Accurate design of megadalton-scale two-component icosahedral protein complexes

Abstract

Nature provides many examples of self- and co-assembling protein-based molecular machines, including icosahedral protein cages that serve as scaffolds, enzymes, and compartments for essential biochemical reactions and icosahedral virus capsids, which encapsidate and protect viral genomes and mediate entry into host cells. Inspired by these natural materials, we report the computational design and experimental characterization of co-assembling two-component 120-subunit icosahedral protein nanostructures with molecular weights (1.8–2.8 MDa) and dimensions (24–40 nm diameter) comparable to small viral capsids. Electron microscopy, SAXS, and X-ray crystallography show that ten designs spanning three distinct icosahedral architectures form materials closely matching the design models. In vitro assembly of independently purified components reveals rapid assembly rates comparable to viral capsids and enables controlled packaging of molecular cargo via charge complementarity. The ability to design megadalton-scale materials with atomic-level accuracy and controllable assembly opens the door to a new generation of genetically programmable protein-based molecular machines.

The remarkable forms and functions of natural protein assemblies have inspired many efforts to engineer self- and co-assembling protein complexes (1–24). A common feature of these approaches, as well as the structures that inspired them, is symmetry. By repeating a small number of interactions in geometric arrangements consistent with the formation of regular structures, symmetry reduces the number of unique interactions and subunits required to form higher order assemblies (2, 25). Symmetric complexes can be designed to form through self-assembly of a single type of protein subunit or co-assembly of two or more distinct types of protein subunits. Multi-component materials possess several important advantages, including the potential to control initiation of assembly by mixing independently prepared components. This property could allow, for example, assembly to be performed in the presence of cargo molecules in order to package the cargo inside the designed nanomaterial. Thus far, only relatively small, 24-subunit two-component tetrahedra have been designed with high accuracy (20, 26). Packaging substantial amounts of cargo will require larger assemblies, such as those with icosahedral symmetry; icosahedra possess the highest possible symmetry of any polyhedron in three-dimensional space and, consequently for the purpose of packaging, generate the maximum enclosed volume for symmetric assemblies formed from a given size protein subunit (27, 28).

We set out to design two-component icosahedral protein complexes capable of packaging macromolecular cargo via controlled in vitro assembly. The two-fold, three-fold, and five-fold rotational axes present within icosahedral symmetry provide three possible ways to construct such complexes from pairwise combinations of oligomeric building blocks; we refer to these architectural types as I53, I52 and I32 (fig. S1). The I53 architecture is formed from a combination of twelve pentameric building blocks and twenty trimeric building blocks aligned along the five-fold and three-fold icosahedral symmetry axes, respectively (Fig. 1, A–E; I53 = Icosahedral assembly constructed from 5mers and 3mers). Similarly, the I52 architecture is formed from twelve pentamers and thirty dimers (Fig. 1F), and the I32 architecture is formed from twenty trimers and thirty dimers, each aligned along their corresponding icosahedral symmetry axes (Fig. 1G). To generate novel icosahedral assemblies, 14,400 pairs of pentamers and trimers, 50,400 pairs of pentamers and dimers, and 276,150 pairs of trimers and dimers derived from X-ray crystal structures (tables S1–3) were arranged as described above, with each building block allowed to rotate around and translate along its five-fold, three-fold, or two-fold symmetry axis. These degrees of freedom (DOFs) were systematically sampled to identify configurations suitable for interface design, as assessed by several parameters, including the size and secondary structure content of the newly formed interface, as well as the backbone geometry between pairs of contacting residues. Protein-protein interface design calculations were then carried out on the resulting 66,115 designs of type I53, 35,468 designs of type I52, and 161,007 designs of type I32. The designs were filtered based on a variety of metrics, including interface area, predicted binding energy, and shape complementarity (29). 71 designs of type I53, 44 of type I52, and 68 of type I32, derived from 23 distinct pentameric, 57 distinct trimeric, and 91 distinct dimeric protein scaffolds, were selected for experimental characterization (fig. S2–5, table S4).

Fig. 1. Overview of the design method and target architectures.

(A–E) A schematic of the design process illustrated with the I53 architecture. (A) An icosahedron is outlined with dashed lines, with the five-fold symmetry axes (grey) going through each vertex and three-fold symmetry axes (blue) going through each face of the icosahedron. (B)12 pentamers (grey) and 20 trimers (blue) are aligned along the 5-fold and 3-fold symmetry axes, respectively. Each oligomer possesses two rigid body DOFs, one translational (r) and one rotational (ω) that are systematically sampled to identify configurations (C) with a large interface between the pentamer and trimer (D) suitable for protein-protein interface design; only the backbone structure and beta carbons of the oligomers are taken into account during this procedure. (E) Amino acid sequences are designed at the new interface to stabilize the modeled configuration. (F) The I52 architecture comprises 12 pentamers (grey) and 30 dimers (orange) aligned along the five-fold and two-fold icosahedral symmetry axes. (G) The I32 architecture comprises 20 trimers (blue) and 30 dimers (orange) aligned along the three-fold and two-fold icosahedral symmetry axes.

Codon-optimized genes encoding each pair of designed sequences were cloned into a vector for inducible co-expression in E. coli, with a hexahistidine tag appended to the N- or C-terminus of one subunit in each pair. The proteins were expressed at small scale, purified by immobilized metal-affinity chromatography (IMAC), and clarified lysates and purification products subjected to gel electrophoresis under denaturing conditions to screen for soluble expression and co-purification of the hexahistidine-tagged and non-tagged subunits (fig. S6A). Designs appearing to co-purify were subsequently analyzed by non-denaturing gel electrophoresis to screen for slowly migrating species as an additional indication of assembly to higher order materials (fig. S6B). Those found to both co-purify and assemble were expressed at larger scale and purified by IMAC followed by size exclusion chromatography (SEC, fig. S7). Ten pairs of designed proteins, four I53 (I53-34, I53-40, I53-47, and I53-50), three I52 (I52-03, I52-32, and I52-33) and three I32 designs (I32-06, I32-19, and I32-28), yielded major peaks by SEC near the elution volumes expected based on the diameters of the design models (Fig. 2, table S4). Two other designs, I53-51 and I32-10, also appeared to form large, discrete assemblies, but their structures could not be verified by subsequent experiments (Supplementary Text, fig. S8 and S9).

Fig. 2. Experimental characterization by size exclusion chromatography and small-angle X-ray scattering.

Computational design models (left), SEC chromatograms (middle), and SAXS profiles (right) are shown for (A) I53-34, (B) I53-40, (C) I53-47, (D) I53-50, (E) I52-03, (F) I52-32, (G) I52-33, (H) I32-06, (I) I32-19, and (J) I32-28. Design models (shown to scale relative to the 30 nm scale bar) are viewed down one of the 5-fold symmetry axes with ribbon-style renderings of the protein backbone (pentamers are shown in grey, trimers in blue, and dimers in orange). Co-expressed and purified designs yield dominant SEC peaks near the expected elution volumes for the target 120-subunit complexes and X-ray scattering intensities (grey dots) that match well with profiles calculated from the design models (green). Alternative configurations of the designs, generated by translating and/or rotating the oligomeric building blocks in the design models about their aligned symmetry axes by +/− 10 Å and/or 20 degrees, respectively, generally fit worse with the SAXS data than the original design models (the range of values obtained from fitting the alternative configurations is shown with light blue shading).

Small-angle X-ray scattering (SAXS) performed on the SEC-purified samples indicated all ten designs form assemblies similar to the intended three-dimensional configurations in solution. The experimentally measured SAXS profiles are feature-rich and distinct, with multiple large dips in scattering intensity in the region between 0.015 Å⁻¹ and 0.15 Å⁻¹, each of which is closely recapitulated in profiles calculated from the design models (Fig. 2). In order to further evaluate how accurately and uniquely the design models match the experimental data, each was compared to a set of alternative models generated by systematically perturbing the radial displacements and rotations of the building blocks in each design by +/− 10 Å and/or 20 degrees, respectively. The vast majority of alternative configurations were found to produce worse fits to the experimental data than the original design models (Fig. 2), suggesting that the materials assemble quite precisely in solution.

The information provided by SAXS about the overall ensemble of structures observed in solution for each design was complemented and corroborated by visualization of individual particles by negative stain electron microscopy (EM). Micrographs of I53-34, I53-40, I53-47, I53-50, I52-03, I52-33, I32-06, and I32-28 show fields of particles with the expected size and shape of the design models, and particle averaging yields distinct structures clearly matching the models (Fig. 3). The large trimeric and pentameric voids observed in the I52 and I32 averages, for instance, closely resemble the cavities in projections generated from the corresponding design models when viewed down the three-fold and five-fold symmetry axes, respectively. The turreted morphology of the I53-50 and I52-33 design models and projections, resulting from pentameric and dimeric components that protrude away from the rest of the icosahedral shell, are also readily apparent in the corresponding class averages. Although the results from SEC and SAXS strongly indicate I52-32 and I32-19 form assemblies closely matching the design models in solution, both appear to be unstable under the conditions encountered during grid preparation, yielding broken particles not suitable for further EM analysis (fig. S10).

Fig. 3. Characterization of the designed materials by electron microscopy.

Left: raw negative stain electron micrographs of co-expressed and purified (A) I53-34, (B) I53-40, (C) I53-47, (D) I53-50, (E) I52-03, (F) I52-33, (G) I32-06, and (H) I32-28. All raw micrographs shown to scale relative to 100 nm scale bar in panel (H). Insets: experimentally computed class averages (roughly corresponding to the five-fold, three-fold, and 2-fold icosahedral symmetry axes; left) along with back projections calculated from the design models (right). Each inset box width corresponds to 55 nm.

To further evaluate the accuracy of our designs, X-ray crystal structures were determined for one material from each of the three different architectural types: I53-40, I52-32, and I32-28 (Fig. 4 and table S5). Although the resolution of the structures (3.5 to 5.6 Å) is insufficient to permit detailed analysis of the side chains at the designed interfaces, backbone-level comparisons show the inter-building block interfaces were designed with high accuracy, giving rise to 120-subunit complexes that match the computational design models remarkably well. Comparing pairs of interface subunits from each structure to the design models yields backbone root mean square deviations (r.m.s.d.) between 0.2 and 1.1 Å, while the r.m.s.d. over all 120 subunits in each material ranges from 0.8 to 2.7 Å (Fig. 4, A–C and table S6). With diameters between 26 and 31 nm, over 130,000 heavy atoms, and molecular weights greater than 1.9 megadaltons, these structures are comparable in size to small viral capsids and, to our knowledge, the largest designed biomolecular nanostructures to date to be verified by X-ray crystallography (fig. S11).

Fig. 4. Crystal structures, assembly dynamics, and packaging.

Design models (top) and X-ray crystal structures (bottom) of (A) I53-40, (B) I52-32, and (C) I32-28. Views shown to scale along the 3-fold, 2-fold, and 5-fold icosahedral symmetry axes. Pentamers shown in grey, trimers blue, and dimers orange. R.m.s.d.s are between crystal structures and design models over all backbone atoms in all 120 subunits. (D) In vitro assembly dynamics of I53-50. (Top) Schematic illustration. (Bottom) Normalized static light scattering intensity (detector voltage, solid lines) plotted over time after mixing independently expressed and purified variants of the I53-50 trimer and pentamer in a 1:1 molar ratio at final concentrations of 8, 16, 32, or 64 μM each (blue, orange, grey, and black lines, respectively). Intensities measured from SEC-purified assembly at 8, 16, 32, or 64 μM concentrations indicated with dashed horizontal lines and used as the expected endpoint of each assembly reaction. The midpoint of each reaction is marked with a dashed vertical line. (E) Encapsulation of supercharged GFP in a positively charged I53-50 variant. (Top) Schematic illustration. (Bottom) Superose 6 chromatograms and SDS-PAGE analysis of packaging/assembly reactions performed in buffer containing: (Top Panel) 65 mM NaCl, (Middle Panel) 1 M NaCl, or (Bottom Panel) 65 mM NaCl with a trimer variant without mutations to positively charged residues. In each case, the same buffer used in the packaging/assembly reaction was also used during SEC. Absorbance measurements at 280 nm (black) and 488 nm (green) are shown. Each SEC chromatogram was normalized relative to the 280 nm peak near 12 mL elution volume. Locations of 37, 25, 20, and 15 kDa molecular weight markers on SDS-PAGE gels are indicated by horizontal lines.

The multi-component composition of the materials presents the possibility of controlling their assembly through in vitro mixing of independently produced building blocks (20). Taking advantage of this feature, the assembly kinetics of an I53-50 variant (fig. S12A) with improved individual subunit stability was investigated by light scattering (Supplementary Materials). SEC-purified components were mixed at concentrations of 64, 32, 16, or 8 μM and the change in light scattering monitored over time (Fig. 4D). Assembly is roughly halfway complete within 1 minute at 64 and 32 uM, 3 minutes at 16 uM, and 10 minutes at 8 uM. Similar assembly time scales have been observed for several viral capsids (30, 31). Since our design process focused exclusively on structure without any consideration of kinetics, these results raise the interesting possibility that the rate of assembly of these viral capsids has not been highly optimized during evolution.

The ability to assemble the materials in vitro potentially enables the controlled packaging of macromolecular cargoes. To investigate this possibility, the trimeric and pentameric components of an I53-50 variant with several mutations to positively charged residues on the interior surfaces of the two components (Supplementary Materials) were successively mixed with a supercharged GFP with a net charge of −30 (32), and encapsulation was evaluated using SEC followed by SDS-PAGE of relevant fractions (Fig. 4E and Supplementary Materials). When both the packaging reaction and SEC were performed in a buffer containing low (65 mM) NaCl, GFP(−30) and both I53-50 components co-eluted from the column at the same elution volume previously observed for unmodified I53-50 (Fig. 2D). Mixtures of GFP(−30) with only one of the two components eluted at later volumes, indicating that the observed co-elution requires assembly of I53-50 (fig. S12, B–D). When the packaging reaction was carried out with buffer containing high (1 M) NaCl or using a variant of the trimeric component lacking the positively charged residues on the interior surface, little to no co-elution was observed (Fig. 4E), demonstrating that packaging is driven by the engineered electrostatic interactions between the I53-50 interior and GFP(−30). High salt incubation resulted in disassociation of packaged GFP (fig. S12E), as also observed for an evolved variant of a naturally occurring protein container that packages cargo via electrostatic complementarity (33, 34). Based on measurements of fluorescence intensity and UV/Vis absorbance, we estimate approximately 7 to 11 GFPs are packaged per icosahedral assembly in 65 mM NaCl, occupying roughly 11 to 17% of the interior volume (Supplementary Materials).

How do the architectures of our designs compare to those of virus capsids and other icosahedral protein complexes found in nature? In the nomenclature introduced by Caspar and Klug (27), our designs can be considered T=1 assemblies in which the asymmetric unit is a heterodimer comprising one subunit from each of the two components. The most similar naturally occurring structures of which we are aware are Cowpea Mosaic Virus (CPMV) and related 120-subunit capsids with pseudo T=3 symmetry. Like our I53 designs, CPMV is composed of 60 copies each of two distinct protein subunits, with one type of subunit arranged around the icosahedral 5-folds and a second type of subunit arranged around the 3-folds (fig. S13). However, the two subunits of CPMV are composed of three similar domains occupying spatially equivalent positions to those found in T=3 assemblies formed from 180 copies of a single type of protein subunit (35, 36). Our I53 designs display no such underlying pseudosymmetry and therefore cannot be considered pseudo T=3. Furthermore, we are not aware of any natural protein complexes characterized to date that exhibit I52 or I32 architectures. Our designs thus appear to occupy new regions of the protein assembly universe, which have either not yet been explored by natural evolution or are undiscovered at present in natural systems.

The size and complexity of the materials presented herein, together with the accuracy with which they assemble, push the boundaries of biomolecular engineering into new and exciting territory. The large lumens of our designed materials, combined with their multi-component nature and the ability to control assembly via mixing of purified components, makes them well suited for encapsulation of a broad range of materials including small molecules, nucleic acids, polymers, and other proteins. These features, along with their potential for precisely engineered chemical or genetic modifications, make them attractive starting materials for the design of functional protein nanomaterials for applications in targeted drug delivery, vaccine design, and bioenergy.

One Sentence Summary.

A computational approach enables design of 120-subunit icosahedral protein cages capable of packaging macromolecular cargo.