Direct observation of mobility state transitions in RNA trajectories by sensitive single molecule feedback tracking

Observation and tracking of fluorescently labeled molecules and particles in living cells reveals detailed information about intracellular processes on the molecular level. Whereas light microscopic particle observation is usually limited to two-dimensional projections of short trajectory segments, we report here image-based real-time three-dimensional single particle tracking in an active feedback loop with single molecule sensitivity. We tracked particles carrying only 1–3 fluorophores deep inside living tissue with high spatio-temporal resolution. Using this approach, we succeeded to acquire trajectories containing several hundred localizations. We present statistical methods to find significant deviations from random Brownian motion in such trajectories. The analysis allowed us to directly observe transitions in the mobility of ribosomal (r)RNA and Balbiani ring (BR) messenger (m)RNA particles in living Chironomus tentans salivary gland cell nuclei. We found that BR mRNA particles displayed phases of reduced mobility, while rRNA particles showed distinct binding events in and near nucleoli.

localization for SNR = 6 (black symbols), SNR=8 (white symbols), SNR=15 (grey symbols). SNR conditions simulated by adding appropriate Poisson noise to high SNR (~400) experimental data. (a) Normalized covariance values for PSF templates from below (circles) and above (squares) the focal plane. (b) Axial localization metric according to (eq. 1). The metric is obviously not dependent on the SNR. (c) PSF width along x (squares) and y (circles) as determined by fitting of a 2D Gaussian peak. (d) Δw = w xw y yields a linear calibration metric, which is again not dependent on SNR. (e) Second moment of the intensity distribution along x (squares) and y (circles). f) Δw still yields a linear metric for the axial localization, but the slope is dependent on the SNR.  Table 3). Towards the nuclear envelope, the fraction of slower moving particles increased.    (Fig. 4), 1-5 were found in a single trajectory close to the nuclear envelope (Fig. 5).

Supplementary Note 1: Real-time tracking DLL description
Real-time image analysis and tracking code was implemented in terms of a DLL written in C++. The DLL was called immediately after an image frame had been transferred to PC memory. Parameters passed to the DLL are listed in Supplementary Table 1. Pre-allocated arrays were used to store tracking parameters, PSF templates and tracking results during the experiments. The general outline of the DLL is presented in Supplementary Fig. 4 and the source code as well as accompanying files and documentation is provided in Supplementary Material 1. During the first call to the DLL in an experiment, tracking parameters and PSF templates were loaded from ASCII files and stored in the previously mentioned arrays.
Particle tracking consisted of six steps: 1. Loading parameters from an ASCII file or PC memory.
2. Candidate identification by detection of an intensity peak in a subimage around either initial, user-supplied coordinates or a previous particle localization. The size of the subimage was (i) small while following a particle to avoid trajectory confusion or (ii) large while searching for the first localization of a trajectory to increase real-time tracking duty cycle.
3. Centroid calculation for lateral localization and either second moment or normalized covariance calculation for determination of axial position relative to focal plane.
4. Particle verification based on peak intensity and either peak width or covariance values.
5. Storing tracking results and parameters in PC memory or, at the end of the experiment, in an ASCII file.
6. Return axial coordinate of the particle being tracked in terms of a piezo stage voltage.
To achieve robust tracking especially at low SNR, the algorithm may not be too sensitive on the one hand to avoid interpreting background noise fluctuations as signals resulting from fluorescence emission. On the other hand, occasional large axial particle displacements as well as fluctuations in the photon detection rate must be tolerated. Best results were obtained by leaving particle detection thresholds moderately high (e.g. minimum covariance value 0.35 -0.45) and at the same time allowing for gaps in trajectories during real-time tracking by waiting for a particle to reappear within n wait = 2-5 frames after the algorithm had lost it. This enabled recovering long trajectories using more sensitive particle detection parameters during post-processing without generating excessive amounts of false positive localizations during real-time tracking. If a trajectory was not continued after n wait frames, the stage was set to proceed at its initial "home" position. At the end of an experiment, all arrays and tracking parameters were saved to ASCII files for further evaluation.
Commented source code for the tracking DLL, auxiliary files for setting parameters and providing template information as well as a PDF document with instructions on how to use the source code to set up a tracking experiment are available for download at http://www.chemie.uni-bonn.de/pctc/kubitscheck/downloads .

Supplementary Videos
Videos presented in the supplementary material were assembled using MATLAB and ImageJ.
Two types of videos were created: 1. To present tracking data, raw image data were overlaid with trajectory information obtained from the post-processing analyis (colored lines connecting localizations), a timestamp indicating the time since the start of the experiment in milliseconds and an indicator of the current stage position. A square plotted on top of the image data indicates the subimage, in which the particle was searched for during tracking.
2. For a schematic representation of the tracking data, 3D plots of the trajectories were overlaid with surface renderings of reference structures (e.g., nuclear envelope or reconstructed vesicle surface). For the representation of time series, the current particle localization is indicated by a red sphere. localizations. Data were acquired and displayed at 63 Hz frame rate. Image field (40.96 µm)².
mRNA particle labeled with an oligonucleotide carrying a single Atto647 dye in the nucleoplasm (red). NTF2-AF546 was co-injected to stain the nuclear envelope (green). Single trajectory of 3.6 s duration tracked at 62.5 Hz displayed at 20 Hz. Image field (30.7 µm)².
mRNA particle tracking in the nucleoplasm. 15.6 s trajectory raw data and tracking data overlay. 3D Plot of mRNA particle trajectory. Image field (30.72 µm)².
Video 5: mRNA tracking nucleoplasm trajectory reconstruction 3D plot of 10 long mRNA particle trajectories and surface renderings of a polytene chromosome (cyan) and the nuclear envelope (green).
rRNA particle tracking at the nucleolus. Raw data overlayed with trajectory and reference information (nuclear envelope, green) and 3D rendering of reference structures (nuclear envelope, green; nucleolus, cyan). Nucleolus reconstructed from negative contrast in NTF2-AF546 channel (green), compare upper panel. Image field (30.72 µm)².