High-fidelity structured illumination microscopy by point-spread-function engineering.

Standard fluorescent sample preparation

Fluorescent microspheres of 100-nm-diameter and commercial

Argo-SIM slide were employed as standard samples to quantitatively evaluate the fidelity of reconstruction algorithms. Carboxylate-modified microspheres (0.1 µm, yellow-green fluorescent 505/515, F8803) were purchased from Thermo Fisher Scientific (MA, USA) and diluted 100 times before use. Commercially available coverslips (~150 μm thick) with 24 × 60 mm (Cellvis, USA) were carefully cleaned using the procedure in ref. 11 . A silicone sheet (GBL665201-25EA, Sigma-Aldrich, USA) with a 9-mm diameter well was attached to the coverslip. The fluorescent solution was dispensed onto the coverslip and imaged in PBS buffer. The “Star” and “2D matrix of rings” patterns in Argo-SIM were used as typical “line” and “ring” structures.

Cell culture and labeling

COS-7 and U2OS cells were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China) and cultured in an incubator at 37 °C and 5% CO 2 . The COS-7 cells were cultivated in a DMEM medium (Thermo Fisher Scientific, USA) supplemented with 1% penicillin G, streptomycin (Sangon Biotech, China), and 10% fetal bovine serum (Thermo Fisher Scientific, USA). The U2OS cells were cultivated in McCoy’s 5A medium, modified (Thermo Fisher Scientific, USA), supplemented with 1% penicillin G, streptomycin (Sangon Biotech, China), and 10% fetal bovine serum (Thermo Fisher Scientific, USA). Cells were transiently transfected using Lipofectamine 2000 (Thermo Fisher Scientific, USA) as per manufacturer’s protocol. The mEmerald-Tubulin-N-18 vector (plasmid #54293, Addgene, USA), mEGFP-Lifeact vector (plasmid #54610, Addgene, USA), mEmerald-Caveolin vector (plasmid #54025, Addgene, USA), and mEmerald-ER-3 vector (plasmid #54082, Addgene, USA) were used to label the microtubule, microfilament, caveolae, and endoplasmic reticulum, respectively. Cell vesicles were labeled by the CD63-EGFP vector, which was constructed by inserting Homo sapiens CD63 cDNAs into pEGFP-n1 vector (Clontech, USA). Twenty-four hours after transfection, the cells were detached using trypsin-EDTA (Thermo Fisher Scientific, USA), seeded onto poly-L-lysine-coated 35-mm glass-bottom dishes (Cellvis, USA), and cultured in an incubator at 37 °C and 5% CO 2 for an additional 24 h before the experiments. For live cell imaging, the complete medium was replaced by HBSS solution (Thermo Fisher Scientific, USA) containing Ca 2+ and Mg 2+ but no phenol red. For fixed cell imaging, the complete medium was removed and cells were fixed with 4% paraformaldehyde for 10 min at room temperature. After fixation, cells were washed twice by PBS buffer. Both live and fixed cells were imaged in PBS buffer.

SIM imaging

SIM experiments were performed using commercial SIM microscopes, namely DeltaVision OMX SR (GE Healthcare in Issaquah, Washington, USA) and N-SIM S (Nikon Corporation, Tokyo, Japan), as well as a custom-built two-beam interference SIM microscope (Fig. S6 ). The custom-built SIM was constructed around a commercial inverted fluorescence microscope (IX83, Olympus Life Science, Japan) with a TIRF-oil-immersion objective (UAPON 100×, NA = 1.49, Olympus Life Science, Japan). A 488-nm, 500-mW semiconductor laser (Genesis MX488-500 STM, Coherent, USA) was used for excitation, a quad-band total internal reflection (TIRF) filter block ( TRF89902 -EM, Chroma, USA) was employed for imaging, and a sCMOS camera (ORCA-Flash 4.0 V2, Hamamatsu, Japan) was used as the detector. To generate illumination patterns with different periods, a ferroelectric liquid-crystal spatial light modulator (SLM, SXGA-3DM, Fourth Dimension Displays, UK) was employed as the grating. Microspheres with 100 nm diameter, microtubules, and microfilaments in live COS-7 cells were imaged using the custom-built setup in TIRF-SIM mode. Microtubules and vesicles in live U2OS cells, and endoplasmic reticulum in fixed U2OS cells were imaged in the setup under conventional SIM mode with incident beam angle smaller than the critical angle of TIRF. Reconstruction parameters: for microspheres data, emission wavelength ( λ em ) = 515 nm while that for microtubule, microfilaments, vesicles, and endoplasmic reticulum data was 525 nm, and the single pixel size of the detector was calibrated to 65 nm/pixel. The “2D matrix of rings” and “Star” patterns in Argo-SIM, microtubules in fixed COS-7 cells, and microfilaments in fixed U2OS cells were imaged on the DeltaVision OMX SR with the parameters: NA = 1.42 (oil immersed), excitation wavelength ( λ ex ) = 488 nm, emission wavelength ( λ em ) = 527 nm, and the single pixel size of the detector was calibrated to 78.6 nm/pixel. In addition, caveolae in live U2OS cells were imaged on the N-SIM S with the parameters: NA = 1.49 (oil immersed), excitation wavelength ( λ ex ) = 488 nm, emission wavelength ( λ em ) = 525 nm, and the single pixel size of the detector was calibrated to 60 nm/pixel.

Image reconstruction Commercial

SIM reconstruction software packages, including GE SoftWoRx and Nikon NIS-Elements, and open source packages, including fairSIM, SIMToolbox, and iterative deconvolution procedures in ref. 8 , were used for comparative SR image reconstruction. Images labeled ‘GE | OMX’ were reconstructed with SoftWoRx, and the Wiener constants were 0.005 by default. Images labeled ‘Nikon | N-SIM | 2D’ were reconstructed with NIS-Elements. Images labeled ‘FairSIM’ and ‘RL-SIM’ were reconstructed with the Wiener-SIM and RL-SIM in fairSIM, respectively. The adopted OTFs were theoretical approximate OTFs (‘dampening’ factor = 0.3), and Wiener constants defaulted to 0.1. Images labeled ‘SIMToolbox | Wiener’ and ‘Map-SIM’ were reconstructed with the Wiener-SIM and Maximum a posteriori probability SIM (Map-SIM) in SIMToolbox. Images labeled ‘re-Wiener’, ‘TV-SIM’, and ‘Hessian-SIM’ were reconstructed with the Wiener-SIM, TV and Hessian denoising procedures in ref. 8 . As a comparison with HiFi-SIM, we have also implemented a traditional Wiener-SIM. In the implementation, raw data preprocessing, and reconstruction parameter estimation use the same methods as in HiFi-SIM, whereas the recombination of spectrum components follows the traditional Wiener deconvolution procedure (Supplementary Note 1 ). Images labeled ‘Wiener’ were reconstructed with the traditional Wiener-SIM implemented by us.

Quantification of the fidelity of SR images

To quantitatively evaluate the fidelity of SR images reconstructed by HiFi-SIM, two typical patterns (rings and lines) of known real structures in the Argo-SIM slide were employed as standard samples for 2D-SIM imaging (Figs. 3 and S9 ). Because structures of the samples were known (rings and lines), raw data with high modulation and high SNR were collected with the exposure time of 50 ms, and SR images with minimal artifacts were reconstructed thereafter. The residual noise in the SR image was eliminated by setting thresholds, and clean SR images were obtained as the ground-truth models (Fig. S10 ). Error maps and corresponding RMSE values, between the SR and ground-truth images, were displayed to evaluate the fidelity of reconstruction algorithms (Fig. 3c ). Furthermore, the structure similarity index measure (SSIM) was used to quantitatively evaluate the fidelity of SR images, defined as documentclass[12pt]{minimal} usepackage{amsmath} usepackage{wasysym} usepackage{amsfonts} usepackage{amssymb} usepackage{amsbsy} usepackage{mathrsfs} usepackage{upgreek} setlength{oddsidemargin}{-69pt} begin{document}$${mathrm{SSIM}}(I^{{mathrm{SR}}},I^{{mathrm{GT}}}) = frac{{(2mu _{{mathrm{SR}}}mu _{{mathrm{GT}}} + C_1)(2sigma _{{mathrm{SR}},{mathrm{GT}}} + C_2)}}{{(mu _{{mathrm{SR}}}^2 + mu _{{mathrm{GT}}}^2 + C_1)(sigma _{{mathrm{SR}}}^2 + sigma _{{mathrm{GT}}}^2 + C_2)}}$$end{document} SSIM ( I SR , I GT ) = ( 2 μ SR μ GT + C 1 ) ( 2 σ SR , GT + C 2 ) ( μ SR 2 + μ GT 2 + C 1 ) ( σ SR 2 + σ GT 2 + C 2 ) where μ SR and μ GT are the mean values of images I SR and I GT , respectively; σ SR and σ GT are the standard deviations of I SR and I GT , respectively; and σ SR , GT is the cross-variance between images I SR and I GT . C 1 and C 2 are used to avoid division by a small denominator and set as C 1 = 0.05 and C 2 = 0.05. To quantitatively analyze the influence of Wiener constants, the initial optimization wiener constant in HiFi-SIM (denoted w 1 ) was set to 0.9 and 1.2; the deconvolution wiener constants of fairSIM (denoted Wiener parameter) and HiFi-SIM (denoted w 2 ) were set to 0.005, 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, and 0.5, respectively. SoftWoRx only set a wiener constant of 0.005 for reconstruction. Ten different regions of interest (ROIs) (53 × 53 pixels), containing ring structures, from the SR images (Fig. S 10b ) were selected to calculate the SSIM values between ROI images and corresponding ground-truth images (Fig. 3f ).

Supplementary information Supplementary Information Supplementary code, test data, and User guide