Structures are built from components, whether the timber and brick of a house or the amino acids of a protein. Such disparate examples reveal two general strategies by which components can be arranged to form a functional structure. Artificial structures are, generally, designed: once a blueprint has been drawn, components are prefabricated and then fitted together so that each piece contributes to the overall design. Biological structures have, generally, evolved: there is no overarching design; rather, each component has evolved to contribute to the whole, and the whole has influenced the evolution of each component (the nonsensical arguments of Michael Behe aside). What happens, then, when these two strategies collide? In the case of artificial structures, you get the elegant geodesic domes of a Buckminster Fuller. In the case of biological structures, you get the equally elegant Platonic shapes of Ned Seeman of New York University. [1]

While many researchers look at the canonical B-DNA double helix and see a molecule that is at best structurally tedious, Seeman saw an architectural marvel. Just as Legos are preformed so that the studs and holes snap together, the chemistry of the natural nucleotides evolved so that hydrogen bond donors and acceptors could "snap" together to form isosteric Watson-Crick pairings. Just as structurally simple Legos can be assembled into complex and impressive structures, Seeman believed it would be possible to assemble DNA molecules into structures than that were more interesting than an extended helix.

The first step was to generate the building blocks from which larger arrays might eventually be constructed. Seeman presages much of his later experimental work in an early publication in the Journal of Theoretical Biology, where he examines the types of junctions and lattices that might be formed from paired DNA strands, and the constraints on sequence that are required to prevent the branches of these structures from migrating by exchanging hydrogen-bonding partners on different strands. [2] The simplest extension of the helix would be a junction, the meeting of three or more helices. Three- and four-arm junctions have been constructed by synthesizing and annealing oligonucleotides that can simultaneously form base pairs with more than one complementary strand (figure 1). [3, 4]

Figure 1 DNA junctions and antijunctions

These junctions' connectivity and flexibility were assayed by ligating together the termini of the junction arms. Of course, the four-arm junctions were similar in nature to intermediates observed during DNA recombination (Holliday junctions), so by studying the physical biochemistry of these junctions it has proven possible to better understand the physical biochemistry of recombination. [5] Most recently, direct evidence for the isomerization of Holliday junctions has been obtained by observing the differences in the hydroxy radical cleavage footprints of a Holliday junction constrained in an energetically unfavorable conformation versus the same junction released from constraints. [6] Five- and six-arm junctions have also been constructed, although the length of the arms had to be increased relative to three- and four-arm junctions to ensure the structures' stability [7] . Finally, while we typically think of junctions in terms of the helical lines intersecting at a central point, it is also possible to construct DNA antijunctions in which the helices are perpendicular to the nexus, and DNA mesojunctions that combine these two features (figure 1). [8]

To construct higher-order nucleic acid architectures, the flexible helical arms of the junctions had to be constrained. Structural limits were imposed in a number of ways. First, the nature of the junctions can influence the rigidity of the orientation of the arms. The number and type of single-stranded residues that form the junction have been shown to orient the helical branches preferentially. [9] Similarly, double-crossover structures (two serial Holliday junctions) have been shown to be much stiffer than single-crossover structures. [10] Second, the arms can be constrained by ligation. For example, single-stranded regions that trail from helices have been ligated to form constrained structures such as trefoil or figure-eight knots. [11] Just as sculptures are supported by armatures or, more appropriately, cathedrals are built with removable buttresses, a comlementary oligonucleotide "strut" was used to template the closure of the knots. A whimsical Borromean ring, a simpler three-member version of the five interlocked rings in the Olympic standard, has also been constructed. [12] The helical arms can also be joined to one another. In perhaps the most famous example of DNA nanoconstruction, Seeman generated an interlinked DNA molecule with the topology of a cube. [13] The cube was a masterpiece of man-made strategy: two circular faces were first generated, then ligated so as to create a third face. The "top" and "bottom" strands fronting these faces were then cyclized to form the full cube (figure 2). By employing a solid-phase synthesis strategy, a truncated octahedron has also been constructed. [14]

By constraining the basic building blocks of structure, it was hoped that repeating structures - nucleic acid arrays - could eventually be generated. Since it has already been shown that the structure of the building blocks can be controlled by altering sequence and assembly protocol, it may prove possible to in turn alter the supramolecular architecture of an array. Such a feat would herald a new era in nanotechnology in which self-assembly rather than machining might lead to the construction of nanoscale or microscopic devices. Some progress toward this goal has been made. Four-arm junctions have been ligated to form a closed quadrilateral. [15] Somewhat more rigid three-arm junctions were mixed in hopes of forming a closed hexamer, but again generated a quadrilateral [16].

While the controlled assembly of DNA Lego into complex architectures may yet be achieved, it is also possible that DNA Lego is more amenable to the self-assembly strategies normally employed by biology. In this respect, it is interesting to note that the few supermolecular structures that have so far been generated have been designed to self-assemble, rather than constructed piece-by-piece, as were the DNA cube and quadrilaterals. The DNA decamer CGACGATCGT can hybridize with itself to form a octamer duplex with two residue (CG) overhangs at either end. When this decamer is crystallized (a self-assembly process), the sticky ends pair with one another to form helices that traverse the crystal and pack against one another [17]. Similarly, Eric Henderson and his coworkers at Iowa State University have designed a simple decamer that can form an extended linear quadruplex array that can reach astounding (over 1,000 nm) lengths [18, 19] (figure 3).

Figure 3 Self-assembly of G-wires

Both of these examples detail the self-assembly of regular molecular architectures. Paradoxically, it may be possible to achieve more complex arrays by sacrificing regularity and control. For example, in collaboration with Eric Henderson we have synthesized branched oligonucleotides that template the synthesis of branched G-wires. Depending on the ratio of "linear" to "branched" building blocks, extensive DNA arrays with differing connectivities but irregular interstices can be created.

Ned Seeman's insight that DNA could be viewed as Lego has reached maturity. The incredible breadth and depth of experiments carried out by Seeman's lab over the last decade have made it possible to seriously consider engineering DNA nanostructures. What remains to be seen is whether greater control over the building blocks and their assembly will lead to the clean lines of a Frank Lloyd Wright, or whether letting the building blocks seek their own level during self-assembly will lead to the tortured nightmares of a Hans Ruedi Giger.