pubs.acs.org/NanoLett
Three-Dimensional Tracking of Single Fluorescent Particles with Submillisecond Temporal Resolution Manuel F. Juette†,‡,§ and Joerg Bewersdorf*,† †
Department of Cell Biology, Yale University School of Medicine, New Haven, Connecticut 06520 United States, Department of Biophysical Chemistry, University of Heidelberg, Heidelberg, Germany, and § Department of New Materials and Biosystems, Max Planck Institute for Metals Research, Stuttgart, Germany
‡
ABSTRACT Observing dynamics at the nanoscale requires submillisecond time resolution. Notably, in studying biological systems, three-dimensional (3D) trajectories of fluorescently labeled objects such as viruses or transport vesicles often need to be acquired with high temporal resolution. Here, we present a novel instrument that combines scanning-free multiplane detection at 3.2 kHz frame rate and single photon sensitivity with optimized beam-steering capabilities. This setup enables ultrafast 3D localization with submillisecond time resolution and nanometer localization precision. We demonstrate 3D tracking of single fluorescent particles at speeds of up to 150 nm/ms over several seconds and large volumes. By focused excitation of only the particle of interest, while avoiding confocal pinholes, the system realizes maximum detection efficiency at minimal laser irradiation. These features, combined with the avoidance of stage movement, provide high live-sample compatibility for future biomedical applications. KEYWORDS Single particle tracking, fluorescence microscopy, 3D, trajectory, nanoscopy, live-cell imaging
T
rant photodiodes)11-13 most applications have focused on the utilization of fluorescence due to advantages such as small, genetically encoded labels and the possibility of multicolor experiments. Predominantly, widefield imaging, often in combination with total internal reflection, has been used to simultaneously track multiple fluorescent particles such as proteins,8,14 lipids,5,11 vesicles,15 polynucleotides,16 or viruses17 inside cells. This technology has recently been expanded to 3D tracking using multiplane detection18-20 or astigmatism,21,22 approaches that have also been demonstrated successfully in photoactivation localization nanoscopy.23,24 These techniques, however, suffer from the speed limitation imposed by reading out full frames, or at least large areas, of the charge-coupled device (CCD) camera chip at every time step. Temporal resolution has therefore typically been limited to tens of milliseconds with these techniques. Additionally, as axial scanning is not feasible when tracking multiple particles in parallel, their depth detection range is limited to typically 1-2 µm,25 and particle trajectories are terminated when a particle moves outside this range. Another class of techniques has solved these issues by limiting tracking to a single particle at any time. Two approaches have been realized so far.26 In the first case, a laser focus is scanned in a circular or more complex trajectory centered on the particle.27-29 Changes in the particle position result in a characteristic signal modulation, where the amplitude and phase indicate the distance and direction of the translation with respect to the center position of the scanning trajectory. As a consequence, the laser focus position is corrected for the detected motion. In an alterna-
he key to our understanding of biology lies in the nanoscale regime. Electron microscopy has revealed the basic architecture of the cell’s organelles, leading to a wealth of biological discoveries. More recently, modern fluorescence imaging techniques have reached the sub-100 nm resolution scale, often in three dimensions.1,2 These new methods make use of fluorescent labels and can consequently provide insights into cellular structures in vivo. Multiple colors can be used to specifically label selected targets of interest. The challenge now is speed; while a cell’s morphology, looking at the micrometer-level, typically changes on the second to minute time scale, details on the nanoscale rearrange within milliseconds. Brownian motion is exemplary of this phenomenon: a 10 µm diameter object (about the size of a B cell in suspension) diffuses in water at 37 °C on average by about 200 nm per second, a movement barely resolvable in low-magnification microscopy. On the nanoscale, a 40 nm diameter sphere (e.g., a synaptic vesicle) is displaced under the same conditions on average by nearly 100 nm per millisecond, a distance 2.5-fold its diameter on a time scale hardly accessible in standard microscopy. To study subcellular dynamics, particle tracking microscopes have therefore been developed over the last decades.3-10 While techniques using light scattering and transmission have been demonstrated successfully (in some cases at impressive speeds taking advantage of fast complementary metal oxide semiconductor (CMOS) cameras or quad* To whom correspondence should be addressed. E-mail: joerg.bewersdorf@ yale.edu. Received for review: 08/13/2010 Published on Web: 10/12/2010 © 2010 American Chemical Society
4657
DOI: 10.1021/nl1028792 | Nano Lett. 2010, 10, 4657–4663
tive realization, the beam position is kept stationary on the particle, but fluorescence is split up and detected by several point detectors such as avalanche photodiodes (APDs).30 These are arranged in one or two detection planes to provide the necessary signal for 2D or 3D localization. Signal differences between the different detectors indicate a translation of the observed particle. The sample stage is shifted accordingly to compensate for this movement. Alternatively, the laser focus can be parked at a fixed position in the sample, eliminating the feedback mechanism at the expense of a decreased tracking range.31 While these single particle tracking techniques feature, in most cases, a speed and depth range advantage over methods based on conventional widefield imaging, they have given up the concept of taking a conventional image of the observed particle. Especially in a cellular setting, this poses the risk of misinterpreting the collected abstract signals. Two particles approaching each other or high or inhomogeneous background, for example, can easily escape detection, resulting in artifacts in particle localization. Here, we present a new particle-tracking microscope that addresses the challenges described above by a novel concept combining aspects of CCD camera-based multiplane imaging and feedback-driven single particle tracking in a hybrid instrument. In short, the fluorescence of a particle illuminated by a focused laser beam is detected in two detection planes on an EM-CCD camera. Real-time analysis of the particle image allows immediate detection of particle motion. This information is used to actively steer the laser beam to follow the particle. Because of fluorescence descanning similar to a beam-scanning confocal microscope, the particle image stays in the same small detection areas on the camera chip. This allows reading out only three to five lines of the camera chip to image the particle at all times, which can be achieved at 3.2 kHz frame rate. Combined with fast piezo actuators for beam steering, this enables submillisecond 3D particle tracking at particle speeds of well over 100 nm/ms. The schematic setup of the instrument is shown in Figure 1. A 488 nm diode-pumped solid state laser (Sapphire 488 LP50, Coherent Inc.), whose power is controlled by an acousto-optical tunable filter (AOTFnC-400.650-TN, AA OptoElectronic), is spatially filtered by a pinhole for higher beam quality, collimated, and directed into a commercial inverted microscope stand (Leica DM-IRBE) via a dichroic beam splitter (FF503/670-Di02, Semrock) and a piezo-driven tilt mirror (S-330.2SL, Physik Instrumente). An f ) 200 mm lens and the tube lens of the microscope stand image the piezo mirror into the back aperture of the objective (water immersion, Plan Apo 63×/1.2NA, Leica Microsystems). Tilting the piezo mirror results in a translation of the laser focus in the sample by up to approximately (5 µm in the x and y directions. The objective is mounted on a fast piezo actuator (PIFOC P-726, Physik Instrumente) for axial tracking of up to 100 µm. Fluorescence is detected by the same objective. The fluorescence passes the piezo mirror to correct for lateral © 2010 American Chemical Society
FIGURE 1. Schematic of optical setup. A 488 nm diode-pumped solid state laser is coupled into an inverted microscope stand and focused into the sample by a 63×/1.2NA water immersion objective. An acousto-optical tunable filter (AOTF) controls the laser power. High beam quality is achieved by a pinhole. The laser focus can be moved in the focal plane of the sample by tilting a piezo mirror in the beam path. Axial translation of the laser focus in the sample is achieved by the piezo-actuated objective holder. For test and calibration purposes, the sample is mounted optionally on a 3-axis piezo stage. Fluorescence light emitted from a particle in the laser focus is collected by the objective, descanned by the piezo mirror, passes through a dichroic mirror to separate it from the laser illumination, and is imaged by a fast EM-CCD camera chip after being split by a 50:50 beam splitter and passing through a bandpass filter. Because of the longer path of the reflected beam, two axially separated planes of the sample can be imaged synchronously.
movement, which results in a nearly stationary detection beam. Fluorescence light is then transmitted through the dichroic mirror and imaged onto the back-thinned EM-CCD camera (Ixon DU-860, 128 × 128 pixels, Andor Technology) via a 50:50 neutral beam splitter cube. The light transmitted through the beam splitter cube is detected on one side of the chip; after being reflected by a second mirror, the reflected light is detected in the same pixel lines but on the other side of the chip to avoid overlap between the two images. The f ) 500 mm lens in front of the camera results in a ∼160-fold magnification, which, due to the path difference of the reflected and transmitted beams, leads to an axial difference of ∼750 nm between the detection planes, a distance that had previously been identified as well suited for 3D localization in biplane geometry.23,25 The pixel size corresponds to 150 nm in the sample. Five lines of pixels therefore represent a 750 nm wide area, providing a good overview of the immediate vicinity of the particle. Five lines of the camera are read out at up to 3.2 kHz frame rate. Two 5 × 5 pixel large regions of interest (ROIs) are extracted and the signal is analyzed in real-time by a computer to determine deviations of the particle position from the center of the laser focus (see Supporting Information). On the basis of this analysis, correction terms are fed back to the piezo systems to correct for the detected particle movement. This cycle is repeated over the length of the measurement. Particle trajectories are assembled from the 4658
DOI: 10.1021/nl1028792 | Nano Lett. 2010, 10, 4657-–4663
recorded piezo positions determined by the integrated strain gauge resistors (piezo mirror) and capacitive sensor (objective piezo) and corrected for deviations from the center positions as detected in the camera frames in a postprocessing step. The read out ROIs and the piezo sensor information are stored for every time point, which optionally allows for a more detailed analysis following the measurement. To characterize response time and precision of the instrument, a sample containing 200 nm diameter fluorescent beads (F-8774, FluoSpheres 505/515, Invitrogen) attached to the coverslip was mounted on an additional fast 3-axis piezo stage (P-733.3DD, Physik Instrumente) on top of the microscope stage. This allowed introducing well-defined particle trajectories independent of the microscope. The piezo sample stage was operated in closed loop (specified to subnanometer precision) and step functions were programmed into the stage controller. These movements were detected by the instrument observing one of the fluorescent beads on the coverslip. Steps in the x, y, and z directions were monitored separately. Figure 2 shows typical results of three of these experiments. The camera was read out at 3.2 kHz. Every 60 ms the stage was shifted in one or the other direction by 200 nm. The blue symbols and line show the position of the sample stage as measured by the internal capacitive sensors of the stage. From the recorded camera frames, a deviation from the center position (indicated in gray) can be calculated using the algorithm described in the Supporting Information. The sudden movement of the stage results in a sharp response in the particle position in the camera image immediately following the step, as seen by the peak of the gray curve. This triggers a change in the positions of the mirror and objective piezo actuators as displayed in red. The measured response of these elements is delayed by approximately 2 ms for the piezo mirror (Figure 2a,b) and approximately 5 ms for the objective actuator (Figure 2c). In a simple postprocessing step following data recording, the position information derived from the gray curve can be combined with the measured mirror and objective positions (red) to get an accurate representation of the particle movement as shown in orange. Taking the standard deviation of 150 localizations over time ranges where the stage was not moved yields estimates for the obtainable localization precision. For the displayed case, localization precisions of σx ≈ 7 nm, σy ≈ 8 nm laterally and σz ≈ 27 nm in the axial direction were obtained, consistent with the shape of the point-spread function (PSF), which is ∼3-fold longer axially than laterally.25 As expected, the localization precision scaled with N-1/2, where N is the number of detected photons per frame (see Supporting Information). To study the system capabilities for more complex trajectories, we chose to track well-defined trajectories that easily allow for detection of systematic errors. The sample stage was programmed to move in circles of different diameters in the x-y, the x-z, and the y-z plane. Single © 2010 American Chemical Society
FIGURE 2. Step responses in 3D particle tracking. (a-c) The response of the instrument to steps in the x, y, and z direction, respectively. A fluorescent bead attached to a coverslip and mounted on a 3-axis piezo stage is moved in one direction at a time by 200 nm steps as shown by the stage-sensor signal in blue. The particle is observed in biplane mode by the instrument and its position is analyzed. The determined offset from the center position is shown in gray. This signal is fed back to the piezo-driven mirror and objective actuators to track the shifted bead. The sensor position of the corresponding axis, shown in red, is delayed by ∼2 ms (x, y direction) to 5 ms (z direction) relative to the initial step. Combining the mirror and objective position with the detected offset data, however, allows to correct for this delay and to obtain a precise and nondelayed measure of the particle position in 3D as displayed in orange.
beads sticking to the coverslip were tracked with the instrument at 3.2 kHz temporal resolution. Trajectories of single rotations in the x-y and x-z planes are displayed in Figure 3a,b, respectively. Experiments with y-z circles (not shown) yielded results similar to the x-z case. The tangential velocity increased from 5 nm/ms up to 150 nm/ms with the diameter of the circles as indicated in the figure caption. Nearly perfect circular trajectories are obtained in all cases. In z-direction, tracking precision is about 3-fold worse than in lateral direction, however it is still well below the 100 nm radius of the smallest trajectory, which is consistent with 4659
DOI: 10.1021/nl1028792 | Nano Lett. 2010, 10, 4657-–4663
FIGURE 3. Tracking directed motion in 3D. Two hundred nanometer fluorescent beads were attached to a coverslip that was moved in 2D circles (a,b) or a helical pattern (c) by a three-axis piezo stage. The measured trajectories are displayed for (a) x-y circles and (b) x-z circles. The circular shapes are nearly perfectly reproduced for 200 nm diameter circles at 5 nm/ms particle speed (light gray), 500 nm diameter at 10 nm/ms (dark gray), 1.5 µm diameter at 25 nm/ms (green), 3.0 µm diameter at 50 nm/ms (orange), 5.0 µm diameter at 100 nm/ms (red), and 7.5 µm diameter at 150 nm/ms (blue). The gray circles located in the bottom right of (a,b) show magnified views of the smallest trajectories. Panel c shows the tracking of a more complex 3D trajectory, an axially oriented spiral of 1 µm diameter and ∼2.5 µm height. The 3D representation of the spiral is displayed in blue and its projections in gray.
observations in the step response test. The highest tested speed of 150 nm/ms well exceeds the velocities of most known processes of directed motion inside cells. This demonstrates the system’s suitability for a wide range of kinetic studies on the single particle level. In a further example, the three-axis stage was moved in a helical pattern as displayed in Figure 3c. This nicely validates the system’s tracking capabilities of complex 3D trajectories. Next to directed motion, which we tested by the experiments described above, cell biology is dominated by diffusive processes. To further test the feasibility of our method in these applications, we studied the diffusion of 200 nm fluorescent beads in glycerol-water mixtures representing media of different, but well-characterized viscosities. Beads were tracked in solution for ∼3-12 s at 3.2 kHz frame rate, yielding ∼10 000-40 000 time points for each measurement. A laser power of ∼1-5 µW at the sample was chosen, resulting in a bead signal of ∼3000-4000 photons on average per frame. Photobleaching of the beads was negligible (typically
