Automatic fluorescent tag localization II: Improvement in super-resolution by relative tracking

We present an algorithm for the three-dimensional (3D) tracking of multiple fluorescent subresolution tags with super-resolution in images of living cells. Recently, we described an algorithm for the automatic detection of such tags in single frames and demonstrated its potential in a biological system. The algorithm presented here adds to the tag detector a module for relative tracking of the signals between frames. As with tag detection, the main problem in relative tracking arises when signals of multiple tags interfere. We propose a novel multitemplate matching framework that exploits knowledge of the microscope point spread function to separate the intensity contribution of each tag in image regions with signal interferences. We use this intensity splitting to reconstruct a template for each tag in the source frame and a patch in the target frame, which are both free of intensity contributions from other tag signals. Tag movements between frames are then tracked by seeking, for each template-patch pair, the displacement vector providing the best signal match in terms of the sum of squared intensity differences. Because template and patch generation of tags with overlapping signals are interdependent, the matching is carried out simultaneously for all tags, and in an iterative manner. We have examined the performance of our approach using synthetic 3D data and observed a significant increase in resolution and robustness as compared with our previously described detector. It is now possible to localize and track tags separated by a distance three times smaller than the Rayleigh limit with a relative positional accuracy of better than 50 nm. We have applied the new tracking system to extract metaphase trajectories of fluorescently tagged chromosomes relative to the spindle poles in budding yeast.

Automatic fluorescent tag detection in 3D with super-resolution: application to the analysis of chromosome movement

In this paper, we describe an algorithmic framework for the automatic detection of diffraction-limited fluorescent spots in 3D optical images at a separation below the Rayleigh limit, i.e. with super-resolution. We demonstrate the potential of super-resolution detection by tracking fluorescently tagged chromosomes during mitosis in budding yeast. Our biological objective is to identify and analyse the proteins responsible for the generation of tensile force during chromosome segregation. Dynamic measurements in living cells are made possible by green fluorescent protein (GFP)-tagging chromosomes and spindle pole bodies to generate cells carrying four fluorescent spots, and observe the motion of the spots over time using 3D-fluorescence microscopy. The central problem in spot detection arises with the partial or complete overlap of spots when tagged objects are separated by distances below the resolution of the optics. To detect multiple spots under these conditions, a set of candidate mixture models is built, and the best candidate is selected from the set based on chi2-statistics of the residuals in least-square fits of the models to the image data. Even with images having a signal-to-noise ratio (SNR) as low as 5-10, we are able to increase the resolution two-fold below the Rayleigh limit. In images with a SNR of 5-10, the accuracy with which isolated tags can be localized is less than 5 nm. For two tags separated by less than the Rayleigh limit, the localization accuracy is found to be between 10 and 20 nm, depending on the effective point-to-point distance. This indicates the intimate relationship between resolution and localization accuracy.