Viral chimeras decrypt the role of enterovirus capsid proteins in viral tropism, acid sensitivity and optimal growth temperature

Despite their genetic similarities, enteric and respiratory enteroviruses (EVs) have highly heterogeneous biophysical properties and cause a vast diversity of human pathologies. In vitro differences include acid sensitivity, optimal growth temperature and tissue tropism, which reflect a preferential in vivo replication in the respiratory or gastrointestinal tract and are thus key determinants of EV virulence. To investigate the underlying cause of these differences, we generated chimeras at the capsid-level between EV-D68 (a respiratory EV) and EV-D94 (an enteric EV). Although some chimeras were nonfunctional, EV-D94 with both the capsid and 2A protease or the capsid only of EV-D68 were both viable. Using this latter construct, we performed several functional assays, which indicated that capsid proteins determine acid sensitivity and tropism in cell lines and in respiratory, intestinal and neural tissues. Additionally, capsid genes were shown to also participate in determining the optimal growth temperature, since EV-D94 temperature adaptation relied on single mutations in VP1, while constructs with EV-D68 capsid could not adapt to higher temperatures. Finally, we demonstrate that EV-D68 maintains residual binding-capacity after acid-treatment despite a loss of infectivity. In contrast, non-structural rather than capsid proteins modulate the innate immune response in tissues. These unique biophysical insights expose another layer in the phenotypic diversity of one of world’s most prevalent pathogens and could aid target selection for vaccine or antiviral development.

Introduction Enteroviruses (EVs) represent one of the leading causes of human disease worldwide and are associated with a broad spectrum of clinically distinct syndromes. While most infections are benign or asymptomatic, a small proportion may lead to life-threatening disease. Apart from the successful vaccines against poliovirus, there are currently no efficient antiviral treatments or prophylaxes against these common and potentially fatal pathogens.
Classified in the Picornaviridae family, the Enterovirus genus is composed of over 300 different genotypes distributed among 13 species (EV-A to EV-J and RV-A to RV-C), seven of which include human pathogens [1]. These divisions are based primarily on nucleotide and amino acid homology rather than the phenotypic presentation of human infection. Their positive-stranded RNA genome of 7'200 to 7'500 nucleotides is packed into a 30 nm icosahedral capsid. The 5' end of the genome is covalently linked to the small viral protein, VPg, and contains a highly structured untranslated region (5'UTR) composed of cis-acting elements necessary for replication and translation. An internal ribosomal entry site (IRES) is required for cap-independent translation, which gives rise to a precursor polyprotein that is subsequently cleaved into 11 mature proteins by viral proteases. This precursor polyprotein encompasses 3 major regions (P1-P3). The P1 region encodes capsid proteins (1A to 1D or VP4 to VP1), while the P2 and P3 regions encode non-structural proteins (2A to 2C and 3A to 3D). The 3'UTR is marked by a poly-A tail and contains elements involved in the regulation of replication [2] (Fig 1).
Although closely related at a genetic level, viruses within the EV genus have distinct transmission routes and replication sites. They also display remarkably different phenotypic characteristics and associated clinical syndromes [3,4]. Based on their primary replication site, the respiratory or gastrointestinal mucosa, EVs can be divided in respiratory EVs (which also include all rhinovirus (RV) types) and enteric EVs, although gastrointestinal diseases are rarely caused by EV infections [3]. Respiratory EVs are transmitted via the respiratory route, have an infectious tropism restricted to the respiratory tract and rarely disseminate. By contrast, enteric EVs are mostly transmitted via the fecal-oral route, replicate predominantly in the gastrointestinal mucosa from where they can disseminate and infect a wide range of different organs, including the central nervous system. Differences between these 2 groups of viruses also exist in vitro: for example, unlike enteric EVs, respiratory EVs exhibit particle instability at low pH, a lower optimal replication temperature, and a restricted cell tropism [5].
Despite intensive research, the genetic roots of this phenotypic variability are still poorly understood. Acid treatment of RVs was shown to result in a loss of infectivity due to a conformational change in the capsid proteins, leading to the expulsion of VP4 [6]. Studies conducted in the 1970's revealed that acidification followed by neutralization of RV particles produced 2 subviral components ("A", containing RNA without VP4 and "B", with neither RNA nor VP4): a transformation that made the RV particles defective in attachment [7]. There are numerous determinants of viral tropism, but as capsid proteins contain the receptor-binding site, they are thought to play a significant role. Other viral proteins potentially important in this process are those interacting with the host innate defenses as well as elements of the translation and replication machinery [8]. With regard to optimal growth temperature, various mutations across the genome have been described in cold-adapted viruses and shown to impact the growth of EVs in vivo; however, the exact genomic regions providing resistance to higher body temperatures have not yet been defined [9,10]. Similar uncertainty exists regarding the determinants of the diversity in immune responses and complex immune evasion strategies between the various EVs. While innate pattern recognition and evasion strategies have been linked to distinct viral proteins such as 2A, 2C and 3C [11], the precise roles of structural and non-structural proteins in these processes are yet to be elucidated.
The aim of this work is to identify the genetic regions responsible for differences in key determinants of viral viability (namely, acid sensitivity, optimal growth temperature, tissue tropism and induction of innate immunity) through the creation of unique viral chimeras. We thus exchanged the viral capsid proteins between respiratory and enteric EVs. As only intraspecies chimeric combinations are viable at the polyprotein level [12], we selected 2 viruses from the same species that are different in acid sensitivity and tissue tropism: EV-D68 as a Results obtained upon transfection and passage in HeLa cells are indicated on the right, as is the percentage of nucleotide and amino acid sequence identity between the exchanged regions. V, viable construct; X, non-viable construct; V/X, viable unfit construct. B. Non-synonymous adaptation mutations observed in viral stocks after 6 passages in HeLa cells. Sequencing was performed from nt 45 to 7344. Production of viral stocks by transfection at both 33˚C and 37˚C was attempted for all viruses, but only EV-D94 could be recovered at 37˚C. For EV-D94, the stock subsequently used for phenotypic assessment is the one prepared at 33˚C. HeLa 33 and HeLa 37: viral stocks transfected and amplified at 33˚C and 37˚C respectively.
representative of respiratory EVs [13], and EV-D94 as a representative of enteric EVs [14]. Unlike most non-RVs EVs, EV-D68 virions are acid-labile, replicate optimally at low temperatures, are spread mainly via the respiratory route and cause mild to severe respiratory disease [13,15]. Because of their physicochemical properties reminiscent of RVs, they were initially misclassified as human RV-87 [13,16]. Other EVs with a preferred respiratory tropism such as CVA-21 or the recently identified EV-C104, EV-C105, EV-C109, EV-C117 and EV-C118, may also share these properties but as most of these EV-C viruses do not grow in standard culture, their phenotypic characterization remains difficult [3]. While recent EV-D68 isolates have shown neurotropic potential in animal models, the tropism of the original prototypic Fermon strain used in this study is restricted to the respiratory tract [17]. The selected enteric virus, EV-D94 is a recently characterized EV-D type that was isolated concurrently from sewage specimens collected in Egypt and from fecal specimens of acute flaccid paralysis (AFP) patients in the Democratic Republic of the Congo [14]. In vitro examinations revealed a stability of the virion under acidic conditions and a remarkably wide tissue and cell tropism [14]. EV-D94 and most EV-D68 strains depend on sialic acid for binding and infection of susceptible cells [18,19]. ICAM-5/telencephalin was also recently described as a receptor required for entry and replication of sialic acid-dependent and -independent EV-D68 strains [20].
We thus exchanged the capsid regions of EV-D68 and EV-D94 and assessed the viability, tropism (in vitro using cell lines and ex vivo with reconstituted human tissues), acid sensitivity, temperature adaptation and finally interplay with host innate immunity of the chimeras. Taken together, this work clarifies the diverse biophysical and immunogenic roles of capsid proteins in EV pathogenesis and may guide the search of therapeutic targets.

Viability, fitness and adaptation of chimeric EV-D68/EV-D94 viruses
Chimeras were generated by interchanging genomic regions between EV-D68 and EV-D94. Exchanged regions comprised either the capsid region alone (P1), both the capsid and protease regions together (P1 and 2A) or VP1. The following 4 permutations were studied (Fig 1): A) EV-D94/D68 P1 which integrated the P1 region of EV-D68 into the framework of EV-D94, B) EV-D68/D94 P1 which integrated the P1 region of EV-D94 into the framework of EV-D68, C) EV-D94/D68 P1-2A which integrated both the P1 and 2A regions of EV-D68 into the framework of EV-D94 and finally D) EV-D94/D68 VP1 with VP1 of EV-D68 integrated into the framework of EV-D94. After in vitro transcription and cell transfection, 2 blind passages were performed and virus viability was assessed by cytopathic effect (CPE) examination and confirmed by immunofluorescence (IF). EV-D94/D68 P1 and EV-D94/D68 P1-2A gave rise to viable replicating viruses, causing CPE in HeLa cells; however, although EV-D94/D68 P1 growth was similar to EV-D94 and EV-D68, only few cells were infected with EV-D94/D68 P1-2A and viral loads measured by quantitative PCR after the first passage was more than 100'000 times higher for EV-D94/D68 P1 than for D94/D68 P1-2A . The full genomes of EV-D94, EV-D68 and EV-D94/ D68 P1 obtained in HeLa cells after 3 passages were sequenced and several adaptation mutations were observed, particularly for EV-D94 ( Fig 1B). Since EV-D94/D68 P1-2A was not fully adapted, we did not analyze mutations of the viral stock by sequencing. No other constructs were viable as shown by absence of CPE, negative IF signal and absence of viral RNA amplification measured by RT-qPCR after 2 passages. The non-viability of EV-D68/D94 P1 was unexpected since the reverse construct was fully fit. To exclude the presence of unnoticed deleterious mutation in the construct, we generated it again using several different approaches; manually, using a fusion PCR-based procedure and also by ordering the full-length construct. A cell adaptation mutation (VP1 H284Y ) observed in the P1 region of the parental EV-D94 virus ( Fig 1B) was also inserted in the chimera to try to boost the potentially unfit virus. However, none of these methods produced viable progeny.

Capsid proteins are the only determinants of acid sensitivity in EVs
Based on viral fitness, EV-D94/D68 P1 was used in functional assays, and together with its parental viruses (EV-D68 and EV-D94), underwent acid sensitivity testing. A standard protocol was used to test particle stability after acid (pH 3) exposition. While EV-D94 infectivity was unchanged, acid pre-treatment of EV-D68 resulted in a total loss of infectivity. Interestingly, the chimera carrying the EV-D68 P1 region (EV-D94/D68 P1 ) was similarly affected by acid pre-treatment. To identify the step of the viral cycle impaired by acid treatment, we performed several assays to test the binding, entry and replication steps based on time post infection ( Fig  2). Whereas EV-D94 infectivity was not significantly impaired at any step by acid treatment, entry (Fig 2A) and subsequently replication (Fig 2B) of EV-D68 and EV-D94/D68 P1 were blocked after acid exposure. In contrast, binding was only partially affected for EV-D68 and EV-D94/D68 P1 (Fig 2C). To assess the specificity of this residual attachment post acid-treatment, the same binding assay was run on cells pretreated or not with sialidase. Sialidase treatment further inhibited attachment of acid-treated EV-D68 (Fig 2D), indicating a significant and specific binding to sialic acid after acid pretreatment. Entry assay: viruses (pre-treated or not with acid) were added to cells for 2 hours at 33˚C and after extensive washing, viral RNA was extracted for quantification by real-time RT-qPCR. B. Replication assay: viruses were added to cells for 1h at 33˚C and after extensive washing, cells were further incubated for 24h. Viral RNA was extracted from total cell lysate and quantified. C. Binding assay: viruses were added to cells for 1 hour at 4˚C to prevent entry. After extensive washing, viral RNA was extracted and quantified. D. Binding assay as in C but with cells pretreated with sialidase. For each panel, data are expressed in percentage compared to untreated controls. ÃÃ

Cellular and tissue tropism of EVs depends exclusively on capsid proteins
Differential tropism in cell-lines. Viral replication in several cell lines was tested to assess in vitro tropism of EV-D68, EV-D94 and EV-D94/D68 P1 . Numerous human and non-human cell lines originating from various organs were infected with the 3 viruses and infection was assessed by CPE examination and IF (S1 Fig). Infections of HeLa (human cervical carcinoma), RD (human rhabdomyosarcoma), A549 (human lung carcinoma), Caco-2 (human colorectal adenocarcinoma), Vero (African green monkey kidney) and SH-SY5Y (human neuroblastoma) cells were carried out. EV-D94 was able to replicate in all cell types, while EV-D68 replicated only in HeLa, RD, A549 and Caco-2 cells but not in Vero or SH-SY5Y. EV-D94/D68 P1 exhibited the exact same cellular tropism as EV-D68. The ability of EV-D68 to infect Caco-2 cells, at even higher levels than EV-D94, was not expected considering the colorectal origin of these cells and the restricted respiratory tropism of this virus.
Differential tropism in respiratory, intestinal and neural tissues. To better replicate viral tropism in humans, the same infection assays were performed on complex tissue models in vitro reconstituted from human primary cells that mimic the upper respiratory tract, small intestine, neurons or neural tissues composed of neurons and glial cells. For respiratory and gut tissue models grown at an air-liquid interface, cells were infected apically with 10 8 viral particles (normalised based on RNA quantification) of each viral stock and, after extensive washing at 4 hpi, apical viral release was measured daily as previously described [21,22]. Residual RNA was measured 4 hpi despite extensive washes in both tissues (Fig 3A and 3B). However, the residual RNA is progressively washed out during the daily collection process, as suggested by the absence of detection of viral RNA at the apical site of respiratory tissues 2dpi with replication deficient viruses (eg UV-irradiated or acid treated viruses) [21].
Between 0.5x10 7 and 10 8 viral particles were used to infect primary neurons or engineered neural tissues [23]; but as these cultures are fragile, the viral inoculum was not washed out and replication was assessed by comparing the amount of cell-associated virus directly after inoculation and 4 days later.
In respiratory tissues, viral production peaked between day 1 and 3 for all viruses. However, viral replication of both EV-D68 and EV-D94/D68 P1 was significantly higher than EV-D94 from day 1 to day 4 ( Fig 3A, 1 st panel). This higher viral replication was paralleled by an inhibition of mucociliary clearance (MCC) (Fig 3A, 2 nd panel). Similarly, we observed a decrease in cilia beating frequency from day 3 in EV-D68 and EV-D94/D68 P1 infected tissues while neither MCC nor cilia beating were affected by infection with EV-D94. These findings can be linked to the massive loss of ciliated cells observed by IF at 4 dpi caused by EV-D68 and EV-D94/D68 P1 as opposed to EV-D94 ( Fig 3A 3 rd panel).
The opposite situation occurred in intestinal tissues where viral loads measured 4 dpi for EV-D94 were around 4 log higher than those of EV-D68 ( Fig 3B). Once again, EV-D94/D68 P1 exhibited the same characteristics of its EV-D68 P1 donor parent strain in the intestinal tissue model. The same replication profile was observed in in vitro-reconstituted neural tissues, where no increase in cell-associated virus was measured 4 dpi with EV-D68 and the chimera EV-D94/D68 P1 in either neuron monolayers ( Fig 3C) or complex 3D engineered neural tissues ( Fig 3D). Interestingly, EV-D94 could infect both tissues, supporting a potential neurotropism of this newly reported EV. The absolute viral loads of EV-D94 measured in the different tissues 4 dpi were significantly higher in intestinal tissues.

Non-structural proteins modulate IFN induction levels
To investigate the implication of structural and non-structural proteins in induction and/or interference with the antiviral response, IFN induction by the different constructs was investigated both in in vitro-reconstituted respiratory epithelia and cell lines. IFN induction was not compared in intestinal or neural tissues as both EV-D68 and the chimera replicated poorly in these tissues. First, infected respiratory epithelia were examined 3 dpi at the mRNA level for IFN-β and IFN-λ in tissue lysates by RT-qPCR ( Fig 4A and 4B), and at the protein level for IFN-λ in supernatant by ELISA ( Fig 4C). IFN-β was not analyzed by ELISA since we previously reported a poor induction of this cytokine in respiratory tissues [22]. Interestingly, IFN induction by EV-D68 and EV-D94 differed significantly both at the protein and mRNA levels; EV-D68 strongly induced both type I and type III IFN, whereas expression levels in EV-D94 and EV-D94/D68 P1 infected tissues were comparable to uninfected controls. This differential induction did not correlate with differences in viral replication ( Fig 4D). Identical infections in RD, A549 and HeLa cells, however, did not induce either type I or type III IFN (S2 Fig).

Capsid sequences determine sensitivity to high temperatures
To identify the key genetic region implicated in temperature sensitivity, in vitro transcripts of EV-D94, EV-D68 and EV-D94/D68 P1 were transfected at 33˚C and 37˚C. While the 3 viruses could be recovered after transfection and passages at 33˚C, only EV-D94 was recovered after We then tried to adapt EV-D68 and EV-D94/D68 P1 stocks obtained at 33˚C to higher growth temperature. The two viruses were able to grow at 37˚C but with limited fitness and they did not shift their optimal growth temperature to 37˚C, despite repeated attempts for EV-D68.
In contrast, EV-D94 viral stocks adapted to 33˚C or 37˚C could be obtained in HeLa cells after transfection of the original infectious clone and only 6 passages at these 2 temperatures. These viruses presented different optimal growth temperature not only in HeLa cells (Fig 5) but also in RD, A549 and Vero but to lesser extent (S4 Fig). Whole genome sequencing was performed on these adapted stocks to identify temperature adaptation mutations. Five non- synonymous mutations (VP1 H284Y, A241T , 2A K47E, P76S and 3D H40Q ) were observed in EV-D94 adapted to 33˚C when compared to the original plasmid construct (Figs 1B and 5A). To find out if some of these mutations were responsible for the improved growth at 33˚C, the VP1 ( Fig  5B) and 2A (Fig 5C) mutations were introduced independently in the original EV-D94 infectious clone. These 2 derivatives showed a similar optimal growth temperature of 33˚C, indicating that none of these mutations was responsible for this phenotype. Since none of these two constructs has the 3D H40Q substitution, an implication of this mutation in the adaptation at 33˚C was also excluded. Of note, presence of only the 2A adaptation mutations considerably reduced viral fitness (Fig 5C) implying that the VP1 mutations play a critical role for growth in HeLa cells. To identify mutations resulting in optimal growth at 37˚C, the genome of the virus adapted at 37˚C was compared with that of the original infectious clone. Two mutations in VP1 (VP1 G94D, P165R ) and 1 mutation in 2A (2A A144V ) were identified (Fig 5D). Introduction of the 2A mutation alone in the original EV-D94 clone resulted in a drastic fitness loss but a similar growth at 33 and 37˚C (Fig 5F). In contrast, introduction of the VP1 mutations reduced viral fitness but did not suppress the growth advantage of the virus at 37˚C (Fig 5E), suggesting a key role of these mutations in the adaptation to 37˚C. To confirm this hypothesis, these 2 VP1 mutations were introduced in the genetic framework of the virus adapted to 33˚C and they drastically shifted its optimal growth temperature from 33˚C to 37˚C (Fig 5G and  S5 Fig).

Discussion
EV genotypes are co-circulating in humans and are responsible for a wide variety of diseases [4]. The phenotypic diversity of their pathogenesis extends to their biophysical characteristics in vitro but is curiously not reflected at the genetic level as is exemplified by members of the EV-D species. By investigating chimeric EV constructs among members of this species, we were able to identify the genomic regions responsible for the key elements of the functional diversity in EV pathogenesis. Using a fully viable chimeric EV-D94 (expressing EV-D68 capsid-encoding genes), we demonstrate that capsid-encoding genes determine viral tropism and acid sensitivity. We also show that EV-D68 capsid presents a major obstacle to adaptation to higher growth temperature while EV-D94 capsid does not. We finally highlight a significant and differential modulation of innate immune response by non-capsid proteins, and thus clarify mechanisms of several layers of EV pathogenesis from biophysical properties to host-pathogen interaction.
The first novel finding in this study is the limited intraspecies compatibility within the EV-D species. Intertypic intraspecies recombination is a frequent event in nature, especially in non-RV EVs. While viable recombinants within the EV-A, -B and -C species have been reported to occur (both in vitro as well as in nature [24,25]), no similar observation has been reported for EV-D. To explore the role of capsid proteins in EV pathogenesis, we created several variants of capsid-chimeras, containing either the full capsid, full capsid and 2A, or VP1 alone. Out of all the constructs, only the exchange of the entire EV-D68 P1 region within the genomic framework of EV-D94 was as viable and fit as the parental viruses. Restrictions of EV-D94 adapted to 33˚C contains 5 non-synonymous mutations relative to the original EV-D94 clone and presents higher titers at 33˚C than at 37˚C. The VP1 (B) or 2A (C) mutations were introduced independently in the original infectious clone and the two derivatives retain an optimal growth at 33˚C. D) EV-D94 adapted to 37˚C presents 3 non synonymous mutations relative to the original EV-D94 clone and presents higher titers at 37˚C than at 33˚C. The VP1 (E) or 2A (F) mutations were introduced independently in the original infectious clone and only the VP1 mutated virus retains an optimal growth at 37˚C. G. Mutations in VP1 of EV-D94 adapted to 37˚C (E,##) were added to the VP1 mutant adapted to 33˚C (B, #) and the viral titers at both temperatures were assessed. Ã P< 0.05, ÃÃ P< 0.01. https://doi.org/10.1371/journal.ppat.1006962.g005 Viral chimeras decrypt the role of enterovirus capsid proteins viability were not further explored; however, it may be hypothesised that it was due to the highly ordered structure of icosahedral capsids where mosaic capsids constructions (within P1), can be expected to be poorly viable as was revealed in previous in vitro studies on poliovirus [26,27]. Further, recent natural recombination events within P1 are virtually non-existent [28]. The fact that P1 from EV-D68 in the genomic framework of EV-D94 gave rise to a viable chimera, whereas its reverse counterpart did not (despite various methods of fabrication) was more surprising. This chimera could not be made viable even with the addition of cell-adaptation mutations in the capsid-encoding EV-D94 region. One possible explanation for this limitation could be an incompatibility at the encapsidation step, which is controlled by 2C ATPase . This enzyme is believed to interact directly with the VP3 capsid protein to incorporate newly synthesised genomes and a highly specific protein-protein interaction has been shown to be an important constraint of recombinant compatibility [29]. Sequencing of our EV-D94/D68 P1 viral stocks revealed a non-conservative mutation in VP3, mapping to the interior of the viral capsid ( S6A Fig) and possibly reflecting a rescue of a defective encapsidation. However, similar compensatory mutations did not occur for the reverse chimera EV-D68/D94 P1 despite numerous transfection attempts. This suggests that the incompatibility between structural and nonstructural proteins of different genotypes may be unilateral. Of note, we did not observe such incompatibilities when we generated intraspecies chimeric RVs [12], although in this case reciprocal chimeras were not created. Other non-structural proteins, such as VPg, also interact with the capsid at critical points of the viral growth cycle [30] and may cause similar limitations to chimeric constructs [31,32]. Finally, one cannot exclude that unrecognized cis-acting elements, critical for EV-D68 replication may lie in the capsid-coding region.
The chimeric virus EV-D94/D68 P1 exhibited identical acid sensitivity to that of its P1 donor virus, EV-D68, proving the essential involvement of capsid proteins in this biophysical characteristic. While acid treatment completely inhibited viral entry, we showed that sialicacid dependent binding was still effective. Our observation that binding of acid-treated EV-D68 virions could be further reduced by sialidase cell-treatment contradicts previous studies in RV-A2, concluding that acidity induces a conformational change in the capsid (forming subviral "A" and "B" particles) that makes it incapable of cell binding [7]. However, recent studies with the same virus established that acid-induced conformational changes were confined to particle "A" transformations, involving the release of VP4 and exposure of N-terminal part of VP1, but that it was not sufficient to provoke RNA release [33,34]. In these conditions, binding is still possible, albeit diminished. Furthermore, the capsid of EV-D68 may present a different conformation change after acidic treatment than the capsid of RV-A2. Nevertheless, despite a retained ability to bind sialic acid, the conformational change induced by acid treatment definitively prevented viral entry. This may rely on the inability of the virus to bind an entry receptor distinct from sialic acid. Alternatively, the irreversible conformational change induced by acidity may preclude proper viral internalisation and/or uncoating.
As the cellular tropism of EV-D68 was completely transferred to the EV-D94/D68 P1 chimera, we were able to conclude that tropism is probably determined by virus-receptor interactions of the capsid proteins. Both EV-D68 and EV-D94 are known to bind sialic acid on cells [19]; while some sialic acid-independent strains have been reported for EV-D68, the Fermon strain is sialic acid-dependent. Additionally, ICAM-5/telencephalin (a neuron-specific molecule but also largely present in the respiratory tract), has been recently proposed as the entry receptor [18][19][20]. Our results suggest the existence of a virus-specific receptor (which could be ICAM-5/telencephalin for EV-D68) responsible for the differential tropism observed between EV-D68 and EV-D94. In addition, acid sensitive viruses may have a facilitated uncoating in compartments with low pH when compared to acid-resistant viruses, and thus rely less on conformational changes induced by capsid-receptor interactions. Accordingly, EV-D68 and EV-D94 probably have different entry mechanisms. Additional experiments in diverse cell lines as well as improved knowledge on EV-D receptor usage will be necessary to better define the different entry pathways of EV-D68 and EV-D94.
As until now no animal model is able to accurately replicate the EV pathogenesis in humans, this makes in vitro-reconstituted human tissue models essential to EV research [35]. Indeed, even in vitro studies have limitations, as can be seen by our observation that the in vivo EV cellular tropism of both EV-D68 and EV-D94 was not reflected in their respective immortalized cell lines. Although ex vivo models do not fully reproduce the complexity of human physiology, the first steps of infection can be still accurately studied. As expected, EV-D68 proved to be an efficient respiratory virus, reaching 1000-fold higher titres than EV-D94 in respiratory tissues. The fact that these high replication levels were transferred to the EV-D94/ D68 P1 chimera further underlines the essential role of the P1 capsid region in the EV-D68 phenotype. This similarity between EV-D68 and the EV-D94/D68 P1 chimera was also seen in ciliated cell toxicity and impaired mucociliary clearance, which is in line with our recently published work on these tissues [22]. Interestingly, EV-D94 did not have the same repercussions and as innate immunity induction of EV-D94/D68 P1 was comparable to EV-D94, replication levels rather than collateral damage by the host immune response likely explain the pathogenesis observed.
In line with the expected tropism of EV-D94, this virus was able to infect both neural and gastrointestinal tissues, supporting its probable role in AFP cases in humans. While a potential causal role of EV-D68 in AFP has been suggested by epidemiological studies, recent murine models have suggested that only recently emerged strains (and not the Fermon prototypic strain used in this study), have this neuroinvasion ability [17] which was confirmed in our study, where neither EV-D68 nor the EV-D94/D68 P1 chimera were able to efficiently infect the neural tissue models. Of note, absolute replication levels vary considerably among tissue types. Though these values are not directly comparable due to different experimental settings linked to differences in tissue composition (monolayer, pseudo-stratified, stratified), EV-D94 was shown to reach the highest viral loads in intestinal tissues. This correlates with preferred replication of enteric EVs in the gastrointestinal tract. Taken together, these experiments show that, in contrast to immortalised cell lines, ex vivo model systems are reliable surrogates to investigate the early stages of EV pathogenesis and, more importantly, reveal the central involvement of capsid proteins in determining tissue tropism.
Since respiratory tissues sustain important and comparable replication of both EV-D68 and EV-D94/D68 P1 , studies investigating the innate immune response were standardised in this model. As EV-D68 was the only virus to induce both type I and III IFN, whereas EV-D94 and EV-D94/D68 P1 did not lead to strong IFN induction either at the protein or the mRNA level, we could conclude that this feature is not determined by the P1 genomic region. Chimeras with exchanges of non-structural proteins could help to identify the elements involved in the innate immune response. Of note, innate immunity was not induced at all in cell lines, showing again the limitations of this model.
Similarly to the other observations, the fact that neither EV-D68 nor EV-D94/D68 P1 could be transfected or adapted to 37˚C, strongly implicates the P1 capsid region in EV thermal stability. Concerning EV-D94, our selected strain (E210) was only passaged once in L20B cells and was thus not adapted to grow in HeLa [14]. Acquisition of improved growth in HeLa cells was nevertheless a very quick process and only 6 passages allowed the emergence of 2 temperature-adapted versions of EV-D94 with 5 mutations accumulated upon growth at 33˚C and 3 at 37˚C. Different combinations of the 33˚C adaptation mutations decreased viral fitness but without changing the virus optimal growth temperature. This indicates that the better growth at 33˚C is not associated with specific viral mutations but rather represents a default condition linked to higher tolerance of the cells to viral infection at this temperature. This may be related to reduced innate immunity at cold temperature as has been previously demonstrated [36,37] and could explain why RNA transfection is more efficient at 33˚C also for non-viral RNA molecules such as siRNA or RNA PAMPs. In contrast, with the same approach, we identified 2 mutations in VP1 (D94G and P165R) that were critical for optimal growth in cells at 37˚C. G94D maps to the VP1 BC loop, while P165R is located in the canyon (S6B Fig). Both the VP1 BC loop and the canyon are known to be involved in virus adaptation [38,39] and receptor binding [40] and the BC loop was further shown to modulate CVB3 thermolability [41]. In agreement with these published data, these 2 mutations were also important for viral fitness in HeLa cells and it is not clear whether the 2 phenotypes are related. Of note, we did not repeat independent adaptation at 37˚C and cannot confirm that these 2 mutations are the only ones conferring improved fitness at higher temperatures. It is also very likely that in the genomic framework of another EV or in another cell type, other mutations may emerge and fill the same function.
To conclude, using recombinants produced in vitro between EV-D68, a respiratory EV and EV-D94, an enteric EV, we proved that capsid proteins are both necessary and sufficient to determine the acid sensitivity and restricted tropism of the respiratory EV, as well as its optimal growth temperature. In contrast, we showed that innate immunity induction is rather modulated by non-structural proteins. The same experimental approach could be effective to explore the role of other EV proteins and exchanges between other pairs of EVs could help to generalize our findings to all EVs. Such approaches open the door to an improved understanding of these highly prevalent but poorly understood pathogens.

Temperature-adapted viruses
RNA of EV-D94 adapted to 33˚C and 37˚C (6 passages) was extracted and reverse-transcribed. PCR amplification of either the P1 or 2A region was performed and the resulting PCR products were cloned in a pCR2.1-TOPO vector (Thermo Fischer Scientific, Waltham, USA). Cloning in the pcDNA3.1-EV-D94 vector was undertaken with either SfoI/BamHI or BamHI/AvrII for P1 or 2A inserts, respectively. Site-directed mutagenesis, as described in [42] was used to add VP1 D94G , VP1 P165R on the 33˚C P1 -adapted EV-D94 mutant with primers RV192-193 and RV194-195. All PCR reactions were performed using the taq Phusion (Thermo Fisher Scientific, Waltham, MA, USA) and absence of mutation was checked by sequencing (Fasteris-DNA sequencing service, Switzerland).

In vitro transcription and transfection
RNA was synthesised using the RiboMAX Large Scale RNA Production System-T7 kit (Promega, USA) either on Pst1 linearized plasmids or on PCR products according to manufacturer's instruction. 2ug of transcribed RNA were transfected in 6-well plates, each well containing 80% confluent HeLa Ohio cells, using the TransMessenger transfection reagent kit (Qiagen, Germany), at either 33˚C or 37˚C.

Tissues
MucilAir tissues were obtained from Epithelix (Geneva, Switzerland) and cultured at the airliquid interface as previously described [21,22]. Intestinal tissues were purchased at MatTek (Ashland, USA) and cultured at an air-liquid interface according to manufacturer's instruction in the provided culture medium. Primary neurons and 3D neural tissues were both engineered from pluripotent stem cells and cultured as previously described [23].

Ethics statement
Respiratory (http://www.epithelix.com/products/mucilair) and EpiIntestinal tissues (https:// www.mattek.com/products/epiintestinal/tissues) were respectively ordered from Epithelix and MatTek, two biotechnology companies. There, the tissues are developed from anonymized samples and after Ethical approval. Neural tissues were also developed from anonymized samples and following ethical approval. The study was conducted according to the Declaration of Helsinki on biomedical research (Hong Kong amendment, 1989), and the research protocol was approved by our local ethics committee.

Viral stocks
Viral stocks were produced by transfection of HeLa Ohio cells incubated at 33˚C or 37˚C. Seven days after transfection, parental and chimeric viruses' supernatants and cells were collected, subjected to 3 freeze-thaw cycles, purified and re-passaged in 80% confluent HeLa cells. When cytopathic effect (CPE) was observed or at a maximum 7 dpi, supernatants and cells were collected and subjected to 3 freeze-thaw cycles before a 2 nd passage. When CPE was observed, supernatants were collected again and subjected to 3 freeze-thaw cycles, purified and aliquoted. Viral stocks were quantified by qPCR and titration, according to the Reed and Muench method [43]. For temperature-adapted EV-D94 stocks, 3 additional passages were achieved the same way. To check for adaptation to viral culture at different temperatures by sequencing, viral RNA was extracted from the stocks, reverse transcribed and PCR amplified with EV-D94 or EV-D68 specific primers (S1 Table).

Infection of in vitro-reconstituted human respiratory epithelia (MucilAir) and small intestine (EpiIntestinal) tissues
MucilAir and EpiIntestinal tissues were infected apically with 100μl of medium containing 10 8 RNA copies of each virus. The viral inoculum was washed out 3 times at 4 hpi. After these 3 washes and then every day, 200μl of medium was applied apically for 20 minutes at 33˚C (or 37˚C for small intestine tissues) for sample collection. Basal medium was also collected daily and replaced with 500μl of fresh medium. RNA was extracted and viral replication quantified by RT-qPCR.

Infection of human primary neurons and 3D engineered neural tissues
Primary neurons and 3D neural tissues were infected with equivalent amount of each virus (0.5x10 7 to 10 8 RNA copies) diluted in 100μl or 40μl of culture medium respectively. Due to tissue fragility, the viral inoculum was not washed out. Tissues (without culture medium) were lysed directly after infection (2 hpi for neurons and 4hpi for 3D neural tissues) and 4 dpi to measure cell associated virus. Total RNA was extracted and viral replication was quantified by RT-qPCR.

RNA extraction, reverse transcription and real-time RT-PCR quantification
RNA was extracted using the NucliSens easyMAG magnetic beads system (BioMérieux, France) according to the manufacturer's instructions. Retro-transcription was performed with Superscript II (Invitrogen) and either random hexamer primers (Roche) or oligo(dT) primers (Invitrogen), as described [42]. RT-qPCR was performed using the quantitative Entero/Ge/08 assay as previously described [44] or the EV-D68 specific assay [45] in a one-step format using the QuantiTect Probe RT-PCR Kit (Qiagen, Switzerland) according to the manufacturer's instructions, in a StepOne Applied Biosystems thermocycler. 10-fold dilution series (from 2.5 Ã 10 8 to 2.5 Ã 10 5 copies/ml) of the in vitro transcribed full-length pBMH-EV-D68 were used as a quantitative reference standard for each run.

Immunofluorescence (IF)
Infected and control cells were washed twice with phosphate-buffered saline (PBS), fixed and permeabilized for 20 minutes in a methanol-acetone mixture (1:1) at -20˚C and then air dried at room temperature during a few minutes. Incubation with the primary mouse J2 monoclonal antibody (Scicons, Hungary), specific for double-stranded RNA longer than 40 bp, and diluted 1:500 in PBS -1% bovine serum albumin was realized for 45 minutes at 37˚C. Intensive PBS washings were performed prior to second incubation with fluorescein isothiocyanate (FITC)conjugated anti-mouse IgG antibody (Light Diagnostics, Merck, USA). Then cells were washed 3 times with PBS, stained with 4,6-diamidino-2-phenylindole at room temperature for 5 minutes and washed with PBS a final time. Coverslips were mounted in Fluoroprep mounting medium (BioMérieux, France) and analyzed using standard microscopy.

IF of Mucilair tissues
At 5 dpi, tissues were washed 3 times with PBS and fixed with 4% paraformaldehyde at room temperature (RT) for 15 minutes. Tissues were then washed 3 times and permeabilized with Perm/wash buffer (BD, USA), then (after 3 more wash steps) the first antibodies were added and incubated for 1 hour at 37˚C. Viral staining was performed using the mAbJ2 diluted 1/500 whereas ciliated cells were stained with the rabbit anti-beta IV tubulin Ab (179504 Abcam, UK, diluted 1/250). The Alexa 594-goat anti-mouse Ab (A11032, Life technologies, USA, diluted 1/3000) and the Alexa 488-goat anti-rabbit Ab (A11008 Life technologies, USA, diluted 1/3000) were used as secondary antibodies and incubated for 1 hour at 37˚C. After rinsing with PBS, tissues were stained with DAPI, washed with PBS and mounted onto glass slides in Fluoroprep (BioMérieux, France). Images were acquired with a Zeiss LSM 700 Meta confocal microscope with a 63.6/1.4 objective, processed by Imaris and are presented in 3D projections.

Mucociliary clearance and cilia beating assessment
Polystyrene microbeads (Sigma 84135) displacement velocity at the apical surface was measured with contrast phase microscopy and image ProsPlus software, as already described [22]. Cilia beating movies were recorded with Sony XCD-U60CR microscope and processed with Sony zcl 142 software and ImageJ.

Acid sensitivity assay
Based on a standard protocol [46], 20 μl of viral stocks were diluted either in 20 μl of 0.1M citrate buffer pH 3.0 or 0.1M of phosphate buffer pH 7.2 and the mixture was incubated for 1 hour at 37˚C. After neutralization by adding 40 μl of 0.1M phosphate buffer pH 7.2 and 120 μl of culture medium, samples were inoculated onto HeLa Ohio cells. Virus binding assay. HeLa cells were seeded at 5x10 4 cells/well in 96-well plates. The following day, medium was removed, cells were washed with cold Hanks' Balanced Salt Solution (HBSS) and 200ul of binding buffer (HBSS containing 0.1% sodium azide and 1% BSA) were added on cells, which were then chilled on ice for 10 minutes. Buffer was removed and 100ul of virus mixture were added. Wells were washed 3 times with 200ul of binding buffer directly after inoculation or after 1 hour incubation on ice to assess residual or specific binding respectively, and after 3 washes with PBS, cells were lysed in 200ul of easyMAG lysis buffer to quantify bound RNA by RT-qPCR. Residual RNA was subtracted to bound RNA and percent of bound RNA was calculated relative to non-treated control for each virus.

Sialidase assay
α2-3,6,8,9 Neuraminidase A (P0722S, New England Biolabs, USA) was diluted 1:100 in a EDB-0.5%BSA buffer, as described [47], added to cells and incubated for 1 hour at 37˚C. Cells were then washed once with PBS and used for the binding assay.

Interferon induction and interferon (IFN) sensitivity assessment
IFN lambda (IFN-λ1/ λ3, IL-29 /IL-28B) was measured in the apical medium of respiratory tissues by enzyme-linked immunosorbent assay ELISA according to manufacturer's instructions (DY1598B R&D, USA). mRNA of IFN-β and IFN-λ in tissue lysate were quantified using Taq-Man Gene Expression Assay (Hs01103582_s1, ThermoFischer, USA) for IFN-β and with primers and probes described previously for IFN-λ [48]. The IFN Ct values were normalized to those of the RNAse P housekeeping gene (3416844, Applied Biosystems/ThermoFischer, USA). Relative quantification was calculated using the 2 -ΔΔCt method [49].

Statistics
Values are expressed as mean (± SEM). Experiments were done at least in biological duplicates and in each experiment, conditions were run in duplicate except for intestinal and 3D neural tissues where biological replicates were done without intra-experimental replicate. Ordinary one-way or two-way ANOVA tests (without matching or pairing) and multiple comparisons were performed with GraphPad Prism 7.02 software. For Fig 3A and 3B, significance was calculated with t-Tests on the area under the curve (AUC) and SE. For