Electrostatic interactions at the five-fold axis alter heparin-binding phenotype and drive EV-A71 virulence in mice

Enterovirus A71 (EV-A71) causes hand, foot and mouth disease epidemics with neurological complications and fatalities. However, the neuropathogenesis of EV-A71 remains poorly understood. In mice, adaptation and virulence determinants have been mapped to mutations at VP1-145, VP1-244 and VP2-149. We hypothesized that heparin-binding phenotype shapes EV-A71 virulence in mice. We constructed six viruses with varying residues at VP1-98, VP1-145 (which are both heparin-binding determinants) and VP2-149 (based on the wild type 98E/145Q/149K, termed EQK) to generate KQK, KEK, EEK, EEI and EQI variants. We demonstrated that the weak heparin-binder EEI was highly lethal in mice. The initially strong heparin-binding EQI variant acquired an additional mutation VP1-K244E, which confers weak heparin-binding phenotype resulting in elevated viremia and increased brain inflammation and virus antigens in mice, with subsequent high virulence. EEI and EQI-K244E variants inoculated into mice disseminated efficiently and displayed high viremia. Increasing polymerase fidelity and impairing recombination of EQI attenuated virulence, suggesting the importance of population diversity in EV-A71 pathogenesis in vivo. Combining in silico docking and deep sequencing approaches, we inferred that virus population diversity is shaped by electrostatic interactions at the five-fold axis of the virus surface. Electrostatic surface charges facilitate virus adaptation by generating poor heparin-binding variants for better in vivo dissemination in mice, likely due to reduced adsorption to heparin-rich peripheral tissues, which ultimately results in increased neurovirulence. The dynamic switching from heparin-binding to weak heparin-binding phenotype in vivo explained the neurovirulence of EV-A71. Author summary Enterovirus A71 (EV-A71) is the primary cause of hand, foot and mouth disease, and it can also infect the central nervous system and cause fatal outbreaks in young children. EV-A71 pathogenesis remains elusive. In this study, we demonstrated that EV-A71 variants with strong affinity to heparan sulfate (heparin) have a growth advantage in tissue culture, but are disadvantageous in vivo. When inoculated into mice, strong heparin-binding virus variants are more likely to be adsorbed to peripheral tissues, resulting in impaired ability to disseminate and being cleared from the bloodstream rapidly. The lower viremia level resulted in no neuroinvasion. In contrast, weak heparin-binding variants show greater levels of viremia, dissemination and subsequent neurovirulence in mice. We also provide evidence that the ability of EV-A71 to bind heparin is mediated by electrostatic surface charges due to amino acids on the virus capsid surface. In mice, EV-A71 undergoes adaptive mutation to acquire greater negative surface charges, thus generating new virulent variants with weak heparin-binding which allows greater viral spread. Our study underlines the importance of electrostatic surface charges in shaping EV-A71 virulence.


Introduction
We further investigated the heparin-binding ability of the EV-A71 variants. EV-A71 variants 133 with VP1-E145 (KEK, EEK and EEI) displayed significant reduction of 29.7%, 18.8% and 134 37.1% in heparin-binding, respectively ( Fig 1D). Similar findings were also observed in a 135 heparin inhibition assay. Pre-treated heparin inhibition of these three VP1-E145 variants 136 significantly reduced RD cell viability to 51.5%, 45.5% and 34.3%, respectively (Fig 1E). 137 Based on these heparin-binding results, the EV-A71 variants were categorized into two groups: 138 strong heparin binder (EQK, KQK and EQI) and weak heparin binder (KEK, EEK and EEI). We further asked why the EQI variant, a strong heparin binder, exhibited a relatively high 154 virulence in mice. Viral genomic RNA from the brains and hind limbs of dead EQI-infected 155 mice were harvested for genome sequencing, revealing that the EQI variant had acquired a 156 VP1-K244E mutation (Fig 2B). This mutation, however, was not present in the EEI-infected 157 mice ( Fig 2C). 158 159

Emergence of EQI-K244E variant resulted in abolished heparin-binding ability and 160 regaining of virulence 161
Since VP1-K244 is a key determinant of heparin-binding [36], we speculated that the 162 emergence of VP1-K244E had abolished the heparin-binding ability of the EQI variant. To 163 investigate the role of VP1-K244E mutation in heparin-binding and virulence in mice, we 164 introduced this mutation into the EQI variant through site-directed mutagenesis. The EQI 165 variant with K244E mutation (termed EQI-K244E) however failed to achieve high virus yield 166 in tissue culture for subsequent in vivo experiments (data not shown). We thus collected EQI-167 K244E and EEI from the brain homogenates (indicated with + ) for subsequent experiments, 168 after confirming the sequences using Sanger sequencing. EQI-K244E + displayed significant 169 reduction of heparin-binding compared to clone-derived EQI and EEI, and EEI + (Fig 3A). To 170 determine the association of heparin-binding and in vivo virulence, we then infected one-day 171 old suckling mice with clone-derived EEI and EQI-K244E + by i.p. administration. Brain 172 homogenate from EQI-infected surviving mice was also harvested and used as a negative 173 control (EQI + ; viral RNA not detected in RT-PCR). At day 4 post-infection, 100% mortality 174 was observed in EEI and EQI-K244E + -infected mice but none of the mice succumbed to EQI + 175 infection (Fig 3B). 176 The hind limb and brain samples from EEI and EQI-K244E + -infected mice were then processed 177 for histopathological analysis, and results supported earlier findings. Immunohistochemical 178 (IHC) examination revealed massive localization of viral antigens in both EEI and EQI-179 K244E + -infected muscles, indicating that skeletal muscle is an important replication site (Fig  180   3C). Inflammation and extensive muscle damage were also observed in the haematoxylin and 181 eosin (H&E)-stained sections of muscle. In addition, viral antigens were detected in neurons 182 mainly distributed in the pons and midbrains (Fig 3D). Mononuclear cells infiltrations were 183 also evident in the cortices. In contrast, no distinctive histopathological change was observed 184 in the mock-infected organ samples. 185 186

High lethality of weak heparin-binding EV-A71 variants correlates with high viremia 187
Strong heparin-binding ability confers the advantage of promoting virus attachment on the cell 188 surface, thus increasing the probability of virus-functional receptor interaction in vitro [9]. 189 However, we have demonstrated that a strong heparin-binding phenotype is deleterious to virus 190 pathogenesis in vivo. To unravel the discrepancy of cytopathogenicity in vitro and in vivo 191 virulence, we investigated virus dissemination in mice. Following i.p. infection, five mice were 192 sacrificed for viral load quantitation in brain and hind limbs. EEI and EQI-K244E + variants 193 replicated to higher titers than EQI in both hind limbs and brain, at 2 and 4 days post-infection 194 ( Fig 4A). 195 To investigate the attribution of viremia to in vivo pathogenesis, three-to four-week-old mice 196 were infected intravenously with EQK, EEI, EQI and EQI-K244E + variants. Blood samples 197 were then collected at 5, 15 and 30 min post-inoculation for viral load quantitation. Both strong 198 heparin-binding variants, EQK and EQI showed rapid viral clearance, with approximately 70% 199 cleared from the bloodstream at 30 minutes ( Fig 4B). Only about 23% of EEI had been cleared The emergence of a weak heparin-binding variant with VP1-K244E mutation is the key 208 determinant of in vivo adaptation and pathogenesis of EQI. We hypothesized that EQI is 209 avirulent without the acquisition of the VP1-K244E mutation in vivo. To reduce mutation rates 210 and restrict generation of viral quasispecies, we engineered the viral RNA-dependent RNA 211 polymerase (RdRp) of EQI to harbor previously identified high-fidelity mutations G64R and 212 L123F (abbreviated as HF in Fig 5A) [39-41] and recombination deficient mutation Y276H 213 (labelled as Rec -) [42]. We employed a luciferase-based replicon system to assess the impact 214 of these mutations on genome replication. As demonstrated in Fig 5B, no significant 215 differences in luciferase activities were observed between wild type EV-A71 Nluc Rep, EV-216 A71 Nluc Rep-HF and EV-A71 Nluc Rep-Rec -, suggesting that these mutated RdRp variants 217 replicate as efficiently as the wild type. Next, we generated and rescued the EQI-HF and EQI-218 Recvirus variants. These EQI-HF and EQI-Recvariants were genetically stable with no 219 reversion of mutations and emergence of VP1-244E observed after a few passages, in addition 220 to indistinguishable plaque morphology to EQI (data not shown). 221 To characterize the impact of increased fidelity and recombination deficiency on in vivo 222 virulence, one-day old suckling mice were infected with EQI, EQI-HF and EQI-Rec -. Half of 223 the EQI-infected mice died by day 12 post-infection (Fig 5C), while none of the mice died 224 following EQI-HF and EQI-Recinfection. A reduced ability of the EQI strain to undergo 225 mutations was associated with loss of in vivo virulence, although we were unable to recover 226 EQI-HF and EQI-Recviruses from surviving mice to definitively show that this was due to 227 lack of the VP1-K244E mutation. 228 229

Emergence of VP1-K244E is important for neuroinvasion 230
To determine if emergence of the VP1-K244E mutation is critical for systemic dissemination, 231 we next examined neurovirulence (the ability to directly infect the CNS) of all the EV-A71 232 variants following direct intracerebral inoculation. A dose of 1 × 10 5 PFU of each of the EV-233 A71 variants was intracerebrally injected into separate litters of one-day old mice (Fig 6). 234 Similar mortality rates (100% mortality at day 4 post-infection) were shown by the EEI and 235 EQI-K244E + variants, the former having been earlier shown to be highly lethal following i.p. 236 infection (Fig 2A) suckling mice was intraperitoneally infected with EQI. The mice were sacrificed at day 3, 7 248 and 11 post-infection or when moribund to harvest hind limbs and brains for next-generation 249 sequencing of the virus population diversity. At day 9 post-infection, two moribund mice were 250 collected. We first screened all the harvested samples using RT-PCR. At day 3 post-infection, 251 none of the collected organs were positive for EV-A71 ( Fig 7A). All five muscle samples 252 collected were positive for EV-A71 at day 7 post-infection, suggested that viruses were 253 replicating in skeletal muscles, but brain samples were negative suggesting that limited virus 254 had disseminated to the brain at this time point. As expected, both muscle and brain samples 255 collected from the moribund mice at day 9 post-infection were positive for EV-A71. None of 256 the remaining mice collected at day 11 post-infection were positive for EV-A71 RNA in both 257 hind limbs and brain. 258 Deep sequencing of the EV-A71-positive organ samples revealed that the VP1-Q145 residue 259 was highly stable with > 99% variant frequency ( Fig 7B). The frequency of VP1-Q145E 260 mutation in these samples was lower than 0.1% (labelled with # in Fig 7B). To our surprise, 7 261 dpi-M5 sample displayed a poor sequencing coverage and therefore was eliminated from the 262 analysis. We observed a sequential transition of K244 to E244 from day 7 to day 9 post-263 infection. At day 7 post-infection, VP1-K244 was predominant in two out of four hind limb 264 samples (M1 and M2), while only a single sample (M4) showed VP1-E244 as the dominant 265 viral population. Notably, we identified a novel substitution, VP1-K244T from the hind limb 266 of sample M3 with high frequency of 90%. As infection progressed, VP1-E244 was solely 267 detected in the organs harvested at day 9 post-infection, further affirming the contribution of 268 Weak heparin-binding is due to loss of electrostatic interactions at the five-fold axis 282 EEI was experimentally proven to be highly lethal in mice. However, we observed that EQI 283 selectively acquires VP1-K244E over the VP1-Q145E mutation to gain neuroinvasive and 284 neurovirulent properties (Figs 3B & 6). We reasoned that VP1-E244 exhibits weaker heparin-285 binding ability compared to VP1-E145, and therefore, could be favorably selected in vivo. We 286 employed in silico analysis to characterize the heparin-binding affinity of VP1-E145 and VP1-287 E244. VP1-98, 145 and 244 are located around the five-fold axis of the EV-A71 pentamer (Fig  288   8A). Based on the electrostatic maps, the five-fold axis of the EQI variant is highly positive-289 charged (Fig 8B), implying strong affinity to heparin. With the VP1-Q145E substitution, EEI 290 has lower positive charges at its five-fold axis. Since the VP1-K244 residue is protruding from 291 the surface of the five-fold axis, substitution of a positively-charged lysine residue to a 292 negatively-charged glutamic acid at this position greatly reduces the electrostatic potential on 293 the five-fold axis and changes the capsid conformation. 294 Simulated docking of 12-mer heparin to EV-A71 residues within a 4Å radius revealed no 295 notable change in interaction energies between EQK (-730.76 kcal/mol) and EQI (-703.71 296 kcal/mol) (Fig 8C). Substitution of the VP1-Q145E mutation in EEI resulted in a drop of 32.34 297 kcal/mol interaction energy when compared to EQK. The EQI-K244E variant showed the 298 weakest heparin-binding ability as the interaction energy drastically dropped to -476.32 299 kcal/mol. Compared to the EQI-K244E variant, uncharged EQI-K244T variants exhibited 300 higher interaction energies of -552.21 kcal/mol and -582.89 kcal/mol, respectively. The energy 301 change was mainly contributed by the VP1-244 residue, with a descending energy order of 302 K244, T244 and E244, which correlates with the scale of strong to weak heparin-binding (Fig  303   8D). 304 When examining the heparin docking of non-synonymous variants detected by NGS, the VP1-305 E167G mutation had no effect on interaction energy compared to EQI ( Fig 8E). Interestingly, 306 variants with VP1-L97R and VP1-N104S mutations showed slight increase in the interaction 307 energy of EQI-K244E variants, suggesting an enhanced effect of heparin-binding. Although 308 EQI-K244E-L97R showed very weak binding strength at the VP1-K244E site similar to EQI-309 K244E, a compensatory effect was seen in L97R site which singly contributes to a more 310 negative interaction energy i.e. strong heparin-binding ( Fig 8F). Meanwhile, VP1-N104S 311 mutation has also increased the heparin-binding of EQI-K244E variant. Taken together, 312 different compensatory mutations emerged to overcome the capsid instability and alter the 313 virus fitness. 314

315
As heparin-binding phenotype has been implicated in virulence of some neurotropic viruses, 316 we studied the relationship between heparin-binding phenotype and mouse neurovirulence in 317 EV-A71. Our data highlighted the key role of electrostatic interactions in shaping heparin-318 binding to confer virulence in mice. Among the weak heparin-binding variants used in this 319 study (KEK, EEK and EEI), only EEI was associated with increased virulence and virus fitness 320 in vivo. Strikingly, EQI, which should be a strong heparin binder, showed high virulence, and 321 we showed that this was due to the mutation VP1-K244E, which conferred weak heparin-322 binding. This VP1-K244E mutation has been previously identified as a mouse virulence 323 The mechanism of pathogenesis associated with heparin-binding is driven by electrostatic In summary, we showed that weak heparin-binding EV-A71 is highly virulent in mice, in 449 contrast with strong heparin binders which show higher replication in vitro due to culture 450 adaptation. This study shows that weak heparin-binding EV-A71 is preferentially selected to 451 disseminate via the bloodstream; in contrast, strong heparin-binding EV-A71 is adsorbed to 452 peripheral tissues and rapidly cleared. The electrostatic surface charges at the VP1 capsid shape incubated for 30 minutes at 4°C, and this was followed by centrifugation and 5 washing steps. 502 The heparin-bound viruses were collected after eluted with elution buffer (0.02 M Tris-HCl, 503 2M NaCl, pH 7.4). Both virus input and output fractions were quantitated using real-time PCR, 504 and the virus binding efficiency was normalized by dividing the output viral RNA copies 505 number over the input viral copies number. 506 507

Evaluation of inhibitory effect of soluble heparin on EV-A71 variants 508
To determine the inhibitory effect of soluble heparin on EV-A71 variants, a virus inactivation 509 assay was performed as previously described [9]. In brief, viruses were incubated with 2.5 510 mg/ml of soluble heparin (Sigma, USA) for an hour at 37°C. The treated viruses were 511 inoculated onto pre-seeded RD cells and incubated at 37°C. Two days later, the cell viability 512 of each infected virus variants was measured using CellTiter 96 Aqueous One solution Cell 513 Proliferation Assay (Promega, USA). The relative cell viability was calculated by normalizing 514 the absorbance value of treated virus samples against untreated virus samples, as compared to 515 the mock-infected samples. 516 517

Mice infection experiments 518
Groups of one-day old ICR suckling mice (n= 9 to 12) were obtained from AEU. Each group 519 of suckling mice were either intraperitoneally or intracerebrally inoculated with 1 × 10 5 PFU 520 of each EV-A71 variant or PBS alone. All infected mice were monitored daily for weight 521 change and health status up to 13 days post-infection. A clinical score was recorded using the 522 following grades: 0, healthy; 1, weak or less active; 2, hunched posture and lethargy; 3, one-523 limb paralysis; 4, two-limb paralysis; 5, moribund or dead. Moribund mice were sacrificed and 524 removed along with any mice found dead. Harvested mice organs were homogenized using 525 hard tissue homogenizing mix (Omni International, USA). RNA was extracted from the 526 homogenates with QIAamp viral RNA mini kit (Qiagen, Denmark). The viral loads in organs 527 were determined using TaqMan fast virus 1-step master mix (ABI, USA). One step RT-PCR 528 was also performed to amplify viral RNA from the organs using MyTaq One-Step RT-PCR kit 529 (Bioline, UK) for Sanger sequencing or deep sequencing. Illumina Miseq (Illumina, USA) 530 next-generation sequencing (NGS) was performed with 150 nucleotide paired end reads and 531 average coverage of at least 20,000 reads. The NGS reads were analyzed using CLC Bio 532 Genomic Workbench (Qiagen). 533 For the virus clearance assay, 3-to 4 week old ICR mice (weighing within 25-35g) were i.p. 534 injected with ketamine/xylazine cocktail prior to infection. Anesthetized animals were then 535 intravenously inoculated with 5 × 10 5 PFU of each EV-A71 variant via the tail vein. At certain 536 timepoints, blood was collected from anaesthetized mice through the retro-orbital plexus with 537 the use of a sodium heparinized hematocrit capillary (Hirsschmann, Germany). The collected 538 whole blood was then used for viral RNA quantitation. 539 Immunohistochemistry 541 Immunohistochemistry (IHC) were performed by the standard ENVISION technique as 542 described previously [110]. Briefly, deparaffinised and rehydrated tissue sections were blocked 543 using standard immunoperoxidase procedure before antigen retrieval (30 minutes, 99°C, Tris 544 EDTA buffer with 0.05% Tween-20). Tissues were then incubated with rabbit polyclonal EV-545 A71 VP1 (GeneTex, USA) at 4°C overnight. After washing, tissues were then incubated with 546 goat-anti rabbit HRP-conjugate (Dako, Denmark) for 30 minutes at room temperature. Tissues 547 were stained using DAB (Dako) and counterstained with hematoxylin (Dako). The tissues were 548 mounted using DPX (Dako) prior to examination under a light microscope. The negative 549 control tissues for IHC included mock-infected ICR mice brain and hind-limb muscle tissues. 550 Isotype control antibodies or normal rabbit immunoglobulin fractions (Dako) were also used 551 to exclude non-specific staining. 552 553 Electrostatic surface charge analysis of EV-A71 structure 554 The EV-A71 structure was visualized using Chimera software (UCSF Chimera version 1.13.1, 555 USA). Electrostatic surface potentials of virus capsid were analyzed using the 'Coulombic 556 surface coloring' function in which the capsid residues were labelled with different colors 557 based on their electrostatic charges. Positively-charged residues were colored blue while 558 negatively-charged residues were colored red.                                    One-day old suckling mice (n=9-12) were inoculated with 1 × 10 5 PFU of different EV-A71 variants by i.p. injection. (A ) A litter of mock-infected mice was used as a control group receiving PBS injection. The clinical scores and percentage of survival of the infected mice groups were monitored daily for 13 days. The severity of clinical symptoms was scored as follows: 0, healthy; 1, weak or less active; 2 hunched posture and lethargy; 3, one-limb paralysis; 4, two-limb paralysis; 5, moribund or dead . Significant differences compared to EQK are labelled ** (P < 0.01) and ***P (< 0.001). EQK, KQK and KEK curves are identical to that of the mock-infected group. The VP1 sequence chromatograms of EQI (B) and EEI (C) populations isolated from infected animal organs are shown , highlighting VP1-98, VP1-145 and VP1-244 (note that VP2-149 is not shown). The emergence of E244 virus isolated from brains and limbs of EQI-infected moribund mice are shown.

Fig 3. Characterization of virulent phenotype of EEI and EQI-K244E + variants
(A) EV-A71 variants collected from the brain homogenates of dead infected mice were compared to clone-derived variants for heparin-binding properties. (B) One-day old suckling mice (n=8-12) were infected with EEI, EQI-K244E + and EQI + through i.p. injection. A control group receiving PBS injection was also included. The clinical scores and percentage of survival are shown over 13 days post-infection. Significant differences compared to WT are labelled as * (P < 0.05) and ***P (< 0.001) . Tissue samples of mice which succumbed to EEI and EQI-K244E + infection were subjected to IHC and H&E staining. Virus antigen was seen in muscle cells (C), along with increased inflammatory infiltrate s. In brains (D), antigen-positive neurons were seen in the midbrain (representative image from an EQI -K244E + -infected brain) and pons (representative image from an EEI-infected brain), with severe inflammation in the cortex shown by H&E staining. Magnification for IHC staining: X40; H&E staining: X20. (A) At selected time points, EQI, EEI and EQI-K244E + -infected mice (n=5) were sacrificed and viral loads were determined from harvested hind limbs and brains using qRT-PCR. Significant differences between viral variants are labelled as ** (P < 0.01) and *** (P < 0.001). (B) Virus clearance from blood was quantitated using qRT-PCR following intravenous inoculation of EQK (WT), EEI, EQI or EQI-K244E + into 3-4 week-old mice. The virus titers were collected at selected timepoints up to 30 minutes. Significant differences between EQK and EQI-K244E + are labelled as * (P < 0.05) and ** (P < 0.01).  Three major factors are responsible for EV-A71 virulence determination, namely virus entry, peripheral dissemination and neuroinvasion. (A) Both strong and weak heparin binders infect humans at the same rate, using the same inoculation route and receptor. (B) Viremia is established upon virus entry. Strong heparin binders are more readily removed from the blood circulation by binding to peripheral tissues due to their high affinity to heparin. Meanwhile, weak heparin binders give rise to higher viremia with better dissemination to other organs, resulting in higher peripheral virulence. (C) Neuroinvasion occurs when virus travels from peripheral motor nerves to the CNS via retrograde axonal transport . Chi-square test χ 2 = 40.3558. P < 0.00001 S1 Table. Comparison of EV-A71 isolate sequences of primary specimens and passaged isolates .

Mock
Strong heparin binders (denoted with asterisks) were more frequently identified from sequencing of passaged EV-A71 than from direct sequencing of primary specimens , suggesting that the virus isolates have undergone heparin-binding adaptation in tissue culture (P < 0.00001, chi-square test ).

Primer
Sequence (  Primer sets used for EV-A71 VP1 sequencing and qRT-PCR are shown. which the luminal side represents the blood capillary while the abluminal compartment represents the brain (A). The in vitro model was exposed to different EV-A71 variants with titer of 1 × 10 5 PFU. The BBB permeability induced by EV-A71 variants were assessed in terms of transendothelial electrical resistance (TEER), with a greater reduction of TEER indicating greater permeability of BBB. The TEER was recorded at 2 hours (B) and 6 hours post-exposure (C) along with non-infected cell controls (white bars) and normalized with TEER values measured before virus exposure. Results are presented as mean ± SD (n=6). Significant differences between viral variants and WT (black bars) are labelled as * (P < 0.05) and **(P < 0.01), using the Student's t test.