Complete Alanine Scanning of Intersubunit Interfaces in a Foot-and-Mouth Disease Virus Capsid Reveals Critical Contributions of Many Side Chains to Particle Stability and Viral Function*

Spherical virus capsids are large, multimeric protein shells whose assembly and stability depend on the establishment of multiple non-covalent interactions between many polypeptide subunits. In a foot-and-mouth disease virus capsid, 42 amino acid side chains per protomer are involved in noncovalent interactions between pentameric subunits that function as assembly/disassembly intermediates. We have individually truncated to alanine these 42 side chains and assessed their relevance for completion of the virus life cycle and capsid stability. Most mutations provoked a drastic reduction in virus yields. Nearly all of these critical mutations led to virions whose thermal inactivation rates differed from that of the parent virus, and many affected also early steps in the viral cycle. Rapid selection of genotypic revertants or variants with forward or compensatory mutations that restored viability was occasionally detected. The results with this model virus indicate the following. (i) Most of the residues at the interfaces between capsid subunits are critically important for viral function, in part but not exclusively because of their involvement in intersubunit recognition. Each hydrogen bond and salt bridge buried at the subunit interfaces may be important for capsid stability. (ii) New mutations able to restore viability may arise frequently at the subunit interfaces during virus evolution. (iii) A few interfacial side chains are functionally tolerant to truncation and may provide adequate mutation sites for the engineering of a thermostable capsid, potentially useful as an improved vaccine.

Assembly and stability of the multimeric proteins that constitute icosahedral virus capsids depend on the occurrence of multiple non-covalent interactions between many polypeptide subunits (1,2). Compared with small oligomeric proteins, the adaptive structural solutions for subunit recognition and stability may be uniquely limited in virus capsids. For example, the many repetitive intersubunit interactions must be strong enough to provide stability against dissociation by extracellular agents, although not so strong as to impair the mechanism of intracellular uncoating. Viral capsids thus provide good models for understanding protein-protein recognition evolved under stringent, even conflicting, selective constraints (3)(4)(5)(6)(7).
Structural studies of many viruses (see Refs. 1-5 and 8 -10 for some reviews and Ref. 11 for the VIPER data base of capsid structures) and structure-function analyses involving mutation of specific capsid residues (see Refs. 12-24 for some representative studies) have greatly contributed to our current understanding of the assembly, disassembly, and stability of viruses. For small protein oligomers, comprehensive alanine-scanning mutagenesis analyses (25) on the role of interfacial residues in association (26 -30) have been carried out, and a computational analysis of protein-protein interactions in icosahedral viruses has been also described (31). However, to our knowledge no systematic mutagenesis approach to experimentally dissect the contribution of each residue in the subunit interfaces to the stability or functionality of a virus capsid or any other large protein assembly had been attempted to date.
Foot-and-mouth disease virus (FMDV), 1 a picornavirus for which much structural and functional information is available, provides a simple model for the dissection of the molecular determinants of assembly, stability, and disassembly of viral capsids and other large protein complexes. Moreover, FMDV causes the economically most important disease of farm animals (32). The recent epizootics in the United Kingdom, with estimated losses that could exceed £10 billion in that country alone (33), have provided a stark remainder within the European Union of the devastating economic and social damage FMDV may cause worldwide. One important problem of the current FMD vaccines, based on chemically inactivated virions, is their very low stability. Vaccines based on particles with improved thermostability would be highly desirable in the prevention of FMD and other viral diseases.
Assembly of the picornaviral capsid proceeds in several steps (34). The capsid proteins VP0 (1AB), VP3 (1C), and VP1 (1D) are translated as a polyprotein precursor (P1), may fold cotranslationally (2), and are proteolytically processed to yield the mature protomer. Five protomers are assembled to form a pentameric intermediate, and finally, 12 pentamers are assembled to form the icosahedral capsid (Fig. 1). After encapsidation of the RNA genome most VP0 molecules are processed to give VP4 (1A, the N terminus of VP0) and VP2 (1B). Disassembly of the FMDV virion in vivo begins with its dissociation into pentamers (35) by acidification in the endosomes (36). Mild heating of FMDV virions also leads to irreversible dissociation into stable pentamers (34), an event that appears as the main cause for the need of a cold chain to preserve FMD vaccines. Analysis of the crystal structure of the FMDV capsid (37)(38)(39)(40)(41) indicates that the pentameric intermediate subunits interact mainly through a relatively limited number of electrostatic interactions; a role of His-142 of VP3 in the acid-induced disassembly of FMDV has already been demonstrated (20). In the present study, the functional importance of each amino acid side chain involved in noncovalent interactions between subunits in a virus protein shell has been tested by alanine-scanning mutagenesis of the FMDV capsid. The results reveal severe intolerance to variation due to the involvement of many interfacial residues in capsid stability and in other steps of the virus life cycle, and identify those tolerant residues that could be replaced for the rational engineering of a thermostable capsid.

EXPERIMENTAL PROCEDURES
Viruses and Plasmids-FMDV C-S8c1 is a plaque-purified derivative of serotype C isolate C 1 SantaPau-Sp70 (42). Plasmids pO1K/C-S8c1, pO1K/⌬3242, and p3242/C-S8c1 have been described (43). Use of a cDNA clone entirely derived from a type C virus had been hampered by its low infectivity. Instead, pO1K/C-S8c1 was derived from an infectious cDNA clone obtained by Beck and co-workers (44) from FMDV O 1 Kaufbeuren. pO1K/C-S8c1 contains a cDNA copy of an infectious chimeric FMDV genome coding for the capsid proteins and protease 2A of C-S8c1, a chimeric 2B protein and all of the nonstructural proteins (2A and 2B excepted) of O 1 Kaufbeuren. Plasmid pO1K/⌬3242 corresponds to pO1K/C-S8c1 with a segment of 3242 bp (between the two NgoMIV restriction sites) deleted. This segment includes the entire region coding for the capsid proteins of FMDV C-S8c1. Plasmid p3242/C-S8c1 contains this same 3242-bp segment inserted in a vector derived from plasmid pGEM-5Zf(ϩ) (43).
Site-directed Mutagenesis, Subcloning, and DNA Sequencing-Sitedirected mutagenesis of FMDV capsid residues was carried out on plasmid p3242/C-S8c1 by the inverse PCR method using the QuikChange system (Stratagene) and pairs of 25-to 36-mer oligonucleotides. The mutations were checked by dideoxynucleotide-based automated sequencing. No second site mutations were found. The mutagenized 3242-bp NgoMIV segments were subcloned in pO1K/⌬3242 to obtain pO1K/C-S8c1 plasmids including the chosen mutations. As a control, a non-mutated 3242-bp NgoMIV segment derived from the original p3242/C-S8c1 plasmid was subjected to the same steps. The entire subcloned 3242-bp segment of a subset of the pO1K/C-S8c1 mutant plasmids, of non-mutated pO1K/C-S8c1, and of a revertant plasmid (obtained by back-mutating Ala-3146 to the original residue) was sequenced. Again no second site mutations were found.

Transcription of Viral RNA and Electroporation of Eukaryotic
Cells-pO1K/C-S8c1 and the mutant plasmids were linearized by digestion with HpaI, purified, and dissolved in RNase-free water. Infectious FMDV RNA was transcribed from the linearized plasmids by using the Riboprobe in vitro transcription system (Promega). The mixture contained 40 mM Hepes, pH 7.7, 6 mM magnesium acetate, 2 mM spermidine, 10 mM dithiothreitol, 1 unit/l RNasin, 2.5 mM each of ribonucleoside triphosphates, 25 ng/l linearized plasmid DNA, and 0.5 unit/l SP6 RNA polymerase and was incubated for 2 h at 37°C. The RNA concentration was estimated by agarose gel electrophoresis. About 2-4 ϫ 10 6 BHK-21 cells from 60 to 80% confluent monolayers were resuspended in 0.5 ml of electroporation buffer containing 21 mM Hepes, pH 7.05, 137 mM NaCl, 5 mM KCl, 0.7 mM Na 2 HPO 4 , 6 mM glucose, and transferred to a Gene Pulser Cuvette (Bio-Rad) with a 0.4-cm electrode gap. FMDV RNA (3 g) was added and electroporated using a Bio-Rad Gene Pulser set at 280 V, 250 microfarads, 400 ohm (two pulses). The cells were plated in a P60 Petri dish containing 4 ml of Dulbecco's modified Eagle's medium ϩ 10% fetal calf serum and incubated at 37°C. At 4 h post-transfection the medium was discarded, and 2 ml of fresh Dulbecco's modified Eagle's medium ϩ 2% fetal calf serum was added. The incubation was continued for up to 90 h, with aliquots removed at specified intervals. The progeny virus suspension was clarified by centrifugation, and aliquots were stored at Ϫ70°C. For nearly every transfection experiment, RNA was used immediately after transcription. Use of the same RNA either freshly prepared or frozen and thawed yielded very similar results. The same amount of RNA was used for every mutant. In each experiment the same amount of parent FMDV RNA, and no RNA, was used as a control.
Titration and Amplification of Viruses and Extraction of Viral RNA-Virus titers were determined in plaque assays. Mutant virions obtained at very low titers were amplified by a single passage in BHK-21 cell monolayers at the highest possible multiplicity of infection. The progeny virus suspension was collected immediately after the cytopathic effect was observed, clarified, aliquoted, and frozen at Ϫ70°C. RNA derived from virions collected from the supernatants of transfected or infected cell cultures was extracted using Trizol (Invitrogen) and precipitated with ethanol. The RNA was reverse-transcribed to DNA and amplified by reverse transcriptase-PCR as described previously (46), and the relevant segments were sequenced.
Radioactive Labeling and Purification of Virions-FMDV virions were metabolically labeled with [ 35 S]methionine (Redivue Pro-mix, Amersham Biosciences) and purified by sedimentation through a sucrose The icosahedral symmetry axes are indicated (5ϫ, pentagon; 3ϫ, triangle; 2ϫ, oval). B, structure of a pentamer in the capsid. One biological protomer is represented as a black ribbon model, and its 42 amino acid residues involved in interactions with neighboring pentamers (Table I) are represented as gray space filling models. The remaining four protomers are represented as gray backbone models.
cushion followed by sucrose gradient centrifugation, essentially as described (47). The fractions containing full virions (sedimentation coefficient 140 S) were extensively dialyzed against PBS and clarified by centrifugation. The purified virions were free of contaminants, as judged by overloaded SDS-urea-PAGE and Coomassie Blue staining.
Capsid Dissociation Assays and Measurement of Rate Constants-Aliquots (0.3 ml) of 35 S-labeled, purified virus were incubated at a constant temperature for different amounts of time, transferred to ice, loaded in 10 -30% sucrose density gradients, and centrifuged at 4°C in an SW40 rotor (Beckman Instruments) at 18,000 rpm for 18 h. The gradients were fractionated in 0.5-ml aliquots, and the radioactivity was determined using a liquid scintillation counter. In a second type of assay, virus suspensions obtained by transfection of infectious mutant RNA were diluted in Dulbecco's modified Eagle's medium ϩ 2% fetal calf serum to a concentration of about 1000 plaque-forming units/ml. 100-l aliquots in thin walled PCR tubes were incubated at a given temperature (42°C unless indicated otherwise) for different amounts of time, and the remaining virus titers were determined in plaque assays. Non-mutated C-S8c1 virus obtained in parallel transfection experiments was used in each heat-inactivation experiment as a positive control. The experimental data were fitted to a first-order exponential decay by using the program Kaleidagraph (Abelbeck Software) and Equation 1, where C is the virus titer at time t; C 0 is the virus titer at t ϭ 0; and k is the dissociation rate constant. Structural Analyses of Capsid Subunit Interfaces-A silicon graphics work station and the Protein Data Bank coordinates for FMDVs of serotypes C (38) and O (37, 39) were used. The crystallographic models were analyzed using the programs InsightII (Biosym Technologies) and RasMol (48). Contact and solvent accessibility and modeling of mutations were done with the program Whatif (49) using the coordinates of all possible pairs of contacting subunits with different symmetry within the capsid.

Analysis of Interpentameric Interactions within the FMDV
Capsid-In the FMDV capsid, interpentamer interfaces involve essentially VP2-VP3 and VP2-VP2 pairwise contacts (37-41) (Fig. 1). In addition, the N termini of three symmetryrelated VP2 from different pentamers form a short ␤-annulus around each 3-fold axis (38). The low electronic density associated with this annulus prevented the identification of the few side chains of this element, if any, involved in interactions. The structure of homologous FMDV isolates of three different serotypes, O 1 BFS (37, 41), which has been solved to very high crystallographic resolution (1.9 Å), C-S8c1 (38) and A22 (40), revealed high conformational similarity and many conserved interfacial residues and interactions and allowed meaningful contact analyses in the structure of our model FMDV virus (C-S8c1) (38). Within the cut-off distances imposed (Table I), a total of 61 residues per capsid protomer appear involved in direct interpentamer interactions. The side chains of 42 of those residues are involved, beyond C␤, in such interactions and constitute the interpentamer interface accessible to mutation (Table I and Fig. 1B). Eighteen of these side chains (only 8 being nonpolar) participate exclusively in van der Waals interactions, with an average of 2.8 contacts per residue. Very few (0 -2) of these are carbon-carbon contacts (the only exceptions being residues Tyr-2200, Gln-3071, and Thr-2022 with 11, 5, and 3 carbon-carbon contacts, respectively). The remaining 24 polar side chains per protomer at the interfaces are involved in many interpentamer polar interactions. These include, per protomer subunit, a total of 18 different side chain-main chain or side chain-side chain hydrogen bonds, 2 strict salt bridges, 6 medium range charge-charge interactions, and a short range charge-helix dipole interaction (Table I). Most of these 24 side chains participate also in a variable but generally small numbers of van der Waals contacts.
In summary, apart from a limited number of main chain-main chain interactions, essentially all of the direct interpentamer interactions in the FMDV C-S8c1 capsid could be eliminated by truncation to alanine of 42 side chains per protomer that define the pentamer interfaces and that are involved mainly in polar interactions.

Effects of the Truncation of the Side Chains Involved in Interpentameric
Interactions on FMDV Infectivity-Each of the 42 interfacial side chains found involved in interpentameric interactions in FMDV C-S8c1 (Table I) was individually truncated to Ala to analyze first its individual role on the completion of the viral life cycle and the production of infectious particles. Mutation to Ala eliminated the targeted side chain beyond C␤ and disrupted any interaction that could involve that side chain, without introducing new interactions and with the lowest probability of altering the conformation of the polypeptide backbone (25,50). The effect of the introduced TABLE I Side chains involved in non-covalent interactions between pentamer subunits in the FMDV C-S8c1 capsid The cut-off distance for van der Waals contacts was set 0.5 Å higher than the sum of the van der Waals radii of the two atoms considered. For hydrogen bonds and salt bridges, the cut-off distance was 3.5 Å. For medium range charge-charge interactions the distance between charged groups is indicated. Ct, C terminus; sc, side chain; mc, main chain.

Residue a
Charge-charge interactions Hydrogen bonds van der Waals contacts a For each residue, the first digit indicates the protein (either VP2 or VP3) and the last three digits the amino acid position according to Ref. 38. mutations on the infectivity of the viral RNA was tested by electroporation of susceptible cells and analysis of the viral progeny (Table II). The non-mutated viral RNA was highly infectious as expected, with complete cytopathic effect being observed at about 35-40 h post-transfection and an average virus titer of 1.6 ϫ 10 6 plaque-forming units per ml. This titer was obtained with samples collected at 45 h and remained more or less constant up to at least 90 h post-transfection. As a test for the complete mutagenesis step, one of the mutant genomes obtained (E3146A) was back-mutated to wild type (A3146E). As expected, the cytopathic effect, virus titer, and plaque size reverted to normal (Table II).
The 42 mutants could be classified into four groups (Table  II). Group I (38% of the mutants) included viruses that could not be recovered at 45 h post-transfection or, in most cases, at longer incubation times. Group II (17%) included mutants that  a BHK cells were electroporated with wild-type or mutant viral RNA as described under "Experimental Procedures." As a measure of the reproducibility of the experiments (including both transfection and titration), the parent (control) viral RNA was independently transfected 13 times. The titer determined was similar every time (within an order of magnitude, the variation mainly due to the variable plating efficiency, which depends on the metabolic state of the cultured cells used). The average titer is indicated. Every transfection experiment with mutants included a parallel transfection with the wild type, which was also titrated, as a positive control. In addition, for a sample of five group I mutants, the transfection was repeated two or three times and yielded the same result every time. Virus titers are given at the indicated times (in hours post-transfection (h p.t.)). The range of titers for mutants in each group is also indicated.
b Viruses that yielded plaques with average sizes of 3-5 or 1-3 mm were considered to have a large or small plaque phenotype, respectively. c C indicates the residue was conserved in all 23 FMDV isolates from 6 serotypes compared. d A weak delocalized cytopathic effect was observed for mutant E2011A. e Every viral progeny population from group I and those populations from groups I-III that reached high titers at long incubation times were sequenced (indicated in parentheses): (mutant), the site-directed mutation was preserved, and no other mutations were found; (revertant), the revertant (wild type) sequence became dominant; (forw.mutation), a different mutation at the same position was fixed; (comp.mutation), the site-directed mutation was preserved but a second mutation at a different position at the interpentameric interface occurred. yielded drastically reduced titers at 45 h (about 0.01-0.1% that obtained with the non-mutated genome). These titers increased to different extents in samples taken at longer times (up to 90 h). Group III (14%) was formed by mutants that yielded highly reduced titers at 45 h (about 0.1% to less than 10% that of the wild type) and did not recover at longer times. The remaining mutants yielded titers about 10 -100% that of the parent virus. In addition, in most cases the plaques of the progeny virus obtained were clearly reduced in size compared with that of the non-mutated virus ( Fig. 2 and Table II). Thus, a remarkable majority of single mutations at the interpentamer interfaces in the viral capsid had a drastic negative effect on the viral life cycle and virus yield.

Fixation of Genetic Reversions and Forward Mutations in Viral Progeny
Populations-Although several mutants yielded nearly normal titers, this could be the result of reversion rather than tolerance to a specific mutation. Thus, RNA from viral populations recovered at 45 h after transfection with any of those mutants was sequenced. In 8 of 13 cases the consensus sequence of the population, as revealed by the very clean densitograms, showed the introduced mutation was still dominant, with no detected sequence heterogeneity. However, in the remaining 5 cases other mutations had been fixed in the population (Table II). Such mutations included the following. (i) Three cases of genotypic reversion. In one of these cases (E3146A) the densitograms showed a double band at the mutated nucleotide position that indicated the population still contained a sizable fraction of the original mutant. In the two others (E2213A and T3190A), only the revertant genotype was apparent. (ii) Two cases of forward mutations at the position which had been originally mutated: in H2021A and K2063A mutations to Thr or Val had been fixed, respectively. These five site-directed mutants probably had a strongly reduced fitness and were easily outcompeted by revertants or other variants in the population. This left only 19% of the residues at the interpentamer interfaces that are tolerant to mutation (group IV, Table II). Two group I mutants recovered high titers after a long posttransfection incubation time. Sequencing of the global population revealed in one case (N2114A) a heterogeneous mixture of the original Ala mutant and two different variants with forward mutations (Ser and Thr) at the same position. In the second case (N2202A) the population was a mixture of the original Ala mutant and of a double mutant that preserved the original Ala mutation but that included also a second mutation in residue Lys-2063 (to Ile), also at the interpentamer interface, which may phenotypically compensate for the deleterious effect of the original substitution. Interestingly, at least some of the forward mutations that restored viability (N2114T and H2021T) introduced residues that were found in other FMDV field isolates of the widely divergent serotypes SAT2 and SAT3.
The conservation in FMDV of the residues at the pentamer interfaces was evaluated using an alignment of the capsid protein sequences corresponding to 23 isolates from six of the seven FMDV serotypes (Table II). The degree of conservation of the critical or tolerant interfacial residues was 59 or 37%, respectively, numbers that are clearly above or below the average conservation when all VP2 ϩ VP3 residues were considered (45%). However, several interfacial residues critical for the viability of C-S8c1 were not conserved, and several of the substituted residues may not conserve the corresponding interpentamer interactions (see under "Discussion").

Effects of the Truncation of Side Chains Involved in Interpentameric Interactions on the Thermal Stability of Virions-
The kinetic stability of the FMDV C-S8c1 capsid against its irreversible dissociation by heat was first analyzed by sedimentation in sucrose gradients using radioactively labeled virions. Non-heated purified virus yielded a peak with a sedimentation coefficient of about 140 S, corresponding to full virions, and another peak of about 12 S corresponding to free pentamers (due to some dissociation during storage of the virus at 4°C). Heating to a constant, moderate temperature for increasing amounts of time provoked a sharp decrease of the 140 S peak and the simultaneous increase of the 12 S peak. Disappearance of the 140 S peak could be well fitted to a single first-order exponential, consistent with the simple dissociation of virions into pentamers (Fig. 3A). The dissociation rate constant (k off ) of the purified virion at 45°C in PBS was 0.10 min Ϫ1 .
Because the low viability of most of the mutant virions prevented their production and purification to sufficiently high yields, a more simple approach to test the stability of virions, which required only minute amounts of non-purified virus, was validated. Unlabeled virus samples were heated at a moderate temperature (42°C), and the infectious virions remaining were quantitated in plaque assays. The number of infectious particles with incubation time could be well fitted to a single exponential (Fig. 3B), consistent with the first-order dissociation of the capsid into pentamers, and paralleled that observed when dissociation was directly tested in sucrose gradients. As a further control, the dissociation process for the non-mutated virus and one randomly chosen mutant was also tested at a higher temperature (50°C). As expected, the dissociation rate constant increased, but the ratio between dissociation constants was similar (not shown), which is consistent with the same process being observed at both temperatures. For viable viruses of group IV, including some that still preserved the sitedirected mutation and others that had reverted or forward mutated, the rate constants obtained were not significantly different from the wild-type rate, as expected (Table III). The apparent k off obtained for E3146A was slightly higher, but this may be a consequence of this population being a mixture of mutant and revertant genotypes. In contrast, the k off values obtained for all but one of the tested viruses from groups I-III clearly differed from that of the non-mutated virus, consistent with these interpentamer mutations affecting the stability of the capsid. D3069A yielded a rate constant similar to that of wild type, but sequencing of the amplified population showed it had genotypically reverted. Remarkably, three of the mutants with reduced viability (R2018A, T2110A, and L2051A) yielded k off values substantially lower than the wild type indicating higher and not lower stability (Table III). An attractive hypoth-  esis (under study) is that the increased stability may impair uncoating, thus reducing viral yield.

Effects of Mutations in Viral Steps Prior to Protomer Processing and Pentamer
Assembly-The substantial effects on virus yield of some of the interfacial mutations remained difficult to explain by invoking just a difference in particle stability. For example, mutations I2014A or Y2098A did not allow recovery of progeny virus, yet they participated in just one interpentamer van der Waals contact each. This suggested that additional effects in other steps of the viral life cycle could be also occurring. As a first approach to test this possibility, immunofluorescence analyses were carried out on viral RNA-transfected cells (Fig. 4 and Table III). The neutralizing mAb used (5C4) recognizes a discontinuous epitope present on the virion surface (38) and on the unassembled capsid protomer (51) but not on the isolated capsid proteins. A reduced fluorescence intensity would indicate that some step prior to the proteolytic processing of the capsid protomer and the assembly of pentamers could be impaired. As expected, mutants Q2057A and K2096A from group IV (no substantially reduced virus yields) showed normal accumulation of protomers. Some viruses with mutations of residues involved in infectivity and capsid stability (T2023A, D3148A, R2018A, T2026A) showed no intracellular defect, at least prior to the late assembly steps. However, many mutants showed clearly reduced levels of capsid or capsid subassemblies, and fluorescence was observed to accumulate in defined cytoplasmic locations ( Fig. 4 and Table III). Several residues, including some whose mutation led to no progeny virus at all (like Ile-2014 and Tyr-2098), or their coding nucleotides, appeared thus involved in viral function(s) previous to protomer and pentamer assembly. Other interfacial residues (e.g. Lys-3193) may participate in both capsid stability and additional undefined viral function(s) (see under "Discussion").

DISCUSSION
The results obtained demonstrate that in this model virus capsid over 80% of the residues involved in intersubunit (pentamer) interactions are critically important for completion of the virus life cycle. These drastic phenotypic effects appear at least partly related to altered capsid assembly and/or stability against dissociation. The results showed that nearly any mutation tested among those which significantly reduced infectivity, affected also the stability of the virion. In addition to this direct evidence, several other observations are consistent with capsid assembly and/or stability being affected by most of the interfacial mutations. The interfacial residues belonging to any of groups I-III do not cluster at any specific location, as could be expected if these mutations affected any localized function, but are scattered all over the large interpentamer interfaces (Fig. 5). Also, if disruption of an interpentamer interaction is involved in the observed phenotypic effect, it could be expected that mutation of any of the two interacting residues will have similarly critical effects. Such prediction was generally found correct.
Surface accessibility analysis showed that only three (about one-third) of the non-critical side chains were buried. Noncritical residues were not involved in salt bridges and contributed to just one hydrogen bond (Thr-2053 to Gln-2057), which is fully exposed to solvent on the capsid internal surface, and only one of the tolerant side chains participated in more than two interpentamer van der Waals contacts. In contrast, close to three-fourths of the functionally important interfacial side chains were essentially buried upon assembly. 17 of them (52%) were involved in interpentamer salt bridges and/or hydrogen bonds, and 8 others (24%) participated in multiple van der Waals contacts and/or could be involved in medium range charge-charge interactions (compare Tables I and II). Truncation of any residue involved in interpentamer salt bridges had the most critical effects and led in several cases to genotypic reversion. Truncation of any residue involved in buried interpentamer hydrogen bonds led to a thousand-fold or higher reduction in titer. The above observations are consistent with capsid assembly and stability being dependent in part on the hydrophobic effect and multiple van der Waals contacts between subunits, but mainly on each hydrogen bond and salt bridge buried at the pentamer interfaces.
The individual contribution of interfacial residues to intersubunit association energies have been theoretically predicted for some virus capsids and the results are available at the VIPER website (11). Our computational approach identified all contact residues, defined as those for which at least one atom is within the cut-off distance from an atom in a neighboring pentamer. Of these residues, only those whose side chain atoms beyond C␤ are involved in contacts (Table I) are, by definition, amenable to mutational analysis by alanine scanning to determine the specific contributions of their side chains. On the other hand, the contact residues included in the VIPER anal-  Table III).
yses were those whose C␤ is within cut-off distance from the C␤ of a residue in a neighboring subunit. Also, the VIPER prediction of association energies considered the complete residue, not just the side chain, and for FMDV it was carried out with a strain of a different serotype (O 1 BFS). Despite these differences, most residues at the interpentamer interfaces were conserved in both strains and were included both in VIPERЈs analysis and in our analysis. For these interfacial residues, a comparison of the predictions of the VIPER site with our experimental phenotypic study has revealed that mutation of the residues predicted as energetically important (association energy between Ϫ1.3 and Ϫ3.5 kcal/mol) led in most cases (61%) to drastic phenotypic alterations, whereas only 28% of them led to minor or no phenotypic effects (groups III and IV). In contrast, mutation of the residues predicted as energetically less important (association energy between 0 and Ϫ1.0 kcal/mol) led to the most drastic phenotypic alterations in a substantially lower, although still remarkable, number of cases (38%), whereas the proportion of mutations with minor or no effects on the phenotype was raised to 38%.
It may seem surprising that such a large proportion of interfacial residues, many of them involved in interactions that could be considered energetically important for protein-protein association but some of them less so, may individually contribute to capsid assembly and stability. One possibility regards the added effects of the many identical interactions present in a virus capsid (in the case of FMDV, each interaction has a multiplicity of 60), and cooperativity effects may be also important. A theoretical equilibrium thermodynamic model by Zlotnick (52) predicted that loss of even one contact with a small binding energy may cause a significant change in the stability of an icosahedral virus capsid, because of the large number of protomers involved. Our experimental results are consistent with this prediction. A second, non-exclusive possibility regards the fact that a number of the interfacial residues tested (or their coding nucleotides) are also involved in steps of the life cycle prior to the assembly of protomers and pentamers. The precise steps affected have not been identified yet. A defect in viral entry or productive uncoating is unlikely, because the cells were originally transfected and, furthermore, the mutated residues were located away from the natural receptor-binding site. All of the capsid coding region could be deleted without any apparent effect on the ability of the RNA to replicate (53). No involvement of capsid sequences in the control of translation of FMDV proteins has been found either. The position of the mutated residues and the putative structure of the unprocessed picornaviral protomer (2) make their participation in polyprotein processing or RNA-protein interactions also unlikely. Few of the side chains analyzed were involved in substantial intraprotein, intraprotomer, or intrapentamer interactions. Thus, putative defects in protein folding and protomer or pentamer assembly would appear difficult to invoke as a general explanation for the results obtained. Further experiments are required for the identification of these additional defects, which could be different for each mutation. Taken together, the results indicate that both an effect on capsid assembly and/or stability and, in many cases, further effect(s) on steps of the viral life cycle previous to pentamer assembly account for the functional importance of the vast majority of side chains involved in intersubunit contacts.
The observed intolerance to individual truncation of most interfacial side chains in the FMDV capsid is consistent with a degree of conservation of critical residues in natural isolates that was found higher than average. However, a number of the residues identified as critical were not conserved in FMDV. Replication of the mutant viral RNAs in cultured cells led not only to simple reversion events but also to the fixation of new amino acid substitutions that restored the infectious phenotype, and that in some cases had been found in distant field isolates of FMDV. Although further study is required, these a The viral populations used in inactivation assays were sequenced to confirm the presence of the original mutation or reveal the fixation of possible reversions/forward mutations during the amplification step. It was found that for N2114A, K2063A, D3069A, and E3146A, other mutations had become dominant in the populations used in these assays. In N2114A, a heterogeneous mixture of N2114S ϩ N2114T was found; in K2063A, the mutant K2063V had become dominant; and populations D3069A and E3146A had reverted to wild type at these positions. NA, not applicable (no virus could be recovered from transfection assays).
b The viruses were subjected to thermal inactivation at a constant temperature (42°C) for different amounts of time, and the kinetic rate constant (k off ) of the first-order inactivation process was determined by fitting the data to a single exponential, as described under "Experimental Procedures" (see also Fig. 3). The inverse of the inactivation rate constant (1/k off ) is indicated. As a measure of the reproducibility of the assay, the average and S.D. of the 1/k off value obtained from 10 independent determinations is given for the parent virus. NA, not applicable (no virus could be recovered from transfection assays).
c The genomic RNAs from the mutant viruses were electroporated in BHK cells, and the capsid viral protein that had been expressed and assembled (as protomers, pentamers or capsids) was determined by in situ immunofluorescence as described under "Experimental Procedures." As a measure of the reproducibility of the experiment, the parent (control) virus was independently transfected 6 times. The immunofluorescence signal obtained was very similar every time. In addition, every transfection/immunofluorescence experiment with mutants included the wild type as a positive control. For a sample of 4 mutants the transfection was repeated 2-4 times and yielded the same result every time. The fluorescence intensity and subcellular localization (see Fig. 4) were categorized as follows: ϩϩ, fluorescence intensity and cellular localization (diffuse throughout the cytoplasm) similar to wt; ϩ, clearly reduced intensity, and localization in small cytoplasmic clusters; Ϯ, intensity just above background, and localization in a few cytoplasmic clusters; Ϫ, signal not significantly different from the negative control (background). ND, not determined. d These mutants did not yield virions and could not be used in inactivation assays. e Previous to the inactivation experiment, these mutant populations were amplified through a single passage to obtain sufficient amounts of virions. All other mutants were used without amplification. observations point to the frequent selection during virus evolution of compensatory mutations at the pentamer interfaces. These would provide the virus with an adaptive solution to preserve assembly-competent protein subunits and stable capsids, even when simple reversion would be difficult because of the accumulation of two or more point mutations (see under "Results"). It may also allow the satisfaction of severe structure-function constraints (3, 6) without compromising the adaptive rapid evolution of RNA viruses (54).
From a biotechnological point of view, the side chains at the subunit interfaces of the FMDV capsid that could be considered functionally less relevant, and thus more adequate for introducing potentially stabilizing interactions through mutation, have been now identified (Table II and Fig. 5). In addition, three alanine mutants were found already more stable than the natural virus. The results provide hope that directed evolution or the rational engineering of just one additional buried salt bridge or hydrogen bond (per protomer) through mutation of some of the few tolerant positions identified may also increase capsid stability. An FMD vaccine based on a chemically inac-  (38). VP1, VP2, and VP3 are respectively colored cyan, yellow, or magenta. The capsid residues involved in interpentamer interactions are represented as space filling models and color-coded according to their effect in the virus yield. Group I (the most critical residues) is colored red, group II orange, group III yellow, and group IV (tolerant residues) green. Residues for which rapid genotypic reversion or forward mutations occurred are colored pink. C, wire frame model of a pentameric subunit in the FMDV C-S8c1 capsid inclined about 50°from the horizontal and viewed from below (i.e. from the virion interior). Residues tolerant to truncation (group IV) and intolerant residues (all except group IV) are represented as space filling models and respectively colored green or red.

FIG. 4.
In situ immunofluorescence detection of viral capsid assemblies in cultured cells. BHK cells were electroporated with RNA from non-mutated virus, with no RNA, or with RNA from the mutants in Table III, as indicated, and formation of capsids or subassemblies was determined using a monoclonal antibody against a discontinuous epitope present in the FMDV capsid. For an interpretation of the results obtained, see Table III and under "Results." tivated virus with such engineered mutations may relax the requirements for a cold chain, without a need to change the well established methods of FMD vaccine production and testing. The capacity of eliciting a broad immune response by the complete viral particles would be preserved, as the several antigenic sites identified in FMDV do not overlap with interpentamer interfaces (6). The use of empty capsids has been considered in the development of new vaccines to avoid the risk of handling infectious virions. Unfortunately, empty capsids of FMDV and other viruses may be less stable than full virions (55), perhaps because of missing nucleic acid-capsid interactions. The engineering of a capsid with improved stability may allow for the compensation for such effects.