Localization STED (LocSTED) microscopy with 15 nm resolution

2.1 Experimental

LocSTED imaging was performed on a homemade CW STED microscope using DPSS lasers as shown in Figure S1. Alexa 555 fluorophores were excited using 532 nm CW laser light (Verdi-V5, Coherent, Santa Clara, CA, USA) and depleted using 660 nm CW laser light (Laser Quantum Opus, Germany), and they were imaged with a Leica objective lens (HC PL APO 63X, NA=1.30 GLYC, Leica Microsystems, Germany). A vortex phase plate (RPC Photonics, Rochester, NY, USA) was placed in the STED beam path in order to obtain a 0–2π vortex phase pattern. The STED and excitation beams were overlapped and aligned using two dichroic mirrors. The fluorescence signal was collected using a bandpass emission filter (605/70, Chroma Technology Corporation, USA) and detected using a single-photon counting avalanche photodiode detector (APD; Micro Photon Devices s.r.l, Italy). The pixel size for sample-stage scanning was set to 2 nm for LocSTED imaging. The pixel dwell times varied from 1 to 10 ms for different measurements. Scanning was performed using a three-axis XYZ piezo stage with a travel range of 100×100×100 μm (P-611.3S Nanocube, Physik Instrumente, Germany) and a position accuracy of 1 nm. The piezo stage was driven by a position controller (E-664 LVPZT position servo controller, Physik Instrumente, Germany). More details on the preparation of samples and imaging buffer are given in Supplementary Information.

2.2 LocSTED data analysis

The images were analyzed using Fiji and MATLAB. A series of images showing individual fluorescence events temporally switched ON and OFF during LocSTED imaging were recorded and then analyzed using a single-molecule localization algorithm custom-written in MATLAB. The final image was reconstructed from a sequence of 100–200 image frames with 50×50 pixels, each pixel with 2×2 nm² in size, and typical pixel dwell times from 1 to 10 ms. Typically, each frame contains sparse traces of blinking fluorophores (single lines) along the fast axis. The temporal blinking properties strongly depend on two parameters, namely on the ROXS buffer and the STED laser power. For image reconstruction, the obtained signals are processed in the following way:

Thresholding: The blinking events in each frame are preselected by applying an image threshold. Candidate pixels are by default pixels whose magnitudes are greater than the average brightness of the frame plus 2.5 times its standard deviation. The pixel candidates that are selected are saved and the ones below the threshold intensity are rejected. In every frame, each line with signals above the threshold is then fitted using a 1D Gaussian model.

Fitting: The candidate pixel’s local environment, which is determined by a 14×1 pixel window centered at the intensity maximum, is fitted using a sum of two Gaussian functions in the x-direction. All fitted frames are added to each other, thus creating an intensity map of localizations. The 1D Gaussian fitting is a preprocessing tool for generating a smoothened binary mask, which accounts for all pixels above the threshold.

Binary image: The sum image of all fitted 1D Gaussians is transformed into a black and white (binary) image. In this case, all Gaussian-fitted traces are summed up. Although Gaussian fitting is done on the original data, again only the Gaussian-fitted areas with intensities exceeding the image threshold are selected. The threshold for the transformation is calculated with inter/intra variance of image segments, which is based on Otsu’s algorithm [34]. Pixel intensities within an image segment must be close to each other while they must be well separated from different image segments.

Image cleaning: The binary image is subjected to cleaning of single white pixels or single black pixels in a white region using morphological operations such as dilation followed by erosion [35]. The image subjection is performed well and combines the sometimes disrupted regions, which most probably belong to one single emitter detected in various frames, into one connected region describing a single emitter.

Region extraction: The fully connected regions in the binary image are then extracted using a connectivity of eight pixel neighbors. The extracted fully connected regions create a logical mask, which is finally applied to the sum of the experimental frames. This mask cuts out regions from the sum of experimental data that can be considered as candidates for the cumulative signal of a single blinking emitter.

Characterization of the peaks: For each region, the maximum brightness, centroid, and standard deviation are calculated.

3.1 Imaging of nanoanchors

In order to estimate the resolution enhancement and localization precision achievable with LocSTED microscopy, well-defined test systems were chosen. As a first model system, we applied LocSTED microscopy to look at biofunctionalized nanoanchors of 56 nm in diameter, where Alexa 555 molecules were selectively anchored (see Figure 2A). The nanoanchors were written with STED lithography and biofunctionalized using laser-assisted protein adsorption by photobleaching (LAPAP) [30]. The area where the Alexa 555 can be anchored is limited to the surface of the nanoanchors. However, the total number of molecules per nanoanchor cannot be determined with CW STED or STORM alone, because the resolution of either of them is not sufficient. Figure 2B shows an SEM image of a nanoanchor which is 56 nm in diameter and a STED image obtained using CW excitation with 532 nm (excitation power: 10 μW) and CW depletion with 660 nm (depletion power: 30 mW). In this case, the corresponding full width at half-maximum (FWHM) of the effective PSF is 65 ± 5 nm.

Figure 2:

Nanoanchors written using STED lithography and biofunctionalized using LAPAP.

(A) Nanoanchors may carrry one or two Alexa 555 tagged Streptavidins, anchored via biotinylated Atto 390 which itself is tagged to the STED-lithographically written nanoanchors via LAPAP. (B) left: SEM image of a nanoanchor with a diameter of 56 nm, center: STED image of a nanoanchor and (right) its corresponding optical FWHM.

In LocSTED microscopy, the scanning area is limited to the donut minimum and measures 100×100 nm² (Figure 3A) similar to MINFIELD-STED [18]. All the molecules within the central minimum – no matter whether it is a single molecule or there are multiple molecules that are lying too close to each other – would appear as an ensemble within the resolution limit of the CW STED microscope of ~60 nm. We utilize stochastic blinking and localize the single molecules to find their actual positions on the nanoanchors and hence achieve a better resolution. Each nanoanchor is individually focused and scanned with an excitation power of 10 μW, first with confocal and then with LocSTED, in a subdiffractional region of interest (ROI) measuring 100×100 nm². The pixel size is 2×2 nm² and the pixel dwell time varies from 1 to 10 ms for all measurements. The molecules appeared to be in the ON state stochastically in some of the frames and completely in the OFF states in others. Over the course of time, each molecule undergoes multiple ON/OFF switching cycles. Numerous blinking events across time t are recorded (Figure 3B).

Figure 3:

Imaging strategy in LocSTED microscopy.

(A) The scanning field in LocSTED microscopy is limited to a subdiffractional extent. This corresponds to the donut central minimum of 100×100 nm² where the depletion beam intenstiy is moderate. Thus, the molecules are not affected by the maximum of the depletion beam, which further helps to reduce photobleaching. A line scan along a blinking event shows that the ON state typically lasts for ton=10 ms. (B) Examples of frames of raw images that reveal blinking events. Typically, 100=200 images are recorded. All scale bars: 25 nm.

In STED microscopy, the photobleaching comes mainly from the high intensity of the depletion beam. Photobleaching of Alexa 555 was significantly controlled using ROXS [23], [24], [25], [31], which was developed in this case by combining ascorbic acid and methyl viologen together with an enzymatic oxygen scavenging system consisting of glucose oxidase, catalase, and glucose [21]. ROXS helps to depopulate the triplet state, which greatly helps to control the photoblinking and to reduce the photobleaching. The donut is used to deplete the molecule (by stimulated emission and by manipulating the OFF time), and hence the position is estimated within the donut minimum. It was seen that the STED beam, in addition to inducing stimulated emission, was also having an effect on the photoblinking [32], [33] of the molecules, as shown in Figure S2. This improves quenching of the effective fluorescence via the STED beam, in addition to stimulated emission. The power-dependent nonlinear behavior of STED with ROXS also shows that the saturation intensity of the fluorophore can be reduced, which means that the STED power required to achieve the desired resolution can be significantly lowered as shown in Figure S3.

The localization of the molecular coordinates from all the image frames, where blinking events appeared, was obtained as described above. All the blinking events are summed up over time t and the final image is reconstructed. The reconstruction of the localized single emitters strongly depends on their stochastic blinking properties and fluorophore bleaching. Figure 4A–C shows the reconstructed images with the sum of all blinking events (left panels), their Gaussian reconstructions (center panels), and line-cuts through the 2D images (right graphs). The average localization errors are noted above the panels. It was found that, in total, ~56% of the nanoanchors show only one localization event, which can be interpreted that, in most of the cases, only one streptavidin was attached to a nanoanchor (sketch in the right graph). About 39% of the imaged nanoanchors show two localization events, spaced by 16 nm, which hints at two streptavidin per nanoanchor. Three localization events were shown by 5% of the nanoanchors. We note that even in this case, the two Gaussians were easily resolved, and the true resolution was probably much better, close to 10 nm. The fact that we did not observe spacing below 15 nm could be explained by the fact that some minimum distance is required between two streptavidin molecules because of steric hindrance or electrostatic repulsion. This, however, does not exclude the possibility that one streptavidin is doubly labeled with two Alexa 555 molecules, which then, however, would lead to a distance less than 10 nm (see below). Several more examples are shown in Figure S4. We also carried out simulations by feeding randomly distributed blinking events to the software. In those cases, no artificial clustering was observed by the algorithm (Figure S5). The temporal blinking properties strongly depend on two parameters, namely on the ROXS buffer and the STED laser power.

3.2 Blinking analysis

Once the single molecules are localized, it is also possible to extract the underlying information by having a closer look at the blinking events individually and analyzing the duration of ON-OFF times, their intensity, or their distribution. This will further help to quantify the exact number of fluorophores in a particular localized area. Figure 5A shows a case where a localization event appeared on a nanoanchor, which in a first approximation could be assumed to be a single molecule. In Figure 5B, the total fluorescence intensity is plotted with respect to the individual lines from the sum image of all the blinking events. The numbering of the lines is indicated in the left panel of Figure 5A. We take the seventh line as an example since it is the most intense. It actually is twice as intense as other events. It turns out that line 7 appears in frame numbers 61 and 83. In frame 61, the blinking event lasted for ~25 ms (orange line in Figure 5C). The event in frame 83 lasted for only 10 ms (blue line) and appeared abruptly while scanning at around 50 ms. The Gaussian reconstruction also shows an unusually broad FWHM of 24 nm (see Figure 5A, right panel). These observations point toward the presence of two fluorophores at the same place, which means a doubly stained streptavidin is observed. Both fluorophores are in the ON state and get recorded at the same time in frame 61. Only one is in the ON state in frame 83 and it also switches on abruptly during the line scan.

Figure 4:

Number of molecules localized on the nanoanchors.

The reconstructed images with the sum of all the blinking events (left panels, the white dashed circles are guides to the eye) and their corresponding Gaussian reconstructions (center panels). The right panels show line scans through the Gaussian reconstructed images. Three different cases are shown: (A) Nanoanchor with a single localization event. The corresponding FWHM of 12.8 nm yields a positioning accuracy of ~5 nm. (B) Two localizations with a separation of 16 nm and (C) three localizations, where the lower left two are spaced by 18 nm, indicated by the arrow. All scale bars: 25 nm.

Figure 5:

Study of individual blinking events of localized molecules.

(A) Sum of all the blinking events (the white dashed circle is a guide to the eye), their Gaussian reconstruction and the corresponding line profile. (B) Total fluorescence intensity per each line. Line numbers are indicated in the left panel of (A). (C) Total fluorescence intensities plotted in detail for the 7th line, which appears bright in frames 61 and 83. The signal is ON for a longer duration of ~25 ms in frame 61 (orange line) and the total fluorescence intensity is double the intensity compared to that in frame 83 (blue line) which also has a shorter duration of ~10 ms.

Simulation was also carried out for localized single molecules when the ON probability is changed, as shown in Figure S6. Figure S7 shows the simulation results with simulated blinking events using different spatial orientations and distances.

The localization error is obtained from centroid calculations of each region, which are the candidates for a full signal (the cumulative signal of a single blinking emitter). The blinking rate is adjusted as best as possible via the ROXS buffer to have the blinking emitters very sparsely distributed within individual frames. However, since blinking is a statistical process, it cannot be excluded that a single emitter is turned on and off repeatedly within one line. Multiple emitters can be rarely observed. Combining multiple times resolved line scans of the same molecule shows a cumulative intensity distribution. Blinking depends also on the intensity of the STED beam, and the ON probability is higher when a fluorophore is in the center of the donut. This increases the resolution. The derived position accuracy for each line would correspond to a temporal fingerprint, but it is not the case if we accumulate all the scans. Figure S8 shows simulated results of the average localization errors obtained in both X- and Y-directions, which are the fast and slow axes, respectively.

3.3 Imaging of DNA origami nanorulers

The second test system we studied was DNA origami nanorulers that carry fluorescent markers on both ends, as shown in Figure 6A. The center-to-center distance between the marks was approximately 50 nm. The DNA origamis carry four biotin molecules and were immobilized on the glass substrate with the help of a BSA-biotin surface incubated with neutravidin solution. Figure 6B shows the CW-STED image of a DNA origami. Only one unresolved, diffuse feature of ~60 nm FWHM is observed. Figure 6C shows the LocSTED images of a DNA origami where the two dye-tagged ends of the nanoruler are clearly resolved and the distance between them is 50 nm. The sum of all blinking events is displayed in the left panel, and the Gaussian reconstructions are displayed on the right panel.

Figure 6:

LocSTED microscopy on a DNA origami nanoruler.

(A) Schematic of the DNA origami nanoruler which is immobilized on a BSA-biotin coated surface, incubated with Neutravidin. (B) The CW-STED image of a DNA origami nanoruler shows one blurred image with a FWHM of ~60 nm. (C) LocSTED image of a DNA origami nanoruler. Left panel: sum of all the blinking events; right panel: corresponding Gaussian reconstructions. A distance of 50 nm is obtained between the two labeled ends. Note the two different scale bars of 200 and 25 nm in (B) and (C), respectively.

Figure 7A shows a detailed geometry of a DNA origami. They were fluorescently labeled with Alexa 555 at both ends, and the horizontal center-to-center distance between the two bunches of binding sites for fluorophores was 50 nm. However, the left end actually carries a bunch of eight binding sites spanning an area of 7 nm×15 nm and the right end carries a bunch of eight binding sites spanning an area of 8 nm×15 nm, approximately. Hence the expected distance between the two fluorescent spots may vary from slightly less than 50 nm up to approximately 60 nm.

Figure 7:

Detailed geometry of a DNA origami nanoruler (A) and a statistical distribution of the distances measured (B).

Figure 7B shows a statistical distribution of all the distances that were measured using LocSTED microscopy. Indeed, most of the measured distances are between 50 and 60 nm as expected from the geometry. Distances shorter than 50 nm can be explained by origamis where the innermost binding sites are fluorescently tagged (resulting in spacings down to 50 – 4 – 3.5=42.5 nm) or from an improper horizontal orientation such as tilt or bending of the DNA origamis. Distances above 60 nm could be due to partially unfolded origamis. More examples are shown in Figure S9.