Coronavirus hemagglutinin-esterase and spike proteins coevolve for functional balance and optimal virion avidity

The receptor-destroying enzyme (RDE) activity of HE facilitates virus release at the end of the infectious cycle but is also important during the very early stages of infection (32, 37). Virions attach to 9-O-Ac-Sia–based receptors via S with fast association and dissociation rates (36). The combination of dynamic S-mediated receptor binding and HE RDE activity results in local receptor depletion, thereby allowing virions to escape from irreversible nonproductive binding to the highly densely clustered decoy receptors in the mucus and glycocalyx. In analogy with influenza A and C viruses (41–44), the repetitive action of binding and catalytic release may translate in virion motility allowing particles to traverse the mucus layer, to pass the glycocalyx, and to reach the entry receptors at the cell surface. We recently demonstrated the importance of HE for infection during (pre)attachment in infection experiments with G-deficient vesicular stomatitis (VSV) particles that were pseudotyped with wild-type BCoV S (pVSV-S). Human rectal tumor 18 (HRT18) cells are abundantly covered with 9-O-Ac-Sias (28) and susceptible to BCoV and OC43. They can also be infected by pVSV-S but the efficiency of infection increases by more than fivefold when enzymatically active, soluble recombinant HE is added to the inoculum (32). Apparently, also in HRT18 cells, the sialoglycans in the glycocalyx act as decoys and their depletion by HE enhances infection.

To study the role and importance of HE during natural β1CoV infection and that of the HE lectin domain especially, we developed a reverse genetics system for BCoV strain Mebus based on targeted RNA recombination (SI Appendix, Fig. S1A) (45, 46). Recombinant wild-type BCoVs (rBCoV) with parental type HE and S, but with accessory ORF4a replaced by the Renilla luciferase gene (rBCoV-Rluc), were readily generated upon seeding acceptor-virus–infected, donor RNA-transfected mouse LR7 cells onto monolayers of feeder HRT18 cells. rBCoV-Rluc arose and within 7 d grew to final titers routinely obtained for wild-type BCoV (∼10⁸ TCID50/mL).

Generating BCoV-Rluc derivatives defective in HE lectin function proved more cumbersome. To abolish the HE RBS, we substituted Phe²¹¹, which is crucial for ligand binding (17) (SI Appendix, Fig. S1B), by Ala. To prevent rapid reversion, this was done by introducing two nucleotide substitutions. Mutant viruses were recovered eventually, but in three of four successful trials, a multistep 160-h rescue did not suffice and an additional 72- to 96-h blind passage was required (Fig. 1A).

Sequence analysis of clonal virus populations, obtained by endpoint dilution, confirmed the presence of the HE Phe²¹¹Ala substitution in all cases. Surprisingly, the purified viruses all suffered single site mutations in S, clustering in domain S1^A (amino acids 15 to 302) in proximity of the RBS (Fig. 1 A–C and SI Appendix, Fig. S2). Two of the trials yielded multiple S variants and some variants (Thr⁸³Ile and Leu⁸⁹Pro) emerged independently in separate experiments (Fig. 1A). The mutations map to three distinct S RBS elements (β1, 3₁₀1, and β5) that are essential for ligand binding (Fig. 1C and SI Appendix, Fig. S2C; nomenclature according to ref. 32). Ile²⁶Ser and Asn²⁷Ser locate in the S1^A β1 element within the N-terminal L1-β1-L2 segment (amino acids 15 to 33) that walls pocket P1 (32, 36). Moreover, in the OC43 S cryo-EM holostructure, the Asn²⁷ side chain hydrogen bonds with the 9-O-acetyl carbonyl (36). Leu⁸⁹Pro in S1^A element 3₁₀1 is immediately adjacent to Trp⁹⁰. The latter is arguably the most critical residue in the RBS as its indole side chain separates the P1 and P2 pockets, and its replacement precludes receptor binding and virus infectivity (32). Finally, Gly⁸²Glu and Thr⁸³Ile substitutions occurred in S1^A element β5 that interacts with the sialate carboxylate through hydrogen bonding with Lys⁸¹ and Thr⁸³.

As measured by solid-phase lectin binding assay (sp-LBA) with S1^A-Fc fusion proteins and bovine submaxillary mucin (BSM) as ligand, all mutations significantly reduced S binding to 9-O-Ac-Sia, albeit to widely different extents. S1^A-Fc binding affinities of the mutants were 500-fold (Asn²⁷Ser) to more than 10,000-fold (Ile²⁶Ser; Gly⁸²Glu) lower than that of parental BCoV S1^A-Fc (Fig. 1D).

Destruction of the HE lectin RBS abolishes HE-mediated virion attachment to 9-O-Ac-Sia receptor determinants and causes a reduction in overall virion binding avidity. However, this is clearly not the main defect. The HE-Phe²¹¹Ala substitution selected for mutations that markedly decreased the affinity of the S RBS, thus reducing overall virion avidity even more. The data can be understood from our earlier observations that loss of HE lectin function also affects HE’s receptor destroying sialate-O-acetylesterase activity (11). The HE lectin domain serves as a regulator of the esterase domain, its function similar to that of the carbohydrate-binding modules that are appended to the enzyme domains of cellular glycoside hydrolases (47). It promotes receptor-destroying esterase activity toward multivalently presented glycotopes, such as present on mucins by keeping the esterase domain in prolonged close proximity of these clustered ligands. In the case of soluble HE (sHE), inactivation of HE lectin RBS decreases esterase activity toward mucin-associated sialoglycans by up to 500-fold. However, in the natural context of the intact virus particle, with spikes and HE molecules closely packed, the effects are more modest. Virion-associated esterase activity is diminished but still substantial because the lectin function of HE is partially taken over by S (11). Nevertheless, the present observations suggest that the destruction of the BCoV HE RBS creates an imbalance between S-mediated virion attachment and HE-mediated receptor destruction that can only be compensated by strongly reducing S affinity. Again, the observations fit a model in which dynamic binding of virions is crucial for infectivity and in which reversibility of attachments confers virion motility thereby enabling virions to dodge decoy receptors and to browse for entry receptors.

To test whether loss of virion-associated RDE activity in rBCoV-HE-F²¹¹A/S^wt/Rluc might be compensated for by adding exogenous soluble HE to the culture medium, we seeded infected/transfected LR7 cells onto HRT18 cell monolayers, supplemented the cell culture supernatant with BCoV HE-Fc (17) to final concentrations of 1 pg to 10 μg/mL, and allowed infection to proceed for 120 h. While in the absence of HE-Fc there was no sign of virus propagation as detectable by immunofluorescence assay, concentrations of exogenous sialate-O-acetylesterase as low as 1 ng/mL to up to 1 μg/mL promoted virus growth (Fig. 2A).

To determine whether these conditions would allow isolation of rBCoV-HE-F²¹¹A without mutations in S1^A, we performed targeted recombination and rescued recombinant viruses by 160-h multistep propagation as before, but now with culture supernatant supplemented with 100 ng/mL HE-Fc (Fig. 2B). Sanger sequence analysis of RT-PCR amplicons showed that all viruses, cloned by endpoint dilution of the 160-h stock (n = 4), coded for mutant HE-Phe²¹¹Ala in combination with wild-type S1^A. To assess the stability of clonal rBCoV-HE-F²¹¹A/S^wt/Rluc, the virus stock was amplified by a low multiplicity of infection (MOI) passage for another 160 h in the presence of exogenous HE-Fc (Fig. 2B). The resulting virus population was analyzed by next-generation sequencing (NGS), which allows for the detection of low-frequency mutants. Sequence variation in HE and S1^A was distributed randomly and did not exceed background levels (<0.15%). More than 99.5% of the viruses coded for HE-Phe²¹¹Ala, while preserving parental type S1^A (Fig. 2B).

With a clonal, virtually pure stock of rBCoV-HE-F²¹¹A/S^wt available, we performed controlled forced evolution experiments. The virus was serially passaged involving three consecutive 120-h multistep propagation rounds in HRT18 cells, but now in the absence of exogenous HE-Fc, with the initial infection performed at an MOI of 0.005 (Fig. 2C). In trial 1, viral titers in passage 1 (p1) increased only slowly to 3 × 10⁴ and 2 × 10⁴ TCID50/mL (measured with or without exogenous HE-Fc, respectively). The withdrawal of exogenous RDE during viral passage immediately selected for mutations in S1^A. Virus cloning by endpoint dilution of the 120-h p1 sample yielded S RBS mutants Asn²⁷Ser, Thr⁸³Ile, and Ile²⁶Ser (Fig. 2C), all three of which had been seen before (Fig. 1A). NGS analysis revealed the true complexity of the p1 population (Fig. 2C) and identified two additional S1^A variants with substitutions (His¹⁷³Tyr and Arg¹⁹⁷Cys) more distal from the RBS (SI Appendix, Figs. S2D and S3 A and B). His¹⁷³Tyr also reduced the relative binding affinity of S1^A-Fc, albeit less dramatically than the other mutations, namely by 30-fold (Table 1). The Arg¹⁹⁷Cys mutation seemingly falls in a separate category and presumably reduces the avidity of S homotrimers by inducing aberrant disulfide bonding and local S1^A misfolding in one or more S monomers (SI Appendix, Fig. S3).

All in all, the p1 population was comprised for virtually 100% of HE-Phe²¹¹Ala mutants, 40% of which in combination with parental BCoV S, the remaining 60% with second-site mutations in S1^A (Fig. 2C and Table 1). Of the latter, the ultralow-affinity variant S1^A-Ile²⁶Ser was the most abundant at 46% and the Asn²⁷Ser variant the least at less than 1%. However, upon a subsequent round of 120-h multistep propagation, the tables were turned with S1^A-Asn²⁷Ser now comprising almost 40% of the p2 population and with the Ile²⁶Ser variant reduced to 0.7%. In addition, four other S1^A variants emerged. One of these had a mutation in S1^A RBS loop L1, Val²⁹Gly, and a relative binding affinity close to that of the Asn²⁷Ser mutant (Fig. 2C and Table 1). We also identified at position 75 a second S1^A Cys-substitution mutant, which like Arg¹⁹⁷Cys, presumably disrupts the RBS through aberrant disulfide bonding (SI Appendix, Fig. S3). Remarkably, two other S1^A variants arose with mutations (Arg⁸⁸Thr and Pro¹⁷⁴Leu) that affected the relative binding affinity only modestly to 0.25 and 0.5 of that of wild-type S1^A-Fc, respectively (Table 1). Even more remarkably, upon further passage these mutants of near wild-type affinity increased to dominate the p3 population, effectively outcompeting variants with low-affinity spikes as well as those with parental spikes (Fig. 2 C and D). However, when the p1, p2, and p3 stocks were cloned by endpoint dilution in the absence of exogenous HE-Fc, only virus variants with low-affinity mutations in S1^A were isolated (Fig. 2C and Table 1). Strikingly, from the p3 stock, the S1^A-Asn²⁷Ser variant was isolated exclusively against all odds (10 of 10 tested; P < 10⁻⁶) when calculated from its frequency in the population (21%). Conversely, virus cloning by endpoint dilution in the presence of exogenous HE-Fc yielded high-affinity S1^A mutants Pro¹⁷⁴Leu (5 of 11) and Arg⁸⁸Thr (4 of 11), parental virus rBCoV-HE-Phe²¹¹Ala/S^wt (1 of 11), and intermediate S affinity variant His¹⁷³Tyr (1 of 11).

Notably, the conditions selected not only for mutations in S but also in HE. Variants with an Ala²¹¹Val substitution in HE emerged in p2, rising to 17% of the p2 end population, to stabilize around this frequency in p3. As a result of this mutation, HE lectin affinity was regained albeit to levels solely detectable by high-sensitivity nanobead hemagglutination assay assay (HAA), while esterase activity toward clustered glycotopes in BSM increased 4-fold as compared to HE-Phe²¹¹Ala, but still remained 125-fold lower than that of wild-type HE (SI Appendix, Fig. S4). Apparently, the increase in HE function, minor as it may be, provides a selective advantage, but apparently one that benefits both low- and high-affinity S variants, because the mutation was found in cloned viruses of either type.

To corroborate our observations, the controlled forced evolution experiment was repeated (SI Appendix, Fig. S5). As compared to the first trial, there was a much faster population built-up already in p1 at 120 h postinfection (hpi) with final titers reaching 4 × 10⁸ and 3.4 × 10⁷ TCID50/mL, when measured in the presence or absence of exogenous HE-Fc, respectively. Surprisingly, in stark contrast to trial 1, the trial 2 p1 population was comprised for about 94% of viruses expressing wild-type BCoV S. Less than 6% consisted of variants with mutations in S1^A, four of low receptor binding affinity (Thr²²Ile, Asn²⁷Tyr, Val²⁹Gly, His¹⁷³Tyr), with the exception of Thr²²Ile, all at positions seen before, and one of near wild-type binding affinity (Pro¹⁷⁴Leu) (Tables 1 and 2). Consistent with our previous findings, however, virus purification through endpoint dilution in the absence of exogenous HE-Fc yielded low-affinity mutants (11 of 11 tested) exclusively (SI Appendix, Fig. S5). If the variants in the trial 2 p1 population were all of equal replicative fitness under the conditions applied, the odds of this result would be less than 1.10⁻¹².

At first glance, the two trials would seem to differ in their outcomes. We offer, however, that the results are in fact consonant and that the main difference is in the speed with which the virus populations increased and evolved. There is an inherent stochastic element to the experimental approach and whether the developing quasi-species undergoes slow-track (Exp. 1) or fast-track evolution (Exp. 2) is likely dependent on the time of advent of the first mutant virus and its properties, for example whether it is an ultralow- (like Ile²⁶Ser) or low-affinity variant (like Asn²⁷Ser). The findings allow for several conclusions. 1) They confirm and firmly establish that loss of HE lectin function selects for mutations in S1^A that reduce S receptor-binding affinity and virion avidity. 2) The possibilities to reduce the affinity of the S RBS through single site mutations are finite. In several independent experiments, substitutions in S1^A occurred at a limited number of positions, albeit not necessarily by the same residue. For example, Asn²⁷ was replaced both by Ser and Tyr. 3) The mutations that reduce S affinity fall into different categories. Most map within or in close proximity of the RBS to affect receptor–ligand interaction directly. Others, like His¹⁷³Tyr and Pro¹⁷⁴Leu, are more distal from the RBS and apparently affect ligand binding indirectly through long range conformational effects. A third type of mutations, quasi-random Cys substitutions, apparently disrupt S1^A folding by promoting nonnative disulfide-bonding. 4) Perhaps most surprisingly, quasi-species developed in which loss of HE lectin function was compensated at the level of the viral population. Minority low-affinity variants, in trial 2, constituting less than 6% of the swarm, not only sustained the replication of high-affinity variants but actually allowed the latter to flourish and amplify to become the majority phenotype. In trial 1, this took multiple passages, and spontaneous mutants with near wild-type S affinity became dominant. In trial 2, residual input virus with wild-type S presumably profited from the early emergence and rapid expansion of the initiating low-affinity variant S1^A-Asn²⁷Ser, already during the first 120 h of multistep propagation.

Among the first mutations fixed upon zoonotic introduction and early emergence of OC43, was a HE-Thr¹¹⁴Asn substitution, which created a glycosylation site at the rim of the lectin domain RBS (11) (Fig. 3A). Glycans attached to HE Asn¹¹⁴ hamper binding to 9-O-Ac-Sia through steric hindrance, causing a 500-fold reduction in HE avidity (Fig. 3B) and a 125-fold in sialate-O-acetylesterase-activity, respectively (Fig. 3C). We introduced the HE Thr¹¹⁴Asn substitution in BCoV, expecting that the glycosylation site would be rapidly lost through any of several single-nucleotide restorative mutations in HE. Indeed, NGS analysis of the virus swarm arising after targeted recombination showed the glycosylation site to be destroyed but only in 10% of the population and exclusively by Ser¹¹⁶Phe substitution (Fig. 3D). This mutation only partially restores HE receptor binding and receptor destruction to 0.125 and 0.17 of that of wild-type HE, respectively (Fig. 3 B and C). In the vast majority of viruses, the newly introduced HE glycosylation site was retained and, instead, low-affinity S1^A mutations were selected again, with S-RBS Thr⁸³ replaced either by Ile (69%)—as seen before (Figs. 1A and 2C and Table 1)—or by Asn (10%) (Fig. 3D). The latter mutation reduces S1^A affinity to 0.008 of that of wild-type.

Virus cloning by endpoint dilution yielded, in three of five isolates, S1^A-Thr⁸³Asn variants with the newly introduced N-glycosylation site in HE still intact (HE-Thr¹¹⁴Asn). Furthermore, a single S1^A-Thr⁸³Ile variant was isolated, but this virus in addition had lost the N-glycosylation site in HE (HE-Thr¹¹⁴Asn/Ser¹¹⁶Phe) (Fig. 3D). The observations led us to entertain the possibility that the mutations in S1^A and HE did not occur independently and that, even in viruses expressing low-affinity spikes, partially restorative mutations in HE would yet provide a selective advantage. To test this, the clonal S1^A-Thr⁸³Asn/HE-Thr¹¹⁴Asn variants were serially passaged. All three viruses independently lost HE glycosylation at Asn¹¹⁴over time and, saliently, again through Ser¹¹⁶Phe substitution exclusively. Even more remarkably, with HE-Ser¹¹⁶Phe mutants gaining dominance, variants emerged that had restored S affinity to (near) wild-type through substitution of S1^A-Asn⁸³ either by Thr or by Ser (Fig. 3E).

For one of the five clonal populations obtained by endpoint dilution, we unfortunately failed to determine its genotype for technical reasons. From the NGS analysis of the p1 population, we deduced that the starting mutant must have been a low-affinity S1^A-Leu⁸⁹Pro variant that, like the S1^A-Thr⁸³Asn/Ile variants described above, quickly lost the HE-Asn¹¹⁴ glycan through an HE-Ser¹¹⁶Phe substitution. Oddly enough, the Leu⁸⁹Pro substitution had not been detected by NGS in the precloning virus stock. Note, however, that this mutation had been selected before twice independently in trials with rBCoV-HE-F²¹¹A (Table 2). Possibly, it arose spontaneously during endpoint dilution procedure. Be that as it may, its in vitro evolution proved informative (Fig. 4A). NGS analysis of a passage p1 population, resulting from 120-h multistep propagation, showed that 100% of the viruses coded for HE-Thr¹¹⁴Asn/Ser¹¹⁶Phe in combination with S1^A-Pro⁸⁹ (53.3%), -Thr⁸⁹ (40.5%), or -Ser⁸⁹ (2%). Note that the relationship between these variants and the course of evolution—from Leu⁸⁹ in the parental recombinant virus to Pro and from Pro to Thr or Ser—is evident from the codon sequences (CTA→CCA→T/ACA; mutations indicated in bold and italicized) and that the Thr⁸⁹ and Ser⁸⁹ substitutions restored S RBS affinity almost to that of wild-type RBS (Fig. 4B). All three variants (S1^A-Pro⁸⁹, -Thr⁸⁹, and -Ser⁸⁹) were readily cloned and isolated by standard endpoint dilution, and propagated independently without a requirement for exogenous RDE. p1 also contained a minor population of viruses with parental S1^A, presumably regenerated from S1^A-Pro⁸⁹, which as for the S1^A-Thr⁸⁹ and Ser⁸⁹ variants would have required only a single nucleotide substitution (CCA→CTA). Apparently, with HE lectin function partially restored, viruses that regained (near) wild-type S affinity had a selective advantage. At the end of passage p2, the low-affinity S1^A-Pro⁸⁹ variants had dwindled to less than 1.5%, S1^A-Thr⁸⁹ had become dominant at 75% and viruses with wild-type S1^A-Leu⁸⁹ had rapidly risen from 2.25% in p1 to 23% (Fig. 4A).

In summary, the reduction in HE receptor-binding affinity resulting from the introduction of an N-linked glycosylation site triggered a series of successive mutations in S and HE. The mutations, first emerging in S, strongly decreased overall virion avidity initially, but mutants were then selected that stepwise, through loss of the HE 114-glycosylation site followed by restorative mutations in S, reverted to near wild-type avidity. Thus, the data directly demonstrate HE/S coevolution. Moreover, the findings suggest that virions are under selective pressure not only to balance receptor-binding and receptor-destroying activities in apparent relation to cell-surface receptor-densities, but also, within these constrains, to optimize virion avidity (for a schematic summary of the findings, see Fig. 4C).

Loss-of-function mutations in the BCoV HE lectin domain had a much bigger impact on BCoV propagation than expected. The prototype OC43 laboratory strain USA/1967 also lacks HE lectin function, yet in HRT18 cells it grows to titers comparable to those of BCoV reference strain Mebus. NGS of OC43 stocks revealed heterogeneity, but no indications for the existence of low S affinity minority variants that would support replication of majority high S affinity viruses. In addition, clonal virus populations obtained by endpoint dilution (10 of 10) all conformed to the S1^A master sequence. We offer that instead OC43-USA/1967 may have reached a viable S/HE balance compatible with efficient in vitro propagation through adaptations in S that reduced receptor-binding affinity or altered receptor fine-specificity. When measured by solid-phase assay with bivalent S1^A-Fc fusion proteins, binding of the S protein of OC43 USA/1967 to BSM, containing both mono- and di-O-acetylated α2,6-sialoglycans, is 16- to 32-fold lower than that of BCoV-Mebus (32). sp-LBA with BSM preparations, selectively depleted for either 9-O- or 7,9-di-O-Sias, showed that BCoV S, like BCoV HE (28), preferentially binds to 7,9-di-O-Ac-Sias (SI Appendix, Fig. S6A). OC43 USA/1967 S1^A may not share this preference. Apparently due to its low affinity, detectable binding to BSM was lost upon depletion of either type of Sia (SI Appendix, Fig. S6A). Moreover, even though BCoV-Mebus S preferably binds to 7,9-di-O-Ac-Sia, monovalent one-on-one binding of the BCoV S1^A domain to α2,6-linked 9-O-acetylated Sia is still threefold stronger than that of OC43-USA/1967 as measured by biolayer interferometry (SI Appendix, Fig. S6B). On a cautionary note, the isolation and complex passage history of OC43-USA/1967 (48, 49) entailed several passages in human tracheal organ culture, suckling mouse brain and many rounds of replication in cultured cells (9), which would have given the virus ample opportunity to adapt to the in vitro conditions. Thus, the binding characteristics of its spike may not faithfully reflect those in circulating field variants. Indeed, OC43 variants in sputum samples, contrary to the OC43 USA/1967, replicate in airway epithelial cell cultures but not in tissue culture cells (50).

Our findings demonstrate that in the prototypic β1CoV BCoV the envelope proteins S and HE are functionally entwined and coevolve. We posit that the same holds for other members of the species Betacoronavirus-1, including its zoonotic descendant human CoV OC43 and related viruses of swine, rabbits, dogs, and horses, as well as for other Embecovirus species, most prominently among which human CoV HKU1. The data lead us to conclude that the respective activities of S and HE in receptor-binding and catalysis-driven virion elution are balanced to ensure dynamic reversible virion attachment and, thereby, efficient virus propagation. In consequence, for the viruses listed above, the roles of S and HE during natural infection cannot be understood in isolation but must be considered in unison.

Using a reverse genetics-based forced evolution approach with BCoV as a model system, we showed that loss of HE lectin function causes an offset in S/HE balance, practically incompatible with virus propagation and spread. With the HE lectin domain as modulator of esterase activity, mutations that decrease or abolish HE RBS affinity reduce virion-associated sialate-O-acetylesterase activity toward clustered glycotopes on hypervalent glycoconjugates (11), such as are present in the mucus and glycocalyx in natural tissues and on the surface of cultured cells. The extent of the resultant defect is such that compensatory second-site mutations in S are selected for that dramatically reduce S RBS affinity, apparently to restore reversibility of binding and virion motility as an escape ticket from inadvertent virion attachment to decoy receptors.

The single-amino acid mutations in receptor-binding domain S1^A were limited to a finite number of positions. They were either within or proximal to the RBS to directly affect protein–ligand interactions, or more distal to reduce RBS affinity through long-range effects or by disrupting local folding through aberrant disulfide-bonding. Whereas the parental recombinant viruses, defective in HE lectin function but with wild-type S RBS affinity, require an external source of receptor-destroying enzyme for propagation, their progeny escape mutants regained propagation-independence by lowering S affinity.

In expanding clonal populations of HE-defective rBCoV-HE-F²¹¹A, propagated in the absence of exogenous receptor-destroying enzyme activity, viruses with reduced S affinity gained a selective advantage initially. Eventually, however, quasi-species developed in which loss of HE lectin function was compensated at the population level. Variants that combined the HE-Phe²¹¹Ala mutation with (near) wild-type affinity S proteins increased to dominate the swarm at least numerically. Still, these high-affinity S variants for their proliferation were strictly reliant on minority low-affinity S variants. This relationship extends beyond cooperativity and group selection described for other systems (51–57) and amounts to a state of dependency. We propose that the virions of the low-affinity minority variants provide aid by serving as a source of exogenous sialate-O-acetylesterase activity. They evade decoy receptors through enhanced reversibility of virion attachment, but this phenomenon increases their motility—whether by sliding diffusion or binding/rebinding—causing them to deplete cell surface 9-O-Ac-Sias, decoy receptors, and functional receptors alike. With increasing concentrations of low-affinity virions in the culture supernatant, high-affinity variants would profit progressively, whereas falling cell surface receptor densities would put the low-affinity viruses increasingly at a disadvantage.

The forced evolution trials performed with rBCoV-HE-F²¹¹A were restricted in course and outcome by design, because full reversion would require simultaneous mutation of two adjacent nucleotides. Moreover, the crucial role of the Phe²¹¹ in ligand binding (SI Appendix, Fig. S1) obviates conservative substitutions (17). Although rBCoV-HE^A211V variants did emerge in two separate experiments, this mutation only marginally increases HE RBS affinity and sialate-O-acetylesterase activity.

In contrast to the HE-Phe²¹¹Ala mutation, the deleterious effect of N-glycosylation at HE-Asn¹¹⁴ can be reversed, completely or partially, through various single-nucleotide substitutions in codons 114 and 116 and would therefore more readily allow for compensatory mutations also in HE. Indeed, serial passage of the rBCoV-HE-Thr¹¹⁴Asn resulted in a succession of mutations alternatingly in HE and S. The order of appearance of these mutations and their effect on protein function indicated that they were not fixed to merely restore the balance between attachment and catalysis-driven virion elution. The HE-Thr¹¹⁴Asn substitution initially selected for second-site mutations that reduced S affinity (Thr⁸³Ile, Thr⁸³Asn, and Leu⁸⁹Pro), but with propagation thus recovered, derivatives rapidly emerged with increased HE lectin and esterase activity through a Ser¹¹⁶Phe mutation. Apparently, this created an HE/S disbalance that in turn favored the selection of viruses with revertant mutations in S that raised S RBS affinity again to wild-type (Thr⁸³ → Ile → Thr; Leu⁸⁹ → Pro → Leu) or near wild-type levels (Thr⁸³ → Ile → Ser; Leu⁸⁹ → Pro→ Thr/Ser). Conjointly, our findings indicate that through an initial sharp reduction in overall avidity, compensatory to loss of HE function, virus particles regained the capacity of eluding nonproductive attachment to decoy receptors but at a fitness penalty. The decrease in S RBS affinity would predictably lower the specific infectivity of virus particles through a decrease in productive host cell attachment. The rapid selection of the HE-Ser¹¹⁶Phe mutation in a low-affinity S background can thus be understood to have increased virion avidity, albeit through HE and rather than through S. HE does have a dual function after all, and in influenza viruses C and D as well as in murine coronavirus-1, it is a receptor-binding protein first and foremost (22, 35, 58–61). Of note, the partial Ser¹¹⁶Phe reversion of HE consistently seen in multiple independent experiments suggests that a return to (near) wild-type lectin and esterase activity along with a low-affinity S would have tipped the scale too much toward catalytic virion release. We posit that in addition to an optimal balance between receptor binding and receptor destruction, the system strives toward optimal virion avidity (Fig. 4C).

Under natural circumstances, the set-point of the S/HE balance would be tailored to conditions met in the target tissues of the intact host. The spontaneous loss of HE lectin function in OC43 and HKU1 may thus be understood to have arisen through convergent evolution as an adaptation to the sialoglycan composition of the mucus in the human upper respiratory tract, and that of the glycocalyx of the respiratory epithelia (11). This change, which would predictably reduce virion-associated receptor destruction and hence decrease virion elution, might have been selected for by low density occurrence of 9-O-Ac-sialoglycans in the human upper airways. In accordance, limited tissue array analyses with HE-based virolectins suggested that these sugars are not particularly prevalent in the human respiratory tract and by far not as ubiquitous as in the gut (28). However, full understanding of how the S/HE balance was reset in OC43 and HKU1 upon their zoonotic introduction and why awaits further analysis of the binding properties and ligand fine-specificity of the S proteins of naturally occurring variants, as well as more quantitative and comprehensive interhost comparative analyses of airway sialoglycomes. As an added complication, virion particles encounter widely different circumstances while traversing the mucus layer, at the epithelial cell surface, during local cell-to-cell dissemination, and during transmission. It is an open question whether this selects for majority phenotypes that can cope individually and independently with each of these different conditions by striking an uneasy compromise with regard to HE/S balance and overall virion avidity, or whether there is loco-temporal selection for swarms of variants that collectively allow the virus population as a whole to overcome each hurdle.

The embecovirus HE gene originated from a horizontal gene transfer event, presumably with an influenza C/D-like virus as donor (17, 62). Like the orthomyxovirus hemagglutinin-esterase-fusion proteins, the newly acquired coronavirus HE protein provided the acceptor virus with an opportunity to reversibly bind to 9-O-Ac-sialoglycans (26). This in turn would seem to have prompted a shift in the receptor-specificity of S through adaptations in S1^A that created a 9-O-Ac-Sia binding site de novo so that virions could now attach to these receptor determinants also via S. The embecoviruses thus adopted a strategy of receptor usage entailing a concerted and carefully fine-tuned activity of two envelope proteins that is unique among coronaviruses, but uncannily similar to that of influenza A viruses. In the latter, the hemagglutinin protein HA, as a pendant of S, mediates binding to either α2,3- or α2,6-linked sialosides, while the neuraminidase (NA), like HE, is a receptor-destroying enzyme but with a substrate fine-specificity that closely matches HA ligand preference (63). For influenza A virus, the existence and biological relevance of a functional balance between receptor binding and receptor destruction is well recognized (64–67). This balance is critical for receptor-associated virus motility through the mucus and at the cell surface (41, 43, 44, 68–70). Complete or partial loss of NA activity—whether invoked spontaneously, through reverse genetics or by viral propagation in the presence of NA inhibitors—selects for mutations around the HA receptor-binding pocket that reduce HA affinity (65, 71, 72).

Furthermore, as proposed here for HE, NA contributes to virion attachment and even compensates for loss of virion avidity in mutant viruses with reduced HA affinity (73). Different from HE, NA may do so via its catalytic pocket, which doubles as a Sia-binding site (74). However, NA also possesses a second Sia binding site (75, 76), which like the HE lectin domain, regulates NA activity and which, in further analogy, is conserved or lost in apparent correlation with host tropism (77–79). Finally, among many other similarities to embecoviruses, influenza A variants with different set points in their HA-NA functional balance may cooperate to support their propagation in cultured cells (53). Our observations establish that there are common principles of virion–sialoglycan interactions that prompted convergent evolution of β1CoVs and influenza A viruses. Although these two groups of viruses essentially differ in genome type and replication strategy, envelope proteins, and receptors, they seem to be subject to the same rules of engagement with respect to dynamic receptor binding, the differences between them constituting variations on a theme. This implies that observations made for the one system are informative for the other. Perhaps more importantly, insight into the overriding principles of virus–glycan interactions may open avenues to common strategies for antiviral intervention.

HRT18 (ATCC CCL244) and mouse LR7 (45) cells were maintained in DMEM containing 10% fetal calf serum (FCS), penicillin (100 IU/mL), and streptomycin (100 µg/mL). BCoV strain Mebus and OC43 strain USA/1967, purchased from the American Type Culture Collection (ATCC), were propagated in HRT18 cells.

A reverse genetics system based on targeted RNA recombination was developed for BCoV strain Mebus essentially as described previously (15, 45, 80). Using conventional cloning methods, RT-PCR amplicons of the 5′-terminal 601 nt and 3′-terminal 9,292 nt of the BCoV strain Mebus genome (reference GenBank sequence U00735.2) were fused and cloned in plasmid pUC57, downstream of a T7 RNA polymerase promotor and upstream of a 25-nt poly(A) tract and a PacI site, yielding pD-BCoV1. From this construct, BCoV ORF 4a was deleted (nucleotides 27740 to 27853) and replaced by the Renilla luciferase (Rluc) gene, yielding pD-BCoV-Rluc. A second pD-BCoV1 derivative, pD-mBCoVΔHE, was created by replacing the coding sequence for the ectodomain of BCoV S (nucleotides 23641 to 27433) by the corresponding MHV-A59 sequence and by deleting the BCoV HE gene (nucleotides 22406 to 23623). The nucleotide sequences of pD-BCoV1, pD-BCoV-Rluc and pD-mBCoVΔHE, determined by bidirectional Sanger sequence analysis, were deposited in GenBank (accession codes: MT939521–MT939523).

To generate a recombinant chimeric acceptor virus, mBCoV∆HE, HRT18 cells were infected with BCoV-Mebus at MOI of 10 TCID50 per cell and trypsinized and resuspended in PBS. An aliquot of this suspension, containing 1.5 × 10⁶ cells in 0.8 mL, was mixed with capped synthetic RNA that had been produced by in vitro transcription using the mMESSAGE mMACHINE T7 Transcription Kit (Thermo Fisher) with PacI-linearized mBCoV∆HE vector as template. The mixture was subjected to two consecutive electrical pulses of 850 V at 20 μF with a Gene Pulser II electroporator (Bio-Rad) and the cells were then seeded on a confluent monolayer of LR7 feeder cells in a 35-mm dish. Incubation was continued at 37 °C, 5% CO₂ for 18 h posttransfection until wide-spread cytopathic effect was apparent. The cell culture supernatant was harvested and cleared by low-speed centrifugation at 1,200 rpm, and mBCoV∆HE was purified by end-point dilution and used to generate stocks for future usage in LR7 cells.

To generate luciferase-expressing rBCoVs with the BCoV HE and S genes reconstituted (i.e., rBCoV^wt or rBCoV-HE-Phe²¹¹Ala) LR7 cells, infected with mBCoV∆HE at MOI of 5, were electroporated as described above with synthetic RNA transcribed from pD-BCoV-Rluc and derivatives thereof. The infected and transfected cells were then seeded on HRT18 cell monolayers in 35-mm plates for up to 160 h. For rescue and propagation of rBCoV-HE-Phe²¹¹Ala without second-site mutations in S, the cell culture supernatants were supplemented with 100 ng/mL of BCoV HE-Fc protein (17). After 5 to 7 d of incubation at 37 °C, samples of the cell culture supernatants were tested for infectivity by transferring them to HRT18 cell monolayers grown on 12-mm glass coverslips in 15.6-mm wells. Incubation was continued for 12 h, after which the cells were fixed with paraformaldehyde and immunofluorescence staining was performed with polyclonal antiserum from a BCoV-infected cow.

mBCoV was titrated and cloned by endpoint dilution on LR7 cells with cytopathic effect as read-out. rBCoVs were titrated and cloned in HRT18 cells. To identify infected wells, cell supernatants were analyzed by HAA with rat erythrocytes (26) and by Renilla luciferase assay (Dual-Luciferase Reporter Assay System, Promega). Titers were calculated by the Spearman–Kaerber formula. Clonal virus populations were characterized by isolating viral RNA from 150-µL aliquots of the cell culture supernatant with the NucleoSpin RNA Virus kit (Macherey-Nagel) followed by conventional RT-PCR and bidirectional Sanger sequence analysis.

Confluent HRT18 monolayers (5 × 10⁶ cells) grown in 25-cm² flasks, were inoculated with rBCoV-HE-F²¹¹A at MOI 0.005 in PBS for 1 h at 37 °C. The cells were washed three times with PBS to remove residual exogeneous HE-Fc and incubation was continued in DMEM + 10% FCS at 37 °C, 5% CO₂ for 120 hpi with samples collected every 24 h (p1). Subsequent 120-h passages were performed by adding 10 µL of supernatant to new cultures of HRT18 cells in 25-cm² flasks.

BCoV HE, either enzymatically-active (HE⁺) or rendered inactive through a Ser⁴⁰Ala substitution (HE⁰), and OC43 S1^A were expressed as Fc fusion proteins in HEK293T cells and purified from the cell supernatant by protein A affinity chromatography, as detailed previously (17, 32). Monomeric S1^A was obtained by on-the bead thrombin cleavage (32). pCD5-BCoVHE-T-Fc vectors (17) encoding mutant BCoV HE derivatives were constructed with the Q5 Site-Directed Mutagenesis Kit per the instructions of the manufacturer.

The production of BCoV S-pseudotyped VSV-ΔG particles, their characterization by Western blot analysis, and infectivity assays in HRT18 cells were as described previously (32).

sp-LBA was performed as described previously (32) with BSM (Sigma-Aldrich), coated to 96-well Maxisorp microtitre ELISA plates (Nunc, 0.1 µg BSM per well), serving as a ligand. Binding assays were performed with twofold serial dilutions of HE⁰-Fc, S1^A-Fc, or mutated derivatives thereof. Receptor-destroying esterase activities of soluble HEs were measured by on-the-plate 9-O-Ac-Sia depletion assays, as described previously (11, 28).

HAA was performed with rat erythrocytes (Rattus norvegicus strain Wistar; 50% suspension in PBS). Standard HAA was done with twofold serial dilutions of HE⁰-Fc proteins (starting at 25 ng per well) as described previously (17). High-sensitivity nanoparticle HAA (NP-HAA) was performed as in refs. 32 and 81. Briefly, self-assembling 60-meric nanoparticles, comprised of lumazine synthase (LS), N-terminally extended with the immunoglobulin Fc-binding domain of the Staphylococcus aureus protein A, were complexed with HE⁰-Fc proteins at a 1:0.6 molar ratio for 30 min on ice. The HE⁰-Fc-loaded nanoparticles were then twofold serially diluted and mixed 1:1 (vol/vol) with rat erythrocytes (0.5% in PBS). Incubation was for 2 h at 4 °C after which HAA titers were read.

Viral RNA from culture supernatants was isolated as described above. HE and S1^A coding regions from viral genome of different virus populations were obtained by RT-PCR with primer sets HE_F 5′-TTA​GAT​TAT​GGT​CTA​AGC​ATC​ATG-3′ and HE_R 5′-TTA​GAT​TAT​GGT​CTA​AGC​ATC​ATG-3′, S1^A_F 5′-ACC​ATG​TTT​TTG​ATA​CTT​TTA-3′ and S1^A_R 5′-AGA​TTG​TGT​TTT​ACA​CTT​AAT​CTC-3′, respectively. Amplicons were processed in the NGSgo workflow for Illumina according to the Instructions for Use (Edition 4), except that the fragmentation was prolonged to 40 min at 25 °C (protocol 3A). Briefly, amplicons were subjected to fragmentation and adapter ligation using NGSgo-LibrX (GenDx). Size selection and clean-up of the samples was performed with SPRI beads (Machery-Nagel). Unique barcodes were ligated to each sample using NGSgo-IndX (GenDx), after which all samples were pooled and subsequently purified with SPRI beads, resulting in a library of fragments between ∼400 and 1,000 bp. The DNA fragments were denatured and paired-end sequenced on a MiSeq platform (Illumina) using a 300-cycle kit (V2). FASTQ files were analyzed in NGSengine (GenDx), which aligned the reads to the reference sequences of HE and S (reference GenBank sequence U00735.2 for BCoV strain Mebus, and NC_006213.1 for OC43 strain USA/1967). For the characterization of each virus sample, amplicons from five independent RT-PCR reactions were analyzed in parallel and mutation frequencies were determined by averaging the results from these five replicates.