As part of the prospective Early Dengue (EDEN) infection and outcome study in Singapore (25), adult patients (age >21 years) presenting at community primary care polyclinics with acute onset fever (>38.5°C for less than 72 h) without rhinitis or clinically obvious alternative diagnoses, were included in the study. For the first cohort, a total of six whole blood samples were collected from each volunteer into EDTA-vacutainer tubes (Becton Dickinson) at recruitment (acute phase) and subsequently over a period of 1 year after disease onset (Table 1). For the second cohort, a total of two whole blood samples were collected into EDTA-vacutainer tubes during the acute phase of the disease and 3–4 months later (Table 2). Patients were diagnosed by DENV-specific RT-PCR (26) and by NS1 test (SD Bioline). The commercially available PanBio Indirect ELISA kit (Inverness Medical, Australia) was used to classify patients experiencing a secondary infection based on their IgG seropositivity in the sample collected within 72h after fever onset.

This study was conducted according to the principles expressed in the Declaration of Helsinki. The research, which involved fever patients enrolled in the study was approved by the Domain Specific Review Board of Singapore’s National Healthcare Group (Domain E) and patients gave written informed consent (DSRB Refs. B/05/013, 2010/00227, and 2011/01770).

Maxisorp plates (Nunc) or half-area plates (Greiner) were used for all ELISAs. For the detection of virus particle-specific antibodies, ELISA plates were coated with PEG-precipitated and UV-inactivated DENV serotypes 1–4. For E protein-specific ELISA, the soluble part of E proteins of DENV1-4 with a His-tag were produced in S2 cells as described previously (27). For EDIII-specific ELISA, EDIII of DENV1-4 with a His-tag was produced in S2 cells, using the strategy described for Yellow Fever Virus in Ref. (28). Proteins were purified using Ni-beads. Half-area ELISA plates (Greiner) were coated at 75 ng/well. For all ELISAs, plates were blocked with PBS, 0.05% Tween 20 (PBST) and 3% skimmed milk. Sera and a standard containing pooled dengue-IgG positive plasma were diluted in blocking buffer on the coated plates for 1 h at RT before washing with PBST. Threefold serial dilutions from 1:200 to 1:48600 were tested. To standardize results between plates, a positive control was included on every plate and a ratio was calculated as follows: ratio of sample x = (OD_{450(sample x)} − OD450_(blank))/(OD_{450(positive control)} − OD450_(blank)). The ratio of the positive control is therefore 1 for all ELISA graphs. For clarity, the ratio of only one serum dilution (1:600) per sample and time point is shown in all Figures. Pooled serum from dengue-naïve healthy donors was used as a negative control and included on all plates. Anti-human IgG-HRP (Sigma) was added at a concentration of 1:2000 and incubated for 1 h at RT. After washing, 3,3,5,5-tetramethylbenzidine HRP substrate solution (Sigma) was added. The color reaction was stopped with 1 M HCl.

For the measurement of neutralization, heat-inactivated plasma samples, heat-inactivated pooled dengue-naïve plasma as a negative control and antibody 4G2 as a positive control were serially diluted in RPMI medium in 96-well plates (serum was diluted threefold over six wells starting with 1:200, mAb 4G2 starting concentration was 5 μg/ml) and a constant amount of virus (MOI 0.1–1) was added. The antibody-virus mixtures were incubated at 37°C for 30 min and then 50 μl of the mixtures were transferred to another 96-well plate containing 200,000 U937 cells (ATCC) stably transfected with human DC-SIGN (MOI 0.1-1) per well. After 2 h of infection, 150 μl RPMI medium containing 10% FCS was added. After incubation over night, the infected cells were fixed and stained intra-cellularly with 4G2-Alexa 647. The percentage of infected cells was quantified by flow-cytometry and data were analyzed with GraphPad Prism software for the calculation of the 50% neutralizing titer (NT50).

All viruses used were produced in C6/36 mosquito cells (ATCC). The following patient isolate strains were used: DENV1-05K2916 [EU081234 (29)], DENV2-TSV01 (30), DENV3-VN32/96 (EU482459), and DENV4-2641Y08 (isolated by Environmental Health Institute, Singapore). DENV3-VN32/96 was a gift from Dr. Cameron Simmons, Oxford University Clinical Research Unit, Viet Nam.

White blood cells were isolated from 8 to 10 ml whole blood using CPT (BD) tubes containing Sodium Citrate as an anti-coagulant. Cells were counted and one million cells were stained directly for flow-cytometry analysis or all cells were frozen in RPMI medium containing 20% FCS and 10% DMSO. Frozen cells were thawed in batches for flow-cytometry analysis. Antibodies were purchased from Biolegend, except for anti-CXCR5-biotin (BD Pharmingen) and anti-CD14-ECD (Beckman Coulter). CXCR5-biotin was labeled with Streptavidin-APC-Cy7 (Biolegend). Cells were stained on ice for 30–60 min and washed with FACS buffer (PBS, 2% FCS, 5 mM EDTA, 0.1% sodium azide). Fixable blue live-dead marker (Invitrogen) was added freshly to the antibody mix and was incubated together with the surface marker antibodies. Cells were washed and fixed with 1% formalin for analysis.

Data were initially analyzed with transMART (31). Data sets found to be significant were further analyzed with Prism Software (version 6 for Mac) and the statistical tests used were indicated in the figure legends. A p value of <0.05 was considered statistically significant.

The DENV-neutralizing activity of patient plasma is the average of binding capacities of a mixture of antibodies with different specificities. Antibodies binding to E protein epitopes and possibly also to prM epitopes can potentially be neutralizing (4, 32). With the aim to better define the specificity of polyclonal antibodies during disease and after recovery, longitudinal samples from 13 patients with primary DENV-2 infection and from 12 patients with secondary DENV-2 infection were collected over a period of 1 year, with the first blood sample being drawn within 72 h after fever onset (Table 1, Cohort 1). Neutralizing antibodies to all four DENV serotypes were quantified for nine primary and nine secondary patients with a flow-cytometry-based assay using U937-DC-SIGN cells. All longitudinal samples per patient were measured together in the same experiment. Since acute sera are highly neutralizing the lowest serum dilution used was 1:200 and the assay was not designed to accurately determine NT50 titers below this dilution. Whereas the response to DENV-2 dominated in both primary and secondary patients, cross-reactivity was observed particularly for the secondary patient group, and this cross-reactivity lasted for up to 1 year after the infection (Figure 1A). Cross-neutralizing activity of polyclonal antibodies is well described (33) and is thought to contribute to protection in individuals with one or multiple previous infections (9, 34), yet the epitopes of the cross-reactive polyclonal antibodies during the period of cross-protection are less well defined. We analyzed the specificity of DENV-binding polyclonal antibodies in plasma using ELISAs coated either with recombinant E protein or with PEG-precipitated, UV-inactivated virus particles (UV-DENV) (Figures 1B,C). As for the neutralization assay, all longitudinal samples per patient were analyzed together in the same experiment. An SDS-PAGE analysis of the recombinant DENV-2 E protein showed two bands, a smaller band for the monomer (40 kDa) and a less prominent band for the dimer (80 kDa) (Figure S1 in Supplementary Material). Recombinant E protein can spontaneously form dimers via the dimerization domain EDII (35) and both monomer- and dimer-specific antibodies are possibly detected in the E ELISA. In contrast to antibodies binding to recombinant E protein, antibodies binding to UV-DENV particles can be specific for E protein monomers, dimers, or for quaternary structures spanning more than one E dimer (36). We found that antibody titers to UV-DENV and to E protein both peaked between day 15 and 25 after onset of fever for the serotype of infection, DENV-2 (Figure 1B), and that antibodies cross-reacting with DENV-1 showed a similar kinetics (Figure 1C). Interestingly, titers against E protein of DENV-2 and DENV-1 increased more rapidly than titers against UV-DENV-2 and UV-DENV-1 in secondary patients.

These data showed that cross-reactive antibodies targeted at E protein increased during acute infection and dropped to almost baseline within 1 year. The different early kinetics between E- and UV-DENV-specific titers further suggested that a majority of B cells secreting antibodies during the acute infection were specific for the E protein monomer or dimer, and that less B cells seemed to be specific for complex viral epitopes, which would only be detected in the UV-DENV ELISA. Similar to the kinetics of the ELISA, both the specific and cross-reactive neutralizing response declined within 1 year.

Patients were classified as dengue-naïve (primary infection) or dengue-immune (secondary infection) based on the pre-existing IgG titer measured in the <72 h sample using the commercially available PanBio indirect IgG ELISA kit (37). Data obtained with our in-house E protein ELISAs in Figure 1 showed that titers specific for E protein increased rapidly between the first (<72 h) and the second time point (4–7 days). This indicated that the early anamnestic response would be more accurate than pre-existing titers to distinguish dengue-naïve from DENV-immune patients and that the E protein ELISA could be more sensitive than the UV-DENV ELISA in distinguishing patients with primary or secondary infections. To test this hypothesis, we compared titers from the two early time points measured with both E- and UV-DENV ELISA. We first compared OD ratios measured in the UV-DENV ELISA <72 h after fever onset with E ELISA titers at the same time point and with UV-DENV ELISA titers measured 4–7 days after fever onset (Figure 2A). UV-DENV OD ratios measured <72 h after fever onset did not show a clear cut-off between primary and secondary patients (as classified based on PanBio kit) and the samples did not cluster at the 4–7 days time point when arranged in the same sequence (Figure 2A). In contrast, when the UV-DENV and E ELISA OD ratio was measured at 4–7 days (Figure 2B), 11 out of 12 secondary cases (patient ID on x-axis in red, based on PanBio kit) had significantly higher E ELISA OD ratios compared to primary cases (patient ID on x-axis in black, based on PanBio kit). The segregation into primary and secondary groups was evident for both DENV-2 and DENV-1 E protein ELISA, showing that day 4–7 antibodies were largely cross-reactive.

In summary, while E and UV ELISA titers were comparable at the <72 h time point, the E ELISA appeared to be more sensitive than the UV-DENV ELISA in distinguishing primary from secondary dengue patients when measured at defervescence (4–7 days after fever onset), in line with the finding that the majority of the anamnestic response seemed to be specific for E protein monomers and dimers.

Several recent publications suggested that EDIII-specific antibodies, although often highly neutralizing and serotype-specific, are rare in the repertoire of dengue-immune individuals (14, 38). We assessed whether, despite their low relative abundance, EDIII-specific antibodies correlated with NT50 or E ELISA titers in our cohort. No convincing correlations between EDIII titers and E titers or between EDIII titers and NT50 were observed (data not shown). Surprisingly, neither primary nor secondary patients generated a strong long-lasting response against DENV-2 EDIII, in contrast to the robust DENV-2 E protein-specific response observed. Patients with a transient DENV-2 EDIII-specific response were found mostly in the secondary infection group (Figure 3A). Similarly, a transient DENV-1 EDIII-specific response was observed in the secondary patient group, and the samples that were positive for DENV-1 EDIII were also positive for DENV-2 EDIII (Figure 3B), suggesting that mostly conserved epitopes in EDIII accounted for the titers. Interestingly, one primary patient developed high titers of DENV-1 EDIII-specific antibodies after DENV-2 infection. The response was highly specific for DENV-1, with no detectable binding to DENV-2 EDIII. Even though a more sensitive Taqman qRT-PCR (39) was performed in addition to the SYBR-green based serotype-specific qPCR we could only detect DENV-2 – and no DENV-1viral RNA in the plasma of this patient. A potential alternative explanation for the DENV-1 specific response could be a molecular mimicry of EDIII with a self-or non-self antigen.

In summary, EDIII-specific serum antibodies were mostly cross-reactive and did not correlate with E protein-specific antibody titers. The EDIII-specific response seemed to vary substantially between patients. The reason for this is unclear, but it is an important point to consider when studying small numbers of patient samples.

Since both the E protein-specific titers and NT50 started to increase as early as 4–7 days after infection (Figure 1) we tested whether there was a correlation between these two readouts, and whether the correlation was consistent over time (Figure 4). For DENV-2, the serotype of infection, a significant correlation between NT50 and E ELISA titers was first seen 15–25 days after fever onset and this correlation was consistent for at least 1 year. In contrast, the correlation between DENV-1 NT50 and DENV-1 E ELISA was only significant 15–25 days after fever onset and not for the later time points (Figure 4A). Only NT50 values calculated from a neutralization curve with a good fit (R² value) were included in the analysis. Day 4–7 time point samples were difficult to analyze in the DENV-1 neutralization assay, even when starting with a lower serum dilution. As a consequence of this the correlation for DENV-1 at 4–7 days did not contain enough data points for interpretation and was only included for the sake of completeness (Figure 4B). However, we could conclude that during early convalescence (15–25 days), E protein-specific DENV-1/2 cross-neutralizing antibodies were likely responsible for the significant correlation between NT50 and ELISA for both homologous and heterologous serotypes. At late time points DENV-1 cross-neutralizing activity and DENV-1 “cross-binding” ELISA titers appeared to decline with different kinetics, resulting in a low correlation between DENV-1 NT50 and DENV-1 ELISA. The correlation in Figures 4A,B included both primary and secondary patients because the number of samples was too small to analyze the two groups separately. To address different correlation outcomes for primary and secondary patient due to different levels of cross-reactivity we compared E ELISA titers measured at the most cross-reactive time point (15–25 days) with E ELISA titers measured at later time points (Figure 4C). Primary patients showed a consistently high correlation between DENV-2 E protein-specific antibodies measured at day 15–25 and DENV-2 E protein-specific antibodies measured at later time points, and the correlation between DENV-2 E protein-specific antibodies measured at day 15–25 and DENV-1 E protein-specific antibodies at later time points was only slightly less consistent. In contrast, secondary patients’ DENV-2 E-specific antibodies measured at day 15–25 did not correlate with DENV-2 and DENV-1 E-specific antibodies measured at late time points (Figure 4C). We concluded that while the cross-reactivity correlations were markedly different between primary and secondary patients the readouts for DENV-2 and DENV-1 ELISAs were similar, suggesting that the different correlations associated with DENV-2 and DENV-1 reactivity observed in Figure 4A were a consequence of changes in the NT50.

In summary, the anti-E protein antibody response in primary patients measured 20 days to 4 months after infection correlated with the response measured 1 year after infection. In contrast, the response in secondary patients measured 20 days after infection was not predictive of the outcome measured 1 year later, suggesting that the E protein-specific cross-reactive antibody response waned within 1 year and that the response became more serotype-specific within this time frame. Correlations of NT50 and ELISA data might be useful to assess serotype-specific and cross-protective immunity as early as 4 months after infection (Figure 4A). However, since the patient cohort analyzed here only included DENV-2 infected patients, it will be important to test whether a similar correlation can be observed in a cohort with different serotypes of previous and new infection.

We next tested, which B cells could account for the rapid increase in DENV-specific antibody titers. Since primary patients do not have memory B cells specific for dengue we assumed that more naive B cells would be activated in primary patients compared to secondary patients, and that the higher antibody titers in secondary patients would be contributed mostly by memory B cells since memory B cells differentiate into plasmablasts more efficiently than naïve B cells (40). We analyzed blood B cells from patients during acute disease (between day 4–7 after fever onset, visit 1) and 3–4 months later (visit 2) by flow-cytometry and measured changes in absolute numbers for naïve B cells, memory B cells and plasmablasts (Table 2, Cohort 2 and Figure 5A). As reported previously by us and by others, a significantly higher number of plasmablasts was detected during visit 1 compared to visit 2, and the number of white blood cells in dengue cases was significantly lower during visit 1 compared to visit 2 (41, 42) (Figure 5B). In the acute phase we also observed a significant drop in memory B cell numbers, one hypothesis for which is that a majority of memory B cells had differentiated into plasmablasts (Figures 5A,B). Interestingly, decreased memory B cell numbers during acute disease were observed for both primary and secondary patients (Figure 5C), suggesting that a large proportion of the activated memory B cells were not DENV-specific. To address whether re-activated memory B cells during secondary infection accounted for long-lasting DENV-specific titers, we compared plasmablast numbers during acute disease with titers measured 4 month after infection (Figure 5D). While a weak correlation existed between plasmablast numbers and DENV-specific titers for secondary patients during acute disease and 4 months after infection, no such correlation was observed for primary cases. This finding indicated that primary infection plasmablasts were activated as “bystanders” and that the number of plasmablasts did not correlate with DENV-specific titers at a later time point.

Fas-receptor (CD95) was significantly up-regulated on both naïve and memory B cells at visit 1 compared to visit 2, exposing the cells to Fas-mediated killing (Figures 6A,B). Loss of CXCR5 and up-regulation of CD95 occurred in parallel, suggesting that the majority of activated B cells did not home back to lymphoid tissues and were prone to die (Figure 6A).

Taken together, dengue infection seemed to systemically activate both naïve and memory B cells based on the increase in CD95-expressing cells and based on the reduction in total cell numbers for both populations.