Article

Fast, single-molecule localization that achieves theoretically minimum uncertainty

Department of Imaging Science and Technology, Delft University of Technology, The Netherlands.
Nature Methods (Impact Factor: 25.95). 04/2010; 7(5):373-5. DOI: 10.1038/nmeth.1449
Source: PubMed

ABSTRACT We describe an iterative algorithm that converges to the maximum likelihood estimate of the position and intensity of a single fluorophore. Our technique efficiently computes and achieves the Cramér-Rao lower bound, an essential tool for parameter estimation. An implementation of the algorithm on graphics processing unit hardware achieved more than 10(5) combined fits and Cramér-Rao lower bound calculations per second, enabling real-time data analysis for super-resolution imaging and other applications.

0 Followers
 · 
99 Views
  • Source
    • "Calculating the CRLB thus provides insight into the amount of information about the bead position stored in any given image. In fluorescence microscopy, the CRLB has been computed to determine the theoretical limit for the localization of point spread functions modeled via 2D Gaussian functions (Smith et al., 2010). We have similarly applied the CRLB to the case of spherical bead localization by replacing the 2D Gaussian with a measured LUT that defines the probability distribution for each pixel in the image. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Numerous biophysical techniques such as magnetic tweezers, flow stretching assays, or tethered particle motion assays rely on the tracking of spherical beads to obtain quantitative information about the individual biomolecules to which these beads are bound. The determination of these beads' coordinates from video-based images typically forms an essential component of these techniques. Recent advances in camera technology permit the simultaneous imaging of many beads, greatly increasing the information that can be captured in a single experiment. However, computational aspects such as frame capture rates or tracking algorithms often limit the rapid determination of such beads' coordinates. Here, we present a scalable and open source software framework to accelerate bead localization calculations based on the CUDA parallel computing framework. Within this framework, we implement the Quadrant Interpolation algorithm in order to accurately and simultaneously track hundreds of beads in real time using consumer hardware. In doing so, we show that the scatter derived from the bead tracking algorithms remains close to the theoretical optimum defined by the Cramer-Rao Lower Bound. We also explore the trade-offs between processing speed, size of the region-of-interests utilized, and tracking bias, highlighting in passing a bias in tracking along the optical axis that has previously gone unreported. To demonstrate the practical application of this software, we demonstrate how its implementation on magnetic tweezers can accurately track (with ∼1 nm standard deviation) 228 DNA-tethered beads at 58 Hz. These advances will facilitate the development and use of high-throughput single-molecule approaches.
    Review of Scientific Instruments 10/2014; 85(10):103712. DOI:10.1063/1.4898178 · 1.58 Impact Factor
  • Source
    • "CBC-mediated clustering images (Fig. 3d, e, Fig. S2) in which localizations with a clustering value above threshold are displayed in green and those below threshold in red show that rejected localizations mostly either form a cloud-like pattern throughout the cell indicative of single or spurious localizations or can be found at the edges of higher intensity clustered areas, thereby defining the border of clusters. after application of the CBC-mediated clustering filter to all images, cluster maps were extracted (Fig. 3f, g, Fig. S3) by applying a 15-nm gaussian blur filter, reflecting the finite experimental localization precision (for this reason, the clusters in the binary map, Fig. 3g, appear slightly larger than in the filtered image, Fig. 3e) (Smith et al. 2010). The area of each cohesive region was then calculated and weighted by its integrated fluorescence intensity to gain information on relative arrestin 3 molecule numbers per cluster. "
    [Show abstract] [Hide abstract]
    ABSTRACT: G protein-coupled receptor activation and desensitization leads to recruitment of arrestin proteins from cytosolic pools to the cell membrane where they form clusters difficult to characterize due to their small size and further mediate receptor internalization. We quantitatively investigated clustering of arrestin 3 induced by potent anti-HIV analogues of the chemokine RANTES after stimulation of the C-C chemokine receptor 5 using single-molecule localization-based super-resolution microscopy. We determined arrestin 3 cluster sizes and relative fractions of arrestin 3 molecules in each cluster through image-based analysis of the localization data by adapting a method originally developed for co-localization analysis from molecular coordinates. We found that only classical agonists in the set of tested ligands were able to efficiently recruit arrestin 3 to clusters mostly larger than 150 nm in size and compare our results with existing data on arrestin 2 clustering induced by the same chemokine analogues.
    Histochemie 03/2014; 142(1). DOI:10.1007/s00418-014-1206-1 · 2.93 Impact Factor
  • Source
    • "We used 'lsqcurvefi t', an existing implementation of Matlab, to do the standard 2D Gaussian fi tting. The C implementation would be much faster (about two orders of magnitude) (Smith et al., 2010). We tested on two computers with different confi gurations (Table S1). "
    [Show abstract] [Hide abstract]
    ABSTRACT: The resolution of single molecule localization imaging techniques largely depends on the precision of localization algorithms. However, the commonly used Gaussian function is not appropriate for anisotropic dipoles because it is not the true point spread function. We derived the theoretical point spread function of tilted dipoles with restricted mobility and developed an algorithm based on an artificial neural network for estimating the localization, orientation and mobility of individual dipoles. Compared with fitting-based methods, our algorithm demonstrated ultrafast speed and higher accuracy, reduced sensitivity to defocusing, strong robustness and adaptability, making it an optimal choice for both two-dimensional and three-dimensional super-resolution imaging analysis.
    Protein & Cell 06/2013; 4(8). DOI:10.1007/s13238-013-3904-1 · 2.85 Impact Factor
Show more

Preview

Download
2 Downloads
Available from