Antibody Responses to Zika Virus Infections in Environments of Flavivirus Endemicity

To examine serological immune responses, we developed a flavivirus-focused microarray comprising mature and immature forms (34) of DENV1 to DENV4 and ZIKV (five Asian isolates and six African lineage isolates), along with recombinant protein antigens from 15 isolates (Table 1 and Fig. 1A). Equivalent densities of recombinant proteins or viruses were deposited by inkjet printing on microarray surfaces coated with a thin layer of nitrocellulose. The printed antigens were evaluated with mouse antisera against 12 flaviviruses, representing antibodies produced by a noninfectious route. The E antigens from all flaviviruses except DENV3 were detected by mouse antibodies (IgG) that were produced in response to the corresponding virus, as shown in Fig. 1B, with evidence of significant E cross-reactivity for JEV antisera. The NS1 proteins from all except DENV1 to DENV4 were detected by the mouse antivirus sera, whereas the M antigen was only weakly recognized at best. We utilized the microarray to examine sera from nonhuman primates (NHPs) (Macaca mulatta) that were challenged subcutaneously with either African or Asian isolates of ZIKV (Table 1; see Fig. S1A in the supplemental material) (33) (Zika Open-Research Portal [https://zika.labkey.com ]). The African and Asian lineages of ZIKV share ∼95% of E-protein amino acid sequences (35) or about the same level of similarity found among E proteins of individual DENV serotypes. Serum antibody binding to virus particles and E antigens of ZIKV, as measured by the microarray, was substantially elevated (Fig. 2A and B) 21 to 28 days postinfection (dpi), while no antibody recognition was observed for NS1 or M proteins (Fig. 2B), suggesting that anti-NS1 and anti-M antibodies represent a small proportion of the humoral immune response to infection during the time points examined. Antibody recognition was more robust for mature virus particles (∼2-fold higher) than for immature virus particles, which may display different conformations of E and M proteins (13). Within the assay, the highest antibody interactions were detected by ZIKV E proteins in comparison to mature ZIKV (Fig. 2A and B). However, a direct quantitative relationship between virus and recombinant protein cannot be determined by these results, because the complex nature of the native virus precludes printing equal molar amounts of available antigen. Only minor differences were observed in antibody responses to individual African and Asian lineage ZIKV antigens for both virus and E protein (Fig. 2), consistent with the conserved amino acid sequences and a single ZIKV serotype, as recently reported by others (36). The E-protein-specific IgM responses were detected by 3 dpi, coinciding with the rise of virus, peaked by 11 dpi, and subsided thereafter (Fig. 2C). The corresponding IgG responses were delayed compared to IgM, consistent with a naive immune response, and displayed increasing levels through 28 dpi (Fig. 2C). Further, there were no apparent differences in the magnitude or kinetics of humoral immune responses to the different amounts of virus used for challenges (Fig. S2), suggesting that levels of IgG and IgM were increasing in tandem with virus replication.

The E proteins of ZIKV and DENV have a high degree of structural similarity that may contribute to shared antibody epitopes. We examined NHPs (M. mulatta) challenged independently with DENV1 to DENV4 (Fig. S1A) (37). For virus, antibodies (30 dpi) from NHPs infected with any DENV serotype were highly cross-reactive to heterologous DENV serotypes and ZIKV (Fig. 3A and Table 1). Further, DENV2 and DENV3 antibodies displayed substantially higher reactivity with the heterologous ZIKV, while IgG from ZIKV-challenged NHPs was more specific for ZIKV at the virus level, with a lower overall level of cross-reactivity toward DENV1 to -4 (Fig. 3A). In contrast with viruses, the E proteins presented antibody recognition profiles that were very specific for the challenge virus (Fig. 3A) and minimal antibody recognition of E proteins from 10 more distantly related flaviviruses (Fig. 4A). DENV-challenged NHPs exhibited the highest antibody binding to the E protein from the DENV challenge serotype, and antibodies from ZIKV-challenged NHPs essentially bound only to ZIKV E antigens. A principal-component analysis (PCA) of antibody bound by DENV and ZIKV E antigens differentiated serotype-specific DENV and ZIKV infection sera due to the higher degree of homotypic E recognition (Fig. 3B). In contrast to the E-antigen results, PCA based on IgG recognition of virus only enabled distinction of ZIKV- from DENV-challenged sera, whereas DENV serotype-specific clusters were not evident (Fig. 3B). Furthermore, antibodies from NHPs challenged with African or Asian lineage ZIKV were not differentiated by E protein or virus (Fig. 2 and 3B). We also considered YFV, both as a nearest neighbor of ZIKV and DENV (Fig. 1A) and because vaccination against yellow fever is common in many countries with a high prevalence of dengue. While serum antibodies from NHPs vaccinated with the 17D YFV strain (Fig. S1A) (38) predominantly recognized the E antigen of YFV (Fig. 4A), a modest level of cross-reactivity was evident with several other E proteins, including those of ZIKV and tick-borne encephalitis virus (TBEV).

The animals in the ZIKV, DENV, and YFV infection studies we examined were domestically bred in isolation from most infectious diseases. Therefore, it was important to compare results from the naive backgrounds of animal disease models with primary infections of humans without documented prior exposures to flaviviruses. Dengue human infection models were recently developed to assess the efficacy of live attenuated DENV vaccines (39). Human challenges with the attenuated DENV2 strain rDEN2Δ30 (40) result in a mild disease, with viremia, rash, and neutropenia. We examined sera collected from flavivirus-naive subjects 28 days after challenge (10³ PFU) with rDEN2Δ30 by subcutaneous injection (Fig. S1B) (40). Among the extended panel of E proteins (Table 1), human antibody responses to rDEN2Δ30 resulted in specific recognition of the E protein from DENV2 and to a lesser degree from DENV4 (Fig. 4A and Fig. S3A), which is most similar to DENV2 among all other flaviviruses (Fig. 1A). Low levels of neutralizing antibodies against other DENV serotypes were previously reported for individuals challenged with rDEN2Δ30 (41). For viruses, extensive human antibody cross-reactivity was again noted for other DENV serotypes and ZIKV strains (Fig. S3A). These results indicated that the NHP DENV challenge model replicated the antigen specificity profile of human antibody responses to primary infection.

Antibody cross-reactivity between flaviviruses could be influenced by homology of sequences and structures, as well as the abundance and degree of cross-reactive antibodies in polyclonal sera. For example, cross-reactivity could be due to a small population of antibodies that exhibit high levels of specificity for heterologous E proteins, or it may be due to a larger population of antibodies that exhibit broad cross-reactivity. Although the highest recognition of the homologous E protein was common for antibodies from primary infections, we observed differences in the amount of total antibody across virus species (Fig. 5). Comparing results obtained with all E proteins, cross-reactive antibodies were not detected for ZIKV, while DENV1 and DENV2 antibodies recognized other DENV serotypes. Antibodies from DENV3-infected and YFV-vaccinated NHPs exhibited the lowest binding to the respective E proteins, while a high level of DENV2 E-specific antibodies interacted with the DENV2 E protein (Fig. 5). The lower levels of DENV3 and YFV antibodies that were specific for the cognate E protein, compared to DENV2 for example, contributed to the appearance of an overall higher level of background cross-reactivity (Fig. 3A and 4A). Vaccination with the live-attenuated 17D strain results in low levels of viremia that mimic a true YFV infection, and titers of specific antibodies are also lower than those in wild-type YFV infections (42). In addition, antibodies from NHPs challenged with ZIKV (33) and DENV (43) exhibited neutralizing antibody titers (33, 43) that directly correlated (R² > 0.99) with the E-antibody recognition pattern we observed (Fig. 5).

The antibodies of flavivirus-naive NHPs and humans to primary flavivirus infection were highly specific to the E protein of the challenge virus. Because increased levels of antibody cross-reactivity would be expected for flavivirus-primed individuals with secondary flavivirus infection, we next examined human sera after one or more flavivirus exposures. In contrast to results obtained with primary infections, IgG from DENV2 or DENV3 infections occurring in Peru prior to the Zika epidemic (here defined as secondary DENV infections) (Fig. S1C) collectively interacted with several E proteins, including those from ZIKV (Fig. 4B), suggesting that antibodies from previous infections, possibly DENV4 (based on the amount of IgG bound), dominated immune responses to other flaviviruses. Despite expectations, sera from secondary DENV2 infections did not correlate with primary DENV2 infections (Fig. S3B), providing additional evidence of previous dengue infection in these samples. Moreover, although principally recognizing the E protein of YFV, antibodies from human 17D vaccinations (Fig. S1D) were less specific than those from primary NHP vaccinations, as E proteins from DENV4 and several other flaviviruses were also targeted (Fig. 4B). It is possible that the less-specific YFV responses were a result of declining antibody titers, as the sera were collected up to 118 days after vaccination. We further noted that serological responses from WNV infections that occurred in North America (Fig. S1D), a region with only a small incidence of dengue, exhibited elevated antibody interactions with E proteins from WNV and a few other flaviviruses, but only a low level of interactions with DENV antigens (Fig. 4B). Finally, we examined primary ZIKV infections from the Dominican Republic (Fig. S1C), a Caribbean country where dengue is endemic. Antibodies from ZIKV infections interacted to a greater extent with E proteins from DENV than from ZIKV (Fig. 4B and Fig. S3A) and also recognized E proteins from several other flaviviruses. It is important to note that levels of total E-specific antibodies from all human flavivirus exposures were significantly reduced compared to levels observed in primary infections (Fig. 5). While maximum E-specific antibody abundance never exceeded the low levels of binding observed for primary YFV and DENV3 exposures, these results suggested that serum levels of anti-E antibodies were predominantly driven by infection histories, and it is conceivable that at least one DENV infection preceded each clinical disease examined with sera from secondary infections.

Given the complexity of the human antibody response from primary ZIKV and secondary DENV infections (Fig. S1C), we attempted both to estimate the probability of previous flavivirus exposures and to identify the likely antecedent virus. We used a supervised machine learning method to classify sera by features of antibody binding to the extended panel of 15 E proteins (Table 1). The support vector machine (SVM) classifier was trained on a positive set of E-specific antibody binding signals from primary flavivirus infections and a negative set of background signals from flavivirus-naive sera. The performance of the SVM was evaluated using a 10-fold cross-validation resampling method, which readily differentiated infected from naive sera and different primary infections, resulting in a total model accuracy of 98.5%. Using a probability cutoff value of ≥0.5, the classifier was used to predict flavivirus exposures that occurred prior to the secondary DENV and ZIKV infections. Four secondary DENV2 sera were predicted to have had a previous DENV4 infection, while high probability for two primary ZIKV sera suggested a previous DENV1 infection (Fig. 4C), which was consistent with clustering based on correlated antibody binding (Fig. S3B). Lower overall probabilities for single virus infections were observed for the remaining secondary DENV and ZIKV samples, and classification to a single group was therefore not possible (Fig. 4C). For example, a secondary DENV3 serum had comparable probability values for DENV2 (0.28) and DENV4 (0.27), suggesting a previous infection with either virus. The inclusion of more-extensive training data sets for primary ZIKV and other viral infections will be important for refining the predictive power of the described SVM method.

The ZIKV and DENV presented in Table 1 were propagated and prepared as previously described (34), with some modifications. Briefly, HEK293T cells were maintained in Dulbecco modified Eagle medium (DMEM), supplemented with 10% fetal bovine serum (FBS) at 37°C and 5% CO₂. The cells were seeded in T125 flasks to 60% confluence and cultured for 12 h. The cells were infected (2 h) with 5 ml of suspended virus stock diluted 1:25 with fresh culture medium. Infectious suspensions were replaced for virus propagation, and culture supernatants were harvested at early (48-h) and late (144-h) time points to obtain immature and mature virions, respectively, while adding fresh HEK293T cells 72 h postinfection to ensure the presence of enough viable cells for sustained proliferation to generate mature virus. Culture supernatants were filtered using prewashed (Super G blocking buffer [Grace Bio-Labs] followed by sterile phosphate-buffered saline [PBS]) 45-μm syringe filters, and precipitated for 12 h (4°C) in PBS containing 8% polyethylene glycol 8000 (PEG8000) (vol/vol). Precipitates of viruses were pelleted by centrifugation (14,000 × g, 1 h, 4°C), resuspended in 300 μl sterile PBS (∼100-fold concentration by volume), snap-frozen in a dry ice-ethanol bath, and stored at −80°C.

Viral RNA for preparation of protein-expressing plasmids was obtained from the following sources: American Type Culture Collection (DENV1 to DENV4), Integrated BioTherapeutics, Inc. (YFV and Japanese encephalitis virus [JEV]); NIAID World Reference Center for Emerging Viruses and Arboviruses (Robert Tesh) (St. Louis encephalitis virus [SLEV], WNV, and Rocio virus [ROCV]). The cDNA templates of E, NS1, and premembrane protein (pM) genes were produced by reverse transcription of full-length viral RNA by using 50 μM oligo(dT)₂₀ primers and the SuperScript III reverse transcriptase first-strand synthesis system (Life Technologies). ZIKV genes (E, NS1, and pM) and E genes from Murray Valley encephalitis virus (MVEV), Powassan virus (POWV), and tick-borne encephalitis virus (TBEV) were synthesized (gBlocks [Integrated DNA Technologies, Inc.] and GeneArt Strings [Life Technologies]), with codon optimization for expression in Escherichia coli. Synthesized genes and cDNA were used as the templates in PCR amplification reactions (50 μl total) with gene-specific primers (final concentration of 2.5 μM) and 2× Phusion high-fidelity PCR master mix with HF buffer (New England BioLabs Inc.). PCR-amplified genes were purified using QIAquick spin column PCR purification kit (Qiagen). NS1 and pM were produced as full-length open reading frames (ORFs), and E genes were truncated to exclude transmembrane domains, as predicted by analysis of amino acid sequences using TMHMM server v.2.0 (Center for Biological Sequence Analysis) (51, 52). Purified DNAs were TOPO cloned into the pENTR/TEV/D-TOPO vector (Gateway Technology, Life Technologies). Sequence-verified entry clones were shuttled into expression vectors by recombination reactions using LR clonase II (Life Technologies). Specifically, the ZIKV-MR766-pM ORF was shuttled into an N-terminal His-labeled maltose-binding protein (HisMBP)-tagged vector (53), while all other flaviviral ORFs were shuttled into the N-terminal 6×His-tagged pDEST17 (Life Technologies). All flavivirus constructs were expressed in E. coli BL21(DE3), propagating bacteria in media containing Luria broth (300 ml) supplemented with 100 μg ml⁻¹ ampicillin and 0.1% glucose. Proteins were induced at mid-log phase with 1 mM isopropyl-β-d-thiogalactopyranoside (IPTG) (EMD Chemicals). Induction conditions were optimized for each protein, and bacteria were grown at either 30°C (2 to 4 h) or 18°C (12 h) prior to harvest by centrifugation. Bacterial pellets were lysed in B-PER reagent (Thermo Scientific) containing EDTA-free 3× Halt protease inhibitor cocktail (Thermo Scientific), 0.2 mg ml⁻¹ lysozyme, 250 U DNase I (Thermo Scientific), and 1 mM IPTG. Protein expression was analyzed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) (Bio-Rad) followed by Coomassie staining, and Western blotting using a mouse anti-His-horseradish peroxidase (HRP)-conjugated polyclonal antibody (Abcam). For purification of insoluble proteins, inclusion body pellets were washed as previously described, with minor modifications (54). Briefly, buffer containing 50 mM Tris-HCl (pH 7.4), 1 M urea, and 1% Triton X-100 was used to wash pellets three times, followed by two washes with Tris-HCl (pH 7.4), with centrifugation at 15,000 × g for 7 min between each wash, and purified pellets were stored at −80°C. Purified inclusion bodies were solubilized in 50 mM HEPES (pH 7.3), 140 mM NaCl, 2 mM dithiothreitol (DTT), and 1% SDS, followed by incubation at 99°C with gentle mixing (5 to 15 min), and centrifugation to remove remaining insoluble protein. Solubilized proteins were analyzed by SDS-PAGE with Coomassie blue staining and by Western blotting using anti-His-HRP-conjugated polyclonal antibody (Abcam). The protein concentration and purity of flavivirus proteins were measured using the Agilent Protein 230 kit and Bioanalyzer 2100 instrument (Agilent Technologies). Purified proteins were stored at −20°C in solubilization buffer, with a final concentration of 25% glycerol.

Recombinant proteins were diluted to 200 ng μl⁻¹ in microarray printing buffer (50 mM HEPES, 140 mM NaCl, 2 mM DTT [pH 7.3]) with glycerol added to a final concentration of 40%. Flavivirus and control proteins were printed onto microporous nitrocellulose-coated slides (Oncyte SuperNOVA; Grace Bio-Labs, Inc.) in replicates (n = 6) using a noncontact inkjet microarray printer (ArrayJet, Glasgow, United Kingdom). The virus preparations were printed with printing buffer containing 50% glycerol, and preliminary experiments were performed with printed virus to optimize antibody binding signals. Frozen virus stocks were gamma irradiated (6 megarads) for inactivation and visualized by electron microscopy to assess quality and quantity. Concentrated virus and a dilution series of bovine serum albumin (BSA) (Sigma-Aldrich) were subjected to SDS-PAGE and stained with Coomassie blue to determine the relative amounts of viral proteins in each preparation. The final printing parameters were established by comparison of virus gradients printed onto nitrocellulose-coated slides, where deposited material was quantified against both an IgG and a BSA standard gradient by SYPRORuby staining (Thermo Scientific), and by comparing signal strength with a pan-flavivirus polyclonal rabbit antiserum specific to an E-domain II peptide that is highly conserved among flaviviruses. The deposited protein antigens were similarly evaluated using SYPRORuby, an anti-N-terminal 6×His monoclonal antibody (Sigma-Aldrich), and the pan-flavivirus rabbit antisera described above. Printed microarrays were desiccated (12 h) and stored frozen (−20°C) until use.

All microarray processing steps were performed at 22°C, protected from light. For IgM detection assays, serum IgG was inactivated using GullSORB (Meridian) prior to performing microarray manipulations. NHP (1:50) and human (1:150) sera, diluted in probe buffer (1× PBS [pH 7.4], 0.1% Tween 20, 1% BSA), were precleared by incubating (1 mg ml⁻¹) with E. coli lysate (Promega) with gentle agitation, followed by centrifugation (17,000 × g, 5 min) to remove the pelleted immunoprecipitates. Microarrays were blocked with Super G blocking buffer (Grace Bio-Labs) for 1.5 h at 22°C and washed three times (for 5 min each time) in wash buffer (1× PBS, 0.2% Tween 20, 1% BSA). The microarrays were incubated for 2 h with E. coli-cleared serum, washed five times for 5 min each time, and incubated for 1 h with either Alexa Fluor 647-conjugated goat anti-human γ-specific IgG (1:1,000) or goat anti-human μ-specific IgM (1:250) secondary antibody (Southern Biotech) diluted in probe buffer. Microarrays were washed three times with wash buffer, rinsed twice with filtered deionized water to remove any residual salts, and dried.

Microarray slides were scanned at 635 nm using a confocal laser scanner (GenePix 4400A scanner; Molecular Devices) using settings below signal saturation. Background-subtracted pixel counts were determined with GenePix Pro 7 software, and outliers among data replicates, identified using a modified Z-score (median absolute deviation of >3.5), were removed. Pixel counts from replicate spots were averaged to obtain mean fluorescence intensity (MFI) and used for subsequent analyses. Relative binding was calculated as RB = (x/x_i)(100) where x is the MFI originating from microarrayed antigens and i is the infecting virus species. Relative binding signals were used in hierarchical clustering analyses (average-linkage Pearson correlation) performed using MeV v4.8.1 within the TM4 software suite (55). Student's t tests, polynomial curve fitting, principal-component analyses, and one-way analysis of variance (ANOVA) with Tukey's posthoc honestly significant difference (HSD) test were performed using OriginPro v9.0 (Origin Lab Corporation).

The support vector machine (SVM) method LIBSVM (https://www.csie.ntu.edu.tw/~cjlin/libsvm/ ), available in the R package e1071 (56), was used for predictions of infection histories with quantile normalized microarray data. An optimal separating hyperplane between data classes was determined with the SVM by maximizing the margin between the closest points and minimizing the classification error. All binary subclassifiers were fitted to the model, and the correct class was identified by a voting mechanism (i.e., the class with the highest probability). We used a radial basis function (RBF) as the kernel function, which is defined by K(u,v) = exp(−γ||u − v||²), where u and v are two data vectors and γ (set at 0.001) is a training parameter that makes the decision boundary smoother as the value becomes smaller. The regularization factor C, set at 100, controls the trade-off between a low training error and a large margin. A grid search was used for selection of C (1 to 1,000) and γ (0.0001 to 1) using 10-fold cross-validation of the training data set and the built-in “tune” function of e1071. The final SVM model was generated using the optimal parameters with complete training data sets. To evaluate the performance of the model, a 10-fold cross-validation was implemented on the training data set, which consisted of a positive set of E-specific antibody binding signals from primary flavivirus infections (n = 32 [human and NHP]) and a negative set of background signals from flavivirus-naive sera (n = 34 [human and NHP]). Based on one-way ANOVA followed by Tukey's range test, the overall antibody binding patterns of DENV2-challenged NHPs and humans were not statistically significantly different (P > 0.05). Therefore, the positive training set consisted of data from both rDEN2Δ30-challenged humans and NHPs. The training data set was randomly divided into 10 equal parts, and each run of cross-validation was comprised of 1/10 as the independent test data set and the remaining 9/10 as the training data set. The performance of the model was calculated as accuracy = (TP + TN)/(TP + FP + TN + FN).

A phylogenetic tree was generated based on E-protein amino acid sequences (Asian-YAP/2007 and African-MR-766/1947 lineage ZIKV selected as representative strains). CLUSTAL W2 (57) was used to generate three multiple-sequence alignments (MSAs), each with a different gap opening penalty (5, 10, 25), Blosum62 as the protein weight matrix, and all other options left as default. T-Coffee Combine (58, 59) was then used to generate a single alignment that had the best agreement of all three MSAs. gBlocks (60, 61) with relaxed settings (small blocks allowed, gap positions allowed within final blocks, and less-strict flanking positions) was used to eliminate poorly aligned positions and divergent regions in the combined alignment, and 202 conserved columns within the alignment were retained. A molecular phylogeny was generated using the maximum likelihood method implemented in the PhyML program (v3.0 a LRT) (62). The Blosum62 substitution model and four gamma-distributed rate categories were selected to account for rate heterogeneity across sites. The gamma shape parameter was estimated directly from the data (γ = 1.564). Tree topology and branch length were optimized for the starting tree and subtree pruning and regrafting selected for tree improvement.

All macaque monkeys used in this study were cared for by the staff at the Wisconsin National Primate Research Center (WNPRC) in accordance with the regulations and guidelines outlined in the Animal Welfare Act, the Guide for the Care and Use of Laboratory Animals (63), and the recommendations of the Weatherall report (64). This study was approved by the University of Wisconsin—Madison Graduate School Institutional Animal Care and Use Committee (Animal Care and Use Protocol G005401). For all procedures (i.e., physical examinations, virus inoculations, ultrasound examinations, blood and swab collection), animals were anesthetized with an intramuscular dose of ketamine (10 ml kg of body weight⁻¹). Blood samples were obtained using a Vacutainer system or needle and syringe from the femoral or saphenous vein.

Research on human subjects was conducted in full compliance with the U.S. Department of Defense (DoD), National Institutes of Health (NIH), federal, and state statutes and regulations relating to the protection of human subjects and adheres to principles identified in the Belmont Report (65). All specimens, data, and human subject research were gathered and conducted for this publication under institutional review board (IRB)-approved protocols.

Three groups of Indian origin Macaca mulatta (three individuals per group) were challenged subcutaneously with a different dose (10⁶, 10⁵, or 10⁴ PFU) of either an Asian (study identification [ID] ZIKV001 and ZIKV004) or African (study ID ZIKV002) lineage ZIKV (Table 1; see Fig. S1A in the supplemental material). Sera were collected prior to ZIKV challenge (day 0) and daily for 10 days (all cohorts), followed by two to three times a week from 11 to 28 days postinfection (dpi) (ZIKV001 and ZIKV002 only) (33) (Zika Open-Research Portal [https://zika.labkey.com ]). Human sera from ZIKV infections were collected from four female patients from the Dominican Republic that developed symptoms of ZIKV (fever, joint pain, headache, conjunctivitis, rash, and muscle pain) in January 2016 (Fig. S1C). Three patients were PCR confirmed for ZIKV infection by the CDC within the first 2 weeks of symptom onset, whereas the remaining patient tested positive for the presence of anti-ZIKV IgG by a microplate enzyme-linked immunosorbent assay (ELISA) (Euroimmun, Inc.), as performed by BocaBiolistics (Pompano Beach, FL), and sera were collected 12 to 31 days after the onset of symptoms.

Sixteen healthy, flavivirus naive rhesus macaques (M. mulatta) were subcutaneously injected with 10⁵ PFU of DENV1 (West Pac 74), DENV2 (S16803), DENV3 (CH53489), or DENV4 (341750) (n = 4 per challenge group; Fig. S1A and 4), derived from low-passage, near-wild-type virus isolates. Sera were collected prior to infection and 30 dpi (37, 43). Sera from human primary DENV2 infections (n = 10; Fig. S1B) were collected as part of a DENV human challenge model originally developed by the Laboratory of Infectious Diseases at the U.S. National Institutes of Health. Sera from participants that had no history or serological evidence of flavivirus infection were collected prior to challenge and 28 days postchallenge with 10³ PFU of rDEN2Δ30. rDEN2Δ30 induced viremia in all participants by 5 dpi (39, 40). Two individuals exhibited elevated binding to flavivirus E prior to challenge with rDEN2Δ30 (Fig. S4) and were further excluded from our analysis. Convalescent DENV sera (n = 7) were collected in Peru by the U.S. Naval Medical Research Unit No. 6 (NAMRU-6) between February 2011 and November 2013. Subjects had febrile illness for 5 days or less and were confirmed to have DENV infections (DENV2, n = 5; DENV3, n = 2; Fig. S1C) by PCR during the acute phase of infection. Sera were collected 14 to 24 days after confirmation of acute infection.

Early immune yellow fever virus antisera from three NHPs (Fig. S1A; NR-29335, NR-29337, and NR-29338; BEI Resources, NIAID, NIH), immunized by subcutaneous injection of 0.5 ml of live, attenuated YFV vaccine (strain 17D [38]), were collected 30 days after vaccination. Human sera from 17D-vaccinated individuals (seven primary and six boosted [Fig. S1D]), collected 14 to 118 days after vaccination, were obtained from the Department of Defense Serum Repository (Silver Spring, MD).

Confirmed WNV-infected human sera (n = 20; Fig. S1D) were collected between 2009 and 2011 at FDA-approved blood donor locations within the United States in accordance with a surveillance protocol performed by the National Heart, Lung, and Blood Institute (NHLBI) Biologic Specimen and Data Repository Information Coordinating Center (BioLINCC) (66). Sera were identified as WNV positive by nucleic acid testing, indicating a current WNV infection at the time of blood donation. WNV-positive donors were then contacted for study enrollment, at which point subjects completed symptom questionnaires and provided subsequent blood samples at several weekly and monthly visits after the initial donation. Each specimen tested positive for the presence of WNV-specific IgM and IgG antibodies (67).

Sera collected by SeraCare Life Sciences, Inc., from healthy U.S. donors (n = 5) were used for negative controls. These sera were selected based on no detected antibodies to human immunodeficiency virus type 1 and type 2, hepatitis A and B viruses, and all flaviviruses used in the microarray.