(page 2 of 2)

A single high-resolution image of a sample under an electron microscope has too much noise to yield accurate molecular representation. Images must be averaged together with their counterparts to reduce noise. Plus, any single molecular assembly imaged will be but one 2-D projection of what is a 3-D object, so the averaging must be done over many possible angles. To build a 3-D model, one must take many images and build a structure by looking at all the different angles of all the different molecular assemblies imaged.

Building a 3-D model is like looking at a piece of sculpture in a gallery. Only by walking around the piece and viewing its various sides and angles can the brain build a mental image of the art and fully comprehend its dimension, perspective, and scale. The same is true using a computer. Only by piecing together many different views of a molecule from a microscope can a computer build a model of the molecular assembly.

And the molecule that is being imaged gets destroyed in the process, so the next image must be captured from some other part of the sample holder grid. This has always required a person to choose different spots on the grid manually. As the number of grid spots goes up, so goes the level of tedium.

"What we really want is 100,000 to 1,000,000 molecule images and that just takes too long to do manually," says Carragher. "Then you want to do 10 different conformational states, 20 different labeling studies, and each time it's going to take three to six months. That's more than the lifetime of a graduate student."

"There are projects," Carragher adds, "projects people just don't do because the manual labor required is just too daunting."

Carragher and Potter, who lead the Automated Molecular Imaging group, are creating algorithms for automated data collection and analysis, which should simplify the technique of electron microscopy and enable throughput to be increased dramatically.

So Long, John Henry

Several years ago, Carragher and Potter suggested that automated data collection and analysis could be developed for EM. A similar goal had been accomplished in x-ray crystallography, and given the need for structural information in our post-genomics proteomics world, automation would represent significant progress.

So Carragher and Potter started developing automated EM algorithms and began writing grants with Milligan to develop these into programs. "It took off from there," says Carragher.

They succeeded in developing software for both the collection and the analysis, which they brought to TSRI when they came last year to form the Automated Molecular Imaging group at TSRI. Milligan helped recruit his long-time collaborators from the University of Illinois at Urbana–Champaign, where they were co-directors of the Imaging Technology Group of the Beckman Institute for Advanced Science and Technology.

Creating the algorithms was not easy. Using the manual technique, a person has to make decisions about where to focus the EM beam and take a picture, looking first at low resolution and then deciding in which areas to collect data at high resolution. For automation to succeed, the computer must do the same thing and use intelligent criteria to search the low resolution image for appropriate targets.

"Even a two-year-old can tell a cat from a dog, but that's a very hard problem for a machine," says Carragher. "But what humans are not good at is doing the same boring thing a thousand times in the dark for weeks."

Carragher and Potter had to write their software to take a low-resolution image, select areas to image in medium resolution, and then analyze that image and strip out targets for high-resolution maps. Then, they had the computers put the data into processing programs and calculate 3-D maps. Recently, they have been testing and refining the programs.

"What we have done over the past year is to show that you can insert a [sample] in the microscope and [calculate] a 3-D map fully automatically," says Potter.

In fact, Carragher and Potter constructed one of the best 3-D maps of the tobacco mosaic virus in under two days. By comparison, the work would have taken several months of work just a few years ago and perhaps several weeks using conventional methods today.

"We can now go from inserting a specimen in the microscope to having a 3-D map in 24 hours," says Milligan, adding that the fear of failure should no longer be a limiting factor for experiments.

Still, the automation is not fully implemented, so one of the immediate goals of the Automated Molecular Imaging Group is to see their software used for practical applications, something that their coming to TSRI will facilitate.

"There are so many people who want to collaborate with us here—it's great," says Potter, adding that within a few months of their arrival they had already found an almost overwhelming number of projects.

"At the moment we need to make the technique very efficient and very general, [and] get it out to the community" says Carragher. "We can do it, and now we want to be able to do it routinely for anybody."

Additional plans include the design of technology that would make EM high-throughput. This includes a robotic specimen handler that Carragher and Potter have been experimenting with that would allow the instruments to be left alone to collect and analyze even larger sets of data.

"You could look at maybe 10 grids overnight," says Potter.

Ready for Tours

Though the shared space of the CarrAmerica B building and the collaborations it fosters within CIMBio and throughout the TSRI campus will be reward enough, there is one more thing that the building provides: ready-made tours.

In the plans that Milligan and the others drew up, they envisioned several people controlling the microscopes and discussing the images as they are collecting data. They also anticipated people peering through the glass wall of the control room, and this has its advantages.

"It's a very easy way to communicate what we are doing," Carragher says to me as we walk past the control room on a recent tour of the facilities. A technician had one of the new microscope's back panels open and was busy fiddling with some wires.

"Is it done yet?" Carragher asks.


1 | 2 |



This model of tobacco mosaic virus at 10 angstrom resolution was generated completely automatically from specimen grid to 3-D map in less than two days.

















For more information: