Bayesian cluster identification in single-molecule localization microscopy data

Abstract

Correspondence should be addressed to patrick. rubin-delanchy@bristol.ac.uk and dylan.owen@kcl.ac.uk Single-molecule identification-based super-resolution microscopy techniques such as photo-activated localisation microscopy (PALM) and stochastic optical reconstruction microscopy (STORM) produce pointillist data sets of molecular coordinates. While many algorithms exist for the identification and localisation of molecules from the raw image data, methods for analysing the resulting point patterns for properties such as clustering have remained relatively under-studied. Here, we present the first model-based Bayesian approach to evaluate molecular cluster assignment proposals which, in this article, are generated by analysis based on Ripley's Kfunction. The method is also the first to take full account of the individual localisation precisions calculated for each emitter. The technique is validated using simulated and experimental data from which we characterise the clustering behaviour of CD3ζ, an important subunit of the CD3-T cell receptor complex required for T cell function, in resting and activated primary human T cells.Conventional fluorescence microscopes produce images of the distribution of fluorophores in the sample convolved with the microscope Point Spread Function (PSF). Due to diffraction, this PSF typically has a width of hundreds of nanometres meaning the resulting image has a resolution, as assessed by the Rayleigh criterion, of ~200 nm. Several strategies now exist to circumvent this resolution limit 1 . Some of these, such as Stimulated Emission Depletion (STED) microscopy, rely on narrowing the excitation spot of a confocal microscope by means of a toroidal depletion beam and the process of stimulated emission 2,3 . Despite the increased resolution, these produce conventional fluorescence images, i.e., arrays of pixels with values representing the fluorescence intensity at those locations. Quantification can be performed in the same way as for conventional microscopes.Another strategy is based on Single-Molecule Localisation Microscopy (SMLM) [4][5][6][7] . This relies on the temporal separation of the excitation of fluorophores in the sample whose PSFs would otherwise overlap at the detector.The position of each fluorophore can then be estimated from the centres of the PSFs. Many algorithms are available to extract the x-y coordinates of the molecules [8][9][10] . Each emitter can be localised to a precision between 10 and 30 nm. Common strategies for the temporal separation of molecules involve intra-molecular rearrangements to switch from dark to fluorescent states or the exploitation of non-emitting molecular radicals 11,12 . These strategies are typically pursued using photoactivatable or photoconvertible fluorescent proteins or small molecule probes coupled with a reducing buffer and immunostaining protocols 13 . We refer to all such strategies as SMLM.Unlike non-pointillist microscopy methods, SMLM imaging does not produce a conventional image. Instead, the raw data is a list of the...

CTCF and cohesin regulate chromatin loop stability with distinct dynamics

“…We then divided the nucleus into partially overlapping 3 μm squares and performed clustering analysis on these squares using a recently reported Bayesian algorithm (Rubin-Delanchy et al, 2015). We used the same prior as published (Rubin-Delanchy et al, 2015) and performed cluster identification and characterized clusters according to their cluster radius and fraction of molecules in clusters as described (Rubin-Delanchy et al, 2015). …”

Stochastic fluorescence switching of nucleic acids under visible light illumination

“…In fact, numerous methods have been developed and employed to suppress or eliminate signals that emanate from endogenous molecules. These methods include chemical means for suppression of endogenous emissions and mathematical analysis that eliminates events deemed unlikely to originate from the fluorescent tag [30][31][32]. The ubiquitous use of these methods has created a widely accepted notion that endogenous molecules lack the capacity for usable fluorescence emission.…”

Minimizing Structural Bias in Single-Molecule Super-Resolution Microscopy

“…This degradation in HD SMLM is fundamentally different from low-density SMLM for which even localizations with large uncertainty provide unbiased information regarding the underlying structure. Besides thresholding, the confidence levels can systematically be used as additional input to increase the accuracy of various quantitative analyses such as Bayesian clustering 43 . Nonetheless, RoSE-C exhibits robustness in measuring the apparent labeling density w.r.t.…”