Inducing Lipid Domains in Membranes by Self‐Assembly of DNA Origami

Self‐assembly of biological molecules and structures is a fundamental property of life. Whereas most biological functions are based on self‐assembled proteins and protein complexes, the self‐assembly of lipids is important for the spatial organization of heterogeneous cellular reaction environments and to catalyze cooperative interactions on/with membranes. Lipid domains or “rafts”, which are known to selectively recruit proteins, play an important functional role in sorting and trafficking of membrane components between subcellular organelles. However, how the recruitment and interactions of proteins in turn contributes to the formation and turnover of these structures has not been systematically addressed, due to the large variety in membrane–protein features and their spatiotemporal dynamics. The small size and transient nature of lipid domains adds to the complexity in visualizing them in living cells. Here, DNA origami is presented as a programmable tool to mimic protein clustering and assembly on membranes and illustrate how nanometer sized lipid domains coalesce into visible domains upon origami self‐assembly in defined patterns. Hence, the local membrane composition can be efficiently regulated by the self‐assembly of peripheral membrane binders. This reinforces the hypothesis that lipid rafts in cells occur as a result of membrane–protein interactions and, in particular, protein self‐assembly.

membrane domain formation also affects the protein conformation and oligomerization. [9] The characterization of these domains in vivo is difficult, because of their small size and highly dynamic nature. [10] Hence, model membranes, mostly in the form of giant unilamellar vesicles (GUVs), serve as a vital tool because of their relative ease of preparation and physical-chemical modulation, as well as their suitability for analysis by light microscopy. [11] Ternary lipid mixtures have been particularly well suited to understand dynamics of micrometer sized domains, as they exhibit spontaneous demixing into two defined liquid phases. [12] These domains show different characteristics, based on their packing efficiency. In general, liquid-ordered (L o ) domains are enriched in phospholipids with saturated acyl chains, sphingolipids, and cholesterol, which make them characteristically more tightly packed than the surrounding bilayer. On the other hand, liquid-disordered (L d ) phase domains preferentially contain phospholipids with unsaturated carbon chains that are more loosely packed. L o /L d phase separation in model membranes has been suggested as the underlying basis for the raft concept in biological membranes. [13] DNA origami has been frequently used as a tool to study selfassembly, as it can be customized to mimic shape and properties of functional biomolecules. [14] Long single-stranded DNA in combination with short oligonucleotides can be assembled to yield custom nanostructures of a defined shape. For this approach, short oligonucleotides, also called staple strands, are sequence specifically bound to predefined segments of a long singlestranded scaffold DNA. Each staple strand hybridizes with several scaffold-strand segments, which ultimately lead to constraining of the latter to helical structures arranged as 2D or 3D arrays as desired. The structures can be fine-tuned based on targeted insertions or base pair deletion within the staple strands in order to achieve twisting or curving of DNA bundles at defined angles and sites. The possibility to functionalize each staple strand in the DNA origami can be used as a template for protein assembly. [15] The size of a single DNA origami structure is inherently limited by the length of the scaffold. Thus, larger assembled structures can be hierarchically achieved from folded DNA origamis via sticky end cohesion and/or blunt end interactions. Constraining the structures to surfaces such as membranes has been shown to be beneficial to assemble large scale 2D DNA origami arrays, [16] which can be achieved by the stepwise control of surface diffusion. [16a-e] This can be realized by changing the cation concentration on the surface, in particular by exploiting the competition between Mg 2+ and Na + in DNAmembrane binding. [15j] Supported lipid bilayers (SLBs) as well as GUVs have been used as soft templates for the self-assembly of DNA nanostructures. [17] To accommodate their lateral diffusion on the membrane, weak adsorption is required that is best implemented by using cholesterol anchors.
Self-assembly of DNA origami under the influence of membrane phase separation has been widely studied, but their effect on membrane characteristics remains unknown. [17] Owing to their exquisite programmability of structure and self-assembly characteristics, we hypothesize that DNA origami represents a perfect toolkit to mimic the influence of self-assembly of membrane proteins on lipid membrane organization. By following origami structures assembling on specific lipid domains, we observed membrane reorganization and the induction of phase separation in previously homogeneous ternary lipid mixtures (Figure 1). The specific lipid composition resembles the small size and transient nature of lipid nanodomains in cellular membranes at physiological temperatures. Intriguingly, membrane domains are likely to emerge from the phase boundary region. In spite of the largely different nature of membraneattached macromolecules used here, our assay provides direct access to the interplay between membrane domain formation and protein assembly as a result of recruitment of proteins to specific membrane environments.

Design of DNA Origami
DNA origami structures are based on different cross-linking strategies, i.e., staple triggered (sticky ends) and Mg 2+ triggered, which were designed using caDNAno software. [18] We chose different designs to ensure inclusivity of the design features in terms of their cross-linking strategies as well as their intrinsic shapes. First, the rectangular one-layer origami structure was adapted and improved from our previously reported design with some modifications. [19] DNA sequences and design features can be found in detail in Figure S1 and Table S1 of the Supporting Information. In particular, the design consists of 24 helices (≈70 × 100 nm) and is modified with more staples for increased binding strength to lipid surfaces through cholesterol antihandles, and additional staples to allow for the cross-linking at desired sides of the nanostructure (Figure 2a). For the immobilization, 16 staples in total were equally distributed at the bottom of the structure, which can then hybridize with the corresponding single stranded DNA of oligonucleotides modified with a cholesterol anchor. The number of staples was increased to facilitate more binding sites to the membrane surface via the cholesterol anchors to ensure stronger binding and further subunit oligomerization of the DNA origami. The cross-linking staples were designed as follows. To gain maximum crosslinking efficiency without too much loss of specificity, four staples with polyA overhangs were placed at each side, which can cross-link with those of another origami using polyT connector strands. In comparison to previous structures, [ 19 ] the overhangs do not have the same directionality but are alternating, which is expected to increase the specificity. The staples used to attach fluorophores to the origami consist of a 19-nt overhang where the corresponding oligonucleotide modified with an imager could permanently bind.
Further, the rod-shaped structures were chosen and modified from previously reported structures, [20] as depicted in Figure S2 and Table S2 of the Supporting Information. In particular, the modifications included staple extensions for fluorescence imaging and immobilization on lipid surfaces with cholesterol handles (Figure 2b). 18 staples needed for the immobilization on the surface were equally distributed on the bottom of the structure, while the staples needed for imaging were placed just on one strand covering the whole length of the rod-shaped origami. The strategy for cross-linking this structure is based on base stacking. Therefore, the ends of the origami at some positions do not include overhangs or loops, thus allowing for better stacking at increased MgCl 2 concentration. The designs of the DNA nanostructures were found to successfully cross-link to form self-assembled superstructures on lipid membranes.

Partitioning of DNA Origami on Phase Separated Membranes
To investigate the lipid domain specificity of DNA origami structures, first we visualized the differential localization of DNA origami on the surface of phase separated membranes (SLBs and GUVs) using confocal laser scanning microscopy. Figure 3 shows representative microscopic images to observe the binding preference of DNA origami (sticky ends, monomeric) on phase separated SLBs. Here, we chose a ternary mixture of lipids to observe the microscopic phase separation into coexisting L o and L d phases. In particular, the lipid composition comprises of DOPC, eSM, and cholesterol at ratios 1:1:1. The L d domain is labeled with Atto-655 DOPE and the L o domain is unlabeled. DNA origami binds to the membrane via cholesterol anchors mediated by Mg 2+ ions. [15h,j] In the presence of buffer with high Mg 2+ (20 mm) concentration, the origami structures  Table S3 of the Supporting Information. Further, in the case of GUVs with a ternary composition of the same lipid mix at ratios 2:2:1 a similar behavior is observed (Figure 3c,d). The phase diagram for this mixture is well established in the literature. [12f ] Monomeric DNA origami colocalizes with L o phase at high Mg 2+ concentration and at L d phase at lower Mg 2+ concentration. Our results are in agreement with previously reported observations. [15h] The binding and self-assembly of origami structure depends on the lipid composition and cation concentration in the buffer. In case of the coexisting phases, the lipids still freely diffuse in the membrane. The observed phase preference has also been reported previously in short DNA oligonucleotides, long double-stranded DNAs as well as nanostructured DNA, which all preferentially bind to the L o phase in the presence of high concentration of divalent cations. [21] This preference results from the difference in charge densities between the L o and L d phases due to their varying degrees of lipid packing. In particular, the L o phase consists of tightly packed saturated lipids resulting in a higher charge density as opposed to the L d phase. Moreover, the L o phase does not exhibit fluidic properties and thus is more energetically favorable for binding of nanosized particles in the presence of divalent cations. [15j] This suggests that the self-assembly depends on the fluidity of the bilayer and hence, the charge density of the bilayer surface. Thus, selfassembly can in principle occur on either domain, based on the divalent cation concentration in the phase separated system. [15j]

DNA Origami Assembly Induces Microsized Lipid Domains in Homogeneous Ternary Lipid Mixture
Because of the clearly distinctive binding of DNA origami on phase separated membranes, we sought to investigate whether the self-assembly of DNA origami could in turn induce local demixing of homogeneous membrane. Here, we chose a lipid composition mimicking cell membrane composition that does not phase separate into visible domains, i.e., DOPC:eSM:chol in the ratio 10:65:25. This lipid mixture instead forms nanoscopic domains, as observed by atomic force microscopy (AFM), in which freely diffusing lipids dynamically exchange between a fluid and a nanodomain-associated state. [1f-g] In particular, lipid domains ranging from 100 to 200 nm size are observed for this lipid composition (Figure 4a). In live cell membranes, direct transient lipid interactions were first observed as nanodomains of the order of 50 nm. [1f ] Thus, with this composition, we not only aim to resemble heterogeneity of the membranes but also the small and transient nature of these domains in the cell membrane.
Upon incubation of SLBs with DNA origami (rectangular sticky ends, monomeric), we observed an apparently homogeneous membrane binding. At 5 mm Mg 2+ concentration, the origami exhibits the preference for the L d domains as discussed previously, but since the nanodomains are distributed all-over the membrane (Figure 4b), this results in a uniform coverage of the membrane by origami without disrupting the nanodomains. Moreover, the origami diffuses rapidly in its monomeric form on the membrane surface (Video S1, Supporting Information).
Upon addition of the cross-linkers (polyT60), the diffusive origami structures on the surface start to form assemblies. The fluid membrane surface allows for their cross-linking and  self-assembly into origami superstructures (here R2), thereby spatially recruiting the membrane phase to which they exhibit a strong binding preference. The indirectly coupled diffusion of lipids and origami results in the emergence of visible L d domains after cross-linking of the structures into superstructures ( Figure S3, Supporting Information). Due to the phase preference of DNA origami based on the respective salt conditions, the cross-linking of DNA origami into micrometersized assemblies presumably drives the nanodomains closer to one another, thereby promoting their coalescence into larger microdomains. These domains become more pronounced over time with the growth of self-assembled origami structures. The anchored DNA origami rapidly diffuses in its monomeric form, but diffusion becomes much slower and a large fraction becomes even immobilized upon cross-linking by polyT crosslinkers. [19] This decrease in mobility further supports the formation of superstructures (Video S2, Supporting Information). The self-assembly of DNA origami decreases the diffusion coefficient of the anchored lipid molecules in the membrane ( Figure S4, Supporting Information) from 1.6 ± 0.2 to 0.8 ± 0.3 µm 2 s −1 . It is to be noted that the diffusion of lipids molecules is slowed more pronounced after overnight incubation (12 h) as compared to 30 min, indicating that the growth of superstructures continues on large time scales. We observed similar results when DNA origami was allowed to diffuse and cross-link on the freestanding membrane of a GUV surface (Figure 4c). While the monomeric origami structure binds homogeneously to the membrane, the cross-linking of membrane-anchored origami drives the L d domains closer to one another and promote their growth accompanying the self-assembly of the origami superstructures. To allow for maximum cross-linking of the origami structures and the resultant domain formation, the samples were again incubated overnight (≈12 h). Along with the slowed diffusion indicating the formation of bigger superstructures over time, we also observe the resultant growth of membrane domains. Hence, the initially homogeneous membrane starts to exhibit macroscopic phase separation (Video S3, Supporting Information). Similar results are also observed when the origami preferentially binds to L o domains.
It is to be noted that we tuned these experiments through the diffusion of origami on either domains. In particular, we used a lower concentration of Mg 2+ to drive the diffusion of DNA origami on L d phase and higher concentrations for L o phase. Since DNA origami readily diffuses over the membrane, it can cross-link on wither domain and exhibit macroscopic phase separation. This effect is the consequence of two processes: the lipid domains L d recruiting the DNA origami at low salt conditions, as well as cross-linking of DNA origami by complimentary base pairs. Since both these two processes are simultaneously enhanced, the lipid domain diffusion is directly connected with the assembly of DNA origami. Thus, the diffusion of lipids is controlled and directed by origami diffusion regulated by their assembly on the membrane. Thus, the lipid domains could be induced by origami self-assembled to form superstructures.

Inducing and Concurrent Patterning of Lipid Domains by DNA Origami
The diffusion of origami structures on the membrane domains allows for their cross-linking to form larger superstructures. This membrane-mediated assembly leads to changes in the domain behavior by segregating nanodomains together. This is analogous to the raft behavior in cell membranes arising as a result of interaction with proteins and their self-assemblies. However, there is an exhaustive array of patterns resulting from the self-assembly of proteins in vivo, and the function of cellular protein networks is directly influenced by the patterns they form. [22] Well-ordered protein aggregates can form spontaneous assemblies resulting into different patterns and shapes, such as viral capsids, tubulins, actin filaments, and flagella. [23] Interestingly, this pattern organization affects the architecture of the cell membrane. [22] For example, actin filaments can organize into dynamic patterns by self-assembly, such as vortices, asters, and stars. However, the plasma membrane lipid order increases when binding to the asters, but does not alter upon interaction with the stars or actin bundles. This influences the membrane organization to different degrees due to difference in membrane anchoring with the resultant selfassembled patterns. [22] Control of pattern formation is thus crucial in artificial biomimetic systems as they can lead to changes in the membrane properties.
Inspired by this, we set out to observe how different patterns of origami superstructures can affect the segregation of membrane domains. For this, we choose origami monomers resulting in different superstructures due to their intrinsic shape as well as cross-linking strategy. In particular, R2 structures can only cross-link at the opposite edges resulting in micrometer-scale striped structures, whereas R4 results into more global structures owing to cross-linking sites on all four sides. Similarly, rod structures LE are taller and skinnier whereas LS superstructures are thick but short. Based on these structural differences, we determine how differently lipid domains arise with the anchored origami superstructures.
First, with anchored origami monomers, the GUV surface does not exhibit microscopic phase separation (Figure 5a). This is because the nanoscaled DNA origami binds to the lipid domains via the cholesterol anchors which is mediated by Mg 2+ ions present in the buffer. In general, lipid headgroups control the partitioning of origami, which is supported by divalent cations. Divalent cations enhance the ability of DNA to interact with membranes composed of zwitterionic headgroups by efficiently screening the electrostatic charge of the DNA when binding to the membrane. This occurs by a notable reorientation of the headgroup toward the DNA easily distinguished from the lipid domains), b) inducing domains with self-assembled origami superstructures on SLBs. Here, non-cross-linked DNA origami (monomeric) binds to the SLB uniformly without preference for either of the lipid domain (top panel) whereas cross-linking of DNA origami results into inducing domains in the SLB by reorganizing the membrane with the anchored origami structures (bottom panel). c) GUVs with origami cross-linked preferentially on either domain, based on respective cation concentration in the buffer (5 mm for L d domains and 20 mm for L o domains). Cross-linked origami results into macroscopic phase separation. Scale bar: 5 µm.

www.advmatinterfaces.de
to allow it to bind. [24] The binding of divalent cations to the membrane is weaker when the degree of unsaturation in the lipid molecule is higher. [25] As a result, DNA origami preferentially binds to L d domains at lower Mg 2+ ion concentration, whereas it is more favorable for binding at L o domains at higher concentrations of Mg 2+ . [15h] Even though the interaction is preferential for either domain, but since the size of this origami is too small, it fails to perturb the nanodomains and does not result into visibly larger domains as also mentioned earlier in Section 2.3.
Furthermore, the self-assembly of DNA origami is aided by incubation of the origami above the phase transition temperature of the membrane. When the membrane is still fully disordered, it is more fluid and allows for the origami to diffuse and cross-link. This is done to allow for the active diffusion of lipids to coalesce with origami. With the decrease in temperature, the domains also grow and match the shape of the superstructures. It is apparent that the growth of domains and self-assembly of DNA origami is coupled with each other and thus takes place simultaneously. Hence, the domains are able to adopt the precise patterns of origami superstructures.
In particular, origami R2 and R4 bind preferentially to L d domains in the presence of 5 mm Mg 2+ concentration and to L o domains at 20 mm Mg 2+ ions (Figure 5b,c). However, for the rod-like origami structure, a higher concentration of divalent ions is required for cross-linking to form assemblies as the rod structures are designed to cross-link with base stacking triggered by an increase of Mg 2+ concentration as already mentioned in Section 2.1. Thus, we see them binding to L o domains at high Mg 2+ concentration of 70 mm (Figure 5d). In all cases, with selective binding of cross-linked structures on membrane domains, the domains follow and copy the shapes of anchored origami superstructures. The cross-linking and self-assembly of DNA origami hinders the diffusion of lipid domains underneath and thus just maps the lipid domains depending on the shape of the superstructures formed. In order to eliminate the effect of changes in Mg 2+ concentration, we measured the lipid diffusion coefficient of the membrane over a range of 0-70 mm Mg 2+ concentration ( Figure S5 and Table S4, Supporting Information). We observe a slight decrease in the diffusion of the lipids, but the effects are negligible as compared to membrane binding with the cross-linked origami. We observe a similar effect in SLBs where the origami is anchored and self-assembled in order to map the domains ( Figure S6, Supporting Information). Thus, our results indicate that the lipid domains in living cells which are transient and nanometer-ranged can segregate as a result of membrane-protein interactions and their respective self-assembly. In principle, lipid rafts might arise from such interactions resulting into functional compartments where lipid compositions could be locally and effectively regulated.

Conclusion
In conclusion, we demonstrate the effect of DNA origami assemblies on lipid phase behavior. Anchored DNA origami shows a preferential binding to either lipid phase depending on the salt conditions. Thus, diffusion of origami on the membrane surface toward self-assembly leads to inducing local phase separation in the membrane. Here, we found that membrane composed of a ternary lipid mixture that does not phase separate into microdomains can spontaneously form domains when cross-linking anchored origami structures on their surface. The chosen lipid composition was found to exhibit nanoscopic domains, which coalesce into microscopic domains upon the diffusion of origami to form superstructures over ≈12 h, i.e., the anchored membrane domain grows from initially present nanoscopic domains with the self-assembly of origami superstructures. Above the phase transition temperature, the lipid domains can still grow with the shape of origami superstructures. The present study could help to understand www.advmatinterfaces.de the dynamic reorganization and domain formation of lipids in heterogeneous cellular membranes, which are assumed to be well above the phase separation temperature, following complex protein assembly and disassembly. Our findings thus confirm the hypothesis that protein recruitment and assembly depend on the membrane phase state, and vice versa. Overall, our results reinforce the idea that protein functionality on membranes is not only controlled by protein-protein interactions, but also by lipid-protein and lipid-lipid interactions.

Experimental Section
DNA Origami Folding and Purification: The DNA origami structures were assembled as described before. [19,20] The single stranded scaffold DNA (p7249, tilibit nanosystems, 10 nm) was mixed with the staple oligonucleotides (eurofins genomics) in a 1:10 ratio in a folding buffer (1xTE, 5 mm NaCl) containing required amounts of MgCl 2 . For the rectangle the buffer consisted of 12.5 mm MgCl 2 and the folding reaction was performed using an initial 5 min melting step at 80 °C, followed by decreasing temperature ramp 60 to 4 °C over 3 h. By contrast, the folding reaction of the rod-shaped origami consisted of 20 mm MgCl 2 with an initial melting step 65 to 60 °C in 1 h and temperature ramp from 59 to 40 °C in 40 h. The structures were purified using a PEG-purification. The formed nanostructures were diluted in a 1:1 ratio with PEG buffer (15% w/v PEG-8000 (89510), 250 mm NaCl) and centrifugated (30 min, 17 900 rcf, 4 °C). After discarding the supernatant, the DNA origami was resuspended in the folding buffer.
SLB Preparation: SLBs were prepared via fusion of Small Unilamellar Vesicles (SUVs) deposited on top of freshly cleaved mica already glued on top of a glass coverslip, as described elsewhere. [15g,20] The lipid mix used is DOPC:eSM:chol in ratio 10:65:25 for homogeneous membranes and 1:1:1 for phase separated membranes, each labeled with 0.001 mol% Atto 655-DOPE. 0.75 mm of the lipid mix prepared in buffer 1 (5 mm tris-HCl, 1 mm EDTA, 5 mm MgCl 2 , 300 mm NaCl, pH 8.0) was sonicated for 10 min. After the lipid mix becomes clear, it was fused on top of mica and incubated at 50 °C for 5 min, which was then rinsed several times with 10 mL buffer. At the end, 300 µL of the sample was kept in the chamber before imaging.
GUV Preparation: GUVs were prepared by electroformation in PTFE chambers with Pt electrodes, as previously described elsewhere [15g,20] with minor modifications. Briefly, 6 µL of lipid mix (2 mg mL −1 dissolved in chloroform) was spread onto two Pt wires and dried in a desiccator for 30 min. The chambers were then filled with 350 mL of an aqueous solution of sucrose (≈350 mOsm kg −1 ) iso-osmolar to the imaging buffer. An AC electric field of 2 V (rms) was applied at frequency of 10 Hz for 1 h, followed by 2 Hz for 0.75 h at 50 °C. The sample was allowed to cool to room temperature for ≈2 h before imaging.
Membrane and DNA Origami Assembly: SLBs were prepared on freshly cleaved mica as described above. Then cholesterol anchors (10 nm) were incubated with SLBs for 10 min, after which the sample was washed gently with Tris buffer containing required MgCl 2 . Selfassembled DNA origami was prepared in buffer with 0.1 nm respective DNA origami, 10 nm imagers, and 100 nm cross-linkers (polyT60) to ensure maximum cross-linked structures. This mix was then incubated with the lipid mix at 37 °C and allowed to cool to room temperature before imaging. For GUVs, a similar protocol was applied. GUVs were incubated with cholesterol anchors (10 nm) and a prepolymerized DNA origami mix in high or low Mg 2+ buffer at a temperature of 37 °C before it was allowed to cool to RT. The osmolarity was maintained at nearly 300 mOsm kg −1 for all the experiments for the GUVs. Samples were images after overnight incubation of the origami with the membrane.
Confocal Microscopy: All confocal images were taken on a Zeiss LSM780 confocal laser scanning microscope using a Zeiss C-Apochromat 20× air or 40 × W objective (Carl Zeiss). Membrane (GUVs and SLBs) labeled with Atto 655 DOPE was excited using a 633 nm He-Ne laser and DNA origami labeled with Atto 488 (or Cy3B) was excited using a 488 nm Ar laser (or 561 nm laser). Images were typically recorded with a pinhole size of 2.6-4 airy units for the channels 512 × 512-pixel resolution. Images were analyzed in ImageJ (http://rsb.info.nih.gov/ij/). Atomic Force Microscopy: Measurements were performed on a JPK Nanowizard 3. AFM images were taken in quantitative imaging mode using BioLever Mini BL-AC40TSC2 (Olympus) or PEAKFORCE-HIRS-F-B (Bruker) cantilevers. The set point force was 0.15-0.25 nN, acquisition speed 66.2 µm s −1 , Z-range 106 nm. Images were first processed in JPKSPM Data Processing (JPK, v6.1.142) performing a line-wise second-degree polynomial leveling followed by another second-degree polynomial leveling with limited data range (0% lower limit, 70% upper limit). Subsequent plane leveling, third-degree polynomial row alignment, scar correction, and cutting off of outliers past 1.2 rms were performed in Gwyddion (v2.58, https://www.gwyddion.net/).
Fluorescence Correlation Spectroscopy (FCS): Measurements were performed using a Microtime 200 system (PicoQuant GmBH, Germany) integrated with an FCS module, a dual SPAD detection unit, timecorrelated single photon counting, and inverted microscope model Olympus X1-71 equipped with a water immersion objective (UPlanApo 60×, NA 1.2, Olympus, Japan). All measurements were carried out with pulsed excitation at 40 MHz. The fluorescence emission was collected by the same excitation objective, focused through a 50 µm pinhole. Raw data were correlated and fitted with custom written Python scripts. 2D diffusion model was used for fitting correlation curves for supported lipid bilayers.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.