Integrated evaluation of antibody responses to mosquitoes and mosquito-borne pathogens using highly multiplexed serology Authors: Melodi Anahtar, Diana Striplet, Eben Philbin, Joshua R. Lacsina, Jesus G. Valenzuela, Michael A. Johansson, Fabiano Oliveira, Laura Willen, Shaden Kamhawi, Auden Cote-L’Heureux, Peter D. Crompton, Tuan M. Tran, Sarah M. Fortune, Daniel E. Neafsey Published: 2026 ABSTRACT Vector-borne diseases (VBDs) are a leading cause of morbidity and mortality worldwide. Antigens driving host immune responses to pathogens and vectors can serve as vaccine candidates and biomarkers of previous exposure, but the immunogenicity of relevant proteomes remains undercharacterized. To comprehensively profile antibody responses to VBDs and the vectors themselves, we developed a highly multiplexed phage display library, VectorScan, containing more than 250,000 peptides derived from diverse arthropod vectors and prevalent vector-borne pathogens. We used phage immunoprecipitation sequencing to screen VectorScan against blood samples with experimental and natural exposures to arboviruses, malaria parasites, and mosquitoes. We analyzed quantitative measurements of peptide seroreactivity to identify epitopes driving viral serotype-level exposure signatures and mosquito sialome antigens. Mosquito-directed antibody responses were quantitatively associated with natural viral and parasite exposure. We leverage VectorScan to define the relationship between the intensity of mosquito exposure and the rate of malaria infection in children, demonstrating the public health utility of a tool for paired vector and pathogen serosurveillance. INTRODUCTION Each year, vector-borne diseases (VBDs) cause more than 700,000 deaths and hundreds of millions of infections around the world (1). Arthropod vectors such as mosquitoes and ticks accomplish pathogen transmission through blood feeding behavior. During feeding, an arthropod not only ingests blood but also injects saliva and any pathogens present in the salivary glands. The proteins in saliva, referred to collectively as the sialome, can enhance pathogen transmission and increase mortality in animal models of infection compared to needle-based inoculation (2) through mechanisms such as host immune activation and immune cell recruitment to the bite site (3, 4). Vaccination with immunogenic salivary proteins has been shown to prevent infection in animal models of major VBDs including Lyme, Zika, and malaria (5–8), and the safety and efficacy candidate antigens are being tested in humans (9, 10). The immunogenicity of salivary components has also motivated their use as population-level biomarkers of vector exposure via serosurveillance (11–13). However, immune responses to individual sialome antigens assayed via enzyme-linked immunosorbent assays (ELISAs) have been too variable to reliably infer individual-level exposure (14, 15). Ultimately, the ability to detect serological biomarkers of exposure to vector-borne pathogens and their vectors simultaneously could be used to connect vector exposure with infection risk, inform public health measures, identify shifts in vector ranges and disease incidence with climate change, train ecological models, and determine new vaccine targets for VBDs. Phage display libraries enable broad surveys of immunogenicity by encoding small peptides into bacteriophages that express the protein on their surface, associating each encoded DNA sequence with the properties of its expressed peptide. These libraries are screened using phage immunoprecipitation sequencing (PhIP-seq) to interrogate antibody responses to hundreds of thousands of peptides 25 to 60 amino acids in length (16, 17). These peptides can be designed to tile through the potentially antigenic protein sequences of an organism or infectious agent of interest, creating a library of potential linear epitopes to which host-derived antibodies against these antigens can bind. Several phage display libraries have been previously developed to query antibody responses to vector-borne pathogens. These include “ArboScan” (18), which is focused on arboviruses, the “Falciparome” (19), which is focused on the Plasmodium falciparum malaria parasite, and others (20, 21). Phage and yeast display libraries have also been designed to query the humoral response to tick and mosquito antigens (22, 23). To facilitate the biomarker discovery process and draw direct relationships between exposure and humoral responses to both pathogens and their vector hosts, simultaneous screening of pathogen- and vector-derived antigens needs to be conducted. To this end, we developed a phage display library called VectorScan to broadly profile the immunogenicity of sialomes from 24 vector species and the full proteomes of 15 viral, bacterial, and parasitic VBDs. In this work, we focus on mosquito antigens as well as mosquito-transmitted arboviruses and malaria parasites that were included in the library. We validate the performance of the phage display library, identify immunogenic regions of vector sialomes and pathogen proteomes, and test the hypothesis that variation in malaria infection rate in Malian children is associated with differences in mosquito exposure. RESULTS VectorScan contains peptides from diverse vector-borne pathogens and arthropod vectors We developed a phage library designed to simultaneously query the antibody response to the proteomes of vector-borne viruses, bacteria, and parasites and the sialomes of diverse mosquito and tick vectors. We enabled interrogation of this broad range of targets by minimizing peptide redundancy using a homology-based approach (see Materials and Methods). From these proteins or homolog clusters, we synthesized a DNA oligo library encoding 253,789 peptides (Fig. 1A; library composition details are in data S1, and sequences are in data S2), which was packaged into T7 bacteriophage as previously described (24). Overview of VectorScan and the PhIP-seq pipeline. (A) VectorScan includes peptides from both pathogens and the sialomes of their vector carriers. This pilot library, focused on mosquito- and tick-borne illnesses, incorporates peptides from mosquitoes (Aedes: 42,484 peptides from 2 species; Anopheles: 88,298 from 17 species; Culex: 26,121 from 2 species), ticks (23,348 from 3 Ixodes species), arboviruses (Alphaviruses: 2701 from 3 species; Bunyaviruses: 632 from 2 species; Flaviviruses: 15,205 from 7 species), bacteria (5660 from B. burgdorferi), parasites (28,411 from P. falciparum and 20,417 from P. vivax), and controls (412 from common human viruses and 100 containing random sequences). The control fraction is not labeled because of scale. (B) In PhIP-seq, small peptides derived from specific antigens are encoded into bacteriophage. This resulting phage library is incubated with samples containing the antibodies of interest (e.g., human serum). Peptides on the phage surfaces bind to antibodies targeting the antigen from which the sequence was derived. These antibody-phage complexes are then pulled down, and the peptide-containing regions of the captured phage are amplified using PCR and sequenced to generate seroreactivity profiles. E. coli, Escherichia coli. Panel (A) created in BioRender. M. Anahtar (2026), https://biorender.com/wm589mv. PhIP-seq (Fig. 1B) was performed largely as published (see data S1 and Materials and Methods for details) (19, 24, 25). Briefly, for every serum sample, antibody capture of antigen-expressing phage was performed, and immunoreactive phage were quantitated by whole-genome sequencing across multiple independent technical replicates. We used a mock-immunoprecipitation approach as described in (26) to define the background reactivity of each peptide caused by nonspecific peptide binding. To define peptides with seroreactivity above the background, the distribution of normalized counts from the phosphate-buffered saline (PBS) wells was used to set a background reactivity threshold for each peptide. For each sample, peptides were deemed “seroreactive” if the technical replicates had counts that were five standard deviations above the background for a predicted uncorrected P ∼ 3 × 10−7. Epitope profiles are reflective of species- and serotype-specific antibody responses to DENV exposure PhIP-seq has been extensively deployed to study exposures to all human viral pathogens (16, 27–29), while other studies using the PhIP-seq approach have focused on a narrower set of immune targets including arboviruses (18) and flaviviruses (20). To assess how the diversity of the VectorScan library affects the ability to detect antigen exposure relative to more focal libraries, we evaluated detection of reactivity to dengue virus (DENV). We screened sera from three nonhuman primates (NHPs), each exposed to one of three distinct DENV serotypes (DENV1, DENV2, and DENV4) (Fig. 2A). Three longitudinal samples were screened from each animal, one from before DENV exposure and two from postexposure time points, allowing us to assess individual trajectories and aggregate DENV reactivity. Epitope profiles in individual samples are reflective of virus-specific antibody responses and can inform serotyping. (A) Sera was sourced from three NHPs, each exposed to a different DENV strain of distinct serotypes (NHP1: infected with DENV1; NHP2: infected with “DENV2”; NHP4: infected with “DENV4”). Three samples were available per NHP: “preexposed” samples collected before infection, “early-immune” sera collected 30 days postinfection, and “late-immune” sera collected more than 120 days postinfection. (B) The set of seroreactive peptides was identified within each NHP. (C) Spearman correlation was used to compare the reactomes of each NHP across the three time points. (D) Measurement of breadth from DENV-derived peptides across time points (n = 3) for each NHP (n = 3). (E) Magnitude of signal across time points for each DENV-derived peptide seroreactive at the early-immune time point in each NHP. (F) Mann-Whitney tests were used to evaluate which species in the library had up-regulated signal at the early-immune versus the preimmune time point. The inset displays the distribution of signal among peptides that were seroreactive in NHP1, with peptides from DENV colored in red. (G) For each animal, the peptides that were seropositive at any time point were mapped to the proteome of the DENV strain used to infect the NHP. The relative position within the proteome is shown on the x axis. The y axis reflects the fold change in RPK between each postexposure time point and baseline. (H) Among the seroreactive DENV peptides, FETs were used to evaluate whether serotype-informative peptides from the infecting strain were overrepresented in each NHP. The strength and direction of the association are indicated by the OR; an OR > 1 suggests overrepresentation. Error bars represent the 95% confidence interval. Padj values were calculated using Bonferroni adjustment. Panel (A) created in BioRender. M. Anahtar (2026), https://biorender.com/wm589mv. In each NHP, just more than 1% (NHP1: n = 2835; NHP2: n = 4059; NHP4: n = 2738) of the peptides from the entire library met our strict definition of seroreactivity. The reactome of each animal was highly unique (Fig. 2B and fig. S1) but with only a moderate degree of correlation within a given NHP (Fig. 2C and fig. S1), likely reflecting qualitative and quantitative changes in peptide seroreactivity following DENV infection. Within each NHP and across all three animals as a group, we observed postinfection increases in both the number of unique seroreactive DENV peptides, which we interpret as response breadth (Fig. 2D), and the magnitude of signal to these seroreactive peptides (Fig. 2E). To leverage the combination of breadth and magnitude, we used a gene set enrichment (GSEA)–style analysis, where for each NHP, we rank ordered the difference in signal for every seroreactive peptide between the pre- and postexposure time points. In each NHP, DENV-derived peptides ranked significantly higher compared to the rest of the library [one-sided Mann-Whitney with Benjamini-Hochberg (BH) correction; NHP1: Padj = 4.51 × 10−63; NHP2: Padj = 1.50 × 10−119; NHP4: Padj = 1.15 × 10−35; Fig. 2F), reflecting strong anti-DENV antibody reactivity at 30 days postinfection and significant (fig. S2), although quantitatively lower (Fig. 2, E and G), reactivity at the later time point compared to the baseline. Most of DENV-derived peptides mapped to envelope protein (E), nonstructural protein 1 (NS1), and nonstructural protein 3 (NS3) (Fig. 2G), in agreement with previous studies (18, 20). By mapping every DENV-derived peptide that was seroreactive postinfection to the infecting serotype, we discerned reactivity specific to the infecting serotype within each NHP [Fig. 2H; DENV1-specific peptides in NHP1: odds ratio (OR) = 2.60, Fisher’s exact test (FET) with Bonferroni correction Padj = 7.57 × 10−09; DENV2-specific peptides in NHP2: OR = 2.88, FET Padj = 1.68 × 10−18; DENV4-specific peptides in NHP4: OR = 3.01, FET Padj = 3.57 × 10−07], indicating that a serotype-specific immune response is detectable. There was also signal enrichment from Zika virus, which shares high genetic and structural homology to DENV (30), in NHP2 and NHP4 (Fig. 2F). However, despite known cross-reactivity challenges among flaviviruses (31–33), there was little enrichment among the other flaviviruses, arboviruses, or any other library constituent (Fig. 2F). Concordant phage display and ELISA seropositivity for CHIKV in field samples We then evaluated the ability of VectorScan to detect exposures to both vector-borne pathogens and associated mosquito vectors in naturally exposed human cohorts. We screened serum collected before 2019 by the US Centers for Disease Control and Prevention (CDC) Dengue Branch from 62 individuals living in dengue-endemic US territories (Puerto Rico and American Samoa). A subset of samples was from subjects who were not acutely ill (no recent febrile illness, n = 44) but screened for DENV immunoglobulin G (IgG), chikungunya virus (CHIKV) IgG, CHIKV IgM, and/or DENV IgM via ELISA to assess historical exposures to both viruses. The remaining samples were from subjects with recent febrile illness (n = 18) who had confirmed CHIKV infection via polymerase chain reaction (PCR) and were positive for CHIKV IgM but negative for DENV IgM (IgG ELISAs were not run on these samples). DENV and CHIKV are transmitted by Aedes spp. mosquitoes, and therefore, a history of DENV or CHIKV infection was taken as presumptive evidence of vector exposure. We categorized subjects on the basis of these lab-based diagnoses (summarized in Fig. 3A) and defined associated differences in antibody responses corresponding to this range of acute and historical arbovirus exposures. We identified seroreactive peptides for each individual as described above and assessed CHIKV and DENV responses. We found significantly higher breadth of CHIKV-derived peptides in the CHIKV IgG+ versus IgG− individuals (Fig. 3B, P < 0.0001, Mann-Whitney). We then applied the GSEA-like analysis to determine how CHIKV responses aligned with the ELISA results. Among the peptides that were seroreactive in any individual, we calculated and rank ordered the difference in the mean of the normalized read counts for CHIKV IgG+ (n = 19, samples with profiles 1 to 3; Fig. 3A) and CHIKV IgG− samples (n = 25, samples with profiles 4 and 5). CHIKV-derived peptides had significantly higher differences as compared to the rest of the library (one-sided Mann-Whitney, P = 5.53 × 10−18; Fig. 3C; with BH correction Padj = 2.87 × 10−16; fig. S3), demonstrating strong concordance between VectorScan signal and conventional ELISA. We further characterized the immunodominant regions driving these antibody responses. We used network graphs to group these peptides on the basis of sequence similarity (Fig. 3D) (30–32), yielding 27 clusters of peptides representing immunogenic regions of the CHIKV proteome (Fig. 3E). The degree of immune reactivity to these regions was proxied by summing the reads per kilobase (RPKs) of seroreactive peptides within each cluster, yielding a quantitative “cluster reactivity score” reflective of peptide breadth and signal depth. Some clusters had reactivity that was relatively personal (e.g., clusters 15 and 16), while others were highly reactive (i.e., had high reactivity scores) in multiple individuals. Two clusters exhibited significantly higher scores in CHIKV IgG+ versus IgG− individuals (Wilcoxon test with BH correction, Padj < 0.05; Fig. 3F). Furthermore, cluster 24 has elevated scores among CHIKV IgM+ individuals (Fig. 3G), which is suggestive of class switching in these responses. Peptides in both clusters overlapped with linear epitopes as predicted by ElliPro [CHIKV PDB (Protein Data Bank): 2XFB; E2: chain B; Fig. 3H] (33). Those with the strongest signals from cluster 24 (fig. S3) mapped specifically to the E2-B domain, which has been shown to elicit neutralizing antibodies (34, 35). Field samples with known CHIKV exposures exhibit antibody responses to CHIKV glycoprotein. (A) Serum samples (n = 62) from arbovirus-endemic areas had preexisting metadata of IgG and/or IgM ELISA seroreactivity to DENV and CHIKV. (B) The breadth of seroreactive peptides was compared between those who were CHIKV IgG+ versus IgG−. P value calculated with a Mann-Whitney test. (C) Peptide-level differences in signal among those who were CHIKV IgG+ versus IgG− were ranked and plotted. CHIKV-derived peptides (purple) are colored, while all other relevant seroreactive peptides are gray. P value calculated from one-sided Mann-Whitney tests. (D) Seroreactive peptides were organized into clusters using network graphing based on sequence similarity calculated with blastp. The normalized counts (RPKs) of each seroreactive peptide in the cluster were summed to yield a “cluster reactivity score” for each sample. (E) Heatmap of scores from each resulting cluster for every sample in the cohort, regardless of ELISA seroreactivity status. (F) Two of 27 clusters (clusters 24 and 25, labeled) had significantly higher cluster scores among those who were CHIKV IgG+ versus IgG−. Horizontal line at Padj = 0.05. FC, fold change. (G) Individual-level reactivity scores from cluster 24. P values reflect a one-way analysis of variance (ANOVA) with Tukey’s multiple correction. (H) The peptides in these two clusters map back to the CHIKV structural polyprotein, specifically E2. AA, amino acid. (I and J) Peptide-level differences in signal among those who were DENV IgG+ versus IgG− using the same analyses as (B) and (C). We performed the equivalent analysis for DENV IgG+ (n = 28, profiles 1 and 4) and IgG− (n = 16, profiles 2, 3, and 5) individuals. The breadth of DENV-derived peptides was not significantly different on the basis of DENV IgG seropositivity (P = 0.53, Mann-Whitney; Fig. 3I), and incorporating measures of magnitude via GSEA-style analysis showed no significant enrichment in signal from DENV-derived peptides (P = 0.10; Fig. 3J). The lack of significance could reflect differences in quantitative sensitivity between the two methods or signal dilution across the serotypes, which would weaken the statistical power of observing signal changes resulting from infection with a single serotype. Antibodies to mosquito-derived antigens are associated with exposure to infected mosquitoes We next evaluated whether measurable differences in antibody responses to mosquito-derived epitopes were associated with a presumptive mosquito exposure. We focused on individuals who were IgM+ versus IgM− for CHIKV. Because CHIKV is transmitted by Aedes spp. mosquitoes, recent CHIKV infection provides presumptive evidence of recent Aedes exposure in IgM+ individuals. The level of exposure to Aedes in IgM− individuals was a priori unknown, as were expected exposures to Anopheles or Culex, both of which are also present in the cohort geographies. Among the 62 subjects, we identified 37,518 mosquito-derived peptides that were seroreactive in at least one subject. All subjects exhibited mosquito reactivity, and there was a significant difference in the breadth of these responses between those who were CHIKV IgM+ versus IgM− (P = 0.0436, Mann-Whitney test; Fig. 4A). To assess the magnitude of the responses to mosquito-derived peptides between groups, we took the sum of RPKs across all peptides that were seroreactive in each sample and observed a trend toward greater responses in the IgM+ individuals (P = 0.11, Mann-Whitney test; Fig. 4B). To identify common targets of mosquito exposure–associated responses despite the heterogeneity of individual reactomes (fig. S1), we again used homology-based clustering of immunogenic peptides, which may map back to a single protein or be shared by several protein orthologs in the library. Mapping seroreactivity to these clusters (via quantification of cluster reactivity scores) again highlighted the extensive interindividual heterogeneity in reactivity (Fig. 4C). Five clusters had significantly higher reactivity in the IgM+ group (Wilcoxon test with BH correction, Padj < 0.05; Fig. 4D), and two had particularly large differences (clusters 1506 and 3428; Fig. 4E). The peptides in these five clusters were most likely to be derived from Aedes spp. (two-sided FET with Bonferroni correction, Aedes: OR = 3.78, Padj = 5.78 × 10−4; Anopheles: OR = 0.28, Padj = 0.0021; Culex: OR = 0.89, Padj = 1; Fig. 4F). Cluster 3428 was driven by Aedes peptides, which have high homology to AAEL013532 (27277: 94.64% identity, E = 7 × 10−32; 34408: 66.67% identity, E = 2 × 10−21), a vector protein described as induced after infection of Aedes aegypti with a range of arboviruses (36). For cluster 1506, peptides 94909 and 147087 have the strongest signal, and both have similarity to Aedes protein AAEL025037 (94909: 59.26% identity, E = 1 × 10−06; 147087: 59.62% identity, E = 3 × 10−06), which is uncharacterized. Mosquito-derived antibody repertoires are highly heterogeneous and are associated with exposure. (A) The breadth of mosquito-derived peptides was assessed on the basis of the number of seroreactive peptides for each CHIKV IgG+ and IgG− sample. (B) The magnitude of the responses from these peptides was evaluated by summing the RPK values for each seroreactive peptide for every sample in these two groups. (C) Mosquito-derived peptides that were seroreactive in any of the CDC samples were clustered on the basis of sequence similarity. Cluster reactivity scores from the resulting 5091 clusters were plotted for each sample, and hierarchical clustering was performed using the pheatmap R package. (D) Volcano plot showing the statistical differences (Wilcoxon tests with BH correction) in the fold change between cluster reactivity scores among CHIKV IgM+ and IgM− individuals. There were two clusters (in red) with Padj < 0.05 (threshold cutoff signified by the dashed line). (E) Log-transformed reactivity scores for these clusters, stratified by ELISA seropositivity. (F) FETs were performed to determine whether any mosquito genera represented in VectorScan were overrepresented among the peptides making up these two clusters. Horizontal dashed line at OR = 1. Error bars represent the 95% confidence interval. (G) Healthy adult volunteers underwent four feedings by uninfected A. aegypti mosquitoes. Plasma samples from day 0, before these successive exposures, and day 44, after all four exposures, were screened with VectorScan. (H) Number of seroreactive peptides derived from mosquitoes in samples from individuals living in Puerto Rico/American Samoa compared to both the day 0 and day 44 time points. P values were calculated using the Kruskal-Wallis test with Dunn’s multiple comparisons. Panel (G) created in BioRender. M. Anahtar (2026), https://biorender.com/wm589mv. The individuals in the CDC cohort were likely to have had recurrent mosquito bites, and their seroreactivity could reflect responses to chronic or acute exposures. To evaluate whether detectable antibody responses result from controlled acute mosquito exposure, we obtained samples from a National Institutes of Health (NIH)–conducted clinical trial in which healthy volunteer adults living in the Washington, DC metropolitan area experienced four 10-min feedings by up to 10 uninfected A. aegypti mosquitoes, with each feeding ~2 weeks apart (Fig. 4G) (37). These samples were tested by ELISA using salivary gland homogenate (SGH) from A. aegypti and Anopheles gambiae. While there was a modest increase in reactivity between day 0 (before exposure) and day 44 (after four exposures) to A. aegypti SGH (P = 0.049, Wilcoxon matched-pairs rank test), the ELISA optical density (OD) from all samples was low relative to the positive control (all had values <50% of the positive control), suggesting that the overall strength of this antibody reactivity was relatively weak (fig. S4). There was no difference in signal against A. gambiae SGH (P = 0.76, Wilcoxon matched-pairs rank test; fig. S4). We evaluated whether the broader range of antigens in VectorScan would demonstrate evidence of reactivity in these day 0 and day 44 samples. The VectorScan profiles of each subject were highly individualized and stable between both time points (median Spearman R of paired samples: 0.75; range: 0.58 to 0.83; fig. S1). All individuals exhibited seroreactivity to mosquito-derived peptides at both time points, but only a single individual showed significant enrichment of mosquito signal via our GSEA-style analysis at day 44 versus day 0 (fig. S4). Furthermore, there was no significant difference in the number of seroreactive mosquito peptides between day 0 and day 44 (P = 0.654, paired t test; Fig. 4H). In combination with the ELISA analysis, these data suggest that the controlled exposures did not elicit a strong antibody response to vector antigens. We next analyzed differences between the vector-targeted responses in the NIH study volunteers who live in the DC area where mosquito activity is seasonal and the CDC subjects living in Puerto Rico and American Samoa, where there is year-round mosquito exposure. We compared the number of seroreactive mosquito peptides recognized in sera from the NIH study volunteers to those from the CDC subjects. The number of seroreactive peptides recognized by each individual spanned a broad range in the CDC cohort with some individuals much higher than the NIH volunteers, but group-level differences were not significant (Padj > 0.05 with Kruskal-Wallis test with Dunn’s multiple comparisons; Fig. 4H). Seroreactivity to Plasmodium- and mosquito-derived peptides is associated with malaria infection rate We next investigated the quantitative relationship between recurrent malaria infections and mosquito exposure in children. We used samples from a longitudinal cohort study conducted in Kalifabougou, Mali, where dried blood spot samples were collected biweekly from children during the 6-month malaria transmission season in 2011 and monthly during the transmission seasons in 2012 to 2016 (38). We previously quantified the infection rate of P. falciparum [i.e., molecularly detected force of infection (molFOI): the number of genetically distinct parasite strains detected in a person over time; Fig. 5A] using a custom multiplex amplicon sequencing assay that yields deep Illumina read coverage of four highly polymorphic P. falciparum parasite antigens (39). In that study, we observed a large range in molFOI among participants that was stable across seasons and not associated with age, sex, or proximity to mosquito breeding sites (40). While a portion of the molFOI variation was attributable to the malaria-protective HBS sickle mutation, we hypothesized that differences in mosquito exposure may be responsible for variation in Plasmodium infection rates. While virtually all participant households reported use of insecticide-treated bednets (40), other behavioral differences and individual factors that modulate attractiveness or exposure to mosquitoes may influence mosquito biting rate (41–43). Characterization of signal P. falciparum– and mosquito-derived peptides shows relationships between seropositivity and malaria infection risk. (A) The molFOI throughout the 2011 malaria season had been previously characterized for 19 Malian children. (B) Number of P. falciparum peptides that were seroreactive in each human sample. *Padj < 0.05 using Brown-Forsythe and Welch ANOVA with Dunnett’s T3 multiple correction. (C) Quasi-Poisson generalized linear models were fit (red lines) to quantify the association between molFOI and the number of seroreactive peptides in January 2012 from P. falciparum. (D) Heatmap showing the cluster reactivity scores of clusters composed of mosquito-derived peptides that were significantly higher in Mali versus the US (two-sided Wilcoxon with BH correction, Padj < 0.05). (E) FETs were performed to determine whether any mosquito genera represented in VectorScan were overrepresented among the peptides making up these significant clusters. The horizontal dashed line represents OR = 1; error bars represent the 95% confidence interval. (F) Quasi-Poisson generalized linear models were fit (red lines) to quantify the association between molFOI and the number of seroreactive Anopheles peptides in January 2012 in each individual. To test this hypothesis, we screened plasma samples collected before and after the 2011 malaria season (May 2011 and January 2012) from a subset of 19 participants representing the range of molFOI that was observed in the previous study (0 to 46; Fig. 5A) (40). We first compared the Plasmodium reactivity of Malian samples to that of CDC and NIH samples where there is no expectation of malaria exposure. A prior serological study of samples from this Mali cohort demonstrated robust responses to a large panel of P. falciparum antigens (44). Accordingly, the number of P. falciparum–derived seroreactive peptides was significantly higher in Malian samples compared to those from the combined CDC/NIH samples (Kruskal-Wallis test with Dunn’s multiple corrections, Padj < 0.05; Fig. 5B). We performed homology-based clustering of the P. falciparum peptides recognized among all human samples and identified clusters with significantly higher scores among the Malian samples (fig. S5). The clusters containing immunogenic regions from FIRA (cluster 204), PHISTc (cluster 330), EMP3 (cluster 102), STEVOR (cluster 750), and EXP-1 (cluster 1961) exhibited the largest difference reactivity between Mali and the CDC/NIH samples. Immunogenic regions within these and other key P. falciparum antigens, and the rate of seroreactivity to these regions (fig. S5), are consistent with those identified through a previous PhIP-seq screen of the P. falciparum proteome (19). We explored whether molFOI was associated with Plasmodium and mosquito peptide seroreactivity in the 19 Malian children. Using the number of unique seroreactive peptides as a measure of the breadth of the immune response, we identified a significant positive association between molFOI and the postseason breadth of P. falciparum seroreactivity (quasi-Poisson, P = 0.001; Fig. 5C), indicating that increased infection incidence was associated with a broadening of the antimalaria immune response and suggesting that a broad antibody repertoire is a correlate of high infection incidence and not associated with protective immunity to infection. We then investigated antibody responses to mosquito-derived peptides, presuming that individuals in the Mali cohort who experienced malaria infection were all exposed to Anopheles spp., the vector of malaria. We clustered the 53,017 mosquito-derived peptides that were seroreactive in any of the samples and compared the magnitudes of the resulting cluster reactivity scores between the Mali (n = 19, from January 2012) and CDC/NIH (n = 76, 62 CDC samples and 14 NIH samples from day 0) samples. We identified seven clusters with significantly higher scores in the Malian samples (Wilcoxon test with BH correction, Padj < 0.05; Fig. 5D). Among these clusters, Anopheles peptides trended toward overrepresentation (two-sided FET with Bonferroni correction, Anopheles: OR = 1.79, Padj = 0.075; Aedes: OR = 0.83, Padj = 1.0; Culex: OR = 0.36, Padj = 0.056; Fig. 5E). Last, we assessed the relationship between molFOI and vector exposure, as measured by the number of Anopheles-derived seroreactive peptides. We observed a significant positive association between molFOI and the number of Anopheles-derived seroreactive peptides in January 2012 (quasi-Poisson, P = 0.027; Fig. 5F). These data are concordant with the hypothesis that differences in mosquito exposure are associated with malaria infection rate. The humoral response to mosquito exposure is broad We observed the high interindividual heterogeneity of mosquito peptide seroreactivity across all human cohorts. To better understand the targets of the antibody response to mosquito exposure, we explored the functional profile of the proteins from which these peptides were derived. Of the 156,878 mosquito-derived peptides in the library, 33.8% of them (n = 53,017) were seroreactive in at least one human sample, and 82.7% (n = 43,869) of those had high enough similarity with another peptide to cluster (Fig. 6A). Among the peptides in those clusters, three of the most highly represented orthogroups are composed of trypsins (OG7_0000395, n = 1175 peptides), proteins with fibrinogen C domains (OG7_0000972, n = 695), and chitinases (OG7_0001525, n = 632; Fig. 6B). Although peptides from proteins in OG7_0000395 were a large fraction of immunogenic peptides, peptides from this orthogroup were not overrepresented compared to the overall library (two-sided FET with BH correction, Padj = 0.24). Rather, peptides from mucin-like proteins (OG7_0073655, n = 170 peptides, Padj = 1.59 × 10−24), insect cuticle proteins (OG7_0002530, n = 245, Padj = 1.13 × 10−16), and ECM-like proteins (OG7_0020517, n = 218, Padj = 2.35 × 10−15) were significantly overrepresented. Recent studies have identified cuticle proteins as insect- and arthropod-derived allergens, but their ability to elicit IgG responses is not well characterized (45–47). Many of the other orthogroups with significant overrepresentation are composed of uncharacterized proteins, suggesting that elements of the humoral host immune response to mosquitoes have been overlooked and deserve further study. Immunogenic mosquito-derived peptides come from a wide range of orthogroups with distinct functional annotations. Each peptide in the library was annotated on the basis of the ortholog group of its protein origin. The (A) frequency of these OrthoMCL annotations and (B) enrichment of these annotations relative to the entire mosquito-derived portion of VectorScan were analyzed for 43,869 peptides. In (A), the bars are ranked in order of size, while (B) is ranked by descending −log10(Padj). The top 10 groups are shown in each panel. DISCUSSION It is estimated that the human antibody repertoire is composed of up to 1018 unique sequences (48). The magnitude, focality, and duration of any given humoral response that gives rise to these sequences may differ markedly on the basis of the type of exposure, ranging from highly targeted vaccine responses to acute pathogen-associated responses to responses to chronic exposures or self-antigens. Moreover, every individual has a unique exposure history on top of their immunogenetic diversity. As a result, the challenge of PhIP-seq is to identify informative patterns of seroreactivity to antigens of interest over a background of substantial interindividual heterogeneity, which has been described in many studies (17, 49–51). To specifically query antibody repertoires against a wide range of organisms responsible for both the pathogenesis and transmission of VBDs, we developed VectorScan, a custom phage display library. Through our validation of this tool, we have added to a growing body of work that demonstrates how interrogating arboviral proteomes with highly multiplexed assays can be used to identify seroreactive peptides that reflect species- and serotype-level arboviral-exposure signatures (18, 20, 52). Furthermore, by simultaneously and broadly profiling antibody responses to pathogens and their relevant vectors, we also observed that field samples from individuals exposed to CHIKV or P. falciparum exhibited significantly different mosquito-derived peptide seroreactivity from those who were pathogen-naïve. Notably, there were Aedes-derived peptide responses associated with CHIKV infection and Anopheles-derived peptide responses associated with P. falciparum infection, pointing to the generation and detectability of antibody responses specific to relevant vector genera. In seeking to profile both pathogen and vector responses simultaneously, we faced several technical challenges. First, because eukaryotic proteomes (i.e., vector and parasite) are much larger than the viromes typically queried with PhIP-seq, the conventional approach of tiling peptides across the proteins of interest yielded library sizes that were not feasible for us to deploy across large numbers of samples. We addressed this by implementing a pipeline for homology-based library design and leveraging the sequences of the resulting peptides to perform cluster-based analyses of seroreactivity. Second, the magnitude and dynamics of the human response to acute pathogen exposure (e.g., DENV), chronic pathogen exposure (e.g., Plasmodium), and recurrent mosquito bites are very different, and there is a much better understanding of ground truth for the former versus the latter. To address this, we explicitly assessed measures of the magnitude and breadth of the humoral response in our analytic pipeline. The design and analytic features of VectorScan performed largely equivalently to other PhIP-seq approaches in the setting of viral infection; measurements of magnitude and breadth, and the GSEA-based integration of the two, were able to detect infection and discriminate serotype information. The patterns of antibody reactivity to DENV-derived peptides across the proteome that we observed in the infected NHPs were consistent with those generated by an earlier study that used a flavivirus-specific phage library (20). These patterns were also reproducible at the epitope level. For example, the arbovirus-focused ArboScan library was used to identify epitopes that were associated with DENV neutralization, such as an epitope within EDIII (envelope domain III protein) that was specifically correlated with neutralization of the DENV1 serotype (18). We also observed that DENV1 serotype–informative peptides containing this epitope had high signal in the DENV1-infected NHP. Furthermore, we detected CHIKV-specific responses in individuals with known CHIKV exposure. The strongest CHIKV-derived signals we observed were from regions with known immunogenicity and from peptides that were predicted by in silico methods to be linear epitopes, further supporting the capacity to reliably query antiarbovirus responses despite the relatively small representation of these organisms in the overall VectorScan library. VectorScan was also able to discern exposure state in the more challenging cases of Plasmodium and vector exposure. In both cases, the breadth, rather than the magnitude, of seroreactivity appeared to be a more quantitative measure of chronic exposure. While it is obvious that somebody who has contracted a VBD has had a mosquito bite, it was not clear that vector exposure or the resulting serologic response would differ between recently infected and uninfected individuals. However, our data suggest that for both CHIKV and Plasmodium, infection is associated with a quantitatively greater response to the cognate vector but not other vector species. We interpret this as supporting the hypothesis that greater vector exposure is a driver of greater infection risk, but it is possible that the humoral response to vector proteins that occurs in the setting of an antipathogen response is boosted by noncognate immune stimulation. Future studies will be required to delineate these two possibilities. This work also demonstrates that previously underappreciated components of the mosquito proteome are immunogenic in humans exposed to mosquito bites, concordant with findings from a recent yeast display study on ticks (8). We found that cuticular proteins, structural proteins, mucins, and multiple other classes of noncanonical salivary proteins were immunogenic. It is possible that cuticular and other structural proteins are introduced into the human host via shedding from the proboscis during probing or that mucins are being co-delivered through saliva after being transported from the midgut to the salivary glands. Regardless of the mechanism, such proteins may be valuable targets for further study in pursuit of vaccine targets that could disrupt transmission of VBDs (e.g., the mosquito-derived epitopes that were found to be associated with recent arbovirus infection). While we took care to include peptides from proteins traditionally used as serosurveillance biomarkers of mosquito exposure, such as AeD7L1 (11) and salivary gland 6 (SG6) (12), we did not observe significant seroreactivity from these proteins. We suspect that this is due to multiple factors. First, some individuals may not have experienced mosquito exposure levels sufficient to generate a robust IgG response. In the samples from the NIH study, where individuals underwent four exposure sessions to A. aegypti, ELISAs against A. aegypti SGH showed a modest increase in antibody reactivity after exposure but weak overall signal. A recent preprint reported that two controlled exposures to A. aegypti were unable to elicit IgG responses against AeD7L1 and AeD7L2 in individuals from the New York state (53). Another preprint demonstrated that hundreds of Anopheles bites are required to elicit minimal levels of antibodies against gSG6-P1 (the most studied diagnostic peptide derived from SG6) (54). The need for high levels of exposure to mosquitoes or pathogen-associated exposures to elicit meaningful immune responses to these proteins may explain their lack of prominence among these NIH samples. However, seroreactivity to such biomarkers was also low among the CDC samples, which were collected in areas with endemic arbovirus transmission, and Malian samples, which were collected in a village with endemic malaria. Therefore, it is also possible that this lack of seroreactivity was due to the inability of PhIP-seq to effectively detect conformational epitopes or due to a lack of posttranslational modifications, such as glycosylation, both of which are noted limitations of this technology. It is also possible that some epitopes were disrupted by the peptide tiling layout. Immunogenicity to commonly used markers may vary by age, ethnic background, geography, and other demographic factors, which are often controlled for on a study-by-study basis, making them more suitable for serosurveillance in specific populations that were not represented in this study. For example, antibodies to gSG6-P1 are most effective as a marker of mosquito exposure in Africa (12). By broadening the set of proteins used to quantify exposure histories via techniques such as PhIP-seq or other highly multiplexed assays, a wider range of biomarkers may be identified and used to quantitatively evaluate individual-level vector exposure. In this study, we had relatively small sample sizes for studies with controlled or inferred natural exposure to pathogens and vectors, particularly for the DENV-exposed NHPs, for which we had samples from only three animals. PhIP-seq studies benefit from large sample cohorts with known exposure histories to strengthen confidence in disease-associated hits, and the importance of increasing the size of healthy control cohorts has been evaluated (51). However, there are a limited number of cohorts with known exposure histories or sophisticated measures of infection and disease as in the Mali cohort. Additional work will be needed to further explore the utility of VectorScan as a diagnostic or serosurveillance tool for other pathogens and vectors represented in the library, including screening samples from larger patient cohorts and including samples from people or animals exposed to Borrelia, tick bites, and other taxa represented in the library. The deeply multiplexed serological screening capacity afforded by technologies such as phage display will accelerate vaccine candidate and biomarker discovery. We anticipate that broad serological screening tools such as VectorScan that link vector exposure to VBD prevalence will be used to understand how shifts in vector population density and range affect public health through epidemiological models. In addition, broad serosurveillance to understand vector exposure will inform ecological models that explore how climate change, natural disasters, and vector interventions influence the geographic range and population density of disease vectors. MATERIALS AND METHODS Sample sourcing We sourced sera from BEI Resources deposited by National Institute of Allergy and Infectious Diseases (NIH) from three NHPs that were each exposed to a different DENV strain from distinct serotypes [NHP1: infected with dengue virus 1 Nauru/West Pac/1974 (“DENV1”); samples: Pre-Immune Dengue Virus Type 1 Sera, NR-41782, Early-Immune Dengue Virus Type 1 Antiserum, NR-30247, Late-Immune Dengue Virus Type 1 Antisera, NR-41786; NHP2: infected with dengue virus 2 Thailand/NGS-C/1944 (“DENV2”); samples: Pre-Immune Dengue Virus Type 2 Sera, NR-41783, Early-Immune Dengue Virus Type 2 Antiserum, NR-29321, Late-Immune Dengue Virus Type 2 Antisera, NR-41787; NHP4: infected with dengue virus 4 Dominica/814669/1981 (“DENV4”); samples: Pre-Immune Dengue Virus Type 4 Serum, NR-41785, Early-Immune Dengue Virus Type 4 Antiserum, NR-29327, Late-Immune Dengue Virus Type 4 Antisera, NR-41789]. “Preexposed” samples were collected before infection, “early-immune” sera were collected 30 days postexposure, and “late-immune” sera were collected monthly from 120 to 390 days postexposure. A preexposed sample was unavailable for the NHP exposed to DENV3, so it was not included in this study. All human samples were approved for use under Harvard University human subject protocols IRB19-1317 and IRB22-1649 and transferred to Harvard University under material transfer agreements as needed. One set of deidentified human plasma samples collected as part of NCT03641339 (37) was from healthy individuals exposed to A. aegypti mosquitoes four times over the course of 6 weeks (the cohort B study arm). Samples were collected before feedings (day 0) and again after all four exposures (day 44). The deidentified Malian samples were collected in May 2011 (“month 0”), before the start of the 2011 malaria season, and January 2012 (“month 7”), after its conclusion. These samples were heat inactivated at 56°C for 30 min in a thermocycler before use. The CDC Dengue Branch in Puerto Rico provided a set of deidentified human plasma samples for VectorScan evaluation. These samples were collected before 2019 from individuals living in Puerto Rico or American Samoa who either (i) were not ill at the time of sample collection or (ii) had recent febrile illness with an acute specimen that was CHIKV+ via PCR. Samples from the healthy individuals were tested by ELISA for anti-DENV IgG, anti-CHIKV IgG, and/or anti-CHIKV IgM by the CDC and subsequently used to perform PhIP-seq. For the febrile group, samples were collected at a clinical follow-up visit ~1 week after the initial CHIKV diagnosis, heat inactivated at 56°C for 2 hours in a water bath, and tested via ELISA for anti-CHIKV IgM before being sent to Harvard University. One sample (TUNE-1, from a nonfebrile individual) had an “equivalent” CHIKV-IgM result, which was treated as negative and grouped within profile 4. Design of the oligo library Vectors Because many vector species are nonmodel organisms, we selected those with well-defined sialomes to identify likely sialome proteins in other relevant but less studied species. First, we curated a set of “seed” sialomes from previous publications representing different genera of vectors that are relevant to human disease consisting of A. aegypti (55), A. gambiae (56), Culex quinquefasciatus (57), Ixodes scapularis (58), and Xenopsylla cheopis (59). Next, we curated a set of “test” datasets by downloading available genome, transcriptome, or UniProt sequences (based on sequences available in October 2022) for various Anopheles, Aedes, Culex, and Ixodes species (see data S1 for details). We used UBLAST to find orthologs of all seed sequences in the test datasets with an E value threshold of 1 × 10−10 and removed duplicate sequences. We also filtered out sequences from X. cheopis from the list of candidates, as we used it to find orthologs of salivary proteins in the test species but focused this pilot library on ticks and mosquitoes only. We removed possible paralogs or redundant sequences by using cd-hit to cluster all sequences at 98% identity, resulting in 209,862 proteins. Because of the size of this initial list, we clustered progressively at 75 and 70% identity to reduce the candidate pool to 65,921 proteins. We then ran each protein through SignalP and filtered for those with a signal peptide, further reducing the pool to 11,839 protein sequences (60). Sequences were listed alphabetically within each FASTA, resulting in increased representation postclustering and posttiling for proteins from A. aegypti, Anopheles albimanus, and C. quinquefasciatus, which were the first species represented within each genus (data S1). Bacteria Full protein sequences were obtained from UniProt for Borrelia burgdorferi (taxonomy ID: 139), resulting in a pool of 9251 proteins. Two strategies were then used to narrow down the pool of candidate proteins. In the first strategy, cd-hit was used to cluster proteins with greater than 98% identity. The resulting 3350 proteins were input into SignalP, and the 544 proteins predicted to be secreted were selected for downstream analysis. In parallel, the genes in the UniProt list were filtered on the basis of gene ontology for various annotations (outer membrane proteins: GO:0019867 and GO:0009279; extracellular region: GO:0005576; flagella: GO:0009288). The resulting 4143 proteins were then clustered with cd-hit on the basis of 98% identity, resulting in 1097 proteins. The protein lists from these two strategies were combined, and identical sequences were clustered using cd-hit to create the final set of Borrelia proteins. The pepsyn package was then used to create tiles using the same methods as described for the viral sequences. Plasmodium Because Plasmodium species have large proteomes that would necessitate the creation of species-specific phage display libraries to represent the full protein space, more stringent protein selection methods were applied to target predicted immunogenic proteins. Previously, Raghavan et al. (19) created a PhIP-seq library (the “Falciparome” library) specifically for P. falciparum, which they screened against samples collected from individuals in the United States and Uganda. Trusting the rigor of their approach, we used PlasmoDB to retrieve the sequences of all genes that yielded seroreactive peptides in their study and mapped them back to UniProt, resulting in 2939 proteins. The gene name for each entry was extracted and used to query all UniProt entries from P. falciparum (taxonomy ID: 5833) and Plasmodium vivax (taxonomy ID: 5855). The resulting proteins from both species were combined, yielding 48,499 sequences, which were then clustered on the basis of 95% identity using cd-hit. The 48,390 proteins representing each cluster were input into SignalP, enabling extreme downselection to 2944 proteins, which were then tiled through with pepsyn as described above, yielding 40,339 peptides. To ensure that all seroreactive peptides from the Falciparome screen were included in the library, we clustered these 40,339 peptides with all seroreactive peptides using a threshold of 75%. All Falciparome peptides that did not cluster were manually included, resulting in a final list of 48,855 peptides. Notably, the sequences used directly from the Falciparome were each 62 amino acids long. Peptides from the Falciparome are included in the counts for the Species: Falciparome/Family: Falciparum/Group: Plasmodium/Pathogen sections of the library. Furthermore, 27 surviving peptides had sequences derived from Anopheles, which were included in the count for the Species: Falciparome/Family: Anopheles/Group: Mosquito/Vector sections of the library. Arboviruses Protein sequences from all 51,136 UniProt entries for the viruses of interest (see data S1) were compiled into a single FASTA file, and cd-hit was used to cluster sequences with 98% identity, distilling this list to 2106 proteins. We tiled through these sequences in their entirety, ultimately resulting in 18,538 arbovirus-derived peptides. Controls To create negative control sequences, we generated 100 random 56-mers (17, 61). For positive controls, peptides from previous phage libraries with high seroprevalence were selected and tiled (16, 17). To ensure the comprehensive coverage of these sequences, the full proteins from which these seroprevalent peptides originated were also included. After protein downselection, we then created 56–amino acid peptide tiles with 28 residues of overlap using the pepsyn package (24, 62). We used cd-hit to reduce the amino acid level redundancy of tiled peptides within each major taxonomic group on the basis of a threshold of 95% identity. The pepsyn package was then used to perform reverse translation of each peptide into a DNA sequence, add sequence adapters containing Eco RI and Xho I digest sites (5′-TGAATTCGGAGCGGT-3′ and 5′-CACTGCACTCGAGA-3′) to each end of the resulting oligo, and recode any potential cleavage sites within the DNA sequences (24). Construction of the VectorScan phage library The resulting oligo pool was synthesized by Twist Biosciences and amplified via PCR using the KAPA HiFi HotStart PCR kit (12 cycles). Forward primer: 5′ GTCCTTAGCAGTCAATGATACGGCGTGAATTCGGAGCGG 3′; reverse primer: 5′ GGTTACACTCCATCAAGCAGAAGACGTCTCGAGTGCAGTG 3′. The amplified library digested with Xho I and Eco RI-HF at 37°C for 60 min. A postdigest cleanup was performed using the Monarch PCR & DNA Cleanup Kit. The oligo phage library was cloned into T7 bacteriophage, packaged, amplified, and used for PhIP-seq using a slightly modified version of previous protocols (25, 63). Briefly, the T7F bacteriophage was provided by S. Elledge at Harvard Medical School. It was amplified in BLT5403 and digested with Sal I-HF and Eco RI-HF. The digest was cleaned up using the Monarch PCR & DNA Cleanup Kit and run on a 2% low melting point gel. The band containing the T7F DNA was extracted and eluted using the Zymoclean Large Fragment DNA recovery kit. The resulting DNA was concentrated using Amicon spin filters, and the concentration was measured via Qubit. The digested vector DNA and insert DNA were ligated using T4 DNA ligase at 16°C overnight in a thermocycler. The ligated product was packaged using the T7 Select Packaging Extract kit, and quantitative PCR was used to confirm that the ligation was successful. The resulting phage library then underwent two rounds of amplification on plates. The resulting lysate was titrated with a plaque assay and subsequently used for PhIP-seq. Two distinct library constructions were performed using this protocol, starting from the same Twist library digest. Construction no. 1 was used to run plate 13 (containing the NHP samples). Construction no. 2 was used to run plates 22, 23, 27, and 28 (containing all the human samples). PhIP-seq procedure We screened each sample with VectorScan in duplicate alongside no-serum controls, where the biological sample was replaced with an equal volume of PBS. First, each relevant well of a 1.1-ml 96-well plate was filled with 1 ml of tris-buffered saline with Tween 20 + 3% bovine serum albumin and rotated overnight at 4°C. The following day, the phage library was prepared by combining the expanded library (~5 × 10 plaque-forming units/ml) with chloramphenicol (50 μg/ml) and kanamycin (50 μg/ml). The blocking solution was removed from the plate, and 1 ml of the prepared library and 10 μl of diluted sample (~2 μg of IgG) were added to each relevant well and rotated overnight at 4°C. Next, immunoprecipitation was performed by adding 40 μl of a 1:1 mixture of Protein A and Protein G Dynabeads (Thermo Fisher Scientific) to each well and incubating with rotation for 1 hour at 4°C. The beads were washed five times with cold radioimmunoprecipitation assay buffer, resuspended in M9LB + carbenicillin in a 2-ml deep-well plate, and added to 1 ml of Escherichia coli (BLT 5403) grown to an OD600 ~ 0.5. The plate was then incubated in an orbital shaker at 37°C until the wells cleared. Sodium chloride (NaCl) was added to each well, and the plate was centrifuged at 3000g for 1 hour at 4°C. Two microliters of lysate was used as input for a PCR to amplify the peptide-encoding inserts. A subsequent PCR2 reaction was used to apply indices to each sample. The indexed samples were pooled and submitted for sequencing. Plate 13 was processed by the Harvard Medical School Biopolymers Facility and run with single-end sequencing. Plates 22, 23, 27, and 28 were sequenced by the Harvard University Bauer Core Facility with paired-end sequencing. To obtain counts for subsequent analyses, the adapters were trimmed from each read using cutadapt and aligned to the DNA sequences in the VectorScan library using bowtie-2 (--very-sensitive option), samtools, and GNU Parallel (64–67). Plates 13 and 23 were aligned locally, and plates 22, 27, and 28 were aligned on the Harvard FAS Research Computing cluster across eight central processing unit cores with 32,000 megabytes of random-access memory per core. For plate 13, multimapped reads were filtered out (using samtools view -b -q 1 -F 4). For paired-end reads, read counts were determined on the basis of concordant pairs (samtools view -b -f 2). Determination and quantification of seroreactive and seronegative peptides Determination of seroreactivity Samples were analyzed on a per-plate basis. For all human samples, samples were excluded from analysis if the raw read counts were poorly correlated or if one of the replicates for the sample did not reach an adequate number of reads. These sample exclusion criteria were not used for plate 13 (which was used to run the NHP samples) because of the small sample size. High correlations between the raw read counts from surviving replicates indicate technical consistency (fig. S1). To quantify signal, peptide counts were first normalized by sequencing depth within each replicate (on the basis of counts per 100,000 reads, RPK). The mean (μ) and standard deviation (σ) across the PBS-only wells was calculated on a per-peptide basis, and a limit of detection (LOD) for each peptide was set at μ + 5σ. Each sample was run in duplicate on the plate; therefore, to be considered seroreactive, we required that both replicates each needed to have at least five counts and an RPK value above the LOD. Next, in the process of examining the sequences of the arbovirus-derived signal from the NHPs, we observed that a subset of seroreactive peptides contained GS-linkers (repeat sequences of glycine-serine residues) comprising much of the peptide sequence. These GS-linkers are an artifact introduced by the pepsyn package because of poor sequence quality from the original UniProt entries. Therefore, we excluded peptides containing the subunit of “GSGGGSGGGSG” to help filter out peptides dominated by these sequences. This downselection was performed across the entire library, decreasing the number of peptides being analyzed from 253,789 to 248,577 (data S1). The largest impact was on the arboviral section of the library. Within a sample, each of these 248,577 peptides was characterized as seroreactive or not seroreactive on the basis of whether both replicates had adequate counts and signal above the LOD. However, a peptide is only considered seronegative if both replicates failed to reach these criteria, rather than because they were inconsistent (i.e., one fulfilled the criteria, while the other failed). Across all samples, a majority of the library had consistent signal, indicative of technical accuracy (data S3). The proportion of inconsistent replicates was higher for the NHPs, likely because this plate was run with single-end sequencing. For each sample, peptides with consistent signal were deemed “seroreactive” or “seronegative” on the basis of whether they had signal above or below the thresholds for noise. These concordant peptides were used as the basis for downstream analyses. Last, to maximize confidence in our results, if longitudinal samples were available for a subject, peptides needed to have concordant signal at all time points for that subject. Signal quantification for downstream analyses If a peptide was seroreactive, the average of RPKs across replicates was used for downstream analyses. If a peptide was seronegative, its RPK value was artificially set to 0 to avoid quantifying what is effectively noise in downstream calculations such as fold change. Spearman correlations were used to evaluate stability and biological consistency in the longitudinal samples (Fig. 2C and fig. S1). To evaluate biological and technical consistency in the nonlongitudinal samples, a subset of 16 samples was run on both plates 22 and 27. The average RPK values from these samples were calculated on each plate and found to be well correlated (fig. S1). As a result, the values from plate 22 were used to perform downstream analyses. Comparison of seroprevalence rates among control peptides We evaluated the validity of this approach by calculating the seroprevalence rates of the positive control peptides that were included in VectorScan. These include peptides from common viruses such as rhinovirus B, respiratory syncytial virus, and enterovirus B. Seroprevalence in the adult NIH and CDC cohorts was in line with previous studies (data S4). GSEA-like analyses NHPs Within each NHP, we first identified peptides that were seroreactive at any time point. We then calculated the fold change in RPK (after averaging replicates, as described above) between the pre- and early-immune time points. A scaling factor of 0.1 was added to the numerator and denominator to avoid division by zero. For each NHP, these differences were ranked from the largest to the smallest (giving peptides with a higher signal at the early-immune time point a higher rank) and log2 transformed to stabilize variance. Each peptide was labeled with its species, and we performed one-sided Mann-Whitney tests for each species represented among the seroreactive peptides for that NHP to determine whether peptides from that species had significantly higher rankings compared to the rest, which would be indicative of increased signal postexposure. To create Fig. 2F, P value correction via BH was applied across all species-wise tests from all animals. The same procedure was followed to analyze the pre- versus late-immune signal. CDC We first identified all peptides that were seroreactive in at least 1 of 62 samples from this cohort. To identify whether a higher CHIKV-peptide signal was associated with CHIKV IgG ELISA seropositivity, we created two categories of samples: CHIKV IgG+ samples (n = 19, samples with profiles 2, 4, and 5; Fig. 3A) and CHIKV IgG− (n = 25, samples with profiles 1 and 3). For each peptide, the mean RPK was calculated for each category, and the fold change between the means was calculated (again, adding 0.1 to each value to avoid division by zero). The same Mann-Whitney procedure was performed as above for the NHPs. To evaluate the association of DENV-derived signal with DENV IgG+ individuals, we used the same base of seroreactive peptides but categorized samples on the basis of whether they were DENV IgG+ (n = 28, profiles 2 and 3) or IgG− (n = 16, profiles 1, 4, and 5) and used the Mann-Whitney pipeline to evaluate enrichment. NIH Within each person, we identified peptides that were seroreactive at either time point. We calculated the fold change in RPK between day 0 and day 44 and followed a similar procedure to the NHPs to transform and rank the differences. Each peptide was labeled with its group (i.e., bacteria-derived versus vector-derived), and one-sided Mann-Whitney tests were used to determine whether peptides from that group had significantly higher rankings compared to the rest. To create fig. S4, P value correction via BH was applied across all group-wise tests from all individuals. Defining serotype-informative peptides and quantifying signal We mapped every DENV-derived peptide in VectorScan to each strain used for inoculation. After performing a threshold analysis of sequence homology (fig. S6), we defined peptides as “serotype-informative” if they had at least 85% identity to one strain and less than 80% identity to the other two strains. On the basis of this definition, 2669 of 5660 DENV-derived peptides from across the DENV proteome (fig. S6) were serotype-informative (DENV1: n = 892; DENV2: 1192; DENV4: 585). Quantification of cluster reactivity scores For each relevant cohort, the peptides that were seroreactive in at least one sample were used to create a blastp database (example command: makeblastdb -in CHIKV_peptides_of_interest.faa -out db -parse_seqids -dbtype prot). blastp was then run with the “-max_hsps 1” option. Query-target matches with a bit-score ≥40 were organized into clusters in igraph on the basis of Louvain community detection. Clusters with ≥2 members were considered to represent an epitope-containing motif and kept for downstream analysis. For each sample, the RPKs from seroreactive peptides in each cluster were summed. Samples were grouped on the basis of exposure history or other relevant categorizations. Two-sided Wilcoxon tests were performed with BH correction. Clusters with a Padj < 0.05 were considered significant. Modeling relationships between seroreactive peptides and molFOI Peptide breadth was estimated by summing up the number of seroreactive peptides derived from Anopheles or P. falciparum in each month 7 sample. Because this breadth is a count value, the “glm” function in R was used to fit a quasi-Poisson model (which was selected to account for overdispersion) between the molFOI of each sample and peptide breadth from the organism of interest. Protein-level annotation of mosquito-derived peptides Using the sequences from the 11,839 vector-derived proteins that were tiled through to create the vector portion of VectorScan, we performed DIAMOND mapping to identify OrthoMCL entries for each protein (using OrthoMCL version 7.0) (68). The Anopheles peptides pulled from the Falciparome library were separately mapped to their respective orthogroups. This yielded orthogroup annotations for 156,817 of 156,878 (99.96%) of mosquito-derived peptides in the cleaned library (i.e., the library after exclusion of long GS-linkers, given that only those peptides will have been considered for determination of seropositivity). Peptides that failed DIAMOND mapping had an orthogroup of “−1”. Among the 43,869 mosquito-derived peptides that were seroreactive in at least one human sample and clustered with at least one other peptide, we first determined which orthogroups had the highest representation. To determine whether each orthogroup was overrepresented within this set of peptides, we created a contingency table for each (Table 1). Table 1.Contingency table format for identifying overrepresented orthogroups among seroreactive mosquito-derived peptides.The orthogroup of each mosquito-derived peptide in the library was determined using OrthoMCL. Overrepresentation of each orthogroup was evaluated using FETs.Peptide has annotation XPeptide does not have annotation XIs a hitXYNot a hitZ156,878-X-Y-Z Contingency table format for identifying overrepresented orthogroups among seroreactive mosquito-derived peptides. The orthogroup of each mosquito-derived peptide in the library was determined using OrthoMCL. Overrepresentation of each orthogroup was evaluated using FETs. A two-sided FET was then performed, and multiple comparisons were accounted for using BH correction. Both OR > 1 and Padj < 0.05 were required for an orthogroup to be considered overrepresented. To gain functional information about these immunogenic targets, for the top 10 orthogroups that emerged from the frequency and significance analyses, respectively, the PFam domain or keyword most often associated with that orthogroup was determined on the basis of annotations available in the OrthoMCL database.