In the past decade there has been a large advance in microscopy spurred by the advent of high-speed CCD (charge-coupled device) detectors, inexpensive lasers, and advances in "staining" protocols. One could argue that the major advance provided by whole-genome shotgun sequencing is that, precisely because the entire genome of the underlying organism is known, we can now light up particular proteins in particular cells with GFP (green fluorescent protein) and other recombinant constructs. Most of the mesoscale functioning of the cell and many organism-level processes, such as neural functioning, will be unraveled primarily by direct observation using new microscopy techniques. An essential factor in the success of such a transition will be the existence of algorithms and software that can reliably and automatically extract information from tens of thousands of captured images.
The group leaders—an unparalleled interdisciplinary gathering of researchers—who will converge on Janelia Farms in June 2006 will collaborate on large-scale projects. Shared goals include (1) the detailed in vivo observation of the 300-cell nervous system of C. elegans, in order to infer function of the known neuroanatomy; (2) determining the complete neuroanatomy of a fly brain (100,000 neurons), down to the level of synaptic connections; and (3) a first-order understanding of olfaction and vision in mice. These projects will involve new technology and a great deal of image data. For example, tens of thousands of "sections" must be analyzed to reconstruct the fly brain in three dimensions. While these projects begin to crystallize, my group is focusing on five "warm-up" projects.
Registration of 3D Confocal Stacks of Fly Brains
In collaboration with group leader Julie Simpson, we are developing software to register stained fly brains automatically. To do this, we find a minimal energy warping of space that maps corresponding regions of one fly brain to any other. This is necessary because every fly brain is different in size and morphology, and each experiment must be performed on a different fly. The staining agent is a general contrast agent for neurons. A selected set of neurons, implicated in some functional process, are stained in another color. Interesting computational problems arise when we examine the relationship between two stained sets of targeted neurons.
Tracing Sparsely Stained Neurons in Mice Brains
With group leader Karel Svoboda, we are developing software to track a set of sparsely stained neurons in a set of 3D stacks tiling the entire cerebral cortex of a mouse. If we can trace neurons and identify their dendritic processes, then in principle, over a sufficiently large collection of such data sets, we could build a "stochastic" neural atlas of a mouse that would reveal the connectivity patterns of the neurons.
Analysis of PALM Microscopy Data
With group leader Eric Betzig, we are refining and improving the interpretation of raw data produced by a new photoactivated localization microscope (PALM) developed by Betzig and Harald Hess. In principle, this microscope can use light with a wavelength of 440 nm to localize the position of a molecule to within 2 nm. We are exploring protocols for this microscope that may allow us to observe aspects of the mesoscale structure of cells and to build models of organisms (such as a brain) with a level of detail that extends to the position and character of every individual synapse.
Cell-Level Expression Atlas of C. elegans L1-Adult
With Stuart Kim (Stanford University), we are producing 3D stacks of in situ worm preparations, where all nuclei are stained and the 81 muscle nuclei are also stained with a GFP construct in a different color. Our goal is to be able to segment every nucleus and identify it within the fixed (eutological) cell lineage ofC. elegans. If we can accomplish this with high reliability, then we can observe the cell-by-cell expression of any gene simply by building a nucleus-localized GFP construct for the gene. With between 550 cells in a first-stage larva (L1) and nearly 1,000 in an adult worm, the image analysis problem is difficult; the SIR construct was added as a fiduciary marker to facilitate the task. Alternate or additional markers will be developed if necessary.
In Vivo Monitoring of Cellular Processes During Mitosis
With Tony Hyman (Max Planck Institute, Dresden), we are developing methods for tracking and measuring cellular processes during the first anterior/posterior mitotic division of C. elegans. For example, a GFP-labeled protein involved in the growing cap of a tubulin fiber can be traced as these fibers are spun out from the centrosomes toward the cell wall, giving a readout of growth rate and total flux. Similarly, the position and movement of GFP-labeled centrosomes can be monitored. Our goal is to achieve a fine-grained phenotypic description of the labeled object/process and its degradation/change under a genetic screen of RNA interference knockdowns.
In addition to the five projects above, I spent the first 25 years of my career developing fundamental algorithmic methods for the comparison of sequences (e.g., BLAST, concave gap penalties, sublinear search, and approximate pattern matching) and the reconstruction of genomes from read data (e.g., the overlap/layout/consensus paradigm, interval graph collapsing, the paired-end whole-genome-sequencing protocol and Celera assembler, and most recently the string graph paradigm). I continue to work with a number of collaborators on new methods for solving some of my favorite problems.
CONTACT
