Hybrid derivative of cathelicidin and human beta defensin-2 against Gram-positive bacteria: A novel approach for the treatment of bacterial keratitis

Bacterial keratitis (BK) is a major cause of corneal blindness globally. This study aimed to develop a novel class of antimicrobial therapy, based on human-derived hybrid host defense peptides (HyHDPs), for treating BK. HyHDPs were rationally designed through combination of functional amino acids in parent HDPs, including LL-37 and human beta-defensin (HBD)-1 to -3. Minimal inhibitory concentrations (MICs) and time-kill kinetics assay were performed to determine the concentration- and time-dependent antimicrobial activity and cytotoxicity was evaluated against human corneal epithelial cells and erythrocytes. In vivo safety and efficacy of the most promising peptide was examined in the corneal wound healing and Staphylococcus aureus (ATCC SA29213) keratitis murine models, respectively. A second-generation HyHDP (CaD23), based on rational hybridization of the middle residues of LL-37 and C-terminal of HBD-2, was developed and was shown to demonstrate good efficacy against methicillin-sensitive and methicillin-resistant S. aureus [MIC = 12.5–25.0 μg/ml (5.2–10.4 μM)] and S. epidermidis [MIC = 12.5 μg/ml (5.2 μM)], and moderate efficacy against P. aeruginosa [MIC = 25-50 μg/ml (10.4–20.8 μM)]. CaD23 (at 25 μg/ml or 2× MIC) killed all the bacteria within 30 min, which was 8 times faster than amikacin (25 μg/ml or 20× MIC). After 10 consecutive passages, S. aureus (ATCC SA29213) did not develop any antimicrobial resistance (AMR) against CaD23 whereas it developed significant AMR (i.e. a 32-fold increase in MIC) against amikacin, a commonly used treatment for BK. Pre-clinical murine studies showed that CaD23 (0.5 mg/ml) achieved a median reduction of S. aureus bioburden by 94% (or 1.2 log10 CFU/ml) while not impeding corneal epithelial wound healing. In conclusion, rational hybridization of human-derived HDPs has led to generation of a potentially efficacious and safe topical antimicrobial agent for treating Gram-positive BK, with no/minimal risk of developing AMR.


Results
In vitro antimicrobial efficacy of HDPs. A total of 12 synthetic HDPs were rationally designed and synthesized based on the templates of native LL-37 and HBD-1 to -3 ( Table 1). The MIC values (in μg/ml and μM) of all the tested antibiotics, single and hybrid peptides, in the absence and presence of 150 mM NaCl are summarized in Table 2.
For reference and comparison purposes, the antimicrobial efficacy of the full-length sequence of LL-37 and HBD2 and HBD3 was first determined, though extensive examination was not performed in view of the lack of translational potential for clinical use (due to the cost of synthesis associated with long length of the peptides). LL-37 demonstrated moderate-to-good efficacy against methicillin-sensitive S. aureus [MSSA; MIC = 25.0 μg/ ml], methicillin-resistant S. aureus [MRSA; MIC = 25 μg/ml] and P. aeruginosa PAO1-L (MIC = 50 μg/ml) whereas HBD2 and HBD3 exhibited low efficacy (MIC = 100 μg/ml) against all three organisms. Subsequently, the efficacy of the three truncated versions of peptides (6-12 amino acids; based on the native template of HBD-2, HBD-3 and LL-37) and 6 first-generation hybrid HDPs (based on different combinations of LL-37, and HBD-1 to -3) were examined. All of them did not demonstrate any significant antimicrobial efficacy (MIC ≥ 200 μg/ml) against either Gram-positive or Gram-negative bacteria. Further 3s-generation peptides (derived from CaD2 sequence) were synthesized through rational modification: (1) CaD21: substitution of cysteine with alanine; (2) CaD22: substitution of phenylalanine with tryptophan (to increase hydrophobicity); and (3) CaD23: substitution of phenylalanine and proline with tryptophan (to further enhance hydrophobicity) and substitution of cysteine with leucine. CaD21 and CaD22 demonstrated slight improvement in the antimicrobial efficacy whereas CaD23 exhibited good antimicrobial efficacy against MSSA (MIC = 12.5-25.0 μg/ml), (MRSA; MIC = 25 μg/ml) and methicillin-sensitive Staphylococcus epidermidis [MSSE; MIC = 12.5 μg/ml], highlighting the importance of increased hydrophobicity for antimicrobial efficacy ( www.nature.com/scientificreports/ NaCl), the MIC of CaD23 against MSSA and MRSA increased by two to fourfold, remained unchanged for MSSE, and remained relatively stable (one to twofold increase) against P aeruginosa.
In vitro cell viability and cytotoxicity. Amikacin and CaD2 did not demonstrate any lethal effect on HCE-2 cell viability at 200 μg/ml (Fig. 1A). The IC 50 (concentration that inhibits 50% of the cell viability) of CaD21, CaD22 and CaD23 were > 200 μg/ml, > 200 μg/ml and 54.6 ± 11.7 μg/ml, respectively. This demonstrated that the increased hydrophobicity with tryptophan residues enhanced the antimicrobial efficacy of CaD23 but with increased negative effect on the cell viability of HCE-2 cells. In terms of cytotoxicity, amikacin and CaD2 did not show any sign of toxicity for HCE-2 cells at 200 μg/ml and CaD23 showed 30.4 ± 7.8% cytotoxicity at 200 μg/ml (Fig. 1B). The LC 50 (concentration that kills 50% of the cells) of CaD23 was > 200 μg/ml, which yielded a therapeutic index (defined by LC 50 divided by the MIC value) of > 8 for treating S. aureus ATCC29213. Hemolytic assay demonstrated minimal (7.1 ± 3.0%) hemolytic activity of CaD23 at 200 μg/ml (Fig. 2). A summary of the cell viability, cytotoxicity and hemolytic results of CaD23 is provided in Table 3. In addition, a summary of the therapeutic index of CaD23 (defined as IC 50 , LC 50 or HC 50 divided by the MIC value) is provided in Table 4.
Time-kill kinetics. Time-and concentration-dependent antimicrobial activity of CaD23 was determined against MSSA (SH1000) using time-kill kinetics assay. When CaD23 was used at 2× MIC (25 μg/ml) against SH1000, it was able to achieve significant killing (99.9% or 3 log 10 CFU/ml reduction) with 15 min and complete killing (100%) within 30 min of treatment (Fig. 3). In contrast, amikacin (used at 20× MIC; 25 μg/ml) could only achieve significant and/or complete killing (99.9-100%) of SH1000 within 4 h of treatment, which was 8 times slower than CaD23. The antimicrobial efficacy of CaD23 (25 μg/ml) and amikacin (both 10 μg/ml and 25 μg/ml) were maintained at 24 h' time-point, with no evidence of bacterial re-growth. Similar findings were observed when CaD23 and amikacin were tested against SH1000 in physiological tear salt concentration (150 mM NaCl). Multipassage antimicrobial resistance (AMR) study. Multipassage AMR assays were performed to evaluate the risk of development of AMR of S. aureus ATCC SA29213 against CaD23 and amikacin over 10 consecutive passages (days). S. aureus did not develop any AMR against CaD23 after 10 consecutive passages whereas it developed significant AMR against amikacin, with a fourfold increase in the MIC after the 2nd passage and a 32-fold increase in the MIC after the 10th passage (Fig. 5). This highlights the therapeutic potential of CaD23 as novel antimicrobial agent in the era of emerging AMR.
In vivo safety of CaD23. In view of the known discrepancy between in vitro and in vivo results, the safety of CaD23 was further determined in a murine corneal epithelial wound healing model. When compared to the PBS group, both CaD23 0.03% (or 300 μg/ml) and CaD23 0.05% (or 500 μg/ml) groups did not show any significant difference in the rate of corneal re-epithelialization (all healed within 2-3 days), suggesting that both concentrations were safe for topical application (Fig. 6). However, significant delay in corneal re-epithelialization was observed in the CaD23 0.1% group, with a mean wound size of 28.5 ± 19.9% at day 3 post-injury (p = 0.004).
In vivo efficacy of CaD23. Based on the in vivo safety results, the highest tolerable concentration of CaD23 0.05% was subjected to subsequent in vivo efficacy testing in a murine S. aureus ATCC 29213 keratitis model. As the data was not normally distributed and was log-transformed, the results were reported in median ± inter-  www.nature.com/scientificreports/ quartile range (IQR). When compared to the S. aureus bacterial viability in the PBS group (4.2 ± 1.3 log 10 CFU/ ml), there was a considerable reduction of bacterial viability in the CaD23 0.05% and levofloxacin 0.5% groups by 94% (3.0 ± 2.4 log 10 CFU/ml; p = 0.72) and 98% (2.4 ± 2.2 log 10 CFU/ml; p = 0.08), respectively, though statistical significance was not achieved due to wide standard deviations ( Fig. 7). At day 3 post-treatment, the clinical appearance of the corneas in the CaD23-treated group (1.2 ± 0.45; p = 0.07) and the levofloxacin-treated group (1.0 ± 0.71; p = 0.03) was substantially better than the untreated group (2.2 ± 0.84), based on the ocular clinical scoring (Fig. 7).

Discussion
IK represents a persistent and uncurbed burden on human health at a global level. Topical antibiotics are the current mainstay of treatment for BK but the efficacy is being increasingly challenged by the emergence of AMR 1,2,14 .
In addition, patients with BK often require long duration of treatment and sight-threatening complications may still ensue despite timely intervention 11 , highlighting an unmet need for newer and better treatment. HDPs serve as an attractive class of antibiotics for treating IK based on the following reasons: (1) the broad-spectrum activity of HDPs can provide comprehensive coverage to a wide range of microorganisms, particularly when mixed infection is relatively common (5-20%) in IK 1 ; (2) the rapid antimicrobial action help reduce the microbial load and limit the damage to the cornea more effectively, ultimately have a better chance of preserving the vision, as well as reducing the risk of developing AMR, which is currently an emerging issue in ocular infection 14 ; and  www.nature.com/scientificreports/ > 200 26.6 (6.5) 54.6 (11.7) 69.6 (7.8) > 200 7.1 (3.0) www.nature.com/scientificreports/ (3) HDPs can be used as synergistic or additive agents to the current conventional antibiotics, enhancing the therapeutic index by increasing the antimicrobial efficacy and reducing the dose-related toxicity 42,43 .
In this study, we highlight a body of work in designing and developing human-derived hybrid HDP as topical treatment for BK. Hybridization strategy has been previously employed by several research groups to improve the therapeutic index of HDPs 39,40,44 but the strategy of using human-derived hybrid HDPs (with LL-37 and HBD-2) is first of its kind. The initial concept of developing synthetic HDPs derived from LL-37 and HBD was founded on the observation of the upregulation of these key HDPs at the ocular surface during IK 23,27,33,45 . Furthermore, LL-37 and HBD-2 and -3 were shown to exhibit good antimicrobial efficacy against a range of organisms 34,45 . We have selected the mid-region of LL-37, consisting of residues 18-29 of LL-37 (i.e. KR-12) as part of the hybrid template as studies have demonstrated that KR-12, though much shorter than parent LL-37, exhibited similar antimicrobial activity against E. coli (MIC = ~ 64 μg/ml) with reduced toxicity to host cells 46 . C-terminal of HBD1-3 was used as the other part of the hybrid template in view of its rich content of cationic residues, which have been shown to play an important role in interacting with anionic bacterial membrane and killing of bacteria 47,48 .
Our initial attempt of engineering single linear HDPs (including LL-37 and HBD-2 and -3) and hybrid HDPs (based on LL-37 and HBD-1 to -3) did not yield any compound with good antimicrobial efficacy. However, with systematic SAR analysis and modification, particularly through substitution of proline and phenylalanine with tryptophan residues (to increase hydrophobicity and membrane partitioning), ensued the development of peptide with enhanced antimicrobial efficacy against MSSA (MIC = 12.5-25 μg/ml) and MRSA (MIC = 25 μg/ml), serving as a proof-of-concept of this novel design strategy. Interestingly, increased hydrophobicity augmented the Phosphate buffer solution (PBS) group serves as the untreated control. "0 min" represents the starting inoculum, which is around 6 log 10 CFU/ml. The red dotted horizontal line at ~ 3 log 10 CFU/ml signifies the threshold of significant bacterial killing (defined as 99.9% or 3 log 10 CFU/ml reduction of the bacterial viability compared to the starting inoculum). Data is presented as mean ± standard deviation (depicted in error bars) of two independent experiments performed in biological duplicate. CaD23 (4× MIC) was able to achieve complete (100%) killing of SH1000 within 15 min of treatment whereas amikacin (8× MIC) was only able to achieve complete killing of SH1000 within 4 h of treatment. The antimicrobial efficacy of CaD23 and amikacin was maintained at 24 h post-treatment.

Figure 5.
Multipassage antimicrobial resistance (AMR) assays for CaD23 and amikacin against S. aureus ATCC SA29213 over 10 consecutive passages (days). S. aureus did not develop any AMR against CaD23 after 10 passages whereas it developed significant AMR against amikacin, with a fourfold increase in the MIC after the 2nd passage and a 32-fold increase in the MIC after the 10th passage. Data is presented as mean ± standard deviation (depicted in error bars) of two independent experiments performed in biological duplicate. Some error bars are missing due to small or no standard deviation values. www.nature.com/scientificreports/ antimicrobial activity of HDP against Gram-positive bacteria more than Gram-negative bacteria. This is likely attributed to the different compositions in the bacterial membrane between Gram-positive bacteria, which consist of a thick peptidoglycan layer, and Gram-negative bacteria, which comprises an additional outer membrane with abundance of negatively charged lipopolysaccharide for which cationicity of the peptide plays a more important role 49 . In addition, the efficacy of CaD23 (18 amino acids in length) is at least equal to or stronger than the fulllength parent LL37 (MIC = 25-50 μg/ml), HBD-2 (MIC = 100 μg/ml) and HBD-3 peptides (MIC = 50-100 μg/ ml). Peptides with shorter sequence not only have the advantage of lower manufacturing cost but may also have  www.nature.com/scientificreports/ a lower risk of inducing immunogenicity 25,50 . Moreover, the antimicrobial efficacy of CaD23 is comparable to some of the HDPs that are developed for ocular surface infection, including esculentin-1a (1-21)NH2 51 , RP444 52 , melimine and its derivatives 40,53 , and ε-lysylated Mel-4 54 .
In the time-kill kinetics study, we demonstrated that CaD23 at 2× MIC was able to achieve complete killing of S. aureus within 30 min as compared to 4 h with amikacin at 20× MIC (i.e. 8 times slower than CaD23). We also observed that increasing the concentration of amikacin from 8× MIC to 20× MIC did not expedite its antibacterial action against S. aureus, suggesting that the speed of bacterial killing is more related to the underlying mechanism of action than the concentration of antibiotic. The rapid killing action of CaD23 is likely related to a membrane perturbation effect 55 , which is in contrast to amikacin where it exerts its anti-bacterial activity via intracellular inhibition of the 30S subunit of bacterial ribosome 56 . The rapid and membrane disruptive action of CaD23 is likely accountable for the low risk of S. aureus developing AMR against CaD23 as the bacteria has less time to adapt and require substantial modification of genome to develop effective resistant mechanisms, which are shown in conventional antibiotics such as amikacin.
Based on the LDH cytotoxicity assay, CaD23 was shown to be relatively non-cytotoxic (20-30% toxicity) at 200 μg/ml, with a therapeutic index of > 8-16 against S. aureus (defined by LC 50 concentration divided by MIC value). In addition, the hemolytic activity of CaD23 at 200 μg/ml was only less than 10%. Interestingly, the IC 50 value (based on cell viability assay) was around 50 μg/ml, considerably lower than the LC 50 value. This discrepancy may be due to the fact that some of the cells were metabolically inhibited but the cell membrane was not disrupted, therefore the IC 50 value was lower than the LC 50 value. That said, our in vivo corneal epithelial wound healing study showed that CaD23 did not demonstrate any significant toxicity when used at 500 μg/ml (or 0.05%), which was 10 times higher than the IC 50 value. The observed discrepancy between in vitro and in vivo findings is likely attributed to the inherent dynamic environment of ocular surface with eye blinking, high tears turnover, and drainage, and dilution of the peptide by the tears 57 . This also suggests that in vitro plate-based static assays such as cytotoxicity or cell viability assays could potentially overestimate the in vivo cytotoxicity of drugs that are developed for ocular topical application. The shortcoming of in vitro assays may be addressed by the recently developed novel ex vivo biomimetic model where it could simulate the dynamic and complex interface between the ocular surface and the external environment, allowing a better prediction of the in vivo effect 58 .
Our in vivo BK study demonstrated that CaD23 0.05% (20× MIC) was able to reduce the bacterial viability of S. aureus by 94% when compared to the untreated control group, which was more than the pre-defined endpoint of the study (i.e. 1 logCFU or 90% reduction in the bacterial bioburden). However, the effect was not statistically significant due to the considerably wide standard deviation observed in both the treatment and the control groups. This similarly explained the insignificant improvement in the levofloxacin-treated group, though there was 1.8 logCFU median reduction in the bacterial bioburden compared to the PBS group. Nevertheless, these results serve as a strong proof-of-concept that CaD23 may be employed as a potentially efficacious treatment for treating Gram-positive bacterial keratitis, but a larger sample size will be required to fully ascertain its efficacy. In addition, we had chosen to sacrifice the mice at day-3 post-infection because S. aureus keratitis had been shown to be most severe in C57BL/6J mice at 3-day post-infection, as compared to other stains of mice such as BALB/c and A/J mice. In addition, S. aureus keratitis has been shown to improve spontaneously in C57BL/6J mice at day-5 post-infection and beyond 59 . To investigate the longer-term in vivo antimicrobial efficacy of CaD23 in S. aureus keratitis, other strains of mice may be required to examine this aspect.
One potential approach to improve the therapeutic effect of HDP-derived antimicrobial treatment is to use them in combination with antibiotics as peptide-antibiotic synergism has been demonstrated in several studies 42,43,60 . This attractive antimicrobial strategy not only helps extend the lifespan and broaden the antibacterial spectrum of conventional antibiotics, but also reduce the dose-dependent toxicity associated with HDPs and antibiotics 61 . Recently, our group has also shown that FK16, a cathelicidin-derived molecule, could improve the antimicrobial efficacy of vancomycin, a glycopeptide antibiotic, against PA by eightfold 42 . This is likely ascribed to the different mechanisms of action between FK16 and vancomycin where the membrane disruptive action of FK16 facilitates the diffusion of vancomycin across the bacterial membrane. Hence, future work investigating the potential synergism between CaD23 and commonly used antibiotics for BK would be useful.
At present, our aim is to develop the human-derived hybrid HDPs as topical treatment for BK. However, it is interesting to note that the hemolytic activity of CaD23 was very low (7.1% at 200 μg/ml), suggesting that it can potentially be developed for treating systemic infection. It also reduces the concern of systemic toxicity when it is applied to the ocular surface because systemic absorption of ocular topical treatment can occur, particularly when the drugs are hydrophobic 57 . The considerable disparity between the hemolytic activity and cytotoxicity against corneal epithelial cells could be attributed to the difference in the membrane structure (particularly the cytoskeleton) and the metabolic structures of the non-nucleated cells (e.g. red blood cells) and nucleated cells (e.g. corneal epithelial cells) 62 . Further studies examining the efficacy and stability of CaD23 in serum as well as its interaction with blood proteins will be conducted to determine its therapeutic potential for treating systemic infection.
In conclusion, we demonstrated that rational hybridization of LL-37 and HBD-2 serves as a useful strategy in translating the therapeutic potential of human-derived HDPs. Future work examining the efficacy and safety of combined CaD23-antibiotic therapy would be beneficial. Potential strategies such as N-and C-terminal modifications, introduction of unnatural amino acids and nanoformulation of CaD23 will be further explored with an aim to enhance the antimicrobial efficacy, reduce the toxicity and improve the stability of the peptide 25

Materials and methods
The commercially synthesized peptides and commonly used antibiotics for IK were first examined for their in vitro antimicrobial efficacy against a range of bacteria. In vitro cytotoxicity of these antimicrobial agents was then determined against human corneal epithelial cells (HCE-2, CRL-11135, ATCC, UK) using cell viability assay and cytotoxicity assay, and against human erythrocytes using hemolytic assay. The most promising synthetic peptide, CaD23, was further examined for its time-and concentration-dependent in vitro antimicrobial activity. Finally, in vivo efficacy and safety of CaD23 were evaluated in corneal wound healing and S. aureus keratitis murine models. All the assays described in this study were conducted in biological duplicate and in at least two independent experiments, with appropriate positive controls (PCs) and negative controls (NCs). Continuous values were expressed in mean ± standard deviation (SD), unless specified otherwise.
Design and synthesis of HDPs. A template-based design method was used to design our human-derived HDPs. The native peptide sequences were obtained from an established protein bank database (https:// www. unipr ot. org/). Several human-derived HDPs, specifically HBD-1, -2, and -3, and cathelicidin (LL-37), were subject to testing and hybridization. The physicochemical properties of the designed peptides, including peptide weight, net charge, hydrophobicity (<H>), and amphiphilicity/hydrophobic moment (<μH>, were analyzed using a computational programmes such as PepCalc (https:// pepca lc. com/) and HeliQuest (http:// heliq uest. ipmc. cnrs. fr). Various strategies, including residue substitution and hybridization, were employed for SAR analysis and for improving the therapeutic index (i.e. increasing the antimicrobial efficacy and reducing the toxicity profile). The native and synthetic peptide sequences are shown in Table 1. Synthesis of the single and hybrid HDPs was based on the knowledge of the functional regions of the native templates. Three single, short-sequenced (truncated) peptides, based on the native template of HBD2, HBD3 and LL-37, were first generated. Truncated versions of the parent peptides were engineered as this strategy has been shown to serve as a useful method in improving the efficacy of peptides and reducing the cost of synthesis, which represents a significant translational barrier of peptide-based antimicrobial therapy 25 . The C-terminal region of HBD2 and HBD3 was synthesised and examined in view of the presence of high cationicity (i.e. rich of lysine and/or arginine residues), which are important for the antimicrobial efficacy 26,47,48 . In addition, the middle region of LL-37, same as the KR12 molecule, was synthesized as it has been shown to exhibit efficacy equivalent to the full-length of LL-37 46 .
All antibiotics, including amikacin (an aminoglycoside) and levofloxacin (a fluoroquinolone), were purchased from Sigma-Aldrich, United Kingdom. Both antibiotics were used as positive controls as they were commonly used for treating BK 1,4,63 . The full-length peptides were commercially produced by Anaspec (Cambridge, UK; for LL-37) and PeproTech (London, UK; for HBD-2 and HBD-3). All other peptides were commercially produced by Mimotopes (Mulgrave Victoria, Australia) via the traditional solid phase Fmoc synthesis method. All the synthetic peptides were purified by reverse-phase high performance liquid chromatography (RP-HPLC) to > 95% purity and characterized by mass spectrometry. Determination of antimicrobial efficacy using MIC assay. In vitro antimicrobial efficacy of the antibiotics and designed HDPs was determined using an established MIC assay with broth microdilution method approved by the Clinical and Laboratory Standards Institute (CLSI) 64 . Briefly, the microorganisms were cultured on Tryptone Soya Agar (TSA) and incubated overnight for 18-21 h at 37 °C. Bacterial inoculums were subsequently prepared using the direct colony suspension method 64 . Three to five bacterial colonies were obtained from the agar plate and inoculated into an Eppendorf tube containing 1 ml of cation-adjusted Muller-Hinton broth (caMHB), consisting of 20-25 mg/L calcium ions (Ca 2+ ) and 10-12.5 mg/L magnesium ions (Mg 2+ ). The bacterial suspension was adjusted to achieve a turbidity equivalent to 0.1 OD 600 or 0.5 MacFarland, containing ~ 1.5 × 10 8 colony-forming unit (CFU)/ml, which was then further diluted in 1:150 in caMHB to reach a final bacterial concentration of ~ 1 × 10 6 colony forming units (CFU)/ml. Each treatment (peptide or antibiotic) was prepared in 1:2 serial dilution in 96-well polypropylene microplates (with a final treatment volume of 50 μl per well), followed by the addition of 50 μl of 1 × 10 6 CFU/ml bacteria into each well (with a final bacterial concentration of 5 × 10 5 CFU/ml). As the HDPs are known to be influenced by the salt content 25 , the MIC assay was also performed in the presence of physiological tear salt concentration (150 mM NaCl). The MIC values, defined as the lowest concentration of the antimicrobial agent that prevented any visible growth of bacteria, were determined after 24 h of incubation with treatment.

Range of microorganisms being tested.
Cell viability and cytotoxicity assays. Cell viability and cytotoxicity of antibiotics and peptides were determined against human corneal epithelial cells (HCE-2, CRL-11135, ATCC, Manassas, Virginia, USA) using cell-counting-kit-8 (CCK-8) assay (Sigma Aldrich, Merck Life Science UK Limited, Dorset, UK) and lactate dehydrogenase (LDH) assay (ThermoFisher Scientific, UK), respectively, as per manufacturer's guidelines. HCE-2 cells were cultured and seeded into a 96-well plate at a density of 7.5 × 10 3 cells/well and allowed to attach overnight and grew to 80-90% confluency in keratinocyte serum free medium (KSFM) supplemented with human recombinant epidermal growth factor, bovine pituitary extract, hydrocortisone, and insulin. Once www.nature.com/scientificreports/ the confluency reached 80-90%, the cells were incubated with treatment for 3 h before OD 450 measurement was taken using BMG Clariostar microplate reader (BMG Labtech Ltd., Aylesbury, UK). Appropriate controls were used, including 0.1% Triton X-100 as positive control and KSFM as negative control. Cell viability was calculated using the following formula: [((I treatment -I NC )/(I NC )) × 100; I = intensity] and cytotoxicity was calculated using the following formula: [((I treatment -I NC )/(I PC -I NC )) × 100].
Hemolytic assay. Ethical  In vivo safety of CaD23 (in 0.03%, 0.05% or 0.1%), and PBS (negative control) was first determined in a mouse corneal epithelial wound healing model. In the absence of in vivo pilot data of our designed HDPs, the sample size was calculated based on a previous study 67 . This was designed as a non-inferiority trial to ensure that the HDPs did not affect the wound healing when compared to PBS (control). The non-inferiority margin was set at 10% difference of the wound size between HDPs and PBS at 3 days (deemed as significantly different), with a standard deviation (SD) of 5% (Cohen's d = 2.0), power = 80% and p < 0.05. A minimum sample size of 4 mice/ treatment group was needed.
All mice were randomly allocated to each of the four treatment groups (n = 4 mice/group). Prior to the study, all eyes were examined with slit-lamp biomicroscopy to confirm the health of corneas. Under general and topical anesthesia, the central 2 mm corneal epithelium was gently debrided with sterile Beaver mini-blades, leaving the www.nature.com/scientificreports/ basal lamina intact. Each treatment was applied immediately after wounding, then 4 times a day at 3-h interval for 3 days (total dose of treatment per mouse = 14). Corneal epithelial defect was assessed using a cobalt-blue filter-equipped slit-lamp biomicroscopy and photography with staining with topical sodium fluorescein 1% at baseline (immediately post-debridement) and daily up to 3 days post-treatment. Fluorescein-stained images of the corneal wound defect were analyzed using ImageJ software (https:// imagej. nih. gov/ ij/) 68 . The main outcome measure was the wound size at the end of day 1, 2 and 3 [expressed as % of the original wound size in mean and standard deviation (SD)]. Difference in the wound size between groups was analyzed using one-way ANOVA with Dunnett's post hoc test (PBS as the control group).
In vivo S. aureus keratitis study. Based on the in vivo safety data, the highest tolerable concentration of CaD23, 500 μg/ml (0.05%), was used in the subsequent S. aureus keratitis murine model. Levofloxacin 0.5%, a commonly used antibiotic for BK in clinical setting 4 , and PBS were used as the positive and negative controls, respectively. In the absence of in vivo pilot data, the sample size was calculated based on a previous study 65 . To detect an effect size of 1 LogCFU (or 10 times) difference in the bacterial load (significant antimicrobial efficacy) between HDPs (mean = 5 logCFU) and PBS (mean = 6 logCFU; NC), with a SD of 0.5 logCFU (Cohen's d = 2.0), power = 80% and p < 0.05, a minimum sample size of 4-5 mice/group is required. All mice were randomly allocated to each treatment group (n = 5 mice/group). Slit-lamp examination was performed before the start of experiment to confirm the health of corneas. Under general and topical anesthesia, the central 2 mm corneal epithelium was gently removed with sterile Beaver mini-blades. 10 μl of ~ 1 × 10 8 CFU/ ml of ATCC SA29213 was applied topically onto the cornea and the lid was held shut for 1 min. At 6 h postinfection, 10 μl of treatment was applied directly onto the infected corneas with a dose regimen of 4 times a day at 3-h interval for 3 days (total dose of treatment per mouse = 12). The eyes were monitored daily using slit-lamp biomicroscopy and photography. Ocular clinical scoring was adapted from a previous method described by Clemens et al. 52 with minor modifications: (a) 0: Clear cornea or minimal opacity, partially covering the pupil; (b) + 1: Mild opacity, partially/fully covering the pupil; (c) + 2: Dense opacity, partially covering the pupil; (d) + 3: Dense opacity fully covering the pupil; and (e) + 4: Corneal perforation or phthisis.
At the end of day 3, all animals were sacrificed, and the infected eyes were enucleated. The whole corneas were subsequently dissected and homogenized in 1 ml of sterile PBS using sterile glass micro-beads. The homogenised infected corneal tissue suspension was serially diluted in 1:10 and plated on TSA plates in triplicates for enumeration of CFU after 24 h incubation at 37 °C. The main outcome measures were the ocular clinical scoring and the residual bacterial load at 3-day post-treatment (expressed as log 10 CFU/ml, which was the same as log 10 CFU/cornea) and the difference among groups was analyzed using one-way ANOVA with Dunnett's post hoc test (PBS as the control group). www.nature.com/scientificreports/