Structure–Antimicrobial Activity Relationships of Recombinant Host Defence Peptides Against Drug‐Resistant Bacteria

ABSTRACT

Host defence peptides (HDPs) represent a valuable class of antimicrobial agents with the potential to address the growing threat of antimicrobial resistance (AMR). Here, we have studied recombinant constructs based on a combination of HDPs fused to the GFP protein and multidomain proteins combining three or four HDPs in a single polypeptide, referred to as first and second generation antimicrobials, respectively. These recombinant peptides were tested against Gram‐positive and Gram‐negative bacteria associated with healthcare infections. In addition, in silico studies provided insight into the antimicrobial structure–activity relationships of these biomolecules. For the first generation of antimicrobials, amphipathicity mainly explains the average antimicrobial activity against the Gram‐positive strains. In the case of the Gram‐negative bacteria, it depends on the quantity and the exposed area of the Ser and Thr amino acids. For the second generation of antimicrobials, the order of domains is crucial to act against Gram‐positive strains, preferably by positioning the most bioactive domain against the Gram‐positive pathogen at the ends.

Recombinant constructs of Host defence peptides (HDPs) represent a valuable class of antimicrobial agents to address the growing threat of antimicrobial resistance (AMR). Structure–activity relationship (SAR) studies demonstrate that amphipathicity mainly explains the average antimicrobial activity against the Gram‐positive strains, and the exposed area of alcoholic side chain amino acids influences the bioactivity against the Gram‐negative bacteria. For multidomain proteins, the order of HDPs domains is crucial to act against Gram‐positive strains.

1. Introduction

Therapeutic peptides represent a promising class of safe, selective and efficient pharmacological agents (Fosgerau and Hoffmann 2015; Henninot et al. 2018), which can be produced using both chemical and biological methods (Wang et al. 2022). Peptide‐based therapeutics represent a valuable class of antimicrobial agents with the potential to address the growing threat of antimicrobial resistance (AMR) (Mookherjee et al. 2020), which is a pressing global health concern. AMR occurs when microbes evolve mechanisms to evade the effects of antimicrobial agents, rendering existing treatments ineffective. The misuse and overuse of antibiotics in clinical, agricultural settings and the food chain have accelerated this process, leading to a rise in multidrug‐resistant (MDR) pathogens (O'Neill 2014; Lee Ventola 2015; Ayukekbong et al. 2017; Wu‐Wu et al. 2023).

Peptide‐based therapeutics can target pathogens with high specificity while minimising the development of resistance compared to conventional antibiotics (Mookherjee et al. 2020). They can target microbial vulnerabilities that are less prone to mutation and can potentially overcome resistance mechanisms that have rendered conventional antibiotics obsolete (Hancock and Sahl 2006). Despite their potential, these biomolecules face challenges in terms of stability, bioavailability and immunogenicity (Goles et al. 2024). Thus, exploring strategies to address these limitations, such as peptide modifications, recombinant protein production, formulation technologies and delivery systems, are crucial for the successful development of these biomolecules as future drugs (Fosgerau and Hoffmann 2015).

Several natural and synthetic antimicrobial peptides have shown efficacy against a wide range of pathogens, including bacteria, fungi and viruses. For instance, host defence peptides (HDPs) such as LL37 and related peptides (Krishnamoorthy et al. 2023), as well as defensins (Zasloff 2002) have demonstrated potent activity against both Gram‐positive and Gram‐negative bacteria. These peptides often disrupt microbial membranes or interfere with essential cellular processes, making it difficult for pathogens to develop resistance (Hancock and Sahl 2006). Importantly, akin to the success story of insulin production, which evolved through phases of human and porcine production before finally being produced recombinantly (Zaykov et al. 2016), HDPs are undergoing a similar journey due to advances in their production using recombinant technology and demonstrating significant efficacy in combating pathogenic microorganisms (Roca‐Pinilla et al. 2022; López‐Cano et al. 2023; Nazarian‐Firouzabadi et al. 2024).

The bioactivities of HDPs mentioned above can be attributed to both their three‐dimensional structure and diverse physicochemical properties (Zamora, De Souza, et al. 2020; Zamora et al. 2023). Understanding these properties is crucial for the design and optimisation of HDP‐based therapies (Torres et al. 2019). Some of the main physicochemical properties of HDPs include cationicity, meaning they carry a net positive charge at physiological pH, which facilitates their interaction with the negatively charged microbial membranes, leading to membrane disruption and cell lysis (Hancock and Sahl 2006). Lipophilicity also plays a significant role in the interaction of HDPs with microbial membranes because HDPs often possess both hydrophobic and hydrophilic regions, allowing them to insert into lipid bilayers and disrupt membrane integrity (Yin et al. 2012; Cherry et al. 2014). Amphipathicity and secondary structure are other fundamental properties of HDPs that contribute to their mechanism of action and effectiveness against microbial pathogens (Santos et al. 2016). An amphipathic structure (including α‐helices and β‐sheets) consists of regions with both hydrophobic (water‐repellent) and hydrophilic (water‐attracting) properties within the same molecule, which allows HDPs to interact with and disrupt microbial membranes while minimising interactions with host cell membranes that carry an overall electric charge approximately neutral. In this context, modulating the amphipathic and cationic properties is crucial for enhancing the bioactivity of HDPs while minimising toxicity caused by low selectivity (Brogden 2005; Bechinger 2009; Bechinger and Gorr 2017; Wang et al. 2017).

The convergence of the structural information of HDPs and their physicochemical properties, along with the evidence of their antimicrobial activity, has catalysed a plethora of investigations into the study of structure–activity relationships. These studies have demonstrated that the three properties mentioned above—cationic nature, hydrophobicity and amphipathicity—are critical determinants of HDPs bioactivity (Lin et al. 2012; Neubauer et al. 2017; Torres et al. 2019; Ciulla and Gelain 2023). However, from a quantitative perspective, the models and correlations derived have not been entirely satisfactory (Ando et al. 2010; He and Lazaridis 2013; Lipkin and Lazaridis 2015; Zhu et al. 2017; Ahmed and Hammami 2019). This underlines the importance of delving deeper into leveraging the wealth of information offered by the diverse structures of HDPs and employing refined models to describe crucial properties such as lipophilicity and amphipathicity of peptides (Wegener 2001; Wang et al. 2005; Sani and Separovic 2018). While modern machine learning models exhibit the capability to differentiate HDPs from non‐HDPs (Porto et al. 2017, 2022) and identify antimicrobial peptides encrypted within other proteins based solely on sequences and properties like the hydrophobic moment (Lee et al. 2016; Ma et al. 2022; Cesaro and de la Fuente‐Nunez 2023; Cesaro et al. 2023; Huang et al. 2023; Coelho et al. 2024; de la Fuente‐Nunez 2024; Wan et al. 2024), further exploration is warranted to include intrinsically disordered regions that challenge these computational models (Bello‐Madruga and Torrent Burgas 2024) and fully harness the richness of structural information, physicochemical properties and bioactivities of HDPs.

In this study, our objective was to provide a comprehensive analysis of the structure–antimicrobial activity relationships of a novel class of recombinant HDPs against both Gram‐positive and Gram‐negative bacteria, particularly in the context of healthcare‐associated infections. We place particular emphasis on in vitro synergistic effect, structural features, and computed physicochemical properties of the produced recombinant peptides, with the latter being calculated using a state‐of‐the‐art model for lipophilicity and amphipathicity.

2. Materials and Methods

2.1. Computational Section

2.1.1. Structure Modelling

Peptide structures for representation of first generation recombinant constructs and second generation antimicrobials (see Figure 1) were modelled using the ColabFold Server (Mirdita et al. 2022) which gives an accelerated prediction of peptide structures by combining the homology search of MMseqs2 with AlphaFold2. The resulting structural models (rank 1 model for each peptide) were saved in their PDB formats and employed for the representations used in this work.

FIGURE 1.

Representation of the first generation antimicrobials (top). (a) The antimicrobial constructs are constituted of regions in common (GFP, Green fluorescent protein in green; L, linker sequence (GGSSRSS) in red; N‐t, N‐terminal region with the Met amino acid in red; C‐t, C‐terminal region with Cys amino acid in red; H6, 6 histidine (H6)‐tag in cyan) and the HDP‐based differentiator domain (grey). (b) The HDP domains correspond to HβD2 (orange), HβD3 (light green), LAP (magenta), HD5 (yellow) and LL37 (blue). Representation of the second generation antimicrobials (bottom) which are constituted of regions in common (C‐t, C‐terminal region with Cys amino acid; H6, 6 histidine (H6)‐tag; L, linker sequence (GGSSRSS); N‐t, N‐terminal region with Met amino acid) and the combination of HDP domains. (c) HD5‐LL37‐HβD3 polypeptide, (d) HD5‐LL37‐HD5‐LL37 polypeptide and (e) HD5‐LAP‐LL37‐HβD3 polypeptide.

2.1.2. Determination of Sequence‐Based and Context‐Dependent Lipophilicity of Host Defence Peptide Domains

The first and second generation antimicrobials were designed to have fragments in common, and a discriminant region named the HDP‐based sequence (see Figure 1). Therefore, for lipophilicity computations, we focus on the HDP domains. The structural information of the HDPs was obtained from the PDB database (Burley et al. 2019): human α‐defensin 5 (HD5, PDB code 2lxz (Wommack et al. 2012)); human β‐defensin 2 (HβD2, PDB code 1fd3 (Hoover et al. 2000)); human β‐defensin 3 (HβD3, PDB code 1kj6 (Schibli et al. 2002)); and human cathelicidin (LL37, PDB code 5nmn (Sancho‐Vaello et al. 2017)). However, for the β‐defensin lingual antimicrobial peptide (LAP), it was simulated using AlphaFold2 (Mirdita et al. 2022) since its structure is not deposited in the PDB.

The sequence‐based lipophilicity of the HDP domains in both the first and second generation antimicrobials, down to the residue level, was computed using the Grand average of hydropathicity (GRAVY) (Stothard 2018) and using the values for individual amino acids reported in the ProtL scale to pH = 7.4 (logD _7.4) (Zamora et al. 2019; see Figure S1).

An additional approach, called context‐dependent lipophilicity, was used to take advantage of the wealth of chemical information deposited in the bioactive structure of the HDP domains. In this regard, the ProtL scale (Zamora et al. 2019), which is a pH‐dependent and structure‐based lipophilicity scale, was utilised. It is based on the IEFPCM/MST continuum solvation method (Luque et al. 2003), wherein the lipophilicity of each amino acid is calculated, considering its specific structural features and the pH of the medium. This methodology was implemented as reported in our previous work (Zamora et al. 2019) according to Equation (1).

$$\log {\left(\raisebox{1ex}{${\mathrm{D}}_{\mathrm{pH}}$}\!\left/ \!\raisebox{-1ex}{${\mathrm{P}}_{\mathrm{N}}$}\right.\right)}^{\text{peptide}}=\sum _{\mathrm{i}=1}^{\mathrm{N}}\left({\mathrm{\lambda }}^{\mathrm{i}}·\log {\left(\raisebox{1ex}{${\mathrm{D}}_{\mathrm{pH}}$}\!\left/ \!\raisebox{-1ex}{${\mathrm{P}}_{\mathrm{N}}$}\right.\right)}^{\mathrm{bb}+\mathrm{sc}}+{\mathrm{\lambda }}^{\mathrm{i}}·\log {\left(\raisebox{1ex}{${\mathrm{D}}_{\mathrm{pH}}$}\!\left/ \!\raisebox{-1ex}{${\mathrm{P}}_{\mathrm{N}}$}\right.\right)}^{\mathrm{cg}}+{\alpha }^i+{\beta }^i+{\gamma }^i\right)$$ (1)

Equation (1) applies to any peptide composed of N amino acids, where ${\lambda }^i$ stands for the fraction of solvent‐exposed surface area (SASA) of the amino acid, including the backbone (bb), side chain (sc) and capping groups (cg) according to the local structural environment of the HDP domain. For our purposes, the SASA was determined using NACCESS program (Hubbard et al. 1991). The ${\beta }^i$ factor accounts for a correction due to the burial of the side chain of hydrophobic residues (Ala, Leu, Ile, Val, Pro, Phe, Trp, Met and Tyr) from water to a lipophilic environment. This contribution has been estimated to be 0.023 kcal mol⁻¹ Å⁻² according to the studies reported by Moon and Fleming for the transfer of nonpolar side chains from water into a lipid bilayer (Moon and Fleming 2011). Therefore, the ${\beta }^i$ term had been estimated from the fraction of the buried side chain concerning the fully buried side chain (for details, see Data S1 of Ref. 55).

In this version of the calculation of context‐dependent lipophilicity of amino acids, two correction factors were introduced: (i) an adjustment in the ${\alpha }^i$ term and (ii) an additional correction factor, ${\gamma }^i$, which will be detailed as follows. Initially, the parameter ${\alpha }^i$ introduced a correction to the hydrophobic contribution when the backbone participates in a hydrogen bond (HB). However, in the new algorithm proposed in Equation (1), the factor ${\alpha }^i$ also incorporates the hydrophobic contribution due to the side‐chain intramolecular hydrogen bonds. This contribution can be estimated to amount, on average, to 0.73 (logP units) per HB (Pace et al. 2014). Finally, the ${\gamma }^i$ factor accounts for the enhancement of the amino acid's lipophilic character due to the formation of salt bridges. This electrostatic interaction has been experimentally shown to increase the transfer of oppositely charged side chains in model peptides in n‐octanol/water systems by up to ca. 4 logP units relative to glycine (Wimley, Gawrisch, et al. 1996). Additionally, it stabilises the presence of charged residues in highly hydrophobic environments, such as transmembrane helices (Musafia et al. 1995; Jayasinghe et al. 2001). Stabilisation in both immiscible solvent systems and transmembrane proteins has demonstrated a consensus value of about 1 logP unit (approximately 1 kcal/mol) compared with the cost of transferring these ionic side chains individually without interacting with each other (Duart et al. 2022). In this framework, for consistency with our calculations, Table 1 presents the n‐octanol/water partition coefficient of noninteracting pairs of oppositely charged amino acid side chains. On average, the transfer of these pairs of side chains without interacting and relative to glycine in the ProtL scale (Zamora et al. 2019) is around −9 logP units. This suggests that to stabilise these pairs in hydrophobic environments by at least 1 logP unit (0.5 logP units by residue), the interaction must favour the transfer to n‐octanol by at least 10 logP units. In this way, the value adopted for ${\gamma }^i$ in each amino acid forming a salt bridge is approximately 5 logP units. For special cases involving a three‐body interaction, where the same charged side chain participates in two salt bridges, a ${\gamma }^i$ value of 7 logP units was established for the amino acid that interacts twice to achieve an additional global stabilisation of 2 logP units (1 logP unit by pair). These extra logP units compared to a two‐body interaction are supported by evidence indicating that three‐body intermolecular interactions in proteins, for example, cation–cation interactions, exhibit greater stability than the simple additive assumption (Pinheiro et al. 2017). Indeed, experimentally quantifying this extra stability provides an opportunity to design experiments in biphasic systems as well as in vitro and whole‐cell systems (Duart et al. 2022).

TABLE 1.

Matrices of the n‐octanol/water partition coefficient of noninteracting pairs of oppositely charged amino acid side chains (logP _I) relative to Gly obtained from the ProtL scale. The logP _I of the ionisable side chain is shown in parentheses.

Amino acid side chains (logP I)	Asp (−5.4)	Glu (−3.1)
Lys (−3.7)	−9.1	−6.8
Arg (−4.3)	−9.7	−7.4

Consequently, in this work, the new correction factors, ${\alpha }^i$ and ${\gamma }^i$, were applied in the calculation of lipophilicity of the HDP domains using a pH = 7.4. Table S1 shows the type and number of intramolecular side‐chain hydrogen bonds (IMHB) and intramolecular salt bridges (IMSB) present in the five HDP domains and Figures S2–S5 show the structural detail of these interactions within HDPs. In addition, Figure S6 depicts the lipophilicity profile of the HDPs where the final logD _7.4 for each amino acid is reported. In addition, the CSV files are uploaded to GitHub: https://github.com/cbio3lab/SAR_RECOMBINANT_HDPs detailing each of the lipophilicity calculation models, adding the correction factors one by one, until the final value is obtained according to Equation (1).

2.1.3. Determination of Sequence‐Based and Context‐Dependent Amphipathicity of Host Defence Peptide Domains

The sequence‐based and context‐dependent amphipathicity was calculated for HDP regions with a well‐defined hydrophilic and hydrophobic face. Specifically, for HβD2, HβD3 and LAP (see Figure 2) the calculation relied on the α‐helix stretches. However, for HD5 and LL37, the hydrophilic and hydrophobic regions spanned virtually the entire sequence (see Figure 2). Equation (2) presents the method for calculating amphipathicity (Amp):

$$\mathrm{Amp}=\sum _i^N\log {\left(\raisebox{1ex}{${\mathrm{D}}_{\mathrm{pH}}$}\!\left/ \!\raisebox{-1ex}{${\mathrm{P}}_{\mathrm{N}}$}\right.\right)}^{\text{Hydrophobic}}-\sum _j^N\log {\left(\raisebox{1ex}{${\mathrm{D}}_{\mathrm{pH}}$}\!\left/ \!\raisebox{-1ex}{${\mathrm{P}}_{\mathrm{N}}$}\right.\right)}^{\text{Hydrophilic}}$$ (2)

where the logD _pH or logP _N stands for the sequence‐based or context‐dependent (see Equation 1) lipophilicity of i and j amino acids on the hydrophobic and hydrophilic faces, respectively.

FIGURE 2.

(a) Representation of the amphipathic nature of five HDP domains in the first generation antimicrobials. HDP stretches exhibiting polar and apolar sides are shown in blue and yellow surfaces, respectively. To the right of each HDP, the amphipathic region is extracted and the amino acids on the polar and apolar sides are indicated in blue and yellow letters, respectively. (b) Computed context‐dependent amphipathicity to pH = 7.4 of first generation antimicrobials.

Let us highlight that the methodology employed in this algorithm draws upon the principles of the model proposed by Eisenberg and collaborators (Eisenberg et al. 1982) for calculating the hydrophobic moment. However, it offers the advantage of applying to any type of structure, not limited to α‐helices. Furthermore, it exploits the advantages of other vector methods (Reißer et al. 2014; Pillong et al. 2017), as the logD values are computed considering the three‐dimensional arrangement of amino acids in the peptides. An additional benefit is that it does not rely solely on sequence‐based lipophilicity values but incorporates structural and energetic details. These elements are integrated into Equation (1), making this novel algorithm a state‐of‐the‐art approach for determining peptide amphipathicity.

2.1.4. Molecular Docking Studies

Molecular docking studies were performed using the local 3D Zernike descriptor‐based protein docking (Venkatraman et al. 2009; Christoffer et al. 2021) for pairwise comparisons between the HDP domains. The submitted structures were HD5 (PDB code 2lxz), HβD2 (PDB code 1fd3), HβD3 (PDB code 1kj6), LL‐37 (PDB code 1nmn) and LAP (simulated structure with AlphaFold2). The programme generates several models for the pair of interacting domains, up to 50,000. The models are assessed by four scores: GOAP, DFIRE, ITScore and Rank Sum.

The GOAP (Generalised Orientation‐dependent All‐atom Potential) is a physics‐based scoring function that evaluates the binding energy of protein–protein complexes which consider both the shape complementarity and the energetics of the interaction between protein residues including van der Waals forces, electrostatic interactions and hydrogen bonding (Zhou and Skolnick 2011). The DFIRE (Distance‐scaled Finite Ideal‐gas Reference) is a statistical potential‐based scoring function derived from a large database of known protein structures and their interactions which calculates the energy of interaction between residues based on their pairwise distances and types (Zhou and Zhou 2002). The ITScore (Interface and surface‐based scoring function) is a knowledge‐based scoring function that considers various geometric and physicochemical properties of protein surfaces to assess their compatibility for binding (Huang and Zou 2011). Finally, Rank Sum represents a consensus scoring approach established on the sum of the ranks from the three previous scores and was the score used for selecting the best model of each pair of interacting domains. The results of these scores and the relative score based on the weakest pair of interacting domains according to Rank Sum are noted in Table S2.

2.2. Experimental Section

2.2.1. Recombinant Proteins

Mature sequences of human‐beta‐defensin 2 (HβD2), human‐beta‐defensin 3 (HβD3), lingual antimicrobial peptide (LAP), human defensin 5 (HD5) and cathelicidin LL37 were fused to the green fluorescence protein (GFP) gene through the linker sequence SGGGSGGS in the case of first generation antimicrobials (see Figure 1; López‐Cano et al. 2023). In the case of second generation antimicrobials, HDPs were fused with each other using the same linker and without the need of using GFP, resulting in the multidomain proteins HD5‐LL37‐HβD3, HD5‐LL37‐HD5‐LL37 and HD5‐LAP‐LL37‐HβD3 (see Figure 1; López‐Cano et al. 2023). Both generations of antimicrobials were finally fused to a C‐terminal 6 histidine (H6)‐tag and sequences were codon optimised by GeneArt (Life technologies, Regensburg, Germany). The final constructs were cloned into pET22b (Amp^R) vector and transformed by heat shock in competent Escherichia coli BL21 (DE3) cells.

Transformed E. coli BL21 cells were grown in Luria‐Bertani (LB) broth with ampicillin at 100 μg/mL in 2 L shake flasks with 500 mL of LB at 37°C and 250 rpm. The recombinant protein production was induced by 1 mM isopropyl‐β‐d‐thiogalactopyranoside (IPTG) when cultures reached an OD600 = 0.4–0.6. After 3 h of induction, the cells were harvested by centrifugation (6200 × g, 15 min, 4°C) and the resultant pellets were stored at −80°C until purification.

Pellets coming from the production step were resuspended in binding buffer (500 mM NaCl, 20 mM Tris, 20 mM imidazole) with PMSF protease inhibitor. The suspensions were then disrupted at 20 KPsi (Constant Systems CF1 disruptor) and centrifuged for 45 min at 15,000 × g at 4°C. Then, the supernatant was filtered with 0.2 μm membrane filters. Finally, proteins were purified by Immobilised Metal Affinity Chromatography (IMAC) in ÄKTA Start (GE Healthcare) using 1 mL HisTrap chelating HP columns (GE Health care). The purified proteins were dialysed in 0.01% acetic acid and their final concentration was measured by Qubit 4 (Thermo Fisher).

2.2.2. Antimicrobial Activity

A broad screening bactericidal assay and minimal inhibitory concentration (MIC) analyses were performed on methicillin‐sensitive Staphylococcus aureus (MSSA, ATCC‐3556), methicillin‐resistant Staphylococcus aureus (MRSA, ATCC‐33592), methicillin‐resistant Staphylococcus epidermidis (MRSE, ATCC‐35984) and Pseudomonas aeruginosa (ATCC‐10145).

2.2.3. Broad Screening Bactericidal Assay

BacTiter‐Glo Microbial Cell Viability assay (Promega) was used to determine the bactericidal activity of HDPs. Briefly, an overnight (O/N) culture of the selected pathogenic strain was reinoculated in fresh Brain–Heart Infusion (BHI) broth and grown at 250 rpm and 37°C until exponential growth was reached. The resultant culture was diluted in 10 mM KPi buffer at 10⁵cfu/mL following an OD vs. cfu/mL correlation equation for each strain. Then, the diluted pathogens were separated in different eppendorfs and centrifuged. The resultant pellets were resuspended with either acetic acid 0.01% (negative control) or 5 μM of HDPs treatment and disposed of in a sterile 96‐well plate of polypropylene (Costar). The microplate was incubated for 5 h at 37°C. After incubation, 100 μL of each well was transferred to a 96‐well polystyrene microplate and then mixed with 100 μL of BacTiter‐Glo reagent. The microplate was shaken for 30 s and then incubated for 5 min at room temperature. Finally, the luminescence was measured in a microplate reader, LumiStar (BMG LABTECH). The bacterial survival was assessed by comparing the luminescence of the treated wells with the negative controls.

2.2.4. Minimum Inhibitory Concentration (MIC)

The MIC of HDPs was assessed using a modified broth microdilution method. Overnight cultures of MRSA, MSSA, MRSE or P. aeruginosa in BHI broth were diluted to a concentration of 106 cfu/mL in fresh Mueller Hinton broth (MHB‐II) 10% (MHB‐II, Sigma‐Aldrich). This diluted MHB‐II is used so as not to interfere with the activity of cationic peptides. Meanwhile, a two‐fold serial dilution in 0.01% acetic acid was performed with each peptide tested, taking into account that the peptide concentrations had to be ×3 concentrated compared with the desired final concentrations. Subsequently, 73 μL of the bacterial suspension was mixed with 37 μL of each peptide dilution and added to a 96‐well polypropylene microtiter plate (Costar). As a negative control, 73 μL of bacterial suspension and 37 μL of 0.01% acetic acid were added together in some wells. After adding all the treatment conditions and controls, the plate was incubated at 37°C for 24 h. Bacterial survival was measured by BacTiter‐GloTM reagent as explained in the section before. The MIC was considered as the minimum HDP concentration that totally inhibited bacterial growth.

2.2.5. Haemolytic Activity

For the haemolysis assay, the HDPs were diluted to a final concentration of 10 μM, 5 μM, and 2.5 μM in 0.005% acetic acid buffer. Blood was collected in EDTA tubes from calves. 1 mL of blood was resuspended in PBS (pH = 7.4) at a final volume of 4 mL, and then the solution was centrifuged (1500 × g, 5 min and 4°C). After centrifugation, the plasma layer was removed and the red blood cells (RBCs) were washed in PBS and centrifuged (1500 × g, 5 min and 4°C) for three rounds. When the last wash was completed, the RBCs were again resuspended in PBS at a final volume of 1 mL. A cell count was then performed to determine the cells/mL of RBC solution. The RBCs were stained with trypan blue and counted in a Neubauer chamber with a microscope. Once the concentration of the RBCs stock was known, it was diluted to a final concentration of 5 × 10⁷ cells/mL in PBS. Subsequently, 100 μL of the final RBC solution was added to 100 μL of each HDPs dilution. A negative control was also performed by adding 100 μL of the RBC solution to 100 μL of 0.005% acetic acid buffer, as well as a positive control by adding the same volume of RBC solution to 100 μL of 1% Triton X‐100 diluted in 0.005% of acetic acid buffer. All conditions were done in triplicate and incubated at 37°C for 1 h. After the incubation, the samples were centrifuged (1500 × g, 5 min and 4°C) and 100 μL of the supernatants were transferred to a flat‐bottomed 96 well plate (Eppendorf). The haemoglobin released due to toxicity was measured by absorbance at 405 nm by a microplate reader (LumiStar). The analysis of the results was represented in % of haemolysis.

3. Results

This study reports a detailed analysis of the structure–antimicrobial activity relationships of recombinant HDPs against Gram‐positive and Gram‐negative bacteria in healthcare‐associated infections. The recombinant constructs are the result of the combination of different HDPs with the GFP protein and an H6‐tag, and multidomain proteins combine three or four HDPs in a single polypeptide with an H6‐tag (see Figure 1), hereinafter referred to as first and second generation antimicrobials, respectively. To the best of our knowledge, this study represents a pioneering exploration of structure–activity relationship (SAR) analysis within this particular class of recombinant peptide‐based antimicrobial agents.

In the following sections, we will delve into their antimicrobial activity, paying special attention to the physicochemical properties and structural patterns of these biomolecules, which underlie their diverse bioactivities.

3.1. Physicochemical Properties of First Generation Antimicrobials

The most extensively used physicochemical properties to characterise antimicrobial peptides (AMPs) are net charge, lipophilicity and amphipathicity (Eisenberg et al. 1982; Kim et al. 2005; Bechinger 2009; Yin et al. 2012; Cherry et al. 2014; Henriksen et al. 2014; Santos et al. 2016; Wang et al. 2017). These are used to explain the bioactivity and toxicity of HDPs (Subbalakshmi et al. 2000; Ruiz et al. 2014; Santos et al. 2016; Oddo and Hansen 2017; Greco et al. 2020; Zamora, De Souza, et al. 2020; Khabbaz et al. 2021; Salem et al. 2022; Krishnamoorthy et al. 2023). Taking into account that the first generation antimicrobials designed have fragments in common (N‐t, L, GFP, and H6 sequences; see Figure 1; Roca‐Pinilla et al. 2022; López‐Cano et al. 2023), we focus on their discriminant region, the HDP‐based sequence, to calculate the physicochemical properties detailed in Table S3.

Table S3 demonstrates that all five first generation antimicrobials exhibit a net positive charge at pH = 7.4, a crucial attribute for their interaction with the predominantly negatively charged bacterial membrane (Yin et al. 2012; Rai and Qian 2017). Table S3 also shows for the five HPDs domains analysed in this study (excluding common segments) the lipophilicity determined from the sequence using the Grand average of hydropathicity (GRAVY) score alongside our amino acid lipophilicity scale (ProtL), represented as the n‐octanol/water distribution coefficient (logD _7.4) (Zamora et al. 2019). Let us mention that these calculations rely on a straightforward cumulative method to assess the lipophilicity of peptides and proteins.

Concerning the cumulative method utilised for computing peptide lipophilicity (sequence‐based logD _7.4), it is evident that all five HDPs are predominantly hydrophilic. This observation stems from the fact that this type of sequence‐based calculation provides information for a nonstructured peptide (MacCallum and Tieleman 2011; Peters and Elofsson 2014; Simm et al. 2016). Given that these distribution coefficients exceed by an order of magnitude those reported for small molecules (organic compounds with molar masses less than 700 Da) (Avdeef 2003; Ruiz et al. 2022) a common approach to reporting the lipophilicity in peptides is by means of the average of hydropathicity values (e.g., GRAVY score and average sequence‐based logD _7.4). Likewise, these findings point out that all five HDP domains are water‐associated, with particular emphasis on the HβD3 and LL37 peptides. As expected, our calculation as well as that reported by the GRAVY score are correlated (see Figure S1).

On the other hand, our computations incorporate an additional approach, called context‐dependent logD _7.4 (see Table S3), which considers both the three‐dimensional structure information of the HDPs and the pH (see 2.1 Computational Section for details). The acidity of the medium in which the antimicrobial assays were conducted was considered, which closely resembled the physiological pH (ca. 7.4). Hence, each amino acid will possess a distinct lipophilicity value that works like a fingerprint (see lipophilicity profiles of the HDP domains in Figure S6). The application of this method for calculating lipophilicity has proven highly valuable in elucidating the experimental binding affinities of MHC‐bound peptides (Zamora et al. 2019) and assessing the stability of the SARS‐CoV‐2 nonstructural protein 9 (NSP9) dimer (de Araújo et al. 2021). Notably, this approach leverages the three‐dimensional information of the proteins and peptides under investigation, which cannot be captured through a basic summation based solely on the sequence of the peptide.

In this framework, the context‐dependent approach revealed that two HDP domains (HβD3 and LAP) were remarkably hydrophilic due to the three‐dimensional distribution of their amino acids (see lipophilicity profiles of the HDP domains in Figure S6). On the other hand, three HDP domains were shown to be mainly hydrophobic named HD5, HβD2 and LL37. LL37 exhibits significantly higher lipophilicity attributed to the existence of intramolecular hydrogen bonds (IMHBs), which are known to enhance the permeability of small molecules and drugs (Kuhn et al. 2010; Ermondi et al. 2014; Caron and Ermondi 2017; Caron et al. 2017; David et al. 2021), PROTACs (Hornberger and Araujo 2023; Apprato et al. 2024) and peptides (Goetz et al. 2014; Whitty et al. 2016; Guha et al. 2019), a property closely linked to lipophilicity. Experimental evidence has shown an increase in lipophilicity of around 1 logP unit due to IMHBs (Caron and Ermondi 2005, 2017; Borges et al. 2017; Caron et al. 2018). Failing to account for this factor can indeed lead to a decrease in the performance of computational models (Işık et al. 2020; Zamora, Pinheiro, et al. 2020). In the α‐helix structure of LL37, the polar amide backbone is shielded by intramolecular hydrogen bonds. Moreover, the existence of side‐chain intramolecular hydrogen bonds (see Table S1 and Figures S2–S5) further enhances the lipophilicity of the HDPs. Finally, the presence of charged residues is effectively neutralised by intramolecular salt bridges (IMSBs, see Table S1 and Figures S2–S5), further enhancing its lipophilic character. As outlined in the computational methods section, no published lipophilicity scale to date incorporates both IMHBs and the influence of IMSBs in its calculation into a pH and context‐dependent lipophilicity scale of amino acids. However, experimental evidence suggests that salt bridges can elevate lipophilicity (Wimley, Gawrisch, et al. 1996; Wimley, Creamer, and White 1996) and stabilise charged side chains in highly hydrophobic regions with low dielectric constants such as cell membranes (Jayasinghe et al. 2001).

It is important to note the significant shift in lipophilicity observed in LL37 when the structural context is taken into account, as this contradicts the high hydrophilicity suggested by both GRAVY and our average sequence‐based logD _7.4. This underlines the relevance of context‐dependent approaches together with correction factors based on intramolecular interactions which have been experimentally observed. Consequently, the lack of correlation between GRAVY and our structure‐based calculation is notable (see Figure S1) and underscores the limitations of solely sequence‐based methods. Indeed, our structure‐based lipophilicity calculation aligns with the high hydrophobicity observed in HPLC experiments for LL37 (Dutta et al. 2015; Hemshekhar et al. 2019).

Figure 2 presents the calculated amphipathicity (see Computational Section for details) of the first generation antimicrobials. Commonly, this descriptor is calculated by the hydrophobic moment (μ) using the helical‐wheel projection, but it is only applicable to helical structures (Schiffer and Edmundson 1967; Zelezetsky and Tossi 2006). For other types of structures, the approaches applied rely on defining polar and apolar sides and using lipophilicity scale values to determine the hydrophobicity difference in these regions (Reißer et al. 2014; Edwards et al. 2016; Pillong et al. 2017). In this work, we aim to exploit the structural richness of HDPs. Accordingly, as detailed in the methods section, the computation of our descriptor for amphipathicity depends directly on the calculated pH and context‐dependent lipophilicity of the amino acids contributing to the formation of both the hydrophilic and hydrophobic faces (see Figure 2 and lipophilicity profiles in Figure S2).

Since amphipathicity must be calculated on structures that possess a well‐defined hydrophilic and hydrophobic side, our calculation focused only on peptide stretches (α‐helix regions) exhibiting this characteristic for HβD2, HβD3 and LAP (see Figure 2). Furthermore, this approach is supported by experimental studies indicating that the disulfide bridges of β‐defensins are not essential for antimicrobial activity (Taylor et al. 2008; Ciulla and Gelain 2023). Consequently, these regions were excluded from the calculation of amphipathicity. On the contrary, for HD5 and LL37, the two faces span virtually the entire sequence (see Figure 2).

Figure 2b reveals that the most amphipathic peptides are represented by HD5 and HβD3, followed by LAP and HβD2, and finally, the least amphipathic peptide corresponds to LL37. It is worth mentioning that the amphipathicity calculated using the sequence‐based lipophilicity values (see Table S4) follows a similar order for the HDP domains, with the clear exception of the amphipathicity calculated using the structure‐based lipophilicity values for LL37. The significant decrease in the structure‐based amphipathicity observed for LL37 can once again be attributed to the abundance of intramolecular hydrogen bonds and salt bridges present in its structure. These structural features effectively neutralise the polarity of oppositely charged residues, resulting in the polar layer being less polar overall. This contrast would not be apparent if only the raw contribution of its sequence were considered. This fact highlights the critical importance of calculating lipophilicity and amphipathicity while taking into account the peptide's structure, as well as the specific interactions that influence these descriptors.

The physicochemical properties computed in this section, utilising our model of lipophilicity and amphipathicity dependent on pH and local context, will serve as the foundation for exploring the structure–antimicrobial activity relationships of recombinant HDPs against drug‐resistant Gram‐positive and Gram‐negative bacteria in the subsequent sections.

3.2. Structure–Antimicrobial Activity Relationships of First Generation Host Defence Peptides Against Gram‐Positive and Gram‐Negative Strains

In our previous work (López‐Cano et al. 2023), HDPs produced as single molecules fused to a carrier protein as GFP protein (see Figure 1) presented important advantages over individual HDPs, such as cutting down the toxicity on the producer cell and protecting the HDPs from host proteases. Furthermore, they showed promising bioactivities against healthcare‐associated infectious agents, including three Gram‐positive pathogens, such as methicillin‐susceptible Staphylococcus aureus (MSSA), methicillin‐resistant S. aureus (MRSA) and methicillin‐resistant Staphylococcus epidermidis (MRSE), and one Gram‐negative strain represented by MDR Pseudomonas aeruginosa . In this section, we aim to delve into the structure–antimicrobial activity relationships of the HDPs against Gram‐positive and Gram‐negative strains.

Table S5 exposes the reduction of colony‐forming units (log10 cfu/mL) of the different first generation constructs against Gram‐positive strains (MSSA, MRSA and MRSE). On average, the main reduction in the planktonic form of Gram‐positive pathogens fell on the HβD3 recombinant peptide, then HD5, followed by HβD2 and LAP and finally LL37. Table S6 shows the minimal inhibitory concentration assay which determines the minimal concentration of an HDP required to inhibit pathogen growth. Similar to the results in Table S5, on average, the most and least active recombinant peptides against Gram‐positive strains were HβD3 and LL37, respectively. Considering the results of Tables S5 and S6 together, it is observed that there is a differential antimicrobial activity between the recombinant HDPs. Therefore, we explored the influence of the calculated lipophilicity and amphipathicity (see Table S3 and Figure 2b) on these results.

Lipophilicity has been identified as a critical descriptor in determining the bioactivity of peptides. For example, it plays a pivotal role in determining the bacterial killing rate of AMPs and influences the development of bacterial resistance (Zhang et al. 2022). Figure 3a depicts the relationship between the computed average context‐dependent lipophilicity versus the planktonic assay (average reduction of colony‐forming units [log10 cfu/mL]) and average minimal inhibitory concentration (MIC, μM) of Gram‐positive strains (MSSA, MRSA and MRSE). In general, the peptides included in this work exhibiting higher context‐dependent hydrophilicity tend to demonstrate better antimicrobial activity in average reduction of colony‐forming units (Figure 3a, left). A similar trend is observed for the average MIC except for LAP (Figure 3a, right). Previous studies have reported poorer correlations with lipophilicity when employing an additive approach based on constant lipophilicity values for each amino acid (He and Lazaridis 2013), which aligns using either the GRAVY score or our sequence‐based approach (see Figures S7 and S8, respectively).

FIGURE 3.

(a) Relationship between computed average context‐dependent lipophilicity versus (left) average reduction of colony‐forming units (log10 cfu/mL) and (right) average minimal inhibitory concentration (MIC, μM) of Gram‐positive strains (MSSA, MRSA, and MRSE). The r represents the Pearson correlation coefficient. The p values of the correlations are 0.04 and 0.19 with a 95% confidence interval, respectively. (b) Relationship between computed amphipathicity versus (left) average reduction of colony‐forming units (log10 cfu/mL) and (right) average minimal inhibitory concentration (MIC, μM) of Gram‐positive strains (MSSA, MRSA, and MRSE). The r represents the Pearson correlation coefficient. The p values of the correlations are 0.04 and 0.07, respectively.

Figure 3b reflects an encouraging correlation and trend between the wide screening antimicrobial assay and the determination of the MIC of the recombinant HDPs with the calculated context‐dependent amphipathicity. Peptides with a higher amphipathic character exhibit superior antimicrobial activity against Gram‐positive pathogens. Interestingly, incorporating the presence of intramolecular interactions such as hydrogen bonds and salt bridges into the algorithm for calculating lipophilicity appears to capture the experimental evidence of decreased bioactivity with lower amphipathicity. Although all peptides studied exhibited some level of these interactions (see Table S1 and Figures S2–S5), LL37 demonstrated a greater quantity, significantly influencing the final value of lipophilicity and, consequently, the amphipathicity value. Let us mention that previous studies have shown the influence of amphipathicity on the bioactivity of antimicrobial peptides, mainly in helical structures, but correlations have been discrete (Edwards et al. 2016; Santos et al. 2016; Wang et al. 2017). In this manner, our findings substantiate the appropriateness of our approach for computing context‐dependent lipophilicity and amphipathicity. Despite the increased complexity associated with requiring a structural template for the HDPs, this is offset by the encouraging results achieved in this work.

For the Gram‐negative MDR Pseudomonas aeruginosa , Table S7 compiles the plain bactericidal activity of the first generation constructs. In general, all peptides were able to kill the strains in this assay except LL37, which has a limited effect. Similarly, Table S8 shows the inability of the LL37 peptide to prevent the growth of the MDR P. aeruginosa, although in this MIC assay, it was possible to distinguish the bioactivity of the other peptides where the most active was the HD5, then HβD2 and LAP followed by HβD3.

From a structure–activity relationship point of view, Figure S9 shows that lipophilicity does not convincingly explain the results presented in Tables S7 and S8. On the other hand, amphipathicity seems to show some trend (see Figure S9, right). This fact imposes the distinct activity of recombinant HDPs toward Gram‐positive and Gram‐negative strains, which enhances their value in terms of selectivity for the development of new antimicrobials (Wang 2017). Indeed, our results align with previous studies indicating that the bioactivities of α‐helical AMPs correlate with retention times in HPLC (an experimental observable related to charge, hydrophobicity and amphipathicity) in S. aureus (Gram‐positive bacteria) but not in E. coli (Gram‐negative bacteria) (Kim et al. 2005).

Considering the general composition of the outer membrane of Gram‐negative bacteria, which mainly consists of lipopolysaccharides (Strahl and Errington 2017; Carey et al. 2022), led us to question whether the presence of alcoholic side chain amino acids (Ser and Thr) in the amphipathic regions (see Figure 4a) of the recombinant peptides would influence its bioactivity due to its possible favourable interaction through intermolecular hydrogen bonds with the polysaccharides of the outer bacterial membrane. Accordingly, Table S9 collects the results of the analysis of the number and the solvent‐accessible surface area (SASA) of Ser and Thr residues present in the amphipathic structures of the HDPs analysed in this work. Interestingly, the most active peptide, HD5, has the highest amount of these amino acids, three Thr and three Ser. On the other hand, the less active ones, LL37 and HβD3, contain the least number of these residues. Let us mention that, in the case of LL37, although it has a Ser residue in the α‐helix region, it is buried, so it may not be a major contributor to intermolecular hydrogen bonding interactions with the bacterial membrane.

FIGURE 4.

(a) Representation of alcoholic side chain amino acids (ROH AAs = Ser and Thr) in the bioactive regions of the first generation recombinant HDPs. (b) Relationship between the minimal inhibitory concentration (MIC, μM) against Gram‐negative MDR P. aeruginosa and (left) the number (n) of alcoholic side chain amino acids (ROH AAs = Ser and Thr) and (right) the number of alcoholic side chain amino acids corrected by SASA fraction (n_SASA). In the graph to the left, a linear model was modelled to fit the data, while on the right, an exponential decay where α ~2.43, ln (MIC)_min ~ 0.21, ln (MIC)_max ~ 2.42 according to the best fit. The root‐mean‐square error (RMSE) is calculated relative to the experimental values of ln (MIC) assuming a linear (left) and exponential decay model (right). The r represents the Pearson correlation coefficient (left) and ρ the Spearman correlation (right). The p values of the correlations are 0.06 and 0.05, respectively.

Figure 4b depicts the structure–antimicrobial activity relationships considering the number of serine (Ser) and threonine (Thr) amino acids in the amphipathic regions of the HDPs and the number of those amino acids corrected by the SASA fraction (n_SASA) considering the total SASA as the sum of each SASA of the 11 residues noted in Table S9. The results show that there is a tendency that the greater the number of these residues in the amphipathic region, the greater the bioactivity against the Gram‐negative bacteria. Using only the number of alcoholic amino acids, the relationship appears to resemble a linear model (as depicted in the graph on the left in Figure 4b). However, when weighting the number of ROH AAs according to SASA, an exponential decrease with an improved correlation and root‐mean‐square error (RMSE) is observed. To the best of our knowledge, this observation has not been previously reported in other structure–activity relationship studies; however, it should be mentioned that our findings are supported by the effective antimicrobial activity of alterins, a class of marine cyclolipopeptides from Crassostrea gigas oyster haemolymph, against Gram‐negative bacteria (Offret et al. 2022). The cycloheptapeptide alterine family contains five nonproteinogenic amino acid residues, three of which have a modified side chain with amine and alcoholic groups. These structural modifications enable the heptapeptides to interact with the lipopolysaccharides in the outer membrane of Gram‐negative bacteria (Desriac et al. 2020). Consequently, our results represent a significant advancement and open up new avenues of research for enhancing the development of novel antimicrobials targeting Gram‐negative pathogens.

3.3. Haemolytic Activity of First Generation Recombinant HDPs

One of the main drawbacks of antimicrobial peptides is related to haemolytic activity. Erythrocyte lysis is closely related to the amphipathic properties of HDPs according to the literature (Oddo and Hansen 2017; Salem et al. 2022) and hence we proceeded to analyse four out of the five first generation recombinant HDPs since the LAP and HβD2 exhibited similar amphipathic properties (see Figure 2b). Table S10 supports previous observations on AMPs, where one of the most amphipathic domains (HβD3) showed haemolytic activity at concentrations as low as 2.5 μM, while the least amphipathic peptide (LL37) showed no activity at all even at a fourfold higher concentration.

It is of utmost importance to note that this result further validates the incorporation of intermolecular interactions such as hydrogen bonds (IMHBs) and salt bridges (IMSBs) in the calculation of proper context‐dependent lipophilicity and amphipathicity. Without considering these interactions, the computed hydrophobicity for LL37 (sequence‐based logD _7.4 = −94.52, see Figure 2b) would have resulted in a significant amphipathicity value (sequence‐based Amp = 60.05, see Table S4), which would contradict the experimental findings of bioactivity and haemolytic activity.

3.4. Structure–Antimicrobial Activity Relationships of Second Generation Host Defence Peptides Against Gram‐Positive and Gram‐Negative Strains

Based on the individual analysis of the first generation of recombinant HDPs and their structure–activity relationships, we decided to explore a new generation of recombinant polypeptides. These combine three or four HDP domains (see Figure 1) to evaluate their potential as new antimicrobial agents while simultaneously unravelling structural details that determine their bioactivity. These tailored antimicrobials have demonstrated recombinant host viability and present some advantages over the first generation constructs, as they do not require the use of inactive protein carriers such as GFP, allowing the inclusion of only bioactive domains in the new construct. (Roca‐Pinilla et al. 2022; López‐Cano et al. 2023).

For these second generation HDPs, we established combinations that include amphipathic HDP domains with bioactivity against Gram‐positive strains (such as HβD3 and HD5), along with the most active domain against Gram‐negative strains (HD5). Additionally, to balance the potential toxicity resulting from haemolytic activity, we have chosen to intercalate the LAP and LL37 domains. Furthermore, although not evaluated in this study, LL37 has been demonstrated to possess excellent antibiofilm activity (Krishnamoorthy et al. 2023; López‐Cano et al. 2023), which compensates for its lack of activity in the assays included in this study.

From a computational point of view, the calculation of the local context‐dependent lipophilicity and amphipathicity for SAR and QSAR studies in the second generation constructs establishes a complex task. First, although our previous studies have shown the value of determining lipophilicity by considering pH and structural details of proteins (Malik et al. 2016; Zamora et al. 2019), there are no previous reports related to amphipathicity, at least in biomolecules as elaborate as those proposed in this study. Second, there are no three‐dimensional structures deposited in databases because these novel structures have been designed in our laboratories and are stored as purified soluble proteins. Finally, the available models rely on computational tools for predicting the 3D structures of proteins, in which complex biomacromolecules can be observed (see Figure 2), preserving the bioactive structures of the individual HDPs.

A practical approach to analyse the amphipathicity descriptors in these recombinant polypeptides could be based on the individual bioactive domains as discussed in the previous sections. Interestingly, all the second generation multidomain peptides have at least one of the most amphipathic and bioactive HDPs against Gram‐positive bacteria (e.g., HβD3 and HD5, see Tables S3 and S4); however, Tables S11 and S12 show that their antimicrobial activities are different, suggesting that the order and number of HDP domains used directly influence the bioactivity in a nonadditive manner.

Tables S11 and S12 reveal that the polypeptide HD5‐LAP‐LL37‐HβD3, combining four different HDP domains, was inactive against both the planktonic and growing forms of Gram‐positive strains, whereas HD5‐LL37‐HβD3 (three HDP domains) showed a modest activity. On the other hand, HD5‐LL37‐HD5‐LL37 (four HDP domains) was considerably more active than the other two polypeptides, at least in the MIC assay (see Table S12). Since the individual amphipathicity of HDP domains does not seem to explain the bioactivity in these polypeptides, this led us to analyse the arrangement of amphipathic domains within second generation contracts.

Figure 5a represents the order, number and amphipathicity of HPD domains used to build the recombinant second generation polypeptides. Based on the amphipathicity values, we can define two domains with high amphipathicity (HD5 and HβD3), one with a moderate value (LAP), and another with a low value (LL37). Curiously, only the polypeptides with alternating high/low amphipathic domains in the full combination of HDPs (HD5‐LL37‐HβD3 and HD5‐LL37‐HD5‐LL37) were those with promising bioactivity. This fact leads us to propose that in cases of polypeptide combinations, the optimal approach to preserve bioactivity is through a combination of high and low amphipathicity stretches in tandem. When this alternation is disrupted (e.g., HD5‐LAP‐LL37‐HβD3), bioactivity decreases drastically (see Tables S11 and S12). As a consequence, the polypeptides HD5‐LL37‐HβD3 (odd alternating high/low amphipathic domains) and HD5‐LL37‐HD5‐LL37 (even alternating high/low amphipathic domains) seem to stabilise the bioactive structures of the individual HDPs by preserving the antimicrobial activity against Gram‐positive bacteria.

FIGURE 5.

(a) Scheme of the functional HDP‐based domains used to design the three recombinant polypeptides. Blue, grey and yellow blocks represent domains highly, moderately, and with low amphipathicity, respectively. (b) Scheme of second generation antimicrobials using combinations of HD5, LL37 and HβD3. The red asterisk represents the most active domain against MRSE (see Table S15). Blue and yellow blocks represent HDP‐based domains highly and with low amphipathicity, respectively.

As mentioned above, the lack of alternating high/low amphipathic domains in HD5‐LAP‐LL37‐HβD3 polypeptide confers lower bioactivity compared with the HD5‐LL37‐HD5‐LL37 construct. This led us to hypothesise a possible incompatibility, understood as a favourable interaction, of some HDP domains when combined in the same polypeptide. To evaluate this possibility, molecular docking studies were performed using the local 3D Zernike descriptor‐based protein docking (Venkatraman et al. 2009; Christoffer et al. 2021) for pairwise comparisons between the HDP domains: HD5, LAP, LL37 and HβD3 (see Table S12). Figure 6 shows that the lowest interaction score (see 2.1 Computational Section for details) occurs in the coupling between HD5 and HβD3 which was used as the relative score for the other pairs. Impressively, the docking score for the HD5‐LL37 pair is only slightly larger than the reference (ca. 10%) and this pair is present in HD5‐LL37‐HD5‐LL37, the most active multidomain peptide. This fact further corroborates the stability provided using these domains. Based on their docking score and amphipathic patterns, they present a promising template with weak interactions between them. By contrast, the docking score for the LAP‐LL37 pair is significantly higher than the reference (ca. 40%) and, in agreement with experimental results, is the pair present in the intern region of the polypeptide HD5‐LAP‐LL37‐HβD3, the least active of all those studied in this work. The strong relative molecular interaction for LAP‐LL37 domains is triggered by hydrophobic interactions between the interfaces of the HDP domains (see Figure 6b).

FIGURE 6.

Docking for pairwise HDP domains. (a) Docking Score for the interaction between HDP domains (HD5, LAP, LL37 and HβD3) relative to HD5‐HβD3 complex docking score and (b) the complex between LAP and LL37 showing the interface zone which is steered by hydrophobic interactions (dashed green lines). Hydrophobic residues of LAP and LL37 are depicted in magenta and blue, respectively. Pymol was used to generate the representations (Schrödinger & DeLano (2020); Retrieved from http://www.pymol.org/pymol).

Tables S13 and S14 once again confirm the inactivity of HD5‐LAP‐LL37‐HβD3, this time against the Gram‐negative pathogen. This is surprising considering that the domains HD5 and LAP, especially the former, were active in the first generation constructs. Likewise, HD5‐LL37‐HβD3 (odd alternating high/low amphipathic domains) and HD5‐LL37‐HD5‐LL37 (even alternating high/low amphipathic domains) showed antimicrobial activity, particularly the second one. This observation supports the hypothesis that using alternating high/low amphipathic domains with a low docking score for the interaction between them (as seen in Figure 6a) stabilises the bioactive structures of the multidomain constructs to some extent. Among these, the paired high/low amphipathic domains present in HD5‐LL37‐HD5‐LL37 exhibit the most success. Therefore, HD5, being the most active HDP against P. aeruginosa and present twice in the sequence of this polypeptide, renders it the most active against this pathogen.

Overall, bearing in mind that the best single HDP against the Gram‐negative bacteria and three Gram‐positive strains was the HD5 (MIC = 1.25 μM) and the HβD3 (MIC = 1.57 μM), respectively, allows us to discern that the even alternating high/low amphipathic domains present in HD5‐LL37‐HD5‐LL37 confer to this recombinant polypeptide an optimised and unified antimicrobial activity against both Gram‐negative (MIC = 1.19 μM) and Gram‐positive strains (MIC = 2.04 μM on average).

3.5. Impact on the Antimicrobial Activity of the Host Defence Peptides Distribution Within a Multidomain Recombinant Protein

In the previous section, the HD5‐LL37‐HβD3 polypeptide presented a modest activity in both the Gram‐negative bacteria and Gram‐positive strains presumably because of their odd alternating high/low amphipathic domains. Moreover, as it presents only three HDP domains, it allows for a more thorough analysis of possible combinations of these domains, thus enabling the discovery of structural patterns based on the order of the domains that dictate their bioactivity. For the sake of simplicity, the three novel multidomain proteins: HβD3‐LL37‐HD5, HD5‐HβD3‐LL37, and LL37‐HD5‐HβD3 were designed (see Figure 5b) and tested against MRSE (Gram‐positive) and P. aeruginosa (Gram‐positive) bacteria (see Figure S10).

Table S15 shows that the original polypeptide HD5‐LL37‐HβD3 and two new recombinant peptides HβD3‐LL37‐HD5 and HD5‐HβD3‐LL37 exhibit the same antimicrobial activity against MRSE. In the case of HD5‐LL37‐HβD3 and HβD3‐LL37‐HD5, this is to be expected, since they essentially retain the odd alternating high/low amphipathic domains. Similarly, HD5‐HβD3‐LL37, despite not exhibiting the amphipathic alternation between domains, demonstrates the same biological activity as the two previous constructs. Conversely, LL37‐HD5‐HβD3 presents 100% less activity than the first three polypeptides.

To further check the structure–antimicrobial activity relationships of these second generation HDPs, we analysed the potential interactions between the domains in the four constructs depicted in Figure 5b (HD5‐HβD3, HD5‐LL37, and HβD3‐LL37). The molecular docking values (see Figure 6a) indicate that these scores are all similar. In fact, they even represent the pairs of interactions with the lowest score. Thus, they do not allow us to discern the loss of activity in LL37‐HD5‐HβD3.

Comparison of the order of combinations of HD5, LL37 and HβD3 in Figure 5b details that the antimicrobial activity against MRSE seems to depend on the position of the most active domain (HD5, see Table S6) in the multidomain construct. Therefore, the presence of the HD5 domain in the extreme position of the multidomain peptide ensures similar bioactivity (see red asterisks in Figure 5b), presumably because it exposes its amphipathic nature, necessary for its bioactivity against Gram‐positive bacteria. Otherwise, when it is in internal positions, as in LL37‐HD5‐HβD3, it causes the loss of bioactivity.

Finally, as demonstrated in the previous sections, the antimicrobial activity against Gram‐negative bacteria is mainly dependent on the presence of the HD5 domain, due to the presence of Thr and Ser residues, and not on the internal or extreme position, and since it is present in all the constructs, no difference was observed between them (see Table S15).

3.6. Comparing the Activity of Different Host Defence Peptides Produced as Single Molecules and Combined in the Same Polypeptide Sequence

To further inspect the advantages of second generation polypeptides over the first generation constructs, we compare the antimicrobial activity of HD5‐LL37‐HD5‐LL37 and HD5‐LAP‐LL37‐HβD3 multidomain proteins (second generation proteins) with the mixture of their HDPs produced as single molecules against MRSE. Let us recall that HD5‐LL37‐HD5‐LL37 was the most active second generation polypeptide while HD5‐LAP‐LL37‐HβD3 was inactive against both Gram‐positive and Gram‐negative strains (see Tables S11–S14).

Figure 7 Minimal inhibitory concentration (MIC, μM) in the different cases analysed in this work. On one hand, for the HD5‐LL37‐HD5‐LL37 polypeptide and the mixture of its stand‐alone peptides, the minimal inhibitory concentration against MRSE bacteria was essentially the same (MIC = 3.75 μM). According to Table S6, we would expect an average MIC amounting to 4.63 μM using an additive model for the individual HDP domains; however, the enhanced bioactivity points to both a synergistic effect as well as a compatibility of the use of these domains either by using the individual HPDs in a mixture or by combining them in the same polypeptide. This observation further supports the use of alternating high/low amphipathic domains with a low docking score for the interaction between them (as seen in Figure 6a) to design bioactive structures of the multidomain constructs.

FIGURE 7.

Minimal inhibitory concentration (MIC, μM) of HDPs produced as single molecules (filled red circles) and combined in the same polypeptide sequence (filled black triangles) against MRSE.

On the other hand, for HD5‐LAP‐LL37‐HβD3 the multidomain protein had a better bioactivity (MIC = 7.50 μM) than the mixture of its HDPs (MIC = 15.00 μM). Despite the higher bioactivity of the multidomain protein in this latter case, the expected average MIC is approximately 3.72 μM considering the additive model for the individual HDPs produced as single molecules (see Table S6), which means that the synergistic effect is lost. The lack of alternating high/low amphipathic domains in the HD5‐LAP‐LL37‐HβD3 polypeptide confers lower bioactivity compared with the HD5‐LL37‐HD5‐LL37 construct; nevertheless, this fact is not sufficient to explain the breakdown of the synergistic effect in the mixture of stand‐alone HD5, LAP, LL37 and HβD3 domains. Significantly, this experiment allows us to corroborate the effect of the predictions obtained from molecular docking scores of HDPs interacting domains. The loss of synergy in the mixture of individual peptides for HD5, LAP, LL37 and HβD3 domains shows that it is not only problematic to have domains such as LAP‐LL37 in a multidomain peptide, but the effect is even more pronounced by having the domains in separate peptides. Their strong relative interaction (approximately 40%, see Figure 6a) hinders the design of functional mixtures or a multidomain polypeptide against Gram‐positive and negative strains, even though their individual domains are bioactive against pathogens.

Overall, multidomain proteins allow the creation of efficient combinations of HDPs, avoiding their production fused to nonfunctional carrier proteins. These potential new antimicrobial agents are a very promising therapy with high antimicrobial capacity, which can exert synergistic effects, having the same or even better antimicrobial activity against MRSE than the combination of the same HDPs, but expressed as single proteins.

4. Discussion

The rational design of antimicrobial agents based on HDPs represents a promising strategy with the potential to address the escalating threat of AMR, a pressing global health concern. The success of designing bioactive peptides relies on modulating and accurately describing relevant physicochemical properties such as charge, lipophilicity and amphipathicity. Mastering these properties can accelerate the discovery of pathogen‐selective peptides and enable their efficient incorporation into artificial intelligence algorithms to search and screen promising sequences encrypted in a proteome.

The application of a state‐of‐the‐art approach to computing lipophilicity and amphipathicity, which incorporates pH and context‐dependent features such as intramolecular hydrogen bonds and salt bridges, along with peptide–peptide docking studies and experimental antimicrobial assays, has enabled us to establish encouraging structure–antimicrobial activity relationships of recombinant host defence peptides against Gram‐positive (MSSA, MRSA, and MRSE) and Gram‐negative ( P. aeruginosa ) bacteria in healthcare‐associated infections.

This study included four defensins (HβD3, HβD2, LAP and HD5) whose mechanism of action primarily involves interactions with bacterial membranes—and considering that their tertiary structures are similar despite low amino acid sequence similarity (Taylor et al. 2008). We propose that their distinctive bioactivity can be explained and modulated by key physicochemical features along with structural motifs which play a critical role in defining their function. By analogy to the widespread use of physicochemical descriptors as metrics for assessing molecular quality in small‐molecule drug discovery, our findings can similarly guide the evaluation and optimisation of defensins.

The results point out that both first and second generations of recombinant HDPs constitute new promising agents against these drug‐resistant bacteria. In the first recombinant proteins, we were able to determine that lipophilicity, and to a greater extent, amphipathicity can predict the average antimicrobial activity against the Gram‐positive strains analysed in this work. In the case of the Gram‐negative bacteria, the results indicate that the bioactivity of first generation polypeptides depends mainly on the quantity and the exposed area of the alcoholic side chain amino acids (Ser and Thr) on the amphipathic stretches of the HDP domains. To the best of our knowledge, this observation has not been previously reported in defensins. Consequently, our results represent a notable advancement and open up new avenues of research for enhancing the development of novel antimicrobials targeting Gram‐negative pathogens.

Our results align well with previous studies in the field of quantitative structure–activity relationship (QSAR) modelling of antimicrobial peptides (AMPs), which consistently highlight the critical role of cationic amino acids—such as lysine, arginine and histidine—in facilitating electrostatic interactions with the negatively charged bacterial membranes. In particular, the amphipathic regions shown in Figure 2 display a notable enrichment in arginine residues compared to lysine. This is relevant given that arginine‐rich peptides have been associated with both broad‐spectrum antimicrobial and anti‐inflammatory activity, while generally maintaining low cytotoxicity towards human cells (Ciulla and Gelain 2023).

For the second generation of recombinant multidomain proteins, our findings reveal that using combinations of even alternating high/low amphipathic domains with a low docking score for the interacting domains permits designing antimicrobial agents with synergistic effects compared to mixtures of stand‐alone HDP domains against Gram‐positive strains. We also demonstrate that the order of these domains in the multidomain protein is crucial to act against Gram‐positive strains, preferably by positioning the most bioactive domain against the Gram‐positive pathogen at the ends. Finally, for Gram‐negative bacteria P. aeruginosa , the effect of domain order on multidomain peptides composed of the same HDPs is not decisive for its bioactivity; it is only necessary that the most active HDP against these pathogens is present. Altogether, we have presented a concise guide based on physicochemical properties and structural patterns that enable the design of antimicrobial agents using recombinant HDPs, and we have also demonstrated that multidomain proteins allow the construction of efficient combinations of HDPs, avoiding their production fused to nonfunctional carrier proteins.

Considering the growing role of artificial intelligence (AI) in the discovery and optimisation of antimicrobial peptides (AMPs) as novel antibiotics, our study offers new insights that could enhance the performance and interpretability of machine learning (ML) models in this field. Specifically, we demonstrate the value of incorporating structure‐dependent, computable physicochemical descriptors—such as lipophilicity, amphipathicity and charge distribution—which can serve as robust input features for classification and regression tasks. In addition, we emphasise the importance of integrating experimental minimum inhibitory concentration (MIC) data across a range of pathogens to improve biological relevance and predictive accuracy. This aligns with recent recommendations in the literature that highlight the need for standardised, high‐quality experimental data sets to train and validate AI‐driven AMP discovery frameworks (Brizuela et al. 2025).

5. Conclusions

The results support the suitability of the pH and context‐dependent approach to calculate the lipophilicity and amphipathicity of recombinant HDPs, aiming to establish successful structure–activity relationships that can be implemented in the improvement and design of new antimicrobial agents based on peptides. We found that for the first generation of antimicrobials, amphipathicity mainly explains the average antimicrobial activity against the Gram‐positive strains. In the case of the Gram‐negative bacteria, it depends on the quantity and the exposed area of the Ser and Thr amino acids. For the second generation of antimicrobials, the order of domains is crucial to act against Gram‐positive strains, preferably by positioning the most bioactive domain against the Gram‐positive pathogen at the ends. In consequence, our work proposes metrics based on physicochemical properties and structural patterns that regulate the rational design of new generations of proteins, particularly those that combine multiple HDP domains.

This study provides a comprehensive analysis of the structure–antimicrobial activity relationships of recombinant HDPs, guiding the rational design of next‐generation antimicrobials effective against both Gram‐positive strains MSSA, MRSA and MRSE as well as Gram‐negative bacteria MDR P. aeruginosa .

Future studies will address the implementation of the novel approach to calculate physicochemical descriptors such as lipophilicity and amphipathicity along with docking scores and structural patterns into machine learning models for the search of new bioactive peptides encoded in various proteomes and prioritise the exploration of these predicted bioactive domains against pathogenic bacteria.

Author Contributions

Sergi Travé‐Asensio: validation, methodology, formal analysis. Aida Tort‐Miró: methodology, validation, formal analysis. Silvana Pinheiro: methodology, validation, software, data curation. Elena Garcia‐Fruitós: conceptualization, investigation, funding acquisition, writing – original draft, methodology, validation, visualization, writing – review and editing, formal analysis, project administration, supervision, resources. Anna Arís: conceptualization, investigation, funding acquisition, writing – original draft, methodology, validation, visualization, writing – review and editing, formal analysis, project administration, supervision, resources. William J. Zamora: conceptualization, investigation, funding acquisition, writing – original draft, methodology, validation, visualization, writing – review and editing, software, formal analysis, project administration, resources, supervision, data curation.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: mbt270204‐sup‐0001‐Supinfo.docx.

Travé‐Asensio, S. , Tort‐Miró A., Pinheiro S., Garcia‐Fruitós E., Arís A., and Zamora W. J.. 2025. “Structure–Antimicrobial Activity Relationships of Recombinant Host Defence Peptides Against Drug‐Resistant Bacteria.” Microbial Biotechnology 18, no. 9: e70204. 10.1111/1751-7915.70204.

Data Availability Statement

All lipophilicity and amphipathicity tables presented here are hosted at: https://github.com/cbio3lab/SAR_RECOMBINANT_HDPs.