NATUREIVol 440/16 March 2006 ARTICLES often distorts aspect ratios so that inter-helix gaps cannot be inferred each original from the aspect ratio of a single rectangle. A range of aspect ratios created by mixing appropriate subsets of these strands. In this way, aplied a gap size from 0.9 to 1.2 nm; later designs assume I nm. any desired pattern can be made ong stacked chains with dozens of rectangles. Such stacking was ox In principle, a variety of DNA modifications-for example, biotin almost completely abolished by omitting staples along vertical edges. (Fig. Id inset, Supplementary Note S6), designed to avoid dimeriza On the other hand, stacking across the seam of an unbridged tion at high concentration, were added to the middle of 32-mer rectangle(as in Fig. Ic)kept 65%of structures(S=40)well-formed; staples at the position of merges made during design. Depending on the rest showed some degree of dislocation at the seam. Other defects, the merge pattern, the resulting pixel pattern was either rectilinear such as the intentional omission of single staples, could be visualized with adjacent columns of hairpins on alternate faces of the shape, or as 5-10-nm holes However, sharp tips and high tapping amplitudes staggered and nearly hexagonally packed, with all hairpins on the were required; repeated scanning created holes difficult to dis- same face In AFM images labelled staples give greater height contrast tinguish from holes due to missing strands. This effect also increased (3 nm above the mica) than unlabelled staples(-1.5 nm), which uncertainty when stoichiometry was varied. When staple excesses of results in a pattern of light" Iand dark 0' pixels. Several patterns approximately 100: I and 9: 1 were used, the frequencies of 5-10-nm(Fig. 3), each with -200 pixels, illustrate the generality of this holes(a few per rectangle) were indistinguishable. At 2: 1, rectangles technique. were similar; perhaps a greater fraction were malformed. At 1.5:1 Yields of patterned origami similar to those of unpatterned rectangles formed but had holes up to -10% of their area in size. At a origami; for the pattern in Fig. 3a, 91%(S=85)of rectangles were 1:1 ratio, <1% of structures were rectangular well-formed. Because rectilinear patterns imaged poorly, only stag To demonstrate the creation of arbitrary shapes, a five-pointed star gered patterns were examined quantitatively. Distances measure was designed with 1.5-turn spacing, 32-mer staples and a linear between pairs of I' pixels in alternating columns( two pixel widths ather than circular scaffold(Fig. 2c). Designed assuming a 1.5-nm 11.5+0.9 nm, mean + s.d., n=26)and adjacent rows(one pix inter-helix gap(the work was carried out before the gap for 1.5-turn height: 6.6 0.5 nm, n= 24)are consistent with the theoretically acing was measured), the stars are somewhat squat(Fig. 2c, upper expected pixel size of 5.4 nm x 6nm. Most defects take the form of AFM image). Still, the stars show that the width of a shape may be 'missing pixels; that is, pixels that should image as'I's but image as approximated to within one DNA turn. Many of the structures O's instead 94% of I pixels(of 1,080 observed)were visualized bserved were star fragments(Fig. 2c, lower AFM image), and only Whether missing pixels represent real defects or artefacts of imaging 11%(S=70)were well-formed. The low yield of stars(and squares, is unknown; sequential AFM images nally showed"I pixels ee above)may be due to strand breakage occurring during BsrbI that later converted irreversibly to 0 pixels, suggesting tip-induced digestion or subsequent steps to remove the enzyme; when untreated damage. Stoichiometric errors, synthetic errors, or unwanted sec- circular scaffold was folded into stars, 63%(S=43)were well- ondary structure are not implicated for any particular strand, as the formed. To show that DNA origami need not be topological disks, position of missing pixels appeared random(Fig. 3b, f and g) and that scaffolds can be routed arbitrarily through shapes, a three- Stacking of shapes along blunt-ended helices provides an uncon hole disk was designed(Fig 2d). Although the shape approximated is trolled mechanism for the creation of larger structures(Fig. 3b) mmetric, the folding path is highly asymmetric and has five distinct Instead of removing staples on the edge of a rectangle to avoid seams. Unlike the rectangles, which rarely break or fold, three-hole stacking(as described previously ), 4-T hairpin loops(four thymines disks exhibit several characteristic deformations( Fig 2d, lower AFM in a row, Fig. le, inset) or 4-T tails can be added to edge staples image); still, 70%(S=90)were well-formed (Fig. 3e, f); stacked chains of 3-5 rectangles still formed(Fig. 3g) DNA origami is not limited to the approximation of shapes by but 30% of rectangles(S=319)occurred as monomers(Fig. 31). raster fill: some shapes can be created more exactly by combining Without hairpins, all rectangles occurred in aggregates( Fig. 3h) distinct raster fill domains in non-parallel arrangements. Figure 2e Controlled combination of shapes was achieved by designing shows a triangle built from three separate, 2. 5-turn spacing rec- extended staples that connected shapes along their edges. To create angular domains; only single covalent bonds along the scaffold hold a binding interaction between two particular edges, extended staples the domains together. But the desired equiangular triangles(upper were designed by merging and breaking normal staples along these AFM image) were rarely observed(<1%, S=199). As seen in the edges(Supplementary Note S7). Starting with sharp triangles, this lower AFM image, stacking caused rectangular domains of separate approach was used to create finite(hexagons; Fig 3n, P, q) as well as triangles to bind; this effect and the flexibility of the single-bond periodic structures(triangular lattice; Fig. 30, r-u). I note that the joints at the vertices may account for the ease with which these successful combination of shapes(unlike the successful formation of triangles deform. To solve these problems, sharp triangles, built individual shapes) is in principle very sensitive to the concentrations (Fig. 26). The slanted edges of the trapezoid cing, were designed of extended staples(which should ideally be equal to that of the from trapezoidal domains with 1.5-turn meet at the triangle scaffold). Poor stoichiometry may play a role in the poor yield of vertices and allow the addition of bridging staples along these hexagons(<2%, S=70)and lattices(not measured) interfaces. Sharp triangles remained separated and equiangular (Fig. 2f, lower AFM image); 88% were well-formed (S=78). Even Discussion when bridging staples at the vertices were not used, a large number of The scaffolded self-assembly of DNA strands has been used to create well-formed(55%, S=22). These weakened' linear structures" and proposed as a method for creating arbitrary sharp triangles provided the most stringent test of the estimated patterns.. But the widespread use of scaffolded self-assembly, and nter-helix gap, because too high or low an estimate would have in particular the use of long DNA scaffolds in combination with caused gaps or overlaps between trapezoids. Gaps of 10 nm occasion- hundreds of short strands, has been inhibited by several misconcep ally appeared but overlaps were never observed, suggesting that I nm tions: it was assumed that(1)sequences must be optimized to avoid may be a slight underestimate of the inter-helix gap secondary structure or undesired binding interactions,(2)strands must be highly purified, and(3) strand concentrations must be Patterning and combi precisely equimolar. These three criteria are important for the In addition to binding the DNA scaffold and holding it formation of many DNA nanostructures and yet all three are ignored staple strands provide a means for decorating shapes with in the present method. For example, M13mp18 is essentially a patterns of binary pixels. Given a shape, the original set of natural sequence that has a predicted secondary structure which is aken to represent binary 0s; a new set of labelled staples, one for more stable(lower in energy) than similar random sequence 2006 Nature Publishing Group© 2006 Nature Publishing Group often distorts aspect ratios so that inter-helix gaps cannot be inferred from the aspect ratio of a single rectangle. A range of aspect ratios implied a gap size from 0.9 to 1.2 nm; later designs assume 1 nm. Whatever the exact value, it is consistent: aspect ratios were invariant along stacked chains with dozens of rectangles. Such stacking was almost completely abolished by omitting staples along vertical edges. On the other hand, stacking across the seam of an unbridged rectangle (as in Fig. 1c) kept 65% of structures (S ¼ 40) well-formed; the rest showed some degree of dislocation at the seam. Other defects, such as the intentional omission of single staples, could be visualized as 5–10-nm holes. However, sharp tips and high tapping amplitudes were required; repeated scanning created holes difficult to dis￾tinguish from holes due to missing strands. This effect also increased uncertainty when stoichiometry was varied. When staple excesses of approximately 100:1 and 9:1 were used, the frequencies of 5–10-nm holes (a few per rectangle) were indistinguishable. At 2:1, rectangles were similar; perhaps a greater fraction were malformed. At 1.5:1, rectangles formed but had holes up to ,10% of their area in size. At a 1:1 ratio, ,1% of structures were rectangular. To demonstrate the creation of arbitrary shapes, a five-pointed star was designed with 1.5-turn spacing, 32-mer staples and a linear rather than circular scaffold (Fig. 2c). Designed assuming a 1.5-nm inter-helix gap (the work was carried out before the gap for 1.5-turn spacing was measured), the stars are somewhat squat (Fig. 2c, upper AFM image). Still, the stars show that the width of a shape may be approximated to within one DNA turn. Many of the structures observed were star fragments (Fig. 2c, lower AFM image), and only 11% (S ¼ 70) were well-formed. The low yield of stars (and squares, see above) may be due to strand breakage occurring during BsrBI digestion or subsequent steps to remove the enzyme; when untreated circular scaffold was folded into stars, 63% (S ¼ 43) were well￾formed. To show that DNA origami need not be topological disks, and that scaffolds can be routed arbitrarily through shapes, a three￾hole disk was designed (Fig. 2d). Although the shape approximated is symmetric, the folding path is highly asymmetric and has five distinct seams. Unlike the rectangles, which rarely break or fold, three-hole disks exhibit several characteristic deformations (Fig. 2d, lower AFM image); still, 70% (S ¼ 90) were well-formed. DNA origami is not limited to the approximation of shapes by raster fill: some shapes can be created more exactly by combining distinct raster fill domains in non-parallel arrangements. Figure 2e shows a triangle built from three separate, 2.5-turn spacing rec￾tangular domains; only single covalent bonds along the scaffold hold the domains together. But the desired equiangular triangles (upper AFM image) were rarely observed (,1%, S ¼ 199). As seen in the lower AFM image, stacking caused rectangular domains of separate triangles to bind; this effect and the flexibility of the single-bond joints at the vertices may account for the ease with which these triangles deform. To solve these problems, ‘sharp triangles’, built from trapezoidal domains with 1.5-turn spacing, were designed (Fig. 2f). The slanted edges of the trapezoids meet at the triangle vertices and allow the addition of bridging staples along these interfaces. Sharp triangles remained separated and equiangular (Fig. 2f, lower AFM image); 88% were well-formed (S ¼ 78). Even when bridging staples at the vertices were not used, a large number of sharp triangles were well-formed (55%, S ¼ 22). These ‘weakened’ sharp triangles provided the most stringent test of the estimated inter-helix gap, because too high or low an estimate would have caused gaps or overlaps between trapezoids. Gaps of 10 nm occasion￾ally appeared but overlaps were never observed, suggesting that 1 nm may be a slight underestimate of the inter-helix gap. Patterning and combining DNA origami In addition to binding the DNA scaffold and holding it in shape, staple strands provide a means for decorating shapes with arbitrary patterns of binary pixels. Given a shape, the original set of staples is taken to represent binary ‘0’s; a new set of labelled staples, one for each original staple, is used to represent binary ‘1’s. Patterns are created by mixing appropriate subsets of these strands. In this way, any desired pattern can be made. In principle, a variety of DNA modifications—for example, biotin or fluorophores—could serve as labels. Here, ‘dumbbell hairpins’ (Fig. 1d inset, Supplementary Note S6), designed to avoid dimeriza￾tion at high concentration, were added to the middle of 32-mer staples at the position of merges made during design. Depending on the merge pattern, the resulting pixel pattern was either rectilinear, with adjacent columns of hairpins on alternate faces of the shape, or staggered and nearly hexagonally packed, with all hairpins on the same face. In AFM images labelled staples give greater height contrast (3 nm above the mica) than unlabelled staples (,1.5 nm), which results in a pattern of light ‘1’ and dark ‘0’ pixels. Several patterns (Fig. 3), each with ,200 pixels, illustrate the generality of this technique. Yields of patterned origami were similar to those of unpatterned origami; for the pattern in Fig. 3a, 91% (S ¼ 85) of rectangles were well-formed. Because rectilinear patterns imaged poorly, only stag￾gered patterns were examined quantitatively. Distances measured between pairs of ‘1’ pixels in alternating columns (two pixel widths: 11.5 ^ 0.9 nm, mean ^ s.d., n ¼ 26) and adjacent rows (one pixel height: 6.6 ^ 0.5 nm, n ¼ 24) are consistent with the theoretically expected pixel size of 5.4 nm £ 6 nm. Most defects take the form of ‘missing pixels’; that is, pixels that should image as ‘1’s but image as ‘0’s instead. 94% of ‘1’ pixels (of 1,080 observed) were visualized. Whether missing pixels represent real defects or artefacts of imaging is unknown; sequential AFM images occasionally showed ‘1’ pixels that later converted irreversibly to ‘0’ pixels, suggesting tip-induced damage. Stoichiometric errors, synthetic errors, or unwanted sec￾ondary structure are not implicated for any particular strand, as the position of missing pixels appeared random (Fig. 3b, f and g). Stacking of shapes along blunt-ended helices provides an uncon￾trolled mechanism for the creation of larger structures (Fig. 3b). Instead of removing staples on the edge of a rectangle to avoid stacking (as described previously), 4-T hairpin loops (four thymines in a row, Fig. 1e, inset) or 4-T tails can be added to edge staples (Fig. 3e, f); stacked chains of 3–5 rectangles still formed (Fig. 3g), but 30% of rectangles (S ¼ 319) occurred as monomers (Fig. 3i). Without hairpins, all rectangles occurred in aggregates (Fig. 3h). Controlled combination of shapes was achieved by designing ‘extended staples’ that connected shapes along their edges. To create a binding interaction between two particular edges, extended staples were designed by merging and breaking normal staples along these edges (Supplementary Note S7). Starting with sharp triangles, this approach was used to create finite (hexagons; Fig. 3n, p, q) as well as periodic structures (triangular lattice; Fig. 3o, r–u). I note that the successful combination of shapes (unlike the successful formation of individual shapes) is in principle very sensitive to the concentrations of extended staples (which should ideally be equal to that of the scaffold). Poor stoichiometry may play a role in the poor yield of hexagons (,2%, S ¼ 70) and lattices (not measured). Discussion The scaffolded self-assembly of DNA strands has been used to create linear structures17,18 and proposed as a method for creating arbitrary patterns18,19. But the widespread use of scaffolded self-assembly, and in particular the use of long DNA scaffolds in combination with hundreds of short strands, has been inhibited by several misconcep￾tions: it was assumed that (1) sequences must be optimized20 to avoid secondary structure or undesired binding interactions, (2) strands must be highly purified, and (3) strand concentrations must be precisely equimolar. These three criteria are important for the formation of many DNA nanostructures and yet all three are ignored in the present method. For example, M13mp18 is essentially a natural sequence that has a predicted secondary structure which is more stable (lower in energy) than similar random sequences NATURE|Vol 440|16 March 2006 ARTICLES 301