Polyprotein sequences of the Yellow Fever 17DD vaccine virus (YFV-17DD, Genbank: U17066.1 and DQ100292.1), wild strains of Yellow Fever virus (YFV Asibi, Genbank: AWG42197.1; YFV Outbreak, Genbank: MF538784.2, MF370533.1 and MF405338.1; YFV Pre-Outbreak, Genbank: JF912190.1, JF912188.1), Zika virus (ZIKV, Genbank: KU926310.2, MF352141.1, KX197205.1 and KU527068.1), Dengue virus serotypes 1-4 (DENV1-4, Genbank: JX669466. 1, KP188543.1, KP188551.1, JX286516.1, JX669477.1, KP188556.1, GU131865.1, GU131877.1, GU131844.1 and KJ596662.1), Chikungunya virus (CHIKV, Genbank: KY704952.1, KY704954.1 and KY704947.1), Mayaro virus (MAYV, Genbank: KM400591.1, KY618136.1, KT818520.1, KY618131.1 and KY618132.1) and Oropouche virus (OROV, Genbank: KP795103.1, KP795100. 1, KP795088.1, MG747607.1, AF484424.1, AWU67042.1, AWU67043.1) were acquired on the NCBI Nucleotide database (https://www.ncbi.nlm.nih.gov/nucleotide/).

All collected sequences were used as input to NetCTL 1.2 web software (http://www.cbs.dtu.dk/services/NetCTL/) for prediction of peptides with restriction to HLA-A1, HLA-A2, HLA-A3 or HLA-A24. Peptide prediction was based on MHC-I affinity, peptide cleavage and transport by transporter associated with antigen protein complex (TAP) (21). Only 9-mer peptides with Combined Score values above to 0.75 were considered immunogenic for CD8⁺T cells and maintained for further analysis. This evaluation encompassed four of the most prevalent MHC-I haplotypes in Brazilian population (22).

Evaluation of the HLA-A*02-restricted immunodominant peptide repertoire was performed using PEPperCHIP^© (PEPperPRINT^©, Heidelberg, Germany) custom peptide microarray ( Supplementary Figure 1 ). All predicted peptides for YFV-17DD virus, YFV wild strains, DENV1-4, ZIKV, CHIKV, MAYV and OROV with combined score values above 0.75 were included in this microarray, including non-overlapping peptides. Supplementary Figure 1 shows a schematic compendium for the detection of HLA-A2-peptide binding using PEPperCHIP^© microarray (PEPperPRINT^©, Heidelberg, Germany).

For the microarray, 1,350 peptides were adsorbed in duplicate on the inside of spots located on a microarray glass slide (75.4 mm x 25.0 mm x 1 mm). The first step of the assay consisted of pre-labeling the microarray glass slide with anti-murine IgG1/Cy3 secondary antibody (BD Biosciences, California, USA), for exclusion of background reactivity. For this, the microarray slide was incubated with standard buffer [phosphate-saline buffer (PBS) + 0.05% Tween20; pH 7.4] for 15 minutes at room temperature. After complete removal of the buffer, a new incubation was performed with blocking buffer [PBS + 0.05% Tween20 + 1% bovine serum albumin (BSA); pH 7.4] for 30 minutes at room temperature. After complete removal of the buffer, the microarray slide was incubated with anti-murine IgG1/Cy3 secondary antibody (BD Biosciences, San Jose, CA, USA) previously diluted (1:5,000) in staining buffer [‘standard buffer ‘+ 10%’ blocking buffer ‘]. Incubation was maintained for 45 minutes, at room temperature and protected from light. After complete removal of the secondary antibody, the microarray slide was washed three times in standard buffer, air-dried and digitized.

After the background staining, the microarray slide was incubated with standard buffer at room temperature for 15 minutes. After complete removal of the buffer, 1µg/mL, 10µg/mL and 30µg/mL of HLA-A*02:01 dimer (HLA-A2:β2M:Ig Protein) (23) diluted in staining buffer were added to the microarray glass slide and incubated for 16 hours at 2-8°C. After complete removal of the HLA-A*02:01 dimer, the microarray slide was washed three times using standard buffer and incubated with anti-murine IgG1/Cy3 secondary antibody (BD Biosciences, San Jose, CA, USA) previously diluted (1:5,000) in staining buffer. Incubation was maintained for 45 minutes, at room temperature and protected from light. After complete removal of the secondary antibody, the microarray slide was washed three times in standard buffer, air-dried and digitized.

Throughout the assay, all incubations were performed under constant agitation in an orbital agitation system (140 rpm). The PEPperCHIP^© microarray slide (PEPperPRINT^©, Heidelberg, Germany) was digitized on the Affymetrix 428 Array Scanner device (Thermo Fisher, California, USA). Data regarding HLA-A2:β2M:Ig reactivity to adsorbed peptides was pre-analyzed using PepSlide^® Analyzer (PEPperPRINT ^©, Heidelberg, Germany) and quantified in 16 bit grey-scale images. Results were expressed in fluorescence intensity.

Statistical analyses were performed using the GraphPad Prism v.8.0 software (GraphPad Software, California, USA). Non-parametric data distribution was confirmed by the Shapiro-Wilk test. Wilcoxon test was applied for statistical comparisons involving only two groups. Multiple comparisons were performed using the Kruskal-Wallis method followed by Dunn’s post-test, in the case of unpaired data, and the Friedman method with Dunn’s post-test to compare paired data. Chi square test was applied for the comparison of unpaired data regarding number of reactive peptides. Spearman’s rank correlation test was applied for the correlation of in vitro reactivity and in silico prediction of cytokine induction. Differences were considered statistically significant at p <0.05. Principal component analysis (PCA) was performed using Minitab 19.

Immunodominance and haplotype restriction are very important parameters for the design of multiepitope proteins applicable to the differential diagnosis and vaccine development. The analysis which aimed at understanding in silico similarities and divergencies and HLA restriction for immunodominant arboviral peptides revealed that overlap amongst the arboviruses were found for peptide binders of different haplotypes: HLA-A1 (n=1,005), HLA-A2 (n=1,693), HLA-A3 (n=1,514), and HLA-A24 (n=999). A higher number of HLA-A*02:01-restricted peptides were predicted for YFV-17DD, YFV wild strains, DENV 1-4, ZIKV, CHIKV and MAYV, when compared to other HLA-A haplotypes. For OROV, peptide restriction to HLA-A3 was prevalent, however an elevated number of HLA-A*02-restricted peptides was also observed for this arbovirus ( Figure 2 ). Considering this results, HLA-A*02-restricted peptides were selected for further evaluations.

Arboviral epitopes which overlap amongst viruses were found for peptide binders of different viral families. As demonstrated by Figure 3 , overlapping was demonstrated by the number of peptides from YFV wild strains, ZIKV, DENV 1-4, CHIKV, MAYV and OROV that correspond to YFV-17DD peptides (namely the number of matches). After selection of HLA-A*02:01-restricted-peptides, an identity matrix between 17DD-YFV and ZIKV, DENV 1-4, CHIKV, MAYV and OROV was created to identify overlapping peptides. From over 1700 epitopes, only three YFV-17DD peptides with combined score ≥0.75 were common to one or more arbovirus. Namely, peptide YFV_17DD_1995_127 presented complete overlapping with DENV2_PE1_116 and DENV3_BR5_112; peptide YFV_17DD_1995_42 presented complete overlapping with DENV3_BR5_25, and peptide YFV_17DD_1995_76 presented complete overlapping with DENV1_SJRP1_62, DENV3_BR5_55, DENV4_62 and ZIKV_RN1_56. No epitopes predicted for CHIKV, MAYV or OROV were shared by YFV-17DD vaccinal strain ( Figure 3A ).

To expand the assessment of the peptide signature shared among the selected arboviruses and YFV-17DD, peptides exhibiting at least 50% of similarity with epitopes predicted for YFV-17DD were selected ( Figure 3B ). From 244 peptides matches of complete or partial overlapping (>50% identity), 195 belonged to other flavivirus (ZIKV n=36; DENV1 n=46; DENV2 n=48; DENV3 n=35; DENV4 n=30), 31 were from togaviruses (CHIKV n=16; MAYV n=15) and 19 matched peptides were found for OROV (peribunyavirus). As expected, wild strains from yellow fever virus presented an elevated number of matched peptides with 17DD vaccinal strain (YFV Asib n=163; YFV Outbreak n=162; YFV Pre-Outbreak n=168) ( Figure 3C ). The same tendency was observed for identity average. Peptides from ZIKV and DENV serotypes 1-4 had between 61.15% (DENV1) and 67.65% (DENV3) of similarity with YFV-17DD peptides. Means of identity for CHIKV, MAYV and OROV remained close to baseline, ranging from 55.60% (CHIKV) to 56.18%. These observations were a clear contrast to peptides from YFV wild strains, that presented average identity close to 90% (93% in YFV Asib; 89.65% in YFV Outbreak and 89.36% in YFV Pre-Outbreak), which can be further observed as the noticeable inversion of the violin plot ( Figure 3D ). Other prediction parameters were also evaluated and are shown in Supplementary Figure 1 .

Intending to extend on the understanding of immunodominant peptides from Arbovirus, in vitro assessment of peptide-HLA-A*02:01 reactivity was evaluated. For that, a peptide microarray was designed and generated as shown in Supplementary Figure 2 (PEPperPRINT^©, Heidelberg, Germany). The peptide array employed fixed concentrations of peptides (1nM) and three different concentrations (1μg/mL, 10μg/mL or 30μg/mL) of a dimeric protein containing HLA-A*02:01 domain fused to β2-microglobulin (HLA-A2:β2M:Ig Protein) produced in house. Global analysis of the fluorescence intensity for peptide-HLA-A*02:01 binding indicated a dose-response effect regarding the concentration of the HLA-A*02:01 dimeric protein with fixed concentrations of peptide in the array. The same pattern was observed for peptides from all arboviruses taken together and also for peptides classified by viral families (Flaviviridae: YFV, DENV1-4 and ZIKV; Togaviridae: CHIKV and MAYV; Peribunyaviridae: OROV), as indicated in the upper panels of Figure 4 .

The absolute number of HLA-A*02:01-reactive peptides in all three concentrations of HLA-A*02:01 dimeric protein remained stable in assessment of all arboviruses taken together and for viruses belonging to the Flaviviridae and Togaviridae families. For OROV, it was possible to observe an increase in the number of reactive peptides in the concentration of 30μg/mL as compared to a concentration of 10μg/mL of HLA-A*02:01 dimeric protein ( Figure 4 - middle panels).

Despite the stability in the number of HLA-A*02:01-reactive peptides in two of the three viral families analyzed, the trend of the dose-response effect was confirmed by the evaluation of the mean fluorescence intensity. In this case, there was a progressive increase in the reactivity of peptides derived from arboviruses analyzed together and for peptides derived specifically from the Flaviviridae family. As for the peptides from Togaviridae family, the reactivity was higher for the concentrations of 10μg/mL and 30μg/mL, but no difference was observed between these two concentrations. As for peptides derived from OROV (Peribunyaviridae:family), the highest reactivity was observed for the concentration of 30μg/mL of HLA-A*02:01 dimeric protein ( Figure 4 - lower panel).

Quantitative analysis of HLA-A*02:01-reactive peptides derived from arboviruses were further evaluated and demonstrated that non-overlapping immunodominant epitopes display the highest HLA-A*02:01 affinity binding assessed in vitro. First, data analysis demonstrated that DENV strains show the highest number of reactive peptides among Flavivirus for all three HLA-A*02:01 dimer concentrations ( Figure 5A ). Considering all arbovirus sequences studied, the number of DENV-derived peptides with HLA-A*02:01 reactivity was the highest in two of the three HLA-A*02:01 dimer protein concentrations, mainly at 30μg/mL. In addition, CHIKV, next to DENV, presented the highest number of reactive peptides at 10μg/mL ( Figure 5A ). On a second note, Figure 5B shows that a significant higher number of non-overlapping peptides presented reactivity when compared to YFV-17DD overlapping peptides. This was observed for peptides derived from all assessed arboviruses except for YFV wild strain derived peptides, for which most of HLA-A*02:01 reactive peptides presented species-specific overlapping with YFV-17DD sequences, as expected ( Figure 5B ).

Of note, the assessment of HLA-A*02:01-reactive peptides across virus polyproteins highlighted non-structural proteins as “hot-spots” for HLA-A*02:01-restricted peptides ( Figure 6 ). The identification of HLA-A*02:01-reactive peptides along viral polyproteins indicated that non-structural proteins encompass the highest HLA-A*02:01-restricted peptides binders. This effect was observed when peptides reacted at all three concentrations of HLA-A*02:01 dimeric protein. The global fluorescence intensity for HLA-A*02:01 binding indicated a significant increase in the reactivity of peptides for non-structural protein in a dose-dependent manner as indicated in Figure 6 .

Hydrophobicity of peptides is a crucial parameter to understand protein binding affinity (30–32). Therefore, assessment of hydrophobicity was performed for the YFV-17DD polyprotein in order to understand the influence/correlation of this parameter with immunodominance results observed in silico and in vitro. Data analysis indicated the presence of major hydrophobic sites in the final segment of non-structural protein 1 throughout 2a (Ns2a) and in non-structural proteins 2b (Ns2b), 4a (Ns4a) and 4b (Ns4b). Conversely, structural proteins from YFV-17DD such as the capsid and membrane presented only restricted areas with elevated hydrophobicity ( Figure 7A ). The establishment of global median for HLA-A*02:01-peptide binding affinity using in silico score allowed the identification of peptides with high, medium and low affinity score values in all arboviruses. The alignment of representative YFV-17DD peptides with different ranges of HLA-A*02:01 predicted in silico binding scores revealed that peptides with high and medium affinity scores are derived from non-structural Ns2b (YFV-17DD_1995_1), Ns4a (YFV-17DD_13 and YFV-17DD_1995_74), and Ns1 (YFV-17DD_1995_68). As for the peptides with low HLA-A*02:01 affinity scores, these sequences were derived from YFV Ns3a and Ns5 ( Figure 7A ). Regarding peptides with complete overlap amongst YFV-17DD and other flaviviruses, two peptides were derived from Ns3 (YFV-17DD_1995_7, observed in DENV1, DENV3, DENV4 and ZIKV; and YFV-17DD_1995_127, common to DENV2 and DENV3) and one peptide was derived from the Ns5 portion (YFV-17DD_1995_42, also found in DENV3) ( Figure 7A ).

For the assessment of amino acid patterns within peptides of different HLA-A*02:01 affinity scores, two sequences of high, medium and low affinity scores were aligned for the identification of a consensus sequence, highlighting all conserved residues among the assessed arboviruses ( Figure 7B ). For peptides with high HLA-A*02:01 affinity score, the presence of leucine in position two and valine in position nine were predominant. As for sequences with relative low HLA-A*02:01 affinity score, a remarkable presence of leucine in position 9 was observed ( Figures 7B, C ), occurring in 23 of the 24 aligned peptides ( Supplementary Figure 3 ). A distinct pattern of conserved amino acids was not observed for peptides with medium HLA-A*02:01 predicted affinity score. However, sequence logo analysis allowed the identification of an intermediary profile, with a slight increase of leucine in the ninth position and a decreased proportion of this same amino acid in the second amino acid position ( Figure 7B ). Only peptides with high score of HLA-A2 binding presented both leucine in the second position and valine in the ninth position simultaneously, in a frequency of 37.50% ( Figure 7C ).

Next, the predicted ability to induce cytokine production induced by peptides with complete or partial overlap with YFV-17DD, as well as peptides from the YFV vaccinal strain was also evaluated. For that, Interferon-gamma (IFN-γ) and Interleukine-4 (IL-4) predicted scores were calculated and contrasted amongst viral families. Data analysis showed that induction of IFN-γ was notably higher among peptides from almost all assessed arboviruses. Only DENV1, MAYV and OROV presented negative median values (cut-off = 0) ( Figure 8A ). IL-4 induction by peptides of all arboviruses presented negative median values. Additionally, ZIKV, DENV 1-4, CHIK and OROV presented 75% of peptides with negative valued score for IL-4 induction ( Figure 8A ).

Comparative analysis of IL-4 and IFN-γ induction within viruses confirmed that peptides from all arboviruses presented a substantial bias of peptides towards Th1 IFN-γ immune response axis. Overlapping peptides followed the same trend. This observation is confirmed by the fact that all three peptides common to YFV vaccinal strain and one peptide from Flavivirus family were predicted as IFN-γ inducers but not IL-4 inducers ( Figure 8B ). Overall, peptide-HLA-A*02:01 in vitro reactivity was directly correlated to IFN-γ score (p=0.01206; R=0.005365) and inversely correlated to IL-4 score (p=0.0005874; R=0.004536), as displayed in Figure 8C . In addition, principal component analysis (PCA) showed that cytokine induction may be triggered according to peptide biochemical properties. While peptide charge, isoelectric points and hydrophobicity seem to cluster closer to the peptide IFN-γ induction power, IL-4 induction seems to be closely related to peptide hydrophilicity, molecular weight and net hydrogen ( Figure 8D ). Considering that peptide charge is influenced by acid residues, frequencies of aspartic acid (D) and glutamic acid (E) was assessed in peptides capable of inducing IFN-γ production. The results confirmed that peptides considered to be IFN-γ inducers presented lower frequencies of aspartic acid (D) and glutamic acid (E) when compared with non-inducer peptides ( Figure 8E ).