Modeling of the Human Rhinovirus C Capsid Suggests a Novel Topography with Insights on Receptor Preference and Immunogenicity

Abstract

Features of human rhinovirus (RV)-C virions that allow them to use novel cell receptors and evade immune responses are unknown. Unlike the RV-A+B, these isolates cannot be propagated in typical culture systems or grown for structure studies. Comparative sequencing, I-TASSER, MODELLER, ROBETTA, and refined alignment techniques led to a structural approximation for C15 virions, based on the extensive, resolved RV-A+B datasets. The model predicts all RV-C VP1 proteins are shorter by 21 residues relative to the RV-A, and 35 residues relative to the RV-B, effectively shaving the RV 5-fold plateau from the particle. There are major alterations in VP1 neutralizing epitopes and the structural determinants for ICAM-1 and LDLR receptors. The VP2 and VP3 elements are similar among all RV, but the loss of sequence “words” contributing Nim1ab has increased the apparent selective pressure among the RV-C to fix mutations elsewhere in the VP1, creating a possible compensatory epitope.

Introduction

The human rhinoviruses (RV) comprise 3 species, RV-A, RV-B and RV-C, in the Enterovirus genus of the Picornaviridae (1). Collectively, these positive sense RNA viruses are the most frequent cause of the common cold. Between 50-85% of asthma exacerbations are due to RV infections (2-4) and RV-induced wheezing illness in infants corresponds with a high risk of developing childhood asthma (5). Many infectious properties of the RV link directly to their virion structures. Like all picornaviruses, the capsids are icosahedral (pseudo T=3), composed of 60 copies each of four structural proteins, VP1, VP2, VP3 and VP4. The three largest proteins, VP1-3, assume similar 8-stranded, anti-parallel β-barrel motifs, despite being formed from very different sequences (Fig 1). Protomer units of VP1-4 are derived from a common polyprotein precursor. Assembly is nucleated around the RNA during infection into particles with 5-fold, 3-fold and 2-fold axes of symmetry. The shapes and surface extensions inherent to individual VP1-3 confer strain-specific properties of immunogenicity, receptor binding and drug susceptibility to each RV isolate. The short VP4 proteins, cleaved from intermediate precursor VP0, become myristoylated (N-terminus) as an assembly prerequisite, and ultimately localize with protomer symmetry, inside the capsid, adjacent to the packaged RNA.

Figure 1. RV capsid arrangement.

(A) Particles have 60 crystallographic subunits with VP1 (blue), VP2 (green) and VP3 (red) and VP4 (internal) proteins arranged in icosahedral symmetry. (B) A canyon-like depression circles each 5-fold axis from which a hydrophobic drug-binding pocket extends into the interior of each VP1. (C) Subunit ribbon diagram of A16 shows protein contributions to the north and south walls of the canyon. A biological assembly protomer would include this VP1, a VP2+4 precursor (VP0), and the VP3 from the adjacent (clockwise) crystallographic subunit. Figure modeled after Hadfield et al 1995 (40).

The 99 original serotypes of RV-A+B were defined by immunogenic cross-reactivity (6). But now more routinely, related isolates from all 3 species are binned as “genotypes” if their VP1 nucleotide relationships exceed 87% identity (7, 8). The RV-A (79 types) and RV-B (30 types) are well studied at the structural and clinical levels. All these utilize either ICAM-1 (98 “major” types) or LDLR (11 “minor” types) as their cellular receptors. The molecular nuances of these interactions have been described by co-crystallization and EM studies. The RV-A+B that make up the major and minor groups conserve surface footprints that explain how and why particular isolates use their respective receptors to interact with cells (9).

In 2006 the discovery a new RV species surprised the molecular and clinical communities. The RV-C are clearly rhinoviruses, but unlike RV-A+B, they are not readily propagated in typical cell culture systems, including WI-38, WisL, BEAS-2B, A549 and HeLa lines (9). The 51 recognized RV-C types (as binned by sequence analysis) were identified by PCR while fishing through patient samples for other RV. The new isolates have special clinical relevance since it is now recognized the RV-C are associated with up to half of infections in young children (9, 10). They grow readily in both the lower and upper airways, tolerate higher growth temperatures (11), and use cell receptors not common to the RV-A+B (9). Complicated procedures have amplified some RV-C in mucosal organ cultures, but this technique is difficult and requires primary human donor samples (9). Parallel work with air-liquid interface (ALI) cultures is promising (11, 12), but neither technique has yet to produce enough virus for extensive biological or structural studies. Instead, RV-C information is projected by comparative sequencing. There are now full or nearly full genome datasets from ~68 isolates, with deeper information for the VP1 protein (~300 additional seqs). The genome of rhinovirus C15 (strain W10), an early isolate, was cloned into cDNA, and the resulting transcripts proved infectious to HeLa cells (9). Consequent inhibition assays with this (albeit low titer) virus showed that ICAM-1 and LDLR antibodies failed to prevent C15 from attaching to cells. This confirmed the RV-C use unique receptor(s) (9), although their basic biology is similar to other RV. Identification of this new receptor(s) is hampered by the lack of primary organ cultures and inconsistencies among donor samples. The same problems make antiviral tests difficult.

To gain more insight into the RV-C, we turned to computational structure prediction methods. Homology, protein threading and ab initio modeling are three kinds of these applications. Homology modeling uses sequence identity from robust alignments to search determined structures for likely homologs. Threading does not require outside sequences, but best-fits a protein to a database of known structure templates. Ab initio relies neither on homologs or existing structures and instead computes the lowest energy conformation(s) inherent to given protein. Multiple isolate determinations of full-length RNA genomes, representing 141 RV types, have been sequenced and aligned (1). There are atomic resolution structures for A1, A2, B3, B14 and A16 protomers, along with numerous derivatives embodying capsid mutations and drug-bound complexes. The extensive data cohort was ideal for homology (MODELLER, ROBETTA) and threading methods (I-TASSER) when applied to predictions of the C15 capsid. As reported here, the strength of these techniques, backed by deep, refined alignments of all known rhinovirus isolates, allowed construction of a highly predictive model that can guide antiviral drug targeting as well as receptor discovery until an actual RV-C structure is resolved experimentally. The data converge on the finding that all RV-C have major, conserved deletions in the VP1 protein, shaving significant mass from the 5-fold virion plateau, altering the canyon and immunogenic profiles, and presenting novel receptor interfaces and chimeric hydrophobic pockets for drug binding.

Results and Discussion

C15 Model Development

I-TASSER implements multiple threading algorithms and iterative structure assembly simulations to find optimal sub-fragments within a database of structures or within a user-specified structure (13, 14). Full-length models are assembled by excising fragments using replica-exchange Monte Carlo simulations. The functions and ligands are inferred with a final structure-structure alignment program. Outputs consistently score well in CASP competitions (15-18).

C15 was chosen to represent this species because an isolate (W10) was cloned into infectious cDNA and was available as virus to test predictions (9). The C15 VP1-4 share 77% amino acid identity and equivalent indels among all known RV-C (consensus sequence). The respective proteins of 278, 265, 235, and 67 amino acids were assessed independently against the full I-TASSER database, including the known set of all determined picornavirus structures. The program picked A16 as the statistically best model for all VP1-3 sub-fragments (not shown). To refine this assessment, A16 and B14 files, with and without bound pleconaril as part of the structures, were resubmitted as specified templates. Each output (VP1-3) was evaluated according to I-TASSER metrics (13, 14). Again, native A16 (1aym) had the highest scoring confidence levels (C-score), and TM-scores, with RMSD relative to C15, in the 2-3 Å range (Table 1). Given that C15 and A16 share only 44-59% amino acid identity throughout their capsids (discounting indels) these are exceptionally good structure fits. All PDB depositions for the RV, especially for A16 and B14, register large portions of the VP4 sequence as disordered. I-TASSER was able to extrapolate a complete C15 VP4 only by using other enterovirus templates. Therefore, this portion of the model is clearly of lower quality (Table 1).

Table 1.

C15 I-TASSER Confidence Scores

	% identity1	% aligned2	RMSD3	E-value3	Z-score3	-ln(E)3
VP1	44	95.34	1.4	2.65E-14	35.53	31.26
VP2	59	94.34	1.84	7.28E-14	31.86	30.25
VP3	45	99.57	0.79	1.92E-13	30.82	29.28
VP4	62	43.94	3.09	6.15E-02	2.49	2.79

¹Percent pairwise amino acid identity between A16 and C15.

²MAMMOTH (21) evaluated C15 output PDB files from I-TASSER and MODELLER as in Methods. “aligned” is the % of residues with common C-α backbone coordinates (≤ 4 Å) in both models.

³Comparative output values for each protein are indicated. “RMSD” is root-mean-square deviation (in angstroms).

To evaluate whether similar results would be obtained by other prediction methods, the VP1-4 of C15 was submitted to MODELLER (19). This homology modeling program begins with a pairwise input alignment and calculates an initio model containing all likely non-hydrogen atoms that satisfy a spatial restraints methodology. Then, estimates of loops are calculated de novo, or relative to the input template if there is reasonable similarity. Regardless of the selected template (A16 and B14, with or without pleconaril) the resultant C15 models looked similar to each other and each scored well according to internal benchmarks (not shown). The VP1 C15 sequence was also submitted to ROBETTA (20), another online suite that combines “template-based” homology modeling with ab initio methods. Again, from among all possible homologs, the program selected A16 (1aym) as the preferred template. The I-TASSER, MODELLER and ROBETTA results were aligned to each other using MAMMOTH (21) or the MAMMOTH-multi suit (22). Among all outputs, there was 94-96% agreement for the C15 VP1 (Table 2) indicating strong similarity regardless of algorithm. However, since I-TASSER predicted the entire capsid length, albeit with tentative veracity for VP4, subsequent analyses were based on those predictions. That is, an I-TASSER model of C15 relative to A16 for VP1-3, and relative to the whole I-TASSER database for VP4.

Table 2.

I-TASSER, MODELLER and ROBETTA Comparison

MAMMOTH1	VP	% aligned1	RMSD2	E-value2	Z-score2	-ln(E)2
ITAS vs MOD	1	95.34	1.4	2.65E-14	35.53	31.26
ITAS vs MOD	2	94.34	1.84	7.28E-14	31.86	30.25
ITAS vs MOD	3	99.57	0.79	1.92E-13	30.82	29.28
ITAS vs MOD	4	43.94	3.09	6.15E-02	2.49	2.79

MAMMOTH-mult3	VP	% aligned1	RMSD2	E-value2	Z-score2	-ln(P)2
A16 vs ITAS	1	92.09	1.91	0.6E-13	32.15	30.52
A16 vs MOD	1	92.09	1.90	0.6E-13	32.15	30.52
A16 vs Rob	1	91.01	3.10	0.8E-13	31.73	30.13
ITAS vs MOD	1	95.68	1.40	0.2E-13	33.53	33.53
ITAS vs Rob	1	93.53	2.84	0.3E-13	32.70	31.04
MOD vs Rob	1	94.96	2.69	0.2E-13	33.25	33.25

¹MAMMOTH (21) evaluated C15 output PDB files from I-TASSER (ITAS) and MODELLER (MOD) as in Methods. “aligned” is the % of residues with common C-α backbone coordinates (≤ 4 Å) in paired structures.

²Comparative output values for each protein are indicated. “RMSD” is root-mean-square deviation (Å).

³MAMMOTH-mult (22) evaluated VP1 PDB files from A16 (1aym) and C15 VP1 models generated by ITAS, MOD and Robetta (Rob).

The preferred model was then evaluated by ProQ, a structure quality predictor (23). It scored in the “extremely good” range (5.287) for the LGscore, and in the “fairly good” range for the MaxSub (0.361). As benchmarks for these values, parallel submission of the A16 parent (1aym) as resolved to 2.15 Å resolution, also scored “extremely good” for LGscore (5.998) and “fairly good” (0.44) for MaxSub. The projected C15 stereochemistry was checked for feasibility with PROCHECK, part of the PDBSum program suite (24). Calculations for the main-chains of the model showed key parameters (bad non-bonded interactions, C-α tetrahedral distortion, hydrogen bond energy and overall G-factor) to be within the normal range for a typical 2 Å resolution structure. Ramachandran plot quality assessment reflects a structure with between 2.5-3.0 Å resolution and the peptide bond planarity was above the normal range (~3-8 degrees standard deviation) at around 11 degrees. The side chain properties were all “normal” (or “better”) relative to a typical 2 Å resolution structure. The tested parameters included standard deviation of the chi-1 gauche minus torsion angles, standard deviation of the chi-1 trans torsion angles, standard deviation of the chi-1 gauche plus torsion angles, pooled standard deviation of all chi-1 torsion angles and standard deviation of the chi-2 trans torsion angles. Therefore, the selected model is strong and feasible. Apart from VP4, which remains unresolved in most datasets, it is highly consistent with measurable parameters of experimentally determined RV. Proteins for A16 and C15 (model) are shown in Fig 2. The orientation and shapes of the β-barrels are almost indistinguishable.

Figure 2. Structure comparison.

Ribbon diagrams for A16 (gray) and C15 (blue, green, yellow, red; models) are labeled with features, using the nomenclature of Rossmann et al 1985 (29). The structures were formatted, oriented and rendered in MacPyMOL (35).

Topography of the C15 Model

Genome alignments are available for many enteroviruses, and are especially deep for isolates in the RV species (1). The alignments were founded on superimposition of known structures, particularly for the capsid regions and therefore, analogy (same function) and homology (same lineage) are clearly predicted. Indeed, within every viral protein, including VP1-4, there are benchmark residues that unambiguously delineate every internal β or α element. Among the RV, large relative indels are rare. Outside of VP1 and a few discontinuities that are potential sequencing errors (e.g. C19 EU840728), the only serious length variance is at the N-terminus of the 3A protein, where the RV-B are longer than the RV-A (2-3 aa) and the RV-C (9 aa).

Given this conservation, the VP1 indels are striking. The β-barrel cores superimpose, but the collective βB-βC, βD-βE and βH-βI loops of all RV-C are shorter by ~22 aa relative to the RV-A, and ~28 aa relative to the RV-B. For C15 specifically, the βB-βC loop is 10 aa shorter than A16 and 13 aa shorter than B14 (Fig 3). The corresponding βD-βE loops are 8 and 10 amino acids shorter. The βH-βI loops are 4 aa shorter. As these elements supply virtually all of the mass to the 5-fold virion plateau, the C15 model predicts the capsid surface must change radically. Full reconstructions (Fig 4A, Fig 6E) and “roadmap” topographic projections (Fig 4B) highlight the impact. Relative to A16, B14 and every other resolved enterovirus structure (61 in the PDB) the 5-fold plateau of C15 is effectively gone. In its place, the missing mass is so severe as to cause a depression rather than a projection at the 5-fold. Depth-cued cross-sections measure the loss at up to 20 Å in plateau height (Fig 4C). Because of poorly resolved VP4 in existing structures it cannot be anticipated whether this “shaving” creates an overall thinner protein shell at the 5 fold, but most certainly it creates a different physical surface over at least 1/3 of the particle, including all territory “north” of the canyon.

Figure 3. Core structure elements.

The VP1-3 sequences for A16, B14 and C15 are illustrated to scale showing the color-coded locations of α and β segments using the nomenclature of Rossmann et al 1985 (29). The A16 and B14 elements are according to determined structures. The C15 elements are inferred by analogy in sequence alignments. B14 loops encoding neutralization escape mutations (Nim) are highlighted as are key, relative C15 deletions in VP1 (e.g. Δ13).

Figure 4. Surface topography.

(A) Protomer PDB files for A14, B14 and C15 (model) were extrapolated to full icosahedral capsids using UCSF Chimera (38). The color scale illustrates the particle radius, spanning 130 Å (blue) to 165 Å (orange). (B) In parallel, radially depth-cued “roadmaps” show the surface topographies for respective crystallographic subunits as calculated by RIVEM (39). The applied color scale is the same as A. (C). Cross sections of the particles in A, through equivalent 5-fold axes, compare the protein depths of the canyon region and 5-fold plateau. The structures were aligned, displayed and rendered using UCSF Chimera (38).

Figure 6. Immunogenic sites.

(A) A B14 (4rhv) biological protomer highlights residues mapped with escape mutations to a panel of neutralizing monoclonal antibodies, as defined by Sherry et al 1986 (28). (B) The neutralizing immunogenic sites (Nims) cluster on the virion surface (intense colored residues) as part of continuous surface loops (strong outlines). The Nim2 and Nim3 epitopes are partially discontinuous with contributions from VP1 (blue) and non-adjacent segments of VP2 (green) and VP3 (red). These loops define the sequence fragments queried for Nim conservation in Table 3. (C,D,E) A16, B14 and C15 particles were rendered as in Fig 4 except that color was assigned by protein type (as in B) and a semi-transparent sphere was added, masking the topography below a cutoff of 155 Å. The remaining brightly colored features mark residues above this height, and for B14, include all mapped Nim escape mutation sites.

The “south wall” of the C15 canyon, in contrast, and including the 2-fold and 3-fold regions, is more similar to A16 and B14. Side-by-side (Fig 2), it is hard to identify VP2-3 differences that overtly affect the mass or orientation of a β, α or loop element. Among all RV (N= 349), the longest VP2-3 indels are just 1-2 aa, and these tend to be type-, rather than species-specific (Fig 3). The canyon itself is maintained, as evidenced by the depression (blue) in the center of each roadmap icosahedral unit (crystallographic subunit). Additional shallow residues extend north-west and north-east of the central depression, ringing around the 5-fold and completing the canyon (Fig 4B). All 3 viruses have secondary depressions in the 2-fold region. This interface is formed by VP1 and VP3 interactions with contributions from the N-terminus of VP2 (25). As there are no indels covering any interface residues, and strong conservation at the sequence level, the C15 model preserves all relevant contacts at both the 2-fold and 3-fold. It should be noted though, that the structure files for A16 and B14 are missing the N-terminus of VP2 (8-9 aa) and up to half of VP4 (29-40 aa). If the respective surfaces were to be influenced by inclusion of these full proteins, it isn't apparent from roadmaps.

C15 Surface Residues

Virion surfaces are under evolutionary pressure for immunological diversity and receptor binding. The charge distributions for A16 or B14 are typical of their species (Fig 5A,C). Uncharged or mildly polar (yellow: Asn, Gln, Ser and Thr) residues cluster around the 3-fold region, while the 2–fold conserves acid (red: Asp, Glu) and basic (blue: Arg, Lys, His) patches. The north wall of the canyon, edging the plateau, is strongly charged, contributing contacts for ICAM-1 and LDLR. Receptor footprint conservation within the RV-A+B is well described (26, 27), as is the lack of conservation of these same residues in an alignment of RV-C (9). The C15 model not only emphasizes the truncation of the required receptor binding sites (i.e. VP1 βB-βC loop), it shows that the residual VP1 sequence(s) display a very different charge pattern in addition to new topography (Fig 5E). The depression at the 5-fold for all RV-C is anchored with a basic His or Lys (VP1-161) then rimmed with uncharged amino acids in a pattern dissimilar to RV-A+B. Many of these newly exposed VP1 residues are conserved in every RV-C (Fig 5F, black), or present in >90% of the known sequences (dark gray). The model and the alignments converge on the idea of a unique receptor footprint, similar among all RV-C, and likely to map around the unusual shaved VP1 configuration at the 5-fold.

Figure 5. Surface character.

(A,C,E) Roadmap (39) surface depictions for A16, B14 and C15 are color-coded by amino acid types. The color bins includ, acidic (Asp, Glu), polar (Asn, Gln, Ser and Thr), neutral (Ala, Cys, Ile, Gly, Leu, Met, Phe, Pro, Trp, Tyr and Val) and basic (Arg, His and Lys) residues. (B,D,F) Similar roadmaps depict species surface residue conservation. Capsid alignment positions were queried for conservation (% amino acid identity) relative to each species’ consensus sequence. The observed identity, encoded in continuous grayscale (black = 100% conserved, white = 0% conserved), is superimposed on the A16, B14 and C15 residue positions.

RV-C Immunogenicity

Neutralizing immunogenic sites (Nims) for B14 were originally identified in an historic study correlating the structure of the particle with monoclonal antibody escape mutations (28). Continuous and discontinuous epitopes take their names from the VP contributing the dominant surface loop (Fig 6A,B). Nim1a and Nim1b include fragments of the VP1 βB-βC and βH-βI loops. Nim2 pairs the βE-αB “puff” loops of VP2 with the βG-βH “FMDV” segment of VP1. Nim3 is formed by an insertion in the βB of VP3, with contributions from the βB-βC loop and the nearby VP1 COOH tail (29). Not surprisingly, in all resolved RV structures these regions display as raised features (Fig 6CD), allowing antibody contact without compromising the underlying core, the canyon, or receptor binding motifs. In C15, the physical loops for putative Nim2 and Nim3 analogs are similar to A16 and B14 (e.g. Fig 3). But the VP1 loops that would present Nim1a and Nim1b are removed by the species-wide deletions (Fig 6C).

Experimentally, the immunogenicity for any RV-C is unknown and the typing of current isolates is based solely on full-length VP1 pairwise identities. However, as the alignments contain multiple representatives of each RV-A+B; it is possible to record how loop sequences vary for known serotypes, and by inference, to extrapolate potential epitopes for the RV-C. Statistical screening depends on the concept that immunogenic sites fix mutations more readily than the genome as a whole. But given that 141 RV types are already defined, the formative selective pressures cannot be subtle. The alignment segments encoding VP1-3 surface elements were queried for the number of unique sequence “words” observed among the isolates (Table 3). For the RV-A+B in the mapped Nim sites, the sequence variability (observed words) was frequently greater than the number of types within the species. For example, the Nim2(b) loop residues 2155-2167 in VP2 record116 sequence variants among 79 types of RV-A. In the RV-B, there are 37 variants within 30 types. Many of these added sequences, however, covary within a single type, so that overall, about 93-96% of the measured words are unique to a given type, and predictive of it. The Nim-type predictions showed consistent values of 85-98% among the measured RV-A+B surface loop segments regardless of which Nim was queried. As expected, regions that are not under immune selection, fix mutations at lower frequencies. Fewer words (10-12) and lower type-correlations (64-77%) were observed in words from a core β segment (e.g. VP1 βG) or from a non-capsid region of the 3D pol. Here, multiple types frequently share the same sequence words. For perspective, 61% of all RV-A alignment positions and 66% of all RV-B alignment positions share >90% aa identity within their species, regardless of whether the residues are averaged from the P1 (capsid) or P23 (non-structural) regions.

Table 3.

Surface Sequence Variation versus Isolate Type

		Observed sequence “Words”1			“Words” Predict RV Type2
Segment	B14 residues	RV-A3	RV-B4	RV-C5	RV-A3	RV-B4	RV-C5
VP2 Nim2(a)	2135-2143	115	21	32	85 %	80 %	94 %
VP2 Nim2(b)	2155-2167	116	38	38	96 %	93 %	97 %
VP3 Nim3(a)	3055-3064	106	36	36	92 %	96 %	97 %
VP3 Nim3(b)	3071-3079	103	31	35	98 %	95 %	95 %
VP3 COOH	3227-3236	68	15	36	95 %	82 %	94 %
VP1 Nim1ab	1083-1095	115	43	0	90 %	83 %	(deletion)
VP1 Nim1a	1135-1144	76	33	66	85 %	95 %	61 %6
VP1 βG	1186-1198	12	10	7	66 %	68 %	62 %
VP1 βG-βH	1205-1212	77	22	36	90 %	77 %	98 %
VP1 COOH	1279-1289	113	39	38	86 %	93 %	95 %
3D-pol	318-331	10	12	21	64 %	72 %	76 %

¹Unique sequences (words) in same alignment segment

²Words conserved within, but not between types (percent)

³208 isolates, 79 types

⁴74 isolates, 30 types

⁵67 isolates, 32 types

⁶7 of 10 residues deleted

The RV-C are less well conserved. Only 48% and 51% of the alignment positions share >90% identity among the P1 and P23 proteins as a whole, respectively. But in the putative Nim2 and Nim3 sites, including the VP1 and VP2 COOH ends (also on the C15 surface), the correlation with type was 94-97%. These high values suggest the compared words are legacies of strong evolutionary lineages, presumably immunologic, among all RV-C isolates. If true, it means the raised VP2 and VP3 surface topographies (Fig 4, Fig 6E) are likely to be authentic Nims for the RV-C. Moreover, recombination in the P1 region must be an infrequent means of epitope swapping because the diverse RV-C surface segments seem to hold true to type, as assigned through the VP1 as a whole. Surprisingly though, the type-specific word conservation also held true for the RV-C VP1 βG-βH loop, a surface segment contributing to the south wall of the canyon (Fig 2). In aphthoviruses and cardioviruses, which do not have canyon features, this “FMDV loop” is a dominant, continuous, immunogen (30, 31). In the RV-A+B, it is much less so. Relative to the other Nims, there are fewer words observed here among the RV-A+B isolates and lower covariance with type (Table 3). Only a single βG-βH residue (B14 1210) has been identified with antibody escape mutations (28). For these species, immunogenic access to this segment is limited by the steep north wall of the canyon (32). The RV-C however, vary this region equivalently to other surface loops, and for 98% of the isolates, the specific sequence is predictive of type. In contrast to the RV-A+B structures, the C15 model anticipates the shaved 5-fold region should expose more of this surface (Fig 6E), thereby strengthening selective pressure to fix mutations as a variable immunogen. The enhanced RV-C epitope potential is recorded in the βG-βH words and may reflect a degree of compensation for the structural loss of Nim1a and Nim1b. If this is a continuous, exposed surface epitope, as predicted by the model, antibodies to the segment should neutralize RV-C, perhaps more potently than for the equivalent RV-A+B segments.

Insights into RV disease

Isolates in RV-C species are clearly linked to more childhood illnesses involving lower airways than the other RV (10, 33). As a rule, the RV-B tend to limit their replication to sinus tissues and the upper airways, and the majority of RV-A+B are sensitive to culture temperatures above 33-35°C (34). The RV-C readily infect upper and lower airways and are stable to growth at 37°C (11), a property that is capsid-dependent. The lower airway infections by these viruses can be severe, accounting for the majority of asthma attacks resulting in children's hospitalization (33). The profound structure changes in the 5-fold region, predicted by the C15 model necessitate a receptor preference that is not shared by the RV-A+B and preferred growth locale could be part of this. But these same changes could easily make the capsid more stable and less flexible, allowing replication at higher temperatures.

Materials and Methods

Sequences and Alignments

Refined alignments (349 seqs) of complete genome RV RNA sequences are based on foundation superimposition of determined structures as described (1). Clustal and profile fits added to this base all recent genome length (i.e. >6000 base) RV isolates available from GenBank. Placement required every indel to maximize identity within RV types and species. A translated polyprotein alignment with species, type and accession numbers is available in *.meg and *.fasta formats from http://virology.wisc.edu/acp/aligns/. The set includes RV-A (79 types, 208 seqs), RV-B (30 types, 74 seqs), RV-C (32 types, 67 seqs), and EV outgroups (4 species, 10 seqs). A subset alignment with just VP1 proteins was extracted then augmented with an additional ~500 RV-A+B+C datasets, ensuring representation for the 51 known types of RV-C, and adding depth to the RV-A+B within this gene. Current RV nomenclature (8) designates the species letter (A, B or C), and type number (e.g. A16). Strain designations are unique to each accession number.

I-TASSER

The I-TASSER threading method and online web site (http://http://zhanglab.ccmb.med.umich.edu/I-TASSER/) have been published (13, 14). The VP1-4 protein sequences of C15 strain W10 (GQ323774) were extracted into separate files. The VP1 was threaded in runs against: the entire I-TASSER template database (a subset of Protein Databank); A16 (PDB ID:1aym); A16 with pleconaril (1ncr); B14 (4rhv); and B14 with pleconaril (1ncq). The VP2 and VP3 sequences were threaded against A16. The VP4 was threaded against the full template database. C-score, TM-score and RMSD were recorded for each model (Table 1). A biological protomer was constructed by superimposing the selected C15 models on their A16 (1aym) homologs using the alignment function of MacPyMOL (35). The orientations were exported then concatenated manually into a single PDB file. Helix and sheet lengths (Fig 3) were predicted relative to known elements in the refined alignments.

MODELLER

Independent C15 protomer models (VP1-4) were calculated (locally on a MacBookPro) using MODELLER version 9v8 (19), according to templates, 1aym (A16), 1ncr (A16 plus pleconaril) and 1ncq (B14 plus pleconaril). For each run, the complete protomer sequence was aligned against C15 with the “align2d” and heteroatoms (e.g. pleconaril) were accounted for. Each output included five different iterations, which were then compared to the source template with the “discrete optimized protein energy” algorithm “DOPE” (36).

ROBETTA

The C15 VP1 sequence was submitted to the Robetta server (http://robetta.bakerlab.org/) for Ginzu domain prediction using default settings (20). A full 3D prediction was requested.

Model Evaluation

The PDB file for the preferred I-TASSER C15 protomer model was submitted relative to the MODELLER protomer to MAMMOTH (MAtching Molecular Models Obtained from THeory) (21), for plausibility analyses as implemented with the online web site (http://predictioncenter.org). Using standard parameters, the suite aligns input structures, calculates the optimal local similarity of the protein backbone, and then fills in the residues. The outputs are scored based on this alignment (percent of residues aligned, RMSD, E-value, Z-score and –ln(E)). ProQ (23) was also run online (http://www.sbc.su.se/~bjornw/ProQ/ProQ.cgi). It is a neural-network-based program that predicts the quality of a model or determined structure by extracting features (i.e. atom-atom contacts), evaluating their plausibility, and assigning scoring functions LGscore and MaxSub. The C15 VP1 PDB files from I-TASSER, MODELLER and ROBETTA were also compared to each other and to native A16 (1aym) using MAMMOTH-multi (22), in similar procedures (http://ub.cbm.uam.es/servers/mammoth/mammothmult.php). The sterochemical quality of C15 (ITASSER) was also gaged with PROCHECK, part of the PDBSum suite as available online (http://www.ebi.ac.uk/pdbsum/) (24). The coordinates were refined for steric clashes (between and within putative protomers) with CHIRON (37). Protomer and virion coordinates for the full, preferred C15 model are available from VIPERdb (http://viperdb.scripps.edu/) using the accession code: hrvc.

Molecular Graphics

Protomer illustrations were oriented and rendered in MacPyMOL (35). UCSF Chimera (38) created full capsid structures and pentameric assemblies from protomer PDB files. Virion surface “roadmaps” displaying 3D topographies used the Radial Interpretation of Viral Electron Density Maps (RIVEM) program. RIVEM illustrates residues occupying the surface of a virus by reading standard PDB coordinates and plotting them onto stereographic spheres (39). For comparative roadmaps, A16 (1aym), B14 (4rhv) and C15 (model), protomer files were aligned to “standard Rossmann” (29) orientation. The HOH ATOM lines were removed in MacPyMOL. The amino acids were renumbered into standard “picornavirus format” identifying the protein chain and sequence position (e.g. residue 1 of VP1 becomes 1001, residue 1 of VP2 is 2001, etc.). RIVEM (run locally in LINUX CENTOS 5, 32 bit) then created radial coordinate surface maps with matching color scales according to virion radii (Fig 4B). User specified properties such as residue charge, hydrophobicity, chain identity (e.g. Fig 5A,C,E) can be selected within the program. Roadmaps showing sequence conservation used PDB files with % identity embedded in the B-factor column, according to the residue-specific values calculated from the full, refined alignments.

Highlights.

- We have modeled the full capsid structure of rhinovirus C15

- The surface of the RV-C particle is significantly different from other rhinoviruses

- The immunogenicity and receptor binding sites of RV-C15 are modeled