PBS (phosphate buffered saline), Tween 20, 4-nitrophenyl phosphate (pNPP) and goat antibodies to human IgMs were purchased from Sigma-Aldrich; bovine serum albumin (BSA) from Roche; low-fat milk powder from Regilait; Maxisorp ELISA plates from NUNC. Buffer A contained 0.1% Tween 20 in PBS; buffer B, 5% (w/v) low-fat milk powder in buffer A; buffer C, 1% (w/v) low-fat milk powder in buffer A; buffer D, 10% diethanolamine, pH 9.8, 10 mM MgSO₄, 20 μM ZnCl₂.

The plasmids encoding the H6-ED3-PhoA hybrids have been described 41 . Table 1 gives the origin of the viral ED3 domain and the corresponding segment of the envelope protein. Table 2 gives the number of residue changes between the ED3 domains of any two DENV serotypes, and also between the ED3 domain of any DENV serotype and the consensus ED3 domain (DENVc) 63 64 . The productions of the H6-ED3-PhoA hybrids in the periplasmic space of E. coli and their purification from periplasmic extracts through their His-tag were performed essentially as described 41 . The fractions of purification were analyzed by SDS-PAGE in reducing conditions. The purest fractions were pooled, snap-frozen and kept at −80°C. They were homogeneous at >95%.

The last column gives the residues of the viral E protein present in the H6-ED3-PhoA hybrid. The codons in the recombinant genes were synonymous but not necessarily identical with those in the original viral genomes.

DENVc, consensus ED3 domain.

A first set of human serums (Group 1) was collected within the normal activity of the National Reference Center (NRC) for Arboviruses, Institut Pasteur de la Guyane, French Guiana. These serums were collected from patients who displayed clinical symptoms of dengue and whose infection by DENV was confirmed by laboratory methods. A second set of human serums (Group 2) was collected in the context of a clinical study (DENFRAME project) that was performed in French Guiana 65 . These serums were collected from patients who displayed clinical symptoms of dengue but were diagnosed as negative for DENV infection. In the following, we designate these serums as DENV-uninfected serums. Both Group 1 and Group 2 serums consisted of a series of blood samples collected for the first one during the viremic phase of the disease (from day 0 to day 4 after fever onset) and for a second one during the early convalescent phase (day 5 or later). The samples were characterized by standard diagnostic methods that included virus isolation on mosquito cells and/or viral RNA detection by RT-PCR, non-structural NS1 protein detection, as well as MAC-ELISA and IgG-specific indirect ELISA using virus-infected suckling mouse brain (SMB) extracts as antigens. For the serums of Group 1, the DENV serotype that was responsible for the disease, was identified by RT-PCR and the presence of IgMs against the infecting DENV serotype was ascertained from a MAC-ELISA that used the corresponding SMB extract as antigen. None of the serums was infected simultaneously by two or more DENV serotypes. The serums of Group 1 scored as positive for IgG to DENV in an indirect ELISA performed on an additional sample collected during the convalescent phase of the disease (day 15 or later). For the serums of Group 2, all the tests were negative. The methodologies for the collection of the serum samples, the collection of the associated clinical data, and the characterization of the serums have been described previously in detail 65 . Data on the primary or secondary nature of the infection were not available. We obtained informed consent from the patients for the use of the Group 2 serum samples in a previous clinical study, as described 65 . The constitution of the above human serum collections (Group 1 and Group 2) and their use for the present study were approved by the Institutional Review Board of Institut Pasteur and a regional ethical committee (Comité de Protection des Personnes, Ile-de-France 1).

The MAC-ELISAs of the present study were performed in 96-well flat-bottom microtiter plates with a volume of 100 μL/well. The plates were sensitized with antibodies to human IgMs as follows. A goat antibody to human IgMs (1.0 μg/mL in PBS) was loaded in the wells of the plate. The plate was incubated overnight at 4°C for the reaction of adsorption. The wells were washed with buffer A (three times), blocked with buffer B for 1 h at 37°C, and then washed as above. The serums and recombinant antigens were diluted in buffer C. The serums were diluted 100-fold, a dilution at which they nearly saturate the IgM binding sites in the sensitized wells 66 . Wells were loaded with the diluted serums or with buffer C as a blank sample, and the plate was incubated for 1 h at 37°C for the reaction of antibody capture. The wells were washed as above and then loaded with the solution of recombinant antigen. The plate was incubated for 1 h at 37°C for the binding reaction. The wells were washed as above and the bound antigen, (H6-ED3-PhoA)₂, was detected by addition of 5 mM pNPP in buffer D and measurement of A _405nm after 3 h at 25°C. Each experimental data point was performed at least in duplicate and the corresponding signals were averaged. The serum specific signal was obtained by subtracting the signal of the blank sample from the signal of the serum sample.

The curve fits were performed with Kaleidagraph (Synergy Software), which gives Pearson’s coefficient of correlation, R _P. The mean values, standard deviations (SD) and standard errors (SE) were calculated with the same program. The results of the MAC-ELISAs were analyzed through ROC curves, which are equivalent to Wilcoxon statistics 67 68 . A ROC curve relates the False Positive Fraction (FPF = 1 - specificity) to the True Positive Fraction (sensitivity) of a test when the threshold varies. The area under the ROC curve (AUC) is an unbiased measure of the test accuracy and the difference AUC - 0.5 is the discrimination power or more simply discrimination of the test. An AUC value of 1.0 represents a perfect test whereas a value of 0.5 represents a worthless test. A rough guide for classifying the accuracy of a test has been proposed: 0.90-1.0 = excellent, 0.80-0.90 = good, 0.70-0.80 = fair, 0.60-0.70 = poor, 0.50-0.60 = fail. Semi-parametric ROC curves were computed with the LABROC4 program 69 as implemented in the Web based calculator JLABROC4 70 .

To analyze the serotype specificities and cross-reactivities of IgMs directed against the ED3 domains of the dengue viruses, we used a collection of 179 well-characterized human serums (see Methods). These serums belonged to two groups. The first group included four categories of infected serums, i.e. 18 serums infected by DENV1, 24 by DENV2, 18 by DENV3 and 30 by DENV4 respectively, for a total of 90 DENV-infected serums. Samples of these serums reacted positively in a MAC-ELISA that used a whole homotypic DENV antigen. The second group contained 89 DENV-uninfected serums. In a first step, we assayed each of the 179 serums in duplicate in four MAC-ELISAs that used four (H6-ED3-PhoA)₂ dimers as antigens. The ED3 domains of these four reagents came from each of the four serotypes of DENV and we will designate these reagents as R1 to R4 in the following. The mean signal values and associated SD and SE of the MAC-ELISAs performed with R1 to R4 on the five categories of serums above are reported in Table 3. In a second step, we combined the serums in different sets that we considered as positive (+) or negative (−) for the hypothesis under test and analyzed the results of the MAC-ELISAs with ROC curves. More specifically, we derived a unique parameter for each test from these analyses, the precision (AUC) or discrimination (AUC – 0.5), and compared them between tests.

R1 to R4, (H6-ED3-PhoA)₂ dimers of serotypes DENV1 to DENV4, respectively. The first and second columns give the serotype of the infecting DENV and the number of samples for each category of serums. None, DENV-uninfected serums. The other entries give the mean A_405nm value and associated SD and SE (in parentheses) for MAC-ELISAs performed on the serums of the first column with reagents R1 to R4.

In a first step, we analyzed the capacity of the four reagents, R1 to R4, to discriminate between IgMs of patients infected by a DENV serotype, either homotypic or heterotypic to the reagent, and IgMs of DENV-uninfected patients. The serums of the infected patients were considered as positive and the serums of the uninfected patients as negative. The discrimination between the IgMs of the infected serums and the IgMs of the uninfected serums was the highest when the reagent and infecting DENV were homotypic. These homotypic discriminations were excellent for DENV1, good for DENV2, fair for DENV3 and poor for DENV4 (diagonal of Table 4). Therefore, the serums of patients infected by any of the DENV viruses formed IgMs against the homotypic ED3 domain and could be recognized by the homotypic reagent.

R1 to R4, (H6-ED3-PhoA)₂ dimers of serotypes DENV1 to DENV4, respectively. The first column gives the serotype of the infecting DENV for the (+) serums. DENV1-4, the serotype of the infecting virus could be any one of DENV1 to DENV4. The DENV-uninfected serums were always taken as (−) serums. Each entry gives the accuracy and associated standard error of a MAC-ELISA that used the R_i reagent (i = 1, …, 4) as an antigen and was evaluated on the (+) and (−) serums.

Reagent R1 discriminated between the IgMs of DENV2-, DENV3- or DENV4-infected serums and IgMs of DENV-uninfected serums with excellent (DENV2), good (DENV3) or fair (DENV4) accuracies (column 1 of Table 4). Therefore, the IgMs of serums infected by DENV2, DENV3 or DENV4 cross-reacted with the ED3 domain from DENV1 (ED3.DENV1). By a similar reasoning, we concluded that the IgMs of serums infected by DENV1 or DENV3 cross-reacted with ED3.DENV2 but not the IgMs of serums infected by DENV4 (column 2 of Table 4). The IgMs of serums infected by heterotypic DENV did not cross-react significantly with ED3.DENV3 and ED3.DENV4 (columns 3 and 4 of Table 4). We concluded that there exist inherent cross-reactivities between the ED3 domain of a given DENV serotype and the human IgMs directed against DENV heterotypes.

Table 4 enabled us to compare the accuracies of MAC-ELISAs performed with R1 to R4 when run on the same two sets of serums, infected and uninfected. For example, the discrimination between the DENV1-infected serums and the DENV-uninfected serums was excellent for R1, good for R2, poor for R3 and nil for R4 (row 1 of Table 4). Similarly, the discrimination between the DENV2-infected and DENV-uninfected serums was excellent for R1 and R2 and nil for R3 and R4 (row 2 of Table 4). The discrimination between DENV3-infected and DENV-uninfected serums was good for R1, fair for R2 and R3, and nil for R4 (row 3 of Table 4). The discrimination between DENV4-infected and DENV-uninfected serums was fair for R1, nil for R2 and R3 and poor for R4 (row 4 of Table 4). The discrimination between DENV_j-infected serums and DENV-uninfected serums by the R_i reagent was not linearly correlated with the number n_i,j of residue changes between ED3.DENV_i and ED3.DENV_j (i ≠ j) (R _P = 0.54). These comparisons suggested that the levels of discrimination depended on the specific couple of DENV serotypes and not on the sequence differences between the heterotypic ED3 domains. Whether and how the levels of discrimination of the same two sets of (+) and (−) serums by the different reagents might be related to the strengths of the IgMs reactivities is considered in the Discussion section.

The homotypic R_i reagent and some heterotypic R_j reagents could discriminate between DENV_i-infected serums and DENV-uninfected serums (i ≠ j). How did these homotypic and heterotypic discriminations compare? The discriminations by R1 were higher than or equal to those by the homotypic reagents for any infecting serotype, DENV2, DENV3 or DENV4 (compare the numbers in column 1 and in the diagonal of Table 4). The discriminations by R2 were roughly equal to those by the homotypic reagents in the same conditions. In contrast, R3 and R4 did not recognize serums that were infected by heterotypic DENVs. These results were not due to the serums and their IgMs since the serums were identical for the four reagents. They were necessarily due to differences in the antigenic properties of the ED3 domains of the four DENV serotypes since the (H6-ED3-PhoA)₂ constructions were exactly identical except for this domain.

In a second step, we analyzed the accuracy with which a given R_i reagent discriminated between DENV_i- and DENV_j-infected serums, with i ≠ j (Table 5). The R1 reagent discriminated between the DENV1-infected serums and the DENV2-, DENV3- or DENV4- infected serums with good accuracies (column 1 of Table 5), despite important cross-reactions between ED3.DENV1 and the IgMs of serums infected by heterotypic DENVs (column 1 of Table 4). The R2 reagent discriminated between DENV2-infected serums and either DENV3- or DENV4-infected serums with fair to good accuracies but it did not discriminate between DENV2- and DENV1-infected serums (column 2 of Table 5), in accordance with the cross-reaction pattern (column 2 of Table 4). The R3 reagent discriminated between DENV3-infected serums and either DENV1- or DENV2-infected serums with poor to no accuracies (column 3 of Table 5), despite weak cross-reactions between ED3.DENV3 and the IgMs of DENV1- or DENV2-infected serums (column 3 of Table 4). In contrast, R3 discriminated between DENV3- and DENV4-infected serums with a good accuracy, in accordance with the absence of cross-reactions between ED3.DENV3 and the IgMs of DENV4-infected serums. The R4 reagent did not discriminate between DENV4- and DENV1-infected serums (column 4 of Table 5), despite the absence of cross-reactions between ED3.DENV4 and the IgMs of DENV1-infected serums (column 4 of Table 4). R4 discriminated between DENV4-infected serums and either DENV2- or DENV3-infected serums with good accuracies, in accordance with the absence of cross-reactions between ED3.DENV4 and the IgMs of such serums. A more detailed analysis of the data in Tables 4 and 5 showed that the discrimination by a given R_i reagent between homotypic DENV_i- and heterotypic DENV_j-infected serums was not correlated with the discriminations by R_i between DENV_j infected serums and DENV-uninfected serums (not shown). Otherwise stated, the R_i reagent may not discriminate between the DENV_i- and DENV_j-infected serums, even if the IgMs of the DENV_j-infected serums do not cross-react with ED3.DENV_i. Conversely, the R_i reagent may discriminate between the DENV_i- and DENV_j serums even though the IgMs of the DENV_j-infected serum cross-react with ED3.DENV_i. Moreover, the discrimination by the R_i reagent between DENV_i- and DENV_j-infected serums was not correlated with the number of residue changes between ED3.DENV_i and ED3.DENV_j (not shown). Thus, sequence differences could not predict the discrimination between two serotypes by a given reagent.

R1 to R4, (H6-ED3-PhoA)₂ dimers of serotypes DENV1 to DENV4, respectively; na, non applicable. The serotype of the infecting DENV for the (+) serums was always identical with the serotype of the reagent. The first column gives the serotype of the infecting DENV for the (−) serums. Other DENV, all DENV serotypes except the serotype of the reagent. All others, all serums (infected and uninfected) except those infected by the same DENV serotype as the reagent. Each entry gives the accuracy and associated standard error of a MAC-ELISA that used the R_i reagent (i = 1, …, 4) as an antigen and was evaluated on the (+) and (−) serums. Note that the expressions (+) serums and (−) serums refer to the discrimination under statistical analysis and not to the infected or uninfected character of the serums.

For each R_i reagent, we calculated its discrimination between serums infected by any of the four DENV serotypes (DENV-infected serums) and DENV-uninfected serums, i.e. the set of positive serums was constituted by all the DENV_j-infected serums with j = 1, …, 4. This discrimination was good for R1, fair for R2 and nil for R3 and R4 (Table 4, row 5). The discrimination between the DENV-infected and DENV-uninfected serums by R_i was strongly negatively correlated with the number of residue changes between the consensus ED3 domain and the ED3.DENV_i domain (R _P = 0.98; compare rows 5 of Tables 2 and 4). Thus, ED3.DENV1 behaved like a consensus ED3 domain.

For each reagent R_i (i = 1, …, 4), we calculated its discrimination between serums infected by the homotypic DENV_i and serums infected by any of the three heterotypic DENVs, i.e. the set of negative serums was constituted by all the DENV_j-infected serums with j ≠ i. This discrimination was good for R1 and fair for R2, R3 and R4 (Table 5, row 5). We calculated its discrimination between serums infected by the homotypic DENV_i virus and serums uninfected by any DENV or infected by any of the three heterotypic DENV_j (j ≠ i). This discrimination was excellent for R1, good for R2, fair for R3 and poor for R4 (Table 5, row 6). Thus, each R_i reagent could recognize the homotypic DENV_i-infected serums with at least some accuracy.

The ability to analyze the specificities and cross-reactivities of human IgMs towards the different serotypes of the ED3 domain depends on the existence and use of a reliable test for the interaction between IgMs and ED3 domains. Here, we used a MAC-ELISA with (H6-ED3-PhoA)₂ dimers as antigens. The use of such antigens raised two questions: was the ED3 domain correctly folded in these dimers and was it accessible to antibodies? The ED3 domain has one disulfide bond and the PhoA monomer has two disulfide bonds. Previously, we have shown that the isolated ED3 domain can be produced in a correctly folded state in the periplasmic space of E. coli, where the formation of the disulfide bonds is efficiently catalyzed 71 . In particular, site-directed mutagenesis experiments and the crystal structures of the complexes between the ED3 domains from the four serotypes of DENV and the scFv fragment of the broadly neutralizing monoclonal antibody mAb4E11 have shown that the epitope of mAb4E11 is discontinuous, conformational, and included within the ED3 domain 29 72 . PhoA is enzymically active only as a dimer. We have shown that the H6-ED3-PhoA hybrids have both their ED3 and PhoA portions correctly folded and active when they are produced in the E. coli periplasm. This result was obtained by measuring the specific activity of the hybrids for the dephosphorylation of pNPP in vitro and assaying their binding to immobilized mAb4E11 in an indirect ELISA, revealed with their intrinsic phosphatase activity 66 .

In the H6-ED3-PhoA hybrids, the C-terminal residue of ED3 (residue 400 of the E protein) is linked to residue Val7 of the mature PhoA through a flexible linker tripeptide Thr-Ser-Gly 66 . The (H6-ED3-PhoA)₂ dimers recognize both mAb4E11, as recalled above, and cell receptors 41 73 . Therefore, the ED3 domain should be at least as accessible in the (H6-ED3-PhoA)₂ dimers as in the full and infectious DENV virus, where it interacts with the other domains of the E protein and its C-terminal residue is linked to the transmembrane region of the E protein and faces the lipid membrane 74 .

Thus, the (H6-ED3-PhoA)₂ dimers can be directly produced in a soluble, correctly folded, multimeric state in the periplasmic space of E. coli. They can be produced and purified in a homogeneous state from periplasmic extracts. They constitute self-sufficient reagents since the ED3 antigen and PhoA reporter enzyme are covalently linked within the same molecule. MAC-ELISA based on such dimers involve a low number of steps or manipulations. The antigen is dimeric and therefore can bind its target through an avidity phenomenon. Such dimers have already been used to detect weak interactions between the ED3 domains of various flaviviruses and either cell receptors or human IgMs 41 71 73 .

The analysis of a test with a ROC plot gives a measure of its capacity to distinguish between two alternatives, the positive and negative cases. A ROC plot gives the accuracy (AUC) value of a test independently of the event frequencies in the test samples and of the decision criterion, i.e. threshold value. To measure the accuracies of the MAC-ELISAs that used the (H6-ED3-PhoA)₂ dimers as antigens, it was necessary to constitute collections of well-characterized human serum samples, i.e. serums whose DENV-infected or uninfected status was reliably established, whose infecting DENV serotype was known, and which contained significant amounts of IgMs against the infecting DENV serotype. These characterizations are described briefly in the Methods section and in more detail elsewhere 65 . They depended themselves on specific tests and thresholds. However, because they were based on sensitive, well-established, standardized, redundant methods that differed from that under analysis, we considered the classifications of the human serums in our collection as absolute.

An AUC value of x means that a randomly selected serum with the properties of the (+) group has a test value larger than that of a randomly chosen serum with the properties of the (−) group 100x% of the time 67 . Therefore, if a MAC-ELISA with reagent (H6-ED3-PhoA)₂ can discriminate between the serums of the (+) and (−) groups, this discrimination implies that the interaction with the ED3 domain is “stronger” for the IgMs of the (+) group than for the IgMs of the (−) group, where the word “stronger” may include both concentration and avidity components.

Our results confirmed that the serums of human patients infected by DENV contain IgMs directed against the ED3 domain 46 47 48 . They showed that the IgMs of human serums infected by a given DENV_i serotype cross-reacted with the ED3 domains from other DENV_j serotypes (i ≠ j). Therefore, they extended similar observations, previously made with whole-virus antigens, to the small ED3 domain 15 16 17 18 19 . They also extended similar observations, previously made for IgGs, to IgMs 75 . The exact patterns of cross-reactivities depended on the infecting DENV serotype.

The ED3.DENV_i domains could discriminate between the IgMs of serums that were infected by the homotypic DENV_i virus and the IgMs of serums that were uninfected by DENV, in our MAC-ELISA assays. The levels of discrimination varied with the serotype, in the decreasing order DENV1 ≥ DENV2 ≥ DENV3 > DENV4 (diagonal of Table 4). This conclusion on IgMs is reminiscent of published data on IgGs: (i) Several studies have shown that many mouse monoclonal antibodies that are specific for the DENV complex of flaviviruses, have affinities for the four serotypes of the ED3 domain and neutralization potencies of the four serotypes of DENV in the same decreasing order as above 25 29 76 . (ii) The content of human serums in IgGs against DENV after a primary infection is higher during an infection by DENV1 and lower during infections with DENV2 and DENV3 when assayed by an indirect ELISA with recombinant ED3 domains as antigens 6 . (iii) In tetravalent strategies of immunization against dengue simultaneously with attenuated strains, or E proteins, or ED3 domains of the four DENV serotypes, lower levels of neutralizing antibodies are induced against DENV3 and especially DENV4 in comparison with DENV1 and DENV2 in mice and humans 59 62 77 78 . These comparisons between our results on IgMs and published data on IgGs suggest that the AUC value might be somehow related to the “level” or “strength” of the interaction between the serum IgMs and ED3 domain under assay, in our specific experimental conditions. If such a relation was valid, we could conclude that the serum IgMs reacted the strongest with ED3.DENV1 and the weakest with ED3.DENV4, whatever the infecting serotype (rows of Table 4).

The R1 reagent gave the highest discrimination between serums infected by any one of the DENV serotypes and uninfected serums. In particular, R1 gave a discrimination equal to or higher than that of the homotypic reagent, between serums infected by a given DENV serotype and uninfected serums (Table 4). These results showed that ED3.DENV1 behaved like a consensus or universal ED3 domain for the detection of IgMs directed against any DENV serotype. They were reminiscent of published data showing that, in human secondary infections, the serum titer in IgGs against ED3.DENV1 is higher than those against the three other ED3 serotypes, regardless of the infecting DENV serotype 6 .

We showed that some R_i reagents could discriminate between human IgMs directed against the homotypic DENV_i and IgMs against heterotypic DENV_j viruses (j ≠ i). The levels of discrimination by R_i between IgMs directed against the two different DENV serotypes were not correlated with the levels of discrimination by R_i between DENV_j infected serums and DENV-uninfected serums, i.e. with the profiles of cross-reactivities. They were not correlated with the numbers of residue changes between ED3 domains. Noteworthily, R1 could discriminate between IgMs against DENV1 and IgMs against DENV2, DENV3 or DENV4, whereas R2, R3 and R4 could not discriminate between IgMs against the homotypic DENV and IgMs against DENV1 (Table 5). Any ED3 domain interacted more strongly with the IgMs against the homotypic DENV than with the IgMs against heterotypic DENVs, with one exception. Each ED3 domain interacted as strongly with the IgMs against DENV1 as with the IgMs against the homotypic DENV (row 1 of Table 5). Fine structural properties might explain these discriminations between serotypes. The R_i reagents had also significant levels of discrimination towards IgMs directed against the heterotypic DENV_j viruses (j ≠ i), taken as a whole (line 5 of Table 5), although not high enough for diagnostic tests. Data on the discrimination between DENV serotypes by ED3 domains in a MAC-ELISA had not been reported previously to our knowledge.