## Main

Infectious diseases, such as viruses, bacteria and toxins, present unique spatial patterns of antigens on their surfaces1,2,3. Existing weakly binding ligands (called ‘binders’ herein) that interact with these epitopes can be linked to a scaffold to improve multivalent binding avidity4,5,6,7,8. However, some viruses, such as dengue (DENV), present complex geometric patterns that cannot be addressed by existing scaffolds because they are not as precise in ligand spacing or provide limited control over the scaffold shape and ligand valency9,10,11,12. Furthermore, previous synthetic scaffolds (polymers, dendrimers, nanofibres, nanoparticles and lipid nanoemulsions) have shown some toxicity. Designer DNA nanostructures13,14,15,16,17 are able to overcome previous limitations because they can be designed into various non-toxic18,19,20,21 and biologically stable21,22,23,24,25 two-dimensional (2D) or 3D platforms to control ligand spacing, valency and spatial arrangements14,26,27. Biocompatible DNA nanostructures have exhibited blood circulation times of up to 24 h (ref. 21) but are eventually removed by the liver21 and kidney28,29.

We demonstrate our strategy by directly targeting DENV through a star-shaped DNA scaffold. This architecture was strategically designed to multivalently arrange binders in a 2D pattern mirroring the complex spatial distribution of dengue epitopes with nanometre precision. Although previous studies have shown that optimizing bivalent interactions is important in protein binding30, cell signalling31 and antibody spatial tolerance using DNA nanostructures32, to our knowledge the importance of multivalent, 2D spatial pattern matching has not yet been explored. To this end, we have included a comparison with scaffolds of different shapes and flexibilities when targeting DENV.

Gold standard diagnostic methods for DENV rely on enzyme-linked immunosorbent assay (ELISA) or quantitative polymerase chain reaction with reverse transcription (RT–qPCR)33,34,35,36. These methods are lengthy, costly and require technical laboratory training. On the therapeutic front, prevention or treatment of viral infection typically relies on neutralizing antibodies (NAbs) against target epitopes on the viral surface. Production of NAbs can be triggered by vaccines or active viruses in the host. However, NAbs may induce antibody-dependent enhancement (ADE) of infection (for example, with DENV37), where antibodies generate increased viral infectivity in vivo, or may not provide protection against new epidemics as a result of genetic drift (for example, antigenic drift in influenza38). A customizable molecular scaffold strategy capable of incorporating sensing motifs with pathogen-specific binders and patterns may address these issues on both fronts.

## Results

### Design and characterization of the DNA star

The envelope protein domain III (ED3) viral epitope on the DENV surface is organized into a complex polyhedral pattern39 with alternating clusters of trivalent or pentavalent ED3 sites. By connecting the clusters of ED3 sites linearly, we determined that a star shape, consisting of an interior pentagon connected to five exterior triangles, would provide an optimal multivalent scaffold (Fig. 1a). The viral diameter of DENV is 50 nm (ref. 39) and vector distances were measured using PyMOL. The distance between adjacent trivalent–trivalent clusters was 15.1 nm and the distance between adjacent trivalent–pentavalent clusters was 14.1 nm. The following equation was used to calculate the orthodromic distances needed to design the scaffold:

$$\begin{array}{lll}{\mathrm{Orthodromic}}\;{\mathrm{distance}} &=& {\mathrm{viral}}\;{\mathrm{diameter}}\\&& \times \sin ^{ - 1}\left( {\frac{{\mathrm{cluster}}\;{\mathrm{inter}}\hbox{-}{\mathrm{distance}}}{{\mathrm{viral}}\;{\mathrm{diameter}}}} \right)\end{array}$$

The orthodromic centre-to-centre distances between trivalent–trivalent and trivalent–pentavalent clusters were 15.3 nm and 14.3 nm, respectively. Based on this structural information, the star-shaped DNA scaffold (DNA star) was then designed to not only match the local, orthodromic distances between ED3 clusters but also mirror the global pattern of ED3 clusters aligning with the star’s ten vertices (Fig. 1b).

The DNA star was constructed using the tile-based self-assembly of 21 DNA oligonucleotides (strands). Each of the ten external edges is 42 bp (base pairs) long and connected to the internal edges through four arm junctions at the inner vertices. Each of the five internal edges carries a single-stranded DNA (ssDNA) region that can form a hairpin (stem-loop) structure with a 7 bp stem and 6 bp loop. When the hairpin unzips into ssDNA, it can stretch each internal edge to a distance that fits the distance between adjacent trivalent clusters. These hairpins provide molecular beacon-like motifs for viral detection and offer local structural flexibility to ensure binding of viral targets under various solution conditions and temperatures40.

The formation of the complete DNA star (lane ‘Star’, Fig. 1c) was first characterized by 4% non-denaturing polyacrylamide gel electrophoresis (PAGE), with one of each unique component strand and partial star complex included for reference. Atomic force microscopy (AFM) imaging further confirmed formation of the star (Fig. 1d). For individual DNA nanostructure assemblies, non-denaturing gel electrophoresis is a valid and more reliable method for characterizing structural formation and yield (for example, see previous studies41,42,43). Our non-denaturing PAGE showed a distinct, major band corresponding to each of the 2D architectures, confirming that DNA star structures were formed with high efficiency (Fig. 1c). It is also worth noting that the AFM probe can damage or displace the thin DNA tile-based structures during the scan (Supplementary Fig. 1). Additionally, AFM may display the overall shape of DNA star but cannot fully resolve the details of the structure at the atomic level due to the limited resolution.

### DNA star-based DENV sensing

The DNA star was functionalized for DENV detection and inhibition by hybridizing a well-characterized, ED3-binding aptamer44 at each of the 10 vertices of the DNA star to form a star–aptamer complex that geometrically matched and targeted ED3 clusters (Fig. 2a and Supplementary Fig. 2). The aptamer exhibited weak binding to DENV, as determined by surface plasmon resonance (Supplementary Fig. 3). However, the binding strength increased as aptamers were placed onto scaffolds that increasingly matched the pattern of ED3 sites on DENV with the DNA star showing the greatest binding avidity (Supplementary Fig. 3). 6-FAM (fluorophore) and BHQ-1 (quencher) carrying strands were hybridized to each inner edge, flanking the hairpin to turn the DNA star–aptamer complex into a viral sensor. In the absence of DENV, FAM and BHQ-1 molecules would be brought together by canonical Watson–Crick base pairing in the stem, much like a molecular beacon45. However, unlike a molecular beacon, which gives a fluorescent readout of target nucleic acid hybridization events, here the hairpins are pulled apart and converted to ssDNA as a result of strong aptamer binding to ED3 sites. The potent multivalent interactions, promoted by the matched geometric aptamer–ED3 pattern, separated the FAM fluorophores from BHQ-1 quenchers to afford a fluorescent readout.

The DNA star–aptamer sensor was able to directly detect DENV virions with high sensitivity, affording a limit of detection (LoD) of 1 × 102 p.f.u. ml−1 and 1 × 103 p.f.u. ml−1, respectively, in human serum and plasma (Fig. 2c). To exclude the possibility that the series of DENV-mixed DNA star sensing signals obtained within 1–2 min resulted from an automatic (non-target-triggered) separation of the fluorophores from quenchers, we measured fluorescent readouts over time at different temperatures. Fluorescent measurements over time showed that the FAM fluorophores remained quenched during the duration required for sensing and functioned at different temperatures (Supplementary Fig. 4). To examine the specificity of our sensor, we tested it against another virus available to us, the adenovirus. If the star–aptamer complex does not recognize the virus through proper binding, the complex will remain quenched (Fig. 2b). The experimental results verified that the star–aptamer complex designed for DENV detection is not able to sense the adenovirus, even at high concentrations (Fig. 2d).

To make a comparison between the DNA star-based and gold standard methods under similar conditions, we carried out RT–qPCR and ELISA assays to respectively detect DENV RNA and NS1 antigen (Supplementary Fig. 5). The more sensitive method, RT–qPCR, showed a LoD of 1 × 103 p.f.u. ml−1 after we processed different concentrations of DENV-containing human serum (Supplementary Fig. 5), demonstrating the superior sensitivity of our new approach.

In summary, the LoD of our sensor is well below the viral concentration (>1 × 105 p.f.u. ml−1, Supplementary Table 1) in patients on day 0, or the onset of illness, when fever and a variety of symptoms start to occur and the virus begins to become very pathogenic46. Thus, detection before this date is highly beneficial to the patient and to the timely screening and control of pandemic outbreaks within surveillance and diagnostic networks.

### Bivalent and linear 1D sensors

To investigate if 2D spatial pattern matching is critical for sensing, we constructed two 1D scaffolds (Fig. 3a). The first was functionalized with two aptamers (bivalent) while the second was functionalized with six aptamers templated by a flexible linear DNA scaffold (abbreviated as linear). In both structures, the strand(s) used to create the DNA star were reused to maintain the same length, sequence, initial aptamer spacing and flexibility wherever possible. Scaffold formation was characterized through 10% non-denaturing PAGE (Supplementary Fig. 7). The flexible linear sensor began detecting DENV with a better sensitivity than the bivalent sensor only at a serum sample containing 1 × 105 p.f.u. ml−1 of the virus. However, both sensors failed to provide comparable sensitivity to the DNA star when standardized under the same fluorophore–quencher pair concentration (Fig. 3c). These control experiments confirm that 2D spatial pattern recognition is important for detection.

### Hexagon-centred and heptagon-centred 2D sensors

Next, we determined if a correctly matched 2D pattern was required for proper sensing. For that purpose, two additional 2D DNA nanostructures, hexagon- and heptagon-centred star complexes, were constructed and then characterized using AFM and a 4% non-denaturing PAGE (Fig. 3a and Supplementary Fig. 7). The hexagon-centred, six-point star (hexagon) was made by adding an additional, equivalent triangle to the original five-point star design. Similarly, a heptagon-centred, seven-point star (heptagon) was made by adding another equivalent triangle to the hexagon design. We observed that the hexagon provided good sensing capabilities (LoD of 1 × 103 p.f.u. ml−1) while the heptagon could not detect the virus even at a DENV serum sample concentration of 1 × 105 p.f.u. ml−1 when standardized with the same fluorophore–quencher pair concentrations.

The positive sensing provided by the hexagon was expected, because when we analysed DENV surface ED3 patterns, we also recognized a hexagon-shaped distribution of ED3 clusters when centring the DENV virion on a trivalent cluster (Fig. 3b); this additional pattern differed from the pentavalent cluster-centred star shape (Fig. 1a). The fact that the hexagon can serve as a good positive control sensor proves the need for pattern matching in detection. The sensitivity of the hexagon sensor was tested, providing a LoD of 1 × 103 p.f.u. ml−1 (Fig. 3d). Although the heptagon carried more aptamers than the hexagon, the sensor resulted in poor to no detection. This comparison is particularly interesting as it demonstrates that a small mismatch of 2D geometry, under the same bivalently correct spacing and with higher ligand valency and density (as compared to the hexagon), leads to dramatically reduced sensing capabilities.

### In vitro DENV inhibition

Encouraged by the previous results, we investigated the in vitro inhibition of DENV through standard antiviral, plaque-forming half-maximum effective concentration (EC50) assays. Briefly, the DENV viral particles were incubated with different concentrations of each inhibitor in human serum and the remaining infectivity was determined by a plaque reduction assay (Fig. 4a). The dose-dependent inhibition of DENV for the monovalent aptamer and each of the scaffold–aptamer complexes was examined (Fig. 4b). The EC50 value of the DNA star–aptamer complex for DENV infection inhibition was 2 nM, whereas the EC50 value for the monovalent aptamer was 15 µM (Fig. 4c,d). These results demonstrate that the DNA star–aptamer multivalent inhibitor was 7.5 × 103-fold more effective than the monovalent aptamer for the in vitro inhibition of DENV infection in human serum.

We also examined the EC50 values of the other control scaffolds, where a lower EC50 value demonstrates more potent inhibition (Fig. 4b–d). In correlation with our DENV sensing data, the hexagon scaffold provided the second most potent inhibitory effect, with an EC50 value of 10 nM, while the heptagon scaffold exhibited poor inhibition, with an EC50 value of 440 nM. The flexible linear scaffold provided better inhibition, at an EC50 of 90 nM, than the bivalent scaffold, at an EC50 of 130 nM. Additionally, as previous studies predicted, potentially matching the spacing could increase the inhibitory effect as all the structures proved to be more potent than the monovalent aptamer.

These results point out that stronger inhibition was achieved when the spatial pattern was properly matched. Interestingly, the 2D heptagon scaffold exhibited worse inhibitory potency than the 1D bivalent and flexible linear scaffolds. A mismatch of a fixed geometric pattern, while keeping the same initial spacing, leads to reduced inhibition potency when compared to flexibly matching the spacing. In summary, the more the binders match the 2D pattern of binding sites, the stronger the inhibition as aptamers bound to the star scaffold show superior performance compared to the other scaffold shapes.

### Confocal imaging

Time-lapsed, live confocal imaging demonstrated that DENV virions lose their cell internalization ability after being bound by the DNA star–aptamer complex (Fig. 5). For the assay, the HepG2 line was chosen as it is derived from the liver, an organ DENV infects, and was cultured in medium containing human serum. Membrane-bound structures, including the cell membrane, were stained with Dil (yellow) while cell nuclei (blue) were stained with Hoechst. DENV virions were labelled with DiD (red) and the DNA star–aptamer complex was labelled with the 6-FAM fluorophore (green), quenched in the absence of DENV. When the green DNA stars and the red virions co-localize, the merged signals appear yellow. The Dil stain and 6-FAM dye also exhibit fluorescence spillover due to slightly overlapping excitation and emission spectra. The finer details between labelled structures can be resolved when viewing the merged image as separate fluorescence channels (Supplementary Fig. 8).

Unbound or DNA star-bound virions were introduced to the labelled cells (Fig. 5, Supplementary Fig. 9 and Supplementary Videos 1 and 2). In the unbound condition, DENV accumulated into the cells over time. In the DNA star-bound condition, DENV accumulation was drastically reduced and virions were generally electrostatically repelled from cells. Even if some DNA star-bound virions reached the cell membrane, they were unable to enter cells over the time course of an hour.

A cross-section (indicated by the white dotted line in Fig. 5a) accompanies each panel, with an eye symbol denoting the viewing direction. The view was reconstructed by images taken from different planes (Z stacks) during imaging. Volume reconstruction using Imaris software confirmed viral accumulation in the unbound condition and viral entry inhibition when bound to the DNA star (Fig. 5b).

## Discussion

By targeting a challenging exemplar platform, the DENV flavivirus, we have demonstrated the importance of integrating a structurally defined DNA nanoarchitecture with precise, multivalent spatial pattern-recognizing properties. Sensors and inhibitors recognizing the DENV pattern identity allow them to work effectively. When aptamers are arranged into an optimal 2D shape, they can be induced to fit the correct DENV pattern. In contrast, linear complexes, such as the bivalent–aptamer complex, initially have neither optimal shape nor optimal spacing, so they have weak affinity to ED3 sites. Moreover, for those that bind transiently, they have trouble staying on because the hairpin region also has affinity to the base pair. Without the optimal shape identity to keep these hairpin interactions from forming, sensing and inhibitory abilities suffer.

We also observed that an incorrect shape will be more detrimental than having no shape at all, as observed with the heptagon scaffold. We speculate that the heptagon–aptamer complex can bind, in the best-case scenario, bivalently to two ED3 clusters, but that leaves certain sites unbound and capable of cell internalization (Supplementary Fig. 7). In addition, a bound heptagon–aptamer complex would sterically prevent more heptagon–aptamer complexes from binding while linear scaffolds do not experience a steric block to such a degree.

Our DNA star sensor shows superiority over current gold standard DENV detection methods. The reduced sensitivity of RT–qPCR can result from the low amount of starting material (one genome copy per viral particle), RNA extraction process and instability of the extracted RNA. The star sensor, on the other hand, has direct access to the unprocessed viral sample. Up to two stars can specifically bind to a single viral particle, translating to a 10-fluorophore label (five per star). Additionally, the immunoglobulin-M (IgM) ELISA or IgM rapid test cannot provide early viral sensing as antibody production in the body requires several days after initial infection. So, no gold standard method for DENV detection can achieve the same sensing capacity in terms of cost, ease, sensitivity and speed (see Supplementary Fig. 6, Supplementary Table 1, and Note for Supplementary Table 1 for a detailed comparison). As is standard with viral infection screening, secondary confirmation of infection by some of the standard methods (for example, viral isolation or nucleic acid sensing) should still be employed after the initial precautionary screening in order to make a full documentation and evaluation of an epidemic outbreak.

The performance of the DNA star during in vitro DENV inhibition shows promising therapeutic potential. As is the case for various infectious diseases (for example, the malaria parasite47, DENV48, Zika49 and influenza A8), interactions with negatively charged glycosaminoglycans (GAGs) are required for host cell invasion. Thus, the DNA architecture not only relies on specific multivalent interactions for binding DENV but also electrostatically traps and isolates virions from the host cell plasma membrane and GAGs via the negative charges of the scaffold. Moreover, other monovalent aptamer combinations have been used as antiviral agents for human viruses50. Thus, our scaffold-based strategy can be used to arrange these same aptamers into arrangements that increase specific, on-target binding while reducing off-target binding based on pattern identity.

As binders specific to other binding sites can be evolved or synthesized, our strategy could be widely applicable, and such scaffolds could be used to efficiently detect and inhibit a wide-range of pathogens (Fig. 6), modulate cell signalling by analysing the distribution of cellular receptors, or organize previous drugs into more potent configurations. Although DNA nanostructures have shown certain stability in vivo21,29, the component DNA strands can be chemically modified to further improve its stability.

### Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

## Online content

Any Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41557-019-0369-8.