Amino acid sequences of the PEDV S and M proteins were retrieved from all sequences in the NCBI database. After filtering with Sublime Text 3 to remove incomplete and repeated sequences, 560 and 924 sequences of the M and S proteins were obtained, respectively. The sequences of each protein were next grouped and phylogenetic trees were generated. Sequences of the M protein were classified into 5 groups, while sequences of the S protein were classified into 14 groups ( Figures 1A, C ). The consensus sequence of each group of both proteins was next generated and aligned as shown in Supplementary Figures 1, 2 . Percent similarity among different groups was in the range from 32% to 99.71% for the S protein and from 20.35% to 99.55% for the M protein. The M and S consensus sequences were subsequently aligned with representative sequences of each genotype and subgenotype of PEDV and a phylogenetic tree was generated using the SeaView program. Consensus sequences of groups M1–M4 were clustered in genotype G2 and only consensus sequence of the M5 group was clustered in genotype G1 which is genetically related to CV777 ( Figure 1B ). For the S protein, only the consensus sequences of groups S2 and S14 were clustered in genotype G1, while all the other sequences were clustered in genotype G2 ( Figure 1D ).

We employed immunoinformatics to screen for B-cell epitopes from diverse PEDV strains. Linear B-cell epitopes have been suggested to be correlated with surface accessibility, flexibility, coil probability, antigenicity, and hydrophilicity (31, 32); thus, we utilized multiple immunoinformatics tools to predict linear B-cell epitopes and peptide characteristics. BepiPred-2.0 was used to predict linear B-cell epitopes. In parallel, IUPred was used to predict intrinsically unstructured proteins inferring coil probability, while other features including accessibility, antigenicity, and hydrophilicity were predicted using the methods of Emini, Kolaskar and Tongaonkar, and Parker, respectively ( Figure 2A ).

To search for a prediction tool/method that gives the best performance in terms of accuracy and epitope coverage, we aligned and compared all predicted peptides obtained from each prediction tool with the reference epitopes. Four well-known B-cell epitopes of the PEDV S protein, namely, i) peptide M (MQYVYEPTYYML) (23), ii) S1D5 (PVLVYSNIGVCKSGSI) [with the core epitope SS2 (YSNIGVCK)] (22), iii) S1D6 (SGSIGYVPLQDGQVKI) [with the core epitope SS6 (LQDGQVKI)] (22), and iv) 2C10 (GPRLQPY) (24, 25), were exploited as reference epitopes ( Figure 2A ). Mapping of the reference epitopes with predicted peptides revealed that none of the prediction tools could positively predict all four reference epitopes ( Figures 2A, B ); we, thus, created three new prediction methods, designated methods A, B, and C, for identifying and selecting linear B-cell epitopes as described in Figure 2C . In method A, epitope is identified based on the peptide with the following characteristics: i) predicted to be B-cell epitope by BepiPred, ii) composed of coil structure predicted by BepiPred and/or IUPred, and iii) accessible peptide predicted by BepiPred and/or Emini method. For method B, the epitopes are identified based on the peptides predicted to be antigenic and hydrophilic by the methods of Kolaskar and Tongaonkar and Parker, respectively. For method C, the peptides predicted to be antigenic and accessible by the methods of Kolaskar and Tongaonkar and Emini, respectively, are considered B-cell epitope. By using our prediction methods, the reference epitopes S1D6 and 2C10 were positively predicted by all three methods, while epitope S1D5 and peptide M were positively predicted only by methods B and C, respectively.

The consensus sequences from all 14 groups of the S protein and 5 groups of the M protein were subject to linear B-cell epitope prediction. Putative B-cell epitopes were identified and selected based on prediction methods A, B, and C, and prediction of the M protein group 3 (M3) is shown as a representative ( Figure 3A ). Note that the predicted peptides that overlap or are located in close proximity were combined into one long peptide. Prediction of the M protein yielded seven putative B-cell epitopes, designated M-1 to M-7 ( Figures 3A, B and Table 1 ). Prediction of B-cell epitopes on the S protein from one representative group (out of 14 groups) using our methods is shown in Supplementary Figure 3 . Predicted peptides from 14 S consensus sequences were aligned, and a total of 50 peptides located in 33 regions of the S protein, designated S1B and S2B for the epitopes located on the S1 and S2 subunits, respectively, were predicted as B-cell epitopes ( Figure 4 , Table 1 ). In the COE region, three epitopes, namely, S1B-15, S1B-16, and S1B-17, were identified. Most of the B-cell epitopes predicted from consensus sequences of different groups are located on the same regions although some residues in the epitope are variable among different groups. Impressively, our immunoinformatics methods could identify all published epitopes in the S protein but not in the M protein as indicated in Table 1 and Supplementary Figures 4, 5 .

We further analyzed the strain coverage of each predicted epitope as information toward epitope conservancy could be beneficial to the design and development of a vaccine, particularly a universal vaccine. For the M protein, six out of seven predicted epitopes cover higher than 70% of 560 accession numbers with four epitopes (M-3, M-4, M-5, and M-6) having strain coverage greater than 90% ( Figure 5A ). Analyzed with 924 accession numbers of the PEDV S protein, 34 out of 50 predicted epitopes of the S protein exhibited strain coverage greater than 70% ( Figure 5B ). Among these 34 epitopes, 24 showed strain coverage higher than 80%, of which 8 and 16 are located on the S1 and S2 subunits, respectively, suggesting that sequences in the S1 subunit are more variable than those in the S2 subunit.

To further validate the predicted linear B-cell epitopes, ELISA was performed. PEDV-positive sera were prepared from F1 and F3 pigs raised in the farms and naturally infected with PEDV. Neutralizing activity of the pig sera was investigated using neutralization assay and only serum samples with neutralizing antibody titers at least 32 (reciprocal serum dilution) were further used in ELISA analysis. A total of 12 F1 pigs and 11 F3 pigs showed neutralizing activity that passes the criteria ( Figure 6 ) and were subject to ELISA. Control sera were prepared from 10 uninfected F3 pigs raised in the animal research facility, and their neutralizing antibody titer was confirmed to be lower than 32 by neutralization assay ( Supplementary Table 1 ).

Three well known epitopes, SS2, SS6, and 2C10 (22, 25), were included in ELISA as reference epitopes. Reactivity of the peptides with their target antibodies was determined based on statistical analysis comparing OD450 value of the F1 and F3 pig sera to that of the control group (p < 0.05). All seven peptides from the M protein showed reactivity with sera from at least one pig strain ( Figure 7A ). Out of 50, 48 peptides of the S protein showed reactivity with the sera from at least one pig strain ( Figure 7C ). Immunodominant and subdominant epitopes were next defined. In reactivity with a particular peptide, if all sera from both F1 and F3 pigs exhibited OD450 value higher than OD450 mean of the control group + 2SD, such peptide was considered an immunodominant epitope. On the other hand, if all sera from both F1 and F3 pigs exhibited OD450 value higher than OD450 mean of the control group + 1SD, such peptide was considered a subdominant epitope. Based on these analyses, the epitopes S1B-1, S1B-2.2, S1B-3, S1B-9.2, S1B-14.2, S1B-19.1, S1B-19.3, S2B-22, S2B-25, S2B-29, and S2B-30.2 on the S protein and the epitope M-2 on the M protein were categorized as immunodominant epitopes, while the epitopes S1B-2, S1B-5, S1B-9.1, S1B-15, S1B-16, S1B-19.2, S2B-21, S2B-24, S2B-26.1, and S2B-32 were categorized as subdominant epitopes ( Figures 7A–D and Table 2 ). Two subdominant epitopes, S1B-15 and S1B-16, are located within the COE domain. Furthermore, we demonstrated that antibody response in the sera with high neutralizing titer (NT ≥ 64) and low neutralizing titer (NT = 32) against some epitopes was significantly different (p < 0.05) as indicated in Figures 7A , C . This information may suggest the correlation between antibodies targeting these epitopes and neutralizing activity of the serum.

Based on the B-cell epitopes we identified in the present study, some have already been reported. The epitope S1B-19 (aa. 719–772) is a part of the S1D domain (aa. 636–789) containing neutralizing epitopes (34). The S1B-19.1 epitope (aa. 719–730) overlaps with the known epitope S1D SE16 (722-SSTFNSTREL-731) (33), while the S1B-19.3 epitope (aa. 748–772) consists of two known epitopes, SS2 (22) and SS6 (22). The S1B-8 epitope (aa. 203–211) corresponds to the known epitope, peptide M (23), while the epitope S2B-33 (aa. 1,345–1,379) contains the 2C10 epitope (25). The epitope S1B-14.1 (aa. 457–477) entirely maps in the neutralizing epitope (aa. 432–481) (12). In the COE region, all three epitopes (S1B-15 to S1B-17) are parts of the conformational epitope (572-TLQSVNDYLSFSKFCVSTSLLASACTIDLFGYPEFGSGVKFTSLYFQFTKGELITGTPKPLEGVT-636) (12) as underlined. Additionally, the S1B-16 epitope also has three amino acids (GYP) overlapping with the C2-1 epitope (TSLLASACTIDLFGYP) (18). Besides these known epitopes, other B-cell epitopes identified in our study are considered novel. Thus, epitopes S1B-1, S1B-2.2, S1B-3, S1B-9.2, S1B-14.2, S2B-22, S2B-25, S2B-29, and S2B-30.2 can be recognized as novel immunodominant epitopes in the PEDV S protein. For the M protein, one B-cell epitope named M14 (9) has been reported; however, this epitope does not overlap with the epitopes identified in our study. Therefore, all seven B-cell epitopes in the M protein we identified are novel epitopes.

Neutralization ability of the antibodies against each epitope were further tested using neutralization–inhibition assay as shown in the schematic representation in Figure 8A . The assay was conducted with serially diluted serum and 100 TCID₅₀ of PEDV. Neutralization inhibition mediated by a peptide was determined at the endpoint neutralizing antibody titer, at which the amount of antibodies is only sufficient to neutralize the virus and depletion of neutralizing antibodies will result in the loss of the neutralization potential of the serum. Four sera including F1 1-3, F1 1-4, F1 1-5, and F3 1-10 were used in the assay. Peptides corresponding to the known epitopes SS2, SS6, and 2C10 were used as references, and irrelevant peptides N(MHC) and E(CTL) were included as negative controls. All three reference peptides exhibited their ability to inhibit the neutralizing activity of the sera, while irrelevant peptides did not ( Figure 8B ). Notably, peptides from the M protein were tested with only two sera, F1 1-4 and F1 1-5. While the F1 1-4 serum lost its neutralizing activity when incubated with peptides M-1, M-2, M-3, M-4, and M-6, the neutralizing activity of serum F1 1-5 was depleted in the presence of peptides M-1, M-2, M-5, M-6, and M-7 ( Figure 8B , Supplementary Figures 7, 8 ). A combined result from both sera suggested that all seven epitopes of the M protein had the potential to be neutralizing epitopes. However, the three M peptides, namely, M-1, M-2, and M-6, could inhibit the neutralizing activity of both sera tested, suggesting them to be the most promising targets in the M protein for virus neutralization.

In the presence of the S peptides, four pig sera showed different patterns of neutralization inhibition as summarized in Table 2 , Supplementary Figures 6–9 . In Figure 8B , neutralizing epitopes were concluded based on the combined results of all four sera. The peptide that inhibited the neutralizing activity of at least one serum was considered a neutralizing epitope. Almost all peptides, except S1B-7, S1B-11, S1B-14.2, S1B-19.1, S1B-19.2, and S2B-22, could inhibit the neutralizing activity of at least one serum. Hence, epitopes S1B-1, S1B-2.2, S1B-3, S1B-9.2, S1B-19.3, S2B-25, S2B-29, and S2B-30.2 posed as both neutralizing and immunodominant epitopes. Importantly, antibodies recognizing the three epitopes in the COE domain, which are S1B-15, S1B-16, and S1B-17, were demonstrated with neutralization ability. Among the peptides capable of inhibiting the neutralizing activity of the serum, peptides S1B1, S1B-2.2, S1B-5, S2B-26.1, S2B-27, S2B-29, S2B-30, and S2B-33 showed neutralization inhibition in all sera tested, strongly confirming them as promising targets for PEDV neutralization. While most of the former epitope identifications focus on the S1 subunit, our work demonstrates that the S2 subunit indeed harbors multiple immunodominant as well as neutralizing epitopes, representing targets for both vaccine development and antibody detection.

Surface representation of all epitopes is depicted on the prefusion structure of the trimeric S protein of the PEDV strain USA/Colorado/2013 (PDB: 6VV5) (35) using PyMOL 2.3.4. As shown in Figure 9A , most of the epitopes are exposed on the surface, by which surface accessibility is one main feature associated with B-cell epitopes. Although coil is generally recognized as one main characteristic of the linear B-cell epitopes, B-cell epitopes identified in our study were found to be composed of various structures including coil, alpha helix, and beta sheet ( Figure 9B ). However, close-ups of the immunodominant epitopes and neutralizing epitopes in the COE revealed that these epitopes are either partly or entirely composed of coil structure and some epitopes also consist of other structures either alpha helix or beta sheet ( Figures 9C, D ). Close-ups and the secondary structure of all the other epitopes are shown in Supplementary Figure 10 .

Amino acid sequences of the PEDV S and M proteins were retrieved from NCBI (https://www.ncbi.nlm.nih.gov/protein/). The retrieved protein sequences were analyzed individually based on a defined name and length using Sublime Text 3 program. Based on sequence similarity, the amino acid sequences of the S and M proteins were grouped using alignment and phylogenetic tree tools in the SeaView program (version 4). In the SeaView program, these amino acid sequences were aligned using MUSCLE and grouped using a tree based on distance analysis, NJ method, ignore gap, and bootstrap 1,000 replication. The consensus sequence of each group was generated using the MUSCLE method in Unipro UGENE program (version 1.24.2; September 1, 2016) (49, 50). These consensus sequences were then subjected to epitope prediction.

Immunoinformatics tools were exploited to predict B-cell epitopes and properties of the amino acid residues. Firstly, linear B-cell epitope prediction was performed using BepiPred-2.0 (http://www.cbs.dtu.dk/services/BepiPred/), which predicts linear B-cell epitopes from a protein sequence, using a random forest algorithm trained on epitopes and non-epitope amino acids determined from crystal structures, followed by performing sequential prediction smoothing (51). In our study, the residues with the threshold score of epitope probability above 0.5 were considered as parts of a B-cell epitope. Putative B-cell epitopes were selected based on the region with at least six consecutive residues predicted to have epitope probability above 0.5. Notably, BepiPred-2.0 also predicts and provides accessibility and coil probability of each amino acid residue. In addition, other properties of the proteins were also characterized using multiple immunoinformatics tools provided by IEDB (http://tools.iedb.org/bcell/), except IUPred. IUPred (https://iupred.elte.hu/) was used to predict intrinsically unstructured proteins which infers to coil probability (52). The method of Emini (53) was used to predict surface accessibility, while the method of Kolaskar and Tongaonkar, the semi-empirical method (54), was used to predict antigenicity determinant. Hydrophilicity of the protein was predicted using the method of Parker. The results of prediction from each method were aligned and three methods for identifying and selecting candidate B-cell epitopes were created based on the four known B-cell epitopes derived from PEDV protein as described in the result part.

As epitopes were predicted based on consensus sequences of each group of the PEDV S (14 groups) and M (5 groups) proteins, we further analyzed the conservancy of our predicted epitopes. Epitope conservancy was analyzed by calculating the percent strain coverage of the predicted epitopes in comparison to all of the sequences in the group using the formula below.

Peptides corresponding to the predicted epitopes were chemically synthesized (Mimotopes, Australia). All synthetic peptides include i) 57 predicted epitopes; ii) 3 known reference B-cell epitopes, SS2 (YSNIGVCK) (22), SS6 (LQDGQVKI) (22), and 2C10 (GPRLQPY) (25); and iii) irrelevant peptides, N(MHC) (LADSYEITY) and E(CTL) (AVYTPIGRLY). Synthetic peptides were dissolved to the concentration of 3 nmol/μl (3 mM) as a peptide stock in sterile distilled water containing 0.1% acetic acid. Peptide stocks were aliquoted and stored at −20°C until used.

All procedures of sample collection from pigs raised in the farms were performed in accordance with the guidelines of the Institutional Animal Care and Use Committee (IACUC) of Institute for Animals for Scientific Purpose Development and carried out under the regulation and permission of King Mongkut’s University of Technology Thonburi IACUC protocol no. KMUTT-IACUC-2019/018. For the control group, pigs were raised in the animal research facility of Mahidol University. All procedures were carried out under the regulation and permission of the Faculty of Veterinary Science, Mahidol University IACUC protocol No. MUVS-KA-2021-01-01.

Blood samples (10 ml) were collected from pigs naturally infected with PEDV. Infected pigs were observed and diagnosed by a veterinarian based on signs and symptoms of PED. Pigs showing PED signs or had been diagnosed with PED within the past 4 weeks were used as subjects. Bloods were collected from two different pig strains: F1 sow (Danish Landrace, age 35–40 weeks old) and F3 fattening pigs (Danish Landrace × Large white × Danish Duroc, age 7–12 weeks old). Blood was collected from 13 female F1 pigs and 14 either male or female pigs raised in three different farms. Control sera were prepared from 10 uninfected F3 pigs (age 7 weeks old) raised in the animal research facility of Mahidol University. Bloods were left to clot at room temperature for approximately 2 h, followed by centrifugation at 1,500×g for 10 min in a refrigerated centrifuge to separate the serum. Serum was aliquoted and separated to new tubes and stored at −20°C until used.

PEDV used in this study is a lineage of PEDV CV777, designated ACC_PEDV (classification is shown in Figures 1B, D ). Vero cells were cultured in complete DMEM medium supplemented with 10% fetal bovine serum (FBS, HyClone) and 1% penicillin/streptomycin (Pen/Strep, Gibco), and then incubated at 37°C with 5% CO₂. The seed virus was sequentially propagated in Vero cell with PEDV medium [DMEM supplemented with 0.02% yeast extract (HiMedia), 10 μg/ml of trypsin (Sigma), and 0.3% tryptose phosphate broth] (tryptose, 20 g/L; dextrose, 2 g/L; sodium chloride, 5 g/L; disodium hydrogen phosphate, 2.5 g/L). To prepare the PEDV stock, Vero cells were cultured in 10 T-175 flasks. When the cell confluency was over 90%, the old medium was removed and the cells were infected with PEDV resuspended in 20 ml of PEDV medium. When the cytopathic effect (CPE) reached 70%, the medium was harvested and pooled. Sodium chloride (NaCl) was added to the medium to a final concentration of 0.5 M and polyethylene glycol (PEG, Sigma) was then added to a final concentration of 8%. After an overnight incubation, PEDV was harvested by centrifugation at 8,000 rpm for 20 min. The viral pellet was next resuspended in 5 ml of endotoxin-free PBS, aliquoted in a small volume to new tubes, and stored at −80°C. Virus titer was determined using 50% tissue culture infectious dose (TCID₅₀) method. Virus suspension was diluted in 10-fold serial dilutions and pipetted into 10 wells of confluent Vero cells. After 3 days post-infection, PEDV-infected cells were observed and counted using immunofluorescence staining assay. TCID₅₀ was next calculated.

Three days post-infection, Vero cells in 96-well plate were fixed with 1% formaldehyde at RT for 15 min and then washed with PBS once. Cells were permeabilized with cold 90% methanol and then incubated at 4°C for 5 min. Cells were washed once with PBS and blocked with 2% FBS/PBS for 1 h at RT. The blocking solution in each well was replaced with 50 µl of mouse anti-PEDV (Median Diagnostic) diluted 1:1,000 in PBS. Following an incubation for 2 h at RT, the cells were washed twice with PBS. Secondary antibody, donkey anti-mouse IgG conjugated with Alexa Fluor^® 594 (Abcam) diluted 1:1,000 in PBS, was added to the cell and incubated for 1 h at RT. The cells were washed twice with PBS and then the nucleus was stained with DAPI (Sigma) diluted 1:1,000 in PBS and incubated at RT for 30 min, followed by washing once with PBS. Fluorescence images were directly examined under an inverted fluorescence microscope (Olympus DP74).

All pig serum samples were 2-fold serially diluted with DMEM containing 1% Pen/Strep to the dilution ranging from 1:32 to 1:4,096. PEDV at the concentration of 200 TCID₅₀/ml was prepared using PEDV medium. PEDV (10 TCID₅₀ in 50 μl) was mixed with 50 μl of diluted serum or medium (control) and incubated at 37°C with 5% CO₂. After, a 1-h incubation, the mixture of 100 μl was transferred into 96-well plates containing approximately 90% confluent monolayer of Vero cells. Each condition was performed in duplicate. After a 3-h incubation, the medium was removed and replaced with new PEDV medium. At day 3 post-infection, immunofluorescence staining assay was performed and the number of infected cells was counted. Endpoint neutralizing antibody titer was determined based on the serum dilution that completely inhibited virus infection. As all of the sera from the control pig did not neutralize PEDV at dilution 1:32 and higher, we thus set the cutoff of neutralizing antibody at neutralization titer 32 (reciprocal dilution titer). Only sera with neutralization titer ≥32 were subject to ELISA and neutralization and inhibition assay.

Synthetic peptides (0.75 nmol in 50 μl PBS/per well) were added into 96-well ELISA microplates (Greiner Bio-One). After an overnight incubation at 4°C, plates were washed three times using PBS containing 0.05% Tween 20 (PBST). Plates were blocked using 100 μl of PBST containing 5% FBS and incubated with agitation at room temperature (RT) for 1 h. Pig sera diluted 1:50 in PBST containing 1% FBS were added to the plates. After a 2-h incubation at RT, the plates were washed three times with PBST, and goat anti-pig IgG HRP antibody (Abcam) diluted 1:10,000 in PBST containing 1% FBS was added to the well (60 μl/well), followed by an incubation at RT for 90 min. After a three-time wash with PBST, the TMB substrate (BioLegend) was added (70 μl/well) and the plate was incubated in the dark at RT for 30 min, allowing the color to develop. The reaction was stopped by adding 35 μl of 1 M sulfuric acid (H₂SO₄), and the optical density was measured at a wavelength of 450 (OD450) (Multiskan FC, Thermo Scientific).

Neutralization–inhibition assay was conducted as previously described by Shi et al. (42) with some modifications. Sera from three F1 pigs (F1 1-3, F-1 1-4, and F1 1-5) and one F3 pig (F3 1-10) were subject to the assay. Serum was 2-fold serially diluted with DMEM containing 1% antibiotic to the dilution ranging from 1:32 to 1:256. PEDV was diluted in PEDV medium to 2,000 TCID₅₀/ml. Synthetic peptides were mixed with the diluted serum to a final concentration of 200 nmol/ml in a volume of 50 μl. After an incubation at 37°C for 1 h, PEDV (100 TCID₅₀ in 50 μl) was added to the peptide–serum mixture. After an incubation at 37°C for 1 h, the mixture (100 μl) was transferred to the Vero cells grown in 96-well plates. Following an incubation at 37°C with 5% CO₂.for 3 h, the virus suspension was removed from the well and a new medium was added into the wells. In addition, the conditions that the Vero cells were incubated with i) medium alone, ii) serum alone, iii) PEDV alone, and iv) the mixture of PEDV and serum were also included in the assay as control conditions. All conditions were performed in duplicate. Following a 3-day (F1 1-4 and F1 1-5) or 5-day (F1 1-3 and F3 1-10) incubation, PEDV infection in Vero cell was investigated using immunofluorescence staining assay. Neutralization inhibition mediated by a peptide was determined at the endpoint neutralizing antibody titer.

Putative neutralizing B-cell epitopes were aligned and compared with full-length S protein of PEDV USA/Colorado/2013 strain for the identification of epitope position. To localize each epitope within the trimeric PEDV S protein, the prefusion structure of the PEDV spike named 6VV5 (35) was used as a model. All predicted epitopes were depicted using PyMOL 2.3.4 program.

The statistical significance of the samples in different groups was analyzed using SPSS 22 for Windows software (SPSS, USA). Data analyses were performed using non-parametric Mann–Whitney U test. The difference between the two groups was determined based on p <0.05. Immunodominant and subdominant epitopes were defined with the following criteria. In response to a particular peptide, if all sera from both F1 and F3 pigs exhibited OD450 value higher than OD450 mean of the control group + 2SD, such peptide was considered an immunodominant. On the other hand, if all sera from both F1 and F3 pigs exhibited OD450 value higher than OD450 mean of the control group + 1SD, such peptide was considered a subdominant epitope.