Rolling out DNA nanostructures in vivo

The cellular machinery responsible for maintaining and propagating life is perhaps the ultimate nanoscale device and is frequently (albeit only simplistically) mimicked in the design of artificial nanoscale objects and devices. Either out of a misguided and subtle vitalism, or because there is indeed something special about living systems, the “polymers of life,” proteins and nucleic acids, have become one of the most used buildings blocks in efforts to design and assemble nanoscale devices. DNA has proved to be especially adept at creating diverse nanostructures because of the programmable interactions that exist between complementary sequences, its predictable 3D structure (at least as a duplex), and its relative chemical stability (1). Scores of DNA nanostructures have now been reported in the literature, from an austere Aristotelian cube (2) to a more hair-down Smiley (3). To date, these DNA structures have for the most part been assembled and characterized in vitro, in large measure because the greater control over environmental conditions that is available in vitro may be necessary to promote accurate self-assembly. Additionally, nearly all of the DNA nanostructures that have been described rely on defined, synthetic DNA pieces, something that would not necessarily be available in vivo. Conversely, limitations on chemical synthesis means that DNA nanotechnologists have generally only designed structures that can be built from units of no more than 150–200 nt.

The fact that the DNA double-helix is both stable and uniform makes it ideal for passing on genetic information: pairing bases and polymerases always see much the same thing. However, ssDNA can form a variety of interesting intramolecular and intermolecular base pairs. Properly designed ssDNA sequences can yield complex 2D and 3D architectures (4, 5). These 2 properties are diametrically opposed to one another: a replicating DNA is structurally boring, whereas an interesting DNA architecture may have difficulty replicating. Previous attempts to break this conundrum have involved DNA production and nanostructure assembly in vitro. For example, single-stranded phage DNAs have been organized by synthetic oligonucleotides into 2D structures (3), and long, single-stranded synthetic DNAs have been replicated by PCR in vitro and again aligned by oligonucleotide “struts” into a 3D structure (6).

Although these were landmark demonstrations that nucleic acid nanostructures can both replicate and self-assemble, naked DNA can be more readily manipulated than can DNA in a cellular environment. Nonetheless, it was at least formally possible that a DNA nanostructure could replicate and assemble wholly in vivo. New work in this issue of PNAS from the laboratories of Hao Yan and Ned Seeman (7) takes this possibility 1 step closer to reality. Two critical questions were addressed: (i) can DNA nanostructures be produced by cellular replication and (ii) can DNA nanostructures be formed inside a cell?

To generate ssDNA for the formation of nanostructures, in vivo rolling circle amplification was exploited (Fig. 1A). Two different dsDNA sequences were cloned into a phagemid vector containing a bacteriophage M13 origin of replication. In their single-stranded form, these sequences had been shown to fold via intramolecular interactions into an immobile Holliday junction or a paranemic cross-over motif, respectively (8, 9). This phagemid construct was propagated in bacterial cells, and superinfection of a helper phage that expresses the replication machinery led to selective copying of 1 strand of the phagemid vector via rolling circle replication. ssDNA was extracted from the bacterial cultures and cleaved into unit lengths via restriction enzymes and complementary oligonucleotides. The in vivo-synthesized DNAs could then be assessed for the ability to assemble into nanostructures in vitro. Whereas previous attempts to replicate DNA nanostructures required additional, synthetic oligonucleotides for assembly, these sequences could fold into nanostructures on their own, contiguously. For the immobile Holliday junction nanostructure, endonuclease VII digestion and hydroxyl radical cleavage results to probe the backbone structure were fully consistent with the formation of the programmed, intramolecular cloverleaf structure. For the paranemic cross-over structure, electrophoretic mobility analysis in nondenaturing gels and hydroxyl radical cleavage patterns showed that the in vivo-replicated DNA folded the same as the identical sequence made from in vitro-synthesized DNA.

Fig. 1.

In vivo replication and assembly of DNA nanostructures. (A) The in vivo replication scheme used by Lin et al. (7). dsDNA sequences corresponding to either an immobile Holliday junction or a paranemic cross-over molecule (green) were cloned into a phagemid vector and transformed into E. coli cells. Infection by a helper phage initiated in vivo rolling circle replication of ssDNAs encoding the nanostructures. (Left) Isolated ssDNAs are assayed for their ability to fold into the desired structures. (Right) Alternatively, site-selective cross-linking using trimethylpsoralen (TMP; red) provided the first demonstration that an immobile Holliday junction nanostructure could assemble within E. coli host cells. (B) Possible future applications of in vivo DNA nanotechnology. Using a system similar to that described by Lin et al., it may possible to generate in vivo replicating nanostructures that can assemble into higher-order structures through intermolecular base-pairing (orange). By including as part of the nanostructure an aptamer that can recognize, say, a membrane-associated protein (red), it may be possible to generate an “artifical cytoskeleton” that organizes the cell surface.

However, assessing whether the nanostructures could form in vivo turned out to be a nontrivial task. Taking advantage of the sequence-specific, cross-linking properties of psoralen, Lin et al. (7) provide the first evidence for the formation of an immobile Holliday junction in vivo. For these experiments they used a version of the immobile Holliday junction whose sequence was modified so that cross-linking could only occur at 1 position if the 4-way junction structure was formed (Fig. 1A). After permeabilization, treatment by psoralen, and UV-induced cross-linking of Escherichia coli cells containing the replicating phagemid vector, the DNA was purified and the in vivo cross-linked products were compared with those derived from in vitro cross-linking experiments. Lin et al. found identical cross-linking patterns between the in vivo and in vitro cross-linking, although in both cases the presence of the intercalating cross-linker induced a different stacking arrangement of the duplex arms than was usually found in this type of junction. These experiments provide the first indication that rationally designed DNA nanostructures can assemble in a cellular environment.

However, the replication and assembly of DNA nanostructures in vivo begs the question of what such are structures really good for. Seeman (10) originally proposed the creation of crystalline 3D DNA arrays as a means of organizing proteins for structure determination. Building from that inception point, 2D DNA arrays that organize proteins (11), RNA aptamers (12), and nanoparticles (13) have been developed and have potential applications as biosensors. Porous 3D DNA nanostructures can encapsulate or adsorb a variety of proteins and have potential applications to drug delivery (14, 15). DNA nanotubes have been used to template the growth of metallic nanowires (16) and have potential applications in molecular electronics. And Seeman's original notion is also beginning to be realized, with DNA nanotubes now being used to improve residual dipolar coupling constants in NMR experiments with membrane proteins (17). Although the massive parallelism of DNA computers has not yet come to pass, DNA nanostructures have algorithmic assembly properties in and of themselves (18, 19), and it seems likely that nucleic acids can be programmed to self-assemble or morph into particular architectures based on their environmental interactions.

Rationally designed DNA nanostructures can assemble in a cellular environment.

Therefore, the ability to replicate DNA nanostructures in vivo may have quite practical consequences. While the frontiers of DNA synthesis continue to be pushed forward, the routine synthesis of oligonucleotides is still most efficient below ≈200 nt. An in vivo replication system could therefore produce nanostructures that required much larger single-stranded molecules, up to several kilobases or more. Multiple cross-over junctions or interwoven structures that involved different kinds of nanoscale primitives could be constructed, replicated, and would fold into extensive, modular architectures. Indeed, it may now be possible to generate paranemic cross-over nanostructures in vivo that have an improved ratio of full-length to truncated products relative to previous in vitro replication experiments (7, 9). The system described by Lin et al. (7) also appears to be fully scalable: the larger the bacterial culture used, the greater the yield of ssDNA. Large bioreactors are now routinely used for recombinant protein expression, and the recovery of large amounts of nanostructure-forming DNA from such production-scale cultures can be readily imagined.

Most importantly, though, the work of Lin et al. (7) finally allows the somewhat artificial world of DNA nanotechnology to begin to interface with the basic unit of biology and evolution, the cell. For example, nanostructure functionality can be directly selected for in the context of a cell. Although currently there is no obviously cellular functionality of DNA nanostructures, it is possible to imagine many advantages to having a programmable nanostructure within a cell. Nonprogrammable (or weakly programmable) nanostructures such as the cytoskeleton are of immense importance in the organization of the cell surface and signal transduction. Signal transduction can be engineered by reprogramming the interactions of enzymes with their scaffolds (20). An “artificial cytoskeleton” (Fig. 1B) might allow researchers to impose completely new scaffolds and interactions on these enzymes, perhaps via the previously demonstrated expedient of coupling aptamers to the nanostructure.