Multiple signal classification algorithm for super-resolution fluorescence microscopy

Abstract

Single-molecule localization techniques are restricted by long acquisition and computational times, or the need of special fluorophores or biologically toxic photochemical environments. Here we propose a statistical super-resolution technique of wide-field fluorescence microscopy we call the multiple signal classification algorithm which has several advantages. It provides resolution down to at least 50 nm, requires fewer frames and lower excitation power and works even at high fluorophore concentrations. Further, it works with any fluorophore that exhibits blinking on the timescale of the recording. The multiple signal classification algorithm shows comparable or better performance in comparison with single-molecule localization techniques and four contemporary statistical super-resolution methods for experiments of in vitro actin filaments and other independently acquired experimental data sets. We also demonstrate super-resolution at timescales of 245 ms (using 49 frames acquired at 200 frames per second) in samples of live-cell microtubules and live-cell actin filaments imaged without imaging buffers.

Figure 1. Results for regions A–C of in vitro sample 1.

The mean image of 10,000 frames is shown in a. (b,c) The mean image and MUSICAL image for region A. (d) Shows the profile of section A1 shown in c. (e–g) The mean image, STORM image, and MUSICAL image for region B. (h–j) show the mean image, STORM image and MUSICAL image for region C. (k) shows the profile of section C1 shown in i,j. (l) shows the average PSFs derived from groups AA and BB of line segments shown in c,g, respectively. Scale bars: 1 μm (a); 250 nm (b,c,h–j); and 500 nm (e–g).

Figure 2. MUSICAL results for in vitro sample 2.

The first, second and third columns correspond to the image stacks acquired using laser powers 10.3 W cm⁻² (a,e,i,m,q), 40.2 W cm⁻² (b,f,j,n,r) and 205.6 W cm⁻² (c,g,k,o,s), respectively. The fourth column (d,h,l,p,t) shows result of MUSICAL using 49 frames only from the image stack acquired with 205.6 W cm⁻² excitation power. (q–s) Compares the intensity profiles of mean image and MUSICAL at the sections E1, E2 and E3, respectively. (t) Compares MUSICAL result at E3 using entire stack of 10,000 images with MUSICAL result at F3 using 49 frames only. Scale bars: 2 μm (first and second rows); and 1 μm (third and fourth rows).

Figure 3. Results on Data-SMLM data sets and synthetic example SynSTORM for comparison of MUSICAL with SMLM techniques.

(a) Tubulins high density (500 frames), (b) tubulins long sequence (15,000 frames) and (c) TubulinAF647 (9,990 frames). PALM and STORM results in b are taken from. In b, MUSICAL image (in green) is overlaid with PALM image (in magenta) and STORM image (in magenta), respectively, for comparison. Similarly, MUSICAL image (in magenta) is overlaid with STORM image (in green) in c.

Figure 4. Comparison of MUSICAL with other super-resolution methods that perform statistical analysis of blinking statistics rather than SMLM.

(a) shows comparison for region A of in vitro sample 1. Overlay of MUSICAL image with 3B and deconSTORM is given for comparison in addition to their individual results. (b) Comparison of MUSICAL results with 3B and deconSTORM for the tubulins high-density data of Data-SMLM. MUSICAL can resolve two closely lying microtubules (cyan coloured arrow) which are unresolvable by 3B and deconSTORM. (c) Comparison results for synthetic example SynEx2 which is noise-free. (d) Shows comparison results for synthetic example SynEx2 with Poisson noise statistics and SBR ratio 3. Includes the computation time for each method below its result.

Figure 5. Result of MUSICAL for live-cell microtubules sample 1.

Mean image of 49 frames is shown in a. Mean image and MUSICAL result for region A identified in a are shown in b,c. Mean image and MUSICAL result for region B identified in a are shown in d,e. Zoom-in of region C identified in e is shown in f. Scale bar, 200 nm. Sub-figure g shows the closest maxima from points along the cyan line in f in the directions n₁ (blue) and n₂ (red).