Ultrafast superresolution fluorescence imaging with spinning disk confocal microscope optics

INTRODUCTION

Fluorescence microscopy is now an essential tool for the cell biologist. However, most subcellular structures are <200 nm and blurred by diffraction (Abbe, 1873). Superresolution (SR) techniques, such as photoactivatable localization microscopy (PALM; Betzig et al., 2006), stochastic optical reconstruction microscopy (STORM; Rust et al., 2006), and stimulated emission depletion microscopy (STED; Hell and Wichmann, 1994), can extend the resolution down to ≤50 nm, but usually at the expense of image acquisition time. Therefore their applications are rather limited to fixed specimens. In living cells, some cellular structures, such as endosomes and other vesicles, move at velocities >3 μm/s (Allen et al., 1982). Thus the image acquisition time needs to be <<30 ms for 100-nm-resolution imaging.

From this viewpoint, PALM and STORM are not suitable for observing the rapid dynamics of subcellular structures in living cells because they reconstruct a single SR image from at least thousands of raw images, requiring >10 s of imaging for a single frame. STED can be accelerated, but faster image acquisition requires orders of magnitude stronger illumination than conventional imaging, which risks photodamage. As a result, the practical limit of STED image acquisition time is longer than a few microseconds/pixel, meaning 10 s for a 1024 × 1024–pixel field of view (Huang et al., 2009).

A third SR technique, structured illumination microscopy (SIM; Heintzmann and Cremer, 1999), can extend the resolution to 120 nm, twice the resolving power of conventional wide-field (WF) microscopy. Although its resolution is worse than that of the other two techniques, SIM can achieve faster frame rates with little photodamage and relatively easy multicolor imaging. One study reported a frame rate up to 11 Hz (Kner et al., 2009). However, this may be the best achievable frame rate because SIM requires 9–15 raw images to reconstruct a single SR image. A related technique, multifocal structured illumination microscopy (MSIM; York et al., 2012), a parallelized form of image scanning microscopy (Sheppard, 1988), generates an image with ∼150-nm resolution at ∼1-Hz frame rate. Recently an improved implementation of MSIM, termed instant SIM, was reported (York et al., 2013). A 100-Hz frame rate with ∼150-nm resolution was achieved by using several optical elements, such as two sets of microlens array and pinhole array; however, expertise is required to align all these optical elements correctly.

Here we report another implementation of SIM using spinning disk confocal microscope optics, which we named the spinning disk superresolution microscope (SDSRM). We demonstrate that the commercial spinning disk confocal microscope can be upgraded into a SR microscope simply by replacing the disk. It can achieve the same resolution as SIM, 120 nm, but with a frame rate of 10 ms/frame. This is the fastest frame rate reported for SR imaging (York et al., 2013) and is fast enough to image fine structures in living cells moving at up to 6 μm/s.

RESULTS

Principle of SDSRM

The basic principle of our SR method is illustrated in Figure 1 and fully discussed in Supplemental Text 1. Here we use a one-dimensional model and omit the magnification of the imaging system for simplicity. In WF microscopy, the sample is illuminated homogeneously, and the image contains only the signals within the cut-off frequency of the optical system (f_c), which determines the diffraction limit of the resolution. In conventional SIM (Figure 1A), a raw image is taken with illumination modulated sinusoidally. SR signals—namely, the signals of the structure beyond f_c—are overlaid as Moiré interference patterns with a frequency shift by f₀, the spatial frequency of the sinusoidal illumination pattern. The SR signals are demodulated by digital processing using three raw images taken with illumination patterns phase shifted by 120° each. Thus the cut-off frequency of SIM is extended to f_c + f₀. Usually, the same objective lens is used for imaging and illumination, so that f₀ can be designed up to f_c (f₀ ≤ f_c). Hence, f_c + f₀ ≤ 2f_c. Namely, at most twice-better resolution can be achieved with SIM (Figure1C and Supplemental Figure S1A).

In SDSRM, a stripe pattern with opening width w and period p is used as a modulator for illumination and demodulator for detection (Figure 1B). The stripe pattern is shifted continuously, and the image is integrated on the camera. Thus the same confocal mask pattern serves as both modulator and demodulator, and frequency-shifted SR signals are already returned to the original position optically. Although this optical configuration is the same as the conventional line confocal (CF) microscope, the design of the stripe pattern is different. In conventional CF, p is much larger than w to minimize cross-talk (Shimozawa et al., 2013). Therefore the amplitude of the SR signals is much weaker than the WF signal (Figure 1D and Supplemental Figure S1B) because it is proportional to (p/w) sin²[(w/p)π] (Supplemental Text 1). Thus the effective cut-off frequency (f_c′) of CF is limited to , meaning only -fold increase in resolution (Sheppard, 1988). SR signals between f_c′ and f_c + f₀ are buried in the noise and need statistical estimation for recovery with an algorithm such as Richardson–Lucy deconvolution (York et al., 2012). Instead, we designed the stripe pattern as w = 80 nm and p = 270 nm on the sample plane to increase the SR signals. This para­meter is close to the theoretical maximum (w/p = 0.35) and increases the SR signal more than five times the conventional CF stripe pattern. Furthermore, we designed a digital high-pass filter to further enhance the SR signals (Figure 1E, Supplemental Text 2, and Supplemental Figure S2). Thus we can recover an image with fully double the resolution from a single raw image of line CF with the designed stripe pattern.

Instrumentation of SDSRM

On the basis of this principle, we modified a spinning disk microscope (Figure 2A). The disk, light source, projection lens, and camera were changed from the commercial system.

For the scan head, we used the disk scanning unit (DSU; Olympus, Tokyo, Japan), whose disk is easy to replace. The stripe pattern of the disk was designed to match the condition discussed. Here w and p are 500 nm and 5 μm on the sample plane, respectively, with the commercial DSU for 60×/numerical aperture (NA) 1.4 objectives. Both w and p were narrowed to 80 and 270 nm on the sample plane when projected through a 60× objective (Figure 2, B and C). This custom-designed disk was made by using electron lithography on a low-reflection chrome coat that covered a fused silica substrate (DU-DSR1-SP; Prototype Development Division, Olympus).

The Olympus DSU system normally uses a mercury lamp as the illumination, but the light throughput for the illumination is only 14% with our custom-designed disk mainly due to the opening ratio and the diffraction efficiency of the disk pattern. We therefore replaced the light source with a laser (488 nm, 500 mW; Sapphire 488HP; Coherent, Santa Clara, CA). The laser beam was expanded with a lens to fill the aperture, and the speckle noise was removed by a rotating diffuser (light stirrer; NANDN, Tokyo, Japan).

The image was projected onto a charge-coupled device (CCD) camera (CoolSNAP HQ; Photometrics, Tucson, AZ) or an sCMOS camera (ORCA FLASH 4.0v2; Hamamatsu, Hamamatsu, Japan) via a magnifier to satisfy the Nyquist condition for up to 100-nm resolution. The SR components in the raw image were amplified by Fourier-space filtering using the Custom Filter command of ImageJ (National Institutes of Health, Bethesda, MD). The filter (Supplemental Figure S2E) was designed as discussed in Supplemental Text 2.

The resolution of the system was confirmed using latex beads (Supplemental Figure S3). In our hands, a backilluminated electron-multiplying CCD camera (iXon897; Andor, Belfast, United Kingdom) failed to extend the resolution even after the digital filtering. The SR components seemed to be weakened through the potential cross-talk between the adjacent pixels.

Proof-of-principle experiments with a low-NA objective

As a proof-of-principle experiment, we examined a microtubule sample under four conditions: conventional WF with NA 0.5 objective (Figure 3, A–C), CF with the same NA 0.5 objective followed by deconvolution (Figure 3, D–F), SDSRM with the same NA 0.5 objective (Figure 3, G–I) and conventional WF with NA 0.95 objective (Figure 3, J–L). Here a mercury lamp was used for illumination, and the images were taken with a CCD camera (CoolSNAP HQ).

Because the diffraction limit is proportional to the inverse of the NA (Abbe, 1873), the NA 0.95 objective yields an image with 1.9-times-higher resolution than the NA 0.5 objective. Nonetheless, the image obtained by the SDSRM with the NA 0.5 objective was essentially the same as the image obtained by conventional WF with the NA 0.95 objective, indicating that the SDSRM is a real SR, which improves resolution by a factor of ∼2 compared with conventional WF. Note that deconvolution of the CF image did improve resolution over WF but failed to achieve full 2× resolution (Figure 3, D–F). We tried several different deconvolution software packages, but all gave similar results (Supplemental Figure S4).

SR live imaging of microtubules

We next examined whether our SDSRM can resolve biological structures <200 nm with a high-NA objective (UPLSAPO60XO, NA1.35; Olympus) in living cells. Microtubules were stained by expressing enhanced yellow fluorescent protein (EYFP)–tagged β-tubulin. The raw images were taken by a CCD camera (CoolSNAP HQ) at 5 frames/s with laser excitation 20 W/cm² at the sample. As shown in Figure 4, the width of a single microtubule image is blurred to 260 nm (full-width at half-maximum) in a conventional WF image, which is improved by a factor of 2 to 120 nm with the SDSRM (Figure 4C). A pair of microtubules with 250-nm distance was not separated in the conventional WF image but was clearly separated with the SDSRM (Figure 4D). Furthermore, the dynamics of microtubules was recorded at 5 frames/s (Supplemental Movie S1).

Fast SR live imaging of endosomes

The initial experiments with the CCD camera described earlier confirmed that SDSRM can achieve ∼120-nm resolution, twice better than with conventional WF microscopy, even with the live biological specimen. Next we tried a faster frame rate by replacing the camera with sCMOS camera (ORCA FLASH 4.0v2), which can achieve a frame rate up to 100 frames/s, faster than the video rate. A silicone oil immersion objective lens (UPLSAPO60XS, NA1.3; Olympus) was used to minimize the spherical aberrations by the refraction index mismatch.

As the proof of principle of fast SR imaging, we chose endosomes and lysosomes, which are known to be highly dynamic. They often move at >1 μm/s, and imaging at faster than the video rate is required for capturing their dynamics without motion blur. The dynamics of lysosomes was observed with lysosomal-associated membrane protein 1 (Lamp1)–EYFP. The images were taken with 20-ms exposure. A 100-ms-interval time series of two lysosomes is shown in Figure 5. They sometimes move at as fast as 10 μm/s; the lumen is still clearly visualized, demonstrating that the rapid frame rate enabled SR imaging without motion blur (Supplemental Movie S2). Similarly, the rapid dynamics of recycling endosomes including a fusion event was clearly visualized by 100-Hz SR live imaging (10-ms exposure) of Rab11a-venus (Figure 6 and Supplemental Movie S3).

SR imaging of mitochondrion-derived vesicles

Finally, we applied the SDSRM for the imaging of recently reported but poorly characterized organelles, mitochondrion-derived vesicles (MDVs). They are proposed to be formed by budding of the mitochondria outer membrane and to mediate the transport of damaged proteins from mitochondria to peroxisomes (Neuspiel et al., 2008). However, it has been difficult to examine the structure and dynamics of the outer membrane of mitochondria due to the limited resolution of light microscopy. We therefore examined the mitochondrial outer membrane first by staining with anti-TOMM20 antibody (Figure 7). The lumen of the mitochondria is barely visible with conventional WF imaging but is clearly observed with SDSRM as well as STED (Figure 7D). Unexpectedly, the SR image of mitochondrial outer membrane showed that the size of the lumen of mitochondria is highly variable even in a single cell. We also found thin protrusions from thick mitochondria. Only the outer membrane is apparently protruded, and the lumen is thin and unclear even with the STED SR image (Figure 7E). In addition to the mitochondria and the thin protrusions, many vesicular or tubular structures were also stained with TOMM20 antibody. They would correspond to MDVs as reported previously (Neuspiel et al., 2008).

We next expressed TOMM20-mEmerald and used the SDSRM with exposure 30 ms and excitation 20 W/cm². The 120-nm-resolution live imaging at video rate with SDSRM enabled us to observe not only the mitochondrial fusion event, but also the rapid dynamics of MDVs (Figure 8 and Supplemental Movie S4). The protrusions of the outer membrane were highly dynamic: elongating, shrinking, and branching. Sometimes a thin tubular vesicle was released from the protrusion (yellow arrowhead in Figure 8), which suggests that the protrusion would be pinched off to form MDVs.

DISCUSSION

In this study, we demonstrated that the spinning disk confocal microscope has potential as an alternative implementation of SIM. Conventional SIM requires several modulated images to retrieve SR components by demodulation in the Fourier space, but it has higher photon efficiency because the signal is retrieved without a mask. In contrast, the confocal aperture works as a mask to demodulate the SR components in the confocal microscope. To obtain stronger SR component signals, a smaller confocal aperture is required. However, this comes at the expense of signal intensity because with a smaller confocal aperture, most of the photons from the sample are rejected at the confocal aperture. Here we overcome this problem by specially designing the mask pattern on the disk. The opening was narrowed to λ/5 while keeping the aperture ratio as 30%. This aperture ratio is more than three times higher than for the conventional spinning disk confocal, and this high light throughput enabled 100-Hz SR imaging with moderate excitation power (20–40 W/cm²). Owing to this high aperture ratio, however, the optical sectioning capability is compromised. The stray light from the out-of-the-focus specimen cannot be rejected efficiently, and the axial resolution is close to that of conventional WF imaging. The axial resolution can be enhanced by reducing the aperture ratio but at the expense of light throughput or signal intensity. A longer exposure time would be required.

Even with the high-light-throughput disk pattern, stronger excitation light was required than for typical fluorescence live-cell imaging. Twice-better resolution in two dimensions means one-fourth signal intensity per pixel, and only 30% of the signal passes through the disk to reach the camera. Thus an at least 12-times stronger signal is required. Furthermore, even higher signal-to-noise ratio is required than for conventional WF imaging because the weak SR components are enhanced by high-pass filtering, which can also enhance the noise. In this study, therefore, most of the images were acquired with excitation light intensity of 20–40 W/cm², which is nearly 100 times stronger than for typical fluorescence live-cell imaging (0.1–1 W/cm²). Note, however, that this excitation beam light intensity is comparable to that for typical SIM imaging and >100 times weaker than for STED. Furthermore, the excitation light intensity can be further reduced for light-sensitive samples by elongating the exposure time, although the temporal resolution is compromised.

As applications of live-cell SR imaging, we observed the dynamics of microtubules, lysosomes, recycling endosomes, and mitochondria. The short exposure time (<30 ms) enabled SR live imaging without motion blur, and the fine structures of rapidly moving organelles were captured, such as the fusion and fission events of the outer membrane of mitochondria and the rapid dynamics of mitochondrial protrusions and MDVs. To our knowledge, this is the first visualization of MDV formation from mitochondrial protrusions. The frame rate of our system is mainly limited by the rotation speed of the disk (1800 rpm), which would be possible to accelerate 10 times to achieve a frame rate of 1 kHz. This would enable imaging of rapid processes such as neurotransmitter release at the synapse (Südhof, 2013)

In summary, the spinning disk confocal-based implementation of SIM described here is easy to implement and robust to use. An existing spinning disk confocal can easily be upgraded. Thus we hope that this method will not only open the way for SR live-cell imaging with high temporal resolution, but also help to popularize SR microscopy.

MATERIALS AND METHODS

Microscopy setup for SDSRM

The microscope setup for SDSRM (Figure 2A) is based on the disk-scanning confocal microscope system, which includes an inverted microscope base (IX81; Olympus) and a disk-scanning unit (IX2-DSU; Olympus). A combination of a cooled CCD (CoolSNAP HQ) and a mercury lamp with a light guide (U-HGLGPS, Olympus; not shown in Figure 2A) was used for the initial experiments (Figure 3 and Supplemental Figure S3). The light source was then replaced with a laser (Sapphire488HP; Coherent, Santa Clara, CA) and a square-core fiber (ST175SQE(12); Mitsubishi Cable, Tokyo, Japan) with a vibrating motor (not shown in Figure 2A; used for images in Figure 4) or a single-mode fiber (kineFLEX; Qioptiq, Asslar, Germany) with a rotating diffuser (light stirrer; NANDN; used for images in Figures 5–8). Finally, the CCD camera was replaced with an sCMOS camera (ORCA FLASH 4.0v2) for faster frame rates (for Figures 5–8).

The disk-scanning unit has a spinning disk and a filter-set wheel changer. The spinning disk is put on the primary image plane for confocal imaging. The disk was made by using electron lithography on a low-reflection chrome coat that covered a fused silica substrate (DU-DSR1-SP; Olympus). The width and the pitch of the confocal slit apertures were 80 and 270 nm, respectively, as discussed in the main text.

For comparisons with WF, the disk was removed and a neutral density filter with a transmission factor of 6% (ND6) was put into the illumination path to adjust the image brightness. The image was projected onto the camera via a 5× magnifier and taken by 2 × 2 binning mode. Thus the pixel size of the specimen is 43 nm with a 60× objective lens, satisfying the Nyquist condition for up to 100-nm resolution. The software package MetaMorph (Molecular Devices, Sunnyvale, CA) was used to control the system and for acquisition of the image.

Cell culture and transfection

The cells were obtained from Health Science Research Resources Bank (Osaka, Japan) and maintained at 37°C in DMEM with 10% fetal bovine serum (Life Technologies; Thermo Fisher Scientific, Waltham, MA) at 5% CO₂. The cells were grown on glass coverslips (No. 1S; Matsunami, Osaka, Japan) for immunofluorescence staining. For live imaging, cells were transfected with TransFectin (Bio-Rad, Hercules, CA) and plated onto a glass-bottom dish (No. 1.5; MatTek, Ashland, MA). The medium was changed to L15 medium with 10% fetal bovine serum but without phenol red (Life Technologies) before imaging.

Immunofluorescence staining

For microtubule staining (Figure 3), the cells were fixed with 2% paraformaldehyde and 0.1% glutaraldehyde in 0.1 M cacodylate buffer and 4% sucrose, followed by permeabilization with 0.5% Triton X-100. Nonspecific binding was blocked with 2% bovine serum albumin (BSA, Fraction V; Roche, Schweiz, Switzerland) before staining with anti–β-tubulin antibody (MAB3408; EMD-Millipore, Billerica, MA) and Alexa Fluor 488–labeled secondary antibody (Life Technologies). For mitochondrial staining (Figure 7), the cells were fixed with cold methanol (−20°C) and stained with anti-TOMM20 antibody (ab56783; Abcam, Cambridge, MA), followed by Alexa Fluor 488–labeled secondary antibody. The stained cells were mounted with ProLong Gold (Life Technologies).

STED imaging

For confirmation of the subdiffraction structures, some samples were observed with an STED microscope system (SP8-STED 3X; Leica Microsystems, Wetzlar, Germany) with the following settings. The acousto-optic tunable filter was 3% for 480 nm (excitation) and 95% for 592 nm (depletion). The gate for the detector was opened from 1.5 to 6 ns after the excitation pulse. The sample was scanned at 400 Hz with 8× line averaging. The pinhole was closed to 68 μm (0.5 Airy unit at 520 nm).

FOOTNOTES

ACKNOWLEDGMENTS

REFERENCES

• Abbe E (1873). Beiträge zur Theorie des Mikroskops und der mikroskopischen Wahrnehmung. Arch Mikroskop Anat 9, 413–418. Crossref, Google Scholar
• Allen RD, Travis JL, Hayden JH, Allen NS, Breuer AC, Lewis LJ (1982). Cytoplasmic transport: moving ultrastructural elements common to many cell types revealed by video-enhanced microscopy. Cold Spring Harb Symp Quant Biol 46, 85–87. Crossref, Medline, Google Scholar
• Betzig E, Patterson GH, Sougrat R, Lindwasser OW, Olenych S, Bonifacino JS, Davidson MW, Lippincott-Schwartz J, Hess HF (2006). Imaging intracellular fluorescent proteins at nanometer resolution. Science 313, 1642–1645. Crossref, Medline, Google Scholar
• Heintzmann R, Cremer CG (1999). Laterally modulated excitation microscopy: improvement of resolution by using a diffraction grating. Proc SPIE 3568, 185 –196. Google Scholar
• Hell SW, Wichmann J (1994). Breaking the diffraction resolution limit by stimulated emission: stimulated-emission-depletion fluorescence microscopy. Opt Lett 19, 780. Crossref, Medline, Google Scholar
• Huang B, Bates M, Zhuang X (2009). Super-resolution fluorescence microscopy. Annu Rev Biochem 78, 993–1016. Crossref, Medline, Google Scholar
• Kner P, Chhun BB, Griffis ER, Winoto L, Gustafsson MGL (2009). Super-resolution video microscopy of live cells by structured illumination. Nat Methods 6, 339–342. Crossref, Medline, Google Scholar
• Neuspiel M, Schauss AC, Braschi E, Zunino R, Rippstein P, Rachubinski RA, Andrade-Navarro MA, McBride HM (2008). Cargo-selected transport from the mitochondria to peroxisomes is mediated by vesicular carriers. Curr Biol 18, 102–108. Crossref, Medline, Google Scholar
• Rust MJ, Bates M, Zhuang X (2006). Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM). Nat Methods 3, 793–796. Crossref, Medline, Google Scholar
• Sheppard CJR (1988). Super-resolution in confocal imaging. Optik (Stuttg) 80, 53–54. Google Scholar
• Shimozawa T, Yamagata K, Kondo T, Hayashi S, Shitamukai A, Konno D, Matsuzaki F, Takayama J, Onami S, Nakayama H et al. (2013). Improving spinning disk confocal microscopy by preventing pinhole cross-talk for intravital imaging. Proc Natl Acad Sci USA 110, 3399–3404. Crossref, Medline, Google Scholar
• Südhof TC (2013). Neurotransmitter release: the last millisecond in the life of a synaptic vesicle. Neuron 80, 675–690. Crossref, Medline, Google Scholar
• York AG, Chandris P, Nogare DD, Head J, Wawrzusin P, Fischer RS, Chitnis A, Shroff H (2013). Instant super-resolution imaging in live cells and embryos via analog image processing. Nat Methods 10, 1122–1126. Crossref, Medline, Google Scholar
• York AG, Parekh SH, Dalle Nogare D, Fischer RS, Temprine K, Mione M, Chitnis AB, Combs Ca, Shroff H (2012). Resolution doubling in live, multicellular organisms via multifocal structured illumination microscopy. Nat Methods 9, 749–754. Crossref, Medline, Google Scholar