Neuroscientists working to understand how cells in the brain communicate with one another dream about one day having a precise map charting exactly how each cell connects to the next. Such a map would be a powerful tool for learning how the brain processes information and controls behavior, as well as recognizing how neuronal communication breaks down during disease. But producing that map with today's research tools is a staggering task.
So far, researchers have determined the complete cell-to-cell connections for the brain of just one organism: C. elegans, a tiny flatworm with 302 neurons. With more complex organisms, the pattern of those connections becomes increasingly intricate and difficult to trace. The human brain, for example, is a tangle of about 100 billion neurons-each one communicating with thousands of its neighbors. Just one cubic millimeter of brain tissue contains about 50,000 neurons, many packed together too closely to discriminate individual cells using standard light microscopy. And according to HHMI investigator Xiaowei Zhuang, a useful wiring diagram must illustrate which neurons are connected to one another, but also provide information about how the connections function. This means researchers would need to be able to identify which signaling molecules reside at individual connections.
Neuroscientists are very imaging sophisticated: they are already using the best there is so far. Still, they hit limits. That's why when they see a new technology, they recognize its potential.
The technical challenges are immense, but Zhuang says recent advances in several fields make it the right time to tackle the problem. She is leading a team of collaborators that includes creators of some of the most powerful and innovative new methods for imaging, data analysis, and sample preparation. The team, which has received a new HHMI Collaborative Innovation Award, seeks to develop the technology that will be required to map all the neural connections in mammalian brains. The team will work together to unite and refine their techniques, then construct wiring diagrams for two important regions of the mouse brain.
Neuroscience is a new arena for Zhuang, whose HHMI lab at Harvard develops imaging techniques to monitor individual biological molecules and cells. But she says neuroscience is a field in which new imaging technology can have an immediate impact. “Advanced imaging techniques often have their earliest applications in neuroscience. Neuroscientists are very imaging sophisticated: they are already using the best there is so far. Still, they hit limits. That's why when they see a new technology, they recognize its potential”
Recently, Zhuang's lab developed one such technology-a new method of light microscopy that allows researchers to see objects far smaller than those that can be discerned with conventional optical microscopes. The technique, known as stochastic optical reconstruction microscopy (STORM), uses fluorescing probes that can be turned on and off to let researchers zero in on structures separated by distances smaller than the resolution limit of traditional light microscopes. STORM allows users to track and distinguish millions of molecules with nanometer-scale resolution.
Zhuang explains that without the super-high resolution that STORM offers, it would be impossible to thoroughly map a mammalian brain. “We think of the mouse brain as a gigantic piece of material to be imaged—it's a centimeter cubed. The resolution required is extremely high, because we want to resolve each individual neuron and each individual synapse. Some of the neuronal wires—the axons and dendrites—can be as thin as several tenths of a nanometer. They can be densely packed against each other, so if you do not have high resolution, you cannot tell one wire from the other. The synapses are also very small and can be densely packed, so without high resolution you cannot tell one connection from another in some of the synapse-dense regions.”
“We didn't invent STORM specifically for this purpose,” Zhuang says, “but it looks like an ideal match.” The approach will be particularly powerful, she says, when used to image the vividly colored brains of genetically engineered mice created in the Harvard labs of two of her collaborators on the project, Jeff Lichtman and Joshua Sanes. Each individual neuron in the brains of these mice-known as Brainbow mice-is genetically programmed to exhibit one of approximately 100 distinct hues. When visualized under a light microscope, nearby neurons can be distinguished by color, greatly facilitating the tracing of neuronal connections.
The fluorescent probes used to color the neurons, however, are not well suited to STORM imaging, so Lichtman and Sanes will develop a new generation of Brainbow mice with new labels. Meanwhile, Zhuang's lab will expand the color palette that STORM can detect, so the technique can be used to simultaneously visualize the multicolored neurons and localize specific molecules at the junctions between them.
While these techniques should allow the team to see the appropriate level of detail in the brain, each STORM image can reveal only a tiny area. An isolated snapshot of a single section provides little understanding of how neurons fit into the complete network, and even a large stack of those sections is meaningless if the pieces cannot be fit back into their original context. So the team will employ several strategies to ensure that it can reconstruct the brain's complete three-dimensional wiring structure from its many millions of sections of thinly sliced brain tissue.
First, the team will use a tool developed in the Lichtman lab called an automatic tape-collecting lathe ultramicrotome (ATLUM) to prepare sections of the brain for imaging. This instrument rotates the tissue on a lathe, slices it into a continuous ribbon, and places the ribbon on a flexible sealed tape, ensuring that no tissue is lost. “We need this cutting technology to allow us to connect the sections without seams,” Zhuang explains.
The team will complement its STORM imaging of the Brainbow mice with a new three-dimensional imaging platform developed in collaborator Stephen Smith's lab at Stanford University. The high-throughput platform, called array tomography, involves building, staining, and imaging arrays of large numbers of tissue sections, such as those prepared by ATLUM. “These arrays can be stained and unstained many times with antibodies that target different kinds of molecules,” Zhuang explains. This will allow the team to develop a comprehensive view of the molecules present in each neuron and each synapse. “So this way, you can not only see the neurons, but you can understand the properties of the neurons and the connections between the neurons,” Zhuang explained. “For example, you can see whether they are excitatory or inhibitory.”
Imaging even a small region the mouse brain with the proposed level of detail will generate an enormous amount of data, so the project demands extremely efficient analytic techniques. “It is often said that the human eye is the best imaging analysis machine,” Zhuang notes. “But as sophisticated as our eyes and brains are, with this amount of data, it is impossible for us to go in and trace each neuron manually.” So the team has recruited Sebastian Seung, an HHMI investigator at MIT, to create computer algorithms that can make sense of the information. Seung is a leader in developing machine learning techniques that trace neurons' axons and dendrites in microscopy images and recognize the synapses at which the cells connect. Seung's lab will adapt their algorithms so that can be used to analyze the colorful STORM-generated images of Brainbow mice.
Each of the technologies the team plans to use will require optimization to ensure that it is compatible with the others and powerful enough to meet the project's ambitious goals. But within a few years, the team expects that it will be ready to deploy its strategy to map two regions of the mouse brain - a specific part of the retina and an area of the cerebral cortex. By detailing the connectivity and properties of all of the neurons in these regions, the team hopes to reveal new information on how the brain's structure underlies its function. More importantly, they hope to encourage a broader effort within the neuroscience community to use their technologies to map the connections among all 75 million neurons in the mouse brain.