Top: How to grow a ninja-star dendrimer. The Matrix-esque glyph to the left of each stage is the reaction graph, or programming instructions, for making it.
Above, inset: The key to the reaction graphs. Each circular “node” represents a DNA hairpin. Input domains are triangles; output domains are circles. Inaccessible domains are filled with solid colors. Binding to an accessible domain will open the inaccessible ones. Arrows between the nodes indicate what binds to what.
Above: The dendrimer is built from five kinds of A hairpins and four kinds of B hairpins that assemble themselves in sequence according to the reaction graphs.

Dances with DNA

Nature is a software engineer par excellence. By rearranging protein and RNA building blocks, nature programs myriad molecules to synthesize, haul, detect, and regulate one another. Now scientists are trying their hands at it, and in the January 17 issue of Nature, a group of Caltech researchers published examples of their own molecular programs. Associate Professor of Applied and Computational Mathematics and Bioengineering Niles Pierce, senior postdoctoral scholar Peng Yin, grad student Harry Choi, and research technician Colby Calvert showed how molecules of DNA only ten nanometers in length could be directed to perform specific tasks unaided—without external energy sources, temperature changes, or enzymes.

Biomolecular engineers have assembled DNA molecules into stable patterns, like planar crystals, wireframe cages, tubes, smiley faces, and maps of North America. Pierce and his colleagues concentrate on the motion of the interacting DNA molecules. To see the distinction between these approaches, consider the difference between a choreographer (Pierce) and a cheerleading coach. The coach primarily cares where the cheerleaders end up in a human pyramid: stronger, heavier people go on the bottom, while more agile, lighter people are at the top. He doesn’t care how they get there: all’s well that ends well. By contrast, when directing dancers, the choreographer cares most about how the dancers move across the floor and who they partner with. Where they end up is of less importance. “The trajectory the molecules take is actually the goal of our programs, and the destination is just the by-product: it’s what you get when the function is complete,” says Pierce.

The dancers are short “hairpins” of DNA that fold back onto themselves. Each hairpin has three domains—one input and two output—that can interact with domains in other hairpins by matching the “letters” in one strand with the letters in another. In the alphabet of DNA, A pairs with T and G matches C; the hairpins contain between 50 and 100 letters. The hairpin’s input domain is initially available to pair up with other DNA molecules, while its output domains are inaccessible. Once a matching piece of DNA binds to the input domain, the hairpin pops open and the output domains are exposed. Output domains of open hairpins can then seek out the input domains of closed hairpins and open those molecules. The ensuing cycle becomes a molecular square dance with hairpins exchanging partners according to the design of the bioengineers.

All of these exchanges occur without Pierce having to add energy to the system. So what makes them go? “The basic feature of the hairpin is that it’s initially trapped in a high-energy state,” says Pierce. This state is similar to a mousetrap that has been set and baited. Until a mouse trips the trigger, the trap is stable and doesn’t move. But within the spring of the trap, there is energy waiting to be released when the unsuspecting mouse goes for the cheese. A piece of DNA binding to the input domain triggers the hairpin to release the stored energy locked up in the inaccessible output domains—when the output domain pairs up with still other pieces of DNA, the entire system goes to a lower-energy state.

By designing how each hairpin domain pairs with its fellows, the Caltech team can harness this energy to make the molecules perform the specific task they want. This part of the design process is the most difficult, and requires the team to model the physics of the hairpins. Using algorithms developed by Pierce’s group, the researchers ensure that when the hairpins are mixed together, they interact appropriately so that no hairpin runs off with another hairpin’s dance partner.

To showcase the hairpins’ capabilities, Pierce and his colleagues “wrote” four different programs. In each case, the hairpins were designed not to interact until an initiator molecule was introduced to the system. For the first program, the initiator triggered the hairpins to self-assemble via a specified sequence of “handshakes” into branched structures with multiple arms shooting out from a central point, like three- and four-armed starfish. Upon completing this assembly process, the initiator then disassembled from the structure to catalyze the formation of more starfish. In the second program, the hairpins assembled into a tree-like pattern called a dendrimer, growing from the root of the tree to the leaves. Another program demonstrated a phenomenon called autocatalysis, in which a chemical reaction—in this case, the production of fluorescent pairs of hairpins—feeds on itself. After the initiator was added to the solution, the test tube would begin to glow, getting brighter and brighter exponentially. The most dramatic example was a DNA “walker” that used its DNA “legs” to lurch along a DNA track one step at a time. The walker was inspired by the protein kinesin, which glides along protein microtubules in cells to move molecular freight. “Years ago, I was amazed when introduced to the programmable chemistry of kinesin. I decided then and there that I wanted to be able to engineer that kind of molecular complexity. We still have a long way to go,” says Pierce.

Pierce foresees these hairpins being put to use as molecular sensors or nanomechanical drugs. Molecular instrumentation could detect small changes within cells, like the switching on of a gene within a developing embryo, producing a fluorescent signal for scientists to read and analyze. He hopes that programs like the exponentially glowing one could develop into cheap technologies that would amplify the presence of a miniscule amount of an interesting molecule into a detectable signal. “Instead of thinking of instrumentation as something expensive that your experiment resides within, we want to design exquisite instruments that you embed within your system of study,” says Pierce. Programmable molecules may also eventually lead to dynamic drugs that use one input domain to pinpoint cancer cells, triggering an output domain to kill them.

For Pierce, the work represents a step toward the long-term goal of developing a compiler for biomolecular functions that would allow bioengineers to write molecular programs the way that computer scientists write electronic ones. A compiler is the software that translates high-level programs written in a language like C++ into the binary instructions the machine actually executes. As a first step, the team has developed graphical representations of their hairpins that are used in schematics, called reaction graphs, to describe each step in a program—for example, an output domain on strand C binding to an input domain on strand D in step four, only to unbind again later in preparation for the next cycle. These reaction graphs are not unlike the flow charts beloved of computer programmers. As a software package, the molecular compiler would translate an engineer’s design ideas into reaction graphs and then translate those graphs into a specific set of DNA hairpin sequences to be synthesized. “We want to liberate the molecular engineer from having to think about the detailed structural features of the molecules and instead focus on the functional behavior of the system,” says Pierce.

“In designing a compiler, there’s work for many different fields: computer science, applied mathematics, control and dynamical systems, chemistry, and physics,” says Pierce, who is teaming up with other researchers at Caltech and elsewhere on the project. “Building a molecular compiler is a very daunting challenge, but progress in the field has been pretty dramatic in the last five years, so a primitive first-generation compiler is probably now within reach,” says Pierce. Even so, the day when dancing molecules detect and kill cancer cells in humans is probably still far in the future. —MT