Here, we present a protocol for single particle tracking image analysis that allows quantitative evaluation of diffusion coefficients, types of motion and cluster sizes of single particles detected by fluorescence microscopy.
Particle tracking on a video sequence and the posterior analysis of their trajectories is nowadays a common operation in many biological studies. Using the analysis of cell membrane receptor clusters as a model, we present a detailed protocol for this image analysis task using Fiji (ImageJ) and Matlab routines to: 1) define regions of interest and design masks adapted to these regions; 2) track the particles in fluorescence microscopy videos; 3) analyze the diffusion and intensity characteristics of selected tracks. The quantitative analysis of the diffusion coefficients, types of motion, and cluster size obtained by fluorescence microscopy and image processing provides a valuable tool to objectively determine particle dynamics and the consequences of modifying environmental conditions. In this article we present detailed protocols for the analysis of these features. The method described here not only allows single-molecule tracking detection, but also automates the estimation of lateral diffusion parameters at the cell membrane, classifies the type of trajectory and allows complete analysis thus overcoming the difficulties in quantifying spot size over its entire trajectory at the cell membrane.
Membrane proteins embedded in the lipid bilayer are in continuous movement due to thermal diffusion. Their dynamics are essential to regulate cell responses, as intermolecular interactions allow formation of complexes that vary in size from monomers to oligomers and influence the stability of signaling complexes. Elucidating the mechanisms controlling protein dynamics is thus a new challenge in cell biology, necessary to understand signal transduction pathways and to identify unanticipated cell functions.
Some optical methods have been developed to study these interactions in living cells1. Among these, total internal reflection fluorescence (TIRF) microscopy, developed in the early 1980s, allows the study of molecular interactions at or very near the cell membrane2. To study dynamic parameters of membrane protein trajectories obtained from TIRF data in living cells, a single particle tracking method (SPT)is required. Although several algorithms are available for this, we currently use those published by Jaqaman et al.3 that address particle motion heterogeneity in a dense particle field by linking particles between consecutive frames to connect the resulting track segments into complete trajectories (temporary particle disappearance). The software captures the particle merging and splitting that result from aggregation and dissociation events3. One of the output data of this software is detection of the particles along the entire trajectory by defining their X and Y positions in each frame.
Once particles are detected, we apply different algorithms to determine the short timelag diffusion coefficient (D1-4)4,5. By applying the Moment Scaling Spectrum (MSS)6,7,8 analysis or by fitting the 'alpha' value by adjustment of the Mean Square Displacement (MSD) to the curve9, we also classify the particles according to the type of trajectory.
Analysis of spot intensity in fluorescence images is a shared objective for scientists in the field10,11. The most common algorithm used is the so-called Number and Brightness. This method nonetheless does not allow correct frame-by-frame intensity detection in particles in the mobile fraction. We have, thus, generated a new algorithm to evaluate these particle intensities frame-by-frame and to determine their aggregation state. Once the coordinates of each particle are detected using U-Track2 software3, we define its intensity in each frame over the complete trajectory, also taking into account the cell background in each frame. This software offers different possibilities to determine the spot intensity and the cell background and, using known monomeric and dimeric proteins as references, calculates the approximate number of proteins in the particle detected (cluster size).
In this article, we describe a careful guide to perform these three steps: 1) detecting and tracking single particles along a video of fluorescence microscopy using U-track; 2) analyzing the instantaneous diffusion coefficient (D1-4) of those particles and the type of movement (confined, free, or directed) of particles with long trajectories by MSS; 3) measuring the spot intensity along the video corrected by the estimated background fluorescence for each spot. This allows cluster size estimation and identification of the photobleaching steps.
The use of this protocol does not require specialized skills and can be performed in any laboratory with cell culture, flow cytometry and microscopy facilities. The protocol uses ImageJ or Fiji (a distribution of ImageJ12), U-track3, and some ad hoc made routines (http://i2pc.es/coss/Programs/protocolScripts.zip). U-track and ad hoc routines run over Matlab that can be installed in any compatible computer.
1. Preparation of Biological Samples
2. Selection of Images and Creation of Masks
3. Tracking the Particles
4. Calculation of the Diffusion Coefficients and Classification of Trajectories
5. Calculation of Cluster Ssize Through the Particle Density
NOTE: Be sure that all the scripts are invoked from the directory of the video being analyzed (in the example shown, VideoName/Serie1).
The use of this protocol allows the automated tracking of particles detected in fluorescence microscopy movies and the analysis of their dynamic characteristics. Initially, cells are transfected with the fluorescently-coupled protein to be tracked. The appropriate level of receptors presents on the cell surface that allows SPT is obtained by cell sorting (Figure 1). Selected cells are analyzed by TIRF microscopy that generates videos in a format that can be studied with the tools described in this protocol (Supplemental Video 1).
The videos generated cannot be directly analyzed, and independent frames of each video are required. The use of ImageJ generates files that can be interpreted using Matlab software such as video frames in .tiff format or masks. U-track tracks the particles seen in the selected video and saves the information in Matlab (.mat) files (movieData.mat, Channel_1_detection_result.mat, channel_1.mat, Channel_1_tracking_result.mat), see Figure 2.
As shown in Figure 2, the calculation of the diffusion coefficients and classification of trajectories commands, generates different files (diffusionCoefficients.txt, diffusionCoefficientsMobile.txt, diffusionCoefficientsShort.txt, trajectoryClassificationLong.txt, etc). The information contained in these files are best managed using Excel and Prism software. Information that can be obtained from this analysis includes:
The step 5 of the protocol analyzes the intensities of each particle along the trajectory. The script analyzeSpotIntensities will print on screen all the information analyzed. For each spot and frame, the script shows the coordinate of the spot in that frame (x,y) in pixel units, the estimate of the background fluorescence (k0), the raw spot intensity in the 3×3 patch, the corrected intensity (calculated as the raw intensity minus its background), the maximum value of intensity observed in the patch, and the region number within the mask where this spot is located at this frame. An example of the kind of output produced is
spot=43 frame=184
x=78.0397
y=72.5395
k0=1571
spotRawIntensity=5550.1111
spotCorrectedIntensity=3979.1111
maxCorrectedSpotIntensity=6243
maskRegion=2
All this information is stored in a log file (results/TrackingPackage/tracks/log.txt) and a table that can be read from Excel (results/TrackingPackage/tracks/spotIntensitiesByFrame.txt). After analyzing each spot, the script prints the average intensity along the trajectory and the majoritarian region within the mask
spot=43
meanCorrectedSpotIntensity along frames=4762.303
majoritarian region=3
This information is stored in the log file above and a table that can be read from a spreadsheet (results/TrackingPackage/tracks/meanSpotIntensities.txt). As an example, mean spot intensities (msi) for every particle along their trajectory in cells stimulated with different conditions is shown in Figure 4B. We may now use this information to roughly estimate the size of the fluorescent cluster. As a spot's mean corrected intensity is related with the number of fluorescent proteins present in this spot, and the fluorescence of a monomer can be measured through a similar but independent experiment, directly calculate the number of receptors per particle. The frequency distribution of receptor number per particle using as reference the intensity of the monomeric protein expressed in the same cells is shown in Figure 4C.
The gather diffusion and intensity command create two files: one called diffusionCoefficientsAndIntensitiesShort.txtand another called log_diffusionCoefficientsAndIntensitiesShort.txt in the directory results/TrackingPackage/tracks. Both files contain 1) the spot index, 2) the diffusion coefficient, 3) the intensity, and 4) the region number within the mask. These files can be read from a spreadsheet.
Similarly, the gather trajectory classification and intensity command will create two files one called trajectoryClassificationAndIntensitiesLong.txt and another log_trajectoryClassificationAndIntensitiesLong.txt in the directory results/TrackingPackage/tracks. The first one contains 1) the spot index, 2) the movement type, 3) the spot first moment, 4) the intensity, 5) D1-4 and 6) the region number within the mask. This file can be read from Excel.
A summary of all the files generated using this protocol is shown in Figure 2. Other analysis that can be performed using this protocol includes the comparison of the dynamic parameters of small vs bigger spots, i.e. monomers vs oligomers, variations on these dynamic parameters induced by ligands, inhibitors, membrane composition, alteration of signaling pathways, etc. A complete analysis of CXCR4 behavior in response to its ligand CXCL12 under different experimental conditions has been performed using this protocol13.
Figure 1: Cell sorting. Expression of GFP in Jurkat cells transfected with CXCR4-AcGFP before and after cell sorting. Cell expressing low levels of GFP (GFPlow) are selected and employed for TIRF experiments.
Figure 2: Summary of the files generated. The figure shows all the files generated using the Matlab routine described. Please click here to view a larger version of this figure.
Figure 3: Classification of trajectories. Number of the different types of trajectories corresponding to cells treated with different stimuli. (A) Percentage of immobile spots, (B) percentage of long trajectories and (C) type of movement of the long trajectories.
Figure 4: Diffusion coefficient, mean spot intensities and number of receptors per particle. (A) Distribution of the short diffusion coefficients (D1-4) values for each spot in response to different stimuli (I, II, II and IV). Red line represents the median value for D1-4. (B) Mean spot intensities (msi) values for each spot along its first 20 frames in response to different stimuli (I, II, II and IV). Red line represents the mean intensity value (SD). (C) Percentage of receptors/particle as extracted from intensity distribution of the individual particles, taking as bin width the monomeric protein intensity value. Please click here to view a larger version of this figure.
Supplemental Figure 1: Opening a TIRFM file in Fiji or ImageJ. Options that appear on the Bio-formats window upon dragging and dropping a .lif video. Please click here to download the figure.
Supplemental Figure 2: Select series. Images present in the .lif video. (A) Window that permits selection of the series to be analyze, including multichannel images. (B) Result example of series selection, including multichannel image (left) and corresponding video (right). Please click here to download the figure.
Supplemental Figure 3: Split channels. (A) Window capture of the ImageJ commands needed for splitting channels of a multichannel image and (B) result example of the channel split. Please click here to download the figure.
Supplemental Figure 4: Merge Channels. (A) Merging different channels in a single image. (B) Channel selection for the merging. (C) Result example of the channel merge. Please click here to download the figure.
Supplemental Figure 5: Synchronize windows. (A) Localization in the ImageJ menu of the commands required for image synchronization and (B) resulting window. Please click here to download the figure.
Supplemental Figure 6: Select the region of interest. (A) Selection of the window of interest using the rectangular selection tool and (B) result of the image crop. Please click here to download the figure.
Supplemental Figure 7: Save the video. Save the region of interest as an image sequence and parameters for the Save as Image sequence window. Please click here to download the figure.
Supplemental Figure 8: Segmentation. (A) Open the Segmentation editor plugin in the ImageJ's plugins menu. (B) Add labels/materials for the segmentation. Please click here to download the figure.
Supplemental Figure 9: Mask design. (A) Selection of the appropriate labels and definition of the green mask. (B) Selection of the red mask. Example of two masks corresponding to two labels. (C) Save the masks as a .tiff file. Please click here to download the figure.
Supplemental Figure 10: Matlab working directory. Selection of the correct directory containing the series to be analyzed. Please click here to download the figure.
Supplemental Figure 11: U-track main menu. (A) Movie selection, (B) movie edition and (C) channel Settings menus. Please click here to download the figure.
Supplemental Figure 12: Select the type of object to track. Select tracking of single particles. Please click here to download the figure.
Supplemental Figure 13: Detection of particles. (A) U-track, control panel. (B) Example of settings for the detection of particles. Please click here to download the figure.
Supplemental Figure 14: Example of results for particle detection. Different windows that include a viewer menu, movie options menu and the movie. Please click here to download the figure.
Supplemental Figure 15: Tracking menu. Example of settings. (A) Tracking, (B) setting – frame to frame linking and (C) gap closing, merging and splitting menus. Please click here to download the figure.
Supplemental Figure 16: Example of results for particle tracking. Please click here to download the figure.
Supplemental Figure 17: Track analysis menu. Example of settings. Please click here to download the figure.
Supplemental Figure 18: Verification of track analysis. Screen displayed after track analysis, including a video window showing the particles detected and their corresponding tracks. Please click here to download the figure.
Supplemental Figure 19: Calculation of diffusion coefficients. Histograms of the diffusion coefficients calculated (left) and the mean squared displacement (MSD, right). Please click here to download the figure.
Supplemental Figure 20: Calculation of diffusion coefficients including the different fitting modes. Histograms of the diffusion coefficients calculated (left) and the mean squared displacement (MSD, right) for the confined model. Please click here to download the figure.
Supplemental Figure 21: Intensity Profiles. Example of spot intensity along its trajectory (blue line) and background (red line). Please click here to download the figure.
Supplemental Figure 22: Background for each frame. Left: Sample cell image. Middle: Automatically detected cell. Right: Area automatically selected as background. Please click here to download the figure.
Supplemental Video 1: Example of a typical TIRF video microscopy showing the presence of particles with different intensities and types of movements. Please click here to download the video.
Supplemental Material 1: Files containing all the protocol scripts employed in the ad hoc routines employed for the classification of trajectories and analysis of the cluster size. Please click here to download the materials.
The described method is easy to perform even without having any previous experience working with Matlab. However, Matlab routines require extremely accuracy with the nomenclature of the different commands and the localization of the different folders employed by the program. In the tracking analysis routine (step 3), multiple parameters can be modified. The "Setting Gaussian-Mixture Model Fitting" window (step 3.8) controls how U-track will detect single particles on the video. This is done by fitting a Gaussian mixture model as described in3. One of the key parameters for this fitting defines a filter to help identifying local maxima. The success of this operation depends on the image contrast and the noise present in the images. The first parameter (Alpha value for comparison with local background) controls the confidence of a maxima being a real spot, while the second helps to reduce noise during the identification of spots. Important parameters in the "Tracking" step (3.9) are the number of frames to close gaps (that is a track may span over frames in which the particle is not actually seen, but it is seen before these frames and after these frames; in the example, 0), and the number of frames that a track must span in order to be considered as a successful track (in our example, 20; this parameter is related to the camera acquisition speed and the number frames in the track must be such that it allows the calculation of the diffusion coefficient). It is also important to decide whether to merge and split segments (in the example these two possibilities were chosen). Then, the parameters for the Step 1 in Cost functions (frame-to-frame linking) must be set. This function controls how the particles are tracked along the frames. The most important parameters at this point is the selection of the Brownian search radius, which control how far each spot is expected to be in the next frame. In our example we chose 0 and 5 as lower and upper bounds, respectively. Note that these parameters are very specific to the nature of the particles being tracked, and that they may have to be tuned in each specific case. Particularly important are the scaling power in the Brownian and Linear search radii. These scaling powers depend on the kind of movement of the particles (free or confined diffusion). In the example, the values (0.5, 0.01) were chosen for free diffusion. For the specific documentation of these parameters, see3.
In the calculation of the diffusion coefficients command (step 4.4), the upper bound of the diffusion coefficient of immobile particles can be known from previous experiments or by running the calculateDiffusion function on a separate project with immobile particles (purified monomeric fluorescent protein) and seeing the diffusion coefficients reported. This function takes two extra parameters: `outputSuffix´, which is added to the output filenames and by default takes the empty value, and `plotLength´, which by default is 13 and is the number of time lag (seconds) in which the diffusion calculation is performed. The fitting mode 'alpha' implies an adjustment of the MSD to the curve
that is, an offset (MSD0) and a power function of the time lag. The exponent of this power function, , determines whether the movement is confined (0 < α < 0.6), anomalous (0.6 < α < 0.9), free (0.9 < α < 1.1), or directed (α > 1.1). For a review of this kind of analysis the reader is referred to Manzo et al.9
The calculation of the diffusion coefficients4 produces two output figures (see Supplemental Figure 19). The first one shows a histogram of the diffusion coefficients calculated (note that at this point, there is an independent diffusion coefficient for each trajectory). The mean and 95% percentile of these coefficients are shown in the Matlab console. The second plot shows the MSD (red curve) and the number of steps considered to calculate it as a function of the time lag for those trajectories considered to be mobile (those whose diffusion coefficient is above the threshold given in the command line). The average and standard deviation of the mobile trajectories is computed (these are the values for the trajectories in a single cell). In the example, the exponent is 0.59 meaning that the movement is confined. For this curve fitting, the program reports the uncertainty associated to each one of the parameters (shown as the standard deviation of each one) and the goodness of fit (a perfect fit would reach zero). At this point and after checking the value of alpha we may decide to fit the MSD with a different function:
if 0 < α < 0.6 (confined)
if 0.9 < α < 1.1 (free)
if α > 1.1 (directed)
Anomalous trajectories cannot be fitted with a different function. In case of confined particles you may calculate the confinement size (in microns) as in Destainville et al.14
A good indicator of a correct fitting is that the goodness-of-fit should decrease from the alpha fitting to the final fitting (in the example, the goodness fit falls from 0.30474 to 0.15749, Supplemental Figure 19-20). calculateDiffusioncreates two files in the "resultsTrackingPackagetracks" inside the series directory called "diffusionCoefficients.txt" and "diffusionCoefficientsMobile.txt". These files contain three columns: 1) the index of each input trajectory, 2) their corresponding diffusion coefficient, 3) the majoritarian region in the mask where this track belongs to (the regions are obtained from the file "mask.tif" that we generated in Steps 2.7; if this file does not exist, then this column is not present). You may use this file and the analogous files generated for other cells to analyze the distribution of the diffusion coefficient for a set of cells under similar experimental conditions.
In the calculation of the diffusion coefficient for short trajectories (steps 4.7-4.8), the last parameter 'Short' is a suffix added to the output filename so that you may analyze different subsets of trajectories without overwriting the diffusionCoefficients.txt files. The actual name of the output files is "diffusionCoefficients<Suffix>.txt" and "diffusionCoefficientsMobile<Suffix>.txt". If no suffix is given, as in the general analysis performed above, then an empty suffix is assumed. The model parameters (D, v, and MSD0) may differ from those fitted to all trajectories. Most remarkably, the standard deviation of the parameters as well as the Goodness fit normally increase. The reason is that short trajectories are more unstable and a reliable fitting is more difficult.
The analysis of long trajectories (step 4.9) classify them into confined (1), free (2), or directed (3) according to their first moment and its location with respect to the 2.5% and 97.5% percentiles of the first moment of 500 random paths with Brownian motion, the same diffusion coefficient and length as the trajectory being analyzed and simulated by Monte Carlo. If the first moment of the path being analyzed is below the 2.5% of the first moments observed in the simulations, the path being studied is classified as confined. If it is above the 97.5% of the simulated first moments, it is classified as directed; otherwise, the path is classified as Brownian. In the command employed, 113.88e-3 is the pixel size in µm, 0.0015 is the upper bound of the diffusion coefficient of immobile spots measured in µm2/s, and 'Long' is the suffix for the output filename. The first column of this file ("trajectoryClassificationLong.txt") is the trajectory number, the second is its classification (1=confined, 2=free, 3=directed), and the third is the trajectory first moment.
The script for calculating the intensity of each particle (step 5.1) is highly flexible and allows tracking the fluorescence intensities in many different ways (each one well suited for different situations like tracking immobile spots of a control experiment or tracking highly movable spots over a cell with variable fluorescent background). The script will analyze all trajectories along all frames. For each spot at each frame it will measure the intensity of the pixels around the spot (it analyzes a square patch around the spot, by default of a size 3×3 pixels), estimate the spot background and calculate the fluorescence difference between the spot and the background. The percentage of photobleaching particles and the number of fluorescent particles in a single cluster, seen as a single spot, can be estimated in this way.
This automated method (gatherTrajectoryClassificationAndIntensity) produces information on multiple parameters (spot intensity, lateral diffusion, type of motion), that can help to study the relation between spot size and its dynamic at basal conditions, and how different treatments can modify these parameters.
The main limitation of the method is that it requires cell transfection with the fluorescent protein to be tracked. Usually transfection leads to protein overexpression, a fact that hinders single protein tracking. A cell sorting step must be included to select cells expressing a number of receptors that allow detection and tracking of single particles by SPT-TIRF microscopy. This method could also be employed for tracking biomolecules labelled with quantum dots or Fab fragments. The described protocol requires a number of controls that must be previously analyzed in order to ensure that the conclusions driven from the analysis are correct. First of all, it is critical to determine the appropriate expression conditions that allow detection and tracking of single particles. Movies with densities of ~ 4.5 particles/µm2, corresponding to 8,500-22,000 receptors/cell, were used to analyze the spatio-temporal organization of the cell membrane receptor13.
It is important to establish the minimum detectable diffusion coefficient by using purified monomeric AcGFP proteins or fixed cells. In both cases it is assumed that there is no diffusion and therefore we estimate that the diffusion values of particles analyzed in these conditions correspond to immobilized particles and used to discriminate between mobile and immobile trajectories13.
The analysis we have presented here is a general trajectory analysis tool, that can be applied to the analysis of diffusion of receptors by superresolution, to the analysis of diffraction limited images and to the analysis of cellular movement by standard microscopy. The main advantage of this method with others previously described, such as analysis of number and brightness11, is that it allows the evaluation of the mean intensity values of each individual spot along its trajectory, taking into account the time of survival of the AcGFP monomeric protein before photobleaching. The survival time of a molecule before photobleaching will strongly depend on the excitation conditions.
The authors have nothing to disclose.
We are thankful to Carlo Manzo and Maria García Parajo for their help and source code of the diffusion coefficient analysis. This work was supported in part by grants from the Spanish Ministry of Science, Innovation and Universities (SAF 2017-82940-R) and the RETICS Program of the Instituto de salud Carlos III (RD12/0009/009 and RD16/0012/0006; RIER). LMM and JV are supported by the COMFUTURO program of the Fundación General CSIC.
Human Jurkat cells | ATCC | CRL-10915 | Human T cell line. Any other cell type can be analyzed with this software |
pAcGFPm-N1 (PT3719-5)DNA3GFP | Clontech | 632469 | Different fluorescent proteins can be followed and analyzed with this routine |
Gene Pulse X Cell electroporator | BioRad | We use 280V, 975 mF, for Jurkat cells. Use the transfection method best working in your hands. | |
Cytomics FC 500 flow cytometer | Beckman Coulter | ||
MoFlo Astrios Cell Sorter | Beckman Coulter | Depending on the level of transfection, cell sorting may not be required. You can also employ cells with stable expression of adequate levels of the receptor of interest. | |
Dako Qifikit | DakoCytomation | K0078 | Used for quantification the number of receptors in the cell surface. |
Glass bottom microwell dishes | MatTek corporation | P35G-1.5-10-C | |
Human Fibronectin from plasma | Sigma-Aldrich | F0895 | |
Recombinant human CXCL12 | PeproTech | 300928A | |
Inverted Leica AM TIRF | Leica | ||
EM-CCD camera | Andor | DU 885-CSO-#10-VP | |
MATLAB | The MathWorks, Natick, MA | ||
U-Track2 software | Danuser Laboratory | ||
ImageJ | NIH | https://imagej.nih.gov/ij/ | |
FiJi | FiJI | https://imagej.net/Fiji) | |
u-Track2 software | Matlab tool. For installing, download .zip file from the web page (http://lccb.hms.harvard.edu/software.html) and uncompress the file in a directory of your choice | ||
GraphPad Prism | GraphPad software |