Direct-laser writing
Light from a two-photon laser is used to trigger a chemical reaction that causes polymerization of a photosensitive material. The nonlinear excitation occurs at the focal point leaving other regions unaffected by the laser light. This phenomenon reduces solidification resolution to below the diffraction limit of the applied light. By selectively solidifying the photo resin, 3D geometries are produced layer-by-layer. Samples were fabricated by 3D direct-laser writing (Nanoscribe GmbH, Photonics Professional GT) using a commercially available photoresist (IP-Dip, Nanoscribe GmbH)30. The ultraviolet curable liquid resist was polymerized via two-photon absorption at a wavelength of 780 nm. The designed structures were written layer-by-layer with 100 nm slicing distance between the individual layers. Computer-aided design of the fiducial structures were produced using Autodesk fusion 360 and exported as stereolithography (.STL) files.
Cell culture, fixation, and immunostaining
COS-7, HeLa and HEK293T cells were cultured in DMEM (Life Technologies, 11885-084) supplemented with 10% FBS and 1% penicillin–streptomycin. All cell fixation and immunostaining were done directly on coverslip/chambers with prefabricated photopolymerized fiducials.
For COS-7 cells, a coverslip was attached to an 8-well chamber (ibidi 80841) and washed with PBS. Approximately 10,000 cells were added per well and incubated for 72 h at 37 °C. Cells were fixed in 4% paraformaldehyde solution (PFA, Sigma), washed using a solution of PBS and incubated with 5% BSA/PBS for 1 h to avoid nonspecific binding. Cells were permeabilized with a 0.1% solution of Triton X-100. Cells were incubated overnight at 4 °C in a 1:1000 dilution of anti-alpha-tubulin antibody (DM1A, Sigma-Aldrich), washed with PBS and incubated for 45 min with a secondary antibody labeled with Alexa Fluor Plus 647 (A32728, Life Technologies) using a 1:500 dilution.
For HeLa cells, a coverslip was attached to an 8-well chamber (ibidi 80841) and washed with DMEM. Approximately 10,000 cells were added per well and incubated for 24 h at 37 °C. The cells were labeled using a 1:50,000 dilution of CD47 (Invitrogen, B6512) conjugated to Alexa Fluor Plus 647 (A32728, Life Technologies).
HEK293T cells were deposited directly within the dish containing the microwells for the soSPIM imaging. First, DMEM media was added to the microwells by placing the chamber in a vacuum chamber for 30 min. Approximately 10,000 cells were deposited in the microwells by centrifugation at 50 × g for 1 min and incubated at 37 °C overnight. Fixation was performed using 4% PFA. Cells were permeabilized with a 0.1% solution of Triton X-100, washed with PBS and incubated with 5% BSA/PBS for 1 h. A 1:500 dilution of CD47 (Invitrogen, B6512) was added for 30 min, washed with PBS and incubated for 45 min with a secondary antibody labeled with Alexa Fluor Plus 647 (A32728, Life Technologies) using a 1:200 dilution in 5% BSA/PBS. The sample was washed using PBS, and then fixed using 4% PFA for 15 min.
Unsupervised data acquisition
To perform autonomous acquisitions, we used coverslips with arrays of fiducial structures. Fiducials were produced with a regular spacing which matched the field-of-view, thus ensuring at least one per area. The geometry and distribution of the fiducial array was added to the acquisition software which enables automatic recognition of fiducials within the array. The diffraction pattern for individual fiducials was also added to the software. The diffraction profile is a look-up table made up of the average radial profile as a function of depth. Direct-laser writing removes the large range of sizes commonly found in other fiducials (e.g., latex beads) and thus, allows for comparable axial look-up tables. Prior to determining a new fiducial’s position, we typically use an average of 10–30 diffraction profiles with the same geometries.
Initially the surface was scanned in a xy raster pattern. The software can automatically identify a fiducial once it enters the field-of-view. Initially, the xy position of a fiducial is estimated, within ~1 pixel, by using cross-correlation. Fiducials are then identified by their spacing, predicated geometry and distribution throughout the slide. After a fiducial is identified their 3D position is determined via their individual look-up table.
At each fiducial, the surface was located by iteratively performing z-stacks. In each iteration, the axial range of the stack was reduced. Once the surface was found, an axial profile of the fiducial was acquired by performing a z-stack over a pre-determined axial range by stepping the nano-positioning stage in 5 nm increments. Variations in print design and real shape of the beads is not a problem since the interference pattern of the bead in question is measured before the experiment. The rotational orientation of the array (sample chamber) was determined by calculating the distance between fiducials distributed throughout the slide. For single-molecule imaging, cellular structures or regions-of-interest (ROI) were selected and the software automatically acquired images for each location sequentially.
The accuracy of the automatic surface detection routine is 35 nm (Supplementary Fig. 5) and was determined by recording the height difference between our autonomous focusing method and the position with maximum signal of individual molecules deposited onto the coverslide. Errors from the estimated position of the surface and the surface itself can arise from a variety of factors such as surface flatness. In practice, this error can be reduced by manually (re)focusing onto the fluorescence from the sample prior to imaging. The focus-locking routine, which maintains the sample stationary, is activated after the fiducial is found on the surface and the performance is not influenced by the variations of the automatic surface detection routine.
3D focus locking
Real-time focus locking was performed by repositioning the sample using the nano positioning stage at a frequency of 20 Hz. Applying an equal but opposite amplitude suppressed mechanical movement. Focusing onto a different z-position was done by offsetting the lock-in plane across the diffraction pattern12. To change the lateral position, an offset (∂x, ∂y) is added to the lock-in position. The software automatically focus-locked onto the new location. For complex assemblies, such as the vertical wall, each individual sub-element covered an axial range. The structures were designed such that the axial profiles of each element overlap in z, but remain separate in x/y. The full focusing range is determined by the number of sub-elements and the axial range of the diffraction pattern.
Calculations were performed by using a camera-based particle tracking software implemented in LabVIEW31. We used real-time data processing in a graphics-processing unit (CUDA, NVIDIA GTX 1080). The lateral positions were determined by correlating linear intensity profiles with their mirror profiles10,32. The axial position was determined by comparing the radial intensity distribution to a pre-recorded look-up table. To enhance the signal from the diffraction patterns, we used a nonlinear dynamic fixed pattern noise correction31. This corrects for the nonlinear pixel response, read-out noise and inhomogeneous illumination. CMOS technology allows for an increased frame rate by selectively recording a ROI. The diffraction patterns from the fiducial structures were captured at ~400 Hz during the single-molecule acquisitions (10–15 Hz).
To ensure high contrast in brightfield, we selected high N.A. objectives. These were previously inspected by the manufacture using wavefront aberration technologies to ensure the lowest possible asymmetric aberration and superior optical performance required for super-resolution imaging.
The diameter of the fiducial can influence performance. To determine the best fiducial dimension for a given objective/magnification, we empirically evaluate each imaging objective using an array of spherical fiducials with an increasing range of sizes (0.5–10 µm diameter). Larger particles (>5 μm diameter) diffract more and give rise to more pronounced diffraction rings. which facilitates accurate calculations. However, larger particles may require a larger field of view and fitting the profile slows down the acquisition and/or calculations, particularly for multi-fiducial tracking.
Microscopy
soSPIM module4 was mounted on a conventional inverted microscope (Nikon Ti-E). Illumination lasers with wavelengths of 405 nm (200 mW; Vortran), 488 nm (200 mW; Coherent), 561 nm (200 mW; Vortran) and 640 nm (150 mW; Coherent) were passed through clean-up filters, combined via dichroic beam mirrors and coupled into optical fiber (Thorlabs, P3- 405BPM-FC-2). The lasers were collimated and aligned into a beam–steering system, which consists of a x1.5 beam expansion, an x and y-axis galvanometric mirror (Pangolin SCANMAX 506) and a tunable lens (EL-30-10, Optotune). These were conjugated to the back focal plane of the objective (1.27 NA, x60 CFI Nikon) so that the excitation beam was always emitted parallel to the objective optical axis4. The sample was mounted on a nano-positioning stage (Mad City Labs, LP50-200) atop a linear translation stage (ASI, LS50). The fluorescence was captured by the same objective and imaged using a CMOS camera (Hamamatsu Orca Flash 4). The transmission profile was recorded by illuminating fiducials within the wells with a 440 nm LED (CoolLED), separated from the fluorescence emission via a dichroic (Semrock Di03-R473-t3), filtered (Semrock, FF01-440/40-25) and focused (f = 300 mm) onto a CMOS camera (Allied Vision, Manta).
TIRF microscopy was achieved by displacing the laser, focused on the back focal plane of the objective, towards the periphery of the objective10. Briefly, laser lines of 405 nm (Vortran, Stradus 405–100), 488 nm (Vortran, Stradus 488–150) and 637 nm (Vortran, Stradus 637–180) were filtered, expanded ×10 and focused onto the back aperture of a 100 × 1.49 NA TIRF objective (Nikon, CFI Apochromat) using an achromatic lens (f = 200 mm), via a dichroic beamsplitter (Chroma, ZT488/640rpc). Displacement of the laser towards the edge of the objective was performed by moving the focusing lens with a mirror assembled on a translation stage (M-423-MIC; Newport). The sample was mounted on a nano-positioning stage (Mad City Labs, LP50-200) atop a micro-positioning stage (Mad City Labs, MicroStage). The fluorescence was captured by the same objective and focused (f = 400 mm) onto an EMCCD camera (Andor, iXon 897). For astigmatism based axial localization, a weak cylindrical lens (f = 500 mm) was placed before the EMCCD. The fiducial structures were illuminated with an infrared LED (Mightex LCS-0850-02-22), filtered (Semrock, FF01-842/56-25) and focused (f = 200 mm) onto a CMOS camera (Allied Vision, Manta).
Highly inclined and laminated optical sheet (HILO) microscopy6 was performed using the same optical setup used for TIRF microscopy, as detailed above. Laser beams are focused on the back aperture of the objective which is translated towards the periphery of the objective. HILO creates a light sheet, providing optical sectioning capabilities superior to widefield imaging at depths above the coverslip.
All three acquisition methods used a single objective for the delivery of the laser excitation to the sample and subsequent acquisition of the fluorescence emitted.
An environmental control box was installed to control temperature fluctuations of the microscope. A series of Peltiers (TEC3-6; Thorlabs) with temperature transducers (AD590; Thorlabs) were regulated individually (TED200C; Thorlabs). Temperature and humidity were registered independently from the temperature control units (TSP01; Thorlabs). The temperature and humidity have a standard deviation of 0.02 °C and 0.88%, respectively. Before imaging, we allowed the samples to acclimatize for ~15 min.
Direct STORM imaging
All STORM imaging was performed in a closed chamber with buffer containing reducing and oxygen-scavenging compounds. The stained cells were imaged in PBS with the addition of 50 mM mercaptoethylamine, 5% glucose (wt/vol) and oxygen-scavenging enzymes (0.5 mg ml–1 glucose oxidase (Sigma-Aldrich), and 40 mg ml–1 catalase (Roche Applied Science)33. The ensemble fluorescence was converted into the desired density of single molecules by using the 640 nm laser at high power. For TIRF imaging, reduction was performed by briefly placing the setup into HILO and performing a z-stack with high laser power. For soSPIM imaging, we increased the power delivered via the light sheet and performed a z-stack prior to refocusing on the plane of interest using the printed fiducials as references. Prior to reducing dyes into a dark state, a standard TIRF or soSPIM light-sheet image was acquired. For image acquisition the laser power was reduced, and a UV laser (405 nm) turned on to aid switching. Approximately 10,000–20,000 images were recorded and analyzed.
Image processing and single-molecule localization
Fluorescent images were recorded and saved in a 16-bit TIFF format. To localize molecules, the point spread function of an individual molecule within each frame was fitted to a 2D or 3D elliptical Gaussian profile. STORM images were analyzed with the Picasso software34 or ImageJ (NIH) with the ThunderSTORM plugin35. Single-particle tracking was performed using the TrackMate plugin36 in ImageJ (NIH).
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
