Influence of hydrophobic moieties on the crystallization of amphiphilic DNA nanostructures

Diverse applications, including photonics,¹ electronics,² catalysis,^3,4 molecular sieving,^5,6 sensing,⁷ and drug delivery,⁸ could benefit from the ability to reliably program the self-assembly of nanostructured materials with long-range order. DNA nanotechnology^9,10 offers distinct advantages when it comes to designing nanoscale building blocks for self-assembly, owing to the specificity of Watson–Crick base pairing, the facile chemical modification of DNA, and the decreasing cost of its synthesis. Since its foundation by Seeman,^11,12 DNA nanotechnology has produced various examples of periodic materials in one,^13–20 two,^{10,15,20–36} and three dimensions.^{10,20,34,37–49} Three-dimensional crystals are particularly valuable for applications,^{44,45,50–54} but have also proven to be the most challenging to build, having only been demonstrated by Seeman and co-workers⁴⁰ more than two decades after initial proposals.^11,12

Two classes of DNA building blocks have been shown to form 3D crystals: small-nanoscale motifs, consisting of a handful of oligonucleotides,^40–42 and larger DNA origami.^43–46 The former include tensegrity triangles,⁵⁵ which were the first DNA structures shown to crystallize⁴⁰ and have since undergone design updates to embed molecular cargoes,^50–52,56 program environmental responsiveness,⁵⁷ and enhance stability.⁵² More recently, further small motifs built around Holliday junctions have also been shown to crystallize.^{41,42,58–60} DNA origami with various geometries, such as scaled-up tensegrity triangles,⁴³ octahedral, cubic and tetrahedral frames,^44,46 or rods,⁴⁵ have been shown to support 3D crystallization. While origami crystals cannot match the angstrom-level resolution of those assembled from small nanostructures,⁴² the much larger unit cells and greater design versatility enable the implementation of valuable features and functionalities, such as the inclusion of nanoparticles^43,44 and enzymes,⁵⁴ the templated growth of superconductive materials,⁵³ and the construction of photonic crystals.⁴⁵ 

For both nanoscale and origami building blocks, self-assembly has been mediated by base pairing and/or stacking, and in all cases structural rigidity and precise bond directionality were recognized as necessary for crystallization.^40,43 Deviating from this paradigm, we have introduced a new type of flexible DNA-based building block, where crystallization is induced by nonspecific, nondirectional micellization of hydrophobic moieties.^61–64 These simple motifs, which we dubbed C-stars, consist of fixed branched junctions with each arm tipped by a cholesterol molecule, rendering the nanostructures amphiphilic [Figs. 1(a) and 1(b)]. C-stars have been shown to reliably form ordered frameworks with programmable lattice parameter and porosity,⁶² which have been modified to support triggered assembly/disassembly,^62,65 reversibly capture and release proteins⁶² and, recently, served as the basis of RNA-producing artificial cells⁶⁶ and responsive nanoparticles capable of permeabilizing lipid bilayers and trapping swimming bacteria.⁶⁷ 

We hypothesized that long-range order in C-star aggregates emerges from the topological constraints imposed by the flexible DNA junctions, cross-linking, and coordinating cholesterol-rich micelles.^61–63 This hypothesis is supported by the evidence that changing the number of cholesterolized “arms” alters the crystal structure,⁶¹ which is however robust against many (also substantial) design alterations that preserve topology.^62,66 We further hypothesized that a preferred valency of cholesterol micellization may play an important role in the emergence of crystals: After all, the micelle population in a crystal must be monodisperse.⁶¹ 

Here, we aim to clarify the influence of micelle size (valency) and stability on the ability of C-stars to crystallize. To this end, we keep junction topology and flexibility fixed and empirically explore design variations aimed at altering micellization behavior, which we achieve by changing the number of hydrophobic moieties, their chemical nature, and the structure of the DNA motifs close to the hydrophobic sites.^68,69 While these alterations affect bulk micellization substantially, only minor effects are noted on crystal structure. Our findings thus suggest that junction topology remains the dominant factor dictating crystal geometry. Finally, we show that the changes in micelle structure and stability translate into differences in the ability of the crystals to uptake hydrophobic cargoes, which we demonstrate using model drug penicillin V. The latter result showcases how hydrophobic moieties and neighboring portions of the nanostructures can be engineered to control useful functionalities without impacting crystal structure.

As shown in Fig. 1(a), the C-star designs used in this work consist of four-way fixed junctions, formed by four “core” oligonucleotides with unique sequence (blue). Connected to each of the four “arms” are strands functionalized with hydrophobic moieties (orange), which provide amphiphilic character. In previous studies, each arm was tipped by a single cholesterol moiety, separated from the adjacent double-stranded (ds) segment by a TEG spacer and two unpaired adenine nucleotides [Fig. 1(a)].^61–63 As discussed below, in this study we explore design variations affecting the hydrophobic moieties and their immediate vicinity [compare Fig. 1(e)]. An unpaired thymine separates neighboring arms at the junction, providing flexibility.⁶³ As previously reported,^61–63 to induce self-assembly, samples were prepared by mixing all single-stranded (ss) components in stoichiometric ratios (C-star concentration 5 μM), incubating at 90 °C for 20 min. and slowly annealing (−0.01 °C min⁻¹) down to room temperature [Fig. 1(b)]. At high temperature, ssDNA coexists with micelles formed by the hydrophobized strands. Once a sufficiently low temperature is reached, nucleation and growth of aggregates are observed following the formation of DNA junctions that cross-link the micelles.⁶¹ 

We have previously demonstrated that this protocol leads to the emergence of crystalline aggregates in samples of four-arm C-stars [Fig. 1(c)] with body centered cubic (BCC) structure confirmed by small-angle x-ray scattering (SAXS).^61–63 While the relatively low resolution of the crystals (∼50 Å) prevents us from solving the structure in single-crystal diffraction experiments, we hypothesized that the materials feature periodic arrays of micelle-like hydrophobic cores coordinated by the DNA junctions, for instance, as depicted in Fig. 1(d) and confirmed by substantial indirect evidence.^61–63 

Crystallization and BCC symmetry were found to be robust against many design variations that preserved the four-arm topology,^61–63 such as changes in arm-length between 21 and 46 base-pairs (bps), which allowed us to prescribe the lattice parameter over a wide interval (184–342 Å),⁶² and the addition of functional elements along the arms.^62,66 The latter included chemical moieties for the reversible capture and release of proteins,⁶² DNA toeholds underpinning the crystal disassembly,⁶² and even long overhangs that served as templates for the transcription of RNA constructs.⁶⁶ Amorphous phases were observed with the four-arm topology when excessively long arms were used, possibly due to entropic effects destabilizing the micelles,⁶² or for designs and/or cationic conditions that stabilized rigid, cross-stacked configurations of the central junctions, thought to be incompatible with unit-cell symmetry.⁶³ Rigid four-way junctions lacking unpaired bases separating the arms [compare Fig. 1(a), where one unpaired base is present] displayed an unidentified SAXS spectrum, either corresponding to low-symmetry (non-BCC) unit cells or coexisting crystal phases.⁶³ Non-BCC (unidentified) spectra were also recorded for three or six arm designs with flexible junctions.⁶¹ 

However, non-BCC crystals have never been observed with four-arm designs, when sufficient junction flexibility was allowed by introducing at least one unpaired base between neighboring arms.

Experiments to date, therefore, hint at a picture in which the topology of C-star monomers is a critical factor in determining the equilibrium crystal structure. However, we have also noted that micelles formed by cholesterolized DNA nanostructures in bulk tend to be surprisingly monodisperse,⁶¹ as recently confirmed by Liu et al.⁷⁰ Because C-star crystals have higher resolution (⁠$∼5$ nm) compared to the length scale of the building blocks (⁠$∼20$ nm),⁶² the micelle-like cores must necessarily have consistent coordination within the crystals. It is thus natural to hypothesize that a preferred valency for the micelles may play an important role in crystallization. In other words, the observed BCC symmetry may be favored by the tendency of the micelle-like cores to display a specific valency.

To test this hypothesis, here we explored design variations aimed at tuning the preferred size (valency) and stability of the micelle-like cores in C-star aggregates, which we achieved by changing the shape and flexibility of the building blocks in the vicinity of the hydrophobic modifications and/or altering hydrophobic moieties themselves.

Specifically, we identified three properties, programmable by design, which we deemed likely to influence the self-assembly of hydrophobe-DNA micelles: (1) steric hindrance of DNA around the hydrophobic micellar core, (2) details of cholesterol–cholesterol interaction, and (3) the properties of the hydrophobic molecule itself. To study the effect of steric hindrance, we took two approaches. The first was to change the length of the unpaired adenine spacer between the end of the dsDNA arm and the cholesterol molecule connected to the 3′ terminus. Here, we introduced ssDNA spacers of zero, two, four, and six As with the corresponding designs labeled 3′-0A-chol, 3′-2A-chol, 3′-4A-chol, and 3′-6A-chol, respectively. In addition, we further modulated steric hindrance by including unpaired bases at the non-cholesterolized terminus, which we tested in two designs where both DNA strands terminated with two unpaired As but only one, either 5′ or 3′, was cholesterolized (5′-2A-chol:3′-2A and 5′-2A:3′-2A-chol, respectively). In order to study the effect of cholesterol–cholesterol packing, we tested a design in which the C-star arms were terminated with two cholesterol molecules at both the 3′ and the 5′ ends, with both termini also featuring a 2A spacer (5′-2A-chol:3′-2A-chol). We expected that this double modification may influence the specifics of the cholesterol–cholesterol interactions (e.g., stacking⁷⁰) within the hydrophobic core of the micelle. Finally, to see how the properties of the hydrophobic molecule itself altered its micellization, we replaced the cholesterol molecule with tocopherol (5′-2A-toc). The two hydrophobic tags differ in terms of structure, flexibility (with tocopherol featuring a longer hydrophobic chain compared to cholesterol), and molecular weight (430 Da for tocopherol and 387 Da for cholesterol), all characteristics expected to influence micellization.

The correct assembly of all the nanostructures was confirmed by agarose gel electrophoresis (AGE) using non-functionalized designs as shown in supplementary material Fig. 1. The sequences of all ssDNA components are reported in Table 1 of the supplementary material, while the list of oligonucleotides used for each design is given in Table 2 of the supplementary material.

First, we assess the effect of the aforementioned design variations on the ability of the hydrophobe-modified C-star “arms” to micellize. To directly investigate micellization, rather than higher-order self-assembly, here we study samples consisting of linear DNA duplexes mimicking individual C-star arms [Fig. 2(a)]. We characterize the bulk self-assembly of these DNA-amphiphiles using dynamic light scattering (DLS), as summarized in Figs. 2(a)–2(c). Control samples lacking the hydrophobic modifications were used as reference.

DLS intensity distributions, shown in Fig. 2(b), are similar for all non-functionalized control duplexes (blue lines), both in terms of mean hydrodynamic diameters [D_H, see Figs. 2(a) and 2(c)] and distribution width, indicating lack of unwanted aggregation between the constructs. Micellization is observed in all functionalized samples (red lines), highlighted by the emergence of a peak at larger D_H. In most single cholesterol designs, micelles coexist with a monomeric population of constructs, whose abundance relative to the micelle population is greater for designs with 2A spacers on both termini. In designs with a single cholesterol and variable A-spacer length, the relative abundance of monomeric species decreases with increasing spacer length, with 3′-6A-chol displaying a unimodal size distribution associated with the micelles. It is therefore apparent that the design of the spacer region between the cholesterol and dsDNA duplex has a significant effect on micelle stability, with shorter spacers and terminal ssDNA overhangs having the largest destabilizing effect. Expectedly, micelle stability is enhanced in the double-cholesterol design 5′-2A-chol:3′-2A-chol, where a unimodal D_H distribution is observed. The micelle peak is sharper compared to single cholesterol designs, hinting at a more monodisperse population. The tocopherol-labeled design also displays a unimodal micelle distribution, with a width comparable to that recorded for single cholesterol designs. As summarized in Fig. 2(c), mean micelle hydrodynamic size is consistently between 3 and 4 times that of the corresponding non-hydrophobized duplexes. Micelle D_H is similar across all designs, except for single cholesterol—double-spacer constructs 5′-2A-chol:3′-2A and 5′-2A:3′-2A-chol, which produced micelles with larger D_H. While constructs 5′-2A-chol:3′-2A and 5′-2A:3′-2A-chol are longer than most other designs, they have the same length as the double-cholesterolized design 5′-2A-chol:3′-2A-chol [see diagrams and refer to Fig. 1(e)], suggesting that the larger D_H may result from a higher (mean) number of construct per micelle. In addition, it is also possible that the comparatively weaker hydrophobic interactions produce less-densely packed micelle cores for the single cholesterol designs compared to 5′-2A-chol:3′-2A-chol, which may contribute to the larger D_H.

Although less direct than DLS measurements carried out on native samples, electrophoretic mobility measured by AGE supported the observed influence of spacer design and hydrophobic modification on micellization. As summarized in Fig. 2(d) (and corresponding supplementary material, Fig. 2), all non-functionalized control constructs display a very similar migration distance, which is strongly reduced in their functionalized counterparts, probably as a combined effect of micellization and interaction of the hydrophobic moieties with the hydrogel matrix. While no clear double bands are observed to match the bimodal hydrodynamic size distributions seen in DLS, single cholesterol designs consistently show broader and more smeared bands, which likely contain unresolved micellar and monomeric populations. As with DLS measurements, double-cholesterol constructs produce a sharper band, indicative of a more monodisperse population. Differences in electrophoretic mobility are noted between (double) cholesterolized constructs, which do not straightforwardly map onto the hydrodynamic size trends seen with DLS. However, given that the electrophoretic mobility is also influenced by molecular shape and overall charge, which less directly reflect on D_H, these differences between DLS and AGE measurements are not particularly surprising. Of particular note is the case of construct 3′-2A-chol, which displays hydrodynamic size similar to that of other single cholesterol motifs [Fig. 2(c)] but has significantly lower electrophoretic motility. We speculate that the oddity in the AGE result (observed in independent repeats) may arise from a peculiar micelle shape or less dense packing, the latter resulting in a lower charge density. Similarly, the single-tocopherol construct produced a sharp band only slightly slower than the corresponding unmodified control, hinting that micelles observed in DLS may be destabilized in AGE experiments, possibly as a result of the lower concentration used in electrophoresis experiments (see “Methods”), which may drop below the critical micellar concentration (CMC) for the tocopherol constructs. Consistent with this hypothesis, we observe the presence of (larger) micelles in AGE when the solution is supplemented with divalent cations (magnesium), known to increase the CMC of tocopherol-DNA constructs⁷¹ (supplementary material, Fig. 3).

Regardless of qualitative differences in the hydrodynamic size and electrophoretic mobility trends, both DLS and AGE confirm that altering the details of the constructs at or near the hydrophobic site influences their micellization behavior in terms of preferred size, polydispersity, and stability, as intended. We now ask the question: Do these changes influence the crystallization behavior of C-stars?

To answer this question, we used SAXS to assess the nanoscale structure of materials self-assembled from all C-star designs discussed above. The 2D “powder” diffraction patterns shown in Fig. 3(b), obtained from highly concentrated aggregate samples (see “Methods”), display multiple diffraction rings, hinting at an underlying crystalline structure of all the aggregates. Fits of the radially averaged intensity profiles, depicted in Fig. 3(a), confirm that all tested samples adopt a body centered cubic (BCC) crystal structure, similar to all crystals formed by four-arm C-stars we studied previously.^61–63,66 See Fig. 4 of the supplementary material and the “Methods section” for information on peak fitting and crystal identification.

Somewhat unexpectedly, these results demonstrate that the details of the hydrophobic modification and DNA-linker morphology, despite having a clear influence on micelle size and stability, fail to alter crystal symmetry, which thus appears to be uniquely imposed by the four-arm nanostructure topology,⁶¹ given sufficient junction flexibility.⁶³ Nonetheless, differences are observed between the diffraction patterns in terms of resolution, determined as the d-spacing associated with the Bragg peak observed at the largest q vector (d = 2π/q). Most notably, the sample prepared with double-cholesterol anchors displays the highest resolution at d ∼ 43.5 Å, possibly owing to more stable micelle formation. While not comparable with the values obtained for DNA motifs whose crystallization is mediated by base-pairing,^40–42 this resolution is quite remarkable in a system where self-assembly is induced by nonspecific hydrophobic forces. Some samples, notably 3′-0A-chol, and possibly 5′-2A-chol:3′-2A and 5′-2A:3′-2A-chol, show traces of an amorphous phase coexisting with the BCC crystal, producing smoother modulations in the scattering intensity. These three samples are among those displaying the greater proportion of non-micellized constructs in bulk DLS experiments [Fig. 2(b)], suggesting that the amorphous phase may emerge as a result of weaker hydrophobic interactions between the arms.

The values of the lattice parameter, a, correlate well with the length of the double-stranded DNA arms, which slightly differ across designs, as observed previously.⁶² 

Finally, we note that upon inspection with bright field microscopy, aggregates in nearly all samples display a polyhedral shape consistent with the BCC symmetry, namely, that of a rhombic dodecahedron.⁶¹ Exceptions are samples 3′-0A-chol and 5′-2A-toc, where aggregates are spherical despite the underlying crystalline structure. This observation is not new or unexpected, as we have previously reported similar behaviors in C-star designs where the SAXS diffraction patterns displayed low resolution and/or coexistence of crystal and amorphous phases,^62,63 as we see for 5′-2A-toc and 3′-0A-chol, respectively.

Our previous research revealed that C-star crystals can be loaded with hydrophobic cargoes, for instance rhodamine B, which presumably localizes within the hydrophobic micelle-like cores of the aggregates.^62,65 In this study, we demonstrated that making substantial changes to the hydrophobic modifications does not influence crystal geometry, suggesting that micelle-like sites retain the same coordination in all tested crystal samples. It follows from this observation that the characteristics of the individual micelles must be quite diverse between implementations in terms of physical size and arrangement of molecules. For instance, the double-cholesterol design 5′-2A-chol:3′-2A-chol would be expected to have twice the number of cholesterol molecules in each hydrophobic pocket compared with single cholesterol motifs, while designs that differ in A-spacer structure should have different density and arrangement of unpaired nucleotides in the immediate vicinity of the micellar regions. The characteristics of micelle-like structures would also be expected to differ between cholesterol and tocopherol-modified C-stars. In this section, we test whether the expected differences in micelle properties influence the ability of the crystals to uptake hydrophobic molecular cargoes.

As model payload, we consider hydrophobic antibiotic penicillin V, conjugated to a BODIPY FL fluorescent dye. The crystals were soaked in solutions with different antibiotic concentrations, ρ_PV, and the degree of accumulation of the drug-fluorophore conjugates within the aggregates was determined by computing the ratio, ξ, between the fluorescence intensity measured inside and outside the crystals from confocal images (see “Methods”). Representative micrographs together with extracted ξ values are shown in Fig. 4 [panels (b) and (a), respectively].

Differences in ξ observed between designs at fixed ρ_PV reveal that the partitioning of penicillin V strongly depends on micelle morphology. At lower antibiotic concentrations (ρ_PV ≤ 20 μM), a considerable rise in ξ is detected in samples featuring two single-stranded AA overhangs, regardless of whether only one DNA terminus (5′-2A-chol:3′-2A, 5′-2A:3′-2A-chol) or both termini are functionalized with cholesterol-TEG (5′-2A-chol:3′-2A-chol). We speculate that the greater degree of accumulation noted may arise from (i) the unpaired As reducing packing density close to the core of the micelles, hence leaving more room for the drugs and/or (ii) favorable stacking interactions between the unpaired As and either Penicillin V or the BODIPY FL modification. All single cholesterol-tagged designs lacking two AA spacers show significantly lower accumulation at low ρ_PV. Differences between cholesterolized designs are less prominent at high ρ_PV, probably due to binding-site saturation. Aggregates formed from tocopherol-labeled C-stars show the lowest degree of accumulation at all ρ_PV values. The difference noted between 5′-2A-toc and 3′-2A-chol, which share the same 2A-spacer (although connected to different DNA termini), indicates that the chemical nature of the hydrophobic tag also plays an important role in regulating encapsulation. In particular, is it is possible that the stronger stacking interactions supported by cholesterol, compared to tocopherol, may favor interactions with the aromatic rings in penicillin V or BODIPY FL.

In summary, we have investigated the impact of altering hydrophobic modifications and their immediate vicinity on the ability of amphiphilic DNA junctions (C-stars) to crystallize. In previous studies,^61–63 we had hypothesized that, alongside the number of hydrophobized DNA arms (topology), preferential (bulk) valency and stability of micelles formed by the arms may have an impact on determining crystal geometry. Surprisingly, we observe that this is not the case, as several designs that display notable differences in bulk micelle size and stability produce crystals with the same BCC geometry. This evidence confirms that the nanostructure topology is the single most important factor determining the emergent crystal phase in C-stars—a notion that will help inform the rational design of novel amphiphilic DNA phases. While the characteristics of the hydrophobic tags do not influence crystallization, they naturally result in changes to the hydrophobic pockets within the crystals. We demonstrate that these changes translate into differences in the ability of the materials to uptake a model hydrophobic payload—a penicillin V/BODIPY FL complex. The possibility to easily tune the encapsulation capabilities of C-star materials independently of their crystal structure adds to the arsenal of functionalities afforded by the material. In particular, in the context of drug delivery or bottom-up synthetic biology, it could be useful to retain a defined microstructure, which could be critical to regulate diffusion of macromolecules⁶⁶ and material resilience to degradation while at the same time optimizing the system for uptaking different hydrophobic payloads. Finally, our previous work demonstrated the ability of assemblies of C-stars with one single cholesterol modification (on each arm) to permeabilize and/or rupture the membrane of synthetic lipid vesicles.⁶⁷ We envision that the replacement of cholesterol with tocopherol moiety or increasing the number of cholesterol molecules per arm would allow better control over this process, by enabling targeting of membranes with specific lipid composition⁷² or by tuning disruption efficiency.

DNA structures were designed using the nucleic acid design and analysis software NUPACK.⁷³ All oligonucleotide sequences are listed in Table I of the supplementary material. DNA strands with 5′ tocopherol and 5′ cholesterol modification were purchased from Biomers and Eurogentec, respectively. All the remaining oligonucleotides were purchased from Integrated DNA Technologies (IDT). Non-functionalized strands were purified by the supplier using standard desalting, while chemically modified strands were purified using high-performance liquid chromatography (HPLC). The dehydrated DNA strands were reconstituted in syringe-filtered (0.22 μm pore size, polyethersulfone, Millex) 1× Tris–EDTA (TE) buffer (10 mM Tris, 1 mM EDTA, pH 8.0, Sigma-Aldrich) upon receipt. The absorbance at 260 nm, measured with a Thermo Scientific Nanodrop 2000 UV–Vis spectrophotometer, together with the extinction coefficients provided by the supplier were used to determine concentrations of the reconstituted DNA.

Samples used to assess folding of individual motifs with agarose gel electrophoresis (AGE) were prepared from non-functionalized strands to prevent aggregation. For this purpose, all the oligonucleotides modified with a hydrophobic moiety were replaced by their equivalents lacking the chemical modification.

For the micelle size and stability assay with AGE and DLS (dynamic light scattering), samples containing both the non-functionalized and functionalized strands were prepared.

All the required oligonucleotides (supplementary material, Table 2) were mixed in 200 μl DNase-free Eppendorf tubes to obtain the final DNA concentration of 5 μM for nanostars and 20 μM for micelle-forming and non-aggregating nanostar arms in TE + 300 mM NaCl buffer. To enable the correct assembly of predesigned structures, prepared solutions were thermally annealed in a Techne TC-512 thermal cycler via initial incubation at 95 °C for 5 min followed by temperature decrease to 20 °C at the rate of −0.05 °C min⁻¹. Prior to annealing, samples for the DLS micelle size study were filtered through 0.22 μm pore size Millex filters. Annealed samples were stored at 4 °C and used within 24 h.

Samples of amphiphilic DNA crystals were prepared by mixing the required oligonucleotides (supplementary material, Table S2) at stoichiometric ratios in TE buffer supplemented with 300 mM NaCl in 500 μl DNase-free Eppendorf tubes, to achieve the final nanostar concentration of 5 μM in 60 μl solutions. Prepared mixtures were then loaded into borosilicate glass capillaries (inner section of 4 × 0.4 mm², CM Scientific), pre-cleaned through sonication in 2% Hellmanex III water solution (Hellma Analytics) for 15 min followed by two 15 min sonication cycles in ultrapure water (Milli-Q) to remove the surfactant. Sample-loaded capillaries were capped off with a small volume of mineral oil (Sigma-Aldrich) and sealed permanently onto microscope cover glass slides (Menzel Gläser, 24 × 60 mm², No. 1) with two-component fast-drying epoxy glue (Araldite). Sealed capillaries were placed in a custom-made, Peltier-controlled water bath and slowly cooled down from 95 to 20 °C at the rate of −0.01 °C min⁻¹ to allow crystallization, after an initial incubation at 95 °C for 30 min.

AGE was used to assess the correct folding of non-functionalized DNA motifs (see Fig. 1 of the supplementary material).

Free nanostar samples, prepared according to the aforementioned protocol, were diluted in Tris-borate EDTA (TBE) buffer (pH 8.3, 89 mM Tris-borate, 2 mM EDTA, Sigma-Aldrich) and mixed with a loading dye (Thermo Fisher Scientific) at 5:1 sample:dye ratio to obtain the final DNA concentration of 75 ng μL⁻¹.

Agarose gels were prepared at 1.5 wt. % agarose (Sigma-Aldrich) in TBE buffer and precast with SYBR safe DNA gel stain (Invitrogen Thermo Fisher Scientific) to the typical thickness of 5 mm. Each well was filled with a small volume (10 μl) of diluted sample containing 750 ng of DNA. The two outermost wells carried a 100 bp DNA reference ladder (Thermo Fisher Scientific) at the same mass concentration. A potential of 75 V, equivalent to 3 V cm⁻¹, was applied for 120 min to allow DNA migration. The imaging of gels was performed with a GelDoc-It imaging system.

AGE and DLS were used to determine the stability (AGE) and size (AGE and DLS) of micelles formed by nanostar arms functionalized with various hydrophobic moieties (see Fig. 2 and supplementary material, Figs. 2 and 3).

AGE was performed as in the protocol described above, using the same electrophoresis setup. For the experiment aimed at testing the effect of cation identity on the micellization of tocopherol-modified DNA (supplementary material, Fig. 3), TBE buffer used for electrophoresis was additionally subsidized with 10 mM MgCl₂.

DLS data were collected using a Malvern Zetasizer Nano ZSP analyzer, equipped with a 633 nm He–Ne laser and the scattering angle being fixed at 173°. For these experiments, 100 μl volumes of samples prepared according to the aforementioned protocol were loaded into an ultralow volume quartz cuvette (ZEN2112, Malvern). For each sample, three measurements consisting of 15 data runs were taken. Throughout the data acquisition process, temperature was kept at 25 °C using an inbuilt Peltier heating block.

SAXS measurements on DNA aggregates [Figs. 3(a) and 3(b)] were taken at the I22 beamline of the Diamond Light Source. A radiation wavelength of λ = 1 Å and beam cross-section of ∼300 × 100 μm² were used. The accessible q-range was 0.005–0.18 Å. The q-scale was calibrated using silver behenate.

Amphiphilic DNA crystals for SAXS measurements were prepared as described above. Aggregates were extracted from glass capillaries and concentrated by centrifugation and supernatant removal, to a final nanostar concentration of 100 μM. The concentrated samples were then loaded into x-ray capillaries (1.5 mm external diameter, 0.01 mm wall thickness, Capillary Tube Supplies Ltd.) and left to settle, forming a visible pellet.

Measurements were collected by scanning the pelleted DNA crystals, with one frame taken per location with an exposure time of 100 ms. Collected 2D patterns were reduced to 1D by radial averaging using DAWN.⁷⁴ A minimum of 20 1D patterns from different locations were then averaged for each sample and an arbitrary logarithmic background was subtracted. Peak positions were determined from background subtracted data using a custom MATLAB script. Rough positions were first determined using the inbuilt “findpeaks” function and then refined by fitting each peak with a single Gaussian using the positions from “findpeaks” as initial estimates.

To determine the lattice parameter a, Miller indices were first assigned to each peak by calculating the ratios of the square of the measured peak position q² to the square of the position of the first scattering peak $qmin2$⁠, multiplying them by sequential integers until all the values could be represented as the sum of the squares of three integers h² + k² + l² (note that here a factor of two was enough to achieve this so h² + k² + l² = 2n, where n is a positive integer), and then assigning hkl values that give the sum of squares equal to the expected one (for example, if $q2/qmin2=1$⁠, $2×q2/qmin2=2$⁠, and h² + k² + l² = 2, then hkl can be assumed to be 110). Afterward, the spacing between successive (hkl) planes, termed d-spacing, was calculated using the Bragg equation⁷⁵ and plotted against the extracted hkl values represented as $(h2+k2+l2)−0.5$ (supplementary material, Fig. 4). Finally, the relation

was used to extract a as the gradient of the linear fit of the d vs $(h2+k2+l2)−0.5$ plots.

Bright field microscopy images of DNA crystals shown in Figs. 1(c) and 3(c) were obtained with a fully motorized and programmable Nikon Eclipse Ti-E inverted epifluorescence microscope equipped with a CFI Plan Apochromat λ 40×/0.95 NA dry objective (Nikon), Grasshopper3 GS3-U3-23S6M camera (Point Gray Research), and a Perfect Focusing System (Nikon), preventing sample defocusing during imaging.

Confocal microscopy was utilized to assess the uptake of the antibiotic penicillin V by amphiphilic DNA crystals with different morphology of the hydrophobic region (Fig. 4).

For this experiment, 60 μl volumes of samples of amphiphilic DNA crystals, prepared as described above, were pipetted into silicon incubation chambers (6.5 × 6.5 × 3.5 mm³, Grace Biolabs FlexWells). Each chamber was then supplemented with a small volume (1–10 μl) of a highly concentrated (1.2 mM) fluorescent penicillin V (BOCILLIN FL Penicillin, Invitrogen/Thermo Fisher Scientific) solution and topped up with TE + 300 mM NaCl buffer to the final volume of 120 μl. Afterward, all the chambers were sealed with DNase-free tape (Grace Biolabs FlexWell SealStrips) to prevent evaporation and left for one hour to equilibrate.

Confocal micrographs of prepared samples were recorded with a Leica TCS SP5 laser scanning confocal microscope using a HC PL APO CORR CS 40×/0.85 dry objective (Leica). For excitation of BODIPY FL dye attached to penicillin V, an Ar-ion laser line at 488 nm was used.

See the supplementary material for oligonucleotide sequences and sample composition (Supplementary Tables 1 and 2, respectively), gel electrophoresis images and analysis (Supplementary Figs. 1–3), and SAXS data analysis (Supplementary Fig. 4).

L.D.M. acknowledges support from a Royal Society University Research Fellowship (Grant No. UF160152) and from the European Research Council (ERC) under the Horizon 2020 Research and Innovation Program (Grant No. ERC-STG No 851667–NANOCELL). A.L. and L.D.M. acknowledge support from a Royal Society Research Grant for Research Fellows (Grant No. RGF/R1/180043). M.W. acknowledges support from the Engineering and Physical Sciences Research Council (EPSRC) and the Department of Physics at the University of Cambridge (the McLatchie Trust fund). The authors acknowledge Diamond Light Source for providing synchrotron beamtime (SM24537 and SM29072) and thank A. Smith for assistance in operating beamline I22. The authors thank P. Cicuta for feedback on the project.

The authors have no conflicts to disclose.

Michal Walczak: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Software (equal); Validation (lead); Visualization (lead); Writing – original draft (equal); Writing – review & editing (equal). Ryan A. Brady: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Writing – review & editing (equal). Adrian Leathers: Investigation (equal); Writing – review & editing (equal). Jurij Kotar: Methodology (equal); Resources (equal); Writing – review & editing (equal). Lorenzo Di Michele: Conceptualization (lead); Funding acquisition (lead); Project administration (lead); Resources (lead); Supervision (lead); Validation (lead); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are openly available at https://doi.org/10.17863/CAM.92820.