The pre-(PDB ID:1OKE ³⁷ ) and post-(PDB ID:1OK8 ¹² ) fusion conformations of DENV2 were used for MEPP analysis ³⁸ . The major difference in these proteins is a 33Å displacement of domain III, as previously noted ^{9, 13} . Several metrics were used for identifying residues that undergo spatial, electrostatic and secondary structure perturbations. The first analysis computed pairs of residues that have a electrostatic potential difference (EPD) reversal (EPD-R) (> 150 units), were within 8 Å of each other in both conformations and had minimal distance perturbation (<4Å). Residues were marked as (i) completely conserved, (ii) stereochemically equivalent or (iii) not conversed. His317, the residue implicated in pH sensing ^{66, 67} , switches electrostatic polarity with respect to Thr315 ( Table 1). Both His317 and Thr315 are conserved in ZIKV/DENV1-4/JEV/WNV ( Figure 2), and are known epitopes ^{23, 43} . Another pair (Arg286-Arg288) with EPD-R are stereochemically equivalent in ZIKV/DENV1-4/JEV/WNV ( Figure 3), and lie on a putative GAG-binding domain preceding the DI/DIII linker ^{48, 49} . Thr359, which is an epitope for the same MAb that binds Thr315 and His317 ²³ , but is not conserved even among DENV serotypes, is another such residue which has EPD-R with Ser363 ( Figure 2). Thus, barring the pair Thr32-Thr40 ( Table 1), all residues that have an EPD-R with respect to a spatially proximal residue are known to be epitopes, even when not conserved across different viruses.

Next, normalized distance deviations (see Methods) highlight Phe11, Tyr299, Ser7, Arg9, Glu26, Arg188, Glu13 and Gln316 as residues with the largest spatial perturbations ( Figure 4a). Barring Ser7, all residues are completely conserved in ZIKV/DENV1-4 ( Figure 2). A N-terminal peptide (DENV3,4-12 VGVGNRDFV) that enhances immunogenicity for CD8+ T cells when expressed from modified vaccinia Ankara includes Phe11, Ser7 and Arg9 ³⁹ . Arg9 and Glu13 are also epitopes of other antibodies ³⁵ . This particular study also showed that the N8R substitution DNA vaccine had a more neutralizing and protective effect than wild-type immunized sera, both in vitro and in vivo ³⁵ . Arg9 is part of a salt bridge with Glu368 which maintains the structure of the E-protein in the pre-fusion state ^{10, 66} . Tyr299 is part of the epitope for the cross-reactive neutralizing MAb DENV1-E102 ⁶⁸ . Arg188 is essential for infectivity, and is neutralized by DC4 Fab ⁴⁰ , while the monoclonal antibody DD18-5 recognized residue Glu26 in DENV4 ³⁵ , a residue predicted by the Pepitope server ³⁴ . Thus, all spatially perturbed residues identified by MEPP are known epitopes.

The following residues have significant cumulative EPD deviations ( >150 EPD units) with other residues within 6Å - Thr32, Thr40, Lys160, Lys247, Asp249, Arg288, Met297 and His317 ( Figure 4b). His317, Arg288, Thr32 and Thr40 have been discussed above. Several residues identified by this electrostatic feature do not have known references in current literature. Of these residues, Lys160 and Met297 are not conserved in ZIKV/DENV1-4, while Asp249 is conserved in DENV, but not in ZIKV ( Figure 2). However, Thr32 and Thr40 are two conserved residues ( Figure 2) with EPD deviations, leading to an EPD-R as described above ( Table 1).

Subsequently, analysis of α-helices in the pre- and post-fusion DENV2-E protein revealed α1 is slightly perturbed post fusion, increasing in length by one residue (Asp215) compared to the pre-fusion α1 ( Figure 5, Table 2). Asp215 is important for infectivity ⁴² , a proven ⁴¹ and predicted ⁴⁷ epitope, and a membranotropic region of the E protein (peptide 29) ⁶⁹ .

The secondary structures from DENV and ZIKV E-proteins were extracted using DSSP ⁵⁵ , and analyzed using PAGAL ⁴⁴ . The ZIKV-E protein (PDB ID:5IREA, length=501 residues) has six α-helices and thirty β-sheets (see SSEinfo.zip in Dataset 1). The Edmundson wheels ⁷⁰ for these α-helices in the stem region ( Table 2) shows their amphipathic cationic nature ( Figure 6). Interfacial hydrophobicity plays a critical role in cell entry of viruses ⁷¹ . The membranotropic α3 and α4 in DENV4 ⁶⁹ has been studied extensively through mutational studies of the hydrophobic face ^{62, 63} . Another strategy using peptide mimetic (residues 412 to 444, named DN59) derived from these helices showed inhibition of flaviviruses by releasing genomic RNA ^{72, 73} . A similar study based on peptide mimetic of residues 419-447 (comprising the conserved stretch following α3 and α4) inhibited viral entry ⁷⁴ . These peptides were most effective at inhibition when three residues (442-444) were mutated to tryptophan, the most hydrophobic residue according to the Wimley-White whole residue hydrophobicity scale ⁷⁵ . An interesting feature of α4 is the complete conservation of residues on the charged surface - Ser439, Gly436, Lys432, Gly439 and H435 in DENV ( Figure 6), while the hydrophobic face is much more variable. Only Asn428 is not conserved ( Figure 2).

ZIKV and DENV are Class II fusion viruses that deploy β-sheet-rich domains to destabilize membranes ⁶ . The charged features of these β-sheets emphasizes β21 in ZIKV (294:KCRLK, preceding domain I/III linker) as distinctive, since it has three positively charged residues ( Figure 7a). Two arginine residues on this putative GAG-binding domain ^{48, 49} , stereochemically equivalent in ZIKV/DENV1-4/JEV/WNV ( Figure 3), was identified by MEPP as having a significant electrostatic polarity reversal after membrane fusion ( Table 1). This residue pair (Arg286–Arg288) remains on the β-sheet post-fusion. This cationic ‘hotspot’ might be the target of small anti-viral anionic peptides ^{31, 32} . A separate study focused on mutations in the DI/DIII linker demonstrated that a compensatory mutation in α3 (DENV-E Q400H) restored virus-like particle assembly disrupted by a mutation (DENV-E Y299F). Interestingly, DENV-E Q400 is not conversed even among DENV serotypes ⁷⁶ , and Tyr299 is distant from α3 ( Figure 7b).

The phylogenetic tree for these flaviviruses derived from the multiple sequence alignment (MSA) of the E-protein shows that TBEV and YFV are related, and distant from ZIKV/DENV1-4/JEV/WNV (see Supplementary material Figure 1). TBEV and YFV were excluded from the MSA. Excluding TBEV and YFV shows that ZIKV, JEV and WNV have a loop near the glycosylation site which is missing in DENV ( Figure 2). However, both ClustalW ⁷⁷ and MAFFT ⁵⁷ ) failed to align the glycosylation residues correctly. Replacing stereochemically equivalent residues (see Methods) corrected this alignment, and also gave a better visualization of conservation and differences ( Figure 3).

A recent study on the DENV3-specific human monoclonal antibody 5J7 demonstrated a very potent neutralizing effect through the binding of envelope proteins (PDB ID:3J6U) ⁵⁰ . Interacting residues were determined based on a distance of 8Å since side-chain densities were not resolved (cryo-EM Fab resolution was 9Å). Table 1 in the study reported that T35 from the heavy chain of 5J7 (PDB ID:3J6UH) interacts with four residues (Q52, Q131, E133, N134) from the DENV-E protein (PDB ID:3J6UC), and with K307 and K308 from another E-protein of the same complex. While T35 was within 10 Å for Q52, Q131, E133 and N134, the data on K307 and K308 could not be reproduced since T35 was found to be at a much larger distance from K308 in all three subunits (see Supplementary material Figure 2). The interacting residues of the heavy chain (PDB ID:3J6UH) and the light chain (PDB ID:3J6UL) with other subunits, as computed in this study (see Supplementary material Table 1 and Supplementary material Table 2, respectively). Distance-sorted interacting residues indicates Thr51 in the DENV3-E protein (chain C) is closest to the heavy chain (H-chain) ( Table 3). This explains the specificity of 5J7 to DENV3, since Thr51 is found only in DENV1 and DENV3. Another interacting residue, Thr223, is not conserved in any other DENV or ZIKV virus ( Figure 2). A different study using only the DENV3 domain III identified K307 and K308 as binding sites for mAb 14A4-8 in DENV3, but also included other domain III residues (K325, A329, G381 and I387) not present in 5J7 binding of DENV3 ⁷⁸ . It was recently shown that a domain III-specific antibody protected mice from ZIKV infection ⁷⁹ .

Spatial congruence of catalytic residues in the active site of functionally equivalent proteins, even with no sequence homology ⁸⁰ , has been long established ⁸¹ . Further, electrostatic potential difference (EPD) ^{52, 53} was also shown to be conserved in cognate pairs of active site residues in these active sites ^{54, 82, 83} . Comparison of apo and holo structures quantifying the spatial and electrostatic perturbations after ligand binding was shown to identify critical catalytic residues in several enzymes ³⁸ .

In the current work, this basic postulate was extended to posit that perturbed residues in viral envelope proteins during fusion with the host membrane are good candidates as epitopes for vaccines (MEPPitope). Specifically, computational methods ^{38, 44, 55} were used to analyze spatial, electrostatic and secondary structure perturbations between a pre- ³⁷ . and post-fusion ¹² DENV2-E protein. These residues are overwhelmingly conserved in ZIKV and all DENV serotypes ( Figure 2), and are known epitopes ^{23, 35, 39– 43} . While perturbation was found to be a good predictor of an epitope, not all epitopes are perturbed. For example, the current study did not identify any residues in the fusion loop, the target of several neutralizing antibodies ^{61, 84– 87} , or Thr51/Thr224 ( Table 3) that is an epitope of a potent neutralizing antibody ⁵⁰ . The structure of ZIKV-E protein with the ligand 2A10G6, a flavivirus broadly neutralizing murine antibody, also reveals the fusion loop as an epitope ⁸⁸ . The hydrophobic fusion loop sequence is highly conserved in all flaviviruses ( Figure 2), demonstrating the importance of sequence alignment as a strategy to identify epitopes ⁸⁹ . The current study identified few perturbed residues in domain III (only His317 and Thr315) as significantly perturbed, consistent with the observation that although antibodies targeted to domain III endow protection and minimize enhancement when present, they are redundant and can be replaced by neutralizing antibodies targeted to other epitopes on the virion ⁹⁰ . This study indicates two residues (Thr32 and Thr40) as a significantly perturbed pair in terms of its electrostatic profile. Thr32 is conserved in all flaviviruses, while Thr40 in all conserved in all except TBEV, where it is the stereochemically equivalent Ser40. There has been no emphasis on these residues as epitopes in previous literature. In summary, the current study presents a computational methodology to extract structural and electrostatic features of envelope proteins that undergo conformational changes during fusion, which correlates well with known epitopes of DENV. Conservation of such residues in ZIKV provides a good strategy to leverage existing knowledge in developing ZIKV specific therapeutics.