Introduction
Norovirus (NoV), a member of the
Caliciviridae family, is the leading cause of viral acute gastroenteritis (AGE) worldwide [
1]. It is associated with approximately 685 million AGE cases [
2] and over 200,000 deaths annually [
3]. NoV-induced AGE imposes a significant economic and clinical burden, with an estimated global loss of $60 billion in healthcare and societal costs [
4].
The NoV virion comprises 3 major proteins: the major capsid protein (VP1), the minor capsid protein (VP2), and a nonstructural polyprotein (NSP), which includes nonstructural proteins such as p48, NTPase, p22, VPg, 3CLpro, and RNA-dependent RNA polymerase (RdRp) [
5,
6]. NoVs exhibit extensive genetic diversity and are currently classified into 10 genogroups (GI–GX) based on the VP1 sequence [
7]. Strains from each genogroup infect specific host species [
5,
7,
8]. Notably, GI, GII, GIV, GVIII, and GIX are associated with human infections [
5], with GII being the most prevalent, accounting for approximately 90% of all reported cases and outbreaks [
5,
8].
To date, no specific therapeutics for NoV-induced AGE exist. Current treatments primarily focus on hydration and supportive care [
5]. Severe cases may require hospitalization for intravenous fluid administration [
9]. However, in low-income countries with limited access to rehydration therapies, preventing NoV-AGE is even more critical [
5]. Preventing NoV infection largely depends on proper hygiene practices, such as thorough hand washing, correct food handling, and minimizing contact with infected individuals [
10]. Although improved hygiene and sanitation are crucial, they may not fully control NoV transmission, particularly in high-density or communal environments [
11]. Vaccination remains the most effective measure to prevent viral spread [
12]. By priming the immune system to recognize and neutralize pathogens, vaccination directly prevents infection and offers protection regardless of environmental conditions [
13]. Currently, no licensed NoV vaccine is available, largely due to the virus's extensive genetic diversity, limited understanding of immune responses, and the lack of an effective cell culture system for viral propagation, which renders conventional live attenuated or inactivated vaccine approaches unfeasible [
14,
15]. Consequently, non-replicating recombinant protein-based vaccines have emerged as a preferred alternative [
15].
Multi-epitope subunit vaccines are recombinant protein-based formulations that incorporate immunogenic regions, or epitopes, derived from 1 or multiple viral proteins into a single construct. This design preserves immunogenicity because only the specific epitopes interact directly with immune cells, including B and T lymphocytes, to stimulate antibody production and activate cell-mediated immune responses [
16]. By eliminating non-essential (non-immunogenic) or potentially immunosuppressive regions, the vaccine minimizes unwanted reactions and optimizes immune targeting [
17].
Traditional development of multi-epitope subunit vaccines has relied on
in vivo and
in vitro methods [
18]. Initially, target proteins are isolated from the pathogen and enzymatically or chemically fragmented to produce smaller peptide segments. These segments are then tested for immunogenicity through animal inoculations or immunoassays assessing antibody production, cytotoxicity, T lymphocyte proliferation, and cytokine release [
18]. However, the advent of immunoinformatics has revolutionized vaccine development by streamlining epitope identification. Modern prediction tools can accurately identify epitopes from viral proteins that are capable of activating B and T lymphocytes based on their sequence and structural properties [
17]. The computational tools employed in this study leverage established correlations between these properties and biological outcomes, as demonstrated in several studies [
19–
23]. This approach enables the prediction of biological functions without extensive wet lab experiments, thereby accelerating vaccine development. Immunoinformatics has already been successfully applied to vaccine design for various pathogens [
24,
25].
This study aimed to design and evaluate a multi-epitope subunit vaccine against NoV using immunoinformatics. The approach involved (1) mapping epitopes across all proteins of human-infecting NoV genogroups, (2) designing the vaccine using the predicted epitopes, and (3) conducting in silico evaluations through a series of sequence- and structure-based tests. This work is vital for advancing NoV vaccine candidates, reducing NoV-AGE-related morbidity and mortality, and achieving a significant breakthrough in global public health.
Materials and Methods
The design and evaluation of the NoV vaccine construct were performed entirely using
in silico approaches.
Figure 1 presents a summary of the methodology employed in this study.
Identification of Conserved Regions
The full-length reference sequences for VP1, VP2, and NSP proteins of human-infecting NoV genogroups (GI, GII, GIV, GVIII, and GIX) were retrieved from the National Center for Biotechnology Information (NCBI) database [
26] (
https://www.ncbi.nlm.nih.gov/) in November 2023. Sequences for each protein were aligned within their respective genogroups using the Multiple Sequence Alignment (MSA) tool in Clustal Omega [
27] (
https://www.ebi.ac.uk/jdispatcher/msa/clustalo). The resulting alignment files were then uploaded to the Protein Variability Server (PVS,
http://imed.med.ucm.es/PVS/) to extract conserved fragments. A Shannon Diversity Index (H) threshold of 1.0 was applied to identify variable residues. This index is calculated as
where p
i denotes the proportion of residues of amino type i, and M refers to the total number of amino acid types present at a specific site. The H value ranges from 0 (indicating a single amino acid type at that position) to 4.322 (indicating equal representation of all 20 amino acids at that position), with an H value=1.0 considered low variability or conservation [
29]. Conserved fragments with ≥9-mer, ≥15-mer, and ≥16-mer lengths were collected.
Epitope Mapping
The 9-mer fragments were uploaded to NetMHCCons-1.1 (
https://services.healthtech.dtu.dk/services/NetMHCcons-1.1/) to identify potential cytotoxic T lymphocyte (CTL) epitopes. NetMHCCons-1.1 predicts CTL epitopes by assessing their binding affinity (BA) to known major histocompatibility complex (MHC) class I molecules using 3 methods: NetMHC (an artificial neural network-based allele-specific method), NetMHCpan (a pan-specific ANN method trained on extensive binding data), and PickPocket (a matrix-based method analyzing receptor-pocket similarities) [
30]. The analysis used 9-mer peptide lengths and reference human leukocyte antigen (HLA) class I molecules [
31], with a screening threshold set at a percentile rank (PR) of ≤0.5 or a half-maximal inhibitory constant (IC50) of ≤50 nM. Selected CTL epitopes were further evaluated using NetCTLpan 1.1 [
32] (
https://services.healthtech.dtu.dk/services/NetCTLpan-1.1) with the same HLA parameters and a PR threshold of ≤1.0. NetCTLpan 1.1 integrates predictions for MHC class I binding (via an ANN method similar to NetMHCpan), proteasomal C-terminal cleavage (using NetChop neural networks) [
33], and TAP transport efficiency (using a weight matrix-based method described by Peters et al. [
34]). The Class I Immunogenicity tool (
http://tools.iedb.org/immunogenicity) was then used with a score threshold of >0 to predict immunogenicity based on amino acid enrichment, positional importance, and MHC binding and processing affinities [
35]. Finally, the CTL epitopes were subjected to additional filtering using a score threshold of ≥0.4 in VaxiJen 2.0 (
https://www.ddg-pharmfac.net/vaxijen/VaxiJen/VaxiJen.html), confirmation as non-allergenic by AllerTOP v2.1 (
http://www.ddg-pharmfac.net/AllerTOP), non-toxicity by ToxinPred (
https://webs.iiitd.edu.in/raghava/toxinpred/multi_submit.php), and favorable water solubility in Innovagen’s PepCalc (
https://pepcalc.com/). VaxiJen employs an alignment-free method based on auto-cross covariance (ACC) to transform protein sequences into uniform vectors, with descriptors for hydrophobicity, molecular size, and polarity [
19]. Similarly, AllerTOP uses ACC to represent sequences with descriptors for hydrophobicity, molecular size, helix-forming propensity, abundance, and β-strand forming propensity [
20]. ToxinPred identifies toxic peptides using a combination of similarity-based, motif-based, and machine/deep learning techniques [
23]. Innovagen’s PepCalc estimates water solubility by evaluating molecular weight, hydrophobicity, and isoelectric point. Epitopes meeting these criteria were selected as the final CTL epitopes.
The 15-mer fragments were uploaded to NetMHCIIpan 4.3 (
https://services.healthtech.dtu.dk/services/NetMHCIIpan-4.3/) to identify potential helper T lymphocyte (HTL) epitopes. NetMHCIIpan 4.3 predicts HTL epitopes by assessing their binding to known MHC class II molecules using an ANN method trained on extensive BA and eluted ligand (EL) data [
36]. The analysis employed 15-mer peptide lengths and reference HLA class II molecules [
37] with a PR threshold of ≤1.0. Selected HTL epitopes were further evaluated for immunogenicity using the CD4 T cell immunogenicity prediction tool (
http://tools.iedb.org/CD4episcore) with a score threshold of ≥40. This tool combines neural networks trained on experimentally validated epitopes and negative peptides to assess amino acid enrichment and positional importance, identifying peptides with the potential to bind T cell receptors and induce a T cell response [
38]. The HTL epitopes were also filtered using VaxiJen 2.0, AllerTOP v2.1, ToxinPred, and Innovagen’s PepCalc as described above. Furthermore, only epitopes predicted to induce at least 1 of the following cytokines were selected: tumor necrosis factor (TNF)-α (using TNFepitope at
https://webs.iiitd.edu.in/raghava/tnfepitope), interferon (IFN)-γ (using IFNepitope at
https://webs.iiitd.edu.in/raghava/ifnepitope), interleukin (IL)-4 (using IL4Pred at
http://crdd.osdd.net/raghava/il4pred/), and IL-10 (using IL10Pred at
https://webs.iiitd.edu.in/raghava/il10pred/). TNFepitope employs alignment-free machine learning models based on composition-based features and alignment-based methods using BLAST to compare peptide sequences [
39]. These methods, trained on datasets of TNF-α-inducing and non-inducing epitopes, score peptides based on their ability to activate TNF-α production. Similarly, IFNepitope, IL4Pred, and IL10Pred use a hybrid approach combining motif-based prediction with Support Vector Machine (SVM) classification [
40–
42] to distinguish cytokine-inducing peptides.
The 16-mer fragments were uploaded to BepiPred 3.0 (
https://services.healthtech.dtu.dk/services/BepiPred-3.0) to predict potential B lymphocyte epitopes. BepiPred 3.0 identifies B lymphocyte epitopes by analyzing protein sequences using ESM-2 protein language model representations and a neural network that accounts for hydrophilicity, flexibility, surface accessibility, and amino acid composition [
43]. The predicted epitopes were cross-validated using SVMTrip (
http://sysbio.unl.edu/SVMTriP/prediction.php) and the ABCPred server (
https://webs.iiitd.edu.in/raghava/abcpred); SVMTrip utilizes SVM classifiers [
44] and ABCPred is based on an ANN [
45]. A threshold of 0.85 was used for ABCPred while SVMTrip was run with default parameters. Epitopes validated by either tool were then submitted to the LBtope server (
https://webs.iiitd.edu.in/raghava/lbtope) to assess prediction confidence. LBtope distinguishes potential B lymphocyte epitopes using datasets of verified epitopes and non-epitopes from the Immune Epitope Database [
46]. Only epitopes with a probability of ≥60% were considered for further analysis. Finally, the linear B lymphocyte (LBL) epitopes were further filtered using VaxiJen 2.0, AllerTOP v2.1, ToxinPred, and Innovagen’s PepCalc as previously described, and those meeting all criteria were selected as the final LBL epitopes.
Vaccine Designing
The AAY linker was employed to connect CTL epitopes. This linker facilitates cleavage and selective binding of the attached epitopes to chaperones or TAP transporters, thereby increasing the number of epitopes available for MHC binding and presentation [
47]. The KK linker was used to join LBL epitopes; pairs of basic residues such as KK serve as cleavage sites for cathepsin B, aiding in the release of LBL epitopes for efficient presentation [
48]. The GPGPG linker connected the HTL epitopes, with regions rich in glycine and proline being associated with β-turns and hypothesized to enhance secondary structure formation and antigen processing [
49]. The HEYGAEALERAG linker was used to connect the 3 epitope groups; it contains 5 specific proteasome recognition sites that promote efficient cleavage during antigen processing [
50]. The combined epitopes were then joined with 3 distinct adjuvants—truncated human β-defensin 2 (thβd2), truncated human β-defensin 3 (thβd3), and 50S ribosomal protein L7/L12 (50SrpL7/L12)—using the EAAAK linker. Both β-defensins and 50SrpL7/L12 are recognized adjuvants [
51–
53] and are frequently employed in multi-epitope vaccine design studies [
54–
61]. The EAAAK linker acts as a structural spacer, controlling the distance between the epitope and adjuvant regions and minimizing potential interference [
62].
Assessment of Physicochemical Properties
The physicochemical properties of the adjuvanted vaccine constructs were evaluated using VaxiJen 2.0, AllerTOP v2.1, and Innovagen’s PepCalc under the conditions described above. Expasy ProtParam [
22] (
https://web.expasy.org/protparam) was used to estimate construct stability and the Grand Average of Hydropathy (GRAVY). ProtParam calculates protein stability using the instability index—which predicts whether a protein is stable
in vitro based on its dipeptide composition as described by Guruprasad et al. [
63]—and the GRAVY score, which indicates overall hydrophobicity or hydrophilicity, computed as:
where H
i is the hydropathy value of each amino acid (based on Kyte and Doolittle’s hydropathy index [
64]), and N is the number of residues in the sequence.
Immune Response Simulation
Host immune response profiles were predicted using C-Immsim (
https://kraken.iac.rm.cnr.it/C-IMMSIM). Simulations involved 3 injections of 1,000 particles administered at intervals of 1, 84, and 168 time steps, with the response analyzed over 300 time steps. C-ImmSim integrates several established immunological theories—including clonal selection, clonal deletion, hypermutation, T lymphocyte senescence, anergy, antigen-dose tolerance in B lymphocytes, and the danger theory [
65,
66]—to simulate immune responses. Immune parameters such as antibody titers, IFN-γ levels, and the populations of B lymphocytes, CTLs, and HTLs were compared across constructs, and the vaccine candidate demonstrating the most favorable combination of physicochemical properties and immune simulation outcomes was selected as the final NoV vaccine candidate.
Tertiary Structure Prediction
The tertiary structure of the final vaccine construct was predicted using AlphaFold v2 (
https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb). AlphaFold v2 employs a deep learning model that uses MSAs, co-evolutionary data, and a neural network to generate highly accurate 3-dimensional structures from the amino acid sequence, iteratively refining predictions through an attention-based mechanism [
67]. The predicted model was submitted to UCSF Chimera 1.17.1 for energy minimization using 10,000 steps of steepest descent; Chimera 1.17.1 applies algorithms such as steepest descent or conjugate gradient to achieve a more stable, lower-energy conformation [
68]. Further refinement was performed with the GalaxyWEB Refine service (
http://galaxy.seoklab.org/refine), which optimizes protein structures by rebuilding side chains, resolving steric clashes, and applying both mild and aggressive relaxation via short molecular dynamics simulations [
69]. Structural quality was evaluated using ERRAT and Procheck in SAVES v6.0 (
https://saves.mbi.ucla.edu); ERRAT assesses the distribution of non-bonded atomic interactions [
70], while Procheck examines stereochemical parameters such as bond angles, torsion angles, and backbone conformation [
71]. The highest-scoring model was selected as the representative structure of the vaccine construct. Additionally, Protein Structure Analysis (ProSA)-web (
https://prosa.services.came.sbg.ac.at/prosa.php) was used to further assess structural quality by comparing the energy profile of the protein structure to a database of experimentally determined structures, providing a Z-score where lower values indicate a more stable conformation [
72]. Finally, ChimeraX 1.9 [
73] was employed to visualize the final vaccine construct.
Discontinuous B lymphocyte Epitope Mapping and Disulfide Engineering
Discontinuous B lymphocyte epitopes within the vaccine construct were predicted using the DiscoTope-3.0 webserver (
https://services.healthtech.dtu.dk/services/DiscoTope-3.0/) with default parameters. DiscoTope-3.0 uses inverse folding structure representations and positive-unlabeled learning to predict B lymphocyte epitopes [
74].
Additionally, Disulfide by Design 2.0 (DbD2) (
http://cptweb.cpt.wayne.edu/DbD2/) was used to identify residue pairs within the vaccine construct that could be mutated to cysteine to enhance structural stability. DbD2 incorporates B-factor values to assess regional flexibility and ranks potential disulfide bonds based on mobility [
75]. The analysis was performed with default parameters, applying a bond energy threshold of ≤2.0 kcal/mol. The instability index of the mutated construct was then determined using Expasy ProtParam and compared to that of the original construct.
Molecular Docking and Dynamics Simulation
The crystal structure of human toll-like receptor 4 (TLR4) was downloaded from the RCSB Protein Data Bank [
76] (
https://www.rcsb.org) (PDB ID: 3FXI). Peptides, water molecules, and extraneous atoms were removed from the structure using PyMOL v2 [
77], and missing residues were modeled using Modeller [
78] (
https://salilab.org/modeller). The final vaccine construct was forcibly docked to the TLR4 structure using ClusPro 2.0 (
https://cluspro.bu.edu/login.php), which performs (1) rigid body docking, (2) root-mean-square deviation (RMSD)-based clustering of the 1,000 lowest-energy structures to identify the largest clusters, and (3) refinement via energy minimization of selected structures [
79]. Forced docking was achieved by applying an attraction parameter to constrain binding at the agonist-binding site of TLR4. Binding residues were identified from the LPS-bound TLR4 crystal structure (PDB ID: 3FXI) using ChimeraX’s Contacts function, selecting residues with a van der Waals overlap of –0.40 or lower as attraction residues to ensure biologically relevant binding interactions.
The BA and per-residue energy decomposition of the complex were determined using HawkDock 2.0 (
http://huanglab.phys.hust.edu.cn/hpepdock). This server calculates BA using the Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) method, which estimates the free energy difference between bound and unbound states [
80]. It computes per-residue energy decomposition by breaking down the total binding energy into contributions from individual residues, identifying key stabilizing or destabilizing interactions [
80].
The stability of the complex was analyzed through molecular dynamicssimulations using the Southampton Interface and Reduction Algorithm for Hydrodynamics force field [
81] in GROMACS 2023.2 [
82] (
https://manual.gromacs.org/2023.2/download.html). The system was prepared by placing the complex in a cubic simulation box with periodic boundary conditions, ensuring a minimum distance of 1 Å between the complex and the box boundaries to avoid interactions between periodic images. The box was solvated with the TIP3P water model, and sodium and chloride ions were added to achieve a 0.15 M concentration. Energy minimization was performed with up to 50,000 steepest descent steps, followed by 2 equilibration phases: an NVT equilibration to gradually stabilize the temperature at 312 K using a modified Berendsen thermostat, and an NPT equilibration to stabilize the pressure at 1 bar using a Berendsen barostat. A 100 ns production run was then executed, generating 50 million time steps of 2 fs each, with snapshots recorded every 100 ps (totaling 1,000 frames). The trajectory’s stability was evaluated using RMSD analysis.
Codon Optimization and In Silico Cloning
The JAVA Codon Adaptation Tool (JCat) (
https://www.jcat.de/) was employed to generate an optimized codon sequence of the vaccine construct for efficient expression in an
Escherichia coli K-12 system. JCat evaluates codon adaptation based on the Codon Adaptation Index (CAI) and adjusts rare or less frequently used codons to improve translation efficiency [
83]. BamHI and XhoI restriction sites were incorporated at the N-terminal and C-terminal ends, respectively. The optimized sequence was subsequently inserted into the pET-28a(+) vector via restriction cloning using the SnapGene tool (
http://www.snapgene.com).
Results
Identification of Conserved Regions
Nineteen NSP, 15 VP1, and 18 VP2 reference proteins were retrieved from the NCBI database.
Table 1 summarizes the distribution of these proteins across the genogroups. Sequence alignment was performed for each genogroup, and the conserved fragments identified from these alignments are detailed in
Tables S1–
S3. These conserved fragments served as inputs for epitope prediction.
Epitope Mapping
Through epitope screening, 22 epitopes were identified from the VP1, VP2, and NSP proteins of the human-infecting NoV genogroups. Of these, 4 were classified as potential CTL inducers, 11 as potential HTL inducers, and 7 as potential B lymphocyte inducers (
Table 2). Most epitopes were derived from NSPs (73%), while the 2 capsid proteins contributed an equal number of predicted epitopes (13% each). All genogroups were represented, with the epitopes covering GI (9%), GII (9%), GIV (32%), GVIII (32%), and GIX (36%).
All predicted CTL epitopes exhibited high affinity for HLA class I molecules and strong CD8
⁺ immunogenicity. Similarly, all HTL epitopes showed high affinity for HLA class II molecules and robust CD4
⁺ immunogenicity. A summary of the physicochemical properties of the T lymphocyte epitopes is provided in
Table 3. Additionally, HTL epitopes induced at least 1 key cytokine associated with NoV infection, such as TNF-α, IFN-γ, IL-4, or IL-10 (
Table 4).
All 7 predicted B lymphocyte epitopes demonstrated high predictive accuracy across at least 3 prediction tools.
Table 5 summarizes the physicochemical properties of the B lymphocyte epitopes. Overall, the epitopes were predicted to be highly conserved, strongly antigenic, non-allergenic, non-toxic, and soluble.
Vaccine Designing
Three vaccine constructs were designed using the identified epitopes combined with 3 different adjuvants: thβd2, thβd3, and 50SrpL7/L12.
Figure 2 illustrates the general schematic configuration of the constructs, depicting the arrangement of epitopes, linkers, and the adjuvant.
Physicochemical Properties Assessment and Immune Response Simulation
All 3 constructs met the thresholds for antigenicity, non-allergenicity, solubility, and stability (
Table 6). The construct adjuvanted with 50SrpL7/L12 demonstrated the most favorable immune simulation profile (
Figure 3). As shown in
Figure 3, all constructs displayed relatively similar CTL populations, IFN-γ levels, and IL-10 levels. In contrast, the thβd3-adjuvanted construct exhibited the lowest antibody titers, HTL and B lymphocyte populations, and IL-2 levels, rendering it the least favored design. Although the thβd2 and 50SrpL7/L12-adjuvanted constructs showed generally similar immune responses, the latter exhibited relatively higher HTL and B lymphocyte populations following the second immunization, leading to its selection as the final vaccine construct. When compared with the immune simulation profiles of multi-epitope subunit vaccines previously designed by Shanthappa et al. [
84] and Ahmad et al. [
85], the 50SrpL7/L12-adjuvanted vaccine construct demonstrated remarkably higher antibody titers, HTL and B lymphocyte populations, and IL-2 concentrations. The final vaccine construct has an antigenicity score of 0.54, an instability index of 27.4 (classified as stable), and a GRAVY score of –0.457.
Tertiary Structure Prediction
Figure 4 displays the tertiary structural model of the 50SrpL7/L12-adjuvanted construct. This predicted model achieved high-quality scores in structural quality assessments (
Table S4;
Figure S1) and was selected for subsequent molecular docking analysis as the best representation of the construct.
Discontinuous B Lymphocyte Epitope Mapping and Disulfide Engineering
A total of 41 residues were identified as discontinuous epitopes using the DiscoTope-3.0 server (
Figure 5A), with these epitopes being uniformly distributed across the construct. Additionally, 4 residue pairs suitable for cysteine mutation were identified using DbD2 (
Figure 5B). These pairs were substituted with cysteine, and the instability index of the modified construct was analyzed. Unexpectedly, rather than decreasing, the instability index increased from 27.4 to 30.1, possibly due to unfavorable structural constraints or disruption of local interactions. Therefore, the original non-mutated construct was retained for further analysis.
Molecular Docking and Dynamics Simulation
The 50SrpL7/L12-adjuvanted construct exhibited strong affinity and stable interactions with the immune receptor TLR4 and its co-receptor, myeloid differentiation 2 (MD2), as demonstrated by molecular docking and molecular dynamics simulation analyses. The complex achieved a binding free energy of –145.46 kcal/mol, a value more negative than that observed for previously docked multi-epitope vaccine constructs [
86,
87]. Moreover, significant residues involved in canonical binding to lipopolysaccharide—such as R264, Y296, K341, and K362 from TLR4, and F121 from MD2—displayed notably high negative ∆G values (
Figure 6B). Additionally, residues not typically involved in canonical binding, including Y292 in TLR4 and Y58, L87, R90, and P151 from MD2, also interacted with the vaccine construct, exhibiting highly negative ∆G values.
During the molecular dynamics simulation, the complex demonstrated stable interactions, as evidenced by a plateau in RMSD around 40 ns before the simulation’s end (
Figure 6C). This stability was further confirmed through visual inspection of the simulation trajectory.
Codon Optimization and In Silico Cloning
The optimized codon sequence of the construct was generated using JCat for potential expression in the
E. coli K-12 strain. This sequence achieved a CAI value of 1.0 and a GC content of 51.8%.
Figure 7 presents a sample of the cloned construct within the pET28(a)+ vector, with the sequence inserted at the XhoI (158) and BamHI (198) restriction sites.
Discussion
Using an immunoinformatics approach, this study successfully designed a multi-epitope subunit vaccine construct against NoV. Three previously designed multi-epitope subunit vaccines against NoV have been reported by Ahmad et al. [
85], Azim et al. [
88], and Shanthappa et al. [
84]. Compared with these designs, the current study introduces critical advancements that significantly enhance the potential for broad immunological protection. Notably, this study included epitopes from all known human-infecting NoV genogroups, whereas Azim et al. [
88] focused solely on GI and GII and Shanthappa et al. [
84] restricted their design to the prevalent GII genogroup. The inclusion of GV by Ahmad et al. [
85]—a murine genogroup—is not relevant for human vaccine development because murine genogroup V does not infect humans [
89]. By targeting all human-infecting genogroups, the study ensures comprehensive coverage of both prevalent and less common genogroups, which is essential for mitigating outbreaks caused by emerging or underrepresented strains.
Another key feature of this study was sourcing epitopes from both capsid and nonstructural proteins. Although Azim et al. [
88] and Shanthappa et al. [
84] focused exclusively on capsid proteins, this study recognized the potential of nonstructural proteins to contribute to immune activation. Although nonstructural proteins are not initially accessible to receptors during early infection, their release following synthesis and cell lysis allows them to interact with lymphocyte receptors and trigger immune responses [
61]. This approach is consistent with previous multi-epitope vaccine designs [
61,
90–
92] that incorporated epitopes from nonstructural proteins.
This study further distinguished itself by prioritizing epitope conservancy. Unlike the studies by Ahmad et al. [
85], Azim et al. [
88], and Shanthappa et al. [
84], which relied on a single protein variant for epitope mapping, this study used conserved fragments obtained from multiple protein variants. This strategy ensures that the designed vaccine effectively targets diverse viral strains and offers a significant advantage in vaccine coverage. Several multi-epitope vaccine designs have also employed conserved regions for epitope mapping to potentially increase vaccine coverage [
93,
94].
The study aimed to activate all 3 arms of the immune response—CTL, HTL, and B lymphocyte responses—which each play a critical role during NoV infection [
89]. CTL responses were ensured by selecting epitopes that are effectively processed for proteasomal cleavage, transported by the endoplasmic reticulum, bind to MHC I molecules, and are recognized by CD8
⁺ receptors. These predictions were facilitated by NetMHCcons, NetCTLpan, and CD8
⁺ immunogenicity tools. HTL responses were secured by choosing epitopes that bind effectively to MHC II molecules and are recognized by CD4
⁺ receptors, as predicted by NetMHCIIpan and CD4
⁺ immunogenicity tools. In addition, these HTL epitopes induce at least 1 key cytokine associated with NoV infection, such as TNF-α, IFN-γ, IL-4, or IL-10. B lymphocyte responses were confirmed using multiple prediction methods based on amino acid properties and validated epitope databases, ensuring high accuracy. Furthermore, all predicted T and B lymphocyte epitopes met thresholds for antigenicity, non-allergenicity, and solubility. Antigenicity is essential for an epitope to be recognized as foreign, thereby triggering an immune response [
95]. Non-allergenicity and non-toxicity assessments are crucial to ensure host safety when the components are processed [
91], and sufficient solubility is necessary to prevent aggregation—a common bottleneck in vaccine development [
96]. All prediction tools used in this study confirmed that the epitope components were safe, immunogenic, and suitable for vaccine development.
The final vaccine construct incorporated a 50SrpL7/L12 adjuvant to activate innate immunity. This adjuvant is recognized for its role in dendritic cell (DC)-based tumor immunotherapy [
53]. It activates TLR signaling by binding to TLR4, which induces both the MyD88 and TRIF signaling pathways, leading to DC activation [
53]. Activated DCs, in turn, stimulate naive T lymphocytes, polarize both CD4
⁺ and CD8
⁺ T lymphocytes, secrete IFN-γ, and induce T lymphocyte-mediated cytotoxicity [
53]. Molecular docking results demonstrated a strong BA and stable interaction between the 50SrpL7/L12 component and TLR4. BA was estimated using the binding free energy (ΔG) calculated via MM/GBSA analysis, and the highly negative ΔG value indicates strong binding [
97].
In contrast to BA analysis, which estimates the strength of the complex based on a single binding pose, the molecular dynamics simulation provided insights into the stability of the complex over time under physiological conditions [
98]. The 50SrpL7/L12-adjuvanted vaccine construct docked to TLR4 exhibited a plateauing RMSD, signifying that the structure reached equilibrium during the simulation [
99]. Visualization of the simulation further confirmed that the vaccine construct remained bound to TLR4 throughout the simulation period. The combined results from molecular docking and molecular dynamics simulations consistently demonstrated that the 50SrpL7/L12-adjuvanted vaccine exhibited favorable binding characteristics with TLR4. In addition to effective TLR4 binding, the vaccine constructs also displayed favorable physicochemical properties, including high antigenicity, non-allergenicity, solubility, and stability. The presence of widely distributed discontinuous epitopes may further contribute to robust immune recognition. A sample cloned construct was designed for expression in the
E. coli K-12 strain; however, alternative expression systems may be considered based on production requirements.
The primary limitation of this study is the inherent accuracy constraint of the predictive models used by the servers. Despite selecting tools with high accuracy, these models can still produce inaccuracies, especially when applied to novel constructs. Future work should focus on experimentally validating these predictions through protein expression in various systems and assessing the construct’s stability, solubility, and immunogenicity through in vitro and in vivo studies. Additionally, exploring multi-epitope designs for other pathogens would broaden the application of this approach. As computational tools improve, refining prediction models and incorporating new data will further increase accuracy.
The computationally designed vaccine construct holds significant potential for clinical applications, provided it is validated through further experimental studies. The construct developed in this study could serve as a strong candidate for future vaccine development, pending successful preclinical and clinical evaluations.
