Articleshttps://doi.org/10.1038/s41557-018-0196-3

1Center for Systems and Synthetic Biology, University of Texas at Austin, Austin, TX, USA. 2Institute for Cellular and Molecular Biology, University of Texas at Austin, Austin, TX, USA. 3Department of Molecular Biosciences, University of Texas at Austin, Austin, TX, USA. 4Department of Chemical Engineering, University of Michigan, Ann Arbor, MI, USA. 5US Army Research Laboratory – South, Austin, TX, USA. 6Department of Materials Science & Engineering, University of Michigan, Ann Arbor, MI, USA. 7Biointerfaces Institute, University of Michigan, Ann Arbor, MI, USA. 8Present address: Division of Biotechnology, College of Life Sciences and Biotechnology, Korea University, Seoul, Republic of Korea. 9These authors contributed equally: Anna J. Simon and Yi Zhou. *e-mail: dtaylor@utexas.edu; andy.ellington@mail.utexas.edu

The assembly of genetically encoded molecules into symmetric, supramolecular structures and materials supports the acqui-sition of complex functionality in natural biological systems,

and would likewise prove valuable in the development of bionano-technologies such as drug delivery, energy transport and biologi-cal information storage1–7. The formation of well-defined, complex, and physically and allosterically coupled assemblies from a small number of genetically encoded components is in turn the product of repeated, symmetric interactions between molecular building blocks3–7. However, while exploration of de novo engineering of artificial biomolecular assemblies has begun, these efforts gener-ally rely on the precise and targeted design of complex molecular interfaces8–18. In contrast, studies of simple synthetic colloids sug-gest that fairly complex structures can be derived based solely on packing and energetic considerations. For example, computational and experimental studies of polyhedral and spherical colloids indi-cate that complementary particle shapes19–24 and simple attractive ‘patches’25–30 alone can enable the formation of complex symmetri-cal assemblies. Shape complementarity has similarly been exploited experimentally to arrange synthetic linear biomolecules, specifi-cally double-stranded DNA strands, into distinct, higher-order structures, suggesting a utility beyond inorganic systems31,32.

Here, we demonstrate a simple, robust strategy to assemble monomeric proteins into entirely synthetic, well-defined oligo-mers. This strategy centres on driving protein–protein interactions by combining engineered, oppositely supercharged variants of an initially monomeric protein. Generally, oppositely charged proteins associate through simple electrostatic interactions33. Previously, this propensity for association has been exploited in artificial bio-systems to engineer binary protein crystals34,35, capsules36–38 and Matryoshka-like cages39 from naturally charged or self-assembling

proteins. We suspected that, on combining pairs of highly oppo-sitely charged, monomeric-engineered protein variants, shape and physicochemical features would also favour assembly along par-ticular geometrically and physicochemically favoured interfaces, producing well-defined architectures rather than amorphous aggre-gates (Fig. 1a). The ability to simply engineer proteins to assemble into such synthetic scalable molecular assemblies via supercharg-ing may prove useful for technologies ranging from pharmaceutical targeting to artifical energy harvesting and ‘smart’ sensing and building materials1,8–18,36–40.

ResultsOppositely charged GFP variants interact. As a model system we used variant pairs of monomeric superfolder green fluorescent proteins (GFPs) containing introduced basic or acidic amino acids on their surfaces, rendering them positively or negatively charged, respectively (Fig. 1b and Supplementary Fig. 1)41,42. Beyond the ability to observe fluorescent properties, the simple barrel shape of GFP potentially enables assembly into several crystalline43,44 and non-crystalline45 geometries. For the positively supercharged pro-tein we used a monomeric GFP variant with a net charge of + 33 at pH 7.4 (GFP+ 33)41, and for the negatively charged protein we used monomeric GFP variants with net charges of − 10, − 17 and − 31 at pH 7.4 (GFP− 10, GFP− 17 and GFP− 31)42. To enable Förster reso-nance energy transfer (FRET) measurements between positively and negatively charged proteins we introduced a Y66W mutation to blueshift the GFP+ 33 chromophore along with three associated sta-bilizing mutations (F146G, N147I, H149D)46–48, ultimately yielding ‘Ceru+ 32’. Denaturing gels (Supplementary Fig. 2) and absorbance/emission spectra (Supplementary Fig. 3) indicated that each protein was expressed monomerically and with the expected fluorescent

Supercharging enables organized assembly of synthetic biomoleculesAnna J. Simon   1,9, Yi Zhou2,3,9, Vyas Ramasubramani   4, Jens Glaser   4, Arti Pothukuchy1, Jimmy Gollihar2,5, Jillian C. Gerberich2, Janelle C. Leggere2, Barrett R. Morrow1, Cheulhee Jung   1,8, Sharon C. Glotzer4,6,7, David W. Taylor   1,2,3* and Andrew D. Ellington   1,2,3*

Symmetrical protein oligomers are ubiquitous in biological systems and perform key structural and regulatory functions. However, there are few methods for constructing such oligomers. Here we have engineered completely synthetic, symmetrical oligomers by combining pairs of oppositely supercharged variants of a normally monomeric model protein through a strat-egy we term ‘supercharged protein assembly’ (SuPrA). We show that supercharged variants of green fluorescent protein can assemble into a variety of architectures including a well-defined symmetrical 16-mer structure that we solved using cryo-elec-tron microscopy at 3.47 Å resolution. The 16-mer is composed of two stacked rings of octamers, in which the octamers contain supercharged proteins of alternating charges, and interactions within and between the rings are mediated by a variety of spe-cific electrostatic contacts. The ready assembly of this structure suggests that combining oppositely supercharged pairs of protein variants may provide broad opportunities for generating novel architectures via otherwise unprogrammed interactions.

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry204

ArticlesNature Chemistry

properties. Solutions containing individual proteins showed no precipitation (Fig. 1c), and dynamic light scattering (DLS) mea-surements indicated diameters of 5.7 ± 1.7 nm, 6.2 ± 1.9 nm, 5.5 ± 1.4 nm and 5.9 ± 1.9 nm for Ceru+ 32, GFP− 10, GFP− 17 and GFP− 31, consistent with the diameter of GFP monomers (Fig. 1d and Supplementary Fig. 4).

While each GFP variant was individually monomeric, combin-ing oppositely charged variants produced larger particles. Mixing equimolar amounts of Ceru+ 32 and any of the negatively super-charged GFPs in 50 mM NaCl, 50 mM Tris pH 7.4 buffer produced a cloudy solution containing fluorescent particles that settled after brief centrifugation at 1,700g relative centrifugal force (Fig. 1c and Supplementary Fig. 4). As expected, mixing pairs of posi-tively (Ceru+ 32/control protein GFP+ 17) or negatively (GFP− 17/ GFP− 31) supercharged proteins produced no visible precipitation. DLS measurements indicated that Ceru+ 32/GFP− 10 and Ceru+ 32/GFP− 17 particles were micrometre-scale and monodisperse, with average diameters of 1,675 ± 251 nm and 1,421 ± 136 nm, respectively (Fig. 1d and Supplementary Fig. 4). Ceru+ 32/GFP− 31 particles were larger and more polydisperse, with an average diam-eter of 2,452 ± 984 nm (Supplementary Fig. 4). DLS measurements of Ceru+ 32/GFP+ 17 and GFP–31/GFP− 17 indicated particle diameters consistent with monomers, 6.9 ± 3.0 nm and 5.3 ± 1.8 nm, respectively, confirming that these like-charged monomers did not associate (Fig. 1d).

Particle size depends on solution conditions. The observed inter-action of opposite- but not like-charged proteins indicates that assembly is mediated by electrostatic interactions. To further inves-tigate these interactions, we measured the dependence of particle size on ionic strength and solute composition. Notably, combining pairs of oppositely charged proteins at higher ionic strength pro-duces smaller particles distinct from both the micrometre-scale assemblies produced at 50 mM NaCl and monomeric proteins: combining Ceru+ 32 and GFP− 17 at 150 mM NaCl produced par-ticles 11.6 ± 1.8 nm in diameter (Fig. 2a). Systematic measurements of particle size versus ionic strength showed phase-transition-like behaviour, with Ceru+ 32/GFP− 17 particle diameters of ~1,300 nm at NaCl ≤ 50 mM and ~12 nm at 100–300 mM NaCl (Fig. 2a). At NaCl > 300 mM, the Ceru+ 32/GFP− 17 particle diameter decreased slightly to ~9 nm, suggesting that at these high-salt con-centrations a fraction of the ~12 nm particles dissociate into mono-mers or smaller oligomers. Combining Ceru+ 32 and GFP− 17 in KCl, MgSO4 and NaCl/MgCl2 solutions produces similar behaviour, with ~12 nm particles produced at moderate ionic strength and ~1,300 nm particles produced at low ionic strength (Supplementary Fig. 5). Formation of the 12-nm-scale particle from the larger par-ticles is reversible: adding high-salt buffer to increase the NaCl con-centration from 50 mM to 100 mM reduced the Ceru+ 32/GFP− 17 particle diameter from 1,280 ± 327 nm to 12.5 ± 2.2 nm (Fig. 2b). This transition occurred rapidly, with the solution becoming clear and the measured particle diameter decreasing immediately upon NaCl addition. This rapid reversibility suggests that the microme-tre-scale particles assemble via discrete interactions between well-defined, folded structures, as opposed to non-specifically via unfolded, irreversible interactions49. Ceru+ 32/GFP− 10 and Ceru+ 32/GFP− 31 particles show a qualitatively similar salt depen-dence as Ceru+ 32/GFP− 17 (Supplementary Fig. 6). However, when compared to Ceru+ 32/GFP− 17, the transition from particles with ~1,300 nm diameter to ~12 nm occurred at higher NaCl concentra-tions, 150–300 mM and 300–450 mM for Ceru+ 32/GFP− 10 and Ceru+ 32/GFP− 31, respectively.

Ceru+ 32/GFP− 17 particle size also depends on protein stoi-chiometry and pH. To characterize the effects of stoichiometry, we measured the diameter of particles formed at 50 and 150 mM NaCl from 0.1 mg ml−1 of either Ceru+ 32 or GFP− 17 and 0.02, 0.05, 0.01 or 0.2 mg ml−1 of the other. At 50 mM NaCl, DLS measurements indicated ~1,300-nm-diameter particles with excess or equivalent GFP− 17 but ~10- to 12-nm-diameter particles with excess Ceru+ 32 (Supplementary Fig. 7). At 150 mM NaCl, DLS measurements indi-cated ~12-nm-diameter particles at 1:5, 1:2, 1:1 and 2:1 ratios of Ceru+ 32 to GFP− 17 and ~7 nm particles at 5:1 ratios of Ceru+ 32 to

d

Precipitate

bSuperfolder GFP–3

a

c

GFP–17Ceru+32

Long-rangeassembly

Monomer

Oligomers

01

Diameter (nm)

Vol

ume

(%)

Ceru+32GFP–17

1,00010010

10

20

30

Oppositelysupercharged

variants

No precipitate

Cer

u+32

/GF

P+

17

Cer

u+32

/GF

P–1

7

Cer

u+32

GF

P–1

7

GF

P-3

1/G

FP

–17

Ceru+32/GFP+17

Ceru+32/GFP–17

GFP–31/GFP–17

+ –

+

+

+–

–

–

+-

+

+-

+

+

+

-

-

-

+

+

+-

-

-

+

+ -

- +

+

+

-

-

-+

+

+

-

-

-

+

+

+

-

-

-

+

+

+-

-

-

+

+

+-

-

- +

+

+-

-

-

+

+

+-

-

-

+

+

+

-

-

-

+

+

+

-

-

-+

+

+

-

-

-

+

+

+

-

-

-+-

+

+

+-

-

- +

+

+-

-

-

+

+

+

-

-

-

+

+

+-

-

-

+- +-

+

+

+--

-

+

+

+

-

-

-

+

+

+

-

-

-

+-

+

+

+

-

-

-

+

+

+-

-

-

-

+

+

+-

-

-

+-

+

+

+

-

-

-

+-

+

+

+-

-

-

Fig. 1 | Oppositely supercharged Cerulean (Ceru) and GFP variants as a model system for charge-mediated protein assembly. a, We hypothesized that engineered, oppositely supercharged variants of normally monomeric proteins might assemble into well-defined architectures including symmetrical oligomers and, potentially, hierarchically into higher-order, larger structures. b, To investigate this hypothesis, we employed supercharged GFP variants engineered by mutating the surface residues of superfolder GFP to basic (blue) or acidic (yellow) residues. Shown are the positively and negatively charged proteins employed for the majority of experiments described in this study: Ceru+ 32, with a charge of + 32 at pH 7.4 and a tryptophan-containing cerulean chromophore, and GFP− 17, with a charge of − 17 at pH 7.4 (refs. 41,42). Superfolder GFP variants with + 33, + 17, − 10 and − 31 charges at pH 7.4 were also used (Supplementary Fig. 1). c, Each supercharged protein alone is easily soluble and does not produce a precipitate. However, combining 0.1 mg ml−1 of Ceru+ 32 and GFP− 17 at 50 mM NaCl produces visible particles that sink upon light centrifugation. Combining the like-charged proteins Ceru+ 32 and GFP+ 17 or GFP− 31 and GFP− 17 does not produce a precipitate. d, DLS measurements of the Ceru+ 32/GFP− 17 particles formed at 50 mM NaCl indicate that they are micrometre-scale and quite monodisperse, with average diameters of 1,421 ±  136 nm. DLS measurements of Ceru+ 32 and GFP− 17 alone and Ceru+ 32/GFP+ 17 and GFP− 31/GFP− 17 indicate diameters of 5.7 ±  1.7 nm, 5.5 ±  1.4 nm, 6.9 ±  3.0 nm and 5.3 ±  1.8 nm, respectively, consistent with the expected size of monomeric proteins.

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry 205

Articles Nature Chemistry

GFP− 17. Although the DLS measurements did not directly indicate the presence of monomeric proteins even at unequal stoichiome-tries, we suspect that they were probably present: DLS generally pro-vides the ‘major’ component in a solution by extrapolating particles’ diameters from the time correlation of scattering patterns. Because the scattering intensity depends on particle diameter to the sixth power, DLS contains a significant bias towards larger components50, suggesting that our measurements may resolve only the largest major component (that is, the ~12 nm particle) or, potentially, a diameter in between two similarly sized but distinct components (that is, the ~6 nm protein and ~12 nm particle). Decreasing pH dissolved both the ~12 nm and ~1,300 nm particles completely to monomers. At 50 mM NaCl, particle diameter decreased from 1,622 ± 556 nm at pH 7.4 to 5.7 ± 2.4 nm at pH 6.5; at 150 mM NaCl, particle diameter decreased from 11.0 ± 3.1 nm at pH 7.4 to 6.0 ± 0.8 nm at pH 6.5 (Supplementary Fig. 8). Fluorescence measurements confirmed that both proteins remain well-folded at pH 6.5, suggesting that dissolution occurs due to disruption of intermolecular interactions rather than individual proteins unfolding (Supplementary Fig. 9). Theoretical calculations predict that decreasing the pH from 7.4 to 6.5 to 6.0 increases the surface charge of Ceru+ 32 from + 32 to + 39 to + 43 and the surface charge of GFP− 17 from − 17 to − 9 to − 4. Experimentally, decreasing the pH from 7.4 to 6.0 increases the zeta potential of Ceru+ 32 from 8.9 ± 0.2 mV to 12.3 ± 0.1 mV and the zeta potential of GFP− 17 from − 15.7 ± 0.4 mV to − 12.0 ± 0.5 mV, supporting these predictions (Supplementary Fig. 9). Taken together, the observations that mixtures of Ceru+ 32 and GFP− 17 assemble into three distinct phases (~6 nm (monomeric), ~12 nm and ~1,300 nm diameter particles) at a broad range of salt, stoichio-metric and pH conditions suggest that the behaviour is governed by discrete, specific intermolecular interactions. We term this strategy of driving defined protein interactions by combining engineered supercharged variants as supercharged protein assembly (SuPrA).

FRET between oppositely supercharged fluorescent proteins depends on solution conditions. To further investigate the compo-sition and electronic structure of the SuPrA particles, we measured FRET between Ceru+ 32 and GFP− 17. Specifically, we measured the ratio of the fluorescent output of GFP− 17 at its emission maximum (515–535 nm) in response to direct excitation at its own excitation maximum of 485 nm to that in response to excitation of Ceru+ 32, at Ceru+ 32’s excitation maximum of 433 nm, defining this quan-tity as the ‘excitation ratio’. Through FRET, excited cerulean ‘donor’ proteins transfer energy to GFP ‘acceptor’ proteins located within a ~10 nm distance, producing GFP fluorescence51. A mixture of non-interacting Ceru+ 32/GFP+ 17 produced an excitation ratio

of 0.19 ± 0.01, identical to the 0.19 ± 0.01 excitation ratio mea-sured in the absense of Ceru+ 32 (Supplementary Fig. 10). Ceru+ 32/GFP− 17 particles at 50 and 150 mM NaCl, however, produced substantially higher excitation ratios of 0.37 ± 0.01 and 0.30 ± 0.01, respectively. These results indicate that Ceru+ 32 and GFP− 17 are spatially close in both ~1,300-nm and ~12-nm-diameter particles.

The excitation ratios of Ceru+ 32/GFP− 17 SuPrA particles showed a qualitatively similar though more graded depen-dence on NaCl concentration than did particle size measured by DLS. At NaCl ≤ 50 mM, the excitation ratios remained at ~0.37 (Supplementary Fig. 11). Increasing the NaCl concentration pro-gressively decreased the excitation ratio until ~450 mM, where the excitation ratios levelled off at ~0.22. Control measurements indi-cated that NaCl concentration changes affect neither Ceru+ 32 nor GFP− 17 fluorescence (Supplementary Fig. 12). This graded transi-tion suggests that increasing NaCl produces subtle shifts in equilib-ria or geometries, which DLS misses. This is not surprising—while DLS generally provides the ‘major’ component in a solution via the time correlation of scattering patterns50, FRET provides the ‘average’ component via total energy transfer from donor to acceptor fluo-rophores51. Taken together, our DLS and FRET data are consistent with subtle but significant shifts in the spectral overlap or packing geometries within SuPrA particles of a given size, or with changes in equilibria between differently sized SuPrA particles. Bulk FRET of Ceru+ 32/GFP− 10 and Ceru+ 32/GFP− 31 particles also showed decreasing excitation ratios with increasing NaCl concentration (Supplementary Fig. 13).

Negative stain electron microscopy indicates a monodisperse, globular, oligomer structure. To investigate the specific archi-tecture of the ~12 nm Ceru+ 32/GFP− 17 particle, we first used negative stain electron microscopy. Raw micrographs of negatively stained Ceru+ 32/GFP− 17 particles formed at 150 mM NaCl indi-cated monodisperse, globular protomers ~160 Å in diameter con-taining internal features (Supplementary Fig. 14). Intriguingly, these particles appeared to show eight-fold symmetry after reference-free two-dimensional (2D) alignment and classification. Using 3D reconstruction techniques, we obtained a structure of this complex at ~18 Å resolution that clearly showed eight globular densities con-sistent with the size and shape of the parent protein of Ceru+ 32 and GFP− 17, superfolder GFP, arranged into a single ring. Each pro-tomer contains two of these symmetric rings stacked on top of one another and offset by about half a subunit, ultimately forming a bar-rel. This structure is roughly equivalent in volume to the ~12-nm-diameter sphere measured by DLS, strongly suggesting that the protomer observed by DLS and electron microscopy are the same.

0

10

2050 mM,Ceru+32

50 mM,GFP–17

0 mM

300 mM

950 mM

10 mM

50 mM

100 mM

450 mM

150 mM

0

10

20

30

Vol

ume

(%)

Ceru+32/GFP–17mixed at 50mM NaCl

Sucessive NaCladdition

[NaCl] dependence Reversibilityba

100 mM

150 mM

1

Diameter (nm)

1,00010010 10,000

300 mM

450 mM

1

Diameter (nm)

1,00010010 10,000

Vol

ume

(%)

Fig. 2 | Ceru+32/GFP−17 particle size depends on NaCl concentration. a, Ceru+ 32/GFP− 17 particle size depends on NaCl concentration in a phase-transition-like manner. At pH 7.4, combining 0.1 mg ml−1 of both Ceru+ 32 and GFP− 17 produces particles ~1,300 nm in diameter at NaCl concentrations at or below 50 mM, particles ~12 nm in diameter at NaCl concentrations at 100–300 mM NaCl and particles ~9–11 nm in diameter at NaCl concentrations at and above 450 mM. At all NaCl concentrations measured, Ceru+ 32/GFP− 17 particle size remains significantly higher than that of single proteins. b, Approximately 1,300-nm-scale particles form reversibly: increasing the NaCl concentration from 50 mM to 100 mM rapidly reduces the particle diameter from 1,280 ±  327 nm to 12.5 nm ±  2.2 nm.

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry206

ArticlesNature Chemistry

To further investigate the structural and molecular interactions underlying protomer formation we employed shape- and charge-based Monte Carlo modelling. Although the negative stain data indicated a specific arrangement of globular densities, the resolu-tion was not sufficient to determine their precise orientations or structural features differentiating Ceru+ 32 from GFP− 17. We hypothesized that we could determine the configuration of Ceru+ 32 and GFP− 17 within the protomer by simulating different test configurations; the true configuration(s) would be expected to be the most stable. We identified a set of test configurations that fit two assumptions. The first was that the protomer contained eight Ceru+ 32 and eight GFP− 17 proteins organized in an alternating pattern to minimize opposite-charge repulsion and maximize like-charge attraction. The second was that, given that the vast major-ity of natural homomeric protein complexes are symmetrical3–7,49, the synthetic protomer likewise contained a symmetric arrange-ment of Ceru+ 32 and GFP− 17 proteins. From these, we identified four distinct candidate configurations differing in the orientations of the Ceru+ 32 and GFP− 17 proteins, with the N-terminal bar-rel tops oriented either towards the outside or the inside of the protomer interface (Supplementary Fig. 15). Each candidate con-figuration was modelled by approximating Ceru+ 32 and GFP− 17 proteins as hard, non-convex polyhedra representing their solvent-excluded surface with point charges placed at each atomic site. The stability of each was assessed over a range of NaCl concentrations using the hard particle Monte Carlo (HPMC) plugin to HOOMD-blue52–54. The total energy at each step was calculated as the sum of the hard particle and screened electrostatic contributions. After 400,000 sweeps, protomer stability was assessed quantitatively via the average radius of gyration Rg between the proteins compris-ing the protomer (that is, the root-mean-squared molecule-to- molecule distance). Negative stain data indicated an Rg of ~6.5 nm in the intact protomer; we defined Rg > 13 nm as a clear threshold for dissociation.

Qualitatively, these simulations produced the expected behav-iour: endpoint Rg increased with increasing NaCl concentration (Supplementary Fig. 15). However, while each candidate con-figuration remained intact at low NaCl (< 38 mM), all dissociated at NaCl > 50 mM, in contrast to the experimental observations in

which the protomers remained intact at NaCl concentrations well above 150 mM. Moreover, the different candidates exhibited nearly identical stability ranges; none showed the increased stability range expected for the ‘true’ configuration. This discrepancy sug-gested that the protomer geometry differed from our hypothesized symmetrical arrangements.

Atomic model of protomer and its interfaces revealed by cryo-electron microscopy. To unambiguously characterize the protomer structure and the molecular basis for its assembly, we performed cryo-electron microscopy (cryo-EM) and obtained a 3.47 Å density map (using the 0.143 Fourier shell correlation criterion). The high quality of the reconstruction allowed us to build an atomic model of the protomer (Fig. 3 and Supplementary Figs. 16–18). Consistent with the negative stain electron microscopy reconstruction, the cryo-EM map contains two stacked rings, each composed of eight monomers with the obvious beta barrel architecture expected for a GFP. However, in contrast to the structure inferred by negative staining, in which each ring contained eight-fold symmetry, the higher-resolution cryo-EM structure clearly shows D4 symme-try, with two distinct orientations for each monomer (Fig. 3 and Supplementary Fig. 17). By examining the side chains of selected amino acids, we could easily assign each monomeric density to either Ceru+ 32 or GFP− 17 (Supplementary Fig. 18). Interestingly, on each ring there are four Ceru+ 32 monomers in the plane of the ring, pointing towards the inner channel, and four GFP− 17 mono-mers nearly parallel to the edge of the octameric ring but tilted ~30° with respect to the Ceru+ 32 monomers. These two types of monomer are arranged in an alternating, repeating pattern on both rings. The two rings are mirror-image reflections offset by ~22.5°, with each Ceru+ 32 monomer contacting a GFP− 17 monomer on the opposite ring (Fig. 4).

Introduced charged residues play a key role in protomer inter-faces. Analysis by PDBePISA55 indicated four distinct protein–pro-tein interfaces within the protomer, each containing electrostatic interactions involving introduced charged residues (Fig. 4, Table 1 and Supplementary Figs. 19 and 20). Within each eight-protein ring, Ceru+ 32 and GFP− 17 interact at two distinct interfaces, which we

GFP–17

Ceru+32a

45°90°

90° 45°

b

Fig. 3 | Cryo-EM structure of the Ceru+32/GFP−17 protomer. a, Cryo-EM reconstruction of the Ceru+ 32/GFP− 17 protomer at 3.47 Å resolution in three different orientations. Ceru+ 32 and GFP− 17 monomers are coloured blue and yellow, respectively. b, Atomic model of the Ceru+ 32/GFP− 17 protomer in the three orientations corresponding to a. Only the top ring is shown on the left panel, with the atomic structure in ribbon docked into the semi-transparent cryo-EM density.

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry 207

Articles Nature Chemistry

have termed the ‘Ceru+ 32 clockwise’ interface (where Ceru+ 32 is clockwise of GFP− 17 when viewed from above) and the ‘GFP− 17 clockwise’ interface (where GFP− 17 is clockwise of Ceru+ 32 when viewed from above) (Fig. 4a). The Ceru+ 32 clockwise interface (Fig. 4b) contains nine contacts between the side of the Ceru+ 32 beta barrel and the top of the GFP− 17 beta barrel, seven of which are between an introduced charged residue on one protein and a native residue on the other. The introduced charged residues Ceru+ 32 R205 and K150 form hydrogen bonds, a salt bridge and a hybrid hydrogen bond/salt bridge with a stretch of native, N-terminal GFP− 17 residues (E6, E7, L8, T10 and V12). Additionally, the introduced charged residues Ceru+ 32 K231 and GFP− 17 D39 form hybrid hydrogen bond/salt bridge interactions with native resi-dues GFP− 17 D77 and Ceru+ 32 R74, respectively. The GFP− 17 clockwise interface (Fig. 4c) contains six contacts between the sides of the Ceru+ 32 and GFP− 17 beta barrels, three of which are between an introduced charged and a native residue. Notably, the

introduced charged residue Ceru+ 32 R191 flips out of its origi-nal orientation to hydrogen bond with the native GFP− 17 residue Q205 (Supplementary Fig. 21). Additionally, the introduced residue Ceru+ 32 K125 forms a salt bridge and hybrid interaction between the native residues GFP− 17 E143 and E173. Between opposite eight-protein rings, Ceru+ 32 and GFP–17 interact at a single inter-face that we have termed the ‘inter-ring’ interface (Fig. 4d). This interface contains six contacts between the side of a Ceru+ 32 in one ring and the top of a tilted GFP− 17. Two of these are between the introduced residue Ceru+ 32 K165 and the native residue GFP− 17 D198, whose side chains form a salt bridge and a hybrid interaction. Three additional contacts form between an introduced charged residue and a native residue: introduced charged residues Ceru+ 32 K165 and GFP− 17 E158 and D194 form a hydrogen bond, a hybrid interaction and a salt bridge with the native residues GFP− 17 L196 and Ceru+ 32 K108 and K167, respectively. Interestingly, neighbouring GFP− 17 proteins in opposite rings also interact at a

GFP–17

Ceru+32

Ceru+32 clockwise GFP–17 clockwise Inter–GFP–17Inter-ring

K165

K167 K108Ceru+32

GFP–17

E158

Y183D194

L196D198

GFP–17

GFP–17

E165

E165

K167

K167

b dc e

a

45°90°

R205R74

E6D39

K150

V12

E7T10

L8

Ceru+32

GFP–17

Y201

D77

G229

K231 R191

R110K125

G175

Ceru+32

GFP–17

Q205 N171

E143 E173

S176

Fig. 4 | inter-protein interactions in the Ceru+32/GFP−17 protomer. a, Surface representation of the atomic model, shown in three orientations corresponding to those in Fig. 3. Interfacial regions (within 7 Å of residues participating in inter-monomer interactions, as detected by PDBePISA55) are shaded red, orange, blue and green, respectively. b–e, Details of the four interfaces between Ceru+ 32 and GFP− 17 in the protomer: ‘Ceru+ 32 clockwise’ interface (b); ‘GFP− 17 clockwise’ interface (c); ‘inter-ring’ interface (d); ‘inter-GFP− 17′ interface (e). Dotted lines correspond to hydrogen bonds. Ceru+ 32 and GFP− 17 monomers are coloured light blue and gold, respectively. Underlined residues are the charged residues introduced to engineer the supercharged variants.

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry208

ArticlesNature Chemistry

small, symmetric interface that we have termed the ‘inter-GFP− 17’ interface. This interface contains two salt bridges between the intro-duced E165 and native K167 residues on the tilted beta barrel of each GFP− 17 (Fig. 4e). In total, the protomer interfaces contain 176 inter-protein electrostatic interactions, 128 of which (73%) contain at least one introduced charged residue. However, while many charged residues play key roles in protomer interface forma-tion, 11 of the total 21 introduced charged Ceru+ 32 residues and six of the 11 introduced charged GFP− 17 residues are not involved in interface formation. This suggests that the relatively large num-ber of introduced charged residues provide supercharged protein pairs with many potential sites for interaction, allowing them to ‘find’ interaction geometries supporting cooperative assembly into symmetrical structures.

The protomer interfaces appear to be mediated entirely by the electrostatic contacts described above. Visual inspection indicates loose packing of non-electrostatically interacting residues near each interface, and PDBePISA predicts moderately unfavourable solvation energies (0.3–3.3 kcal mol−1 per interface) and complex formation significance scores of 0.000 for each interface, indicat-ing that hydrophobic packing plays no role in protomer interface formation. Similarly, the C-terminal His6-tag employed for protein purification also appears to play no significant role in protomer for-mation. The C-terminal Ceru+ 32/GFP− 17 residues, including the His6-tag employed for purification, are oriented towards the centre of the protomer and away from the inter-protein interfaces. While the density corresponding to the His6-tag itself is unresolved, these observations indicate that the His6-tag probably does not contrib-ute to the intermolecular interactions driving protomer formation (Supplementary Fig. 22). Notably, while some GFP variants, includ-ing the parent protein originally found in Aeqorea victoria, dimerize

near their native A206 residue (mutated to V to render each variant used in this work monomeric40,41), this residue and the surrounding interfaces are completely uninvolved in the structure, indicating a novel architecture56,57.

Protomer size is consistent with that measured by DLS, sug-gesting high yield. The size of the protomer structure indicated by the cryo-EM structure is consistent with that measured by DLS at 100–300 mM NaCl, indicating that it forms with high yield. To further estimate the fraction of proteins that assembled into pro-tomers, we calculated the relative proportions of monomer and protomer particles (approximated as ~6 nm and ~12.5 nm spheres) expected to produce the average diameter of 11.6 nm measured at 150 mM NaCl for Ceru+ 32/GFP− 17 particles. This calculation pre-dicted that ~87% of the particles in solution are protomers, which corresponds to ~99% of all Ceru+ 32 and GFP− 17 proteins in the protomer phase. While this prediction may be an overestimate due to the bias of DLS toward the ‘major’, larger components in hetero-geneous solutions50, the agreement between DLS measurements and cryo-EM data indicates high yield.

Oligomer formation is generalizable to other pairs of super-charged fluorescent protein variants. Other pairs of oppositely charged GFP variants form ~12 nm particles at certain NaCl concentrations, suggesting that they also form protomer-like oligo-meric structures (Supplementary Fig. 6). To investigate the extent to which modest changes in sequence affect the ability to form protomers we analysed the structure of the particle formed by mix-ing an alternative positively charged GFP variant, GFP+ 33a, with GFP− 1742. GFP+ 33a contains a largely different set of introduced positively charged residues relative to Ceru+ 32, with differences

Table 1 | interactions between Ceru+32 and GFP−17 within the four types of interfaces in the Ceru+32/GFP−17 protomer. Charged residues introduced to generate the supercharged Ceru+32a and GFP−17b are shown.

interface Acceptor Donor Type Distance (Å)

Ceru+ 32 CW Ceru+32, ARG 205[NH1]a GFP− 17, LEU 8[O] Hydrogen bond 3.25

Ceru+ 32 CW Ceru+32, ARG 205[NH1]a GFP− 17, THR 10[O] Hydrogen bond 2.45

Ceru+ 32 CW Ceru+32, ARG 205[NH2]a GFP− 17, VAL 12[O] Hydrogen bond 3.51

Ceru+ 32 CW Ceru+ 32, GLY 229[N] GFP− 17, ASP 77[OD1] Hydrogen bond 3.56

Ceru+ 32 CW Ceru+ 32, TYR 201[OH] GFP− 17, ASP 77[OD1] Hydrogen bond 3.80

Ceru+ 32 CW Ceru+32, LYS 150[NZ]a GFP− 17, GLU 7[OE2] Salt bridge 3.95

Ceru+ 32 CW Ceru+32, LYS 150[NZ]a GFP− 17, GLU 6[OE2] Hybrid 3.51

Ceru+ 32 CW Ceru+ 32, ARG 74[NH1] GFP− 17, ASP 39[OD2]b Hybrid 3.59

Ceru+ 32 CW Ceru+32, LYS 231[NZ]a GFP− 17, ASP 77[OD2] Hybrid 3.05

GFP− 17 CW Ceru+ 32, ARG 110[NH1] GFP− 17, SER 176[O] Hydrogen bond 3.14

GFP− 17 CW Ceru+ 32, ARG 110[NH1] GFP− 17, GLY 175[O] Hydrogen bond 3.48

GFP− 17 CW Ceru+ 32, ARG 110[NH2] GFP− 17, ASN 171[OD1] Hydrogen bond 2.92

GFP− 17 CW Ceru+32, ARG 191[NE]a GFP− 17, GLN 205[OE1] Hydrogen bond 2.92

GFP− 17 CW Ceru+32, LYS 125[NZ]a GFP− 17, GLU 173[OE1] Salt bridge 3.92

GFP− 17 CW Ceru+32, LYS 125[NZ]a GFP− 17, GLU 143[OE2] Hybrid 2.45

Inter-ring Ceru+ 32, LYS 167[NZ] GFP− 17, ASP 194[O]b Hydrogen bond 3.85

Inter-ring Ceru+32, LYS 165[NZ]a GFP− 17, LEU 196[O] Hydrogen bond 3.26

Inter-ring Ceru+ 32, TYR 183[OH] GFP− 17, LEU 196[N] Hydrogen bond 2.84

Inter-ring Ceru+32, LYS 165[NZ]a GFP− 17, ASP 198[OD2] Salt bridge 3.66

Inter-ring Ceru+ 32, LYS 108[NZ] GFP− 17, GLU 158[OE1]b Hybrid 3.83

Inter-ring Ceru+32, LYS 165[NZ]a GFP− 17, ASP 198[OD1] Hybrid 2.50

Inter-GFP− 17 GFP− 17, LYS 167[NZ] GFP− 17, GLU 165[OE2]b Salt bridge 3.99

Inter-GFP− 17 GFP− 17, LYS 167[NZ] GFP− 17, GLU 165[OE2]b Salt bridge 3.97

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry 209

Articles Nature Chemistry

at 32 residues including three—R191D, R205K and D231K—that contact GFP− 17 residues in protomer interfaces (Supplementary Fig. 19). Negative stain images indicate ring-like oligomeric struc-tures similar to those of the Ceru+ 32/GFP− 17 pairs, suggesting a broad ability of this class of oppositely charged protein variants to assemble into defined oligomers (Supplementary Fig. 23).

Computational discrimination of cryo-EM-derived protomer structure. Monte Carlo simulations of the protomer structure obtained by cryo-EM validated the ability of our computational

model to discriminate between stable and unstable configurations. Specifically, we used the cryo-EM data to construct and simulate protomers with the appropriate protein arrangements (Fig. 5a). The resulting simulations exhibit stability over a much wider range of salt concentrations (Fig. 5b): the protomer with the correct arrange-ment of Ceru+ 32 and GFP− 17 proteins does not denature until a NaCl concentration of ~80 mM. However, this calculation still underestimates the true stability range, suggesting that a truly accu-rate model would require more explicit modelling of complex inter-actions than the simple shape- and charge-based methods applied. Nonetheless, the discriminatory power of the model indicates its potential for quickly answering questions about how particular mutations would affect protomer stability.

Confocal microscopy indicates amorphous, fractal-like structure of ~1,300 nm particles. To explore the ultrastructural features of ~1,300 nm Ceru+ 32/GFP− 17 particles, we imaged the particles with confocal microscopy. The resulting images indicated mod-erately loose, fractal-like amorphous particles generally ~1–10 µ m across (Fig. 6). The amorphous, fractal-like appearance of these particles suggests that these particles formed from the ran-dom association of particles, presumably of Ceru+ 32 and GFP− 17 monomers, protomers or other types of oligomeric particles in low salt regimes58. Fluorescent images taken in the Ceru+ 32 channel, GFP− 17 channel and FRET channel (that is, Ceru+ 32 excita-tion but GFP− 17 detection) are nearly identical, suggesting that Ceru+ 32 and GFP− 17 are evenly distributed throughout the larger aggregates (Supplementary Fig. 24).

DiscussionProtein oligomers perform key functions in natural biological sys-tems, but engineering synthetic oligomers for biotechnological applications has remained challenging. While previous studies have shown that it is possible to design synthetic protein oligomers from a ‘bottom-up’ approach by either modifying extant structures9–13 or through the de novo design of complementary interfaces14–18, we now show that it may be possible to induce the formation of

40 60 80 100 120 140

5

10

15

20

25

Rad

ius

of g

yrat

ion,

Rg

(nm

)

NaCl concentration (mM)

0

Best negative-staininformed model

Cryo-EM-informedmodel

Rg = 6 nm

Rg = 13 nm

a bCryo-EM informedprotomer model

Protein orientations

Fig. 5 | Computational simulations of protomer structure stability. a, Simulation of the Ceru+ 32/GFP− 17 protomer with protein arrangements based on the cryo-EM data. Arrows indicate the Ceru+ 32/GFP− 17 orientation within the protomer. Circles with crosses indicate termini pointing into the page, and circles with dots are pointing out of the page. b, Computationally predicted average radius of gyration versus NaCl concentration of the best negative stain informed protomer model (green triangles) and cryo-EM-informed model (blue circles). Data points and error bars are computed as the mean and s.d. of the radii of gyration of five simulations conducted at each value of NaCl concentration. While the models informed solely from the negative stain electron microscopy denature below 50 mM NaCl, the structure inferred from the cryo-EM data remains stable up to 100 mM. While this still underestimates the experimentally observed stability region, it demonstrates that the simulations successfully distinguish between the actual and other unstable configurations.

20 μm 5 μm

a bCeru+32/GFP–17composite

Closer view

Fig. 6 | Confocal images of micrometre-scale Ceru+32/GFP−17 particles. a, Confocal microscopy images of Ceru+ 32/GFP− 17 particles at 50 mM NaCl indicate patchy, amorphous particles on the order of ~1–10 μ m across. This amorphous, fractal-like geometry is consistent with a random association of oppositely charged proteins, particles or patches on smaller particles58. The image shown is an overlay of the Ceru+ 32 and GFP− 17 channels. Supplementary Fig. 24 shows each channel individually. b, Close-up view of boxed area in a, indicating patchy, fractal-like geometry on the smaller size scale, with punctate features on the order of ~200–1,000 nm in diameter.

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry210

ArticlesNature Chemistry

novel quaternary structures and higher order architectures through the simple process of supercharging. Unlike previously devel-oped approaches, the method we demonstrate here, SuPrA, does not require specific expectations for or knowledge of the eventual oligomeric interfaces. Instead, the wide range of charged resi-dues introduced into the protein variants provide a broad range of physicochemically and geometrically complementary interpro-tein interfaces, ultimately allowing the cooperative assembly of defined, symmetric oligomers. Given that both natural proteins3–8 and synthetic colloids19–30 have been shown to readily assemble into symmetric, higher-order structures, we anticipate that our SuPrA strategy will be generalizable, creating synthetic, scalable architec-tures with other proteins, especially since supercharging has proven possible across a broad variety of proteins38,39,59–61.

Code availabilitySource code for HOOMD-blue is available at http://glotzerlab.engin.umich.edu/hoomd-blue/ and at https://bitbucket.org/glotzer/hoomd-blue/. Specific source codes are available upon request.

Data availabilityThe data generated and analysed in this study, including sequence verification files and the data associated with all figures, are available from the corresponding authors upon reasonable request. The cryo-EM density map of the Ceru+ 32/GFP− 17 protomer has been deposited in the EMDB under accession code EMD-9104. The corresponding atomic model has been deposited in the PDB under accession code 6MDR.

Received: 25 September 2018; Accepted: 26 November 2018; Published online: 14 January 2019

References 1. Whitesides, G. M. & Grzybowski, B. Self-assembly at all scales. Science 295,

2418–2421 (2002). 2. Blundell, T. L. & Srinivasan, N. Symmetry, stability, and dynamics of

multidomain and multicomponent protein systems. Proc. Natl Acad. Sci. USA 93, 14243–14248 (1996).

3. Levy, E. D., Boeri Erba, E., Robinson, C. V. & Teichmann, S. A. Assembly reflects evolution of protein complexes. Nature 453, 1262–1265 (2008).

4. Goodsell, D. S. & Olson, A. J. Structural symmetry and protein function. Annu. Rev. Biophys. Biomol. Struct. 29, 105–153 (2000).

5. André, I., Strauss, C. E., Kaplan, D. B., Bradley, P. & Baker, D. Emergence of symmetry in homooligomeric biological assemblies. Proc. Natl Acad. Sci. USA 105, 16148–16152 (2008).

6. Plaxco, K. W. & Gross, M. Protein complexes: the evolution of symmetry. Curr. Biol. 19, R25–R26 (2009).

7. Bergendahl, L. T. & Marsh, J. H. Functional determinants of protein assembly into homomeric complexes. Sci. Rep. 7, 4932 (2017).

8. Padilla, J. E., Colovos, C. & Yeates, T. O. Nanohedra: using symmetry to design self assembling protein cages, layers, crystals and filaments. Proc. Natl Acad. Sci. USA 98, 2217–2221 (2001).

9. Lai, Y.-T., Cascio, D. & Yeates, T. O. Structure of a 16-nm cage designed by using protein oligomers. Science 336, 1129 (2012).

10. Alberstein, R., Suzuki, Y., Paesani, F. & Tezcan, F. A. Engineering the entropy-driven free-energy landscape of a dynamic nanoporous protein assembly. Nat. Chem. 10, 732–739 (2018).

11. Badieyan, S. et al. Symmetry‐directed self‐assembly of a tetrahedral protein cage mediated by de novo‐designed coiled coils. Chembiochem 18, 1888–1892 (2017).

12. Brodin, J. D., Carr, J. R., Sontz, P. A. & Tezcan, F. A. Exceptionally stable, redox-active supramolecular protein assemblies with emergent properties. Proc. Natl Acad. Sci. USA 111, 2897–2902 (2014).

13. Yeates, T. O. Geometric principles for designing highly symmetric self-assembling protein nanomaterials. Annu. Rev. Biophys. 46, 23–42 (2017).

14. Boyken, S. E. et al. De novo design of protein homo-oligomers with modular hydrogen-bond network-mediated specificity. Science 352, 680–687 (2016).

15. Bale, J. B. et al. Accurate design of megadalton-scale two-component icosahedral protein complexes. Science 353, 389–394 (2016).

16. Hsia, Y. et al. Design of a hyperstable 60-subunit protein dodecahedron. Nature 535, 136–139 (2016).

17. Butterfield, G. L. et al. Evolution of a designed protein assembly encapsulating its own RNA genome. Nature 552, 415–420 (2017).

18. Kobayashi, N. & Arai, R. Design and construction of self-assembling supramolecular protein complexes using artificial and fusion proteins as nanoscale building blocks. Curr. Opin. Biotechnol. 46, 57–65 (2017).

19. Damasceno, P. F., Engel, M. & Glotzer, S. C. Predictive self-assembly of polyhedra into complex structures. Science 337, 453–457 (2012).

20. Paik, T. & Murray, C. B. Shape-directed binary assembly of anisotropic nanoplates: a nanocrystal puzzle with shape-complementary building blocks. Nano Lett. 13, 2952–2956 (2013).

21. Gong, J. et al. Shape-dependent ordering of gold nanocrystals into large-scale superlattices. Nat. Commun. 8, 14038 (2017).

22. Wolters, J. R. et al. Self-assembly of ‘Mickey Mouse’ shaped colloids into tube- like structures: experiments and simulations. Soft Matter 11, 1067–1077 (2015).

23. Boles, M. A., Engel, M. & Talapin, D. V. Self-assembly of colloidal nanocrystals: from intricate structures to functional materials. Chem. Rev. 116, 11220–11289 (2016).

24. Fu, L. et al. Assembly of hard spheres in a cylinder: a computational and experimental study. Soft Matter 13, 3296–3306 (2017).

25. Ye, X. et al. Competition of shape and interaction patchiness for self-assembling nanoplates. Nat. Chem. 5, 466–473 (2013).

26. Zhang, Z. & Glotzer, S. C. Self-assembly of patchy particles. Nano Lett. 4, 1407–1413 (2004).

27. Giacometti, A., Lado, F., Largo, J., Pastore, G. & Sciortino, F. Effects of patch size and number within a simple model of patchy colloids. J. Chem. Phys. 132, 174110 (2010).

28. Pawar, A. B. & Kretzschmar, I. Fabrication, assembly and application of patchy particles. Macromol. Rapid Commun. 31, 150–168 (2010).

29. Gong, Z., Hueckel, T., Yi, G. R. & Sacanna, S. Patchy particles made by colloidal fusion. Nature 550, 234–238 (2017).

30. Duguet, E., Hubert, C., Chomette, C., Perroc, A. & Ravaine, S. Patchy colloidal particles for programmed self-assembly. C. R. Chimi. 19, 173–182 (2016).

31. Woo, S. & Rothemund, P. W. Programmable molecular recognition based on the geometry of DNA nanostructures. Nat. Chem. 3, 620–627 (2011).

32. Gerling, T., Wagenbauer, K. F., Neunerm, A. M. & Dietz, H. Dynamic DNA devices and assemblies formed by shape-complementary, non-base pairing 3D components. Science 347, 1446–1452 (2015).

33. Sheinerman, F. B., Norel, R. & Honig, B. Electrostatic aspects of protein–protein interactions. Curr. Opin. Struct. Biol. 10, 153–159 (2000).

34. Liljeström, V., Mikkilä, J. & Kostiainen, M. A. Self-assembly and modular functionalization of three-dimensional crystals from oppositely charged proteins. Nat. Commun. 5, 4445 (2014).

35. Kostiainen, M. A. et al. Electrostatic assembly of binary nanoparticle superlattices using protein cages. Nat. Nanotech. 8, 52–56 (2012).

36. Seebeck, F. P., Woycechowsky, K. J., Zhuang, W., Rabe, J. P. & Hilvert, D. A simple tagging system for protein encapsulation. J. Am. Chem. Soc. 128, 4516–4517 (2006).

37. Wörsdörfer, B., Woycechowsky, K. J. & Hilvert, D. Directed evolution of a protein container. Science 331, 589–592 (2011).

38. Held, M. et al. Engineering formation of multiple recombinant Eut protein nanocompartments in E. coli. Sci. Rep. 6, 24359 (2016).

39. Beck, T., Tetter, S., Künzle, M. & Hilvert, D. Construction of Matryoshka-type structures from supercharged protein nanocages. Angew. Chem. Int. Ed. 54, 937–940 (2014).

40. Sun, H., Luo, Q., Hou, C. & Liu, J. Nanostructures based on protein self-assembly: from hierarchical construction to bioinspired materials. Nanotoday 14, 16–41 (2017).

41. Lawrence, M. S., Phillips, K. J. & Liu, D. R. Supercharging proteins can impart unusual resilience. J. Am. Chem. Soc. 129, 10110–10112 (2007).

42. Der, B. S. et al. Alternative computational protocols for supercharging protein surfaces for reversible unfolding and retention of stability. PLoS One 31, e64363 (2013).

43. Oh, H. J., Gather, M. C., Song, J.-J. & Yun, S. H. Lasing from fluorescent protein crystals. Opt. Express 22, 31411–31416 (2014).

44. Ormö, M. et al. Crystal structure of the Aequorea victoria green fluorescent protein. Science 273, 1392–1395 (1996).

45. Bolhuis, P. & Frenkel, D. Tracing the phase boundaries of hard spherocylinders. J. Chem. Phys. 106, 666–687 (1997).

46. Rizzo, M. A., Springer, G. H., Granada, B. & Piston, D. W. An improved cyan fluorescent protein variant useful for FRET. Nat. Biotechnol. 4, 445–449 (2004).

47. Goedhart, J. et al. Bright cyan fluorescent protein variants identified by fluorescence lifetime screening. Nat. Methods 7, 137–139 (2010).

48. Goedhart, J. et al. Structure-guided evolution of cyan fluorescent proteins towards a quantum yield of 93%. Nat. Commun. 3, 751 (2012).

49. García-Seisdedos, H., Empereur-Mot, C., Elad, N. & Levy, E. D. Proteins evolve on the edge of supramolecular self-assembly. Nature 548, 244–247 (2017).

50. Hassan, P. A., Rana, S. & Verma, G. Making sense of Brownian motion: colloid characterization by dynamic light scattering. Langmuir 31, 3–12 (2015).

51. Bajar, B. T., Wang, E. S., Zhang, S., Lin, M. Z. & Chu, J. A guide to fluorescent protein FRET pairs. Sensors 16, E1488 (2016).

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry 211

Articles Nature Chemistry

52. Anderson, J. A., Lorenz, C. D. & Travesset, A. General purpose molecular dynamics simulations fully implemented on graphics processing units. J. Comp. Phys. 227, 5342–5359 (2008).

53. Glaser, J. et al. Strong scaling of general-purpose molecular dynamics simulations on GPUs. Comp. Phys. Commun. 192, 97–107 (2015).

54. Sinkovits, D. W., Barr, S. A. & Luijten, E. Rejection-free Monte Carlo scheme for anisotropic particles. J. Chem. Phys. 136, 144111 (2012).

55. Krissinel, E. & Henrick, K. Inference of macromolecular assemblies from crystalline state. J. Mol. Biol. 372, 774–797 (2007).

56. Pédelacq, J. D., Cabantous, S., Tran, T., Terwilliger, T. C. & Waldo, G. S. Engineering and characterization of a superfolder green fluorescent protein. Nat. Biotechnol. 24, 79–88 (2006).

57. Yang, F., Moss, L. G. & Phillips, G. B. Jr. The molecular structure of green fluorescent protein. Nat. Biotechnol. 14, 1246–1251 (1996).

58. Lin, M. Y. et al. Universal diffusion-limited colloid aggregation. J. Phys. Condens. Matter 2, 3039–3113 (1990).

59. Miklos, A. E. et al. Structure-based design of supercharged, highly thermoresistant antibodies. Chem. Biol. 19, 449–455 (2012).

60. Johnson, L. B., Park, S., Gintner, L. P. & Snow, C. D. Characterization of supercharged cellulase activity and stability in ionic liquids. J. Mol. Catal. B 132, 84–90 (2016).

61. Zuris, J. A. et al. Cationic lipid-mediated delivery of proteins enables efficient protein-based genome editing in vitro and in vivo. Nat. Biotechnol. 33, 73–80 (2015).

AcknowledgementsThis material is based on work supported by the US Army Research Laboratory and the US Army Research Office under grant no. W911NF-1–51–0120 to the University of Texas at Austin and under grant no. W911NF-15–1–0185 to the University of Michigan. Computational resources and services for simulation work were supported by Advanced Research Computing at the University of Michigan, Ann Arbor. This work used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation grant no. ACI-1053575 (XSEDE award DMR 140129). A.J.S. is supported by an Arnold O. Beckman Postdoctoral Fellowship. D.W.T. is a CPRIT Scholar supported by the Cancer Prevention and Research Institute of Texas (RR160088).

This work was supported in part by a Welch Foundation grant F-1938 (to D.W.T.). The authors thank N. Wang for help with building the atomic model, B. Dear for helpful discussions regarding interpretation of DLS data, Texas Materials Institute, part of the Material Science Engineering programme at University Texas at Austin, for supporting the management of the DLS, A. Miklos for helpful discussions regarding supercharged proteins and A. Webb for assistance with confocal microscopy, and the Center for Biomedical Research Support Microscopy and Imaging Facility at the University Texas at Austin for supporting the management of the confocal microscope.

Author contributionsA.J.S., Y.Z., V.R., J.Gl., J.Go., A.P., C.J., D.W.T., S.C.G. and A.D.E. conceived and designed the experiments. A.P., A.J.S. and J.Go. designed proteins. A.J.S., B.R.M. and A.P. expressed the proteins. A.P., J.Go. and C.J. performed early optimization of DLS and FRET experiments. A.J.S. and B.R.M. carried out DLS and FRET experiments and analysed the data. J.C.G., J.C.L. and D.W.T. performed the negative stain EM experiments and analysed the data. Y.Z. performed the cryo-EM experiments and atomic model building. A.J.S., Y.Z. and D.W.T. interpreted the cryo-electron microscopy structure and produced the structure figures. V.R. and J.Gl. designed the simulations. V.R. performed simulations. A.J.S. and J.Go. performed and interpreted the confocal microscopy experiments. A.J.S., Y.Z., V.R., J.Gl., S.C.G., D.W.T. and A.D.E. wrote the manuscript, and all authors reviewed and commented on the manuscript.

Competing interestsThe authors declare no competing interests.

Additional informationSupplementary information is available for this paper at https://doi.org/10.1038/s41557-018-0196-3.

Reprints and permissions information is available at www.nature.com/reprints.

Correspondence and requests for materials should be addressed to D.W.T. or A.D.E.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© The Author(s), under exclusive licence to Springer Nature Limited 2019

NATuRE ChEMiSTRY | VOL 11 | MARCH 2019 | 204–212 | www.nature.com/naturechemistry212