A compact STED microscope providing 3D nanoscale resolution


Stefan W. Hell. Tel: +49 (0)551 201-2500; fax: +49 (0)551 201-2505; e-mail: shell@gwdg.de


The advent of supercontinuum laser sources has enabled the implementation of compact and tunable stimulated emission depletion fluorescence microscopes for imaging far below the diffraction barrier. Here we report on an enhanced version of this approach displaying an all-physics based resolution down to (19 ± 3) nm in the focal plane. Alternatively, this single objective lens system can be configured for 3D imaging with resolution down to 45 × 45 × 108 nm in a cell. The obtained results can be further improved by mathematical restoration algorithms. The far-field optical nanoscale resolution is attained in a variety of biological samples featuring strong variations in the local density of features.


Despite its limitations regarding the spatial resolution, far-field fluorescence microscopy enjoys ongoing popularity in the life sciences. Simple sample preparations under ambient conditions and the ability to perform imaging in 3D are compelling arguments in its favour. The introduction of stimulated emission depletion (STED) microscopy (Hell & Wichmann, 1994; Hell, 1997; Klar & Hell, 1999; Klar et al., 2000) has conceptually and practically overcome the diffraction barrier in far-field fluorescence microscopy and thus opened the door for far-field imaging with resolution in the nanometre range (Donnert et al., 2006). The basic idea behind STED and related optical microscopy concepts (Hell & Kroug, 1995; Hell, 1997, 2003, 2007; Hell et al., 2003) is to switch off or inactivate fluorescent molecules in the focal region using a dedicated beam of light (STED beam) featuring a zero at a moving coordinate in space. The result is that only molecules at sub-diffraction proximities to the zero are active and hence contribute to the signal at the given coordinate. More recent approaches in far-field fluorescence nanoscopy do not switch the molecules with spatially structured beams but individually and stochastically in space (Betzig et al., 2006; Hess et al., 2006; Rust et al., 2006). Complementing the earlier approaches, they have substantially broadened the range of applicability of emerging far-field optical nanoscopy.

The central zero of the STED beam is produced by phase modifying the wavefront of a laser beam, which is then spatially overlaid with the focal spot of the excitation beam. The wavelength of the STED laser is chosen such that if molecules become excited, they are immediately quenched to the ground state by stimulated emission. Thus, molecules subject to the STED beam are essentially deprived of their ability to fluoresce or simply switched off, whereas those located within a small range at the zero maintain their fluorescence capability. Consequently, the effective point-spread function (PSF) of the STED microscope is narrowed down to sub-diffraction dimensions, which upon scanning translates into an enhanced spatial resolution, because features closer than the diffraction limit are recorded sequentially in time. The full-width half-maximum (FWHM) of the region from which the fluorescence can be emitted is to first approximation given by (Hell, 2003, 2004; Westphal & Hell, 2005)


where λ is the STED wavelength, n is the refractive index of the medium, α is the semiaperture angle of the objective lens and I is the intensity maximum engulfing the zero. Is is the characteristic intensity required for reducing the fluorescence ability by a factor of two for the dye being used; it typically amounts to 10–30 MW cm−2 when implementing STED microscopy with pulses on the order of 100 ps. It should be noted that STED microscopy does not improve the resolution by narrowing the employed light beams, which remain diffraction limited of course; rather the resolution increase rests on the specific employment of distinct molecular states (a fluorescent and a non-fluorescent one) for the separation of adjacent features.

The use of pulsed lasers is not strictly necessary (Willig et al., 2007), but due to the electronic and vibrational kinetics of the fluorophore, STED microscopy is most efficiently implemented using pulsed lasers with pulse durations smaller than the fluorescence lifetime of a few nanoseconds. In addition, the use of pulse repetition rates low enough to allow a relaxation of the triplet state reduces photobleaching (Donnert et al., 2007a). For this reason, STED microscopy has been notoriously associated with complex setups and expensive, high-maintenance laser systems.

Recent developments in laser technology have allowed revisiting the light source issue in STED microscopy. Initially, pulsed supercontinuum sources were used as a source of fluorescence excitation only (Auksorius et al., 2008) but, eventually allowed a replacement of the STED laser as well. Concretely, a STED microscope realized at a fraction of the cost and complexity of previous setups was introduced, which used a single commercial all fibre-based supercontinuum source and a commercial vortex phase plate for preparing the focal doughnut (Wildanger et al., 2008). The system delivered a lateral spatial resolution in the 30–50 nm range in various colours. Although it required little alignment and afforded long-term stability, the reported compact STED microscope did not yet fully exploit the potential of the supercontinuum laser source. This stems from the fact that the vortex phase plate requires circular polarization. However, because the supercontinuum beam is unpolarized, half of the power was dumped. By adding a second beam path with orthogonal circular polarization, we have now extended the performance of our supercontinuum STED microscope, achieving an all-optics based spatial resolution down to 20 nm. Alternatively, we used the second beam to realize 3D sub-diffraction resolution in a compact single objective lens based STED system.

Materials and methods

STED microscope

Because the basic design of the supercontinuum STED microscope has been detailed elsewhere (Wildanger et al., 2008), the subsequent discussion concentrates on the relevant changes in the setup (Fig. 1). In order to take advantage of the full spectral power density of the supercontinuum laser (SC-450-PP-HE, Fianium Ltd., Southampton, UK), a total of three beams are derived from the source: an excitation beam and two STED beams. The laser output is first filtered to remove the infrared components (λ > 950  nm) of the spectrum and is then split using a polarizing beamsplitter cube to provide two orthogonally polarized but otherwise equal beams. Both beams are individually directed to prism-based monochromators selecting a ∼20-nm-wide spectral band at 700 nm for STED. The STED beams are then coupled into two polarization-preserving single mode optical fibres. The excitation laser beam is derived from the s-polarized beam using a shortpass dichroic beamsplitter (Z670SPRDC, AHF Analysentechnik GmbH, Tübingen, Germany) positioned prior to the monochromator. It is then directed through a bandpass filter (Z570/10×, AHF Analysentechnik GmbH) and coupled into a polarization-preserving single mode fibre.

Figure 1.

Experimental setup: the supercontinuum beam is split into two orthogonally polarized beams (polarizing beamsplitter cube), which are spatially filtered through single-mode optical fibres after the STED wavelength has been extracted (WS). After collimation, the beams pass through separate phase plates (PM 1 and 2) and are recombined at a polarizing beam splitter. A dichroic mirror (DC3) couples the STED light into an objective lens. The excitation beam is extracted from the s-polarized supercontinuum beam by employing a dichroic mirror (DC1). After the desired wavelength has been extracted with an interference filter (EF) the beam is spatially filtered (SMF) and coupled into the microscope by a further dichroic mirror (DC2). A quarter-wave plate in front of the objective lens ensures circular polarization for all beams. The fluorescence light passes through the dichroic mirrors and is focused into a multi-mode fibre, which acts as confocal pinhole and is finally detected with an avalanche photodiode; object scanning is performed with a piezo scan stage. Inset: combinations of phase plates and resulting STED PSFs used to achieve either ultimate lateral resolution (HR) or 3D superresolution (3D).

At the fibre outputs, the two STED laser beams are separately collimated, recombined with a polarizing beamsplitter cube and coupled into the objective lens with a dichroic beamsplitter (Z680SPRDC, AHF Analysentechnik GmbH). Likewise, the excitation light leaving the optical fibre is collimated and is coupled into the objective lens with a second dichroic mirror (Z568RDC, AHF Analysentechnik GmbH). Careful adjustment of the orientation and tilt of a superachromatic quarterwave plate (RSU 1.4.15, B. Halle Nachfl. GmbH, Berlin, Germany) placed in front of the objective lens renders circular polarization for both STED beams.

For the purpose of 3D nanoscopy, two different phase plates were placed into the STED beams to achieve a lateral (xy) and an axial (z) resolution enhancement, respectively (Fig. 1, inset). In this configuration, a vortex phase plate (RPC Photonics, Rochester, NY, U.S.A.) is used in conjunction with a custom-made phase plate introducing a π phase shift in a central disk that covers half of the beam's photon flux. For the experiments targeted at the highest possible lateral (xy) resolution, two vortex phase plates were placed into the STED beams. To ensure the appropriate circular polarization for the s- and the p-polarized beams the helical phase ramps have to be placed in opposite senses of direction. The pulse timing was adjusted by equalizing the optical path lengths of all three beams.

To account for the improved resolution of the new microscope the piezo scan stage was replaced with a model which allows higher scan speeds and provides superior scanning precision (P-733.3DD and controller E-710.2CD with dynamic scan linearization, Physik Instrumente GmbH, Karlsruhe, Germany). The coarse positioning was carried out with two crossed linear stages (M426, Newport Spectra-Physics GmbH, Darmstadt, Germany) mounted on a custom-built elevation (z) stage. Typically, images of 10 × 10 μm up to 20 × 20 μm were acquired with a 10–20 nm pixel size and pixel dwell times of 0.2–1.0 ms. The STED and confocal reference images were recorded sequentially. Under these conditions, the total acquisition time for an image of 10 μm × 10 μm, 10 nm pixel size, 0.2 ms dwell time is 4 min.

Cell culture

The SH-SY5Y neuroblastoma cell line was grown as described previously (Encinas et al., 2000). Cells were seeded on standard glass cover slips to a confluence of about 80%.  A total of 10 μM all-trans-retinoic acid (EMD Biosciences Inc., San Diego, CA, U.S.A.) was added the day after plating. After 5 days in the presence of retinoic acid, the cells were washed three times and incubated with 50 ng mL−1 human brain derived neurotrophic factor (Alomone Laboratories, Jerusalem, Israel) in serum-free medium for >7 days.

Fluorescence staining

Immunostaining of neurofilaments and vimentin was performed with anti-O-glycosylated NF-M [NL6] mouse IgG (Sigma, Munich, Germany) and anti-vimentin [V9] mouse IgG (Sigma), respectively, as primary antibodies and with ATTO 590 sheep anti-mouse IgG (ATTO-TEC GmbH, Siegen, Germany/Dianova GmbH, Hamburg, Germany) as a secondary antibody, respectively. After several washes the cover slip was embedded in Mowiol mounting medium. Immunolabelling of microtubules in PtK2 cells was carried out according to the same protocol but with anti-beta-tubulin mouse IgG (Invitrogen GmbH, Karlsruhe, Germany) as the primary antibody.

Results and discussion

In order to characterize the resolution improvement by the use of a second STED beam, we imaged fluorescent nanoparticles [FluoSpheres®, red fluorescent (580/605), Invitrogen GmbH] dispersed on a microscope cover slide. Unless otherwise noted, 44 nm beads were used for the single-beam experiments, whereas 24 nm particles were used in the dual-beam measurements. Figure 2 shows the comparison between the confocal reference and the STED recordings with a single (A, B) and with two beams (C, D) corresponding to time-averaged STED powers of 2.5 and 5 mW which give rise to pulse energies (peak intensities) of 2.5 nJ (9.2 GW cm−2) and 5.0 nJ (18.3 GW cm−2), respectively. The beads in the STED images exhibit FWHM of (45 ± 7) nm and (35 ± 4) nm, respectively. The example shown in Fig. 2(C) also demonstrates that two beads spaced only 36 nm apart are separated by a pronounced dip in brightness of 30%. When the bead diameters are taken into account, the optical resolution of the microscope can be inferred to be ∼42 nm and ∼31 nm, respectively. This corresponds to a near 1.4-fold increase in resolution, which is in perfect agreement with Eq. (1), which predicts an enhancement by a factor of inline image.

Figure 2.

Red fluorescent beads (Ø= 44 nm) imaged confocally (A) and with one STED beam (B). Red fluorescent beads (Ø= 24 nm) imaged confocally (C, deconvolved) and with two STED beams (D, deconvolved). Scale bar: 500 nm. (E) Line profile along the path indicated in (D): two beads separated by 36 nm are clearly discernible. (F) Histograms of the measured FWHM of 24 nm beads imaged with a single (green, ISTED= 9.2 GW cm−2) and with two STED beams (orange, ISTED= 18.3 GW cm−2).

Figures 3 and 4 show exemplary recordings of immunolabelled neurofilaments and vimentin fibres. In each case, the confocal reference is contrasted with the respective STED image, showing a near 10-fold resolution improvement taking advantage of physical phenomena only. The images of these fibrous structures demonstrate that the improved resolution discerns rich detail about the fibre bundles, including the organization of individual filaments. By measuring the width of single filaments, features down to (19 ± 3) nm in size are found, which is about the size of the antibody construct. Hence, in the immunofluorescence imaging, the resolution is superior to that of the imaging of beads. We ascribe this fact to the superior properties of the dye ATTO 590 at the used wavelengths.

Figure 3.

Fluorescence imaging of immunostained neurofilaments. Bundled neurofilaments imaged with confocal (left) and with STED microscopy (right, data deconvolved). Insets (i) and (ii) show intensity profiles along the paths indicated by the pink and orange arrows, respectively. Although the fibre bundles are blurred in the confocal images individual strands are well resolved in the STED images. Also the organization of the filaments can be clearly discerned. Measurement of point objects indicate a resolving power of (19 ± 3) nm focal plane resolution. Scale bars: 1 μm.

Figure 4.

STED imaging of densely packed immunostained neurofilaments: confocal (left) and with STED microscopy (right, data deconvolved). Insets highlight densely packed fibre bundles that are entirely blurred in the confocal images whereas individual strands are well resolved in the STED images. Scale bars: 1 μm.

The fluorescent marker also showed good photostability. It was possible to image the same regions in the vimentin or neurofilament labelled cells with adequate contrast multiple times. Specifically, the signal dropped to about half of its initial value only after four nanoscale recordings. To investigate whether the observed photostability was a unique feature of the ATTO 590 dye, we prepared a series of neurofilament samples using the same immunofluorescence protocol but with different dyes attached to the secondary antibody: ATTO 590, ATTO 594 (both from ATTO-TEC GmbH), DyLight 594 (Thermo Scientific/Perbio Science, Bonn, Germany) and Alexa 594 (Invitrogen GmbH). A small region of each sample was repeatedly imaged with the same time-averaged excitation power (0.27  μW) and STED power (3.8  mW), corresponding to pulse energies (peak intensities) of 0.3 nJ (4.2 MW cm−2) and 3.8 nJ (13.9 GW cm−2), respectively. All other experimental parameters were kept constant, in particular the excitation and STED wavelengths, pulse lengths and pixel dwell times. In each image, an approximately 3  ×  3  μm2 region containing representative fluorescent features was selected.

The integrated fluorescence from these regions normalized to the first frame is plotted as a function of frame number in Fig. 5(A). Among the investigated dyes, ATTO 590 displayed the best photostability. The bleaching curves follow monoexponential decays, and fitting these decays provide the following fractional loss of fluorescence per image: 17% (ATTO 590), 19% (ATTO 590), 20% (DyLight 594), and 27% (Alexa 594). These numbers indicate that the variations in photostability are relatively small among these dyes and that several commercially available fluorescent dyes are suited for STED microscopy with STED wavelengths around 700 nm.

Figure 5.

Photobleaching experiments. Neurofilaments were immunolabelled with various dyes (DyLight 594, ATTO 594, ATTO 950, Alexa 594) and were repeatedly imaged. The integrated and normalized fluorescence intensity is plotted as a function of the frame number (A). To study the influence of the STED pulse energy the measurement was repeated with a ATTO 590 stained sample using increasing STED laser powers (B). The STED pulse energies and the corresponding lateral resolutions are given in parentheses. All images were acquired with the same pixel size (20 nm) and dwell time (0.2 ms). The excitation intensity was kept constant.

A further series of bleaching measurements was performed with the same dye throughout (ATTO 590) but with increasing STED intensities. The other experimental conditions were maintained, but now the STED power was varied between 0 and 3.8 mW to afford pulse energies in the range of 0–3.8 nJ (Fig. 5B). The experiments show that fading is caused by both the excitation and the STED beam. The photobleaching caused by STED is acceptable but, in total terms, it accounts for most of the observed loss in fluorescence. Although the exact mechanisms underlying the STED photobleaching have not been investigated yet in full, they surely involve photon absorption of excited singlet (S1) (Dyba & Hell, 2003) and triplet state (T1) molecules (Donnert et al., 2007a). Our observation is that the importance of each state in the bleaching pathway of a particular dye strongly depends on the dye and to some extent also on the molecular environment. Nonetheless, the observed photostabilities are remarkable compared to earlier experiments in which >50% loss of fluorescence after a single recording was not uncommon (Dyba & Hell, 2003). We assume that an important ingredient to the observed photostability is the relatively low repetition rate of 1 MHz, which allows an efficient relaxation of the triplet population between successive pulses, thus eliminating the photobleaching pathways via the triplet state (T-Rex effect) (Donnert et al., 2007a). Furthermore, the temporal and spectral structure of the laser pulses obtained from the supercontinuum spectrum may differ from the conventional laser sources used previously and may augment photostability; this hypothesis merits further investigation.

Although the doughnut employed in our configuration improved the lateral resolution by ∼10-fold over that of the confocal microscope, the axial (z) resolution remained limited to ∼600 nm. Because the objects shown in Figs 3 and 4 are rather sparse, the signal-to-background ratio and the spatial resolution are not compromised by the fluorescent background. However, if dense structures rather than isolated filaments or clusters are imaged, both the contrast and the resolution can suffer substantially. One approach to overcoming the background issues is to embed the sample in a polymer resin and cut it into 50–100-nm-thick slices (Punge et al., 2008). This procedure also minimizes photobleaching and allows 3D reconstruction. However, the low photobleaching rates observed in our recordings, prompted us to pursue a purely optical axial superresolution approach by STED. In contrast to the slicing technique, it is non-invasive and leads directly to a 3D stack which needs no further alignment. To this end, we replaced the vortex phase plate in the s-polarized beam with a phase plate that introduces a constant π phase shift in a circular region covering half the area of the back aperture of the objective lens. The resulting STED focal spot features a central zero along with two main maxima, one above and one below the focal plane. To confine the effective fluorescence excitation both laterally and axially, the z-doughnut is incoherently overlaid with the doughnut-shaped STED spot produced by the p-polarized beam (Fig. 1, inset). By varying the intensities of the two STED beams, the aspect ratio of the resulting effective PSF can be adjusted over a wide range, including a nearly isotropic PSF.

The resolution provided by this configuration was again tested on 44 nm fluorescent nanoparticles. An xz-section (10 × 1 μm2, pixel size 20 × 33 nm2) containing a single fluorescent bead was acquired, followed by analysis of two line profiles placed along the optic axis (z) and within the focal plane (x), see Fig. 6(A). By fitting Lorentzian functions to these profiles and taking the 44 nm diameter of the fluorescent particles into account, the FWHM of the effective PSF were estimated to be 45 nm laterally and 108 nm axially corresponding to a ∼175-fold decrease in focal volume compared to the confocal PSF. Three-dimensional imaging was further pursued on immunostained microtubules. A region of 15.3 × 15.3 × 2 μm3 with a voxel size of 20 × 20 × 66 nm3 was scanned. After deconvolution and background subtraction, a 3D reconstruction of the data was created with the visualization software package amira (Visage Imaging, Berlin, Germany). An isosurface rendering of the reconstruction along with a maximum projection is shown on the top of Fig. 6(B) whereas in the lower panel the rendering shows an xz section of the same dataset. The intensity line profile shown in the inset discloses features which extend less than 100 nm in the z-direction. As a result, structures which are 200 nm apart are clearly resolved. Finally, Fig. 6(C) shows five subsequent slices from a different image stack.

Figure 6.

STED microscopy with superresolution in 3D. (A) Resolution measurement on fluorescent nanoparticles (44 nm) reveals 52 nm and 110 nm in the lateral (x-) and axial (z-) directions, respectively. (B) 3D imaging of immunolabelled microtubules: isosurface rendering (top) and maximum intensity projection along the y-axis (bottom) of a stack of images comprising 30 slices. The inset shows an intensity profile at the indicated position. (C) Slices from a different image stack acquired at a 100 nm pitch in the z-direction.


We have demonstrated a compact supercontinuum laser source providing a lateral resolution of 20–25 nm in STED microscopy. The beams can also be prepared to render a 3D STED microscopy resolution up to 45 × 45 × 108 nm in a single-lens setup. The gain in resolution provides object details that are completely indiscernible by conventional or confocal microscopy. In combination with the supercontinuum laser source, various dyes proved remarkably photostable and allowed 3D superresolution imaging by purely optical sectioning. It should be noted that, unlike in fluorescence nanoscopy concepts utilizing single molecule switching, the practically obtained spatial resolution does not depend on the local concentration of fluorophores or on the density of the features to be imaged; the resolution is just a function of the dye and the employed aperture and wavelength. This is particularly obvious in Fig. 4, where neurofilaments are clearly discerned in a single image despite the strong variations in local packing density.

The current laser system operates at a repetition rate of only 1–2 MHz which, by limiting the attainable scan rate, accounts for the relatively long recording times. In the near future, supercontinuum sources operating at 5–10 MHz repetition rate are expected to become available. These systems should accelerate image acquisition by a similar factor and enable comparable or superior resolution with a single STED beam. Another modification will extend the microscope to two-colour operation as has already been demonstrated in more complex setups (Donnert et al., 2007b). In conclusion, STED microscopy using supercontinuum sources has a significant potential for future STED microscopy implementations that will prove invaluable in many applications.


The research leading to these results has received funding from the European Community's Seventh Framework Programme FP7/2007–2011 under grant agreement no. 201837. D.W. acknowledges a doctoral fellowship by the German National Academic Foundation.