sequence to determine if they were real single GFP molecules.

Fig. 5b shows the "drop" image after the first pass of the SFDA_/ Fig. 5c shows the "step" image after the second pass of the SFDA, and Fig. 5d shows the calculated centroid of each identified fluorescent single molecule after the third pass of the SFDA. Fig. 5e shows representative intensity tracks of individual fluorophores (i.e. tracks of the average intensities of diffraction-limited areas minus the average intensities of the respective surrounding ring of pixels) together with their calculated idealized behaviors (solid fitted lines) . Fluorescence data were collected at 10 frames/s. Fig. 5f shows the intensity drop distribution of single GFP molecules bound to the glass surface (n=724) .

The representative intensity tracks in Fig. 5e exhibit a steady level of GFP fluorescence before irreversible photo-destruction. The noise level was dominated by photon counting. We collected about 30 photons per image frame so the expected standard deviation in intensity was about 30^~0-5 (»25%) of the mean intensity. The distribution of the intensity drops (Fig. 5f) measured for many different fluorophores was broader than expected from the noise contents of individual records. The large variation in individual intensities probably arises from the random orientation of the GFP on the surface (since polarized light was used to excite fluorescence) and variations in ^■ local environment.

Detected spots had the narrow intensity distribution, shown in Fig.5f, predicted for a sudden intensity drop at the moment of spot disappearance (photobleaching) . The spot lifetime distribution was mono-exponential, and, as expected, the average lifetime of fluorophores corresponded to the illumination intensity. The average spot intensity was 13.7 and 36.6 counts in a 0.16 μm-² area at respective illumination intensities of 3 and 10- μW.μπf². The respective calculated half lives, τ, were 6.9 and 2.2 seconds for the 3 and 10 μW. μm^~2 illumination intensities. Thus, there is strong evidence that the objects identified by the SFDA were GFP molecules on the glass surfaces, because they (i) were of diffraction-limited size, which means that the actual size of the emitting object is much less than the limit of optical resolution (ii) had a steady state fluorescence intensity corresponding to the known fluorescent single molecule intensity, (iii) exhibited single step photobleaching, and (iv) had a half life before photobleaching .which was linearly dependent on illumination intensity.

Some spots showed two separate photobleaching steps. This is because we used IgG antibodies, some of which could bind two GFP molecules. In this case the SFDA identified the bigger intensity drop because it gave a bigger "step" index, I_s. However, the SFDA could be readily modified to detect double photobleaching objects. A small fraction of GFP molecules exhibited "flickering" or "blinking" behavior as reported previously (Pierce et al . , 1997); the SFDA did not detect such spots because the high temporal noise lowered I_s below threshold.

Single Protein Binding to Internal Cell Membrane

Fig. 6a shows individual TIRFM images taken at 0, 6 and 12 seconds after the start of recording of the lamella of a living mouse myoblast under ^'continuous laser illumination. We found that at a low enough PH123-GFP concentration (-10 nM or 1 molecule μm^"3) , single protein molecules could be seen on the internal side of the basal myoblast membrane as separate spots. The individual spots of light had similar statistical properties to those from the experiment made using purified GFP bound to a glass surface via antibody. However the cell background was much noisier compared ^"to the in vitro studies, making manual detection of fluorescent single molecules a difficult task. To overcome this problem we used the SFDA to find fluorescent objects. exhibiting single molecule behavior. Some representative fluorescent intensity tracks (which plot the average intensity of respective 0.16μm², 5^χ5 pixel areas at each^' 100ms image frame) from detected spots exhibiting single step photobleaching are shown in Fig. 6b. These spots were caused by PH123-GFP molecules which had bound to the cell membrane before recording started. ■ The average lifetime of the detected spots was 2.5s. High cell autofluorescence caused a steady decrease in background level. However, the intensity drop distribution (shown in Fig. 6d) for fluorescent single molecules bound to the cell membrane and identified by the SFDA had a narrow distribution and neither its average value nor its distribution decreased with time, which is as it should be for fluorescent single molecules.

Native cell autofluore^'scent proteins, like flavinoids, caused scattered cell autofluorescence, but we also found a significant fraction of unbound PH123-GFP molecules floating close to the cell membrane. Thus a number of fluorescent objects suddenly appeared on the cell membrane image after the beginning of recording. We used the SFDA "pedestal" index to detect such PH123-GFP molecules landing at the basal membrane during the image sequence. Fig. 6c shows representative fluorescent intensity tracks from landing and subsequently disappearing spots. These events prove that PH123-domains of myosin-X have a limited residency time on the cell membrane and the cells have a significant pool of free-floating molecules.

Fig. 6e shows the decline in average fluorescence of the total area shown (570μm²) during the recording, which has a rate constant of 0.4s^"1 (τ=2.5s, R²=0.98). Under continuous illumination most of the fluorophores on the membrane were photobleached, but some fluorescent molecules were able to dissociate from the membrane and leave the evanescent field effectively instantly (taking account of the relatively slow imaging rate of 10-20 frames s^"1 and the fast diffusion coefficient of free floating protein molecules of 1-10 μm² s^"1, see Dayel at al . , 1999). After a few minutes in darkness a photobleached cell could exhibit almost the same density of fluorescent spots on the membrane as at the start of the previous imaging sequence, demonstrating fluorescence recovery after photobleaching.

In control experiments we used mouse myoblasts transfected with GFP alone, and other fragments of myosin-X, namely FERM-GFP and MyTH4-GFP. We could not find any binding of fluorescent objects to the cell membrane for more than one frame, which suggests that PH123-GFP binds specifically to the cell membrane.

We also tested COS-7 cells transfected with PH123-GFP. We found that the basal membrane of COS cells formed ruffles and was much less firmly attached to the substrata (glass surface) compared to myoblasts. We could not find significant numbers of isolated fluorescent spots on these membranes; most likely because the gaps between the substrata and the basal membrane prevented the evanescent field from^' illuminating the membrane.

Time-Lapse Imaging of Fluorescent Single Molecules on Cell Membranes

In order to distinguish between the processes of photobleaching and dissociation, and therefore estimate the rate of detachment of the PH domains from the plasma membrane, the lifetime distribution of individual fluorescent spots was .measured at different average illumination intensities. To do this we used time-lapse imaging, where the cell surface was illuminated in 5 second illumination-darkness cycles. This enabled us to analyse data from up to a thousand individual fluorophores in each cell, allowing us to collect sufficiently large data sets to ensure statistically valid results. However, by handling the data obtained from each individual molecule separately, we retained the power of the single molecule approach.

We changed the illumination period (duty ratio) from 250 ms to 1.5 s, keeping all other illumination conditions the same. Fig. 7a shows individual TIRFM images taken at 0,- 8 and 16 minutes during a time-lapse recording of one cell that was illuminated for 350ms in every 5s interval, giving an illumination duty ratio of 0.07. The majority of the fluorescent spots appeared on the membrane after the beginning of the record. The overall fluorescence intensity of the cell remained high for' many hundreds of seconds and, as shown in Fig. 7c which is a plot of average cell fluorescence against time, decreased by only 25% over the 1000s recording period. Fig. 7d is a histogram showing the lifetime distribution of 775 individual spots. The distribution was fitted by a single exponential with a rate constant of 0.07s^"1 (τ=14s, R²=0.98). The average laser power density was lOμW.μm^"2

We found that most fluorescent spots which appeared on the cell membrane at the beginning of the imaging sequence, as shown on Fig. 7a, stayed visible on the membrane for relatively long times. Under such conditions we could track fluorescent single- molecules for hundreds of seconds. Some fluorescence intensity tracks are shown in Fig. 7b. The average lifetime of the spots was 14s. Fig. 7e shows an image sequence of one fluorescent single molecule that landed on the cell membrane and stayed i attached for over 140s, together with the fluorescence intensity track of this spot measured at 5s intervals. The apparent rate constant for fluorophore disappearance (k'₀ff, the inverse of their average lifetime) should be the sum of two rate processes: photobleaching and detachment, k_P and k - The photobleaching rate, k_pb, should be -proportional to illumination power, ξv and assuming that the processes are irreversible, k'_off = ξ • k_Pb + k_d. At high laser intensity the dominant process will be ξ . k_Pb while at low intensity (ξ ~ 0) it will be k_d. Such an analysis for PH123-GFP in living cells showed that at zero illumination, k'_off = k_d = 0.05s^"1 (see Fig. 8, which shows plots of rate constant against illumination duty ratio for GFP-PH domain on plasma membrane and GFP-PH domain on anti-GFP antibody) . In contrast, GFP molecules attached via antibodies to glass had an extremely slow detachment- rate, k_d, and therefore k'_off extrapolated to near zero at zero laser power. By analysing the "landing rate" (i.e. the distribution of dark intervals before a_. fluorescent spot appears) , the apparent binding rate, k'_on of PH123-GFP at the plasma membrane was found to be 2 x 10^~3 s^"1. Since the apparent binding rate, k'_on depends upon the concentration of free PH123-GFP, it was necessary to measure the expression level (i.e. concentration) of PH123-GFP in our cells. This was done by comparing the cell fluorescence with calibrated dilutions of free GFP in solution using confocal microscopy. Consistent with this, TIRFM images had on average less than one fluorescent spot per μm², and the distance between individual fluorophores binding to the plasma membrane was greater than their diffraction limited spot size. The second order rate constant for binding, k_b = 0.2μM^_1. s^"1, was close^' to the diffusion limited value.

Lateral Diffusion of PH Domains Attached to Cell Membranes

Real-time and time-lapse recording showed that PH123-GFP molecules have a very limited mobility on cell membranes. The fluorescence intensities collected from the same diffraction- limited areas during spot lifetimes had limited fluctuations, suggesting that spots did not move during recording.

The camera registered only a small fraction of the photons emitted by each fluorophore. Under our levels of illumination, approximately 200-300 photons s^"1 or 20-30 photons per fluorophore per 100 ms frame were registered compared with an expected fluorophore emission rate of 10⁴-10⁵ photons per second. This resulted in a limited spatial resolution, .and so we used temporal 5-point median filtering to reduce noise due to photon counting statistics. Closely spaced PH123-GFP spots were not used for the analysis because neighbouring spots affected tracking.

We found that the apparent Dι_at was equal to 0.027 μm² s^_1 (n=265) for PH123-GFP on the basal cell membrane of living myoblasts and was equal to 0.007 μm² s^"1 (n=835) for GFP molecules attached to a glass surface via antibodies. However, the high level of noise in living^' myoblasts suggests that the real Oχ_at was smaller than 0.02 μm² s^"1. Thus calculation of the true lateral diffusion coefficient for PH 123-GFP may be beyond the capacity ■ of our present system because of the low mobility. of PH 123-GFP.

Discussion

We have shown me that an automatic algorithm can be applied for the detection and tracking of fluorescent single molecules in vitro and- in living cells. It was possible not only greatly to reduce the time of analysis but also automatically to detect and count single molecules in a noisy environment, where true events • are often were masked.

We found that single GFP molecules attached to glass substrata via antibodies are a reliable reference for in vivo single molecule studies, especially for testing hardware and detecting or tracking algorithms .

A fluorescent single molecule conjugated^" to a protein molecule can emit up to a million photons during its lifetime, but only few percent of these will^' be detected by the imaging system. In spite of such low levels of photon detection, fluorescent single molecules have many advantages in live cells research, being discrete objects with known properties. Using the SFDA it was^' possible to identify automatically fluorescent single molecules in noisy cell environments. We found that the point spread function of GFP molecules on the cell membrane was the same as in vitro, having a 400 nm width at half-maximum, which means that fluorophores in both cases were much smaller than the diffraction-limited size. Remarkably uniform and firm attachment of mouse myoblasts to substrata resulted in a very narrow distribution single molecule fluorescence intensity level (Fig. 6d) . Sudden spot disappearance confirmed that most detected objects were fluorescent single molecules. When we changed the illumination duty ratio, the average spot lifetime exhibited a linear dependence on illumination intensity.

We were able to measure binding and dissociation properties of single protein molecules on the cell membrane. We found the residency time of PH123-GFP molecules on the cell membrane was around 20 seconds. Binding of the myosin X PH domains at the plasmamembrane on this timescale might be important for its suspected role in controlling rearrangement of the cytoskeleton . during cell movement (~40s, Haugh, et al., 2000).

Random distribution of PH123 along basal myoblast membrane confirmed that 345 phosphoinositols have no specific location on the cell membrane. Also the negligible lateral diffusion shows that they are virtually immobilized in the cell membrane. These results are similar to the observation of single transmembrane protein E-cadherin molecules in fibroblasts (lino et al, 2001) . It is unlikely that 345 phosphoinositols freely diffuse inside small membrane compartments (Fujiwara et al 2002) because this would broaden the PSF of fluorophores attached to such floating molecules.

When the cell was retracting from the^' area of illumination, the fluorescent spots slid with the same speed in the same direction, but did not change their intensity. . It was found that the intensities of fluorescent single molecules had narrow intensity distributions and most of them were in a narrow focal plane. This confirmed that PH123-GFP molecules are attached to the cell membrane, but not to the actin cytoskeleton where they would have produced a broad distribution of intensities with many being out of- focus (the actin meshwork having a significant thickness of 200 nm in such cells) . We conclude that phosphoinositols on the basal cell membrane .of motile cells binding PH123 were 'attached to the rigid cytoskeleton, preventing their lateral movement in the cell membrane.

While the invention has been described in conjunction with the exemplary embodiment described above, many equivalent modifications and variations will- be apparent to those skilled in the art when given this disclosure. Accordingly, the exemplary embodiment of the invention set forth above arise considered to be illustrative and not limiting. Various changes to the described embodiment may be made without departing from the spirit and scope of the invention. REFERENCES

All the references cited in the above text and listed below are incorporated by reference.

Axelrod, D. 1992. Total internal reflection fluorescence. In Topics' in Fluorescent Spectroscopy. (Lakowicz J. R., Ed.). Plenum Press, New York . 3:289-342.

Axelrod, D. (2001) Traffic, 2:764-74.

Berg, J.S., and Cheney, R.E. 2002. Myosin-X is an unconventional myosin that undergoes itrafilopodial motility. Na ture Cell Biol . 4:246-250.

Cheezum, M.K., Walker, W.F., and Guilford W. H. 2001. Quantitative comparison of algorithms for tracking single fluorescent particles. Biophys . J. 81: 2378-2388.

Dowler, S. , Currie, R.A., Campbell, D.G., Deak, M., Kular, G., Downes, C.P. and Alessi, D.R..2000. Identification of pleckstrin-homology-domain-containing proteins with novel phosphoinositide-binding specificities. Biochem J. 351: 19-31.

Dayel, M.J., Horn, E.F.Y., and Verkerman, A.S. 1999. Diffusion of green fluorescent protein in the aqueous-phase lumen of endoplasmic reticulum. Biophys . J. 76:2843-2851.

Dickson, R. M., Cubitt, A. B., Tsien, R. Y., and Moerner, W. E. 1997. Na ture, 388:355-58.

Fujiwara, T., Ritchie, K.> Murakoshi, H., Jacobson, K., and Kusumi, A. 2002. Phospholipids undergo hop diffusion in compartmentalized cell membrane. J. Cell Biol . 157:1071-1081.

Funatsu, T., Harada, Y., Tokunaga, M., Saito, K., and Yanagida, T. 1995. Imaging of single fluorescent molecules and individual ATP turnovers by single myosin molecules in aqueous solution. Na ture . 374:555-559.

Harms, G.S., Codet, L., Lom erse, P.H.M., Blab, G.A., and Schmidt, T. 2001. Autofluorescent proteins in single-molecule research: Applications to live cell imaging microscopy. Biophys . J. 80:2396-2408.

Harms, G.S., Codet, L., Lommerse, P.H.M., Blab, G.A., Karh, H., Gamsjager, R., Spaink, H.P., Soldatov, N.M.,- Romanin, Ch . , and Schmidt, T. 2001. Single-molecule imaging of L-type Ca2+ channels in live cells. Biophys . J. 81:2639-2646.

Harms, G. S., Cognet, L., Lommerse, P. H. M., Blab, G. A., Kahr, H., Gamsjager, R. , Spaink, H. P., Soldatov, N. M., Romanin, C, and Schmidt, T. 2001. Biophysical Journal , 81:2639-46.

Haugh, J.M., Codazzi, F., Teruel, M., and Meyer, T. 2000. Spatial sensing in fibroblasts mediated by 3'phosphoinositides. J. Cell Biol . 151:1269-1279.

Heim, R., Prasher, D. C, arid Tsien, R. Y. 1994. Proc Natl Acad Sex USA, 91:12501-4.

lino, R., Koyama, I., and Kusumi, A. 2001. Single molecule imaging of -green fluorescent proteins in living cells: E- cadherin forms oligomers on the free cell surface. Biophys . J. 80: 2667-2677.

Insall, R.H. and Weiner, O.D. 2001. PIP3, PIP2, and cell movement - similar messages, different meanings? Dev Cell , 1: 743-747. .Kendal, J.M., and Badminton, M.N. 1998. Aequorea ictoria bioluminescence moves into exiting- new era. Trends in Biotech . 16:216-224.

Lemmon, M.A., Fergusson, K.M., O'Brien, R., Sigler, P.B., and Schlessinger, J. 1995. Specific and high-affinity binding of inositol phosphates to an isolated pleckstrin homology domain. PNAS. 92:10472-10476.

Ludin, B., and Matus, A. 1998. Trends in Cell Biology 8, 72-77.

Mashanov, G.I., Tacon, D., Knight, A.E., Peckham, M., and Molloy, J.E. 2003. Visualizing single molecules inside living cells using total internal reflection fluorescence microscopy. Methods, Academ . Press . , 29:142-152.

Paige, M. F. , Bjerneld, E. J., and Mo^'erner, W. E. 2001. Single Molecules, 2:191-201. Peckham, M., Miller, G., Wells, C, Zicha, D., and Dunn, G. A. ^'2001. Specific changes to the mechanism of cell locomotion induced by overexpression of beta-action. J. Cell Sci . , 114:1367-1377.

Prasher, D. C, Eckenrode, V.K., Ward, W.W., Prendergast, F.G. and Cormier, M.J. 1992. Gene, 111:229-33.

Pierce, D. W., Hom-Booher, N., and Vale, R.D. 1997. Imaging individual green fluorescent proteins. Nature. 388:338.

Sako, Y., Minoguchi, Sh., and Yanagida, T. 2000. Single-molecule imaging of EGFR signaling on the surface of living cells. Na ture Cell Biol . 2:168-172.

Sako, Y., and Yanagida, T. 2003. Single-molecule visualisation in cell biology. Imaging in Cell Bioligy. September : SS1-SS5. Sase, I., Miyata H., Corrie, J.E.T., Craik, J.S., and Kinosita, K. 1995. Real time imaging of single fluorophores on moving actin with an epifluorescence microscope. Biophys . J. 69:323- 328.

Seisenberger, G., Ried, M.U., Endreβ, Th . , Buning, H., Hallek,

M., and Brauchle, Ch . 2001. Real-time single-molecule imaging of the infection pathway of an adeno-associated virus. Science . 294:1929-1932.

Volkmer, A., Subramaniam, V., Birch, D. J., and Jovin, T. M. 2000. Biophys J, 78:1589-98.

Watanabe, N., and Mitchison, T. J., 2002. Single-molecule speckle analysis of actin filament turnover in lamellipodia . Science . 295:1083-1086

Yonezawa, S. et al . , 2000.- Mouse myosin X: molecular architecture and tissue expression as revealed by northern blot and in situ hybridization analyses. Biochem Biophys Res Commun . 271: 526-33.

Claims

1. A computer-based method of identifying single molecule light sources in a series of pixellated images of an optical section of a specimen, the method comprising the steps of: (a) determining a pixel group size which corresponds to the diffraction-limited image area- of a single molecule light source in the optical section; (b) over the series of images, tracking temporal changes in intensity of pixel groups of said size; and (c) identifying single molecule light sources in the images by selecting those pixel groups from step (b) which exhibit a change in tracked intensity corresponding to an expected intensity behaviour of a single molecule ^■ light source in the optical section.

2. A method according to claim 1, wherein the single molecule light sources are fluorescent single molecules.

3. A method according to claim 1 or 2, further comprising an initial step of obtaining the series of pixellated images of the optical section of the specimen.

4. A method according to any one of the previous claims, wherein, in step (b) , the tracked intensity of each pixel group is a temporal running average intensity.

5. A method according to any one of the previous claims, wherein, in step (b) , the tracked intensity of each pixel group is relative to background intensity.

6. A method according to any one of the. previous claims, wherein, in step (c) , the change in tracked intensity is a step change in intensity.

7. A method according to claim 6, as dependent on claim 2, wherein, in step (c) , the intensity behaviour includes a stepwise decrease in intensity caused by fluorophore photobleaching or fluorophore removal from the optical section.

8. A method according to claim 6, as dependent on claim 2, or according to claim 7, wherein, in .step (c) , the intensity behaviour includes a stepwise increase in intensity caused by fluorophore appearance in the optical section.

9. A method of analysing a series of pixellated images of an optical section of a specimen, comprising: identifying single molecule light sources in the images according to the method of any one of the previous claims; and determining the lateral position of the identified single molecules in the optical section.

10. A method of analysing a series of pixellated images of an optical section of a specimen, comprising: identifying single molecule light sources in the images according to the method of any one of claims 1 to 8; and tracking lateral movement of the identified single molecules in the optical section.

11. A method of analysing a series of pixellated images of an • optical section of a specimen, comprising: identifying single molecule light sources in the images according to the method of any one of claims 1 to 8; and collecting statistical data on the identified single molecules.

12. A computer-based system for identifying single molecule light sources in a series of pixellated images of an optical section of a specimen, the system comprising a processor which is operatively configured for: (a) over the series of images, tracking temporal changes in intensity of pixel groups of a size corresponding to the diffraction-limited image area of a single molecule light source in the optical section; and (b) identifying single molecule light sources in the images by selecting those pixel groups from step (a) which exhibit a change in tracked intensity corresponding to an expected intensity behaviour of a single molecule light source in the optical section.

13. A computer-based system according to claim 12, further comprising an image storage device for storing the series of images and communicating the series of images to the processor.

14. A microscope system comprising: a microscope for imaging an optical section of a specimen; an image registering device arranged for registering a series of pixellated images obtained by the microscope; and a computer system according to claim 12 or 13 for • identifying single molecule light sources in the series of images registered by the image registering device.

15. A microscope system according to claim 14, wherein the microscope is adapted to perform total internal reflection fluorescence microscopy.