Extended-Depth 3D Super-Resolution Imaging Using Probe-Refresh STORM

Abstract

Single-molecule localization microscopy methods for super-resolution fluorescence microscopy such as STORM (stochastic optical reconstruction microscopy) are generally limited to thin three-dimensional (3D) sections (≤600 nm) because of photobleaching of molecules outside the focal plane. Although multiple focal planes may be imaged before photobleaching by focusing progressively deeper within the sample, image quality is compromised in this approach because the total number of measurable localizations is divided between detection planes. Here, we solve this problem on fixed samples by developing an imaging method that we call probe-refresh STORM (prSTORM), which allows bleached fluorophores to be straightforwardly replaced with nonbleached fluorophores. We accomplish this by immunostaining the sample with DNA-conjugated antibodies and then reading out their distribution using fluorescently-labeled DNA-reporter oligonucleotides that can be fully replaced in successive rounds of imaging. We demonstrate that prSTORM can acquire 3D images over extended depths without sacrificing the density of localizations at any given plane. We also show that prSTORM can be adapted to obtain high-quality, 3D multichannel images with extended depth that would be challenging or impossible to achieve using established probe methods.

Introduction

Cells are dense, three-dimensional (3D) collections of molecules that are highly organized on the nanometer scale. Observing this nanoscale organization is crucial to understanding how cells function, but conventional fluorescence microscopy methods are unable to resolve objects closer together than ∼250 nm because of the diffraction of light. To overcome this challenge, there has been a sustained interest in the development of super-resolution microscopy techniques that can maintain the high molecular specificity of fluorescence microscopy while allowing researchers to surpass the diffraction limit of visible light (1, 2, 3, 4). Among these, single-molecule localization microscopy (SMLM) techniques such as stochastic optical reconstruction microscopy (STORM) and photo-activated light microscopy in their typical implementations achieve a spatial resolution of ∼30 nm laterally and ∼60 nm axially (1, 3).

A range of approaches have been developed for 3D SMLM (5, 6, 7, 8, 9), but the easiest and most widespread of these uses a cylindrical lens to induce astigmatism in the detection path such that the ellipticity of single-molecule localizations can be used to determine the height of the molecules within the sample (5). However, in typical experiments in which high numerical-aperture oil-immersion objective lenses are used to image into water-based samples, substantial spherical aberration is incurred, and the distal half of the focal depth in astigmatism imaging has notably poorer spatial resolution (10). Although the aberrations can be mitigated by using adaptive optics (11) or index-matching solutions (10), these are either costly or can compromise key focus lock systems that require refractive index contrast at the coverglass-water interface to function. Instead, a simple solution is to only accept localizations from the proximal half of the focal depth and to scan this high-resolution portion of the focal plane through the sample to extend the depth range of imaging (10). Unfortunately, this focal-plane scanning process compromises the spatial resolution by dividing up the total localizations achievable among all focal planes (i.e., it lowers the localization density for any given plane), and it particularly degrades the resolution of focal planes imaged later in the process because they are partly photobleached during the imaging of other focal planes. There is thus a need for extended-depth 3D SMLM methods that are easily implemented and can overcome the combined challenges of spherical aberration and photobleaching.

A second major challenge in SMLM is a lack of dyes across the visible spectrum that have excellent photoswitching for multichannel imaging. Methods using sequential rounds of staining and bleaching partially circumvent this problem by using the excellent fluorophore Alexa Fluor 647 (or its close structural relatives, such as Cy5) multiple times, but full restaining has prohibitively long sample preparation times (12, 13, 14). A different sequential imaging strategy termed DNA-PAINT (points accumulation for imaging in nanoscale topography) generates SMLM data from transient binding and unbinding of fluorophore-labeled oligonucleotides that are complementary to oligonucleotides decorating the specimen (15, 16). Although DNA-PAINT does not require photoswitching and is therefore compatible with many fluorophores across the visible spectrum, its use of diffusing fluorescent probes creates high background signal unless low probe concentration and very low frame rates (∼20 Hz) are used (16, 17, 18). More recently, new methods to streamline these processes utilizing DNA exchange probes have been introduced, making multichannel SMLM more accessible (19, 20), although multichannel imaging continues to be a challenge for many SMLM applications.

Here, we present a method to refresh fluorescent probes on fixed specimens to enable robust, multilayer 3D STORM over extended depths without compromised performance in any focal plane. We term this approach probe-refresh STORM (prSTORM). We further demonstrate that the technique can be extended to multichannel imaging with uniform, high resolution in each channel by multiple uses of the same fluorophore targeted to different structures. Our technique utilizes refreshable DNA probes to circumvent challenges faced when imaging in three dimensions with SMLM. Importantly, prSTORM can be implemented with common superresolution microscopes with no modifications and thus has the potential to benefit the whole community of SMLM researchers.

Materials and Methods

See Supporting Materials and Methods for experimental details.

Results

We developed prSTORM by utilizing custom antibody conjugates that are covalently bound by several (3–4) single-stranded DNA (ssDNA) oligonucleotides (Table S1), each ∼20 nucleotides (nt) in length (21, 22, 23). After immunostaining the specimen and treating it with a postfixative (Table S2), we exposed the specimen to a solution of prehybridized adaptor and fluorescently labeled reporter oligonucleotides (Fig. 1). The ∼80 nt adaptor strand contains a barcode-recognition sequence (BRS) that is complementary to the ssDNA antibody on the specimen as well as three repeats of a reporter-docking sequence (RDS) that is complementary to the reporter strand. Three repeats of the RDS were included to boost signal levels. This design allowed us to modify antibodies with several ∼20 nt oligonucleotides rather than several ∼60 nt oligonucleotides to minimize any reductions of antibody affinity while still targeting a sufficient number of fluorophores to each antibody molecule. The modular use of the adaptor oligonucleotide was also convenient for multichannel prSTORM, as described further below.

Figure 1.

Workflow for prSTORM. A specimen immunostained with DNA-labeled secondary antibodies is exposed to a room-temperature (∼25°C) solution of prehybridized adaptor and fluorescent reporter DNA that is imaged by astigmatism 3D STORM and washed with hot water to remove photobleached adaptor/reporter DNA, and then a new focal plane is selected. This process is repeated many times to build up an extended-depth 3D STORM image. The adaptor oligonucleotide (inset, upper right) contains a barcode recognition sequence (BRS) that binds the antibody oligonucleotide as well as three reporter-docking sequences (RDSs) that can bind up to three fluorescently labeled reporter strands. Scale bars, 2 μm.

Once labeled, the specimen is imaged in switching buffer (a deoxygenated solution containing a thiol; see Materials and Methods) using astigmatism-based 3D STORM, the previous round’s adaptor/reporter oligonucleotides are completely removed through brief exposure (<30 s) to hot water (∼90°C), fresh (unbleached) probes are hybridized at room temperature (∼25°C, ∼10 min), and the specimen is imaged at a new focal plane. The process is repeated for many cycles, with a period of ∼30 min. Because our goal was to keep the higher resolution of ∼300 nm layer of localizations beneath the focal plane, we advanced the focal plane by moving the objective ∼300 nm per cycle (equivalent to ∼210 nm in the cell due to focal shift (10)), taking care to ensure that each layer contains overlapping localizations along the axial dimension so that each slice can be properly aligned axially. After acquiring all layers, we analyzed the separate movies, computationally corrected for spherical-aberration-induced axial localization error (10), extracted the ∼300 nm subfocal layer of localizations from each movie, and aligned all of the sections together to create a single, multilayer 3D image. Overall, we found that the DNA labels on the secondary antibodies induced only mild spatial broadening of features compared to antibody labeling alone (Fig. S1), as has also been seen in other recent work using conventional and DNA-conjugated antibodies (24, 25).

To demonstrate the ability of prSTORM to fully refresh the labels on the specimen without perturbation, we imaged the same specimen 10 times each, with and without refreshing the probes (Fig. 2, a–c; Fig. S2 a). The STORM images obtained with refreshed probes maintained the same number of localizations per round, whereas the nonrefreshed specimen decreased substantially in each round because of photobleaching. We took efforts to make the comparison as fair as possible by using 50,000 frames for each round of imaging, by omitting 405 nm activation that may be variable during the acquisition, and by replacing the switching buffer approximately every 15 min to limit potential interference caused by its decomposition over time (26). The difference between refreshed and nonrefreshed probes is even more pronounced with longer movies (data not shown). In a second experiment, we found that even after 20 cycles of hybridization and melting at ∼90°C, there was negligible loss of signal, and distortions were minimal when compared to those of the first round of imaging, in which the specimen had not been previously exposed to hot water (Fig. S3). Together, these results show that our procedure is able to efficiently refresh fluorophores an unlimited number of times, in principle, without perturbing the nanoscale structure of the specimen.

Figure 2.

Improvement of single-molecule localization density due to probe refresh steps. (a) A 3D STORM image of a BS-C-1 cell immunostained for TOM20 and imaged in a single round is shown (see also Table S2 for additional details). (b) Images of the same boxed mitochondrion in (a) across 10 sequential rounds of imaging with and without probe-refresh steps are shown as well as (c) quantification of localization density across many mitochondria in the same data set (see also Fig. S2a; the SE of the mean was <5% of all mean values and was omitted from (c) for the sake of clarity). (d) Four-layer extended-depth 3D STORM images are shown, with and without probe refresh steps, of the endoplasmic reticulum (ER) in a BS-C-1 cell expressing sec61β-GFP (green fluorescent protein) and immunostained for GFP. A higher number of localizations is achieved with probe-refresh steps i than without, regardless of the scan direction ii–iii. Arrows in iv–vi indicate ER tubules at different depths within the cell, which are only both visible when using probe-refresh steps. Zoomed-in views x–xii show that the hollowness of ER tubules is only evident when using probe-refresh steps (x). The contrast for image panels in (d) was adjusted individually to allow visualization of details in the comparably sparse nonrefresh images. (e) The number of localizations (mean ± SEM) for thick and thin regions from i to iii show that probe-refresh steps increase the number of localizations (localization numbers were normalized to the refresh data; see also Fig. S2b). Scale bars, 2 μm (a), 1 μm ((d), i–ix), 500 nm ((b) and (d), x–xii).

We next performed multilayer 3D STORM with and without probe refresh steps on the same sample (Fig. 2, d and e; Fig. S2 b). For this comparison, we used gradually increasing 405 nm laser activation to obtain the highest number of localizations possible from our sample before photobleaching. This “imaging to exhaustion” approach typically lasted ∼400,000 frames per movie. In the probe refresh case, each of four layers was imaged to exhaustion with movies of ∼400,000 frames, and we were able to reconstruct dense STORM images throughout the ∼1.2 μm thickness of the specimen. In the nonrefreshed case, we divided the ∼400,000 measurable frames among the four layers, and as expected, the nonrefreshed sample was substantially sparser than the refreshed case, particularly for regions imaged later in the process because all layers are illuminated in each round of imaging, even those out of the focal plane. Thus, 3D prSTORM is able to substantially boost localization density over extended depths for multilayer imaging without suffering from loss of localization density due to photobleaching.

We extended our use of prSTORM to 1.5 μm of the endoplasmic reticulum (ER), acquiring 200,000 frames per layer (Fig. 3) as well as 3.2 μm of the mitochondrial network (Fig. S4). Without probe refresh steps, we would be severely hindered by photobleaching, having only ∼40,000 frames or fewer available per layer, which severely limits the achievable density of localizations per layer. The 3D view of the red box (Fig. 3 b) shows the thickest region of the ER, and we see from the x-z and y-z slices that with probe refresh steps, we obtain even axial resolution in all layers. We were also able to resolve holes in the “sheet-like” structures (Fig. 3, c and d), which are consistent with a recent report that suggests ER sheets are actually dense matrices of tubules (27).

Figure 3.

Extended-depth 3D prSTORM image of the endoplasmic reticulum (ER) in COS-7 cells expressing sec61β-GFP and immunostained for GFP (see also Table S2 for additional details). (a) An x-y projection colored in z is shown. (b) A 3D-rendered perspective view (left panel) of the red box in (a) is shown with associated y-z sections (middle column) and x-z sections (right column). Section thickness, 100 nm. (c) A zoomed-in view is shown of the yellow box in (a). Arrows indicate gaps in the ER tubule matrices. (d) A conventional fluorescence image corresponding to (c) is shown. Scale bars, 2 μm (a), 1 μm (b), 500 nm (c and d). Positions are indicated relative to the origin “o” shown in (b).

We found that prSTORM was easily adapted for use with multichannel imaging (see schematic in Fig. S5). To do so, we prepared three different secondary antibodies, each functionalized with a unique ssDNA oligonucleotide, and we purchased three different ∼80 nt adaptor nucleotides with orthogonal BRSs that can hybridize with the three antibody ssDNA oligonucleotides. By using the same RDS on each of these three adaptor oligonucleotides, we were able to use the same fluorescently labeled reporter oligonucleotide for all three adapters. This modularity simplified our approach and only required the use of a single reporter oligonucleotide that was labeled with the excellent STORM fluorophore Alexa Fluor 647. Thus, each channel would be imaged with the superior fluorophore Alexa Fluor 647 rather than with, for instance, multiple spectrally independent fluorophores that inevitably have one excellent fluorophore (typically Alexa Fluor 647 is used in the channel with the most demanding structure to image at high resolution) and one or more fluorophores with notably poorer performance (28). The easily refreshable probes also allowed us to label all three targets simultaneously before the sequential labeling and imaging of each structure, which simplified image registration among multiple channels.

We demonstrate multichannel/multilayer 3D prSTORM imaging of mitochondria, microtubules, and the ER by recording a 200,000-frame movie for each structure in each layer, with 405 nm activation, spanning an axial depth of ∼800 nm (Fig. 4). A high density of localizations was achieved on each structure over the depth of the specimen, giving high-resolution images (Fig. 4, a and b), and we saw little evidence of cross talk between the channels (Fig. 4 c). In a series of control experiments, we found that our washing procedure was able to remove ≥98% of fluorescent signal from the specimen (data not shown), and we observed low cross talk (<2–5%) between any of the channels (Fig. S6). We observed that mitochondria generally localized to regions of the cell with “sheet-like” ER structures and were mostly absent from regions with only sparse ER tubules. In both the x-z and y-z slices, we observe hollow mitochondria. Additionally, we are able to ascertain axial relationships among structures that are not easily discernable from the two-dimensional projections, such as microtubules crossing just above and below ER sheets (Fig. 4, b and c).

Figure 4.

Three-channel, extended-depth 3D prSTORM images are shown of microtubules, mitochondria, and the endoplasmic reticulum (ER) in COS-7 cells expressing sec61β-GFP that were immunostained for α-tubulin, TOM20, and GFP (see also Table S2 for additional details). (a) A three-color x-y projection is shown. (b) A zoomed-in view of the box in (a) is shown with orthogonal views of the dashed cross (100 nm section thicknesses). (c) x-y projections colored in z for each structure (tubulin, ER, and mitochondria, respectively) are shown. The white circle indicates the same sites marked in (b). (d) A zoomed-in view of the box in (a) is shown. The inset shows a cross-sectional profile of the tubulin channel. Scale bars, 2 μm (a), 500 nm (b and d).

Discussion

We have described and demonstrated prSTORM, an approach for extended-depth SMLM that uses probe-refresh steps to circumvent limits otherwise caused by photobleaching of fluorophores on fixed specimens. Our procedure for refreshing the probes is simple to implement, and many cycles of melting, hybridization, and washing produced little to no detectable distortions on the sample. We showed that prSTORM extends the depth achievable by 3D STORM without sacrificing the density of localizations. Although we used astigmatism-based axial localization here, together with the collection of only the proximal half of localizations that are of higher spatial resolution (∼300 nm range) to measure specimens up to 3.2 μm thick, we note that prSTORM can extend the depth achievable with any axial localization technique (29) by eliminating limitations previously imposed by photobleaching. We also showed that prSTORM extends straightforwardly to multichannel imaging over an extended depth range with excellent performance in each of the channels and low cross talk. The application of prSTORM to the imaging of thicker specimens, including tissue sections, may face additional challenges, including more prominent spherical aberration, poorer accessibility of antibodies or oligonucleotides, and higher background signal. As a result, some changes to the protocol or compromises in resolution or speed may be necessary.

DNA-PAINT uses transient binding of diffusing probes and can also be performed for very long periods without photobleaching, so it is worth comparing its performance with that of prSTORM. In DNA-PAINT, the binding rate of reporter oligonucleotides is diffusion-limited and is controlled primarily by adjusting their concentration. Because the freely diffusing reporter oligonucleotides contribute to substantial background that can obscure the signal of a single, bound reporter oligo, the reporter oligonucleotides are typically used at relatively low concentrations. As a result, frame rates for DNA-PAINT are generally quite low (∼20 Hz), such that in a 30 min period, a maximum of ∼36,000 frames could be recorded. In contrast, photoswitching with Alexa Fluor 647 can be performed with frame rates of 100–1000 Hz. Here, with prSTORM, we recorded ∼200,000 frames in ∼17 min (at 200 Hz), with an additional ∼11 min for melting, hybridizing probes, and washing, for a total of <30 min/cycle for 5–6 times more frames. Because the maximal information content attainable is proportional to the number of frames acquired, 200,000 frames can provide more than 5 times as many localizations as 36,000 frames can provide. Scientific complementary metal-oxide semiconductor cameras are now routinely used for ∼1000 Hz STORM and would allow the prSTORM cycle time to be shortened to ∼15 min for 200,000-frame movies or would, for instance, allow 200,000-frame movies to be recorded in each of several regions of interest within 30 min. DNA-PAINT with a confocal microscope of whole cells has recently been reported, but only for a modest number of frames per focal plane (10,000–30,000), presumably because of the low frame-acquisition rate of ∼5 Hz (18). Additionally, Förster resonance energy transfer PAINT probes have been used to decrease the fluorescent background and to allow a faster frame-acquisition rate than that of standard DNA-PAINT (≤70 Hz compared to ∼20 Hz) but with some compromise of the resulting image quality (30). Overall, prSTORM can much more rapidly achieve a high-density readout than DNA-PAINT. However, there may be situations in which the potential high photon/localization capability of DNA-PAINT provides superior results, particularly for relatively sparse samples for which very high-resolution localization is desired (31).

Beyond DNA-PAINT, other methods for sequential multichannel imaging have been demonstrated. In methods based on cyclic immunofluorescence, a specimen is immunolabeled against one protein using Alexa Fluor 647 and imaged, and then the fluorescent signal is removed by photobleaching of fluorophores, chemical bleaching of fluorophores, or removal of the antibodies by salt, heat, etc (12, 13, 14). After quenching the fluorescence, the specimen is removed from the stage, immunostained against the second protein, and imaged. Each round of imaging takes at least an hour because the specimen must be incubated with primary and secondary antibodies, blocked, and quenched again (12, 13). prSTORM and other DNA-barcoding-based methods (32) present a distinct advantage because the specimen can be completely labeled with all antibodies at the benchtop, and then these labels can be read out sequentially using rapid DNA hybridization and melting steps performed on the microscope.

Although the DNA-conjugated antibodies enable prSTORM, they also face some drawbacks. Preparation of the antibodies, although documented well (19, 20, 33, 34), takes a few days and time to gain familiarity with the reagents and procedures. The DNA antibody labels also modestly broaden features (24, 25) slightly beyond the use of secondary antibody labeling alone (Fig. S1). However, this DNA-induced broadening could be reduced in the future through some combination of the following: using shorter hybridizing sequences (∼15 nt), using only two repeats of the RDS, using 3′-labeled reporter oligonucleotides (with adapters), omitting the use of the adapters and using 5′-labeled reporter oligonucleotides that bind directly to the antibody-linked oligonucleotide, or conjugating DNA to primary antibodies. We also note that, like other immunofluorescence techniques involving intracellular targets, prSTORM is not compatible with live cell imaging.

In terms of hardware, prSTORM utilizes onstage exchanges of solutions to refresh probes and thereby circumvents problems due to photobleaching. In this work, we performed these solution exchanges manually using a simple and inexpensive apparatus assembled from commercially available cell culture channels and a fixed-height aspirator that is very easy to set up with components already available in most laboratories (Fig. S7). The few necessary components can be easily brought to a central imaging facility or set up in laboratories with a STORM microscope. Although it is not necessary, the entire process could be automated using computer-controlled focal plane scanning and a computer-controlled perfusion system for solution exchanges, etc., to boost throughput and to reduce human labor.

Conclusions

We have developed and characterized prSTORM, which permits extended-depth 3D superresolution microscopy without incurring limitations resulting from photobleaching. We showed that prSTORM enables high-resolution imaging throughout multiple layers of a specimen without loss of resolution because of reduced localization density, and we also showed that prSTORM can be easily extended to multiple channels with low cross talk. The probe-refresh method would potentially also benefit other 3D SMLM techniques, such as 4Pi interferometry as well as stimulated emission depletion microscopy or structured illumination microscopy, so that a superior fluorophore or wavelength range could be used for multiple channels over extended depths.

Author Contributions

D.L., L.A.G., M.D.H., and J.C.V. designed the experiments. D.L., L.A.G., and M.D.H. performed the experiments and analysis. A.R.H. provided scripts for 3D calibration. D.L., L.A.G., M.D.H., and J.C.V. wrote the manuscript and all authors commented on the work. J.C.V. supervised the project.

Editor: Catherine Galbraith.

Supporting Citations

Reference (35) appears in Supporting Material.