Detecting structural heterogeneity in single-molecule localization microscopy data

Abstract

For decades, the resolution of fluorescent light microscopy has been bounded by Abbe’s diffraction limit to λ/2NA. Super-resolution methods, awarded with the 2014 Nobel Prize in Chemistry, use tricks to overcome this limit. The general idea is to image blinking fluorophores for a multitude of frames, such that each frame only contains a sparse subset of fluorophores. Assuming that single emitters give rise to a sparse subset of diffraction-limited spots, their locations can be determined with nanometer precision. The resolution of the final reconstructed image is limited by the localization precision and incomplete fluorescent labeling. To even further improve the resolution, by increasing the signal-to-noise and overcoming the problem of a low label- ing density, single-particle averaging can be used if multiple copies of the same target particle (e.g. macro- molecular complex) can be imaged. All emerging localization patterns are computationally merged into one super-superresolution image. Despite the increase in resolution, potential structural variation among the particles will blur the particle fusion result and possible (small) subsets of structurally different particles can- not be detected in the reconstruction. We present an a-priori knowledge-free, unsupervised classification method that splits the dataset into conformationally different groups of images prior to the merging process, which can subsequently be fused per class. The implemented algorithms are validated on multiple exper- imental and simulated datasets. We achieved classification performances of 96% on experimental datasets with up to four different DNA origami structures, are able to detect rare classes of mirrored origami’s occur- ring at a rate of 2%, and capture the variation in the ellipticity of nuclear pore complexes. This new classifica- tion tool will allow microscopists to study heterogeneous samples with single-particle averaging techniques and discriminate between different particles, structures or conformations with a high resolution.