Cryo-EM structure of rhinovirus C15a bound to its cadherin-related protein 3 receptor

Significance

Rhinovirus C (RV-C) infections are closely linked to 50 to 85% of hospital-level childhood asthma exacerbations and can lead to significant adult respiratory problems. Currently, there are no effective antiviral treatments or vaccines available. Unlike other common cold rhinoviruses, the RV-C are dependent on cell surface recognition of a unique viral receptor, cadherin-related protein 3 (CDHR3). Human expression of allele forms of CDHR3 determine specific susceptibility to RV-C infections and contribute directly to the severity of asthma phenotypes. With cryo-electron microscopy, we have determined the atomic structures of the first extracellular domains (EC1 and EC1+2) of CDHR3 bound to an RV-C. The new molecular interfaces highlight distinct amino acid interactions that can now be targeted by antivirals or possibly vaccines.

Abstract

Infection by Rhinovirus-C (RV-C), a species of Picornaviridae Enterovirus, is strongly associated with childhood asthma exacerbations. Cellular binding and entry by all RV-C, which trigger these episodes, is mediated by the first extracellular domain (EC1) of cadherin-related protein 3 (CDHR3), a surface cadherin-like protein expressed primarily on the apical surfaces of ciliated airway epithelial cells. Although recombinant EC1 is a potent inhibitor of viral infection, there is no molecular description of this protein or its binding site on RV-C. Here we present cryo-electron microscopy (EM) data resolving the EC1 and EC1+2 domains of human CDHR3 complexed with viral isolate C15a. Structure-suggested residues contributing to required interfaces on both EC1 and C15a were probed and identified by mutagenesis studies with four different RV-C genotypes. In contrast to most other rhinoviruses, which bind intercellular adhesion molecule 1 receptors via a capsid protein VP1-specific fivefold canyon feature, the CDHR3 EC1 contacts C15a, and presumably all RV-Cs, in a unique cohesive footprint near the threefold vertex, encompassing residues primarily from viral protein VP3, but also from VP1 and VP2. The EC1+2 footprint on C15a is similar to that of EC1 alone but shows that steric hindrance imposed by EC2 would likely prevent multiprotein binding by the native receptor at any singular threefold vertex. Definition of the molecular interface between the RV-Cs and their receptors provides new avenues that can be explored for potential antiviral therapies.

Rhinovirus (RV) isolates are small, positive-sense, single-stranded RNA viruses cataloged within the Enterovirus genus of Picornaviridae. The RV-A, RV-B, and RV-C species subdivide more than 170 known genotypes according to clade-specific similarities in genome sequence, serology, and capsid structure (1). As a whole, viruses in the three RV species are the major etiologic agents of common colds but frequently cause much more acute respiratory diseases (2, 3).

Multiple RV-As and RV-Bs have been characterized in detail at the genome and structure levels since their collective isolation more than 50 y ago. The RV-Cs have been described only recently (4–6), but current and retrospective clinical observations now associate these viruses with the most severe forms of virus-induced asthma exacerbations, especially in children (7–13), and it is clear the RV-Cs have physical and disease properties unique to this species. Among these are poor growth in tissue culture, a technical problem that hampered structure resolution until 2015 (14), when cryo-electron microscopy (EM) advances allowed the determination of empty capsids (to 3.2 Å), and full infectious particles (to 2.8 Å) of recombinant isolate C15a from a minimal virus preparation.

All RV infectious virions have 60 biological protomers composed of four capsid proteins—VP4, VP2, VP3, and VP1 (e.g., Fig. 1C)—each unit derived contiguously from a common polyprotein precursor. The icosahedral asymmetric repeat (e.g., Fig. 1B) covers portions of two adjacent protomers within the same fivefold pentamer. Relative to other RVs, the RV-Cs have several atypical features, including a physically prominent, highly immunogenic, finger-like extrusion formed by a VP1 capsid protein insertion (10 to 16 amino acids), located between the icosahedral fivefold vertex and the threefold vertex of each biological protomer. Moreover, the characteristic RV raised fivefold “plateau” is noticeably reduced by large deletions elsewhere in VP1, such that C15a lacks 110 to 175 amino acids of mass per fivefold vertex relative to the RV-As or RV-Bs. A third key difference is that the hollow “pocket” usually formed within the core of an RV VP1 and buried under the floor of the “canyon” is collapsed in C15a and contains no hydrophobic molecule or “pocket factor.” The resultant novel topographies are the presumed causes of altered immunology, cell binding, and receptor interactions unique to the RV-Cs.

Fig. 1.

CDHR3 EC1 complexed with C15a. (A) Domain representation of recombinant EC1 and EC1+2 portions of CDHR3. (B) Surface-rendered density map of the icosahedral complex of C15a:EC1. Projection is along a twofold axis that is perpendicular to the image plane. The coloring is according to distance (Å) from the center of the particle. (C) The C15a capsid proteins within an asymmetric biological unit (blue, VP1 chain A; green, VP2 chain B; red, VP3 chain C; cyan, VP4 chain D) as bound by EC1 (magenta, chain U). The short VP4 protein is internal to the particle.

When initiating an infection, the majority (“major group”) of RV-A and RV-B viruses require intercellular adhesion molecule 1 (ICAM-1) as a cellular receptor (15, 16). A minority (“minor group”) of RV-A instead rely on low-density lipoprotein receptor (LDLR) for cell recognition and binding (17). Both interactions are well-characterized at the structural level. The inherent difficulties with RV-C propagation stem from an alternative requirement for receptors of cadherin-related protein 3 (CDHR3), an airway epithelium protein restricted to the apical surfaces of ciliated cells and thus not expressed in common culture systems (18–20). A single nucleotide polymorphism (rs6967330) encoding a C529Y substitution controls the display density of CDHR3 on airway surfaces (21). The ancestral, albeit minor, “A” allele of this gene (Y529) is among the strongest known genetic correlates for childhood virus-induced asthma susceptibility, manifesting as increased apical surface display relative to protein from the majority “G” allele (C529) and creating a consequent propensity for more frequent and severe RV-C infections (22, 23).

Like that of related cadherins, the sequence of CDHR3 (Fig. 1A) predicts a rod-like arrangement of six linear tandem-repeat extracellular domains (EC1 to EC6), each of about 110 residues, preceded by a signal sequence (residues 1 to 19) and followed by a transmembrane domain (residues 714 to 734) and a cytoplasmic tail (to residue 885). Each repeat sequence is consistent with a cadherin seven-strand β-sandwich fold. The native protein has three mapped N-linked glycosylation sites (N186, N384, and N624) which become further sialylated when the Y529 variant is expressed in tissue culture (24). This particular residue, at the junction of the EC5 and EC6 repeats, presumably helps regulate structurally stabilizing interdomain Ca²⁺ binding, without which cadherins are improperly trafficked and/or withdrawn from cell surfaces (25–27). Recombinant and cell-expressed CDHR3 derivatives show that only the N-terminal EC1 repeat (residues 20 to 126) has a central role in RV-C binding, in a Ca²⁺-dependent, glycosylation-independent manner. The EC2 and EC3 repeats influence this only insofar as some engineered mutations there have the potential to disrupt native protein folding. Soluble recombinant EC1, properly refolded in the presence of Ca²⁺, recapitulates virus–receptor interactions in vitro and can directly inhibit RV-C infection of susceptible cells for several viral genotypes (24).

We have now characterized the molecular interaction between C15a, a growth-adapted variant, and CDHR3 using cryo-EM, redetermining the structure of this virus as complexed with recombinant EC1 of human CDHR3 (resolution of 3.2 Å) and as it is complexed with EC1+2, a similar protein containing the first two cadherin repeat units (resolution of 3.4 Å).

Results and Discussion

Determination of C15a:EC1.

Cryo-EM grids of recombinant virus-receptor complexes were prepared as described in Materials and Methods and SI Appendix, Table S1. Prominent densities for bound EC1 were readily observed on the map, one for each biological protomer (Fig. 1B and SI Appendix, Fig. S1 A and B). Depending on the virus preparation, receptor-bound virions sometimes showed evidence of partial aggregation (SI Appendix, Fig. S2A), and only a fraction of the particles were monodispersed. Nevertheless, an atomic model was readily built from the densities of the capsid proteins VP1 to VP4 (full particles) and then for EC1 (Fig. 1C). The reconstruction reached a global resolution of 3.1 Å when icosahedral symmetry was applied (SI Appendix, Figs. S2B and S3A). Relative to native (Protein Data Bank [PDB] ID code 5K0U) C15a coordinates (14), the virions showed no significant structural changes on EC1 binding (SI Appendix, Fig. S1C); the comparative overall rmsd was 0.32 Å.

The bound EC1 fragment was modeled continuously from L20 to N127, ignoring the short terminal recombinant tags, which were present as partially disordered densities. Like classic cadherin repeats, the core protein has two β-sheets with seven total strands (SI Appendix, Fig. S4). Sheet AGFC and sheet BED run approximately parallel to the longer axis of the protein and tangent to the virus surface. The AGF strands are in closest contact with the capsid (Fig. 2B). This face includes the partial strand A* near the N terminus of EC1, which interacts laterally with strand B, oriented as in classic cadherins (SI Appendix, Fig. S4).

Fig. 2.

EC1-binding site. (A) Surface residues for a C15a icosahedral unit are projected onto a flattened sphere (“roadmap”) and colored according to the distance to the center of the sphere. The black triangle shows unit boundaries as in Fig. 1B. The twofold axis is at the bottom center and perpendicular to the image plane. White contours show the density of bound EC1 at a height of 150 Å. The corresponding (relative) receptor binding sites of a minor group (LDLR, pink) and a major group (ICAM-1, yellow) of RV-A are outlined. (B) EC1 (magenta) binds within a C15a groove formed primarily by VP2 (green) and VP3 (red) capsid proteins from a single protomer unit. The C terminus of VP1 (blue) contributes weakly to the interface. (C) The EC1 virus interface is color-coded by electrostatics (blue, positive; red, negative). Ovals suggest key interaction regions patch#1, patch#2, and patch#3. The proximal virus residues are color-coded by chain as in B. (D) Similar to C, the surface opposite the virus-binding face highlights a proposed CDHR3 multimerization feature.

Novel RV-C Binding Site.

The “major” RV-A and RV-B genotypes interact with ICAM-1 by binding the N-terminal “snout” of the protein deep into the central VP1 5-fold “canyon” feature (15, 16, 28–30). Those RV-As that use LDLR bind this receptor near the VP1 fivefold vertex (17, 31, 32). Both sites span junctions between two biological protomers, centering essentially in one icosahedral asymmetric unit. The EC1 orientation on C15a is quite different (Figs. 1B and 2A). Each EC1 unit localized between a VP1 immunodominant finger-like extrusion and a threefold axis interfacing with one biological protomer. It has close interactions with all three exterior capsid proteins, VP1 to VP3, near each threefold vertex. The protein fits with very good geometrical complementarity into a protomer surface groove formed mainly by VP2 and VP3 (Fig. 2B). The amino end of the EC1 orients toward the twofold axis and the carboxyl end is proximal to the clockwise threefold neighboring site (SI Appendix, Fig. S1). The close packing arrangement predicts likely steric hindrance if more than one full-length CDHR3 were to bind a threefold region during native viral attachment.

EC1 Features.

The archetype adhesion function of classic cadherins is achieved by homotypic or heterotypic EC1 strand-swapping requiring one or two N-proximal tryptophan residue(s) (33, 34). EC1 and EC1+2 recombinant proteins are monomers in solution (24), and there is no homologous N-proximal tryptophan, so CDHR3 must necessarily use an alternative strategy for adhesive binding (35). The new structure shows that the upper face of EC1 away from the virus has a deep groove-like feature between strands D and C (Fig. 2D), exposing at one end the side chain of W76, the sole tryptophan in the first three EC repeats. In previous experiments, mutation of W76 had no effect in vitro on C15 binding, but when the same sequence (W76A) was expressed as full-length CDHR3 in transfected cells, 50 to 80% more virus was captured (24). This suggests that full-length proteins with W76A are more accessible to virus binding, and that they capture measurably more particles. Experiments are underway to test the possibility that this new structure-described groove, with its exposed tryptophan, represents a novel cadherin interaction mechanism, conceivably involving more distal homotypic repeat units or even heterotypic proteins.

For either EC1 or EC1+2 to capture virus, the proteins must be refolded in the presence of calcium (24). By analogy to E-cadherin (34) the EC1-EC2 domain junction was expected to have two or three calcium ions, at least one of which is only loosely held and diffusible if the protein is dialyzed (24). The EM map has density for three calcium ions near the C terminus of EC1, close to the side chains of E33, N34, D93, E95, D125, and N127 (SI Appendix, Fig. S4). The ions bridge loop insertions in strands AB as well as strands EF as they contact the C terminus and thus together stabilize the folding. In the absence of calcium, the consequent strand rearrangements could undoubtedly contribute to the observed inhibition of virus binding (24).

EC1 Binding Face.

The composite VP1 to VP3 virus-binding groove places atoms from 17 capsid residues within 4 Å of 20 EC1 elements in the AGF face (Fig. 2C and SI Appendix, Tables S2 and S3). As in ICAM-1 binding with RV-A or RV-B (36), polar interactions seem to dominate, in that R2234 of VP2 and K3075 of VP3 are in apparent direct contact with “patch#1” and “patch#2” negatively charged regions on the EC1 surface (Fig. 2C). Additional nonpolar C-proximal residues in VP1 (L1275 and I1276) and VP3 (I3058) lie close to the N-proximal hydrophobic short-strand A* of EC1 (“patch#3”), securing both protein tails into the mutual complex.

The contributions of individual EC1 residues to this interface were probed with a panel of mutated proteins in immunocapture assays (24) and with replication inhibition studies similar to those described to test the effects of W76A substitutions. The 16 EC1 codons with probable side chain interactions, as suggested by the structure, were engineered to express one to three separate amino acid mutations (Fig. 3 and SI Appendix, Table S2). Although minor shifts in main chain atoms are unlikely to affect circular dichroism, spectra for each new protein were nonetheless cross-checked relative to wild-type (WT) EC1, to identify overt misfoldings. All 25 sequences passed this test, including separate isolates for several of the proteins, ensuring that the resultant phenotypes were not sample- or concentration-specific artifacts.

Fig. 3.

EC1 mutagenesis. (A and B) EC1 recombinant proteins with the indicated substitutions were tested with C15 in immunoprecipitation capture assays. Percent binding is relative to the WT protein in each gel panel. (C) Additional mutations (L116A and Q117A) were tested in parallel with separate, repeat preparations of I23D and L24D as examples of assay reproducibility.

Consistently in multiple experiments, the most disruptive EC1 substitutions centered on negatively charged D102 in patch#1, aliphatic residues I23/P26 in patch#3, and, to a lesser extent, L116 in patch#2 (Fig. 3; also Fig. 4A and SI Appendix, Tables S2 and S3). The acidic side chain of D102, stabilized by the side chains of T120 and N66, and the buried hydrophobic region of I100 network with VP3 R2234 (SI Appendix, Fig. S5), as well as mutations in D102 or I100, reduced virus interactions to background levels (Fig. 3). Surprisingly, changes in the surrounding patch#1 surface were less disruptive. Some had only a small impact (i.e., T28V, N64A, N99A, Q104A, and V118T), while others did not register with measurable binding phenotypes (i.e., N66A, T120A/V, and Q122A) unless the change was to a sheet-intolerant Gly residue (i.e., T28G and V118G), which may have perturbed the local secondary structure.

Fig. 4.

Mutational relief of infection inhibition. (A) The virus-binding face of EC1 color codes residues according to whether all viruses (red), most viruses (yellow), some viruses (light green), or no viruses (dark green) show “weak” inhibition according to B. Key viral residues interacting with core EC1 red residues are identified in the ovals, as is the percent conservation of viral analog positions among 94 RV-C capsid sequences representing 40 genotypes (SI Appendix, Table S3). Calcium-binding residues are in blue. (B) Each EC1 protein was tested for its ability to inhibit viral replication when added to C02, C15, C41, or C45 virus before infection of fCR3Y cells as described previously (24). Inhibition was calculated relative to infection in the absence of EC1 (WT or mutation). Values are the normalized average of at least three replicates from at least two separate experiments. The SD for any value is ∼3 to 5%. (C) Four mutations were engineered into C15 virus, and the resultant sucrose-purified particles were tested for (relative) percent infectivity and for immunoprecipitation activity with WT recombinant EC1. Percent binding is relative to the WT virus.

The same was true for another Gly change near the aliphatic patch#3, in that H21G at the beginning of the A* strand reduced virus binding significantly, but cognates H21L/Q did not. Nearby, P26A and I23D patch#3 substitutions were again inhibitory, but an adjacent charge change of L24D was not. The carbonyl of P26 is very close to the amide of VP3 N3056, while the P26 R group and nearby I23 contribute strongly to the hydrophobicity of patch#3 that captures VP1 I1276. In patch#2, the general negative polarity positions the Nz of VP3 K3056 near the carbonyl of L116, with contributions from the amide of Q117. Nearby charge conversion to L116D, but not to L116A, reduced some, but not all, of the virus binding. Changes to Q117A/E had no apparent effect, indicating that patch#2 is contributory but probably not as important as patch#1 or patch#3 for overall recognition.

Site Commonalities with Other RV-C.

Among the RV-A and RV-B genotypes that use ICAM-1 and LDLR, there is significant sequence latitude in the virus surface footprints contributing to receptor interactions (1, 36, 37). Soluble CDHR3 EC1, when incubated with virus before cell attachment, is a potent inhibitor of replication for at least four determined RV-C genotypes (24). While there are no Western blot detection mAbs for an RV-C other than C15, binding commonalities relative to alternative genotypes C02, C41, and C45 could still be probed in replication inhibition tests against the panel of EC1 mutant proteins (Fig. 4). Where the binding mode is common, those EC1 mutations unable to tightly saturate a given virus will be displaced by the native cellular receptor to allow infection and genome amplification. We carried out these assessments in triplicate (at least), with several preparations of key proteins, by measuring the amplification of nuclease-resistant genome RNA signals at 24 h postinfection.

As summarized in Fig. 4B and visually in Fig. 4A, the P26A, D102A/N changes, and, surprisingly, L116D were equivalently poor inhibitors for all four viruses. Samples I23D, I100A, V118G/T, and T120A/V also registered poorly with two or three of the viruses but not with all four. Likewise, H21G and T28G locations were not inhibitory when configured as Gly, but alternative substitutions (H21L/Q and T28V), each could inhibit between two and four different viruses. All four viruses fully or partially inhibited N64A, N66A, N99A, Q104A, Q122A, which for the most part behaved like the unmutated (WT) protein.

For the 57 known genotypes of RV-C, there are structure-consistent sequence alignments tabulating species commonalities (1). Key EC1 interface comparisons, summarized in SI Appendix, Table S3, show strong diversity in recorded sequences for these four viruses. This is particularly true for most of the VP3 interface residues except N3056, which is common to 98% of all RV-Cs and also common to 97% of all RV-As and 100% of all RV-Bs. In the RV-Bs, the analogs to VP3 3058 to 3079 comprise portions of the βB strand with its interior and terminal BC loops and create one of the defining neutralizing immunogenic regions, Nim3, on the surface of RV-B14 (38). In other words, most of the EC1 patch#2 interactions involve exceedingly variable viral Nim3 epitope analogs, including VP3 K3075, the core binding partner for this interface region. Conserved among the RV-C at only 68%, the remainder of this species’ sequences (32%) are aliphatic instead of basic (i.e., no others with Arg, His). Indeed, an engineered C15 virus with a K3075A alteration bound EC1, virtually as well as WT virus (Fig. 4C). Therefore, virus-receptor complementarity within patch#2 is apparently quite flexible, accounting for the observed mutational tolerability when contributing EC1 residues or VP3 K3075 were altered.

In contrast, VP2 R2234, the principal partner of EC1 patch#1 D102, is an Arg (71%) or Lys (26%) in nearly every RV-C. The RV-A and RV-B analogs (VP2 βHI loop) are predominantly aliphatic and do not present an equivalent basic residue for CDHR3 interactions. Mutation of VP2 R2234 (to R2234Q) prevented C15 binding to EC1 (Fig. 4C), consistent with the predicted dominance of patch#1. In patch#3, the highly conserved VP3 N3056 interacts with EC1 P26, and mutations here (N3056D/L) in a C15 context inhibit EC1 binding (Fig. 4C). However, there are no parallel RV-A or RV-B sequences with aliphatic cognates to RV-C VP1 L1275 or VP1 I1276, the additional physical partners of EC1 P26 and I23, respectively. VP1 1276 is an Ile, Val, or Leu residue in 97% of all RV-Cs. The combined data and sequence comparisons reinforce the concept that a singular small core of EC1 residues, predominantly D102 in patch#1 and P26/L24 in patch#3, have common, conserved recognition interactions among most if not all RV-C genotypes. However, there must be additional requirements for supplemental, perhaps stabilizing interactions with the surrounding residues, and to a great degree these are genotype-specific and highly flexible, albeit not held in common by sequences of the RV-As or RV-Bs.

Determination of C15a:EC1+2.

As rod-shaped linked repeats, native cadherins are notoriously difficult to isolate as recombinant materials because of the propensity for self-aggregation. With CDHR3, even the nonglycosylated EC1+2+3 recombinant fragment behaves poorly in solution and much larger than a homogeneous monomer (24). Nonetheless, virus must recognize the intact protein irrespective of whether or not it is in higher-order complexes. The EC1 orientation around the virion threefold vertex predicts structural clashes if the protein is extended C-terminally, raising the possibility that a linked EC2 repeat, a requirement in native full-length formats, might reorient EC1 on the virus. The cryo-EM determinations were repeated, this time using C15a and recombinant EC1+2. Again, prominent densities were observed for the EC1 moiety, in an identical orientation as for the previous smaller protein (Fig. 5A); however, the density threshold was significantly lower compared with the EC1 alone (Fig. 5B), indicating that EC1+2 was binding to C15a with a lower occupancy. In turn, this meant the final maps (icosahedrally averaged) were at slightly poorer resolution (3.4 Å; SI Appendix, Figs. S2B and S3B and Table S1). The model for the capsid and EC1 was unequivocal, however. The EC1 moiety of EC1+2 binds to the same footprint on C15a as EC1 alone (Fig. 5A). The relative rmsd between the EC1 datasets was only 0.30 Å. However, the density of EC2 was much noisier due to its flexibility, and it only became continuous after low-pass filtering.

Fig. 5.

EC1+2 and C15a. (A) Surface-rendered map of EC1+2-bound C15a, color-coded by distance to the particle center (same scale as in Fig. 1B). (B) Enlarged view of a threefold vertex. (Top Left) Map threshold level 2.9. (Top Right) Map threshold level 1.5. (Bottom Left) Low-pass filtering to 8 Å. (Bottom Right) Cartoon representation with the threefold axis position indicated by the triangle.

In the icosahedrally averaged map, the EC2 density does not make any contact with the virus surface, although indeed there were clashes with the N terminus of a neighboring EC1 (Fig. 5B and SI Appendix, Fig. S6) suggesting that only one EC1+2, or occasionally two if some flexibility is allowed, is able to bind simultaneously to a threefold vertex. The lower average occupancy was confirmed by localized three-dimensional classification methods, which sorted different binding patterns around the threefold vertices. Subimages centered at each of the 60 binding sites on the icosahedral surface were generated. Each image contained three putative binding sites related by threefold symmetry. When asymmetric 3D classification was performed in RELION (39) (Fig. 6A), four image classes were observed. The first three classes each occupied one of the three binding sites around the threefold vertex, while the fourth class showed no EC1+2 binding (Fig. 6B). Although double binding cannot not be excluded, the fact that these four states account for approximately 80% of the total data suggests that double binding is rare, if even possible. For native CDHR3, then, contact through the extended AGFC face of its EC1 repeat cannot be accommodated by more than 10 events per virus during the process of cell attachment and subsequent entry.

Fig. 6.

EC1+2 occupancy. (A) From the primary C15a:EC1+2 datasets, the region within a radius of 60 Å, centered on an EC1 moiety in individual complexes, was subjected to localized asymmetric classification. Four of five classes from a RELION 3D analysis of these subimages indicated different situations of EC1+2 occupancy (the fifth class being noisier and a mixture of different states). (B–E) Examples of the resulting maps color-coded by distance to the image plane. The corresponding binding states are indicated by adjacent cartoons (EC1 in magenta).

Implications for RV-C Antivirals.

As a whole, the new structures, the composite mutational data, and requirement for stabilizing calcium ions (24) suggest that the EC1-binding face of β-strand AGFC must be rigidly configured for optimal C15a interactions, presenting the side chains of D102, P26, and I23 in particular into a footprint on the virus involving residues K/R2234, N3056, I/V/L1276, and, to a lesser extent, K3075. The contiguous virus face must contribute an acceptable induced fit unique to each RV-C genotype. Reasonably, then, potential antivirals that perturb the EC1 face or adjacent calcium ions are likely to be quite potent against all RV-Cs.

The observed binding position at the threefold vertex, new among the RVs, was found to overlap almost completely with the known surface accessible RV immunogenic site Nim3. The RV-As and RV-Bs bury essential ICAM-1 contacts deep within their “canyons” under selective pressure to avoid exposure of common features to antibodies, although some Nim1 sites do have the potential for a lesser degree of steric interference (38). The accessibility of the EC1 site on C15a predicts that almost any polyclonal response induced to the linear epitope of the exposed VP1 finger, to the adjacent VP2 Nim2 (twofold-facing base of the finger), or to Nim3 residues should preclude CDHR3 interactions with virus. That is, peptide (finger) or inactivated virus vaccines should elicit neutralizing antibodies to the RV-Cs that compete with the surface-exposed receptor-binding footprint. We are currently testing this prediction by mapping a panel of Fabs onto the surface of C15a.

We did not expect monomers of CDHR3 to bind virions. Virtually all cadherins dimerize then multimerize as part of their native function. For mature CDHR3 on the apical surfaces of ciliated cells, the new structure-described back-side groove may participate in adhesive interaction. Although the search for probable binding partners continues, it is known that as primary bronchial cell cultures age or recover from physical wounding, they become less susceptible to RV-C binding and infection (23, 21). Possibly, the EC1 virus-binding face becomes progressively occluded as the protein format matures or multimerizes on cell surfaces. The observed antiviral progression caused by aging or healing would then be physically equivalent to converting the high-display Y529 genetic phenotype into the more protective low-display C529. It is not yet known whether pharmacologic intervention in this process is possible or would have therapeutic value.

Materials and Methods

Viruses.

Infectious recombinant C15a was propagated in transformed HeLa cells (E8 or fCR3Y; CDHR3 Y529) and then purified by sucrose cushion sedimentation as described previously (14, 24). The C15a sequence has engineered mutations in proteins VP1 (T1125K) and 3A (E41K) that increase progeny yields in cell systems (40). For virus-binding experiments, additional preparations of native-capsid recombinant C02, C15, C41, and C45 viruses were isolated after RNA transfection of E8 cells, as described previously (24). Similarly, new sequences of C15 were engineered by two-step PCR to contain R2234Q, N3056D/L, or K3075A. For all preparations, virion concentration after sucrose sedimentation was determined by qRT-PCR of the respective viral RNAs (24). C15a infectivity was further confirmed by plaque assay (40). Standard picornavirus capsid nomenclature uses the first digit of a residue to designate its inclusion in the VP1, VP2, or VP3 protein (e.g., R2234 is VP2 Arg-234). Foundation sequence alignments based on analogous RV capsid residues have been described previously (1).

Recombinant CDHR3.

Bacterial expression, isolation, and refolding of recombinant human CDHR3 EC1 and EC1+2 proteins have been described previously (24). These comprise respectively, residues 20 to 130 and residues 20 to 237 of native CDHR3 linked to amino terminal FLAG tags and carboxyl-terminal 6X His tags. Protein numbering is according to GenBank AIC58018 with domain delineation from a previous structure model (25). Twenty-four derivatives of EC1 (Fig. 4) differing by single amino acid substitutions were engineered into the parental plasmid by two-step PCR with appropriate cDNA primers. These included one to three separate mutations at 16 amino acid sites predicted by the nascent structure to have proximity with C15a. The value of the refolding protocol for each preparation was assessed by circular dichroism using an AVIV Model 420 CD spectrometer and samples (0.1 mM) prepared in 5 mM Tris, 50 mM NaCl, and 2 mM CaCl₂ buffer. Far UV wavelength scans (200 to 280 nm) were performed at 25 °C with a 5-s averaging time in a 0.2-cm path-length quartz cuvette.

Binding and Infectivity Assays.

His tag-dependent immunoprecipitation assays quantitating C15 virus captured by EC1 proteins have been described previously, as have the commercial and laboratory sources for Western blot detection mAbs (24). In brief, EC1 proteins were incubated with virus and with an α-His tag antibody, followed by reactions with protein G Sepharose beads, washing, then elution. The same reference describes viral replication inhibition tests using recombinant isolates of C02, C15, C41, and C45 viruses (24). Typically, virus (3 × 10⁶ PFUe) was incubated (1 h, 25 °C) with or without recombinant EC1 protein (3 μM) in binding buffer (100 μL; 20 mM Tris pH 8.0, 137 mM NaCl, and 2 mM CaCl₂), followed by dilution into Eagle’s medium (250 μL). Inoculation was onto plated, stably transformed fCR3Y cells. After attachment (30 min at 25 °C, 15 min at 34 °C), the cells were washed twice with PBS to remove unattached virus and incubated for 24 h at 34 °C, followed by harvest (lysis in 350 RLT buffer; Qiagen) and assessment of virus titer by qPCR. Inhibition of virus replication was calculated relative to infection in the absence of rEC1 (WT or mutation).

Cryo-EM Data Collection.

Propagated C15a (∼0.1 mg/mL) and purified WT EC1 (4 mg/mL) or EC1+2 (6 mg/mL) were mixed at a volume ratio of 3:1. This approximates a molar ratio of 90 EC1 per asymmetric virion protomer, of which there are 60 per particle. Incubation was at 4 °C overnight, after which 3-μL aliquots of the mixtures were applied to glow-discharged lacey carbon grids (400 mesh copper, lot 200617; Ted Pella). The setting used for glow discharge was 20 mV, 2 min in a vacuum at 0.2 atmospheric pressure. Whatman blotting papers (20 mm; GE Healthcare, catalog no. 1001-020) were used to blot away the excess sample. The grids were blotted for 3 s and then plunged into liquid ethane using a Cryoplunge 3 system (Gatan) placed in a BL2-certified cabinet.

Two separate sample datasets were collected on a Titan Krios microscope operated at 300 kV and equipped with a K2 direct electron detector in superresolution mode (7,676 × 7,420) and with a Gatan energy filter. Automated data collection was enabled by Leginon (41). The nominal magnification was 81,000×, corresponding to a calibrated superresolution pixel size of 0.865 Å. Each exposure was collected as a 60-frame set, with a frame exposure time of 200 ms. The dose rate on the K2 detector was 8 e⁻/pixel/s (for binned pixels), equivalent to a total dose of 32 e⁻/Ų for each movie. The defocus range used for data collection was 1.5 to 3.5 μm.

Image Processing.

A total of 1,041 movies documenting the C15a:EC1 complexes were collected. Relative motion between frames within any movie were calculated and then corrected using MotionCor2 software (42). The contrast transfer function estimation was calculated with CTFFIND4 (43). Initially, a low-threshold, template-matching method in the Appion package was used to box 28,120 particles (44). Subsequently, a two-dimensional classification was performed by RELION (39) with a circular mask of 350 Å. Particles from selected classes were visually inspected, and a total of 14,974 particles were kept as acceptable. This dataset was then divided at random into two halves for further processing, thereby complying with the gold standard refinement (45). Parameters (e.g., center, orientation, scale, defocus, beam tilt) for projection were then refined through approximately 30 iterations using jspr (46), assuming icosahedral symmetry. A Fourier shell correlation (FSC) metric was calculated between the resulting maps of the two subsets. The measured resolution was 3.14 Å according to the 0.143 criterion (47). The two subsets were then combined to reconstruct a final map, which was then sharpened using RELION 3.0 postprocessing procedures with an estimated B factor of −115 Ų.

The pipeline of image processing for the C15a:EC1+2 complexes was similar. A total of 22,981 particles were selected from 2,452 motion-corrected images. After 2D and 3D classifications in RELION (39), 9,607 particles were used in the final reconstruction by jspr, assuming icosahedral symmetry (46). This resolution was 3.3 Å, according to the gold standard FSC. This map was further sharpened using RELION with a B factor of −115 Ų. Local resolution for both maps were estimated using ResMap (48). The localized 3D classification for this complex relied on modification of a published protocol (49). A marker was placed under the EC1 density in the EC1+2 complex using Chimera (50). Subimages that are centered at this marker and symmetry-related positions were extracted from the original particle images (or “parental images”) with relion_localized_reconstruction.py, with a size of 80 × 80 pixels (49). The resulting 576,420 (60 × 9,607) images were then subjected to asymmetric 3D classification in RELION (39), with a circular mask of 120 Å. The relative orientations and centers of these subimages were derived from the alignment parameters of the parental images and were not refined in the classification.

Model Building and Refinement.

The published coordinates of C15a (PDB ID code 5K0U) were fitted into the new density map of the EC1:C15a complex. The fit was very good overall, and only minor local adjustment in Coot was necessary. Any additional remaining density, representing the bound EC1 protein(s), was initially cropped out. Multiple in silico models for this density were proposed according to the MapToModel function of the PHENIX software suite (51). Ignoring the short N-terminal and C-terminal recombinant tags, one of the best-fitting models was selected as the foundation structure. Manual model-building was then carried out in Coot (52). The final models encompassing a protomer of C15a and an EC1 (without tags) were combined and subjected to real-space corefinement using PHENIX (53). Icosahedral symmetry was applied to generate a complete model for the entire complex, followed by another round of real-space refinement to reduce any putative clashes between symmetry-related proteins. For the C15a:EC1+2 dataset, the new coordinates of the C15a:EC1 complex were fit into the map. No specific model was built for EC2 moiety, because its corresponding density had lower resolution than EC1 or for C15a, indicating likely lower occupancy or higher motion. Nonetheless, a real-space refinement with icosahedral symmetry was carried out using PHENIX to identify the relative EC2 orientation.

Data Deposition.

C15a coordinates for virion particles with RNA (PDB ID code 5K0U) and without RNA (PDB ID code 5KZG) have been published previously (14), as has a Robetta-predicted model of the six EC domains of human CDHR3 (25). PDB ID codes have been assigned for C15a:EC1 (EMD-20443, PDB ID code 6PPO) and C15a:EC1+2 (EMD-20458; PDB ID code 6PSF).