Research Articles

Designer nanoscale DNA assemblies programmed from the top down

See allHide authors and affiliations

Science  24 Jun 2016:
Vol. 352, Issue 6293, pp. 1534
DOI: 10.1126/science.aaf4388

Simplifying DNA origami design

Many intricate nanostructures have been made with DNA origami. This process occurs when a long DNA scaffold develops a particular shape after hybridization with short staple strands. Most designs, however, require a difficult iterative procedure of refining the base pairing. Veneziano et al. now report algorithms that automate the design of arbitrary DNA wireframe structures. Synthesizing and structurally characterizing a variety of nanostructures allowed for verification of the algorithms' accuracy.

Science, this issue p. 1534

Structured Abstract

INTRODUCTION

Synthetic DNA can be programmed by using canonical Watson-Crick base pairing to form highly structured nanometer-scale assemblies that rival many natural protein and RNA assemblies in structural complexity. The use of a single-stranded scaffold DNA molecule that traverses the entire DNA architecture offers near-quantitative yield of target DNA-based objects, in addition to full control over their asymmetric three-dimensional (3D) structure and site-specific functionalization for applications that include cellular delivery, nanoscale photonic materials, single-molecule imaging, and structured metamaterials, among others. Design of these versatile DNA assemblies is currently limited, however, to experts who are knowledgeable in the sequence design rules needed to fold a target DNA shape.

RATIONALE

A fully autonomous design procedure that computes the single-stranded DNA (ssDNA) sequences needed to fold an arbitrary 3D shape offers the potential for broadening participation in this powerful molecular design paradigm. Toward this end, we present the algorithmic framework DAEDALUS (http://daedalus-dna-origami.org) that uses a simple surface-based representation of target 3D geometry to automatically generate the ssDNA needed to synthesize the object. Target shapes are rendered with parallel DNA duplexes that are folded to form nanoparticles of high structural fidelity that are also biologically compatible. Our approach realizes the fully automatic, top-down sequence design of nearly arbitrary structured 3D DNA assemblies based only on simple geometric representation, offering nonexperts the ability to design synthetic DNA-based molecular architectures.

RESULTS

We apply our computational algorithm to design a broad palette of 45 complex nanoparticle geometries including Platonic, Archimedean, Catalan, and Johnson solids, as well as more than 10 arbitrarily shaped solids that include both symmetric and asymmetric shapes. We apply asymmetric polymerase chain reaction (aPCR) to generate custom ssDNA scaffolds that result in high-fidelity folding of DNA nanoparticles of diverse shapes and sizes, which are also verified to fold in low salt concentrations due their open wireframe design. We use single-particle cryo–electron microscopy (cryo-EM) to obtain 3D reconstructions of six target objects that are consistent with 3D atomic model predictions within the nanometer-scale resolution obtained, and demonstrate versatility of our algorithm in realizing both rigid and toroidal-like objects. Nanoparticles are demonstrated to be stable in cellular buffers including Dulbecco’s modified Eagle’s medium (DMEM) and fetal bovine serum (FBS). Our aPCR procedure offers facile amplification of target natural or synthetic DNA scaffolds for assays requiring high amounts of material, such as the cryo-EM reconstructions realized here.

CONCLUSION

Inverse sequence design of structured DNA assemblies with full control over asymmetric 3D structure and sequence properties is a long-standing aim of molecular engineering. We present a general solution to this problem that offers the ability for nonspecialists to design and synthesize nearly arbitrary DNA-based nanoparticles using only a simple surface representation of the target object. We provide open-source software together with a versatile synthesis strategy to self-assemble complex nanoparticles that are stable in diverse buffers, including in physiological conditions, which offers the opportunity for their potential use in numerous in vitro as well as biomedical applications.

DNA nanoparticle design, synthesis, and characterization.

(Top) Top-down geometric specification of the target geometry is followed by fully automatic sequence design and 3D atomic-level structure prediction. (Bottom) aPCR is used to synthesize object-specific ssDNA scaffold for folding. Nanoparticle stability is characterized in cellular media with serum and nanoparticle 3D structure is characterized by single-particle cryo-EM.

Abstract

Scaffolded DNA origami is a versatile means of synthesizing complex molecular architectures. However, the approach is limited by the need to forward-design specific Watson-Crick base pairing manually for any given target structure. Here, we report a general, top-down strategy to design nearly arbitrary DNA architectures autonomously based only on target shape. Objects are represented as closed surfaces rendered as polyhedral networks of parallel DNA duplexes, which enables complete DNA scaffold routing with a spanning tree algorithm. The asymmetric polymerase chain reaction is applied to produce stable, monodisperse assemblies with custom scaffold length and sequence that are verified structurally in three dimensions to be high fidelity by single-particle cryo-electron microscopy. Their long-term stability in serum and low-salt buffer confirms their utility for biological as well as nonbiological applications.

DNA nanotechnology offers the ability to synthesize highly structured nanometer-scale assemblies that in principle could rival the geometric complexity found in natural protein and nucleic acid assemblies. The past decade has witnessed dramatic growth in the diversity of structured DNA assemblies that can be programmed from the bottom up to self-assemble into target shapes by complementary Watson-Crick base pairing (17). Scaffolded DNA origami is a particularly powerful means of synthesizing structured DNA assemblies, offering full control over both molecular weight and intricate nanometer-scale structure, with near-quantitative yield of the programmed product that relies on a single-stranded DNA (ssDNA) template (2, 5, 8, 9). Wireframe topologies based on the scaffolding principle have further demonstrated highly versatile control over two-dimensional (2D) and three-dimensional (3D) spatial architecture (1013).

Similar to the challenge of structure-based protein sequence design, which seeks to infer the amino acid sequence needed to fold a target protein structure of interest (14, 15), achieving a general strategy for structure-based design of synthetic DNA assemblies represents a major challenge as well as opportunity for nanotechnology. Although numerous computational design tools exist to aid in the bottom-up, manual programming of scaffolded DNA origami (16), which requires complex scaffold routing and staple design to realize a target geometry based on Watson-Crick base complementarity, only one approach offers a solution to the inverse problem of sequence design based on specification of target geometry (13). However, this approach is only semi-automated and relies on single duplex DNA arms and multiway junctions to represent polyhedral geometries, which may result in compliant and unstable assemblies that are unsuitable for many applications. Moreover, programmed geometries must be topologically equivalent to a sphere, substantially limiting its scope.

As an alternative, here we introduce the fully automatic inverse design procedure DAEDALUS (DNA Origami Sequence Design Algorithm for User-defined Structures) that programs arbitrary wireframe DNA assemblies based on an input wireframe mesh without reliance on user feedback or limitation to spherical topologies. We apply our procedure to design 35 Platonic, Archimedean, Johnson, and Catalan solids; six asymmetric structures; and four polyhedra with nonspherical topologies; each of which is specified using surface geometry alone. Designed sequences are used to synthesize icosahedral, tetrahedral, cuboctahedral, octahedral, and reinforced hexahedral structures by the asymmetric polymerase chain reaction (aPCR) for facile production of single-stranded scaffolds of custom length and sequence. Programmed objects are confirmed by cryo–electron microscopy (cryo-EM), folding, and stability assays to be both high-fidelity structurally and stable under low-salt buffer conditions important to biological as well as in vitro applications. These results demonstrate the broad applicability of our design and synthesis strategy for numerous potential applications in biomolecular science and nanotechnology, including nanoparticle (NP) delivery (17, 18), photonics (19, 20), inorganic NP synthesis (21, 22), memory storage (2325), and single-particle cryo-EM analysis (3, 2628), among others (7, 29, 30). The ability to synthesize nearly arbitrary geometric shapes that are automatically rendered from the top down should enable the broad participation of nonexperts in this powerful molecular design paradigm.

Top-down automatic sequence design

To enable the fully automatic and robust inverse design of programmed DNA assemblies, we chose to render arbitrary geometries as node-edge networks based on the DX-based wireframe motif in which interconnected edges consist of two duplexes joined by means of antiparallel double crossovers (DX) (Fig. 1) (3, 1012, 31). This strategy offered application of our procedure to any closed geometric surface, including nonspherical topologies such as a torus, provided that it can be rendered with polyhedral surface meshes. By means of this approach, the spatial coordinates of all vertices, the edge connectivities between vertices, and the faces to which vertices belong fully specify the target object (32). Standard polyhedron file formats containing this information are converted into this set of arrays, providing input to our scaffold routing and staple design procedure (Fig. 1, step i, and Fig. 2A). Programmed edges are required to consist of multiples of 10.5 base pairs (bp) rounded to the nearest nucleotide (nt), as commonly assumed in DNA origami design to satisfy the natural helicity of B-form DNA. Obeying the natural geometry of DNA ensures that no over- or underwinding in duplexes occurs (33, 34), which may otherwise result in shape distortions that force deviation from the target geometry and would require iterative, ad hoc adjustment of edge lengths and sequence design (13). Notably, our algorithm has no theoretical limitation on the length of scaffold that can be used to program the target DNA origami object.

Fig. 1 Top-down sequence design procedure for scaffolded DNA origami nanoparticles of arbitrary shape.

Specification of the arbitrary target geometry is based on a continuous, closed surface that is discretized with polyhedra. This discrete representation is used (step i) to compute the corresponding 3D graph and (step ii) spanning tree. The spanning tree is used (step iii) to route the ssDNA scaffold throughout the entire origami object automatically, which then enables (step iv) the assignment of complementary staple strands. Finally, (step v) a 3D atomic-level structural model is generated assuming canonical B-form DNA geometry, which is validated using 3D cryo-EM reconstruction.

Fig. 2 Fully automatic sequence design of 45 diverse scaffolded DNA origami nanoparticles.

(A) Face-shaded 3D representations of geometric models used as input to the algorithm. (B) Three-dimensional atomic models of DNA-rendered nanoparticles for (blue) Platonic, (red) Archimedean, (green) Johnson, (orange) Catalan, and (violet) miscellaneous polyhedra generated using the automatic scaffold routing and sequence design procedure (particles are not shown to scale). Miscellaneous polyhedra include (first column) heptagonal bipyramid; enneagonal trapezohedron; small stellated dodecahedron, a type of Kepler-Poinsot solid; rhombic hexecontahedron, a type of zonohedron; Goldberg polyhedron G(2,1) with symmetry of Papillomaviridae; (second column) double helix; nested cube; nested octahedron; torus; and double torus. Platonic, Archimedean, and Johnson solids each have 52-bp edge length, Catalan solids and the first column of miscellaneous polyhedra have minimum 42-bp edge length, and the second column of miscellaneous polyhedra have minimum 31-bp edge length. Thirty of the 45 structures shown have scaffolds smaller than the 7249-nt M13mp18, whereas 15 have scaffold lengths that exceed it (table S2).

Representing the target geometry as a polyhedral mesh that satisfies the preceding design criteria guarantees that a single-stranded scaffold can be routed uniquely throughout the entire object with an Eulerian circuit, without modifications to the target geometry. From the mesh, the graph of the target structure is computed, containing the vertex, edge, and face information (Fig. 1, step i). Scaffold routing is then assigned by using a spanning tree (Fig. 1, step ii) generated with Prim’s algorithm (fig. S1) (35). Each of the edges that is a member of the spanning tree is assigned no scaffold crossover, whereas each remaining edge is assigned one scaffold crossover. For a given graph, every spanning tree therefore corresponds to a unique scaffold routing, where, by default, Prim’s algorithm generates a maximally branching spanning tree (figs. S1 and S2) that has been suggested for folding 2D nets to self-assemble more reliably than linear trees (36). The use of DX arms to represent edges ensures that a solution to the scaffold routing problem is obtained efficiently in solution time that scales as E log V, where E and V are the numbers of edges and vertices, respectively (35). To complete the scaffold routing, a scaffold crossover is placed at the center of each edge that is not part of the preceding spanning tree. From the scaffold crossover positions and spanning tree, the circuit for the scaffold routing is determined, specifying the order in which the scaffold visits each vertex and crossover while ensuring that the scaffold does not intersect itself at vertices (Fig. 1, step iii) (37). A linear scaffold nick position is set to ensure that it is noncoincident with crossovers and other nicks, with the polarity of its routing chosen to be counterclockwise around each face because of the preference of the major groove to orient inwards at vertices (38). Thus, use of our spanning tree approach enables fully automatic conversion of the input polyhedral geometry to full scaffold routing based on the single circuit that traverses each duplex once.

With the scaffold routing determined, staple strands are assigned automatically by using distinct rules for vertex versus edge staples, enabling the assignment of staple strand sequences assuming Watson-Crick base complementarity (Fig. 1, step iv) (32). Vertex staples hybridize to the scaffold in the 10 to 11 bp closest to vertices, comprising 52- and 78-nt staples, the numbers of which are determined by the degree of the vertex (32). Edge staples occupy the intermediate regions, spanning across scaffold crossovers to help rigidify each edge by creating crossovers every 10, 11, or 21 bp (32). Finally, the positions and orientations of each nucleotide are modeled to predict the 3D structure of the NP (Fig. 1, step v, and fig. S3) (32). Critically, in contrast with previous tile-based approaches that used this DX-motif to synthesize NPs of diverse form (3), the use of a ssDNA scaffold that is routed throughout the entire object in our strategy offers near-quantitative yield of the final product in its self-assembly without dependence on relative multiarm junction tile concentrations, as well as full control over DNA sequence. This final feature is essential to biomolecular applications that use spatially specific asymmetric sequence programming for protein or RNA scaffolding, as well as other chemical functionalization.

To test the generality and robustness of our design procedure to be applied to diverse polyhedral geometries, we first applied it to design Platonic solids, which have equal edge lengths, angles, and vertex degrees; followed by geometries of increasing complexity, including Archimedean solids with unequal vertex angles; Johnson solids that include heterogeneity in vertex degrees; and Catalan solids that have unequal edge lengths (Fig. 2). Applicability to asymmetric and nonconvex objects specified on the basis of surface geometry alone (13) was also confirmed (Fig. 2 and fig. S4), in addition to nonspherical topologies, including a nested cube, a nested octahedron, and tori that have not been previously realized experimentally (3, 12) and cannot be solved computationally with existing procedures (13). Fifteen of these structures required scaffolds longer than the 7249-nt M13mp18 (39), for which random sequences of appropriate length were generated (Fig. 2B). Taken together, these examples illustrate the broad ability of our procedure to automatically generate complex scaffold and staple routings for diverse geometries based on top-down geometric specification alone (Fig. 2).

Custom scaffold production with asymmetric PCR

To investigate the synthetic yield and homogeneity of self-assembled objects programmed with our computationally generated scaffold and staple designs, we used the aPCR (40) to generate object-specific scaffolds for folding (figs. S6 and S7) (32). These custom scaffolds minimize excess ssDNA in the final structure, which may result in nonspecific object aggregation or otherwise interfere with folding as well as downstream chemical functionalization (Fig. 3 and fig. S8) (2, 41, 42). Monodispersity of multiple custom linear short scaffold strands synthesized, ranging from 450 to 3400 nt, was first confirmed by gel electrophoresis of aPCR products based either on the M13pm18 ssDNA plasmid or double-stranded DNA (dsDNA) fragments as templates (Fig. 3). Custom scaffolds were used to fold tetrahedra of 31-, 42-, 52-, 63-, and 73-bp edge lengths in addition to an octahedron, two pentagonal bipyramids (42- and 52-bp edge lengths), a cube, a reinforced cube, an icosahedron, a cuboctahedron, and a nested cube (Fig. 4 and figs. S8 to S18) (32). Folded objects were analyzed by single-particle atomic force microscopy (AFM) after purification, which confirmed their folding yields of up to 90% (table S3) and particle homogeneity that is characteristic of scaffolded DNA origami objects (Fig. 2 and figs. S12 to 18). Notably, application of our aPCR approach offers folded sample purity that is similar to that of existing synthesis strategies that use restriction enzymes to generate subfragment scaffolds (Fig. 3 and figs. S6 and S7) (43), yet without dependence on restriction sites and with higher synthetic yield. Redesign of vertex staple nicks to be positioned at crossovers instead of interior segments of duplexes also resulted in increased folding stability (fig. S19) (32). Thus, diverse polyhedral origami objects from 270 kDa to 2 MDa programmed with our top-down, inverse sequence design procedure self-assembled robustly by using scaffolds of custom length and sequence, which may be natural or synthetic.

Fig. 3 aPCR strategy to synthesize custom ssDNA scaffolds.

(A) ssDNA scaffolds of custom length and sequence for each target structure are amplified using either a single- or double-stranded DNA template mixed with appropriate primer pairs consisting of 50× sense primer and 1× antisense primer concentration relative to the scaffold concentration. (B) Amplified ssDNA products are purified and analyzed by agarose gel electrophoresis.

Fig. 4 Folding and 2D structural characterization of scaffolded DNA origami nanoparticles.

(A) Characterization of folding for five platonic solids (52-, 63-, and 73-bp edge-length tetrahedra; 52-bp edge-length octahedron; 52-bp edge-length icosahedron) by agarose gel electrophoresis, AFM and cryo-EM. (B) Characterization of folding for one Archimedean solid (52-bp edge-length cuboctahedron), one miscellaneous solid (reinforced cube with 52- and 73-bp edge lengths), and one Johnson solid (42- and 52-bp edge-length pentagonal bipyramid), using agarose gel electrophoresis, AFM, and cryo-EM. M: DNA marker; sc: custom ssDNA scaffold. Scale bars: 20 nm for AFM and cryo-EM and 10 nm for atomic models.

Cryo-EM 3D reconstruction

Next, we evaluated the structural fidelity of programmed origami objects by using single-particle cryo-EM and 3D reconstruction (Fig. 5 and figs. S20 to S27) (32). Cryo-EM imaging confirmed the abundance of well-folded single DNA NPs of expected sizes and shapes, which were used to generate 3D density maps (Fig. 5). All maps were validated by matching class averages, map projections, and tilted-pair images (figs. S28 to S33) (32). Three-dimensional atomic structures of programmed DNA origami objects were predicted by using a rigid duplex model in which each edge is composed of two parallel duplexes with the central axis of each edge meeting at a single point that is specified in the input graph (32). Quantitative comparison of model predictions with cryo-EM reconstructions revealed favorable agreement, with correlations ranging from 0.73 for the tetrahedron to 0.92 for the cuboctahedron, and resolutions comparable to those of previous wireframe DX designs based on tile assembly that produces polydisperse samples (Fig. 5A and fig. S28) (3). Interestingly, the tetrahedron exhibited outward bowing of its edges (Fig. 5B and fig. S29), which has been previously observed for single-duplex edges (44) and may be attributable to its smaller acute interior angles compared with those of other designed objects, which may result in steric overlap of adjoining DNA duplexes at vertices. Structure-based molecular modeling is of interest to test possible redesigns of the tetrahedral object that may improve agreement with the target, input geometry. Comparison of the cuboctahedron reconstruction with two competing models for vertex geometry suggested that duplex overlap is preferred over backbone stretching in determining overall equilibrium shape (Fig. 5D and figs. S3 and S30).

Fig. 5 Three-dimensional structural characterization of scaffolded DNA origami nanoparticles using cryo-EM reconstruction and comparison with model predictions.

(A) Programmed edges of the 52-bp edge-length icosahedron are straight and vertices are rotationally symmetric, as designed. Cryo-EM resolution is 2.0 nm and correlation with the model is 0.85 (55). (B) Edges of the 63-bp edge-length tetrahedron reveal outward bowing (arrow) attributable to its acute interior angles that might result in steric hindrance. Cryo-EM resolution is 1.8 to 2.2 nm and correlation with the model is 0.72. (C) A 15° right-handed twist is visible at each vertex (arrow) of the 52-bp edge-length octahedron, which suggests that the structure folds as prescribed rather than “inside-out.” Cryo-EM resolution is 2.5 nm and correlation with the model is 0.89. (D) A 52-bp edge-length cuboctahedron has unequal angles between edges that meet at vertices (arrows), which supports a rigid-duplex model in which phosphate backbone stretch is minimized (32). Cryo-EM resolution is 2.9 nm and correlation with the model is 0.92. (E) The addition of 73-bp reinforcing struts to a simple cube of 52-bp edge-length increases its structural homogeneity to produce a 3D reconstruction with 915 particles. With the reinforcement, the particles maintain right-angled vertices (upper arrow). The diagonal edges form a tetrahedral symmetry that exhibits outward bowing (lower arrow). Cryo-EM resolution is 2.7 nm and correlation with model is 0.72. (F) Three-dimensional reconstruction of a nested cube within a cube that has nonspherical topology. The 73-bp edge-length outer cube is connected to a 32-bp edge-length inner cube by eight 31-bp edge-length diagonals. Cryo-EM resolution is 4.0 to 4.5 nm and correlation with the model is 0.74. Scale bars: 5 nm.

Notably, cryo-EM reconstructions suggested that origami objects assembled as designed instead of “inside-out” while satisfying programmed Watson-Crick base pairing from sequence design (32). This result reaffirms the suitability of our sequence design algorithm to choose to point the major groove inward at vertices, which was based on the previous observation that DNA origami folds in this manner (38). These two variants are distinguished experimentally by the designed asymmetry that places a 1-bp overhang on the 5′ end of the scaffold on each edge to keep the scaffold and staple crossovers perpendicular to the helical axis. Although 3′-end overhangs would have also achieved this, our choice of 5′-end overhangs anticipates a right-handed twist of the central hole of each vertex when the structures fold as prescribed, which is supported experimentally for the octahedron in which the chirality of the vertex twist is apparent (Fig. 5C and fig. S31). Notwithstanding, the extent to which the edges twist is not predicted by a simple geometric model that assumes that edges meet straight-on at vertices, deviating from the ~15° right-handed twist observed experimentally.

Although individual DX-based polyhedral edges are expected to be more structurally rigid than single-duplex edges used previously (13), some surfaced-rendered polyhedral objects may still be expected to be structurally compliant because of their overall geometry, as in the case of the simple cube with 52-bp edge lengths that was initially observed by AFM and cryo-EM to fold into heterogeneous single particles (Fig. 2 and figs. S34 to S36). To test the ability of our algorithm to reprogram such flexible objects—to render them rigid for applications including single-particle cryo-EM at subnanometer resolution that requires highly homogeneous structural populations for reconstruction—we applied it to redesign a new, reinforced cube (Fig. 5E and fig. S32). We hypothesized that the observed heterogeneity in the simple cube resided in its vertex flexibilities, in which right-angle vertices can change in concert while maintaining constant edge lengths in shearing modes. To eliminate this compliant mode of deformation, we sought to introduce cross-bars on each face. Reinforced structures folded into highly homogeneous hexahedral objects that enabled single-particle cryo-EM imaging and reconstruction that agreed with 3D atomic model predictions (Fig. 5E and fig. S32). Surprisingly, constraining edge lengths to be multiples of 10.5 bp did not prove restrictive in achieving this target redesign, which required introduction of 73-bp cross-bars across each face (Fig. 5E and fig. S32). Although in principle this agreement may be attributable to the overall symmetry of the object, further bowing of edges may equally have been anticipated on the basis of steric overlap or duplex crowding at vertices. Given that uniform triangular polyhedra or other simple rules do not apply generally to the ultimate aim of rendering mechanically rigid surface geometries, generalizing this preceding redesign strategy from the hexahedron to other objects represents an important future challenge and demonstrates the versatility of our top-down sequence design procedure.

To test the ability of our algorithm to design scaffolded DNA origami objects of nonspherical topology and with internal structure that have not been previously realized experimentally (3, 12) and cannot be designed with existing top-down computational design procedures (13), we synthesized the nested cube (Fig. 2B) and reconstructed its 3D structure using cryo-EM (Fig. 5F and fig. S33). Although the folding yield and cryo-EM resolution of this object were somewhat lower than for other objects (32), possibly due to the flexibility that is intrinsic to cubelike geometries, noted above, the ability to program internal structure substantially broadens the scope for synthesizing such complex scaffolded origami objects (45) in a one-step folding reaction.

Folding and stability characterization

An important limitation of DNA origami, for biological as well as in vitro applications, has been the requirement of high concentrations of either magnesium or monovalent cations for their folding and stability (42, 46), which was recently shown to be alleviated by the use of single-duplex edge meshworks that fold and are stable in physiological buffer and salt conditions (13). Investigation of the folding properties of DX-based objects synthesized here revealed that objects fold effectively in cation concentrations as low as 4 mM Mg2+ and 500 mM Na+ (Fig. 6 and figs. S37 to S48), as well as in phosphate-buffered saline (PBS) alone (figs. S49 to S53). Although these results mirror those for wireframe structures employing single-duplex edges (13), our use of DX arms and multiway junctions (11, 12) here results in structurally stable assemblies of enhanced rigidity that are crucial to many applications. To test the utility of our objects for cellular assays, after folding in TAE-Mg2+, particles were transferred to PBS and Dulbecco’s modified Eagle’s medium (DMEM) containing 0 to 10% fetal bovine serum (FBS), where they were found to be stable for at least 6 hours (Fig. 7 and figs. S54 to S59). When transferred to salt-free solution after folding, however, particles were observed to be unstable, confirming the well-known importance of a minimal amount of added salt for their longer-term stability (Fig. 7 and fig. S58). Interestingly, folding of the 52-bp tetrahedron with the full M13 scaffold instead of its custom scaffold also required higher magnesium concentrations to achieve folding, reaffirming the importance of the use of scaffolds that eliminate excess ssDNA (fig. S8).

Fig. 6 Characterization of scaffolded DNA origami nanoparticle folding in variable concentrations of added salt.

Characterization of folding of the 52-bp edge-length pentagonal bipyramid in increasing magnesium chloride (MgCl2) and sodium chloride (NaCl) concentration using 2% agarose gel electrophoresis and AFM imaging. Critical concentrations for folding are 4 mM MgCl2 and 500 mM NaCl in TRIS-acetate (pH 8.0). M: DNA marker; sc: custom ssDNA scaffold. Scale bars: 30 nm.

Fig. 7 Characterization of scaffolded DNA origami nanoparticle stability in physiological buffer and serum.

Agarose gel electrophoresis and AFM structural characterization of the 52-bp edge-length pentagonal bipyramid after 6 hours in PBS, TAE (without added NaCl or MgCl2), and DMEM buffer with increasing concentration of FBS (0, 2, and 10%) after folding in TAE-Mg2+ buffer (12 mM MgCl2) followed by buffer exchange. Stability is observed for structures in PBS buffer but not in TAE owing to the absence of salt, which demonstrates the importance of a minimal salt concentration for stability. AFM imaging reveals the presence of intact objects after 6 hours in DMEM media in the presence of 2 to 10% FBS despite partial degradation observed in agarose gel electrophoresis. Scale bars: 30 nm.

Outlook

Structure-based, rational design of macromolecular assemblies, including both nucleic acids and proteins, is a long-standing aim of nanotechnology and biological engineering. Unlike proteins, which contain a myriad of specific and nonspecific interresidue interactions that determine their local and global folds, and RNA, which exhibits promiscuity in base pairing and secondary structure, synthetic DNA assemblies are well established to be highly programmable with Watson-Crick base pairing alone (1, 2). In particular, wireframe polyhedral geometries offer the powerful ability to program nearly arbitrary 3D geometries on the nanometer scale, limited only by current size constraints imposed by single-stranded scaffold lengths. This important and versatile class of topologies therefore has broad potential for programming complex nanoscale geometries, including biomimetic systems inspired by viruses and photosynthetic systems, as well as other highly evolved natural macromolecular assemblies. Achieving full automation of inverse sequence design with this versatile wireframe approach has the potential to realize the original vision of Ned Seeman to program nanoscale materials with full 3D control over positioning of all atomic-level groups (1, 47). As a major advance in this direction, we developed a top-down, geometry-driven sequence design procedure that uses a spanning tree algorithm to determine scaffold crossover positions. This enables efficient and unique routing of the single-stranded scaffold throughout any target origami object of arbitrary polyhedral shape and molecular weight, as well as automated staple assignment to enable custom synthesis of programmed origami objects of near-quantitative yield and high-fidelity atomic-level structure. Asymmetric PCR is demonstrated to provide full control over scaffold sequence and length, verified between 449 to 3356 nt, and use of the DX-based design further confers folding capacity and stability under diverse conditions, including cell-compatible buffers. Our strategy to realize the top-down design of nanoscale DNA assemblies offers full control over both 3D structure and local sequence which, together with our broadly usable software and experimental protocols, provides a versatile approach to the design of functionalized DNA objects of nearly arbitrary shape for photonic applications (19, 48, 49), including self-assembled superlattices (50), cellular delivery (18, 51), memory storage devices (2325), and single-particle cryo-EM reconstruction assays (3, 2628) for proteins and RNAs that are not otherwise amenable to crystallography or nuclear magnetic resonance. Possible future generalizations of our approach may include rendering polyhedral networks with edges of arbitrary cross-section (4) for applications requiring further enhanced NP rigidity and closed surface topologies, such as inorganic nanoparticle synthesis (21, 22) and hybrid nanoscale materials (52).

Materials and methods

Top-down sequence design

DAEDALUS is available for use as open-source software (32) and online at http://daedalus-dna-origami.org. Additional details can be found in Supplementary Materials (32) and in documentation provided with the software.

Single-stranded DNA scaffold amplification using aPCR

aPCR was performed with a sense primer concentration of 1 μM, an antisense primer concentration of 20 nM (table S2), 30 ng of M13mp18 ssDNA template or 10 ng of dsDNA gBlocks, 200 μM deoxynucleotide triphosphates mixed in a final volume of 50 μl and 1 unit of Accustart Taq DNA polymerase HiFi. The aPCR program used is as follows: 94°C, 1 min for the initial denaturation; followed by 30 to 40 cycles of 94°C, 20 s; 55° to 59°C, 30 s; 68°C, 1 min per kilobase to amplify. PCR products were run through 1% low–melting temperature agarose gel and were extracted and purified with Zymoclean Gel DNA recovery kit.

DNA origami assembly and purification

In a one-pot reaction, 5 to 40 nM scaffold strand was mixed with 50 to 800 nM of a staple strands mix in a Tris-acetate EDTA-MgCl2 (TAE-Mg2+) buffer [40 mM Tris, 20 mM acetic acid, 2 mM EDTA, 12 mM MgCl2 (pH 8.0)]. Annealing was performed in a thermal cycler with the following program: 95°C for 5 min, 80° to 75°C at 1°C per 5 min, 75° to 30°C at 1°C per 15 min, and 30° to 25°C at 1°C per 10 min. Fresh DNA origami solutions were purified with an Amicon Ultra-0.5 ml centrifugal filter [molecular weight cut-off (MWCO) 100 kDa] to remove excess of staple strands and stored at 4°C.

Stability experiments

After buffer exchange with Amicon Ultra-0.5 (MWCO 100 kDa) DNA origami stability was evaluated for 6 hours in TAE, PBS, or DMEM buffer complemented with FBS.

Agarose gel electrophoresis

Samples were loaded in 2% agarose gel in Tris-acetate EDTA buffer supplemented with 12 mM MgCl2 and prestained with EtBr. Gels were run on a BioRad electrophoresis unit at 4°C for 3 to 4 hours under a constant voltage of 70 V. Gels were imaged with a Gene flash gel imager (Syngene, Inc.), and yield was estimated by analyzing the band intensity with the Gel Analyzer program in the ImageJ software (53).

Atomic force microscopy (AFM)

Samples for AFM were annealed at a concentration of 5 nM in TAE-Mg2+ (12 mM) buffer and subsequently purified before AFM imaging. A 1.5-μl volume of the samples was deposited onto freshly peeled mica (Ted Pella, Inc.), and 10 μl of 1× TAE-Mg2+ buffer were added to enlarge the solution drop to cover the whole mica surface. A 1.5-μl volume of NiCl2 (100 mM) was added to the drop immediately. After leaving about 30 s for adsorption to the mica surface, 70 μl of 1× TAE-Mg2+ buffer were added to the samples and an extra 40 μl of the same buffer were deposited on the AFM tip. The samples were scanned in“ScanAssyst mode in fluid” by using an AFM (Dimension FastScan, Bruker Corporation, Inc.) with SCANASSYST-FLUID+ tips.

Cryo-EM

Fifteen to 25 tubes of samples were freshly annealed at a scaffold concentration of 20 nM in 50 μl of TAE-Mg2+ buffer, pooled, purified, and concentrated with an Amicon column Ultra-0.5 ml centrifugal filter (MWCO 100 kDa) to a final volume of 20 to 30 μl. Two microliters of the freshly concentrated DNA nanostructure solution were applied onto the glow-discharged 200-mesh Quantifoil grid, blotted for 1.5 s, and rapidly frozen in liquid ethane with a Vitrobot Mark IV (FEI). All grids were screened on a JEM2200FS cryo–electron microscope (JEOL) operated at 200 kV with in-column energy filter with a slit of 20 eV. Micrographs of the icosahedron, tetrahedron, cuboctahedron, and octahedron were recorded with a direct detection device (DDD) (DE-20 4k×5k camera, Direct Electron, LP) operating in movie mode at a recording rate of 25 raw frames per second at 25,000× microscope magnification (corresponding to a calibrated sampling of 2.51 Å per pixel) and a dose rate of ~21 electrons s–1 Å–2 with a total exposure time of 3 s. Micrographs of the reinforced cube and the nested cube were recorded on a 4k×4k charge-coupled device camera (Gatan, Inc.) at 40,000× microscope magnification (corresponding to a calibrated sampling of 2.95 Å per pixel) and a dose rate of ~20 electrons s–1 Å–2 with a total exposure time of 3 s. A total of 180 images for the icosahedron, 91 images for the tetrahedron, 101 images for the cuboctahedron, 100 images for the octahedron, 36 images for the reinforced cube, and 152 images for the nested cube were collected with a defocus range of ~1.5 to 4 μm.

Single-particle image processing and 3D reconstruction

Particle images recorded on a DE-20 detector were motion corrected and radiation damage compensated (24). The motion correction was done with running averages of three consecutive frames by using the DE_process_frames.py script (Direct Electron, LP). Single-particle image processing and 3D reconstruction was performed with the image processing software package EMAN2 (54). EMAN2 was used for initial micrograph evaluation, particle picking, contrast transfer function correction, 2D reference-free class averaging, initial model building, and 3D refinement. All initial models for the DNA origami objects were built from 2D reference-free class averages. For final refinement, 3650 particles were used for the icosahedron, 2183 particles for the tetrahedron, 1758 particles for the cuboctahedron, 2678 particles for the octahedron, 915 particles for the reinforced cube, and 2008 particles for the nested cube, applying icosahedral, tetrahedral, octahedral, octahedral, tetrahedral, and octahedral symmetries, respectively. Resolutions for the final maps were estimated with the 0.143 criterion of the Fourier shell correlation curve without any mask. A Gaussian low-pass filter was applied to the final 3D maps displayed in the UCSF Chimera software package (55). Correlation of each map with its corresponding atomic model is calculated by the UCSF Chimera fitmap function.

Tilt-pair validation for the cryo-EM map (56) was performed by collecting data at two goniometer angles, 0° and –10°, for each region of the grid. The test was performed with the e2tiltvalidate.py program in EMAN2. Additional details on the tilt-pair validation are provided in table S3.

Quantitative PCR thermal analysis

Quantitative PCR (QPCR) analyses were performed in a Roche LightCycler® 480. A scaffold concentration of 80 nM was used for the tetrahedron folding analysis, and the concentrations of each strand were adjusted to 1 μM for the three-way junctions. Samples were complemented with 1× final concentration of SYBR Green in a final volume of 20 μl. Fluorescence curves obtained were analyzed with first-order derivatives to identify transition temperatures.

Supplementary Materials

www.sciencemag.org/content/suppl/2016/05/25/science.aaf4388.DC1

Materials and Methods

Supplementary Text S1 to S7

Figs. S1 to S59

Tables S1 to S27

Movies S1 to S7

References (5764)

DAEDALUS Software Package

References and Notes

  1. See supplementary materials on Science Online.
Acknowledgments: Funding from the Office of Naval Research (ONR) N000141410609, the Human Frontier Science Program (HFSP) RGP0029/2015, and National Science Foundation (NSF-EAGER) CCF-1547999 to M.B.; the National Science Foundation (NSF CMMI) 1334109 to M.B. and H.Y.; and the National Institutes of Health (NIH P41GM103832 and P50GM103297) to W.C. is gratefully acknowledged. Patent applications have been filed by the MIT Technology and Licensing Office on behalf of M.B., R.V., and S.R. related to this work. M.B. is scientific founder of Structured DNA Technologies LLC. K. Pan is acknowledged for technical assistance with atomic-level rendering of origami models and with website construction; C. Hill for technical assistance with website construction; V. Kaushik, A. Leed, and M. Adendorff for technical assistance with QPCR; and S. Ludtke for advice on cryo-EM image processing. Author contributions: R.V. designed and implemented the aPCR protocol; conducted the aPCR, the agarose gel electrophoresis characterization, and the QPCR experiments; prepared samples for AFM and cryo-EM; and analyzed the data. S.R. designed the algorithm; coded and implemented the algorithm computationally; analyzed the atomic models and cryo-EM maps; performed the QPCR experiments; and analyzed the data. K.Z. performed the cryo-EM imaging and the 3D reconstruction and analyzed the data. F.Z. performed the AFM imaging and analyzed the data. H.Y. supervised the AFM imaging and interpreted the data. W.C. supervised the cryo-EM imaging and interpreted the data. M.B. designed and supervised the study and interpreted the results. S.R., R.V., and M.B. wrote the manuscript. All authors commented on and edited the manuscript. Data repository: Cryo-EM maps have been deposited to EMDataBank (Tetrahedron EMD-3408; Icosahedron EMD-3409; Octahedron EMD-3410; Cuboctahedron EMD-3411; Reinforced cube EMD-3412; Nested cube EMD-3413).
View Abstract

Navigate This Article