High-density volumetric super-resolution microscopy

Abstract

Volumetric super-resolution microscopy typically encodes the 3D position of single-molecule fluorescence into a 2D image by changing the shape of the point spread function (PSF) as a function of depth. However, the resulting large and complex PSF spatial footprints reduce biological throughput and applicability by requiring lower labeling densities to avoid overlapping fluorescent signals. We quantitatively compare the density dependence of single-molecule light field microscopy (SMLFM) to other 3D PSFs (astigmatism, double helix and tetrapod) showing that SMLFM enables an order-of-magnitude speed improvement compared to the double helix PSF by resolving overlapping emitters through parallax. We demonstrate this optical robustness experimentally with high accuracy ( > 99.2 ± 0.1%, 0.1 locs μm^-2) and sensitivity ( > 86.6 ± 0.9%, 0.1 locs μm^-2) through whole-cell (scan-free) imaging and tracking of single membrane proteins in live primary B cells. We also exemplify high-density volumetric imaging (0.15 locs μm^-2) in dense cytosolic tubulin datasets.

Fig. 1. Encoding the 3D position of single-molecule fluorescence into a 2D image.

ai The standard (2D) point spread function (PSF) can be modified to encode 3D position by phase modulation in the back focal plane (BFP, indicated by gray level) of a widefield microscope. aii Key 3D SMLM techniques include astigmatism, the double helix PSF, and the tetrapod PSF, shown here in an 8 × 8 μm² field-of-view (scale bars are 1 μm). aiii The associated PSF footprints integrated over their entire axial range (color-coded by depth). The loss in lateral resolution at the expense of axial range leads to a lower signal-to-noise ratio. b Schematic of the microlens array used in this SMLFM platform, the PSF in the central perspective view, and the PSF footprint integrated over the entire 8 μm axial range (color-coded by depth). A simplified optical diagram of SMLFM is also shown on the right, where OBJ = objective, TL = tube lens, FL = Fourier lens, and MLA = microlens array. Optical diagrams for all the 3D techniques discussed herein can be found in Supplementary Fig. 2. Pixel size is 110 nm for standard, astigmatism and the tetrapod PSF, and 266 nm for the DHPSF and SMLFM to reflect experimental parameters.

Fig. 2. SMLFM consistently outperforms other 3D-SMLM techniques at correctly identifying and reconstructing single emitters at increasing densities.

a Representative snapshots of simulated raw localization data (100 frames, n = 3) in a 10 × 10 μm² zoomed region for each imaging modality discussed herein (2D, astigmatism, double helix PSF, light field [central view] and tetrapod PSF). The scale bar represents 2 μm. DoF indicates the depth of field achieved by each technique. b Top: Average positive predictive value (PPV) curves for each SMLM technique as a function of emitter density (ρ_loc) at 4000 detected photons, where PPV refers to the number of true positive localizations vs. total number of fitted localizations. Bottom: Average sensitivity curves as a function of ρ_loc at 4000 detected photons, where sensitivity refers to the number of true positive localizations vs. the total number of ground-truth localizations. Light and dark-shaded regions represent the first and second standard deviations from the mean over three repeats of 100-frame simulated datasets. Example simulated data are presented in Supplementary Movies 1 and 2.

Fig. 3. Scan-free SMLFM-STORM imaging of B-cell receptors over whole primary mouse B cells.

a Representative frame of SMLFM localization data showing individual membrane receptors through 7 perspective views in a hexagonal arrangement (total of 50,000 frames per cell, where n = 3 cells). The expanded insert shows 7 fluorescent puncta (AlexaFluor 647) in the central perspective view and two fiducial markers indicated by arrows. Directly below is an illustration of the cell being imaged. b Associated 3D reconstruction of the whole primary mouse B cell (40,000 3D localizations in a 9 μm³ box), c an xy projection, and d a 1 μm thick central clipping to illustrate non-internalization of dye molecules. e Median lateral and axial localization precision (fitting error) for localizations below 60 nm resolution as a function of the number of views used to reconstruct a 3D localization (shading represents interquartile range). f Proportion of 3D localizations below 60 nm lateral precision as a function of the number of views used to reconstruct in 3D. A total number of 3D localizations per view is represented by dots (where n = 3 cells), with the mean value represented by each bar. Experimental data and the accompanying real-time 3D reconstruction are presented in Supplementary Movie 3.

Fig. 4. Scan-free whole-cell 3D SPT of the B-cell receptor on primary mouse B-cell membranes using SMLFM.

a 3D trajectory map of the BCR over a whole primary mouse B-cell totaling 614 tracks color-coded by diffusion coefficient using maximum likelihood estimation. Example isolated trajectories of varying diffusion coefficients are expanded directly below. Histograms showing b individual diffusion coefficients, and c track lengths, from the cell presented in a (bin widths were determined using Freedman-Diaconis' rule). d Diffusion coefficient of the BCR measured by FCS (n = 7 cells) and SMLFM-SPT (n = 5 cells), and the minimum diffusion coefficient measurable by SMLFM-SPT determined with immobilized beads (n = 9 regions). The center of each box represents the median average diffusion coefficient over n repeats (data points from repeats overlaid), with the box bounds representing the 25th and 75th percentile, and the whiskers representing maximum and minimum values. SMLFM-SPT comprises a total of 1806 trajectories over 5 cells with a mean track length of 8 points. Example experimental data and reconstructed 3D trajectories are presented in Supplementary Movie 4.

Fig. 5. dSTORM imaging of AlexaFluor 647-labeled tubulin in a HeLa cell.

a Representative frame of SMLFM localization data showing microtubule-stained HeLa cells through 7 perspective views in a hexagonal arrangement (total of 40,000 frames, imaged to completion). Two fiducial markers are indicated by arrows. b The corresponding super-resolved 3D volume, containing 173,314 localizations, color-coded by depth. c Localization rate over the first 2500 frames indicating a mean 3D localization rate of ~ 22 frame⁻¹ (blue line, rolling average over 100 frames) and an upper limit of ~ 40 frame⁻¹, corresponding to a localization density of ~ 0.075 and ~ 0.15 locs μm⁻², respectively. d Histogram of detected photons per 3D localization. e Line plots of width of 400 nm illustrate the resolution of individual microtubules as indicated by the triangle and diamond in b.