Beyond Icosahedrons: Exploring the Structural Space of Polymorphic HK97 Family Protein Cages

The Myxococcus xanthus encapsulin is a self-assembling HK97-family protein that sequesters iron by encapsulating ferritin-like proteins. In vivo, encapsulin protomer subunits assemble into T1 and T3 icosahedral cages, whose high degree of multivalency has made the system an attractive target scaffold for vaccine development, drug delivery, and biomanufacturing applications [1]. Interestingly, coat protein–coat protein interactions appear to be fairly plastic, as multiple mutations have been identified which do not abort higher-order assembly but rather alter the geometry of the resulting structures. These alternative structures are often overlooked because of their low abundance in wild-type populations; however closer examination may reveal principles for the rational design of protein nanoparticles with controllable, biophysically distinguishable geometries. Here, we revisit an extraordinarily rich cryo-EM dataset of a M. xanthus encapsulin construct containing a 20 amino acid insertion from the SARS-CoV-2 receptor binding domain in the encapsulin E-loop [2]. This insertion was previously described to allow encapsulins, comprised of a single identical subunit, to adopt eight additional assemblies beyond the typical T1 and T3 icosahedral geometries; however it was postulated even at the time of publication that this characterization fell short of being exhaustive. To remedy this, we developed an iterative processing pipeline using readily-accessible tools in CryoSPARC which allows users to approach exhaustive classification of heterogeneous viral-like particle samples [3] (Fig 1). We began with a dataset of 6,742 motion-corrected micrographs which were “rehydrated” from compressed .jpeg files into .mrc format [4]. Templates for particle picking were generated by 2D classification of manually picked 1,024 particles with special attention paid to select particles across the spectrum of sizes and shapes present in the micrographs. Template picking produced 2.6 million particles which were subjected to two rounds of 2D classification into 200 classes to remove contaminants and poorly aligned views. ~1 million clean particles were then grouped by 2D super-classification using a Rebalance 2D job, yielding five morphology-based super-classes. In this case, the morphology grouping acted as an effective in silico size exclusion step. For each super-class, an ab initio reconstruction (one volume per super-class) was generated and refined using homogeneous refinement to produce low-resolution initial maps. To begin dissecting structural heterogeneity within each super-class, particles were subjected to heterogeneous refinement provided with five identical references, then each resulting volume was subjected to further 3D classification into additional five classes (25 classes per super-class; 125 total for the dataset). The symmetry of the resulting cages could then be determined by manual inspection and applied in a subsequent homogeneous refinement to produce a high to moderate resolution map. This extensive classification approach produced several degenerate classes which gave us confidence that we were approaching exhaustive classification of intact protein cages present in the encapsulin dataset. Due to the similarity of many of the cages present within each super-class, we reasoned that particles could be easily misassigned during the initial round of heterogeneous refinement. All particles in each super-class were therefore re-sorted by heterogeneous refinement using all unique symmetrized maps as references, and these were evaluated for homogeneity in “Cleanliness Checks” which involved 3D Classification into 5 classes. If an additional cage was identified in the output of a cleanliness check, it would also be refined with symmetry and used as a reference in another round of heterogeneous refinement using all particles in a super-class. This process was repeated iteratively until no new cages were identified, and the cleanliness check was passed. Once homogeneous classes were obtained, particles corresponding to identical maps across super-classes were combined, and when the symmetry of a cage implied chirality, chiral pairs were separated by heterogeneous refinement with two references of opposite hand, low-pass filtered to 20 Å. Each resulting cage is comprised of a mixture of pentameric and hexameric capsomeres containing of five and six protomers, respectively. These can be represented by differently colored spherical markers placed at the pore of the axial domain and connected to provide an impression of the three-dimensional symmetry in a “constellation model” (Fig. 2a). Following our processing scheme, we identified 30 such cage assemblies produced by a single protomer subunit (Fig. 2b). These occupy much of the available space between the T1 icosahedron with zero hexamer subunits, and the T3 icosahedron with 20 hexamers, all with 12 pentamers in accordance with Euler’s Formula. They possess a range of symmetries including asymmetric, rotational, dihedral, and tetrahedral in addition to the canonical icosahedrons. Further, there are several cases in which cages with multiple symmetries are observed for a given number of hexamers, and multiple conformers are observed for a given order of symmetry (Fig. 2b). The presence of many cages which differ by incorporation of a single hexamer led us to wonder whether they might provide any insight regarding cage assembly. We noticed that cages which differ by a single hexamer could be aligned well on all but one face, and that the differing faces appeared to be related by a rearrangement of one hexamer and two pentamers in the n-1 hexamer cage to make room for the additional hexamer in the n hexamer cage (Fig. 3). This rearrangement was consistent with Endo-Kroto C2 insertions which describe similar relationships between carbon Fullerene cages [5]. Moreover, it became apparent that when multiple cages are observed with n hexamers, they are related to the n-1 hexamer cage(s) by Endo-Kroto insertions at different symmetrically distinct hexamers. This is consistent with a model by single protomer addition in which it is possible for intermediate groups of five protomers can either close into a pentamer or accommodate a protomer addition to become a hexamer throughout a growing cage [6]. The encapsulin construct studied here also appears to penalize lateral association of three hexamers observed in small cages and appears only as weak density when present in the larger 16 and 17 hexamer cages (Fig. 2b). This is in contrast to a wild-type Qβ viral-like particle dataset which we processed similarly to reveal 15 unique cages in which association of three hexamers occurs regularly. Thus, exhaustive classification using our iterative cryo-EM processing pipeline provides insights into the geometric constraints which govern assembly of different protein cages. In sum, near-exhaustive analysis of this highly heterogeneous dataset reveals that a single encapsulin protomer can access a broad and nearly continuous spectrum of cage geometries. The E-loop insertion reshapes this assembly landscape, biasing local capsomer interactions and altering the accessible set of architectures. Beyond this specific construct, our workflow demonstrates that deeply heterogeneous viral-like particle datasets can be systematically deconvolved using accessible tools, enabling structural interrogation of assemblies that are typically overlooked. Such comprehensive characterization provides a foundation for rationally tuning cage geometry in engineered protein nanoparticles. Cryo-EM processing scheme. A sample micrograph from the mutant M. xanthus encapsulin dataset is shown at the top with the final 2D class averages shown to the right. These were then split into super-classes containing extra small (XS), small (S), medium (M), large (L), and extra large (XL) particles. An example of further downstream processing is shown for the M super-class, in which maps are colored by radius from the centroid from red to white to blue to best illustrate the features of each given map. Near exhaustive classification reveals 30 unique cages produced by the mutant encapsulin. a) One map revealed by the processing scheme in Fig. 2 with D2 symmetry and 10 hexamer subunits is shown to the left in gray. Second from the left, hexamer subunits are colored in blue and pentamer subunits are colored in green. Second from the right, the map is made transparent and colored in gray while blue and green spherical markers indicate the center of each hexameric and pentameric subunit, respectively. Furthest to the right, the map is removed and markers of like capsomeres are connected by similarly colored lines to indicate overall symmetry in a constellation model. b) Constellation models for the 30 identified cages, with symmetry groups followed by the number of hexamers present. Subscript numbers indicate members of chiral pairs. Endo-Kroto insertions describe the relationship between cages which differ by one hexamer. A cage with D3 symmetry and three hexamers is shown in blue and a tetrahedral cage with four hexamers is shown in red. The top representations show faces of each cage with corresponding subunit arrangements, while the bottom shows the faces after a 120° rotation which are related by a rearrangement consistent with an Endo-Kroto insertion. Relevant subunit arrangements are shown as cartoon schematics to the right of each map. The maps possess superimposed constellation models in which green represents pentamers and blue represents hexamers as in Fig. 2.