We present an approach for 3D particle fusion in localization microscopy which dramatically increases signal-to-noise ratio and resolution in single particle analysis. Our method does not require a structural template, and properly handles anisotropic localization uncertainties. We demonstrate 3D particle reconstructions of the Nup107 subcomplex of the nuclear pore complex (NPC), cross-validated using multiple localization microscopy techniques, as well as two-color 3D reconstructions of the NPC, and reconstructions of DNA-origami tetrahedrons.
Main textSingle molecule localization microscopy (SMLM) is capable of resolving biological structure at the nanometer scale. However, SMLM image resolution is ultimately limited by the density of the fluorescent labels on the structure of interest and the finite precision of each localization 1, 2 . Recently, methods for obtaining higher precision localizations have been reported, which work by either increasing the number of collected photons per molecule via e.g. cryogenic imaging 3, 4 , or by introducing patterned illumination 5, 6 . The first limitation remains, however, and one approach to boosting the apparent degree of labeling (DOL) and filling in missing labels can be applied when the sample consists of many identical copies of the structure of interest (e.g. a protein complex). In this case, by combining many structures into a single "super-particle", the effective labelling density is increased, and the resulting super-particle has a high number of localizations leading to a significantly improved signal-to-noise ratio and resolution.Previous approaches to this problem can be classified as either template-based or adaptations of existing single particle analysis (SPA) algorithms originally developed for cryo-electron microscopy (EM). Template-based methods 7, 8 are computationally efficient, however, they are susceptible to template bias artefacts. Methods derived from SPA for cryo-EM have previously been adapted 9, 10 and employed to generate 3D volumes from 2D projection data. These approaches are, however, intrinsically 2D to 3D, as they assume that the raw data are projections. Recently, Shi et. al 11 also described a structure-specific method for 3D fusion, although they implicitly assume cylindrical particles and projected the volume onto top views only.