We introduce a parallelized STED microscope featuring m = 4 pairs of scanning excitation and STED beams, providing m-fold increased imaging speed of a given sample area, while maintaining basically all of the advantages of single beam scanning. Requiring only a single laser source and fiber input, the setup is inherently aligned both spatially and temporally. Given enough laser power, the design is readily scalable to higher degrees of parallelization m.
©2011 Optical Society of America
Since the beginning of the 20th century, far-field optical microscopy has been essentially limited to >λ/2 > 200 nm in resolution, with λ denoting the wavelength of light. While electron and scanning probe microscopy have provided much finer details since then, they lack the possibility of non-invasively imaging (living) cells and tissue as well as other transparent materials in three dimensions (3D). In the 1990s it was discovered that the diffraction resolution barrier in far-field fluorescence microscopy can be effectively overcome by employing basic transitions between the states of the fluorophores in use . Specifically, transitions ensuring that fluorescent molecules remain transiently non-fluorescent allowed the separation of nearby features by causing them to emit sequentially. This simple strategy forms the basis of Stimulated Emission Depletion (STED) microscopy [1, 2] as well as of more recent far-field fluorescence nanoscopy techniques [3–11]. While they maintain most of the advantages of far-field imaging, these concepts procure resolution down to the nanoscale.
To separate features by sequential detection, STED microscopy forces a part of the fluorophores to dwell almost exclusively in the ground state . The nearly permanent occupation of the ground state is enforced by the so-called STED beam, whose wavelength and intensity ISTED are tuned such that it is capable of stimulating excited fluorophores instantaneously back down to the ground state. The focal pattern of the STED beam features (at least) a (single) point  or line  of vanishing intensity at which the occupation of the fluorescent state is inherently allowed. By precisely positioning the zero point(s) or line(s) in the sample and adjusting ISTED one can define the coordinate in space where the fluorophores can still assume the fluorescent state, i.e. are not 'switched off' by the enforced occupation of the ground state [9, 10].
The most common STED microscopy configuration is to overlay a diffraction-limited focused excitation beam with a doughnut-shaped STED beam. Although both beams are limited by diffraction, resolution beyond the diffraction limit is possible, because above a certain threshold intensity IS, molecules will almost exclusively remain non-fluorescent; the time they spend in the fluorescent state becomes negligible. The resolution can be increased to subdiffraction dimensions by making sure that the doughnut intensity surpasses the threshold IS at subdiffraction distances to the zero- intensity point, so that virtually all fluorophores are kept non-fluorescent but those residing at the closest proximity to the doughnut zero. The resolution increases in proportion with the square-root of ISTED,m/IS, with ISTED,m denoting the intensity at the (doughnut) maximum engulfing the zero [10, 14]. In fact, any reversible transition between two states can be used to create subdiffraction resolution images , which is why the STED principle has been generalized to any reversible saturable optical linear (fluorescence) transitions (RESOLFT) . Hence, any reference to STED in this paper will equally apply to the more general RESOLFT concept.
It should also be noted that the principles of STED or RESOLFT do not require the imaging of the fluorescence onto a point detector, which is the hallmark of confocal microscopy. Therefore, STED and RESOLFT superresolution need not be implemented as point-scanning or confocal systems. In essence any pattern that uses an intensity minimum (zero) to confine a signaling or a non-signaling state to subdiffraction dimensions can be used to create superresolution images. However, because it facilitates the suppression of background and elegantly provides three-dimensional (3D) imaging, confocal arrangements have been preferred in most STED implementations. Furthermore, since the diffraction resolution problem applies only to features that are closer than λ/2, it is conceptually straightforward to create patterns of multiple local zero-points or zero-lines separated by distances > λ/2. Such patterns automatically parallelize STED or RESOLFT nanoscopy. Nonetheless, to date most STED microscopes are still based on single point-scanning, even though the image field is far greater than λ/2. This is a drawback for imaging large field of views, since an n-fold higher resolution in the x- and the y -direction in the focal plane requires an n × n-fold denser pixelation and hence a recording time longer by this factor.
Effective parallelization has been accomplished by spinning disc arrangements with multiple pinholes, which have permitted to increase the recording speeds in confocal microscopy . Other methods have used cascaded beam splitters to create 64 parallelized beams (LaVision BioTec, Germany). Another way of parallelizing the scanning process is to apply arrays of zero-lines and valleys, which is usually called structured illumination . Within the super-resolution techniques, saturated structured illumination SPEM/SSIM do provide parallelization of subdiffraction imaging, but the reported versions are laid out such that they require data post-processing [4, 18]. Moreover, they are not confocally arranged, meaning that 3D imaging is more difficult to implement. The stochastic super-resolution methods [6–8], which ensure that molecules capable of fluorescent emission are further away than λ/2, are inherently parallelized. Nevertheless, the stochastic nature of single molecule detection and the fact that at most a single emitting molecule is allowed within a λ/2 spatial range, compromises their imaging speed. Providing images based on the nearly simultaneous emission of an arbitrary number of fluorophores from predetermined positions [9, 10], STED and RESOLFT are advantaged with regard to time resolution. Yet, while theoretical studies have been given  and initial steps with RESOLFT parallelized with arrays of line-shaped zeros have been shown , parallelized imaging with these concepts still requires development. In particular, the widely applied STED microscopy has so far not been parallelized.
STED microscopy can be parallelized using arrays of multiple doughnuts  or with arrays of lines [9, 14] ('structured illumination'). The latter is presently less attractive simply because of the large power requirement for the STED beam, which stems from the fact that the maximum intensity ISTED,m has to exceed IS. A challenge in parallelizing STED by implementing multiple doughnuts is to retain the zero intensity at the doughnut center while guaranteeing the alignment with the excitation spots. We now demonstrate a parallelized STED microscopy using m=4 doughnuts realized by beam multiplication, which is scalable to a higher number m of spots. Additional properties are the inherent alignment of both excitation and STED beams by launching them from a single fiber. Both parallelization and the creation of multiple doughnuts are achieved just by polarization optics . The fluorescence originating from the individual excitation spots is spatially separated and collected by m=4 single photon-counting modules. The photons collected in these independent detection channels stem from different positions in the focal plane as determined by the spatial arrangement of the doughnut pattern. Different scanning schemes are demonstrated that combine the individual channels in order to reduce the required recording time by a factor of four. Furthermore a new type of beam-scanning device is presented.
2. Design of the parallelized setup
Parallelization of STED by focused beam multiplication requires a device generating multiple beam pairs (excitation/STED) and the co-alignment of these beams. Furthermore, their wavefronts need to be controlled such that the generated focal intensity spots are nearly diffraction-limited and doughnut-shaped with central zeros for the STED light. Last but not least, in pulsed STED setups such as the one implemented here, one should be able to tune individual laser intensities and the relative delay between the laser sources.
Optical devices, such as Wollaston prisms, can separate light beams into two orthogonally polarized sub-beams that exit the device at a specific angle determined by the devices’ geometry. Both laser wavelengths required for STED can originate from a single fiber to cross the Wollaston prism while remaining perfectly aligned. For large scanning angles or differences in wavelength between STED and excitation wavelength one should account for birefringence dispersion by using achromatic Wollaston prisms . The axial alignment of the beams depends of the micro objectives used. For APO type objectives used in the present system they are usually shifted by 100-200 nm which is well within the 400-500 nm axial resolution and can hence be neglected. Stacking these prisms doubles the number of beams for each prism while maintaining their alignment. The number of beams is therefore scalable depending on the available laser power. Moreover the intensities can be evenly and accurately distributed among the individual beams by rotating the input polarization and individual prisms. We therefore decided to use this type of beamsplitting device in order to minimize losses of the limited laser power and construct a stable parallelized STED setup.
Two approaches [21, 23] have been reported to simplify STED nanoscopy by making excitation and STED beams pass through a common beam-shaping device. Both configurations leave the wavefront of the excitation wavelength largely unaltered and cause the STED beam to interfere destructively at the focal point, thereby creating the desired doughnut shape. Consisting of a chromatic segmented wave plate (SWP) that alters the polarization of individual beam segments in the aperture, the (easySTED) type proposed by Reuss et al. , is polarization-independent with respect to keeping the central intensity minimum of the doughnut as low as possible. Non-circular components will basically lead to asymmetry effects in connection with the random orientation of the molecules. Such effects can be avoided by providing a uniform distribution of polarization directions, e.g. circular polarization. The handedness of the circular polarization is unimportant. In addition, they do not need to be perfectly circular because non-circular components do not fill up the central intensity minimum of the SWP induced doughnut. For all these reasons, we preferred the SWP over the helical vortex phase plate which has been commonly used so far .
The polarization-independence of the beam-shaping device is also important for implementing a single-fiber as a common source for the STED and the excitation light using Wollaston prisms, because a stack of N Wollaston prisms generates 2N beams where 2N-1 beams have left-handed and the other 2N-1 beams right-handed circular polarization after passing a quarter wave plate. In comparison the more commonly used vortex phase plate only works with one handedness. In order to produce the desired auto-aligned spots of light, the SWP is simply placed centrally close to a plane that is optically conjugate to the back focal plane. It can be placed right after the objective lens, because fluorescence can pass without major alterations, which in turn makes the preceding pupil plane accessible for the beamsplitting elements. More versatile optical devices such as spatial light modulators (SLM) obviously exist, that could combine beamsplitting and shaping tasks, but these devices would waste a large part of the available laser intensity for controlling the wavefront amplitude and phase. Moreover, SLMs would not allow for simultaneous and individual control of the different wavelengths in an easySTED configuration.
Scanning was accomplished with a home-built beamscanner, here referred to as Quadscanner, consisting of four galvanometric mirrors oriented in a configuration that essentially creates a parallel displacement of the focused beams in the conjugated image plane as disclosed in the published patent application WO 2010/069987. This in turn displaces the excitation pattern in the sample by an amount set by the magnification factor objective/tube lens combination. The Quadscanner manipulates the beams close to their beam waist (as opposed to regular beamscanning which rotates the beams in a plane conjugate to the pupil plane) thereby inducing fewer aberrations by the limited flatness of light-weight galvanometric mirrors. The two degrees of freedom per scan axis offer a flexible way to control both the lateral position as well as the plane of rotation of the beamscanning. This ensures a standing beam in the SWP plane and thereby guarantees that the beam remains in the center of the SWP. The fluorescence has to be separated from the excitation beam path before passing back through the Wollaston prism to avoid splitting up the fluorescence, which would reduce the signal and lead to crosstalk between the detection channels. For this reason and to make sure that the beams rotate in the SWP plane under the presence of the Wollaston prisms, these are inserted near the conjugated backfocal plane (cBFP), preceding the scan lens and the bandpass filter (Fig. 1 ). Compared to conventional commercial beam scanning devices, which operate in the conjugated backfocal plane, the Quadscanner reduces the required number of optical images, thereby simplifying the setup and increasing its stability.
Each of the excitation foci overlaid with a doughnut-shaped STED PSF leads to independent fluorescent sources at well-defined positions in the focal plane. The signal from each of these channels must be isolated by focusing the fluorescence onto point detectors, after being separated using a beam separating element, or onto a CCD Camera. We decided to use avalanche photodiode (APD) point detection which enables faster recording times. For larger degrees of parallelization, CCD cameras or APD arrays could represent an easier solution. Different imaging approaches can now be taken regarding scanning and recombination of the different channels to obtain faster acquisition due to parallelization. The most straightforward approach is to scan four non-overlapping areas and to recombine the individual frames by stitching them together in a mosaic-like pattern. For larger field-of-views (FOV) one has to create an array of scanning positions in order to avoid multiple scanning of the same positions. A second approach consists of scanning four large area images simultaneously at a four-times shorter pixel dwell time and adding the overlapping regions of the full-scale frames. Besides reducing the recording time, the advantage of such an approach could be a bleaching-reducing effect by reducing deposited energy dose imposed by each of the individual beams and by allowing the relaxation of the molecules from long-lived dark states back into the ground state . Depending on the FOV however, the non-overlapping region can represent a major proportion, which would gradually reduce the useable area. Still another option would be to use a comb like scanning pattern using different prism rotations to produce a line of spots which is not further pursued here.
We constructed a four channel multispot STED setup using Wollaston prisms for multispot generation, a chromatic segmented waveplate (SWP) to shape the focal spots and four single photon detection units for detection (Perkin-Elmer, USA). To further simplify our setup we used a custom laser system (SC450-20-2, Fianium, United Kingdom) with a single laser pump operating at a repetition rate of 18.45 MHz and generating two unpolarized outputs: a supercontinuum spectrum ranging from 520 nm to 720 nm and a high power output used for STED at 745±5 nm. The excitation band (Z633/10 X, Chroma, USA) is selected from the supercontinuum spectrum and recombined with the STED line using a dichroic mirror (SP750, AHF, Germany), after an optical delay maximizing STED. Once coupled into a single fiber, excitation and STED laser beams no longer required any spatial or temporal alignment. The collimated unpolarized beams passed two Wollaston prisms (Jenoptik, Germany) rotated by 45° with respect to each other to generate four uniform beams (see Fig. 1. inset) at an angle of 20 arcminutes, producing a rhombus-like pattern in the sample with a separation of 5.75 μm at an angle of 45°. The offset between the wavelengths due to the prisms’ birefringence dispersion of ~1% can be neglected in the present system. The beams are reflected by a bandpass filter (z625-745 rpc, Custom-Made, AHF), also acting as fluorescence filter in the detection path, before being focused into the homebuilt beamscanner (Quadscanner) consisting of four galvanometric mirrors (Cambridge Technology, USA) placed next to a commercial microscope (DMI 3000B, Leica Microsystems, Germany). Used as beamshaping device, the SWP (B. Halle, Germany) consists of four segments of crystal quartz cut from a single block and assembled with the fast axes oriented as described elsewhere . The thickness is adapted such that a birefringent retardation of 2.5 λ is achieved for 750 nm and 3λ for 635 nm between the slow and fast axis of the crystal. The SWP was placed just before the 1.46 numerical aperture 100x oil objective lens (Leica Microsystems, Germany). The sample was clamped directly to the objective lens by a home-build sample holder to reduce thermal drifts. Excitation was performed with a total time-averaged optical power of 5 μW at the back aperture of the objective lens. The time-average power of the STED beam at the back aperture of the objective lens was 70 mW, corresponding to a pulse energy of 3.8 nJ. This energy, which was distributed equally among the four spots considering transmission losses, yields a time-averaged power density of 4.1 MW/cm2 per STED doughnut in the focal region. This in turn corresponds to an average pulse intensity of 1.1 GW/cm2 of the 200 ps STED pulse in the focal region. For separating the fluorescence in the detection path we used a custom-made silver-coated pyramid to separate the four channels in the image plane and four lenses to focus the fluorescence onto four single photon counting modules (SPCM-AQR-13/14, Perkin-Elmer, USA). Scanning as well as data processing of the collected signal was controlled using a FPGA board (PCI-7833R, National Instruments, USA) and self-built scanning software (LabVIEW, National Instruments, USA). Additional shortpass (SP750, Semrock, USA) and bandpass (HQ690/60x, Chroma, USA) interference filters were placed in the detection path to isolate the fluorescence from reflected excitation and STED light.
Figure 2 . shows the rhombus-like distribution of the STED and excitation light in the focal plane. For one beam pair the overlaid spot of excitation (green) and STED (red) focal spots were obtained by scanning an 80 nm gold bead (BBInternational, UK) across the focal region. We obtained four-leaf-shaped doughnuts for the STED beam at a wavelength of 745 nm due to the fourfold segmentation  while the excitation spots remained largely unaltered. The central intensity of the doughnut resides below signal background which should guarantee for good resolution improvement and fluorescence signal. The measured excitation spots were separated by 5.75±0.07 μm in the long axis at an angle of 45°.
To investigate the resolution of our system, 20 nm fluorescent beads (Invitrogen, USA) filled with Crimson fluorophores (625/645) were imaged both by confocal and STED microscopy in the four separate channels (Fig. 3 ) and recombined using either mosaic-like stitching or large-scale addition of overlapping frames. Please note the distinct rhombus-shaped stitching pattern of the first scanning scheme. In all four detection channels a 5-fold resolution improvement was obtained with the smallest feature imaged on the scale of 35 nm. The second scanning scheme, which was performed with a four times smaller pixel dwell time, provided similar resolution and signal. It is however exposed to the risk of compromising the resolution gain due to drift or residual errors in the offsets. Cross-correlation of the detection channels can minimize the recombination errors. No asymmetries, which could be attributed to molecular orientation effects or polarized excitation, were observed. All detection channels displayed similar detection count rates.
These observations confirm the proper alignment of our system regarding SWP position, beam quality, polarizations and detection. Please also note that the lenses placed in the detection path were adapted such that the detectors openings acted as confocal pinholes. Relatively low STED beam intensities were used to avoid extensive bleaching induced by the joint action of the excitation and the STED beams, which allowed the comparison of the same structure in the different detection channels (Fig. 4 .).
In Fig. 4 we tested the resolution improvement in all four channels over the entire field of view (FOV), approximately 80 x 80 μm2, by taking images using STED and confocal microscopy of tubulin in mammalian (PtK2) cells. We did not observe reductions in signal or resolution towards the image border. The different channels show identical images and resolution and are shifted laterally in the rhombus-like shape as expected (Fig. 4). Four parallel tubulin filaments can be clearly distinguished in the four STED images but cannot be resolved using confocal microscopy. Minor variations are always expected due to the multiple scanning and potential bleaching of each recording and the differing detection efficiencies.
To test the speed-increasing capabilities of our setup we scanned an area of 5.8 x 4 μm2 of vimentin in PtK2 cells, immunolabeled with the organic dye KK114, using STED and stitching the four channels together (Fig. 5 ). The individual frames (5.8 x 4 μm2) were then recombined by stitching, resulting in an effective scanning area of 11.6 x 8 μm2. After the scanning, a second single large scan was taken using only a single beam and the same region identified. The same features were detected with similar resolution as in the stitched image. Minor discrepancies observed were most probably due to previous bleaching. Overlapping of individual subframes can be minimized by carefully controlling scanning angle and area, but small errors may still occur in the border regions if stitching offsets are not carefully calibrated. Multiple scanning in the bordering region can also increase bleaching and lead to artifacts.
Finally we tested the second scanning approach by overlaying four large frames of fixed PtK2 cells with Atto647N-labelled tubulin after adding an offset in each of the channels at a dwell time of 50 μsec and compared it to a single image taken at 200 μsec (Fig. 6 ). The images gave comparable resolution and signal as a single frame at four times longer dwell time. The advantage of this approach is that recombination errors are spread across the entire image and no longer restrained to the stitching borders. However depending on the scanning area, the non-overlapping region can represent a major proportion of the image, which reduces the gain in recording speed. In brief, we demonstrated four times faster scanning capabilities of STED with respect to single beam scanning and acquisition. Importantly using this configuration, parallelization can be scaled up to more than m=4 excitation and STED beams and the acquisition rate can be increased accordingly using the scanning schemes demonstrated here.
We have designed an alignment-free parallelized STED setup that provides m = 4 -fold increased scanning speed with respect to regular single spot STED setups while delivering a similar resolution. Being based on an easySTED setup and incorporating only one laser source makes all STED and excitation beams quasi identical and intrinsically pre-aligned, both spatially and temporally. A stack of 2 Wollaston prisms in combination with an (easySTED) segmented waveplate (SWP) creates m=4 independent subdiffraction fluorescent sources separated by an angle set by the Wollaston prisms. The fluorescence from these detection channels can be recombined in different scanning modes and used to increase the scanning speed by the number of implemented channels.
The chromatic segmented waveplate (SWP), previously used for easySTED, transforms the STED wavelength into a doughnut-shaped spot with a nearly perfect central intensity zero while leaving the excitation wavelength unaltered, irrespective of the input polarization. Only this property enables the use of Wollaston prisms as beamsplitting elements for generating multiple spots for STED nanoscopy, which generates two spatially separated output beams with perpendicular linear polarizations. The benefit is that the relative intensities between the beams can be matched to each other by controlling the polarization of the input. Furthermore the angle between the beams, which directly translates to spot separation in the image plane, can be manufactured to any desired value and remains constant thereafter, i.e. virtually without drift.
We also introduced a novel beamscanning device based on four galvanometric mirrors offering two degrees of freedom for each scanning axis thereby allowing to control both position and plane of rotation of the beamscanning. This Quadscanner is placed close to the first conjugated image plane where each mirror pair effectively displaces the focused beams laterally, which in turn translates into point-scanning in the sample. This configuration makes the scanning lens obsolete and opens the first conjugated back-focal plane for beam-shaping devices, such as the Wollaston prisms in our setup, and allows for more compact and rugged beamscanning STED microscopes.
In conclusion, this paper demonstrated the first experimental parallelization of STED superresolution fluorescence microscopy. Scaling the parallelization further up is mainly limited by the power of the laser sources available, However, switching the fluorescence by transferring the fluorophores between long-lived states requires lower light intensities [5,9,10,14], in which case enough laser power is available to scale parallelization further up using this and related optical schemes.
We thank Ellen Rothermel for preparing the Ptk2 cells. PB was funded by an AFR PhD grant from the National Research Fund, Luxembourg (TR-PHD BFR08-059).
References and links
1. S. W. Hell and J. Wichmann, “Breaking the diffraction resolution limit by stimulated emission: stimulated-emission-depletion fluorescence microscopy,” Opt. Lett. 19(11), 780–782 (1994). [CrossRef] [PubMed]
3. S. W. Hell and M. Kroug, “Ground-state depletion fluorescence microscopy, a concept for breaking the diffraction resolution limit,” Appl. Phys. B 60(5), 495–497 (1995). [CrossRef]
4. R. Heintzmann, T. M. Jovin, and C. Cremer, “Saturated patterned excitation microscopy--a concept for optical resolution improvement,” J. Opt. Soc. Am. A 19(8), 1599–1609 (2002). [CrossRef] [PubMed]
5. S. W. Hell, S. Jakobs, and L. Kastrup, “Imaging and writing at the nanoscale with focused visible light through saturable optical transitions,” Appl. Phys., A Mater. Sci. Process. 77(7), 859–860 (2003). [CrossRef]
6. E. Betzig, G. H. Patterson, R. Sougrat, O. W. Lindwasser, S. Olenych, J. S. Bonifacino, M. W. Davidson, J. Lippincott-Schwartz, and H. F. Hess, “Imaging intracellular fluorescent proteins at nanometer resolution,” Science 313(5793), 1642–1645 (2006). [CrossRef] [PubMed]
12. T. A. Klar, S. Jakobs, M. Dyba, A. Egner, and S. W. Hell, “Fluorescence microscopy with diffraction resolution barrier broken by stimulated emission,” Proc. Natl. Acad. Sci. U.S.A. 97(15), 8206–8210 (2000). [CrossRef] [PubMed]
13. T. A. Klar, E. Engel, and S. W. Hell, “Breaking Abbe’s diffraction resolution limit in fluorescence microscopy with stimulated emission depletion beams of various shapes,” Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 64(6), 066613 (2001). [CrossRef] [PubMed]
15. S. W. Hell, “Increasing the resolution of far-field fluorescence light microscopy by point-spread-function engineering,” in Topics in Fluorescence Spectroscopy, J. R. Lakowicz, ed. (Plenum Press, New York, 1997), pp. 361–422.
16. A. Ichihara, T. Tanaami, K. Isozaki, Y. Sugiyama, Y. Kosugi, K. Mikuriya, M. Abe, and I. Uemura, “High-speed confocal fluorescence microscopy using a Nipkow scanner with microlenses for 3D-imaging of single fluorescent molecule in real time,” Bioimaging 4, 52–62 (1996).
18. M. G. L. Gustafsson, “Nonlinear structured-illumination microscopy: wide-field fluorescence imaging with theoretically unlimited resolution,” Proc. Natl. Acad. Sci. U.S.A. 102(37), 13081–13086 (2005). [CrossRef] [PubMed]
20. M. A. Schwentker, H. Bock, M. Hofmann, S. Jakobs, J. Bewersdorf, C. Eggeling, and S. W. Hell, “Wide-field subdiffraction RESOLFT microscopy using fluorescent protein photoswitching,” Microsc. Res. Tech. 70(3), 269–280 (2007). [CrossRef] [PubMed]
21. M. Reuss, J. Engelhardt, and S. W. Hell, “Birefringent device converts a standard scanning microscope into a STED microscope that also maps molecular orientation,” Opt. Express 18(2), 1049–1058 (2010). [CrossRef] [PubMed]
24. K. I. Willig, S. O. Rizzoli, V. Westphal, R. Jahn, and S. W. Hell, “STED microscopy reveals that synaptotagmin remains clustered after synaptic vesicle exocytosis,” Nature 440(7086), 935–939 (2006). [CrossRef] [PubMed]