Computational screening and molecular modeling of probiotic-derived peptides targeting the conserved HR1 domain of SARS-CoV-2 spike protein

Abstract

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), the causative agent of the global COVID-19 pandemic, relies on its spike glycoprotein to mediate host cell entry via binding to human angiotensin-converting enzyme 2 (hACE2). The S2 subunit of the spike protein contains heptad repeat regions (HR1 and HR2) that form a six-helix bundle fusion core, a critical step for viral–host membrane fusion. This study aimed to computationally screen a library of 318 probiotic-derived bacteriocins to identify potential inhibitors targeting the conserved HR1 domain. Evaluation included historical variants (Alpha, Beta, Gamma, Kappa, Epsilon, Omicron) and currently circulating strains (BA.2.86, EG.5.1, HV.1, JN.1). Methodologies included: sequence conservation analysis via Clustal Omega, protein–peptide docking using HADDOCK 2.4 with Ambiguous Interaction Restraints derived from the HR1–HR2 complex (PDB: 6LXT), safety profiling with AlgPred, ToxinPred, HemoPred, TargetAntiAngio, and IL-4Pred, and molecular dynamics simulations (GROMACS 2020) assessing RMSD, RMSF, and hydrogen bonds. Results identified plantaricin K (2KEG) as a high-affinity binder to HR1 across all tested variants, with favorable in silico safety profiles (non-toxic, non-allergenic, non-hemolytic). Simulation data confirmed stable binding of 2KEG to HR1, supporting its role as a fusion inhibitor. These findings suggest the potential of probiotic-derived bacteriocins, particularly 2KEG, as broad-spectrum antiviral candidates against SARS-CoV-2. Further experimental studies are required to validate these computational predictions.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-29197-7.

Introduction

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of the COVID-19 pandemic, has led to unprecedented global health challenges since its emergence in late 2019. As of 31 August 2025, the World Health Organization (WHO) has documented over 778 million confirmed cases and approximately 7 million deaths worldwide, highlighting the critical need for effective therapeutic interventions (https://covid19.who.int/). Coronaviruses demonstrate broad tissue tropism, affecting multiple organ systems including the respiratory and gastrointestinal tracts, neurological tissues, hepatic cells, and integumentary structures¹. Among the seven known human coronaviruses identified to date, four typically cause mild upper respiratory tract infections, while the remaining three—SARS-CoV, MERS-CoV, and SARS-CoV-2—are associated with severe respiratory pathologies. Taxonomically, the Coronaviridae family comprises four genera—Alpha-, Beta-, Gamma-, and Deltacoronavirus—classified according to serological characteristics and genomic sequence homology². SARS-CoV-2 belongs to the Betacoronavirus genus, which also includes the previously emerged pathogenic members SARS-CoV and MERS-CoV that caused significant outbreaks in 2002–2003 and 2012, respectively³. SARS-CoV-2 is an enveloped, positive-sense RNA virus with a genome of approximately 29.9 kilobases, encoding both structural and non-structural proteins essential for viral replication and pathogenesis⁴. Among these, the spike (S) glycoprotein plays an indispensable role in viral entry by mediating attachment to the human angiotensin-converting enzyme 2 (hACE2) receptor through its receptor-binding domain (RBD) within the S1 subunit^5,6. Following receptor engagement, host proteases cleave the spike protein at the S1/S2 junction, inducing substantial conformational rearrangements that expose the fusion peptide embedded within the S2 subunit. The S2 subunit contains heptad repeat regions 1 and 2 (HR1 and HR2), which assemble into a stable six-helix bundle fusion core—a critical structure that facilitates viral-host membrane fusion and subsequent delivery of the viral genome into the host cell⁷. While the RBD exhibits remarkable variability and is prone to mutations that drive the emergence of immune-evasive variants such as Omicron sublineages EG.5.1, HV.1, BA.2.86, and JN.1, the HR1 and HR2 domains demonstrate significant evolutionary conservation across SARS-CoV-2 variants and among diverse betacoronaviruses^8–10. This structural conservation, combined with their essential mechanistic role in membrane fusion, positions HR1 as a particularly attractive target for the development of broad-spectrum antiviral therapeutics. Current therapeutic strategies have predominantly focused on targeting the RBD, yet the rapid evolutionary pace of SARS-CoV-2 underscores the necessity for alternative approaches aimed at more invariant viral components. Previous investigations have established the efficacy of synthetic HR2-derived peptides in inhibiting viral fusion; however, the exploration of novel, naturally occurring fusion inhibitors remains comparatively limited¹¹.

In this study, we turned to antimicrobial peptides (AMPs) biosynthesized by probiotic bacteria, specifically bacteriocins as a promising source of potential fusion inhibitors. Bacteriocins are ribosomally synthesized peptides characterized by narrow-spectrum antimicrobial activity, high physiological stability, and remarkable specificity^12,13. We aimed to computationally screen an extensive library of 318 probiotic bacteriocins to identify candidates capable of high-affinity binding to the conserved HR1 domain and subsequent disruption of six-helix bundle formation.

Employing an integrated computational approach encompassing molecular docking, molecular dynamics simulations, and comprehensive in silico safety profiling, we systematically evaluated the potential of probiotic-derived bacteriocins to inhibit SARS-CoV-2 fusion. Our screening identified several promising candidates capable of high-affinity binding to the conserved HR1 domain, with one peptide exhibiting particularly favorable interaction profiles and safety characteristics. This work aims to establish a robust computational framework for identifying natural fusion inhibitors and highlights the potential of probiotic-derived compounds as a novel class of antiviral agents against evolving viral pathogens.

The exploration of probiotic-derived bacteriocins as therapeutic candidates offers several distinct advantages beyond their antimicrobial properties. These molecules typically demonstrate enhanced biocompatibility, reduced immunogenicity, and potential for oral bioavailability compared to synthetic peptides. Additionally, their production through microbial fermentation processes provides significant advantages in terms of scalability and cost-effectiveness for large-scale therapeutic development. The ongoing emergence of SARS-CoV-2 variants featuring mutations in the RBD that confer resistance to existing therapies further underscores the critical need for developing alternative strategies targeting conserved viral regions such as HR1.

This study establishes the conceptual and mechanistic basis for our computational exploration of probiotic bacteriocins as novel fusion inhibitors against SARS-CoV-2. In the following sections, we present a detailed methodological framework encompassing: (1) the systematic compilation and curation of a bacteriocin library; (2) rigorous molecular docking protocols to evaluate binding interactions with the HR1 domain; (3) molecular dynamics simulations to assess complex stability; and (4) comprehensive in silico safety profiling. Through this multi-stage computational pipeline, we seek to identify promising bacteriocin-derived candidates with the potential to inhibit viral fusion machinery. The ultimate goal of this work is to provide a robust preclinical foundation for the development of broad-spectrum antiviral agents effective against diverse SARS-CoV-2 variants.

Materials and methods

AMPs and peptide extraction

Our initial peptide library comprised 500 peptides sourced from four primary databases: StraPep (http://isyslab.info/StraPep/)¹⁴, Bactibase (bactibase.hammamilab.org/main.php)¹⁵, Bagel (BAGEL4 (molgenrug.nl)¹⁶, and the Protein Data Bank (PDB) (https://www.rcsb.org/)¹⁷. Among these 500 peptides, 318 of them with proper structure shaped our final peptide library.

To ensure docking compatibility and computational feasibility, all peptides underwent rigorous evaluation using UCSF Chimera v.11 (https://www.cgl.ucsf.edu/chimera/) based on the following criteria:

• Structural integrity: Only peptides with complete backbone atoms and side chains were retained.

• Amino acid composition: Peptides containing non-canonical or heavily modified amino acids (e.g., lanthionine, dehydroalanine, D-amino acids) were excluded to ensure compatibility with standard force fields.

• Length limitation: Peptides exceeding 60 amino acids were omitted due to reduced druggability and increased computational complexity.

• Resolution threshold: For PDB-sourced structures, only those with a resolution of ≤ 3.0 Å were included.

Applications of these criteria resulted in the exclusion of 182 peptides, yielding a final curated library of 318 peptides suitable for docking studies.

Identification of HR1 region and conservation analysis

Multiple sequence alignment of the HR1 domain (residues 912–984) in the SARS-CoV-2 Spike protein was performed using Clustal Omega (version 1.2.4) (https://www.ebi.ac.uk/jdispatcher/msa/clustalo)¹⁸ across six betacoronaviruses known to infect humans: HCoV-299E (Alpha-CoV), HCoV-OC43 (Beta-CoV), HCoV-KU1 (Beta-CoV), MERS-CoV (Beta-CoV), SARS-CoV (Beta-CoV), and SARS-CoV-2 (Beta-CoV)^19,20. The specific genomic or protein accession numbers used for each virus in the alignment were as follows: HCoV-299E (GenBank: YP_009047219.1), HCoV-OC43 (GenBank: YP_009555241.1), HCoV-KU1 (GenBank: YP_443096.1), MERS-CoV (UniProt: K9N5Q8), SARS-CoV (UniProt: P59594), and SARS-CoV-2 (RefSeq: NC_045512.2; residues 912–984). Alignment parameters were set to their default values: Full protein sequences were input, and the output alignment was formatted using the BLOSUM62 substitution matrix. The phylogenetic tree was constructed using the maximum likelihood method embedded within Clustal Omega²¹.

Determination of HR1–HR2 interacting residues

The high-resolution atomic structure of the HR1–HR2 fusion core (PDB ID: 6LXT; resolution 2.2 Å)²² was analyzed using BioVia Discovery Studio Visualizer 2021 (https://www.3ds.com/products/biovia/discovery-studio)²³. Hydrogen bonds and hydrophobic interactions between HR1 and HR2 were identified using the “Non-Covalent Interactions” module with default parameters (H-bond distance cutoff: 3.0 Å, angle cutoff: 120°). Key active residues on HR1 (N928, Q935, Q949, N953, N960) involved in stable hydrogen bonding with HR2 were selected as Ambiguous Interaction Restraints (AIRs) for guiding the High Ambiguity Driven protein-protein DOCKing (HADDOCK) 2.4 docking process.

Preparing the structure of spike protein

The prefusion state structures of the spike protein from 6 old variants of SARS CoV-2 with suitable resolutions were obtained from RCSB PDB (https://www.rcsb.org/)¹⁷ with the following PDB codes: Alpha (8DLJ)²⁴, Beta (8DLL)²⁴, Kappa (7TF3)²⁵, Epsilon (8DLW)²⁴, Gamma (8DLO)²⁴, Omicron (7T9K)²⁶, and spike protein of the investigated circulating SARS-CoV-2 variants including EG.5.1, HV.1, BA.2.86, and JN.1 were modeled using SWISS-Model (https://swissmodel.expasy.org/)²⁷. Variant-specific mutations were incorporated based on sequences deposited in GISAID (https://gisaid.org/)²⁸.

• EG.5.1: XBB.1.5 spike with additional mutations Q52H and F456L.

• HV.1: EG.5.1 spike with additional mutations F157L and L452R.

• BA.2.86: BA.2 spike with > 30 mutations, including V445H, N450D, and L452W.

• JN.1: BA.2.86 spike with the additional mutation L455S.

Model quality was validated using the SAVES v6.0 server (https://saves.mbi.ucla.edu/). Ramachandran plots were analyzed to assess the stereochemical quality by visualizing the dihedral angles (φ and ψ) of amino acid residues in the modeled structures, with > 95% of residues residing in the favored regions indicating good backbone conformation. Qualitative Model Energy Analysis (QMEAN) scores were used to evaluate the global quality and reliability of the models by comparing their geometry to high-resolution experimental structures. Additionally, the ERRAT program was employed to analyze the non-bonded atomic interactions within the models, with scores above 90% confirming a favorable environment for each residue and overall high model accuracy^8–10,29,30.

Molecular Docking analysis

For protein-peptide docking, the S2 subunits from each structure were obtained from the structure closest to the pre-hairpin state of the spike protein. HADDOCK 2.4 web server (https://wenmr.science.uu.nl/haddock2.4/)³¹ was used to perform molecular docking. HADDOCK 2.4 utilizes a flexible docking approach that incorporates information to model biomolecular complexes. This method differs from ab initio techniques as it leverages predicted or known protein interfaces. A crucial aspect of this analysis involved the use of Ambiguous Interaction Restraints (AIR), which were defined as intermolecular distances (diAB) between any atom ‘m’ of an active residue ‘i’ within protein A (miA) and any atom ‘n’ within both active and passive residues ‘k’ (total Nres residues) of protein B (nkB) – the reverse applied to protein B. It is important to note that the diAB distances were limited by a maximum value of 3 Å. The effective distance diABeff for each restraint was computed using the following equation³²:

To establish a comparative reference, HR1 and HR2 were separated into 2 PDB files from 6LXT and docked at the beginning of this stage. Following that, each antimicrobial peptide was docked with HR1 of spike protein from each of the investigated variants including Alpha, Beta, Kappa, Epsilon, Gamma, Omicron, EG.5.1, HV.1, BA.2.86 and JN.1.

The molecular docking protocol commenced with a sampling phase that generated one thousand distinct structures through rigid body docking. Subsequently, a refinement procedure was applied to the top two hundred ranked structures within an explicit solvent environment. The resulting ensemble of structures was then subjected to clustering analysis, utilizing a backbone root-mean-square deviation cutoff of 7.5 Ångströms to define cluster membership, with the ten most populous clusters selected for detailed examination. To validate the methodological approach, the native HR1–HR2 complex (PDB ID: 6LXT) was dissociated and subjected to redocking as an internal positive control. For each specific peptide-variant combination investigated, the primary outcome metric recorded for analysis was the HADDOCK score associated with the lowest-energy representative structure from the most favorable cluster.

Safety profiling

Peptides were screened for potential adverse effects using a consensus approach across five independent tools. A peptide was classified as “safe” only if all five tools unanimously predicted the desired property. The specific tools, their prediction models, and the classification thresholds used were as follows:

ToxinPred: Predicts toxicity based on a support vector machine (SVM) model. Peptides with an SVM score < 0 were classified as non-toxic³³.

AlgPred: Predicts allergenicity using a hybrid approach combining SVM, IgE epitope mapping, and ARG database search. Peptides were classified as non-allergenic if the prediction result was explicitly “Non-Allergen”³⁴.

HemoPred: Predicts hemolytic activity using a random forest model. Peptides with a prediction result of “Non-Hemolytic” were accepted³⁵.

TargetAntiAngio: Predicts anti-angiogenic potential using a random forest model. Peptides with a prediction result of “Non-Antiangiogenic” were accepted³⁶.

IL-4 Pred: Predicts interleukin-4 induction using an SVM model. Peptides with a prediction result of “Non-Inducer” were accepted³⁷.

It is important to note that while these tools are well-established and provide a valuable initial screening (with reported accuracies typically > 80–85% in their original publications), their predictions are based on computational models and are not infallible. False positives and negatives are possible. Therefore, the results obtained from this in silico safety profiling should be considered strong preliminary evidence for prioritization, but not a substitute for subsequent experimental validation in biological assays.

Molecular dynamics simulation

A comprehensive series of all-atom molecular dynamics simulations was executed utilizing the GROMACS 2020.6 software package in conjunction with the AMBER03 force field. To establish statistical confidence, three distinct molecular systems were simulated in triplicate, with each replica extending to 500 nanoseconds and employing unique initial velocity distributions. These systems included the isolated HR1 domain, encompassing residues 912 to 984 of the SARS-CoV-2 Spike protein, the native HR1–HR2 complex as resolved in the 6LXT structure, and the HR1–2KEG complex derived from the lowest-energy cluster identified in the HADDOCK docking analysis. Each system was prepared by solvation within a dodecahedral box filled with TIP3P water molecules, maintaining a minimum clearance of 1.2 nanometers from the protein to the box boundary. System charge was neutralized, and a physiological ion concentration of 150 mM was established through the addition of sodium and chloride ions. The simulation protocol initiated with an energy minimization step using the steepest descent algorithm, proceeding until convergence criteria were met. This was followed by a two-phase equilibration procedure: first, a 100-picosecond simulation in the NVT ensemble to stabilize temperature at 298 K using the V-rescale thermostat, and second, a 100-picosecond simulation in the NPT ensemble to adjust pressure to 1 bar with the Parrinello-Rahman barostat. The production phase consisted of a 500-nanosecond simulation under constant NPT conditions, employing a 2-femtosecond integration time step. Long-range electrostatic interactions were managed with the Particle Mesh Ewald method, while the LINCS algorithm was applied to constrain all bonds involving hydrogen atoms^38,39. Subsequent trajectory analysis focused on key metrics to characterize system behavior, including the root mean square deviation (RMSD) of backbone atoms to evaluate structural stability, residue-specific root mean square fluctuation (RMSF) to assess local flexibility, and the intermolecular hydrogen bonding according to standard geometric criteria. The steps of the research are shown in Fig. 1.

Fig. 1.

The steps of the research.

Results

Structural validation and conservation analysis of HR1 domain

The conservation analysis of HR1 domains across betacoronaviruses revealed striking sequence homology in critical functional regions (Fig. 2). Structural alignment demonstrated that key residues involved in fusion core formation remain invariant despite substantial variation in other spike protein regions. To further elucidate the molecular basis of HR1-HR2 interactions, we analyzed the native fusion core structure (PDB: 6LXT), which revealed five critical HR1 residues (N928, Q935, Q949, N953, and N960) forming stable hydrogen bonds with HR2 (Fig. 3). This evolutionary conservation, observed across all major SARS-CoV-2 variants including recent circulating strains, provides a strong rationale for targeting HR1 as a strategy to overcome viral mutational escape. The six-helix bundle formation between HR1 and HR2 represents a fundamental mechanistic step in viral entry that is essential across the betacoronavirus genus, making it an ideal target for broad-spectrum inhibitor development. Model quality was rigorously validated to ensure reliability of subsequent analyses. Using the SAVES v6.0 server (https://saves.mbi.ucla.edu/), Ramachandran plots confirmed that > 95% of residues were in favored regions for all models, Z score: 0.71, and ERRAT scores exceeded 90.8%, indicating good stereochemical quality and sequence-structure compatibility (Fig. 4).

Fig. 2.

Conservation of HR1 in Betacoronavirus family. (a) Sequence alignment of HR1 region from spike protein of six Betacoronavirus family members capable of infecting humans. (b) Phylogenetic tree of the Betacoronavirus family constructed by the likelihood of their spike protein. This figure was generated using Clustal Omega.

Fig. 3.

HR1 interacting residues with HR2: the blue helix indicates HR1, HR2, and its interacting residues are colored in red, and hydrogen bonds from N928, Q935, Q949, N953, and N960 are shown with black dashed lines.

Fig. 4.

Ramachandran plots obtained from SWISS-MODEL for modeled structures. (a) spike protein of SARS-CoV-2 EG.5.1. (b) spike protein of SARS-CoV-2 HV.1. (c) spike protein of SARS-CoV-2 BA.2.86. (d) spike protein of SARS-CoV-2 JN.1. e, f) Qmean plots, g) ERRAT plot.

Docking results

As a comparison standard and control, the affinity of HR1 for HR2 was determined by docking HR1 and HR2 as indivual PDB files from 6LXT. The docking result from HR1 and HR2 shows that these two regions need − 118.8 score for binding to each other. Figure 5 illustrates Ramachandran plot from natural HR1HR2 complex (6LXT) and after docking structure wich demonstrates insignificant variation between the two. Based on this data 16 peptides with the amount of energy closest to the HR1HR2 complex were separated. Table 1 displays the docking results from these peptides and HR1 from each SARS-CoV-2 variant. Our docking score results range from − 35.0 (Lantibiotic carnocin UI49) to -176.1 (plantaricin K (2KEG)). Based on the results shown in Table 1, three peptides have higher binding affinities to HR1 than the control (HR1 and HR2 binding energy) in most investigated variants. These peptides are Lactococcin-G subunit beta (PDB code: 2JPK), PlnK (PDB code: 2KEG), and Glycocin F (PDB code: 2KUY). Figures 6 and 7 illustrate 2KEG in complex with spike protein from some of the investigated variants of SARS-CoV-2.

Fig. 5.

HR1-HR2 complex natural structure (PDB code 6LXT) and after docking structure obtained from HADDock outcome. (A) natural structure. (B) after docking structure (docking of HR1 and HR2 sepreated from 6LXT).

Table 1.

HADDOCK 2.4 docking scores of peptide–SARS‑CoV‑2 variant interactions this table presents the molecular Docking scores (obtained using HADDOCK 2.4) between a set of peptide structures (PDB IDs) and Spike proteins from multiple SARS‑CoV‑2 variants.

Haddock 2.4 scores
PDB ID	Gamma	Kappa	Omicron	Beta	Alpha	Epsilon	EG.5.1	HV.1	JN.1	BA.2.86
UI49	-36	-38.7	-38.3	-33.9	-37.6	-39.7	-42.2	-39.8	-48.2	-47.5
2JPK	-134.8	-118.9	-138.2	-127.2	-128.6	-126.2	-146.4	-138.9	-168.0	-166.7
2MWT	-119.7	-123.2	-107.0	-127.9	-131.0	-116.3	-125.3	-124.4	-158.2	-159.6
2KUY	-121.7	-139	-133.5	-128.4	-128.8	-127.5	-157.8	-166.6	-135.0	-148.2
1CW5	-127.8	-110.5	-126.5	-119.1	-113.7	-113.1	-130.3	-125.3	-97.1	-108.7
2KEG	-120.6	-121.8	-120.0	-131.5	-119.0	-148.4	-158.3	-176.1	-140.9	-143.6
2DCV	-119.8	-105.8	-102.5	-113.9	-131.9	-113.2	-129.3	-129.6	-124.9	-120.4
1RKK	-113.6	-122.3	-123.3	-125.7	-112.0	-98.7	-106.9	-116.3	-127.3	-131.9
2N4K	-112.5	-115.8	-112.5	-109.9	-111.4	-114.8	-117.5	-122.8	-104.6	-106.6
1LYP	-109.4	-101.6	-103.2	-102.4	-105.3	-96.4	-96.6	-104.9	-90.3	-86.5
1Z64	-107.8	-101.2	-102.9	-104.2	-100.8	-96.8	-104.6	-111.9	-102.2	-102.5
1ZFU	-106.2	-110.1	-105.4	-103.2	-110.0	-103.6	-122.0	-115.9	-110.0	-97.1
1CW6	-104.2	-108.2	-106.5	-117.9	-107.6	-117.5	-119.9	-115.9	-126.9	-122.9
1H5O	-101.5	-106.9	-115.6	-106.6	-89.8	-105.3	-114.4	-115.5	-105.8	-111.6
2MCW	-100.4	-94.7	-91.9	-98.2	-93.4	-95.6	-100.8	-92.9	-100.6	-94.2
2MLU	-95.9	-95.8	-101.4	-103.1	-95.8	-109.8	-117.2	-111.5	-129.3	-129.6
1PXQ	-94.8	-98.4	-97.6	-95.0	-98.8	-97.8	-117.5	-118.3	-114.8	-117.2
Control	-118.8	-118.8	-118.8	-118.8	-118.8	-118.8	-118.8	-118.8	-118.8	-118.8

Lower (more negative) scores represent stronger predicted binding affinity. The “Control” row provides a reference baseline for score comparison. Results highlighted with a bold indicate the strongest binding interaction for each variant.

Fig. 6.

Predicted binding mode of the 2KEG peptide (shown in stick representation) with the HR1 domain (shown as black ribbon) of the SARS-CoV-2 (A) HV.1 and (B) Alpha variant spike protein. Key residues of HR1 involved in hydrogen bonding with the native HR2 are highlighted in red. The peptide binds within the hydrophobic groove of HR1, occupying the same site as the endogenous HR2 peptide, thus acting as a competitive inhibitor.

Fig. 7.

The interaction between peptide and the SARS-CoV-2 BA.2.86 and JN.1 spike protein. (a) 2KEG in complex with SARS-CoV-2 BA.2.86. HR1 is colored in black. Residues colored in red are from HR1 interacting residues with HR2. (b) 2KEG in complex with SARS-CoV-2 JN.1. HR1 is coloured in black. Residues colored in red are from HR1 interacting residues with HR2.

Peptide characteristics and side effects results

According to the results of studies on toxicity, allergy, stimulation of interleukin 4 secretion, antiangiogenic, and hemolytic properties displayed in Table 2, among the 16 peptides that showed high binding affinities, only two of them, PlnK (2KEG) and CAP18 protein (1LYP), have no side effects and can be used. Unlike 1LYP, 2KEG shows high binding affinities to HR1 in most investigated variants. Although the Lactococin-G beta subunit (2JPK) showed high binding affinity in most variants, it is an allergen and induces interleukin-4 secretion.

Table 2.

Peptide characteristics and side effect results.

Peptide code	Toxicity	Allergen	IL-4 inducer	Antiangiogenesis	Hemolytic
2JPK	Non-toxic	Allergen	IL-4 inducer	Non-Antiangiogenic	Non-hemolytic
2N4K	Toxic	Non-allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
1CW5	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
2KEG	Non-toxic	Non-allergen	Non-inducer	Non-Antiangiogenic	Non-hemolytic
2MWT	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
2MLU	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
1PXQ	Non-toxic	Allergen	IL-4 inducer	Non-Antiangiogenic	Hemolytic
2KUY	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
1RKK	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
2MCW	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
1Z64	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
1ZFU	Non-toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
1H5O	Toxic	Allergen	Non-inducer	Non-Antiangiogenic	Hemolytic
2DCV	Toxic	Allergen	IL-4 inducer	Non-Antiangiogenic	Non-hemolytic
1LYP	Non-toxic	Non-allergen	Non-inducer	Non-Antiangiogenic	Non-hemolytic
1CW6	Non-toxic	Allergen	IL-4 inducer	Non-Antiangiogenic	Hemolytic

Significant values are in [bold].

Among the top-scoring peptides, PlnK (2KEG) was selected as the lead candidate for further investigation based on a multi-factorial rationale that extended beyond docking scores alone. While peptides such as 2JPK and 2KUY demonstrated high binding affinity, their predicted allergenic or hemolytic potential (Table 2) rendered them less suitable for therapeutic development. In contrast, PlnK (2KEG) exhibited a consistently high binding affinity across all tested variants (Table 1) while also possessing a pristine predicted safety profile being non-toxic, non-allergenic, non-hemolytic, and non-antiangiogenic. Furthermore, its natural origin as a bacteriocin from the probiotic Lactiplantibacillus plantarum suggests a potentially superior biocompatibility profile. This unique combination of broad-spectrum efficacy, high predicted safety, and natural provenance positioned PlnK as the most promising candidate.

MD simulation

RMSD analysis

RMSD of the backbone atoms was calculated to assess the structural stability of the S(HR1)-2KEG complex during the 500 ns MD simulation. The RMSD values for the S protein (HR1 region) of the HV.1 variant ranged between 0.8 and 1.5 nm, while the 2KEG peptide exhibited a narrower range of 0.2 to 0.4 nm (Fig. 8A). The relatively stable RMSD values for both components indicate that the complex reached equilibrium early in the simulation, with minimal structural deviations thereafter. The lower RMSD of the S(HR1)-2KEG suggests its inherent rigidity and stable binding to the HR1 region, which is critical for maintaining the inhibitory effect on viral fusion. The absence of significant fluctuations further supports the formation of a stable complex, essential for blocking the HR1-HR2 interaction and interfering the formation of the fusion core.

Fig. 8.

A Root Mean Square Deviation over the 500 ns simulation timeline. Bown: 2KEG peptide, black: S(HR1) protein, blue, green, and yellow: three replicate of the S(HR1)-2KEG complex. B: The RMSF plot of the S protein of HV.1 variant in the S(HR1)-2KEG complex, B1: S protein (HR1), B2: 2KEG peptide. C: Number of hydrogen bonds between the HR1 domain and the 2KEG peptide over the simulation timeline. The consistent presence of 3–10 H-bonds indicates stable binding. Fluctuations are due to the dynamic nature of molecular interactions.

RMSF analysis

RMSF was analyzed to evaluate the flexibility of individual residues in the S(HR1)-2KEG complex. The HR1 region of the S protein exhibited RMSF values between 0.1 and 0.3 nm, while the 2KEG peptide showed slightly lower fluctuations, ranging from 0.1 to 0.25 nm (Fig. 8B). The low RMSF values for the HR1 region indicate that its secondary structure remains stable upon binding to 2KEG, particularly in the key residues involved in HR1-HR2 interactions. The minimal fluctuations in the 2KEG peptide further confirm its stable binding to HR1, which is crucial for competitively inhibiting HR2 binding and disrupting the fusion core formation.

H-bonds analysis

H-bond dynamics were monitored to assess the stability of interactions between the HR1 region and the 2KEG peptide. Throughout the 500 ns (500000 ps) simulation, the complex maintained 3 to 10 H-bonds (Fig. 8C), with persistent bonds forming between the peptide and critical HR1 residues. These H-bonds are vital for stabilizing the complex and ensuring sustained inhibition of HR1-HR2 interactions. The consistent presence of multiple H-bonds suggests strong and specific binding, which is essential for interfering the conformational changes required for viral membrane fusion.

The combined analysis of RMSD, RMSF, and H-bonds provides a comprehensive understanding of the stability and inhibitory mechanism of the S(HR1)-2KEG complex. The low RMSD and RMSF values demonstrate the structural integrity of the complex, while the persistent H-bonds highlight the specific interactions that interfere HR1-HR2 binding. These findings suggest that 2KEG effectively locks the HR1 region in a conformation incompatible with fusion core formation, thereby inhibiting viral entry. This mechanistic insight underscores the potential of 2KEG as a broad-spectrum inhibitor against SARS-CoV-2 variants, including emerging strains with mutations in the spike protein.

Discussion

The emergence of SARS-CoV-2 in late 2019 spurred unprecedented global research into its infection mechanisms and treatment strategies. Coronaviruses exhibit high mutability, particularly in the spike protein, which complicates the development of effective vaccines and therapeutics. While the receptor-binding domain (RBD) undergoes rapid evolution driven by immune selection, heptad repeat regions 1 and 2 (HR1/HR2) remain comparatively conserved due to structural constraints essential for membrane fusion^38,44. This conservation makes HR1 an attractive target for broad-spectrum antiviral development against current and future variants.

While targeting the conserved HR1 domain offers a compelling strategy for broad-spectrum activity, this approach must be critically evaluated against the dominant therapeutic paradigm of RBD-targeting. RBD-directed agents, particularly monoclonal antibodies, have demonstrated notable clinical efficacy due to the domain’s accessibility for neutralization. However, this strategic advantage is counterbalanced by the RBD’s high mutability, which has facilitated the emergence of escape variants that rapidly diminish therapeutic effectiveness. In contrast, the HR1-targeting strategy represents a fundamental trade-off: it promises broader variant coverage due to high sequence conservation but encounters distinct challenges regarding molecular accessibility, as the HR1 region becomes exposed primarily during the transient pre-hairpin intermediate of the fusion process. The present investigation’s focus on HR1 with peptide inhibitors therefore represents a deliberate prioritization of durability and breadth against viral evolution, while acknowledging the different developmental hurdles compared to RBD-targeted approaches.

The evolutionary pressure on coronaviruses manifests differently across various protein domains. The RBD, being the primary target for neutralizing antibodies, accumulates mutations rapidly to evade host immunity. In contrast, HR1 and HR2 maintain higher sequence conservation because they participate in essential protein-protein interactions that are structurally constrained any significant alterations in these regions would likely disrupt the precise conformational changes required for membrane fusion. This fundamental difference in evolutionary dynamics provides a compelling rationale for targeting conserved fusion machinery rather than variable surface epitopes.

Our study builds upon growing interest in antimicrobial peptides (AMPs), particularly bacteriocins from probiotic bacteria, as therapeutic agents against viral infections. Unlike synthetic peptides derived from viral sequences (e.g., HR2-mimetics), probiotic-derived bacteriocins offer potential advantages in biocompatibility, safety, and production scalability^40,41. Previous work by Balmeh et al. demonstrated the feasibility of bioengineered probiotic peptides against SARS-CoV-2^42,43, while Zheng et al. showed stapled peptides targeting HR1 could inhibit fusion. Our approach uniquely leverages naturally occurring bacteriocins, specifically plantaricin K (2KEG) from Lactiplantibacillus plantarum, which demonstrated high-affinity binding to HR1 across all tested variants without predicted toxicity or immunogenicity.

Several key findings emerge from our study with significant implications for antiviral development: First, 2KEG consistently bound HR1 with high affinity (HADDOCK scores: -120.6 to -176.1) across ten variants, including circulating strains (EG.5.1, HV.1, BA.2.86, JN.1). Molecular dynamics simulations confirmed stable complex formation, with RMSD values stabilizing below 0.2 nm after 20 ns and maintained hydrogen bonding with critical HR1 residues (N928, Q935, Q949, N953, N960). The persistence of these specific hydrogen bonds throughout the simulation period suggests that 2KEG effectively occupies the HR2 binding site, potentially interfering the formation of the six-helix bundle essential for viral entry.

Second, in silico safety profiling using consensus tools (ToxinPred, AlgPred, HemoPred) predicted no adverse effects, addressing potential concerns about peptide therapeutics’ immunogenicity or toxicity. The favorable safety profile of 2KEG may be attributed to its natural origin, as bacteriocins from probiotic bacteria have co-evolved with mammalian systems, potentially reducing their likelihood of triggering adverse immune responses. Third, the conservation of HR1 across betacoronaviruses suggests that 2KEG could remain effective against future variants, unlike RBD-targeted therapies susceptible to immune escape. This cross-reactive potential is particularly valuable given the continuous emergence of new variants.

Our results align with broader efforts to develop fusion inhibitors targeting conserved viral regions^44–48. For example, EK1, a pan-coronavirus fusion inhibitor, also targets HR1 but requires chemical synthesis and lacks the natural probiotic origin that may enhance 2KEG’s biocompatibility⁴⁵. The probiotic source of 2KEG offers potential additional benefits, including oral bioavailability, compatibility with mucosal delivery (e.g., inhalation), and possible immunomodulatory effects⁴⁹. Notably, probiotics may enhance mucosal immunity and compete with pathogens for epithelial binding sites, potentially providing adjunctive benefits against respiratory viruses like SARS-CoV-2⁵⁰.

It is crucial to acknowledge that the computational evidence presented here, while robust, has inherent limitations. Molecular docking and dynamics simulations, though powerful screening tools, operate under certain approximations regarding force field parameters and system simplification that may not fully capture physiological complexity or complete spike protein dynamics. These findings thus represent a strong preclinical hypothesis requiring experimental validation through cell-based assays and animal studies.

The translation of 2KEG into a viable therapeutic faces several challenges common to peptide-based pharmaceuticals. The linear structure may be susceptible to proteolytic degradation and rapid renal clearance, potentially resulting in a short plasma half-life. Implementation of stabilization strategies—such as backbone cyclization, incorporation of non-natural amino acids, or conjugation with polyethylene glycol (PEG)—could significantly enhance metabolic stability. Furthermore, targeted delivery to infection sites remains crucial; inhalation administration presents a promising route for respiratory infections like COVID-19, potentially maximizing local efficacy while minimizing systemic exposure.

The immunogenicity profile of 2KEG warrants thorough investigation. Although its probiotic origin suggests a favorable safety baseline, repeated administration might still elicit immune responses that could either neutralize the peptide or cause adverse effects. Conversely, this property could be strategically leveraged in vaccine development. From a regulatory perspective, a probiotic-derived peptide therapeutic would navigate a unique pathway intersecting guidelines for biologics and possibly live biotherapeutic products, necessitating comprehensive data on preclinical safety and manufacturing consistency.

However, several limitations must be acknowledged to provide proper context for our findings. Our study is computational; experimental validation of 2KEG’s fusion inhibition efficacy in cell-based assays (e.g., pseudovirus neutralization) is essential. Additionally, while in silico safety profiling is promising, in vivo toxicity and pharmacokinetic studies are needed to confirm suitability for therapeutic use. Future work should explore formulation strategies such as encapsulation or fusion with stabilizing domains to enhance 2KEG’s metabolic stability and delivery to respiratory tissues. The potential for rapid renal clearance and proteolytic degradation, common challenges with peptide therapeutics, must also be addressed in subsequent development phases.

In the broader context of peptide-based therapeutics, researchers have investigated various targets including RdRp, 3CL protease, and the RBD region^51,52, as well as key receptors such as ACE2, TMPRSS2, and GRP78^53–55. The conservation of HR1 across betacoronaviruses and its functional role in membrane fusion make it an ideal target for durable therapeutic design. Notably, the low mutational frequency in HR1, contrasted with the highly mutable RBD, suggests that such inhibitors could remain effective against future variants⁴⁴.

From a mechanistic perspective, 2KEG appears to function as a competitive inhibitor, binding to the hydrophobic grooves of HR1 that normally interact with HR2 during fusion core formation. This inhibition strategy differs fundamentally from antibody-based approaches that target variable surface epitopes. The multi-valent nature of the trimeric spike protein suggests that effective inhibition may require simultaneous binding to all three HR1 domains, a possibility that warrants further investigation.

Collectively, our findings position 2KEG as a promising broad-spectrum candidate targeting the conserved HR1 domain, with potential applications against current and emerging SARS-CoV-2 variants. The probiotic origin of this bacteriocin may facilitate its development as a prophylactic or therapeutic agent, particularly suited for mucosal delivery routes such as inhalation. This investigation highlights the untapped potential of mining natural microbial compounds for antiviral agents and reinforces the strategic importance of targeting conserved viral regions to circumvent variant-driven therapeutic escape. Furthermore, the integration of computational screening with natural product discovery establishes a robust framework for addressing emerging viral threats in an era characterized by rapid pathogen evolution.

Conclusion

This study computationally identified the probiotic-derived bacteriocin plantaricin K (2KEG) (RRSRKNGIGYAIGYAFGAVERAVLGGSRDYNK) as a high-affinity, broad-spectrum inhibitor targeting the conserved HR1 domain of the SARS-CoV-2 spike protein. Through rigorous molecular docking and dynamics simulations, 2KEG demonstrated stable binding to HR1 across ten variants, including currently circulating strains (e.g., EG.5.1, HV.1, BA.2.86, JN.1), with HADDOCK scores surpassing the native HR2 peptide. Consensus in silico safety profiling predicted no toxicity, allergenicity, or hemolytic activity, highlighting its potential biocompatibility.

The conservation of HR1 across SARS-CoV-2 variants and betacoronaviruses suggests that 2KEG could remain effective against future viral evolution, addressing a key limitation of RBD-targeted therapies. Its origin from Lactiplantibacillus plantarum provides distinct advantages, including potential for scalable fermentation production, natural biocompatibility, and possible adjunctive benefits as a probiotic.

These findings position 2KEG as a promising candidate for further development as a broad-spectrum antiviral agent. Future work must prioritize experimental validation, including: In vitro fusion inhibition assays using pseudotyped viruses, Cell-based neutralization tests against live SARS-CoV-2 variants, In vivo efficacy and toxicity studies in animal models.

Ultimately, this work underscores the value of exploring natural antimicrobial peptides from probiotics as a source of novel therapeutics against rapidly evolving pathogens, offering a complementary strategy to conventional vaccine and antibody-based approaches.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

BS, AS, and FJ collected, analyzed, wrote, and interpreted the data regarding the impact of probiotics as a therapeutic food, and GNH discussed the results. NA designed, wrote, edited, and supervised the work. All authors read and approved the final manuscript.

Data availability

Data is provided within the manuscript.

Competing interests

The authors declare no competing interests.