Research Overview


Single-Molecule Localization-based Superresolution

Fluorophores distributed with inter-particle spacing less than 250 nm become difficult to resolve due to diffraction and the overlap of their observed intensity patterns.  In Single-Molecule Localization-based Superresolution (SML-SR), one of many possible methods is used to identify and localize single molecules in a densely labeled sample. If the localization precision is better than the diffraction limit and the density of fluorophores is sufficiently high, a super-resolution image can be constructed from the fluorophore positions. In our earlier work (Lidke, Optics Express, 2005), which was one of the first papers on SML-SR, we showed that only independent fluctuations of fluorophores are needed for enhanced localization of fluorophores clustered below the diffraction limit.  Subsequently, many types of fluorophores and blinking/switching/activation methods have been used to achieve superresolution. The achievable resolution has no hard limit and our group continues to push the limit of what is possible using this concept.

Multi-Color Single Particle Tracking using Hyperspectral Microscopy

Spectral differences in fluorescence probes allow precise localization of multiple proteins separated at distances much less than the diffraction limit. Using the spectral dimension for separation leaves the time dimension available for tracking dynamics in living cells. The K.A. Lidke lab has developed a high-speed hyperspectral microscope capable of tracking up to eight spectral species of fluorescent quantum dots probes with frame rates up to 30 Hz (Cutler, PLoS One, 2013).

This instrument and analysis methods can measure and quantify protein-protein interactions with a combination of length scales (10 nm), densities (up to 10 proteins/mm2) and speed (>30 Hz) that cannot be achieved with any other currently available method. The instrument is a specific optimization of a hyperspectral microscope (HSM), which collects a broad range of the emission spectrum with high spectral resolution for every spatially sampled volume. The high spectral resolution allows the identification and separation of several species of fluorophores that have distinct emission spectral signatures. We make use of the broad excitation spectra of QDs to allow a single 488 nm laser line to excite all QDs and collect emission from 500 to 750 nm. Spectral multiplexing of up to 8 species of QDs allows single particle tracking techniques to probe and quantify the dynamics of a large class of protein homo and hetero interactions.

Image and Data Analysis

Modern Graphics Processing Units (GPUs) have approximately two orders of magnitude higher floating point performance than that of CPUs.  This is largely due to the massively parallel architecture of such cards - GPUs have hundreds to thousands of processing cores.  The K.A. Lidke lab has developed efficient routines for the estimation of fluorophore positions that run on GPU architecture using NVIDIA's CUDA interface.  Our original work, (Smith, Nature Methods, 2010) was the first to use GPUs for this task and demonstrated a maximum likelihood estimation method that could process 105 particle localizations per second.  Implementations on newer GPUs now achieve over 106 fits per second.  This approach takes the most demanding computational task of superresolution analysis and makes it trivial, removing the need for less accurate approximations in order to achieve reasonable speed. 

GPU computation continues to play a central role in several projects.  In 2011 we extended our analysis to perform multiple-particle fitting, which allows the analysis of higher density data (Huang, Biomedical Optics Express, 2011).  The GPU implementation of this approach allows analysis of typical data sets to complete in a few minutes.   A variant of this multi-emitter estimation task is used for the analysis of multi-color SPT data collected by our high-speed hyperspectral microscope.  GPU computation is essential for our approach to using realistic point-spread-functions for 3D superresolution (Liu, Optics Express, 2013).