SEARCH

SEARCH BY CITATION

Keywords:

  • membrane diffusion;
  • plasticity;
  • receptors;
  • single molecule;
  • synapse

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

Single-particle tracking (SPT) applications have been growing rapidly in the field of cell biology, and in particular in neurobiology, as a means of unravelling the involvement of diffusion dynamics of neurotransmitter receptors and other synaptic proteins in the regulation of neuronal activity. Suitable probes and technological improvements make SPT more accessible than it used to be and open up broad applications in cellular biology. In this technical highlight, we give an overview of the experimental approach in SPT. The concepts and results in neurobiology have already been the object of detailed reviews. Here, we focus on a qualitative description of the implementation of SPT, from molecule labelling to acquisition, data treatment and analysis of protein diffusion properties. Constraints, limitations and future developments are discussed.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

Surface protein trafficking has emerged as an important mechanism underlying cell membrane organization. Advances in real-time imaging in living cells have enabled tracking of individual molecules over time and space, and the development of single-particle tracking (SPT; particle refers in a broad way to any probe selectively attached to a molecule of interest). SPT has been used in various fields of cell biology. In neurobiology, emphasis has been put on the dynamic regulation, by lateral diffusion in the plasma membrane, of neurotransmitter receptor density at postsynaptic densities (Meier et al., 2001; Dahan et al., 2003; Tardin et al., 2003; Charrier et al., 2006; Levi et al., 2008). A diffusion-trapping model has been proposed in which receptors can exchange rapidly between the intra- and extrasynaptic compartments with transient stabilization of the receptor at synapses by interaction with scaffold proteins (Choquet & Triller, 2003; Triller & Choquet, 2005). It has been shown that glycine receptors (GlyRs) exchange continuously between various associated states with gephyrin, inside and outside of synapses, and that the association–dissociation dynamics take part in the regulation of the number of receptors at inhibitory synapses (Ehrensperger et al., 2007). Similar results have been obtained for excitatory metabotropic glutamate receptors and AMPA receptors (AMPARs) respectively with Homer and PSD-95 (Serge et al., 2002; Bats et al., 2007). More recently, the physiological significance of these molecular dynamic regulations have been tackled. It has been shown that the lateral diffusion properties play a part in the compensation mechanisms in desensitization of AMPARs (Heine et al., 2008) as well as in the regulation of GlyRs and γ-aminobutyric acid-A receptors (GABAARs) in response to changes in neuronal activity (Levi et al., 2008; Bannai et al., 2009). SPT is thus now one of the complementary methods in neurobiology for the study of the molecular mechanisms involved in the regulation of neuronal activity.

SPT relies on the specific attachment of a probe, usually fluorescent, to the proteins or lipids of interest, followed by the detection of its position as a function of time with 30–100 Hz acquisition frequency and spatial resolution of 10–40 nm. It provides an improved resolution compared to the usual resolution of a light-focusing microscope, which is limited to λ/2NA ≈ 200–500 nm, where λ is the light wavelength and NA the numerical aperture of the lens. Such a molecular resolution has not been reached in approaches used to study membrane dynamics, such as fluorescence recovery after photobleaching (FRAP; Axelrod et al., 1976) or fluorescence correlation spectroscopy (FCS; Haustein & Schwille, 2004; Thoumine et al., 2008). Both FRAP and FCS give averaged mobility of multiple molecules, with a spatial resolution limited by the light diffraction limit (∼200–500 nm depending on the excitation wavelength). However, bulk approaches are still crucial to understanding diffusive behavior, as SPT can always be biased by sampling effects. Furthermore, SPT, FRAP and FCS can be used to probe various time and space scales.

In this review, we aim to provide a general view of SPT. We introduce the various probes and labelling techniques used in SPT as well as their related instrumentation requirements. Extensions to multicolour SPT and tissue imaging are discussed. Finally, we describe the imaging conditions and data analysis, together with the limitations and experimental constraints.

Probes and live detection for single molecules

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

Various strategies have been developed for single-molecule detection in live cells (Fernandez-Suarez & Ting, 2008). They differ in the nature of the probe, the linker used to attach it to the molecule of interest and the detection mode (Fig. 1). They are more or less adapted depending on the location of the protein or lipid of interest (e.g. cytoplasmic, external membrane layer) and the temporal and spatial resolution needed. One of the main constraints in single-molecule detection is the ability to obtain a high signal-to-noise ratio, which depends on the photophysical properties of the probe itself (Fig. 1A) and the optical system used to monitor the signal (illumination mode, collection factor). To date, the most popular probes used to study the movement of plasma membrane proteins are quantum dots (QDs) and cyanine dyes (Cy), combined with wide-field epifluorescence microscopy.

image

Figure 1.  Fluorescent probes for single-molecule tracking. (A) The three main fluorescent probes used for single-molecule tracking are quantum dots (QDs), organic dyes and fluorescent proteins. The quantum dots are the most widely used probes as they give access to emission wavelengths in the whole visible spectrum, high photostability allowing tracking for up to 20 min and high brightness, which provides optimal time and space resolution. However, QDs lead to steric hindrance due to the large size of the tagging complex and their erratic fluorescence blinking add complexity to the tracking procedure. QD properties can be modified through the semiconductor and coating design. Organic dyes have diverse structures and photophysical properties that can be designed through organic synthesis. Their small size make them less perturbative than QD; however, they are less bright and stable making them suitable only for short trajectories of a few seconds. Fluorescent proteins (FPs) are genetically encoded and they are highly specific with controlled one-to-one stoechiometry. Despite their poor photostability and brightness, they have promising potential for sptPALM applications. (B) Main labelling procedures. The successive steps are colour-coded. (a) Antibody (Ab)-based procedure. Surface proteins can be labelled by sequential incubations of (1) primary Ab, Fab–biotin and probe–streptavidin, (2) primary Ab and precoupled Fab–probe, or (3) by direct incubation with the precoupled Ab–probe complexes. The choice of an antibody strategy depends on the primary antibody amenability. (b) Biotin ligase procedure. The acceptor peptide (AP)-tagged protein can be biotinylated either by addition of BirA, biotin and ATP to the cell medium, or by intracellular BirA expression together with the AP tagged protein. (c) NTA procedure. The histidine-tagged protein (His6 or His10) forms a nickel complex with the NTA-probe moiety. This association is reversible and specificity varies from one cell type to another.

Download figure to PowerPoint

Inorganic quantum dots ally brightness and stability

Commonly, QDs are nanometer-sized semiconductor crystals surrounded by a ZnS shell (Michalet et al., 2005). Their photophysical properties have made QDs a major probe for single-molecule tracking (Jaiswal & Simon, 2004). Compared with the usual organic dyes, their very high brightness provides a high signal-to-noise ratio (up to 25), even under standard wide-field illumination. The high flux of photons makes the fluorescent signal nonlimiting and acquisitions can usually be carried out at video-rate frequency, the time resolution being mainly limited by the speed of the camera data transfer. Their narrow emission wavelength, which is dependent on the QD core size, gives access to the whole visible spectrum with high spectral resolution, allowing multicolour imaging. Finally, they have excellent photostability, which allows recordings over periods of time from milliseconds to several minutes. A particular property of QDs is their alternation between ‘on’ and ‘off’ states, known as blinking. This complicates the particle tracking but ensures the identification of single QDs because signals alternate between 0 and 1, and would be fractional in the case of multiple QDs. However, one cannot rule out the possibility that multiple receptors are bound to a single QD, and this point must be considered carefully.

Recent studies have reported that the relatively large size of QDs (15–25 nm) could limit their access to confined compartments such as synaptic clefts (Groc et al., 2007; Howarth et al., 2008). Efforts are being made to reduce QD core size as well as the size of the organic coating layer. This coating layer, which ensures their solubility in water and stability with regard to aggregation, is obtained either by surface shielding chemistry, such as encapsulation in phospholipid micelles and surface coating with amphiphilic polymers, or by exchange surface chemistry, for instance with peptides (Pinaud et al., 2006). With these surface treatments, the QDs present reactive groups that provide various possibilities for coupling to biological molecules. QDs functionalized with biotin, streptavidin or IgG are commonly used and are commercially available.

Organic dyes and background rejection

Organic dyes are much smaller fluorescent probes (< 1 nm) than QDs and can be easily coupled to any ligand. Their small size means they are suitable for tracking molecules that happen to travel in narrow and crowded environments such as synapses (20–30 nm). However, their use in SPT is limited by their short fluorescence stability. In addition, saturating illumination conditions, using laser excitation, are required to maximize the number of photons emitted by single dyes. This high-power illumination accelerates the photobleaching of the dye, which can only be imaged for a few seconds at video rate (30 Hz), and can be deleterious with regards to the cell. Optimization of the fluorescence detection efficiency is necessary for both organic dyes and QDs. It requires: (i) use of appropriate filter sets with high light transmission (> 80%); (ii) microscope objectives with high numerical aperture (e.g. 60×, NA > 1.4); (iii) new generations of electron-multiplying charge-coupled devices (video cameras) allying high sensitivity and fast frame transfer. Indeed, classical wide-field can be used in the case of organic dyes, but difficulties arise from to the short lifetime of the dye and the background noise. The background noise can be improved by the use of total internal reflection fluorescence (TIRF) microscopy (Sako & Uyemura, 2002). Creation of the evanescent wave at the glass–water interface limits the illumination to a thin slice of the sample close to the glass cover-slip, avoiding excitation of molecules that are out of focus. The working depth is then restricted (∼200 nm) to the plasma membrane and the nearby cytoplasmic zone beneath the plasma membrane. Such an illumination configuration greatly improves the signal-to-noise ratio and is now used in many laboratories. An alternative illumination method (HILO) consists of using highly inclined thin illumination (Tokunaga et al., 2008), which penetrates the entire thickness of a cell. Introducing a pinhole in the emission light path (Yang & Musser, 2006) or using patterns of speckles in a rotating disk also reaches confocal-like quality in the XY axis (Ventalon & Mertz, 2005). This last approach uses simple improvements, allowing its application to any wide-field microscope setup. ‘Speckle disk’ devices appear to be a promising tool for SPT, although image processing might need higher speed for live-cell imaging.

Probes for SPT photoactivation localization microscopy (PALM)

Using genetically encoded fluorescent proteins (FPs) ensures the specificity of the labelling with a covalent link to the protein of interest. However, the poor fluorescence signal of FPs and their rapid photobleaching limit their use in SPT. Recent developments in single-emitter super-resolution techniques such as PALM (Betzig et al., 2006), its single-particle tracking variant sptPALM (Manley et al., 2008) and stochastic optical reconstruction microscopy (STORM; Bates et al., 2007) have opened up considerable opportunities for the use of FPs using photoswitchable or photoactivatable fluorescent proteins (e.g. Dronpa, EosFP etc; Fernandez-Suarez & Ting, 2008). The general principle consists of ‘turning on’ these fluorescent proteins using illumination conditions such that, statistically, only well separated single fluorophores will be detected. Iterative illumination is used to sequentially reconstruct the full image from the emission of single fluorophores. In addition to genetically encoded probes, photochromic synthetic molecules can also be used (e.g. rhodamines and cyanines; Fernandez-Suarez & Ting, 2008). These can have photophysical properties superior to those of FPs but are more limited with regards to protein tagging, even though rhodamine dyes are cell membrane-permeant. PALM is a recently developed technique that can be complementary to SPT. While SPT gives direct access to the dynamics of transmembrane proteins it has not yet been applied successfully to intracellular proteins, mainly due to the difficulty of tagging those proteins with QDs. Thanks to the genetically expressed FPs, PALM can circumvent this difficulty and can thus be used for intracellular SPT as well as for super-resolution imaging. Combining PALM with SPT, using a photoactivatable fluorescent protein chimera, Manley et al. (2008) recently resolved the dynamics of individual membrane fluorescent proteins in living cells. In addition, the very high density of the tracked particles (50 per μm2) enables creation of spatial maps of diffusion coefficient, accounting for the heterogeneity of molecular environments within cell compartments.

Photothermal gold particle imaging

A gold particle imaging method for single-particle tracking has been developed as an alternative to fluorescence-based QD microscopy (Boyer et al., 2002; Lasne et al., 2006). It is based on photothermal interference contrast from small metallic particles down to 1.4 nm in diameter. This technique makes use of the advantage of the absorption properties of the metallic particle and the total conversion of this energy into heat. Resulting local changes in the diffraction index around the particle are monitored using specific instrumentation combining heating and probe laser beams. The heat-induced phase modulation of the beam is detected using a photodiode coupled to a lock-in amplifier. The absence of photobleaching issues, minimized size of the particles and easy coupling chemistry make gold nanoprobes a promising alternative to QDs. Importantly, it has been evaluated that the temperature effect due to the local heating should be negligible as the temperature rise is limited to 1.5 K in the case of a 5-nm gold particle (I = 400 kW/cm2; Lasne et al., 2006), and is rapidly dissipated.

Labelling strategies

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

There are different ways to attach exogenous probes (QD, dye, gold particle) to the molecule of interest expressed at the membrane. Natural ligands, peptides, chemical linkage, biotin–avidin or antibodies could be used to link the molecule of interest to the probe. The linker should fulfill several requirements: (i) the labelling site should be on the extracellular part of the target; (ii) the linker should bind specifically and stably to its target; (iii) the linker should be as small as possible to minimize the steric hindrance of the labelling complex; and (iv) the target : linker : probe labelling stoichiometry should be 1 : 1 : 1. These components and labelling conditions should not affect neuron health or signalling pathways.

Antibodies

Antibodies can be used to conjugate probes to endogenous proteins (Fig. 1B; also Bannai et al., 2006; Ehrensperger et al., 2007; Renner et al., 2009). This classic method has been used successfully in many studies on neurotransmitter receptors. However, suitable antibodies are not always available or may have low binding affinity or poor specificity for their antigen. In addition, antibodies are relatively big (13 nm) and have two antigen-specific binding fragments that might lead to crosslinking. Using Fab fragments of the primary antibodies halves the size and valence of the linker. Unfortunately, efficient Fab fragments are not always easy to obtain. When available, antibodies provide a straightforward and efficient way to tag the protein of interest. Alternative labelling strategies to circumvent the disadvantages of antibodies are described below.

Peptide-based linkers

Peptide-based linkers (Fig. 1B) can be genetically encoded and have the advantage of being small (a few amino acids) and of providing specific labelling sites. The Tris-nitriloacetic acid (NTA; Guignet et al., 2007) and acceptor peptide (AP) tag (Howarth et al., 2005) approaches offer interesting features for SPT experiments in living cells.

A poly-histidine tag (Hisn with n = 6–10 amino acids) interacts specifically with the Ni-NTA moiety. This noncovalent labelling method is based on the coordinated interaction between histidine and a nickel ion, chelated by NTA. Interestingly, the design of probes conjugated to multiple NTA recognition units, such as tris-NTA, has clearly improved the binding avidity and stability of the complex in vitro and in cellular systems (Lata et al., 2006). Tris-NTA methodology has been successfully used in a recent study, in which tris-NTA bound to QDs was applied to living cells expressing a His10-tagged protein at the surface, and provided specific, long-lasting labelling (Roullier et al., 2009). However, the specificity of the labelling appears to be quite variable depending on the cell type (G. Gouzer, D. Alcor, M. Dahan and A. Triller, unpublished observation), suggesting that membrane and matrix composition could in some cases provide endogenous interacting sites for Ni-NTA moieties.

The AP tag (15 amino acids) approach requires BirA, a bacterial enzyme, which specifically and covalently binds biotin to the AP tag. The biotin can, in a second step, bind to any streptavidin-coupled probe. Cotransfection of the AP-tagged protein and the BirA enzyme gene (Howarth et al., 2008) makes the labelling protocol less disruptive as it circumvents the incubation of the cells with biotin, BirA and ATP, the target protein being biotinylated by the BirA intracellularly, in the secretory pathway. Combined with monovalent streptavidin QDs, this approach optimizes the size and stoichiometry of the target–probe complex.

These peptide- and FP-based strategies require transfection of exogenous chimeric proteins. Recurrent questions about the transfected chimera need to be resolved case by case. For example, is the fusion protein as functional as the endogenous one? Does the overexpression modify the system or generate some artifact? Chimera behavior should always be compared with the behavior of the endogenous protein and the development of knock-in animals constitutes the best way to minimize overexpression problems.

Developments for single-molecule labelling

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

Control of stoichiometry and size reduction of the labelling complex

On the molecular scale, the movement of labelled membrane proteins is dominated by their transmembrane domain due to the high viscosity of the membrane, which is ∼100–1000 times greater than the viscosity of the extracellular medium. Tagging strategies involving potentially multivalent elements (e.g. secondary antibody QDs functionalized with multiple linkers) can lead to protein crosslinking. Due to the characteristics of protein diffusion in the plane of the membrane, the diffusion coefficient should not be much affected as long as only two proteins are crosslinked. However, in the case of an increase in the size of the aggregates, the distribution of lateral diffusion coefficients might be modified. More subtle alterations implicating rotational (Brokmann et al., 2005) or anomalous (Ribrault et al., 2007) diffusion can be detected in the cases of size increase and asymmetric molecular complexes, respectively. Crosslinking can also lead to artifacts such as deregulation of signalling pathways or association with other proteins such as the cytoskeleton. The target : linker : probe complex should be obtained with a stoichiometry of 1 : 1 : 1. This is a difficult task requiring monovalent linking methods such as Tris-NTA and AP tag, as well as the development of preparation methods to obtain QDs functionalized with a single linker.

Good tagging strategies should reduce the number of labelling steps and reduce the size of the probe complex using smaller linkers and smaller QDs with smaller crystal cores (e.g. InP-cored QDs). In confined conditions such as the synaptic cleft (∼30 nm), the size and valency of the probe complex could influence the speed of diffusion as well as the access of the tagged receptors to synapses. Nonetheless, while the relative diffusion coefficient is indeed affected by probe size, experimental data show that this steric hindrance does not limit the diffusion coefficient because the protein can be accelerated or slowed down by pharmacological treatments (Levi et al., 2008).

Multicolour labelling

Examination of correlated diffusion of two distinct molecules makes it possible to study the dynamics of protein–protein interactions at the single-molecule level. This approach requires orthogonal labelling strategies. Recently (Roullier et al., 2009), the subunits 1 and 2 of the IFNalpha receptor were labelled in COS cells with the His and AP tag strategies, respectively. Simultaneous tracking of QD655 and QD605 attached to each subunit was performed using a dual view system, allowing dual wavelength recording. Results show that IFNAR1 and IFNAR2 could display a concomitant and colocalized slowing diffusion behavior, suggesting that both molecules were at least in the same membrane subcompartment, if not interacting. Co-diffusion of single molecules should make it possible to distinguish colocalization and molecular interactions in that co-movement is a strong signature of the physical link (direct or indirect) between two molecules. It is also a direct way to access the lifetime of the formed complex. However, the detection of single-molecule interactions is limited by the low probability of detecting such events. High density single-molecule labelling or other alternative strategies that could favour the detection of such events have yet to be developed.

SPT in integrated networks

Data collected from dissociated cultured neurons enable the study of molecular mechanisms in the complexity of the cellular environment. However, these in vitro models are limited and can only be used as a first approximation towards a better understanding of the physiology of neuronal networks. Taking advantage of the high brightness and stability of QDs, it should be possible to develop single-molecule live imaging in tissue slices. SPT in slices requires the reduction of probe size so as to improve penetration into the tissue, and improved stability and specificity of the labelling, which might be more stringent than in dissociated cell cultures. Signal-to-noise ratio optimization for QD live imaging in slices will require drastic changes in the imaging system, confocal-like techniques being preferred to classic wide-field imaging. Finally, single-molecule tracking in slices will require the ability to localize the probe accurately on a well identified structure (e.g. dendrite, synapse, glia) and to reconstruct the trajectories in three dimensions (Ram et al., 2008).

Acquisition of single-molecule movies

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

One of the most straightforward ways of carrying out SPT experiments is to use QD-tagged proteins in dissociated cell cultures with a wide-field standard microscope equipped with high quality optics and electron-multiplying charge-coupled devices. Acquisitions commonly require three types of data (Fig. 2A): (i) the bright-field image to visualize the cell of interest, and to check its morphology and the specificity of QD labelling; (ii) the QD movie whose acquisition conditions will be discussed below; and (iii) additional fluorescent snapshots for the localization of the fluorescently labelled structures of interest. At the single-molecule level, movement is recorded as a trajectory composed of elementary steps which are identified by the {X,Y} coordinates of the particle at each time point.

image

Figure 2.  GABAARs tracked on dissociated hippocampal neurons. (A) Recording of QD-GABAAR γ2-subunit movement for ∼40 s. The green signal (see colour in on-line graphics) corresponds to the maximum projection of QD fluorescence over time and the red signal to FM4-64 staining of active presynatpic boutons. Dendrites are localized on the brightfield image. GABAAR was labelled using anti-γ2-GABAAR Ab + Fab-biotin + QD (605 nm)-streptavidin. (B) At each frame, the position of QDs is determined with a subpixel pointing accuracy by fitting the fluorescence signal (thick curve) with a 2-D Gaussian PSF (thin curve), providing a spatial resolution of 10–20 nm depending on the signal-to-noise ratio. (C) Example of a single QD-GABAAR trajectory diffusing in to (green/light portion) and out of (blue/dark portion) the FM4-64-stained synapse. (D) Fluorescence signal from a single fluorophore over time: one-step bleaching. (E) Signal from a single QD: blinking of the QD leads to oscillations between the ‘on’ and ‘off’ states.

Download figure to PowerPoint

As discussed before, single-molecule detection requires efficient elimination of noise. In addition, pointing accuracy of sub-diffraction resolution is critical for the tracking of single molecules positions from frame to frame, as it defines the {X,Y} position accuracy and thus the sensitivity with regards to elementary displacement (Cheezum et al., 2001). Pointing accuracy depends mainly on three parameters. The first one is the point spread function, which is commonly used for sub-pixel localization using two-dimensional Gaussian fit of the fluorescence signal (Fig. 2B). The second parameter is the pixel resolution of the image, which introduces uncertainty in the position of the arrival of photons. Pixel size is typically ∼200 nm for a 60× objective (NA 1.4), which allies high resolution, efficient photon collection and a reasonable field of view to visualize a significant number of easily distinguishable single molecules. The third parameter is the sub-pixel pointing accuracy of the fluorescent probe position, which depends strongly on the signal-to-noise ratio: it falls from 10 to 100 nm when the signal-to-noise ratio drops from 10 to 4 and below (Cheezum et al., 2001).

Taking advantage of QD brightness, combined with electron-multiplying charge-coupled device technology, a high enough signal-to-noise ratio can be reached. It is commonly possible to do full chip video-rate stream acquisitions with an exposure time of 30 ms. For faster acquisition (typically 10 ms), it is possible to restrict the acquisition area so as to allow quicker data transfer. The acquisition frequency, which is defined as the inverse of the exposure time, is an important parameter with many implication for the analysis of molecular dynamics. While the signal-to-noise ratio partially defines pointing accuracy, the acquisition frequency has implications for both pointing accuracy (the probe position is averaged over the exposure time) and sampling of the trajectory. The higher the frequency, the higher the number of individual steps. Therefore, SPT experiments are a compromise between time resolution, spatial resolution and accuracy of the calculated diffusion parameters. The goodness of a diffusion coefficient, calculated from a single trajectory, depends on the diffusion dynamics, the signal-to-noise ratio, the number of sampled coordinates and the frequency of acquisition.

In SPT experiments, trajectories are recorded and subsequently analyzed individually or globally. In order to reconstruct individual trajectories and to avoid possible particle crosslinking (see stoichiometry issues), labelling is usually performed at a low density. However, biological systems are highly heterogeneous and accumulation of numerous individual trajectories is mandatory for a reasonable sampling of the overall dynamic behavior. Also, bias can emerge from uneven labelling of the protein subpopulations. Statistically relevant sampling can be achieved by increasing the density of labelling and the length of the trajectories. The length of the trajectories determines the ability to detect changes of behavior in the diffusion regime of a given tracked particle and also to have long enough portions of trajectories associated with each of these regimes.

Tracking procedure

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

In SPT, the aim is to reconstruct the (x,y,t) trajectory of every single particle (Fig. 2A–C). To be identified as a single molecule, single-step photobleaching will be expected for single organic dyes (Fig. 2D), and transitory ‘dark states’ due to blinking should be observed for QDs (Fig. 2E). Data processing for molecular movement quantification is mainly divided into two steps consisting of single-particle detection followed by trajectory reconstruction, achieved by reconnecting spots from one frame to the next. The pertinent parameters to be extracted from trajectories are the diffusion coefficient, the confinement and the dwell times. These elements reflect biological properties of the plasma membrane and of the molecular interactions. The analysis of traces allows access to modifications which may take place during the recording.

Sub-pixel localization of single molecules

The algorithms used have a major role in the accuracy, efficiency and limits of the technique (Cheezum et al., 2001). To date, the main tracking algorithms that have been used involve cross-correlation (Gelles et al., 1988; Kusumi et al., 1993; Bachir et al., 2006), centroid (Ghosh & Webb, 1994; Hanus et al., 2006) or Gaussian fits (Anderson et al., 1992; Schutz et al., 1997). Each of them might be more or less suitable depending on the experimental system itself. However, when considering a single molecule labelled with a fluorescent probe with sub-wavelength size, then a two-dimensional Gaussian fit on thresholded images provides the best results in terms of localization, even at relatively low signal-to-noise ratio (< 4). Unlike cross-correlation algorithms (based on relative changes in position from one frame to the next), centroid and Gaussian fits are based on the determination of absolute positions and thus their accuracy is independent of the amplitude of the actual displacement. The two-dimensional Gaussian fit is based on deconvolution of the fluorescence signal with the point spread function (Fig. 2B), which describes the response of the imaging system to a point source. Spot size discrimination as well as intensity can be used to identify relevant objects and/or to distinguish nearby objects. However, the intensity criteria can be difficult to apply when using QDs because of the random blinking.

Trajectory reconstruction

Trajectory reconstruction consists of linking the position of a particle at a given frame to its position in the next frame (Fig. 3). This linking procedure can be carried out using efficient automatic algorithms. However, the final trajectories have to be checked manually. Whereas most of the spots might have been tracked successfully, several issues have to be addressed, such as: (i) the possibility of QDs being tracked or detected on cellular fragments, on a glass surface, or on nontransfected cells; (ii) the jump of a trajectory from one QD to another; (iii) the discontinuity of single trajectories that may appear as partial fragmented trajectories. These difficulties can be minimized by working at low particle density (Ghosh & Webb, 1994). Conversely, additional difficulties arise on increasing QD density, as this makes it difficult to distinctly follow nearby QDs and blinking QDs, which leads to periods of dark states during which distinct QDs might be mistaken. Blinking and high densities of tagged molecules can be dealt with by using partial manual reconnection algorithms. However, they are time-consuming and subject to operator subjectivity. Therefore, automation of reconnection is of major importance. It requires robust criteria of assignment (spot sizes, explored distances, intensities etc.) and must be able to reject uncertain trajectories efficiently. Algorithms to circumvent these issues have recently been published, and deal particularly with blinking (Bonneau et al., 2005; Bachir et al., 2006), signal-to-noise ratio and threshold in spot detection (Yoon et al., 2008) and with automation of detection and tracking (Mashanov & Molloy, 2007; Jaqaman et al., 2008; Serge et al., 2008). When the density of tagged molecules is high, a recursive detection algorithm (Serge et al., 2008) can be used to discriminate particles whose fluorescent spots would merge transiently during their trajectories. The difficulty with the linking process is that individual particles can cross over, merge, split, interact, blink (in the case of QDs) or bleach. To solve these problems, a multiple-hypothesis tracking algorithm has been proposed (Jaqaman et al., 2008), which tends to determine the largest nonconflicting ensemble of trajectories via a temporally global optimization incorporating the history of every single trajectory.

image

Figure 3.  Flowchart of single-particle tracking processing. Reconstruction of the trajectories proceeds in three steps, which consist of individual spot detection followed by their sub-pixel localization using a two-dimension Gaussian fit and, finally, reconnection of the spots from one frame (k-1) to the following one (k). Reconnection can be done accurately using automatic algorithms in the case of low density of tagged molecules. Conflicts in the trajectory reconstruction (due to nearby QDs, blinking issues etc.) must be solved either manually or by using automated advanced algorithms. According to the biological questions, trajectories can be analyzed in correlation with some subcellular compartments. Characterization of the movement is finally done on the basis of a Brownian diffusion hypothesis, and calculation of the diffusion parameters provides quantitative analysis.

Download figure to PowerPoint

Being able to carry out SPT at a high density of labelled single molecules is an important challenge in order to obtain statistically relevant samples including minor populations of biologically relevant behaviors. In addition, studying transient interactions at the level of single molecules requires high density of labelling to improve the probability of detecting such events.

Trajectory analysis

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

In the plasma membrane of cells, movement of molecules is imprinted by their random lateral impacts with surrounding molecules due to thermal agitation, resulting in a first approximation to random Brownian diffusion, which is ideally described by Fick’s laws. The diffusion coefficient and mean square displacement (MSD), which expresses as a function of time the average surface explored from an initial position, are commonly used to characterize and quantify the movement.

Local dynamics and trajectory decomposition

In SPT, detailed analysis of the individual trajectories is needed to extract information on the underlying physicochemical mechanisms, in particular by accurately detecting time- and/or space-dependent transitions of diffusion behavior (exemplified in Fig. 4). They reflect changes in the molecular interactions and/or changes in the local membrane structure. Behavioral transitions within an individual trajectory can be identified by plotting the instantaneous diffusion coefficient (Dinst) as a function of time (Fig. 4A). The entry of a receptor into a synapse can reduce Dinst due to a change in the membrane structure or due to its interaction with proteins (e.g. scaffolding proteins). One major difficulty in the identification of those transitions is due to the erratic variations in Dinst and the limited number of data points obtained for each presumed regime. Sub-division of trajectories can rely on compartment identification with specific fluorescent labels. For example, FM4-64 can be used to identify active synapses and fluorescent fusion proteins (e.g. gephyrin, Homer, Munc etc.) to identify inhibitory synapses, excitatory synapses or presynaptic elements. Compartment location is commonly restricted by the diffraction limit, allowing only ∼300 nm resolution, while the trajectory is resolved with 10- to 15-nm spatial resolution. A further limitation is the variability introduced by the determination of the compartment edges, which requires intensity thresholding. Wavelet coefficient-based filtering provides a robust threshold method (Thery et al., 2004), limiting the dependence on the signal-to-noise ratio. Developing new methods to detect transitions within the trajectory combined with hyper-resolution techniques such as PALM or stochastic optical reconstruction microscopy (STORM) is a way of improving compartment identification (Betzig et al., 2006; Bates et al., 2007).

image

Figure 4.  Illustration of QD-GABAAR motion analysis on hippocampal rat neurons at 15 days in vitro (shown in Fig. 2). (A) Dinst as a function of time for the trajectory illustrated in Fig. 2. Time resolution (750 ms) is defined by the sliding window used for Dinst calculation from the short-time MSD. Top lines indicate synaptic and extrasynaptic location. An increase in Dinst, and in the amplitude of its erratic variations when the receptor exits from the synapse can be seen. *Stars indicate periods of QD dark state (blinking) along the trajectory. (B) MSD vs. time for the synaptic and the extrasynaptic portions of the trajectory (upper and lower values, respectively). The first initial points of the plot can be fitted with a linear slope, providing the average diffusion coefficient in the synapse (0.0176 μm2/s) and outside the synapse (0.0282 μm2/s). The curved shape of the lower plot indicates that the movement is confined. The asymptote gives access to the size of the confinement domain (diameter 0.106 μm). (C) Cumulative frequency distribution of the diffusion coefficient for QD-GABAAR on synaptic (n = 119) or extrasynaptic (n = 113) membrane. The synaptic distribution is shifted to the left compared to the extrasynaptic ones, indicating a slowing down at synapses. (D) Distribution of QD-GABAAR dwell times at synapses. The dwell time is defined as the percentage of time spent by a QD at a synapse during the acquisition duration (38.5 s in this case). The ‘immobile’ fraction refers to QDs that reside at synapses during the full acquisition duration.

Download figure to PowerPoint

Quantification of diffusion properties

Trajectory analysis is commonly based on the Brownian diffusion hypothesis and trajectories are usually characterized from the calculation of Dinst, the MSD, the confinement index, the size of the confinement domain and the dwell time along the trajectory (Saxton & Jacobson, 1997). Experience has shown that, within the sampling frequency limit, the diffusion of membrane proteins often differs from simple Brownian diffusion. This has been attributed to protein binding, crowding, constraints imposed by the cytoskeleton, active transport or even local membrane topology and composition. These factors plus statistical fluctuations lead to broad distribution of the diffusion coefficient, which typically extends over two to three orders of magnitude. Trajectories can be sorted on the basis of their compartment location and, by reference to the linear MSD obtained for regular diffusion, on the basis of the MSD curvature (Fig. 4B), which characterizes regular diffusion, oriented motion, confined diffusion and anomalous diffusion.

The MSD is obtained by averaging displacement over time intervals from single trajectories, error in the MSD calculation increasing when there are fewer points in the trajectory (Qian et al., 1991; Saxton & Jacobson, 1997). Dinst can be calculated accurately using a linear fit of the short-term MSD (Fig. 4B) (Kusumi et al., 1993). However, for longer time intervals, the number of data points decreases and, usually, only the first quarter of the MSD can be exploited. This defines the maximal time window accessible for the diffusion characterization and is commonly ∼ 10–20 s. This higher limit has direct implications for the analysis of confined diffusion, which results from the restriction, due to its environment, of the area that a particle can explore (Kusumi et al., 1993; Simson et al., 1995; Feder et al., 1996; Dietrich et al., 2002; Charrier et al., 2006). For a confined particle, the MSD is linear over short times and eventually bends to reach a plateau when the particle has had enough time to reach the edges of the confinement domain. The ability to identify the plateau and to quantify the confinement domain size depends on the characteristic time to reach the plateau with regards to the exploitable time window. In other words, for small confinement areas, high time and space resolutions are needed. However, for larger confinement regions, longer acquisitions are needed to detect the confinement effect on the MSD. For the calculation of the MSD to be usable, the mean spot displacement must be higher than the pointing accuracy. This defines the lower threshold of diffusion coefficient below which the particle is considered to be immobile. At constant exposure time, spatial accuracy decreases with increasing diffusion coefficient (Schnapp et al., 1988). Typically, in the experimental conditions described above, one may have access to diffusion coefficients ranging from 10−4 up to 10−1 μm2/s and confinement areas ranging from tens to hundreds of nanometers. The range of accessible diffusion properties depends on the pointing accuracy, frequency and duration of acquisition, but also on the diffusion coefficient of the probed molecule itself (Schnapp et al., 1988; Ritchie et al., 2005). As a consequence, comparing distributions of diffusion coefficients for acquisitions made with different sets of parameters is not straightforward. This is of particular importance when comparing various organic dyes with QDs, which differ greatly in their photophysical properties. Careful analysis is also necessary when comparing proteins with significantly different dynamic properties, as in the case of receptors at inhibitory and excitatory synapses.

Further consideration of diffusion properties

The above parameters are not sufficient for complete data interpretation on biological grounds. More parameters such as transition frequency (e.g. entry and exit from synapses, transitions from mobile to immobile state) and dwell time should also be considered. Transition frequencies are related to the statistics of the discrete molecular events involved and whether or not they are stationary. Dwell time is related to the stability of the molecular interactions and/or to confinement in a given compartment. Also, while directed motion can be diagnosed from the MSD through a parabolic component, short periods of directed motion are difficult to detect in SPT because movements are generally dominated by erratic Brownian motion. A velocity correlation-based algorithm has been developed to detect transient directed motion (Bouzigues et al., 2007).

Membrane topology (i.e. tubes such as axons or dendrites, dendritic spines etc.) may also influence the dispersion of the diffusion coefficient. Indeed, QD tracking is commonly done on a two-dimensional projection. However, QDs explore three-dimensional objects and diffusion coefficients might be affected by the local structure such as the dendrite shaft or the spines. Separating longitudinal and transverse components of a molecule diffusing into a nanotubule shows that measured anomalous diffusion can result from normal diffusion constrained by the local topology (Wieser et al., 2007). Thus, it is legitimately expected that analysis of the trajectories will benefit from three-dimensional tracking, taking into account the local membrane topology. Various strategies are being developed using cylindrical lenses, stacks of Z-slices, bifocal and defocused imaging (Speidel et al., 2003; Ram et al., 2007; Toprak et al., 2007).

All these analyses are based on the Brownian motion hypothesis. This prerequisite restricts the spectrum of possible dynamic behaviors, but also implies a loss of information as calculation of the diffusion parameters requires averaging of the trajectory over a time window. A major challenge in SPT is the ability to detect transitions between various dynamic behaviors such as Brownian diffusion, confined diffusion, local stops and deceleration or acceleration. The ultimate goal would be to identify nonsubjective criteria and derive analysis methods leading to assumption-free decomposition of single trajectories.

Conclusions and developments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

To get the best out of single-molecule tracking we need to improve experimental conditions and analysis for high-throughput accumulation of data. The sampling issue could be partially solved by acquisitions using a high density of single molecules combined with smart automated tracking algorithms. To improve the sampling over protein subpopulations, and to minimize the impact of labelling on protein properties, alternative probes and linkers have to be developed. These will allow multi-wavelength acquisitions and reduction of probe–linker size, with few labelling steps and increased stability. SPT in slices requires these improvements and the development of 3-D tracking algorithms. Furthermore, one should take into account the effect of local topology on diffusion properties.

Most of the parameters extracted from trajectory analyses are based on the diffusion hypothesis and involve calculation of MSD. From the MSD, it is possible to classify portions of trajectories as directed movement, Brownian diffusion or confined diffusion. This type of classification limits the identification of transitions between movement regimes and it would be of interest to have a diffusion assumption-free analysis of trajectories. The identification of these dynamic regimes is limited by the ability to decompose the trajectory into sub-portions using objective criteria. Commonly, trajectories are decomposed with regards to the identification of specific compartments. Super-resolution techniques such as PALM would improve the accurate localization of those compartments and so would help unravel different dynamic behaviors.

Single-molecule experiments and bulk experiments such as FRAP provide complementary time and spatial scales. A challenge is to be able to put together bulk and single-molecule measurements to develop quantitative models incorporating dynamic regulation with protein and physiological function (Holcman & Triller, 2006; Triller & Choquet, 2008). This type of development requires reliable quantitative data. However, in most studies, relative variations are analyzed and absolute measurements are usually not exportable. This is linked to the difficulty of standardizing experimental measurements, but is also due to the wide variety of data analysis methods and algorithms. It could be of great interest to develop open-source platforms for data analysis as well as standardized commercial instruments designed specifically for single-molecule experiments.

Acknowledgement

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References

This review was based on work which was made possible by the financial support of Inserm, Ecole Normale Supérieure, The Pierre-Gilles de Gennes Fundation and Agence Nationale de la Recherche (ANR- MorphoSynDiff).

Abbreviations
AMPAR

AMPA receptor

AP

acceptor peptide

BirA

bacterial biotin ligase

Cy

cyanine dye

Dinst

instantaneous diffusion coefficient

FP

fluorescent protein

FRAP

fluorescence recovery after photobleaching

GABAAR

γ-aminobutyric acid-A receptor

MSD

mean square displacement

NTA

nitriloacetic acid

PALM

photoactivation localization microscopy

QD

quantum dot

SPT

single-particle tracking

sptPALM

single-particle tracking variant PALM

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Probes and live detection for single molecules
  5. Labelling strategies
  6. Developments for single-molecule labelling
  7. Acquisition of single-molecule movies
  8. Tracking procedure
  9. Trajectory analysis
  10. Conclusions and developments
  11. Acknowledgement
  12. References