Encoding hierarchical assembly pathways of proteins with DNA

Significance

Structural sophistication in nature would not be possible without hierarchical assembly: the concept that an initial building block (e.g., polypeptide) contains all necessary structural and chemical information to determine its assembly along a multistep pathway to generate a complex architecture (e.g., viral capsid). Translating this concept to synthetic systems is an outstanding challenge. Here, we use DNA—the blueprint of life—to direct the hierarchical assembly of proteins. Through DNA design, we can change the directionality of protein assembly and pathway by which protein–DNA conjugates will assemble as well as realize distinct structures by directing assembly along different pathways. These findings will facilitate the assembly of protein–DNA materials with structural complexity more closely approaching that observed in nature.

Abstract

The structural and functional diversity of materials in nature depends on the controlled assembly of discrete building blocks into complex architectures via specific, multistep, hierarchical assembly pathways. Achieving similar complexity in synthetic materials through hierarchical assembly is challenging due to difficulties with defining multiple recognition areas on synthetic building blocks and controlling the sequence through which those recognition sites direct assembly. Here, we show that we can exploit the chemical anisotropy of proteins and the programmability of DNA ligands to deliberately control the hierarchical assembly of protein–DNA materials. Through DNA sequence design, we introduce orthogonal DNA interactions with disparate interaction strengths (“strong” and “weak”) onto specific geometric regions of a model protein, stable protein 1 (Sp1). We show that the spatial encoding of DNA ligands leads to highly directional assembly via strong interactions and that, by design, the first stage of assembly increases the multivalency of weak DNA–DNA interactions that give rise to an emergent second stage of assembly. Furthermore, we demonstrate that judicious DNA design not only directs assembly along a given pathway but can also direct distinct structural outcomes from a single pathway. This combination of protein surface and DNA sequence design allows us to encode the structural and chemical information necessary into building blocks to program their multistep hierarchical assembly. Our findings represent a strategy for controlling the hierarchical assembly of proteins to realize a diverse set of protein–DNA materials by design.

Hierarchical assembly is integral to the structural complexity and function of materials and systems that occur in nature. Muscle tissue (1), amyloid fibrils (2), and collagen networks (3) are all examples of highly organized supramolecular architectures that arise from bottom-up, multistep, regulated assembly processes. The well-controlled sequence of assembly steps along a given pathway and the specificity of interactions between components are critical to the observed structural complexity and diversity (4, 5). While nanoscale hierarchical assembly is prevalent and important in nature, and our ability to control the bottom-up assembly of synthetic nanoscale building blocks has been transformed over the past two decades (6–8), we are still limited in what can be programmed through hierarchical mechanisms (9, 10). This is due to difficulties in defining the number, type, and location of multiple interactions on synthetic building blocks, as well as limitations in controlling the interplay between orthogonal interactions to achieve a desired assembly pathway (11). The development of tools and strategies to program multistep assembly pathways of nanoscale building blocks would redefine how we control the bottom-up synthesis of materials and accelerate the discovery of novel structures with desirable properties and functions (12, 13). In this work, we address this gap by spatially encoding programmable interacting ligands (DNA) onto the surface of chemically addressable building blocks (proteins).

Proteins are an important class of nanoscale building block because of their structural and functional roles in biology. As such, developing methods to synthetically engineer new materials from proteins is a common goal in the fields of synthetic biology, chemistry, and materials science, with diverse applications from catalysis (14) to immune evasion (15) and biological delivery (16). The chemical complexity of protein surfaces defines specific recognition between protein interfaces and is key to the hierarchical assembly processes observed in nature. However, their complex surfaces make it challenging to design protein building blocks that will transform into targeted materials by traversing an intended assembly pathway. While powerful de novo design strategies have been utilized to create proteins with predetermined interfaces and assembly outcomes (17, 18), this approach inherently deviates from the pool of naturally occurring protein building blocks that could be utilized for materials engineering. Other strategies have relied on introducing controlled molecular interactions to the surfaces of proteins ranging from metal coordination chemistries (19–21) to hydrophobic (22) and host–guest interactions (23, 24). Despite significant innovation in manipulating surface interactions through chemical modifications, less attention has been paid to designing protein building blocks that can undergo multistep assembly pathways mimicking those in nature (25–27), because it remains challenging to realize interactions that are simultaneously specific, orthogonal, and have tunable strengths. Indeed, methods to define interaction location and type on the surface of a building block, in conjunction with an understanding of how to control and regulate each interaction independently, are needed to successfully program hierarchical assembly pathways. Although a growing body of literature has examined assembly pathways in the context of protein crystal polymorphism (28, 29), the ability to design directional, multistep assembly processes remains elusive.

In addition to programming the structures of protein assemblies using DNA origami templates with specific, directional interactions (16, 29–33), our group and others have shown that DNA ligands chemically tethered to the surfaces of proteins at specific locations can drive the assembly of proteins into one-dimensional (1D), two-dimensional (2D), and three-dimensional (3D) assemblies and crystals (34–45). Protein mutagenesis has been used to site-specifically encode multiple, orthogonal DNA interactions onto protein surfaces to program directional assembly (46). Furthermore, the programmable recognition properties of DNA surface ligands have been utilized to control the polymerization pathway of proteins (47). However, these examples all rely on a single assembly step to reach their target structure and do not teach us how to create more complex materials from multistep, hierarchical assembly of proteins, such as those observed in nature. Indeed, even when multistep DNA assembly was demonstrated for inorganic nanoparticles, the second assembly step could only be induced by chemical modification of the structure formed after an initial assembly step and the addition of more nanoparticle building blocks to the system (48).

We hypothesized that, if we could define the specificity, strength, and spatial distribution of multiple specific DNA interactions on the surface of a protein, we would be able to synthesize protein building blocks that undergo spontaneous, programmed, multistep assembly processes. Here, by defining the chemical anisotropy of a protein’s surface via mutagenesis, we define DNA interactions spatially, that is, axially or equatorially with respect to the geometry of an anisotropic protein (Scheme 1A). Through careful DNA design, we modulate the relative interaction strengths of the axial and equatorial faces such that assembly via strong interactions in a single direction leads to an emergent, second interaction that can program assembly in an orthogonal direction (Scheme 1B). The emergence of this second interaction is a hallmark of hierarchical assembly observed in nature and is responsible for directing the assembly of proteins along specific, multistep pathways. This study focuses on articulating this concept for programming the assembly of nanoscale building blocks along specific, hierarchical pathways, rather than obtaining arbitrarily high registry in 2D and 3D protein materials.

Scheme 1.

Design of Sp1m chemical surface and proposed hierarchical assembly schemes. (A) Native Sp1 (Left) presents multiple primary amines (lysines and N termini, blue) and no cysteines (red) on its surface. Three mutations were designed to remove two native lysines and introduce one cysteine per subunit. Due to the dodecameric structure of Sp1m, these mutations define the chemical anisotropy across the protein surface with amine residues only on the axial face and cysteines located only on the equatorial face. (B) Proposed assembly schemes for building blocks containing strong or weak surface interactions at their axial or equatorial positions. Strong interactions direct the first stage of assembly, leading to multivalency among weak interactions that direct the second stage of assembly.

Results and Discussion

Design and Synthesis of Sp1m-DNA Building Blocks.

To explore our hypothesis, we selected stable protein 1 (Sp1, Protein Data Bank [PDB]: 1TR0), a symmetric homododecameric protein with pseudo hexagonal-prism geometry (49). Sp1 was chosen as a model system due to its well-defined, anisotropic shape and high symmetry. To align the chemical anisotropy of the protein’s surface to the shape anisotropy of the protein (Scheme 1A), we recombinantly expressed a mutant (Sp1m) with 24 surface-accessible primary amines and 12 thiols located axially and equatorially, respectively (SI Appendix, Table S1). Importantly, this mutant retains the geometry of the native protein as characterized by transmission electron microscopy (TEM, Fig. 1B). The designed chemical anisotropy was then exploited to introduce orthogonal DNA ligands to the axial and equatorial faces (Fig. 1A). In a typical synthesis, the equatorial cysteine residues were first modified with a thiol-reactive hetero-bifunctional crosslinker (Linker 1, Fig. 1C and SI Appendix, Scheme S2) to install azide functional groups. Near-complete (>95%) modification of the cysteine residues was confirmed using matrix-assisted laser desorption-ionization time-of-flight mass spectrometry (MALDI-TOF MS, Fig. 1D). The axial primary amines were subsequently reacted with an amine-reactive hetero-bifunctional crosslinker (Linker 2, Fig. 1C) to install tetrazine functional groups. Although there are two primary amines per monomeric subunit (lysine K74 and N terminus), MALDI-TOF MS analysis indicated high yield (>90%) modification of only a single primary amine per subunit. Hence, only 12 of the 24 surface-accessible primary amines are available for DNA attachment. High-resolution, top-down proteomic evaluation of this species revealed that the N-terminal primary amine was modified, with marginal to no functionalization of K74 (SI Appendix, Fig. S7 and Table S3). We attribute the low reactivity of K74 to its involvement in hydrogen bonding with an adjacent subunit (SI Appendix, Fig. S8).

Fig. 1.

Synthesis and characterization of Sp1m-DNA conjugates. (A) Sp1m (1) was modified with DNA in three steps: (i) cysteines were first modified with Linker 1 (C) through a thiol-maleimide Michael addition click reaction to give Sp1m-N₃ (2); (ii) primary amines were then modified with Linker 2 (C) to generate 3 through reaction with an NHS-activated ester; (iii) TCO- and DBCO-modified DNA were reacted with 3 in one pot to generate an Sp1m-DNA building block (4). (B) Negative-stain TEM of 1. (Scale bar, 50 nm.) (Lower) Comparison of a model of Sp1m with a magnified region from the TEM image. (C) Chemical structures of heterobifunctional Linkers 1 and 2. (D) MALDI-TOF MS confirming the consecutive addition of a single molecule of each linker to each subunit of 1. (E) Denaturing PAGE (Left to Right) protein ladder, unreacted Sp1m (1), and purified Sp1m-DNA conjugate (4). The presence of two bands of approximately equal intensity, at higher molecular weight compared to 1, correspond to a roughly equal mixture of protein subunits with one and two DNA strands.

Having established a synthetic route to prepare Sp1m with two orthogonal functional groups for click chemistry (tetrazines and azides), we attached DNA to the protein surface. It has been shown that the inverse electron demand Diels–Alder reaction between tetrazines and trans-cyclooctene (TCO) (50, 51) is sufficiently orthogonal to the copper-free strain-promoted alkyne-azide cycloaddition reaction between azides and dibenzocyclooctyne (DBCO) (52), such that these reactants may be used simultaneously to achieve selective, multitarget functionalization (53). Therefore, a one-pot reaction was employed to simultaneously conjugate orthogonal TCO- and DBCO-terminated DNA ligands to the linker-modified protein. Denaturing polyacrylamide gel electrophoresis (PAGE) confirmed successful modification of the protein and revealed the attachment of one or two DNA ligands per protein subunit (Fig. 1E and SI Appendix, Figs. S4 and S5). To understand this distribution and to confirm the orthogonal reactivity of the two DNA conjugation reactions, the reactions were conducted separately and analyzed via denaturing PAGE (SI Appendix, Fig. S6). This confirmed that DBCO-DNA ligands react exclusively with the equatorial azides with high conversion (87%, calculated by gel densitometry), resulting in ∼10 DNA ligands (of a possible 12) in the equatorial plane. The TCO-DNA ligands react with lower conversion (59%) but good selectivity, suggesting that ∼3 to 4 DNA ligands (of a possible 6) occupy each axial face of the protein, for a total of 6 to 8 axial DNA (of a possible 12) per building block. We attribute the lower conversion to the proximity of the N termini to each other in the inner portion of the structure, which may lead to steric and electrostatic congestion between the bulky, negatively charged DNA ligands. Given that as few as two closely placed DNA ligands on a protein’s surface can act cooperatively to form interface interactions between proteins (39), we expected that three to four DNA ligands per face would be sufficient to define the axial interaction. Furthermore, the strength of axial or equatorial interactions can be tuned via DNA sequence design, independently from the number of DNA ligands. Overall, this conjugation strategy is highly effective and enabled the preparation of 19 unique Sp1m-DNA building blocks explored in this work.

Directional Assembly Encoded by Strong Axial or Equatorial DNA Interactions.

While the above conjugation strategy controls the spatial distribution of DNA ligands on the protein surface, DNA sequence design allows for the specificity and strength of the resulting DNA–DNA interactions to be programmed. We designed DNA sequences that interact orthogonally, in different directions and at distinct stages, to define a multistep hierarchical assembly pathway driven by the hybridization of complementary DNA (Scheme 1B). To this end, we synthesized building blocks where the axial and equatorial DNA sequences have disparate melting temperatures (T_m), such that directionally specific interactions occur at different temperatures (DNA designs in SI Appendix, Table S2). Specifically, interactions were designed to be either “strong” (T_m >> room temperature, RT) or “weak” (T_m << RT). We hypothesized that, upon cooling, the strong interactions would hybridize first and building blocks would undergo a first stage of assembly. This assembled structure would display weakly interacting DNA ligands in a multivalent fashion, resulting in an emergent interaction with enhanced cooperativity and increased T_m relative to the isolated weak interactions. The emergent interaction would then drive a second stage of assembly and the formation of a complex assembled structure.

To test if our DNA design strategy imparted directionality on the interactions (axial versus equatorial), we initially characterized the assembly outcomes of systems where only strong interactions are present. Temperature-dependent association of Sp1m-DNA conjugates was probed using a donor-quenching Förster resonance energy transfer (FRET)-based technique (Fig. 2 A and B) (39, 54). In a typical experiment, a pair of complementary Sp1m-DNA conjugates was functionalized with cyanine 3 (Cy3)- and cyanine 5 (Cy5)-modified DNA, respectively. As the proteins assemble, the efficiency of FRET from excited Cy3 to Cy5 increases, leading to quenching of Cy3 fluorescence. Therefore, FRET efficiency monitored via the change in Cy3 fluorescence upon cooling from 65 to 20 °C provides a measure of the degree of assembly (SI Appendix, Supplementary Text 5). Initially, strong axial interactions (denoted A_S) were studied using two complementary conjugates, Sp1m-A_SE_NC and Sp1m-A′_SE_NC, with Cy5- and Cy3-modified axial DNA, respectively, and noncomplementary equatorial (E_NC) interactions that will not assemble equatorially. Their temperature-dependent association profile displayed a single transition with a T_m of 57.3 °C and full width half-maximum (FWHM, reference SI Appendix, Supplementary Text 5.3 for details) of 10.8 °C, compared to T_m = 43.4 °C and FWHM = 16.4 °C for the free DNA duplex (Fig. 2B). The increased T_m and decreased FWHM observed for the Sp1m-DNA conjugates, relative to the free DNA duplex, are suggestive of a multivalent and cooperative interaction between proteins.

Fig. 2.

Characterization of the assembly of Sp1m with strong axial (A_S/A′_S) interactions. (A) Scheme showing the donor-quenching FRET experiment. In a typical experiment, a pair of complementary Sp1m-DNA conjugates were functionalized with Cy3- or Cy5-modified axial DNA, respectively. When well separated, excitation of Cy3 results in fluorescence from Cy3 (filled red circle). However, when Cy3 and Cy5 are in close proximity, FRET from excited Cy3 to Cy5 quenches the fluorescence of Cy3 leading to reduced fluorescent signal (empty red circle). (B) Temperature-dependent association of Sp1m-A_SE_NC and Sp1m-A′_SE_NC represented as fraction assembled versus temperature, where the fluorescence intensities at 65 and 20 °C correspond to a fraction assembled of 0 and 1, respectively (details in SI Appendix). (C) Negative-stain and (D) cryogenic TEM micrographs of slow-cooled Sp1m-A_SE_NC and Sp1m-A′_SE_NC. (Scale bars, 150 nm.)

Sp1m-A_SE_NC and Sp1m-A′_SE_NC were then slow cooled (0.1 °C/10 min), and the assembly products were characterized in the dried and native states using negative-stain and cryogenic TEM, respectively (Fig. 2 C and D). These micrographs revealed the formation of polymeric, 1D protein chains, connected through axial interfaces. Remarkably, in the dried state we can resolve polymeric structures containing tens of proteins (Fig. 2C and SI Appendix, Fig. S12), and we observe chains measuring several hundred nanometers long in the native state (Fig. 2D and SI Appendix, Fig. S19), with negligible off-target, nonaxial interactions. Negative-stain TEM of a control sample where only one building block is present (i.e., Sp1m-A_SE_NC) shows no evidence of assembly (SI Appendix, Fig. S13). Taken together, these data support our hypothesis that a strong DNA interaction (defined via sequence design) and the axial functionalization of Sp1m (defined via mutant design and specific functionalization) encodes highly directional interactions between proteins.

Next, the designed strong equatorial interactions (denoted E_S) were interrogated using an identical donor-quenching FRET technique with a pair of complementary Sp1m-DNA conjugates, Sp1m-E_S and Sp1m-E′_S, functionalized with Cy3- and Cy5-modified DNA, respectively (Fig. 3A). As anticipated, the temperature-dependent association profile for Sp1m-E_S and Sp1m-E′_S displayed a single, sharp transition (Fig. 3B). Analogous to the strong axial interactions, this transition has a higher T_m (57.3 °C) and lower FWHM (4.1 °C) compared to the free DNA duplex (35.9 and 14.0 °C, respectively), again suggestive of a multivalent and cooperative interaction between proteins. To assess the directionality of these interactions and characterize the assembly products, Sp1m-E_S and Sp1m-E′_S were slow cooled (0.1 °C/10 min) and observed in the dried state using negative-stain TEM (Fig. 3C and SI Appendix, Fig. S14) and in their native environment using liquid atomic force microscopy (AFM, Fig. 3 D and E), which enabled quantification of assembly height. Both techniques revealed 2D arrays of assembled proteins, connected through equatorial interfaces, suggesting directional interactions in the equatorial plane. Importantly, negative-stain TEM of a control sample comprising only one building block (i.e., Sp1m-E_S) shows no evidence of assembly (SI Appendix, Fig. S15). Moreover, we confirmed the formation of monolayer assemblies using AFM (Fig. 3D and SI Appendix, Fig. S20), which further supports that favorable interactions only exist in the equatorial plane.

Fig. 3.

Characterization of the assembly of Sp1m with strong equatorial (E_S/E′_S) interactions. (A) Schematic of the donor-quenching FRET experiment. (B) Temperature-dependent association of Sp1m-E_S and Sp1m-E′_S represented by plot of fraction assembled versus temperature. (C) Negative-stain TEM micrograph of slow-cooled Sp1m-E_S and Sp1m-E′_S. (Scale bar, 150 nm.) (D) Liquid AFM micrograph of slow-cooled Sp1m-E_S and Sp1m-E′_S. White arrow denotes line used for height profile in E.

Multistage Assembly Encoded by Strong and Weak DNA Interactions.

Having validated our design for encoding strong, directional interactions between proteins and characterized the assembly behavior resulting from these single-step assembly processes, we proceeded to investigate systems that could undergo defined, multistep assembly. Guided by our hypothesis that building blocks with both sufficiently strong and weak surface interactions would be able to traverse a hierarchical assembly pathway that relies on emergent multivalency to induce the second stage of assembly, we designed building blocks displaying axial and equatorial DNA with vastly different interaction strengths, as characterized by T_m (SI Appendix, Table S2). In all cases, the weak interaction comprises self-complementary DNA sequences with a theoretical T_m < 10 °C to ensure negligible association at ambient temperature prior to undergoing the first stage of assembly. To characterize these assembly steps, we again turned to a donor-quenching FRET-based technique to capture their assembly profiles as a function of temperature.

A pair of Sp1m building blocks, Sp1m-A_SE_W1 and Sp1m-A′_SE_W1, were synthesized in which the proteins were functionalized at the axial positions with the previously discussed strong DNA sequences (A_S and A′_S) and at the equatorial positions with a self-complementary weak DNA sequence (E_W1). We modified the equatorial DNA sequences of Sp1m-A_SE_W1 and Sp1m-A′_SE_W1 with Cy3 and Cy5 dyes, respectively, such that upon the formation of 1D protein chains, driven by the strong axial interactions, the proximity of equatorial DNA would increase and thus partial quenching of the Cy3 fluorescence would occur. Further quenching would take place when the 1D structures associate through hybridization of equatorial DNA strands, indicating a second stage of assembly. As a control, an additional pair of building blocks, Sp1m-A_SE_NC and Sp1m-A′_SE_NC, was synthesized whereby the equatorial DNA ligands of Sp1m-A_SE_NC and Sp1m-A′_SE_NC were modified with Cy3 and Cy5 dyes, respectively. The degree of assembly for both systems was determined by measuring the fluorescence of Cy3 upon cooling from 65 to 20 °C (Fig. 4 A–C). The assembly profiles of both sets of building blocks revealed a sharp transition at T_m = 54 °C, consistent with the T_m measured for the assembly of the axial-only system (57.3 °C), that can be attributed to the association of proteins through axial interactions. The discrepancy in T_m is due to the difference in salt concentration between experiments. Additionally, for the building blocks modified with self-complementary equatorial DNA (Sp1m-A_SE_W1 and Sp1m-A′_SE_W1) a second transition occurs. This transition has a T_m of 32.7 °C, which is greater than expected for the free six base-pair (bp) E_W1 duplex (theoretical T_m < 5 °C), indicating a highly cooperative assembly event.

Fig. 4.

FRET-based characterization of temperature-dependent hierarchical assembly processes. (A–C) Hierarchical assembly mediated by strong axial (A_S/A′_S) interactions. (A) Scheme showing the hypothesized assembly outcomes for two pairs of A_S/A′_S building blocks: Sp1m-A_SE_W1 with Sp1m-A′_SE_W1, and Sp1m-A_SE_NC with Sp1m-A′_SE_NC. Temperature-dependent association of (B) Sp1m-A_SE_W1 and Sp1m-A′_SE_W1 and (C) Sp1m-A_SE_NC and Sp1m-A′_SE_NC represented by plots of fraction assembled versus temperature. Both pairs show the first stage of assembly mediated by A_S/A′_S interactions but only with E_W1 is a second stage of assembly observed. (D–F) Hierarchical assembly mediated by strong equatorial (E_S/E′_S) interactions. (D) Scheme showing hypothesized assembly outcomes for two pairs of E_S/E′_S building blocks: Sp1m-A_WE_S with Sp1m-A_WE′_S, and Sp1m-A_NCE_S with Sp1m-A_NCE′_S. Temperature-dependent association of (E) Sp1m-A_WE_S and Sp1m-A_WE′_S and (F) Sp1m-A_NCE_S and Sp1m-A_NCE′_S represented by plots of fraction assembled versus temperature. Both pairs show the first stage of assembly mediated by E_S/E′_S interactions but only with A_W is a second stage of assembly observed.

DNA interactions are greatly influenced by their ionic environment (55), and thus we investigated how this two-step assembly profile would change under different salt conditions. We repeated the cooling experiment at a higher and lower salt concentration (20 mM and 5 mM versus 10 mM MgCl₂, SI Appendix, Fig. S21). Interestingly, in both 5 and 20 mM MgCl₂, the transition at 32.7 °C disappeared, and the assembly profiles display a single transition at 52.0 and 55.2 °C, respectively, but these conditions result in significantly different relative fractions assembled (SI Appendix, Fig. S21B). Assembly driven by axial interactions results in a much greater fraction assembled in 20 mM MgCl₂ compared to lower salt concentrations, suggesting that at high salt concentration, the two assembly steps become concerted and cannot be resolved. At the lowest salt concentration (5 mM), the assembly profile suggests that only the first (axial) stage of assembly occurs and that a salt concentration between 5 and 20 mM is required for both assembly stages to occur and be resolvable. These trends are consistent with the influence of ionic environment on the hybridization of DNA; however, it is notable that the two stages of assembly differ substantially in the extent to which they are influenced by changes in salt concentration, therefore pointing to additional methods to fine tune hierarchical assembly pathways. Overall, this set of experiments provides evidence for a temperature-dependent, programmed, multistep assembly pathway defined by DNA interactions and supports the hypothesis that Sp1m-DNA conjugates assemble first through axial interactions and then through equatorial interactions. Importantly, this second stage of assembly relies on an emergent interaction that is encoded by DNA sequences in the initial building block but is only activated after the first assembly step. This process is akin to the hierarchical generation of tertiary and quaternary protein structures defined exclusively by the information present in the primary amino acid sequence.

Next, we investigated whether a reversed assembly pathway could be programmed by simply switching the relative strengths of DNA interactions at the axial and equatorial positions. Accordingly, a set of building blocks, Sp1m-A_WE_S and Sp1m-A_WE′_S, was synthesized employing the previously discussed strong equatorial complementary DNA sequences (E_S and E′_S) as well as weak self-complementary DNA sequences at the axial positions (A_W). We modified the axial DNA sequences of Sp1m-A_WE_S and Sp1m-A_WE′_S with Cy3 and Cy5 dyes, respectively, where we expected to observe partial quenching for the first stage of assembly (formation of 2D structures through strong equatorial interactions) and further quenching upon subsequent axial interactions during cooling from 65 to 20 °C. To provide a comparison where axial interactions are inhibited, Sp1m-A_NCE_S and Sp1m-A_NCE′_S were synthesized with noncomplementary axial DNA ligands (A_NC) modified with Cy3 and Cy5 dyes, respectively. When comparing the temperature-dependent assembly profiles for these two sets of building blocks, the system containing both interaction types (Sp1m-A_WE_S and Sp1m-A_WE′_S) displays two distinct transitions (T_m = 50.4 and 38.1 °C) whereas the system with A_NC interactions displays only a single transition (50.4 °C; Fig. 4 D–F). We therefore attribute the common transition at 50.4 °C to the initial association of proteins in the equatorial plane to form 2D structures and the unique transition at 38.1 °C to the subsequent onset of axial interactions between these 2D structures. The transition at 38.1 °C is relatively broad, compared to the first assembly step, which may be due to polydispersity in the domain sizes of the 2D structures that associate in this step (SI Appendix, Supplementary Text 9 and Figs. S22 and S23). We hypothesized that the interaction strength in this assembly stage is a function of 2D domain size, and therefore, this dispersity leads to a multiplicity of interaction strengths via axial DNA, leading to the relatively broad nature of the transition at 38.1 °C in Fig. 4E. Together, these experiments support the hypothesis that Sp1m-A_WE_S and Sp1m-A_WE′_S undergo a reversed, thermally controlled, multistep assembly pathway, first associating through equatorial interactions and then via axial interactions.

Programming Structural Outcomes via DNA Design.

We have shown that designing the relative strength of DNA ligands and their spatial arrangement on the protein surface directs assembly along different pathways with distinct assembly outcomes. We next explored whether the assembly outcome could be changed while maintaining the same pathway via DNA sequence design. To that end, we focused on characterizing the structures that arise from an axial-first, equatorial-second assembly pathway. In addition to the previously described system, Sp1m-A_SE_W1 and Sp1m-A′_SE_W1 (Fig. 4), we designed an additional building block, Sp1m-A′_SE_W2, where the equatorial sites of the second building block were modified with a weak self-complementary sequence (E_W2) orthogonal to E_W1. The E_W1 and E_W2 DNA sequences (SI Appendix, Table S2) are identical in length and bp composition to ensure that differences in the assembly outcome result from differences in the presentation of the emergent second interaction, rather than inherent differences in the interaction strength between the two self-complementary DNA designs. The building blocks were slow cooled (0.1 °C/10 min) in two combinations: Sp1m-A_SE_W1 with Sp1m-A′_SE_W1 (Fig. 5A), and Sp1m-A_SE_W1 with Sp1m-A′_SE_W2 (Fig. 5C). The complementarity of A_S and A′_S ensures that, in the latter system, E_W1 and E_W2 are presented alternately (Fig. 5C). TEM characterization of both samples reveals the formation of 1D protein chains, formed via strong axial interactions, that interact with each other, suggesting that these two systems traverse the same assembly pathway. However, the two sets of building blocks give significantly different structural outcomes (Fig. 5 B and D). For the system containing only E_W1-based building blocks, the 1D protein chains have a high propensity to form bundles and fold up on themselves via intrachain interactions (Fig. 5B). However, when one of the building blocks is modified with E_W2, the 1D protein chains instead interact to form elongated filaments (Fig. 5D). Moreover, TEM suggests that registry between the proteins in each chain is better enforced in this sample (SI Appendix, Figs. S16–S18). We hypothesize that the presence of alternating, orthogonally self-complementary interaction areas, spatially encoded on the surface of the 1D chain by DNA, favors interchain association by reducing kinetic trapping via intrachain folding (SI Appendix, Fig. S24). This highlights how two, orthogonal, self-complementary E_W sequences decrease the propensity for the 1D protein chains to fold and bundle and is a key demonstration of how DNA design not only defines a specific assembly pathway but also directs the final structural outcome. Additional control over the final structural outcome, such as achieving higher order or enforcing registry between proteins, can in principle be achieved through further exploration of DNA design parameters, including length, flexibility, absolute and relative interaction strength, position of DNA attachment, and number of DNA strands, among many others.

Fig. 5.

Characterization of assembly outcomes from axial-first, equatorial-second hierarchical assembly processes. (A) Scheme showing 1D protein chains displaying equatorial E_W1 DNA homogenously. (B) Negative-stain TEM micrograph of slow-cooled assembly of Sp1m-A_SE_W1 and Sp1m-A′_SE_W1. (C) Scheme showing 1D protein chains displaying alternating equatorial E_W1 and E_W2 DNA. (D) Negative-stain TEM micrograph of slow-cooled assembly of Sp1m-A_SE_W1 and Sp1m-A′_SE_W2. (Scale bars, 150 nm.)

Conclusion

This work harnesses the programmability of DNA and the chemical addressability of protein surfaces to control the hierarchical, multistep assembly of protein building blocks mediated by multiple, distinct DNA hybridization events. Through functionalization of a protein’s surface with DNA ligands at axial and equatorial positions, we introduced highly directional interactions between specific geometric interfaces. We programmed multistep assembly profiles by defining disparate recognition properties at different locations within discrete protein building blocks, which allows us to control the assembly pathways and structural outcomes. Furthermore, we used DNA to define multiple orthogonal interactions within a single assembly pathway, thereby realizing distinct, protein-based materials as a function of both the type of pathway traversed and the DNA design employed. This principle, in which all information required for hierarchical assembly is encoded into an initial primary structure, has long been exploited by nature to realize sophisticated architectures from amino acid sequences but seldom by using nucleic acids. In contrast to canonical uses of nucleic acids in nature—primarily information storage and sometimes as a template to organize structures—DNA is rarely, if ever, employed as a programmable “bond” to direct complex assembly pathways. These findings show that, through judicious design, one can use DNA to build structures on demand with a degree of hierarchical control atypical for synthetic nanoscale programmable matter but reminiscent of complex structures in nature. In principle, because the information for hierarchical assembly is encoded by DNA, this approach can be applied to any protein or other nanoscale building block, where the surface can be appropriately modified with DNA. These insights reveal how to go beyond a single-step assembly pathway for the bottom-up assembly of nanomaterials and will enable the synthesis of hierarchically structured materials by design.

Materials and Methods

Detailed materials and methods are provided in SI Appendix.

Functionalization of Sp1m with Azide and Tetrazine Linkers.

Maleimide-azide linker (Linker 1) was prepared from azido-PEG3-amine (2 μL) in DMSO (48 μL) and 3-maleimido-propionic NHS ester (2.5 mg) in DMSO (50 μL). The mixture was shaken at 650 rpm at 25 °C for 30 min. The reaction was quenched by addition of Tris (1 M, pH 7, 10 μL) and shaken for a further 5 min. The mixture (110 μL) was added to an aliquot of Sp1m (1, 400 μL, 5 μM) and shaken overnight at 650 rpm at 25 °C. The reaction mixture was purified by size exclusion chromatography (details in SI Appendix, Supplementary Text 3.1), and fractions containing Sp1m-N₃ (2) were pooled, concentrated to 5 μM, and portioned into 1.5-mL Eppendorf tubes in 500 μL aliquots. To each aliquot, a solution of methyltetrazine-PEG5-NHS ester (Linker 2, 0.6 μL) in DMSO (20 μL) was added and thoroughly mixed by pipette aspiration. The solution was shaken at 650 rpm for 20 h at 25 °C. The reaction mixture was purified by size exclusion chromatography, and fractions containing protein were pooled. Sp1m with both azide and tetrazine linkers (Sp1m-2L, 3) was typically reacted with DNA immediately, although Sp1m-2L (3) could be stored at 4 °C for 24 h without loss in reactivity.

DNA Conjugation to Sp1m-2L (3).

DNA conjugation reactions were typically performed on the 0.5, 0.7, or 1 nmol scale with respect to Sp1m-2L (3). A mixture of Sp1m-2L (1 equivalent), TCO-DNA (180 equivalents), and DBCO-DNA (150 equivalents) in Hepes (20 mM, pH 7.4) and NaCl (500 mM) was shaken at 650 rpm for 20 h at 37 °C. Unreacted DNA was removed by washing the reaction mixture three times in a 4-mL centrifugal filter with 20 mM Hepes (30 K molecular weight cutoff, 3,000 × g, 4 °C, 3 min cycles). The reaction mixture was purified by size exclusion chromatography, and fractions containing protein were pooled and stored at 4 °C.

Donor-Quenching FRET Studies.

Combinations of Sp1m-DNA conjugates at 300 nM total Cy3 concentration were mixed (1:1 ratio, 50 μL) and placed in a 96-well plate, heated at 65 °C for 5 min, and then cooled from 65 °C to 20 °C at 0.1 °C/0.5 min using a Bio-Rad CFX96 Touch real time PCR system. All samples were measured in triplicate, and the data reported represents the average of the three runs. Cy3 fluorescence was measured at 0.1 °C intervals.

Assembly of Sp1m-DNA Conjugates via Slow Cooling.

Samples were mixed to a total protein concentration of 100 or 500 nM and then cooled from 60 °C to 21 °C at a rate of 0.1 °C/10 min using a ProFlex PCR system (Applied Biosystems). The resulting assemblies were characterized using negative-stain TEM, cryo-TEM, or AFM (SI Appendix, Supplementary Text 6 and 7).

Data Availability

All study data are included in the article and/or SI Appendix.