An affinity-matured human monoclonal antibody targeting fusion loop epitope of dengue virus with in vivo therapeutic potency

Dengue virus (DENV), from the genus flavivirus of the family flaviviridae, causes serious health problems globally. Human monoclonal antibodies (HuMAb) can be used to elucidate the mechanisms of neutralization and antibody-dependent enhancement (ADE) of DENV infections, leading to the development of a vaccine or therapeutic antibodies. Here, we generated eight HuMAb clones from an Indonesian patient infected with DENV. These HuMAbs exhibited the typical characteristics of weak neutralizing antibodies including high cross-reactivity with other flaviviruses and targeting of the fusion loop epitope (FLE). However, one of the HuMAbs, 3G9, exhibited strong neutralization (NT50 < 0.1 μg/ml) and possessed a high somatic hyper-mutation rate of the variable region, indicating affinity-maturation. Administration of this antibody significantly prolonged the survival of interferon-α/β/γ receptor knockout C57BL/6 mice after a lethal DENV challenge. Additionally, Fc-modified 3G9 that had lost their in vitro ADE activity showed enhanced therapeutic potency in vivo and competed strongly with an ADE-prone antibody in vitro. Taken together, the affinity-matured FLE-targeting antibody 3G9 exhibits promising features for therapeutic application including a low NT50 value, potential for treatment of various kinds of mosquito-borne flavivirus infection, and suppression of ADE. This study demonstrates the therapeutic potency of affinity-matured FLE-targeting antibodies.


Results
Hybridoma preparation and details of HuMAbs. We established eight HuMAb clones from a blood specimen of an Indonesian patient with dengue using SPYMEG cells that belong to a human hybridoma fusion partner cell line 37,38 . The patient was diagnosed with acute dengue fever (2 days after onset); the blood specimen was found to be anti-DENV IgM/IgG-and NS1-positive, indicating a secondary infection. The DENV serotype could not be determined by RT-PCR using RNA extracted from the patient serum 39 . All the HuMAbs were reactive to detergent-inactivated DENV for ELISA, indicating that HuMAbs recognized the E-monomer. All antibodies belonged to IgG1 subtype. Neutralization activity. All HuMAbs exhibited neutralizing activity against all four serotypes of DENV prototype strains and Indonesian isolates ( Table 1, Fig. 1A). DENV-2, in particular, was strongly neutralized; the NT 50 was lower than 0.1 μg/ml. All of the HuMAbs also neutralized JEV. Overall, HuMAbs 1F11 and 3G9 showed stronger neutralizing activity. Thus, these two promising HuMAbs were subjected to neutralization tests  50 values of the HuMAbs. Average of two independent neutralization tests is shown (μg/ml). *The WNV and ZIKV results were determined using SRIPs containing the prM-E genes of each virus. www.nature.com/scientificreports/ using single-round infectious particles (SRIPs) containing prM-E proteins of ZIKV and WNV, due to the unavailability of the infectious viruses 40,41 . Again, 1F11 and 3G9 neutralized ZIKV and WNV particles as efficiently as DENV (Fig. 1B). These data indicated that these HuMAbs were broadly-neutralizing antibodies to mosquitoborne flaviviruses.
Stage of HuMAb neutralization. The flaviviruses can be neutralized by several mechanisms including the inhibition of receptor binding, inhibition of membrane fusion, and aggregation of virion particles 42 . To determine the neutralization mechanism, a time of addition assay using Vero cells and the DENV-2 New Guinea C (NGC) strain was carried out 43 . The antibodies all neutralized DENV-2 during the pre-adsorption assay but not the post-adsorption assay ( Table 2). This suggested that our HuMAbs blocked pre-adsorption steps including viral adsorption but not post-adsorption steps, such as viral membrane fusion or conformation change.
Epitope mapping. The epitope target of an antibody influences its neutralization potency and mechanism. Therefore, epitope mapping was conducted using HEK293 cells transfected with a DENV2 prM-E mutant library, which possess a single alanine substitution at each residue of prM-E (661 mutants in total) 44 . Four antibodies (1C3, 1F11, 3E1, and 3G9) were tested against all the mutants, while the other four antibodies (1C5, 1E5,  www.nature.com/scientificreports/ 2C2, and 2G2) were tested only against selected mutants with an FLE mutation, because of the availability of materials. The data are shown in Supplementary Fig. S1. The critical residues that abolished the binding ability of the HuMAbs tested are presented in Fig. 2 and Table 3. The W101A mutation, a key epitope site for FLE antibodies, was critical to the activity of all eight HuMAbs 26,29,45 . In addition to W101A, L107A and/or F108A were responsible for binding. All of the critical residues were located on the fusion loop of the E protein, which is highly conserved among the flaviviruses. These data substantiated the breadth of neutralizing ability exhibited by the HuMAbs considered in this study. Our antibodies were typical FLE antibodies, which are dominantly produced when a secondary DENV infection occurs. The escape mutant analysis was attempted to further investigate the epitope, by passaging DENV-2 in the presence of the HuMAbs in Vero cells and then sequencing the E protein gene in any surviving viruses. However, we did not identify any mutations in the E protein of surviving viruses. ADE activity. Targeting of the FLE is a typical characteristic of weak neutralizing and high-ADE antibodies 29,30 . We measured ADE activity using semi-adherent K562 cells 46 . Although the HuMAb clones exhibited strong neutralizing activity in Vero cells, strong ADE activity was observed in our assay system (Fig. 3). DENV-2 was neutralized at high concentrations of the HuMAb clones, while the other strains, DENV-1, -3, and -4, were not neutralized. These data suggested that the cloned HuMAbs could contribute to disease enhancing activity.

Recombinant IgG construction and evaluation.
Although the cloned antibodies were potent in terms of low NT 50 in Vero cells, high ADE was observed in K562 cells. However, these antibodies could potentially be used in therapeutic applications, by modifying the effector function and disrupting the ADE activity. Thus, one of three mutations [L234A + L235A (LALA), D265A, and N297A] was introduced to the Fc region of the best neutralizing HuMAb, 3G9, to disrupt the interaction with Fc receptors 47 . We chose to test three mutations, because the effect of Fc-modification during the in vivo animal experiment varied depending on the study   www.nature.com/scientificreports/ group [48][49][50][51][52][53] . These three mutations are described as being the same in terms of the effect on loss of binding to Fc receptors and should not compromise antibody neutralizing potency 54 or shorten the half-life significantly 55 . However, N297A abolishes N-linked glycosylation at N-297, while the LALA and D265A mutants would be fully glycosylated. Additionally, LALA and N297A reduce complement component 1q (C1q) binding, whereas D265A does not compromise complement-dependent cytotoxicity, which is related to virus clearance 20,47 . The difference in the mutations may cause subtle changes in the IgG phenotype, resulting in differences in in vivo efficacy 55 . The recombinant Fc-modified 3G9 antibodies neutralized DENV-2 at comparable level to that of the 3G9-original in Vero cells (Table 4, Fig. 4A). As expected, the recombinant 3G9 antibodies did not show ADE activity in K562 cells, even at a sub-neutralizing concentration (Fig. 4B).

HuMAb protection in vivo.
Because the neutralization and ADE activities in vitro do not always correlate with protection in vivo, we tested the ability of antibodies to protect against DENV infection in animal models. We tested the ability of 3G9, 1F11 (another potent HuMAb), and Fc-modified 3G9 to protect interferon-α/β/γ receptor knockout (IFN-α/β/γR KO) C57BL/6 mice challenged with DENV-3 56 . DENV-3 was chosen because of the availability of the lethal infection mouse model, in which DENV infection causes vascular leakage without showing neurologic disorder 56 . Antibodies (500 μg/mouse) were injected intraperitoneally (i.p.) after one day of virus challenge (Fig. 5A). Mice administered with an isotype control IgG died within 4 days post infection (dpi) (Fig. 5B). 1F11 exhibited almost no protection, while the original unmodified 3G9 significantly prolonged survival, although all mice died within 20 dpi (p < 0.01). All three recombinant 3G9 antibodies significantly enhanced the survival rate when compared with the original 3G9 (p < 0.05). There were no significant differences among the three recombinant antibodies (p > 0.05).  www.nature.com/scientificreports/ To further validate the potency of in vivo protection, another animal experiment was carried out using DENV-2 NGC strain and immunocompetent BALB/c mice treated with mouse monoclonal antibody against type I IFN receptor (Fig. 5C) 57 . Viral load was measured as a readout because this model is not a lethal infection model. There was no significant difference between isotype control and 3G9-original, possibly due to the low amount of administrated IgG (100 μg/mouse) (Fig. 5D). However, 3G9-N297A showed a significant reduction  A group of five mice were infected with 2.0 × 10 6 FFU of DENV-3. Statistical significance was analyzed using a log rank test. *P < 0.01 for the comparison between isotype control and 3G9-original. **P < 0.05 for the comparison between 3G9-original and the Fc-modified 3G9. There was no significant difference among the Fc-modified 3G9 (P > 0.05). (C) Scheme of the animal challenge experiment carried out using immunocompetent BALB/c mice. (D) Viremia levels of the infected mice after the viral challenge. A group of five mice were infected with 1.0 × 10 6 FFU of DENV-2. The limit of detection (L.O.D) was 100 FFU/ml. Viremia levels below the detection limit are plotted as half of the detection limit (50 FFU/ ml). Statistical significance was analyzed using a one-way ANOVA. P < 0.05 was considered significant. n.s. not significant. www.nature.com/scientificreports/ of viral load compared to isotype control (p < 0.01). Although 3G9-N297A did not significantly reduce viral load at 2 dpi (P = 0.11) compared to 3G9-original, the significance was observed at 3 dpi (P = 0.041), demonstrating faster viral clearance. These results indicated that the reduced ADE shown by these modified antibodies promotes their effectiveness in vivo and suggests that recombinant 3G9 antibodies are promising therapeutic agents.
Competition assay using ELISA and ADE assays. Our animal experiments were infection models without ADE. Thus, it remained unclear whether the Fc-modified 3G9 could suppress infection with ADE in vivo. An antibody highly competitive to 4G2 (low-avidity FLE antibody) has been reported to protect mice from lethal DENV infection with ADE 58 . Since we had not established an in vivo ADE infection model, competition assays were conducted. Competitive ELISA showed that 3G9 strongly competed with 4G2 but not with other mouse monoclonal antibodies (7F4 and 15C12, targeting the central part of domain II and the A strand of domain III, respectively) (Fig. 6A) 20 . 3G9 inhibited 4G2 binding by more than 50% when both were applied to the assay at the same concentration. It has been reported that the suppression of ADE caused by pre-existing antibodies in vitro could be an indicator of in vivo protection efficacy 58 . Thus, a competitive ADE assay was performed using 4G2 and DENV2immunized mouse serum, which showed peak levels of ADE at 100 ng/ml and a 1:640 dilution, respectively (Fig. 3, Fig. S2). These dilutions were used, therefore, for the subsequent competitive ADE assay. 3G9-N297A suppressed the peak ADE caused by 4G2 and D2-immunized mouse serum by 50% at only 30 ng/ml (Fig. 6B). A previous study indicated that an antibody that strongly reduced ADE infection (> 50%) at 1000 ng/ml offered good therapeutic efficacy in an in vivo ADE model 58 . The competition intensities found in the current study were higher than the threshold for in vivo protection quoted in the previous report, although direct comparison is impossible.
Immunogenetic analysis. Even though 3G9 is an FLE antibody, it showed high potency in terms of NT 50 values and in vivo protection. Therefore, we analyzed the sequences of the 3G9 VH and VL regions using the IMGT tool to identify the closest VH and Vλ germline genes. The results indicated that the VH gene was derived from IGHV3-23*02 and the Vλ gene from IGLV7-46*01 (Table 5). Somatic hyper-mutation (SHM) rates of the VH and Vλ genes were 14.3% and 7.1%, respectively. 3G9 showed distinct VH germline gene usage and the highest SHM rate of the VH regions than the HuMAbs generated in this study (Table S1). The high SHM rate of the VH regions indicated the levels of affinity-maturation during the secondary infection 50 . The sequences of HuMAbs have been added in Supplementary Table S2. The OD relative to that of the no competition well (without 3G9) is shown. Each data represents the average of two independent experiments. (B) Competition ADE. 100 ng/ml of 4G2 and 1:640 diluted mouse serum, which showed the peak level of enhancement, were mixed with DENV-2 NGC and serially diluted 3G9-N297A. The number of infected cells relative to that of the no competition wells is shown. Each data represents the average of two independent experiments. www.nature.com/scientificreports/

Discussion
To date, no specific therapeutic agent against flavivirus infection has been made available. Considering the current success of antibody therapy against respiratory syncytial virus, Ebola virus, and severe acute respiratory syndrome coronavirus 2 infections, antibody therapy for dengue is a promising target. DENV-neutralizing HuMAbs have been reported using various techniques and patient backgrounds [31][32][33]36,38 . The objective of this study was to establish therapeutic antibody candidates against dengue using the newly developed SPYMEG cell technology 37,38 . Multiple experiments were attempted using blood samples from dengue patients in acute or convalescent phase of infection. We successfully established eight HuMAb clones from a single dengue patient. The patient whose hybridoma was successfully established was in the acute phase of secondary DENV infection, indicating that the B cells were highly activated and IgG repertoire was abundant. The SPYMEG cell fusion method was highly successful when the patient was in the acute infection phase, which was consistent to the results of a previous study 38 . The circulating serotypes and/or genotypes of DENV vary depending on the region and year 59,60 . These antibodies, therefore, could be a potent therapeutic agent for use against DENV strains circulating in Southeast Asia, including Indonesia, which is one of the largest dengue-endemic countries. Several groups have generated potent neutralizing HuMAbs against DENV. In general, highly potent neutralizing antibodies are serotype-specific, bind to E-dimers or quaternary-structure epitopes, and inhibit both pre-and post-attachment steps 61 . In contrast, weak neutralizing antibodies are highly cross-reactive, bind to E-monomer FLEs, and inhibit the only pre-attachment step. Although all the HuMAbs generated in this study recognized typical epitope of weak neutralizing antibodies, they showed considerably low NT 50 values, especially against DENV-2. The NT 50 values were < 0.1 μg/ml, as low as those of previously reported highly potent neutralizing antibodies 61 . However, these values cannot be compared directly, because the neutralizing antibody titer is influenced by the host cell type, degree of viral maturation, virion breathing, and other assay conditions 62,63 . Thus, 4G2 (a low neutralizing FLE antibody) was used to standardize the NT 50 assay. One of the HuMAb clones, 3G9, exhibited a NT 50 that was 1000-fold lower than that of 4G2 against DENV-2 (Table 1). This was comparable to the findings of a previous study on potent neutralizing antibodies 64 , corroborating the suggestion that 3G9 is a promising neutralizing antibody.
High neutralizing potency, especially against DENV-2, is beneficial, because the only licensed dengue vaccine (CYD-TDV) confers low-level protection against DENV-2 65,66 . Cross-reactive neutralization is reported to play an important role in DENV-2 protection, while serotype-specific neutralization is key in protection against other serotypes for those who received CYD-TDV vaccination 67 . The HuMAbs described in this study could compensate for the drawbacks associated with the current dengue vaccine.
3G9 also neutralized other flaviviruses, with NT 50 values of around 0.1 μg/ml. ZIKV is generally less susceptible to FLE antibodies due to high thermostability and less virion breathing 68,69 . A potent neutralizing antibody targeting both DENV and ZIKV could be promising, considering their co-circulation in the environment and the occurrence of ADE in both diseases 10,11 . 3G9 may be applied therapeutically in the treatment of various kinds of mosquito-borne flavivirus infections.
Other groups have reported potent neutralizing antibodies that target FLEs and closely related bc loop. For example, 2A10G6, targeting D98, R99, and W101 motifs within the fusion loop, has a broad neutralizing capability against all four serotypes of DENV and protects mice against lethal challenges from DENV and WNV 70 . 753C12, targeting both the fusion loop and adjacent bc loop, exhibits low NT 50 values 71 . 1C19, targeting only bc loop also exhibits broad and potent neutralization to DENVs 36 . Furthermore, E60, targeting FLE, showed therapeutic potency in vivo 58,72 . The promising results of FLE antibodies with Fc-modification were reported 53,58,73 . These reports substantiate the potential of FLE antibodies.
No viral escape mutant was obtained, even after 5 viral passages in the presence of the HuMAbs in Vero cells. The fusion loop is highly conserved among flavivirus, and an escape mutation at this position would be lethal to the virus. This feature also could be an advantage to the development of therapeutics.
Generally, cross-reactive non-conformational epitopes are less neutralizing, while serotype-specific conformational epitopes are highly potent 61 . Cross-reactive FLE antibodies are induced exclusively at the secondary infection phase 29,30 . Tsai et al. reported that FLE antibodies derived from patients with a secondary infection are more potent than that those from patients with a primary infection 71 . Considering the high SHM rate (14.2%) on the V gene (Table 5), 3G9 was considerably affinity-matured, and, therefore, potent 50 . In addition, as a third or fourth DENV infection is less possible, affinity-matured FLE antibodies play an important role in viral protection 74 .
Although the HuMAbs investigated in this study were promising in intensity and breadth of neutralization, strong ADE was observed. DENV-2 was slightly neutralized at a high concentration of IgG in the ADE assay, coinciding with the high level of neutralization activity in Vero cells. Meanwhile, the other serotypes of DENV were not neutralized. The ADE assay measures the balance of neutralizing and enhancing activities, and, thus, no neutralizing activity was observed 46 . The HuMAb clones were subtyped as IgG1, which show relatively higher levels of ADE activity 75 . These data indicate that modification of the Fc region is mandatory prior to therapeutic application.
Fc-modified 3G9 neutralized DENV at a similar level to the original antibodies, but ADE was not observed, even at sub-neutralizing levels in vitro. In addition, these recombinant antibodies prolonged the survival of the infected mice and reduced viremia level, compared with the original antibody. Some groups have reported that disrupting the Fc-interaction significantly enhances survival rates 50,52,58 , while other groups have demonstrated the attenuation of protection or no significant difference 49,51 . Our study demonstrated an improved survival rate and no difference among the three recombinant antibodies. These data suggested that disrupting Fc-interactions is promising, at least in mice, regardless of the mutations. The results indicated that Fc-modified 3G9 is a good therapeutic candidate. www.nature.com/scientificreports/ While our data suggests the importance of eliminating the ADE function of protective antibodies, a limitation of our study was the lack of animal experiments using an ADE model. Williams et. al. reported that the assay of antibody competition or ability to displace low-avidity FLE antibodies and reduce ADE in vitro could enable the prediction of in vivo therapeutic efficacy with ADE 58 . Our ADE competition assay data revealed that 3G9 competes with 4G2 and polyclonal immunized-mouse serum (Fig. 6). The observed degree of competitiveness was stronger than that shown previously by E60, a potent neutralizing FLE antibody 58 , although direct comparison was impossible. An affinity-matured FLE antibody would be superior to other potent neutralizing antibodies targeting the DI-DII hinge region or DIII, in terms of competing with pre-existing ADE-prone FLE antibodies, which may warrant the suppression of ADE in vivo 58 .
Another limitation of the study is that the DENV circulating in infected humans has a unique morphology, a higher degree of maturity, and less sensitivity to FLE antibodies than those derived from established culture cells 76,77 . It is possible that neutralizing potency of 3G9 is compromised in human infection since 3G9 is an FLE antibody. Although protection in mice may demonstrate therapeutic potency, further neutralizing and ADE tests using DENV in human serum are warranted.
In conclusion, we have described a potent HuMAb, 3G9, which targets FLE. Fc-modified 3G9 displayed neutralizing potency in vivo. The affinity-matured FLE antibody has several features that make it appropriate for therapeutic application, including a low NT 50 value, potential for the treatment of various kinds of mosquitoborne flavivirus infection, competition with pre-existing ADE-prone antibodies, and less potential for viral escape. This study reconfirms the therapeutic potency of affinity-matured FLE antibodies.

Methods
Ethical statement and patient recruitment. Blood samples were collected from patients with dengue at a private hospital in Surabaya, Indonesia. Signed informed consent was acquired from the patients or their parents upon collection of blood samples. This study was approved by the Ethics Committees of Airlangga University (Ethics Committee Approval Number: 24-934/UN3.14/PPd/2013) and Kobe University Graduate School of Medicine (Ethics Committee Approval Number: 784). Diagnosis data were obtained from medical records.
Animal experiments using IF-α/β/γR KO C57BL/6 mice were performed at the National Institute of Infectious Diseases in Japan (NIID), in accordance with the guidelines for the care and use of laboratory animals at the NIID. The study was approved by the Animal Experiment Committee of the NIID (Ethics Committee Approval Number: 115064). Trained laboratory personnel performed the anesthesia of mice via i.p. injection of a mixture of medetomidine, midazolam, and butorphanol during viral inoculation and euthanasia by cervical dislocation.
Animal experiments using immunocompetent BALB/c mice were performed at Kobe University. The study was approved by the Animal Experiment Committee (Ethics Committee Approval Number: P180504). Trained laboratory personnel performed mice anesthesia via isoflurane inhalation during viral inoculation and euthanasia by cervical dislocation.
All the experiments were carried out in accordance with the ARRIVE guidelines (http:// www. nc3rs. org. uk/ page. asp? id= 1357) and all relevant guidelines and regulations. Hybridoma preparation. Approximately 5 ml of blood were obtained from one patient and the peripheral blood mononuclear cells were isolated by centrifugation using a Ficoll-Paque PLUS (GE Healthcare, Uppsala, Sweden). These cells were then fused with SPYMEG cells at a ratio of 10:1, to establish hybridomas, as described previously 37,38 . Cell lines. The SPYMEG cells and established hybridomas were maintained in Dulbecco's modified Eagle medium (DMEM) supplemented with 15% fetal bovine serum (FBS) and 3% BM-condimed (Sigma Aldrich, St. Louis, MO) 18 . Vero cells were cultured in Eagle's minimum essential medium (MEM) supplemented with 10% FBS and 60 μg/ml kanamycin. C6/36 cells were cultured in MEM supplemented with 10% FBS, nonessential amino acids, and 60 μg/ml kanamycin. HEK-293 T cells were maintained in DMEM supplemented with 10% FBS. K562 erythroleukemia cells were cultivated in RPMI 1640 medium supplemented with 10% FBS, 100 units/ ml penicillin, and 100 μg/ml streptomycin. FreeStyle 293-F cells (Invitrogen, Gaithersburg, MD) were cultured in FreeStyle 293 Expression Medium (Thermo Fisher Scientific, Waltham, MA).
Viruses. DENV-1 genotype I, DENV-2 cosmopolitan genotype, DENV-3 genotype I, and DENV-4 genotype II, which are typical isolates found in Surabaya, Indonesia, were used for the antibody characterization 59,60 . Sequence information is available upon request. In addition, the Mochizuki strain of DENV-1, NGC strain of DENV-2, H87 strain of DENV-3, H241 strain of DENV-4, and Nakayama strain of JEV were used as prototype viruses 79 . The DENV-3 strain P12/08, derived from patients infected with DENV-3 in Thailand in 2008, was used in the animal experiments 56 . Viruses were propagated in C6/36 cells and stored at − 80 °C until use.
Hybridoma screening by ELISA. Rabbit serum immunized with the DENV-2 NGC strain was coated onto Nunc Maxisorp ELISA plates (Thermo Fisher Scientific, Waltham, MA). Then, the DENV-2 Indonesian strain (inactivated with Tween 20), hybridoma culture (or DMEM as a negative control), alkaline phosphatase (AP) conjugated anti-human IgG (Abcam, Canbridge, UK), and p-nitrophenyl phosphate (PNPP) (Nacalaitesque, Kyoto, Japan) were serially incubated, and the absorbance was measured at 415 nm. The wells showing higher values than the average + 3SD of the negative controls were considered to be positive. www.nature.com/scientificreports/ Isotyping and quantification of HuMAbs. HuMAbs were isotyped using anti-human IgG1, IgG2, IgG3, or IgG4 by ELISA (Abcam, Canbridge, UK). HuMAbs were coated onto the ELISA plates. Then, murine antihuman IgG (anti-IgG1, IgG2, IgG3, or IgG4), AP-conjugated anti-mouse IgG, and PNPP were serially incubated, and the absorbance was measured at 415 nm. The subclass of each HuMAb was determined following the targeted identification of the first antibody's subclass, which showed the highest value among IgG1 to IgG4. Human IgG was quantified using a Human IgG Quantification kit (RD Biotech, Besançon, France).
Preparation of single-round infectious particles. pCMV-JErep-fullC, a pcDNA3 plasmid containing JEV genes encoding the whole C and all the NS proteins, was transfected into 293 T cells with a prM-E expression plasmid of the WNV NY99 strain or ZIKV MR776 strain, to prepare SRIPs 40,41 . The culture media containing the SRIPs were harvested 3 days post-transfection. The harvested SRIPs were subjected to neutralization tests.
Antibodies and immunized serum. D1-4G2 (anti-E protein, cross-reactive to the flavivirus group; American Type Culture Collection, Manassas, VA) and JE-2D5 (anti-JEV-NS1 protein) were used to detect virus-infected and SRIP-infected cells, respectively (see below) 40 . Two mouse monoclonal antibodies, 7F4 (targeting the central part of DII) and 15C12 (targeting the A strand of DIII), were used for the competition ELISA 20 .
In addition, a mouse polyclonal antibody against DENV-2 (immunized mouse serum) was used for the competition ADE assay. Six-week old BALB/c mice were immunized three times with 100 μg of DNA vaccine (prM-E protein expression plasmid) intratibially, at two-week intervals 78 . Blood samples were collected one week after the third immunization.

Titration of viral infectivity and neutralization test. Infective titers were determined in Vero cells on
a 96-well plate, by counting the infectious foci after immunostaining (see below) and expressed as FFU.
Neutralizing tests were performed as described previously 78 . Briefly, flat-bottom 96-well plates were seeded with Vero cells (2 × 10 4 cells/well). The following day, 100 FFU of virus and serially diluted antibody were mixed and incubated for 1 h at 37 °C, followed by inoculation into the Vero cells. At 24 h post-infection, the cells were fixed and immunostained (see below). The neutralizing antibody titer was expressed as the minimum IgG concentration yielding a 50% reduction in focus number (NT 50 ).
Immunostaining. Immunostaining was performed as described previously 78 . Briefly, infected cells were fixed with acetone-methanol (1:1). These cells were incubated serially with the antibodies described above, biotinylated anti-mouse or -human IgG, a VECTASTAIN Elite ABC kit (Vector Laboratories, Burlingame, CA), and a VIP peroxidase substrate kit (Vector Laboratories, Burlingame, CA).

Time of addition assay.
A time of addition assay was carried out, as described previously 43 . For the preadsorption assay, approximately 100 FFU of virus were pre-incubated with serially diluted HuMAbs for 1 h at 4 °C and then inoculated onto 2 × 10 4 Vero cells on a 96-well plate. Next, the unadsorbed viruses and excess antibodies were washed out with PBS. The cells were then incubated for 24 h at 37 °C, followed by immunostaining and focus counting.
For the post-adsorption assay, virus was added directly to the cells for 1 h at 4 °C. Then, unadsorbed virus was removed by washing the cells with PBS three times, and bound virus was incubated with serially diluted HuMAbs for an additional hour at 4 °C. The cells were then incubated for 24 h at 37 °C, followed by immunostaining and focus counting. The results are expressed in the same way as for the neutralization assay.
Epitope mapping. Epitope mapping was conducted as reported previously 44 . A DENV2 (strain 16681) prM-E expression construct was subjected to high-throughput "Shotgun Mutagenesis, " generating a comprehensive mutation library. Each prM-E residue of the construct was changed individually to alanine (alanine residues to serine). In total, 661 DENV2 mutants were generated (100% coverage of prM-E). HEK-293 T cells were transfected with an expression vector for DENV-2 prM-E or its mutants, fixed with 4% paraformaldehyde, and intracellular MAb binding was detected using a high-throughput immunofluorescence flow cytometry assay. Antibody reactivity against each mutant protein clone was calculated relative to the reactivity of the wild-type protein. Each raw data point was background-subtracted and normalized to the value for reactivity with wildtype DENV2 prM-E. Mutations within clones were identified as critical to the MAb epitope if they did not support reactivity of the MAb (< 20% of the MAb's reactivity with wild-type prM-E) but did support reactivity of other conformation-dependent MAbs (> 70% of reactivity with wild-type).
Generation of escape mutants. The DENV-2 NGC strain was passaged 5 times in the presence of the HuMAbs in Vero cells. The concentration of HuMAbs was increased with passages, starting from NT 50 to 5 times NT 50 . Surviving viruses were sequenced in the E region. ADE assay. ADE activity was measured using semi-adherent K562 cells and expressed as the number of infected cells 46 . Briefly, serial four-fold dilutions of antibody samples were incubated with 100 FFU virus for 2 h at 37 °C in 96 well poly-L-lysine coating plates. The mixture was mixed with 1 × 10 5 K562 cells and incubated for a further 2 days. After immunostaining, viral foci were counted manually. The baseline of the infected cells (without antibody) was 2.0 (100 FFU). The infected cell number fell lower than 2.0 when the virus was neutralized but rose higher when ADE occurred. www.nature.com/scientificreports/ Generation of Fc-modified recombinant antibodies. RNA was extracted from 5 × 10 6 to 10 × 10 6 hybridoma cells using TRIzol reagent (Invitrogen, Gaithersburg, MD). Heavy (H)-and light (L)-chain cDNAs containing the gene encoding the antibody-binding (Fab) region of 3G9 were amplified and sequenced, as previously reported 79 . Then, primer sets were designed to clone the Fab regions of H and L chains into pFUSE-hIgG1-Fc1 and pFUSE2-CLIg-hL2, respectively (InvivoGen, San Diego, CA). Gene cloning was performed following the manufacturer's instructions (primer information is available upon request). Then, three kinds of mutation [L234A/L235A(LALA), D265A, and N297A] that abolished Fc-Fc receptor interaction were introduced into the Fc region using a site-directed mutagenesis kit, following the manufacturer's protocol (TOYOBO, Osaka, Japan) 47 .
Competition assays. Competition ELISA: The DENV-2 NGC strain was coated on ELISA plates, as described above. Then, 1 μg/ml of each type of mouse monoclonal antibody (4G2, 7F4, or 15C12) was mixed with serially diluted 3G9 (four tenfold dilutions starting at 10 μg/ml) in a separate 96-well plate, and 100 μl of the mixture were added to each ELISA plate. The plates were then incubated with AP-conjugated anti-mouse IgG, followed by color development with PNPP. The relative optical density (OD) was expressed as the average OD of each sample divided by the OD of the non-competition control well (without 3G9).
Competition ADE: Serially diluted 3G9-N297A (eight twofold dilutions starting at 2000 ng/ml) solutions were prepared in a separate 96-well plate, followed by the addition of 4G2 or DENV-2 immunized to the concentration that showed peak enhancement of DENV-2 infection in K562 cells. Then, 36 μl of the mixture were used for the ADE assay, as described above 46 . Relative infection was expressed as the average number of infected cells in each sample divided by the number of infected cells in the non-competition control well (without 3G9-N297A, around 1000 cells).

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.