Bioinformatic analysis of epitope-based vaccine design against the novel SARS-COV-2

Background: An outbreak of infection caused by SARS-COV-2 recently has brought great challenge to public health. Rapid identification of immune epitopes would be an efficient way to screen the candidates for vaccine development at the time of an impending pandemic. This study aimed to predict the protective epitopes with bioinformatic methods and resources for vaccine development. Methods : ABCpred and BepiPred servers were utilized for sequential B-cell epitope analysis. Discontinuous B-cell epitopes were predicted via DISCTOPE server. IEDB server was utilized for HLA-1 and HLA-2 binding peptides computation. Surface accessibility, antigenicity, and other important features of forcasted epitopes were characterized for immunogen potential evaluation. Results : A total of 63 sequential B-cell epitopes on spike protein were predicted and 3 peptides (333-338, 648-663, 1064-1079) exhibited high antigenicity score and good surface accessibility in our modeling. 8 residues within spike protein (Gly 184 , Gly 502 , Lys 558 , Pro 561 , Pro 809 , Ser 810 , Lys 811 , Pro 812 ) are forcasted as components of discontinuous B-cell epitopes. The bioinformatic analysis of HLA binding peptides within nucleocapsid protein produced 81 and 64 peptides being able to bind class-I and class-II molecule respectively. The peptide (104-112) has a high immunogenicity score and was predicted to bind a wide spectrum of both HLA-1 and HLA-2 molecules. Conclusions : Linear B-cell epitopes (333-338, 648-663, 1064-1079) and discontinuous B-cell epitopes (Gly 184 , Gly 502 , Lys 558 , Pro 561 , Pro 809 , Ser 810 , Lys 811 , Pro 812 ) on spike protein, T-cell epitope (104-112) within nucleocapsid protein were identified and recommended for developing vaccine against SARS-COV-2.

The pathogen causing the new type of pneumonia was named 2019-ncoV (recently with its name changed to SRAS-COV-2) by WHO(https://www.who.int/publications-detail/surveillance-casedefinitions-for-human-infection). Its genome sequence has been released and reported by Chinese scientists and submitted to GenBank database on 12 Jannuary 2020 [2]. Like SARS-COV and MERS-COV, the other two virus that caused severe epidemic problem in recent years, SRAS-COV-2 also belongs to β-coronaviruses family. Bats are proved to be their natural host [3]. At present, insufficient knowledge on the latency and contagiosity of SARS-COV-2 increased the uncertainty of virus persistency. Specifc therapeutic agents targeting the virus are currently not available [4].Vaccine is still the most expecting and effective approach of prevention. Whereas selection and design of protective immunogens against pathogens rapidly is a major challenge on vaccine development, especially for the newly emerging pathogens [5]. Traditional methods of vaccine development could not meet the needs of the pressing situation in the event of an outbreak [6]. Bioinformatic analysis combined with Big Data of immunoinformatic resources makes the mission possible [7][8][9].
The purpose of this research is to predict B-cell eiptopes on spike protein and T-cell epitopes within nucleocapsid phosphoprotein of SARS-COV-2 by using bioinformatics methods and immunoinformatic resources. Epitopes information presented by this work may aid in developing promising vaccine against SARS-COV-2.

Data retrieval and sequence alignment
Protein sequences of SARS-COV-2 were retrieved from NCBI database. Alignment was performed on EMBL-EBI server with Clustal Omega method [10]. Conserved domains within predicted polypeptides of SARS-COV-2 were analyzed with CD-search in NCBI.

Structure modeling
Experimentally revealed 3D structure of SARS-COV Spike protein was retrieved from Protein Data Bank (PDB ID: 6ACD) [11]. Structure modeling of SARS-COV-2 Spike protein was performed with SWISS-MODEL Homology Modelling server (https://swissmodel.expasy.org/) using SARS-COV Spike protein as template.
Linear B-cell epitope prediction ABCpred[12] and BepiPred[13] were used for B-cell epitope forcast. Threshold of 0.85 was set for the prediction on ABCpred server and 16 residues lengthy epitopes were presented as adequate to induce defensive immune response. Predicted epitopes were highlighted as sphere in SARS-COV-2 spike protein viewed by pymol. We utilized vaxijen2.0 server to analyze the antigenicity of chosen epitopes [14]. Surface accessibility of predicted peptides were evaluated with the protein 3D structure built with SWISS-MODEL Homology Modelling server.
Discontinuous B-cell epitope prediction Prediction of discontinuous epitopes were conducted via DiscoTope 1.1 server [15]. Parameter was set at -3.1 which indicate 95% specificity and 16% sensitivity. This method is based on surface accessibility, residue statistics, and spatial information in a compiled data set of discontinuous epitopes discovered by X-ray crystallography of antigen/antibody complex structure. The contact number, propensity score, and disctope score for each amino acid are provided for conformationbased epitope prediction. Pymol was employed to illustrate the position of predicted epitopes on the 3D structure of SARS-COV-2 spike protein.

T-cell epitope prediction
We used the free online service provided by IEDB to forcast T-cell epitopes within nucleocapsid protein binding to  or HLA-2[17] molecule. A relatively small pool of HLA alleles covering majority of the population, over 97% and 99% for class-I and class-II respectively, were chosen in the analysis [18,19]. The sequences were given in plain format and top 50 rank percentile results were retained for further analysis.

Results
Protein coding features of SARS-COV-2 genome A map of the predicted ORFs is depicted in Fig. 1 based on the genome sequence of virus of Wuhan-Hu-1 isolate. The genomic structure of SARS-COV-2 shares characteristics that also found in other coronaviruses including SARS-COV, MERS-COV, and Human coronavirus NL63. All these coronaviruses contain recognizable ORFs including the replicase (orf1ab polyprotein), surface glycoprotein (spike protein), envelope protein, membrane glycoprotein, nucleocapsid phosphoprotein, and a number of non-structure proteins (nsp). The conserved domains of proteins encoded by SARS-COV-2 genome are summerized in Supplementary Table 1. Spike protein consists of 1,273 amino acids and the majority part exposed outside of the virion [22]. Thus, it is an ideal target for B-cell epitope screening.
Nucleocapsid protein is more conserved in coronaviruses ( Fig. 2B-D). Though unable to induce humoral immunity, nucleocapsid protein in SARS-COV and MERS-COV has been experimentally tested as a robust immunogen to induce cytotoxic T lymphocyte response [23,24], which suggest nucleocapsid protein in SARS-COV-2 could be a good candidate for T-cell epitope prediction.

Sequence analysis of spike protein and nucleophosphoprotein in selected coronaviruses
To better understand the characteristic of SARS-COV-2, we compared its protein sequences with other selected coronaviruses. The total sequence identity and guide tree result was presented in Fig. 2B and Fig. 2C respectively. Consist with recently published study [25], we found that both spike (S protein) and nucleocapside (N protein) in SARS-COV-2 are more closely related to that of SARS-COV.
The protein domains of spike and nucleocapsid protein (Fig. 2D) were depicted based on previous studies on SARS-COV [22,26] and the protein alignment result in the current study (supplementary files: N align.clustal_num; S align.clustal_num). The amino acid sequence identity result confirmed a highly similarity between SARS-COV-2 and SARS-COV. As anticipated, the nucleocapsid protein is more conserved among selected coronaviruses compared to spike protein.

B-cell epitopes recognition
Full length sequence of spike protein was scanned via ABCpred [12] and BepiPred-2.0 server[13] for putative sequential B-cell epitopes. A threshold value of 0.85 in ABCpred server was set when search for linear B-cell epitope with 16 amino acids in length. A total of 28 non-overlapping peptides were identified by ABCpred server. For sequential B-cell epitopes prediction on BepiPred-2.0 server, a threshold value of 0.5 was applied and 35 peptides were predicted. Antigenicity was calculated by Vaxijen 2.0 server and peptides with highest antigenicity score were selected (Table 1). 3D structure prediction of SARS-COV-2 spike protein was conducted as described in the methods. The new PDB file reflecting SARS-COV-2 spike protein structure could be download from supplementary file SPIK.pdb.
Predicted epitopes in Table 1 were highlighted as sphere in pymol (Supplementary Figure 1). While all peptides predicted were exposed on the surface of spike monomer, only seq#2'GCLIGAEHVNNSYECD', seq#5'HVTYVPAQEKNFTTAP', and seq#8'TNLCPF' displayed good surface accessibility in spike trimer ( Fig. 3 and supplementary file: B-cell-epitope-on-trimer.gif), the pattern more likely exist in nature. Among these three peptides, seq#8 'TNLCPF' predicted at position 333 possesses a random coil structure and presented highest antigenicity score. Conformation-based B-cell epitopes were computed on Discotope 1.1 server [15]. A threshold value of -3.1 was chosen for the computation, which corresponds to a specificity of 95% and a sensitivity of 16%. The contact number, propensity score, and disctope score for each amino acid which passed the threshold were presented in Table 2. The position of these residues were viewed with pymol and highlighted as sphere ( Fig. 4). Processing with a combination of B-cell epitope scanning and 3D structure modeling forcasted 3 potent linear epitopes and 8 residues invovled in discontinuous epitopes formation.

T-cell epitopes recognition
In our study, IEDB server was utilized following prediction methods recommended (a combination of ANN, SMM, CombLib, and NetMHCpan EL methods for HLA-1 binding prediction, and a combination of NN-align, SMM-align, CombLib, Sturniolo, and NetMHCIIpan methods for HLA-2 binding prediciton).
For HLA-1 binding peptide prediction, the top 50 rank percentile was retained for further analysis. A total of 81 nonrepetitive peptides with ANN_IC50 value not higher than 500, indicative of stronger than medium binding affinity, were identified. 6 peptides with highest antigenicity score by vaxijen 2.0 were chosen for next step processing. In this step, we screened all HLA-1 molecules being able to bind these peptides (Table 3). Similar strategy was applied for HLA-2 binding peptides prediction on IEDB server and 64 peptides were identified as HLA-2 binding sequences. SMM_IC50 value of 500 and 5000 were indicative of medium and weak binding affinity respectively. 6 peptides with highest antigenicity score were selected for HLA-2 molecule screening and the outcome was presented in Table 4. In the selected peptides pool for HLA binding, 'LSPRWYFYY' was predicted as both HLA class-I and class-II binding peptide. Additionally, this peptide may excel in capability of binding to a large number of HLA molecules as shown in Table 3&4. Partially overlapping region was found in sequence 'FPRGQGVPI' in Table 3 and sequence 'PRGQGVPIN' in Table 4, which suggests the overlapping region, like the peptide 'LSPRWYFYY' being able to bind both HLA-1 and HLA-2, may initiate both CD4+ and CD8+ dependent immune response.

Selected T-cell epitopes features profiling and evaluation
Peptide stability, mutation analysis, toxicity, allergenicity, hydro and physiochemical features were calculated as indicated in methods and the results were presented in Supplementary Table 2. While no peptide listed is toxic, a majority of them are potentially allergenic. To forcast the probability of an immune response induced by the predicted HLA-1 binding peptides, Class I Immunogenicity test was performed and the scores were presented in Table 5. A higher score indicates higher potential of immune response induction. Compared to traditional vaccine development, potent epitopes can be predicted via bioinformatic analysis, which makes the vaccine design straightforward and fast [27]. As the majority of spike protein is exposed outside the virion, it could be an ideal target to search for B-cell epitope. In fact, spike proteins in MERS-COV and SARS-COV have been shown to induce robust immune response [28,29]. In our analysis of SARS-COV-2 spike protein, 3 peptides (Fig. 3) identified from multi-step screening displayed excellent surface accessibility in the 3D structure modeling. The peptide 'TNLCPF' at position 333 produces a random coil structure and has high antigenicity score. Noticeably, this peptide sits in the receptor binding domain (RBD) of spike in SARS-COV-2, which has been proved to mediate the binding to ACE2 on the epithelial membrane of human lungs [30]. It is likely that antibody recognizing this epitope could also neutralize the virus and prevent infection. We noticed a recent published study of epitope prediction [31]. Though they utilized the same approach in B-cell epitopes prediction, their results are different with the current study which could be due to different parameter setting and strategy. In addition, surface accessibility of potent B-cell epitopes were further evaluated with structure modeling in the current study to exclude B-cell epitopes burried inside.

Discussion
Besides B-cell epitope prediction on spike protein, we selected nucleocapside protein as target protein for T-cell epitope computation for the following reasons: First, nucleocapsid protein in SARS-COV and MERS-COV has been experimentally tested as a robust immunogen to induce cytotoxic T lymphocyte response [23,24]; Second, nucleocapsid protein is the predominant protein expressed in the virion during early stage of infection[32]; Third, nucleocapsid protein was detected in the majority of SARS-COV infected patients as early as day 1 of infection [33,34], which suggests nucleocapsid protein-based vaccine may evoke T-cell dependent immune response timely. 81 and 64 peptides within nucleocapsid protein were predicted to bind HLA-1 and HLA-2 molecules respectively. We selected 6 peptides (Table 3&4) with the highest antigenicity score as potent T-cell epitopes. The peptide 'LSPRWYFYY' at position 104 showed an immunogenicity score of 0.36, suggesting a relatively strong capability to initiate immune response. Interestingly, this peptide was forcasted to bind to both HLA-1 and HLA-2 molecules, suggesting a priority of this peptide in vaccine design. Though these peptides were predicted to be non-toxic, we noticed that a large number of these peptides could be allergens. Thus, special attention should be paid to potential allergic reaction during pre-clinical and clinical trial. As nucleocapsid protein is highly conserved between SARS-COV-2 and SARS-COV, available information on nucleocapsid protein-based vaccine against SARS-COV could be helpful.

Consent for publication
All presentation of the data have consent for publication.

Availability of data and materials
All analysed data of this study are included with the manuscript and its supplementary information files.

Competing interests
The authors delcare no conflict of interests

Funding
This work was funded by the National Science Fundation of China [No.81371834, No. 81171597, No. 81828004, and No. 81970746], and Hunan Provincial Natural Sciences Foundation [2018JJ3740]. China, 2019. N Engl J Med 2020:727-733.    Table 3 HLA class-I alleles binding epitopes predicted by IEDB server    Figure 1 Map of the predicted ORFs in the SARS-COV-2 genome sequence. Site of B cells discontinuous epitopes predicted through DISCOTOPE 1.1 server on the predicted structure of SARS-CoV2 spike protein highlighted with cartoon representation.

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.