In his UCLA lab Desmond Smith chops what appear to be pieces of tofu. He is carefully slicing a human brain into small cubes to measure gene expression in three dimensions.
“Microarrays are now a very powerful and widely used tool in genomics,” he says, “but they’ve not been used to look at 3D patterns of gene expression.” Smith and colleagues at the University of Southern California have developed a method that combines microarrays and the image construction mathematics of PET and CT scan technology to create 3D maps of gene expression. The hope is that comparing the same regions of diseased and non-diseased organs will help highlight disease-associated genes that are buried in flat gene profiling.
Although eventually Smith hopes to apply the technique to other organs and even entire organisms, the researchers have begun with the brain because of its tremendous heterogeneity.
“The first thing you have to do is get a brain,” says Smith. “It does involve the destruction of the brain. So obviously if it’s a human being, it’s best to wait till they’re dead.”
The researchers divide the brain into cubes called voxels, the three-dimensional equivalent to the pixel. And a digital camera snaps a photo of each bit of brain. The smaller the voxels, the greater the resolution of the brain’s spatial gene expression. They then measure the level of expression in each voxel with microarrays. “It’s an enormous dataset because we’ve got many voxels and many thousands of genes in each voxel,” says Smith.
The lab then sends this data off to their collaborators at the electrical engineering department at USC. The USC group, led by Richard Leahy, uses a mathematical method called singular value decomposition to analyze the data and reconstruct the voxels into a three-dimensional image of the brain with gene expression levels superimposed. “These images are very similar to those you get from a PET scan or a functional MRI scan,” says Smith. “Areas in red and yellow represent areas where the gene is expressed at a high level and areas in blue represent areas where the gene is expressed in a low level.” Because no two brains are exactly the same shape, Leahy must mathematically morph one brain to the other in order to normalize the data.
So far the researchers have only tested the method on hemisection of brain, dividing it into 24 one-centimeter cubes, representing a 500-voxel resolution of the brain. Now they are working with three-millimeter pieces, equivalent to about a 10,000-voxel brain.
“Our long-term vision is to have a high-resolution atlas of gene expression for the entire human brain available in the public domain, on the Internet,” says Smith. The only limitation is money. “If we had unlimited resources and unlimited funds, I could easily imagine this being completed in two years,” he says.
Based on the sample size needed to get enough RNA for expression analysis, the highest resolution theoretically possible is 325,000 voxels. At $200 per voxel, a complete brain atlas would run $65 million. “Well, of course I’m biased, but I believe it would be money very well spent,” says Smith.
At first the researchers used a ruler and scalpel to cut the brain. But as the resolution improved, that method was not precise enough. “So we’ve devised a crisscross array of sharp cutting blades at right angles to each other which can directly cut the brain into cubes in precise spatial relationship one to the other,” says Smith. Ultimately, he would like to voxelate not just the brain but also the entire mouse. He would also like to fully automate the entire process so that it wouldn’t require human intervention.
“Our dream is that you could just inject, say, a drug or a drug candidate into the entire mouse, let the mouse sit there for six hours, and then just feed him into the ‘voxelator,’ as we call it, and you’d end up in an automated fashion with a gene expression pattern of all 30,000 genes in the entire three dimensions of the mouse,” says Smith. “Obviously that’s a very distant dream, but that’s our long-term goal.”