Antibiotics, Vol. 11, Pages 1710: Associating Biological Activity and Predicted Structure of Antimicrobial Peptides from Amphibians and Insects

1. IntroductionAntimicrobial resistance is an escalating global health concern, with multiple infectious diseases becoming increasingly difficult and expensive to treat. An estimated 1.27 million deaths occurred in 2019 due to bacterial resistance to antibiotics [1], with the World Health Organization (WHO) stating that this number is expected to exceed 10 million by 2050 [2]. Antimicrobial resistance (AMR) develops through mutations in the bacterial genome as well as horizontal transfer of mobile genetic elements such as plasmids [3]. AMR is exacerbated by the routine administration of antibiotics both clinically and in agricultural settings [4]. Despite the steady increase in antibiotic resistant bacteria [5], there is a shortfall of novel antibiotics being developed, with no new classes approved for clinical use since the 1980s [6]. This gap between increasing antibiotic resistance and lagging discovery of new drug classes creates an urgent need for innovative therapeutic discovery methods to produce novel antimicrobials with different mechanisms of action to combat the emerging threat to public health [7].Antimicrobial peptides (AMPs) are short, often cationic and amphipathic, biomolecules that are produced by the innate immune system of all living organisms [8]. They are a functionally and structurally diverse group of compounds that can defend against bacteria, viruses, fungi, and cancer [9]. AMPs can combat bacterial infections directly through interactions with the negatively charged membrane and with intracellular targets such as DNA and RNA [9,10]. In addition, AMPs can function indirectly through inflammatory or immunomodulatory pathways, resulting in the recruitment of immune cells [11]. Because of these diverse mechanisms of action, it has been suggested that it is more difficult for bacteria to develop resistance to AMPs as compared to conventional antibiotics [12,13]. However, bacteria intrinsically resistant to AMPs do exist [14], and resistance to colistin, an AMP-based therapeutic used as a last resort, has been observed [15]. Continuing the search for and characterization of AMPs with varying structure and associated mechanisms of action would increase our arsenal of available therapeutics against drug resistant bacteria.The three-dimensional (3D) structures of AMPs can be classified into several types; α-helical, β-sheet containing, mixed, or linear extended structures [16]. It has been observed that many α-helical and linear extended peptides undergo conformational changes when interacting with the bacterial membrane [17]. The α-helix structure is reported to be the most effective conformation for AMPs to interact with the bacterial membrane, with size, sequence, charge and hydrophobicity all affecting the spectrum and level of the resulting antimicrobial activity [18]. However, the majority of discovered AMPs do not have a known structure, with approximately 25% of all entries in the Antimicrobial Peptide Database (APD3) reporting an experimentally determined structure [19]. Traditional methods for determining the structure of peptides, such as nuclear magnetic resonance (NMR), X-ray crystallography, and cryo-electron microscopy are laborious and expensive in comparison to in silico methods. While the latter only provide predictions, recent advances in artificial intelligence research bring the potential to make them enabling tools for characterizing AMPs. AlphaFold2 is a model that applies deep-learning to predict the 3D structure of proteins with high accuracy, including cases where there are no similar structures [20]. ColabFold has been reported to improve the speed of this prediction by coupling AlphaFold2 with an MMseqs2-based homology search [21].Here, we studied the relationship between predicted AMP structure and observed antimicrobial activity. Specifically, we used ColabFold to predict the 3D structure of a list of 88 putative amphibian and insect AMPs discovered using rAMPage, a homology-based bioinformatic pipeline for in silico discovery of AMPs using RNA-seq reads [22]. We examined the antimicrobial activity of these peptides against two Gram-negative and one Gram-positive bacteria in the WHO’s list of priority pathogens [23]. We observe 51 AMPs in this list, discovered from 16 amphibian and 16 insect species, with antimicrobial activity and low toxicity to porcine erythrocytes and human embryonic kidney cells. Amphibians possess a diverse repertoire of AMPs, which provide a first line of defense as they transition from an aquatic to a terrestrial habitat throughout their life cycle [24]. Further, unlike vertebrates, insects lack an adaptive immune system, and as such their innate immune system, including AMPs, is their only line of defense in combating bacterial infections [25]. Studying these peptides, we report an association between AMPs with a predicted helical structure and broader antibacterial activity against the bacterial isolates of our panel. 3. DiscussionIn the present study, we characterized 51 AMP candidates originally derived from a bioinformatics scan of RNA-seq data from amphibians and insects using rAMPage [22]. Fourteen of these had high antimicrobial activity to at least one of the bacterial species tested. The most bioactive amphibian AMPs, PeNi1, OdMa2 and PeNi9, and the insect peptide TeBi1 were active against all three bacteria. TeRu4 and PaVa1 both demonstrated selectivity for the Gram-negative species tested, with TeRu4 having higher antimicrobial activity against both strains.The AMPs reported herein had low to no toxicity, with only six peptides having hemolytic activity at the highest concentrations tested. This observation is consistent with previous research that suggests that AMPs tend to have selectivity for microbial cells over eukaryotic cells due to differences in membrane composition [31]. Despite this selectivity, some AMPs do disrupt mammalian membranes and cell processes [32]. In the reported list of AMPs, PeNi1, OdMa2, PeNi9 and TeBi1 only displayed high levels of cytotoxicity at concentrations higher than their MICs. The ratio of the toxicity to the MIC of the peptides is described using the selectivity index, with a larger selectivity index indicating preference for bacterial cells [33,34] (Table S3). TeBi1 displayed the most selectivity of these, as it was not toxic to over 50% of cells until 16 to 128-fold its MIC. TeRu4 and PaVa1 were not toxic against either porcine erythrocytes or HEK293 cells. The selectivity indices of TeBi1, TeRu4 and PaVa1 make them good candidates for therapeutic development.The AMPs characterized here vary in their amino acid sequences, with 30 active peptides belonging to four AMP families. AMPs from the Ranidae family can be classified into 14 peptide families based on their amino acid sequence, with AMPs from the same peptide family thought to share a common evolutionary origin [35]. We identified AMPs from three of these families: Brevinin-1, Brevinin-2 and Nigrocin-2 using the families’ signatures. Brevinin-2 peptides are the longest of the three, with a mature peptide length of 33–34 residues [36]. Brevinin-1 and Nigrocin-2 peptides are shorter, with an approximate length of 24 and 21 amino acids, respectively [36]. However, there is considerable variation among individual family members [36]. AMPs from these three families share some common characteristics: presence of a C-terminal disulfide-bridged cyclic heptapeptide, otherwise known as the ‘Rana box’, and antimicrobial activity against both Gram-negative and Gram-positive bacteria [36,37]. Additionally, Brevinin-1, Brevinin-2 and Nigrocin-2 peptides have been found to adopt an amphipathic α-helical conformation in membrane-like environments [36,37,38]. All peptides we identified as members of these families included the Rana box and were predicted to have a helical or mainly helical structure except for the Nigrocin-2 peptide OdMa4. OdMa4 was the shortest of the Nigrocin-2 peptides and terminated after the first cysteine of the Rana box. Additionally, we identified two peptides belonging to the Apidaecin AMP family. Apidaecins are short, proline-rich AMPs produced by insects [30]. These peptides do not typically form α-helices or β-strands [30], as was seen with ApCe1 and PaVa1, which were predicted to have linear extended structures. Similar to other Apidaecins, both ApCe1 and PaVa1 showed antimicrobial activity against Gram-negative bacterial species but not Gram-positives [30].Most of the validated AMPs were predicted to adopt a helical structure by ColabFold, despite helical structures only making up half of the peptide set. In contrast, predicted linear extended peptides made up a smaller proportion of the antimicrobially active peptides compared to their abundance in the overall set. Note that, while there was a tendency of peptides with a predicted helical structure to be bioactive, not all of the helical peptides were active, and vice versa. In addition, we observed that while there were peptides from all structural categories that were active against Gram-negative species, only peptides that had some predicted helical content were active against both Gram-positive and Gram-negative bacteria. AMPs that form α-helices make up the majority of AMPs with known structure, and it is thought to be the most effective structure rendering antibacterial activity against the bacterial membrane [18]. However, this also biases machine learning algorithms such as the one we used to discover the tested peptides in favour of peptides with α-helix structures.The majority of known AMPs do not have an experimentally validated structure [19]. Determination of peptide structures is often accomplished with NMR, X-ray crystallography or cryo-EM; however, this is time consuming and costly in comparison to in silico techniques. In contrast, one can predict the secondary and tertiary structures in a high-throughput manner based on the amino acid sequences of peptides of interest using a tool like ColabFold [21] prior to synthesis and testing. The association between ColabFold predicted structures and antimicrobial activity identified here can be informative in selecting peptides identified in silico as putative AMPs for in vitro validation. Further, AMPs are often chemically modified to improve their stability and bioavailability. The insights into the predicted structure-function relationships identified here could also be used to investigate how chemical modifications may influence antimicrobial activity.One of the limitations of using rAMPage for peptide discovery is that the pipeline uses RNA-seq reads, and thus does not detect post-translational modifications such as amidation. Amidation of the C-terminus is a common post-translational modification of AMPs [39], and can impact the antimicrobial activity and toxicity of peptides [40]. PeNi12 for example, has 100% sequence identity to the mature region of ranacyclin-N from P. nigromaculatus, however PeNi12 was non-inhibitory against our panel of bacteria at all concentrations tested. Ranacyclins have amidated C-termini [41], so the non-amidated carboxyl end may have contributed to the lack of antimicrobial activity of PeNi12. TeBi1 has 100% sequence identity to the C-terminus region of the propeptide of bicarinalin and includes the mature sequence of bicarinalin, however the last three amino acids at the C-terminus of bicarinalin precursor are cleaved off and the peptide is amidated [42]. Additionally, as the peptides were only tested in vitro, we cannot conclude that the peptides that did not exhibit antimicrobial activity are not AMPs as they may have immunomodulatory properties or act on other bacterial strains not tested in the present study. The AMPs identified here may be tested in vivo to determine if they interact with the immune system to combat infections within the host.While ColabFold determines protein structures in silico with high accuracy, peptides in their biological environments are flexible and may adopt different conformations [17]. AMPs that adopt a helical structure at the membrane are often disordered in aqueous environments [17]. Thus, the predicted structure may not be representative of the 3D structure of the peptide when interacting with specific targets. Additionally, AMPs have diverse sequences and targets. Peptides that act on the bacterial membrane may have different structural characteristics compared to those that mainly interact with intracellular components. Future investigations into the mechanism of action of these AMPs would shed light into how the predicted secondary structure impacts their biological function. The bioinformatic characterization of these AMPs is a cost-effective initial step in studying their mechanisms of action.

AMPs are a promising alternative to conventional small molecule antibiotics as they represent a diverse group of molecules with a wide variety of physiochemical properties. Here, we present the discovery and characterization of structurally and functionally diverse AMPs with potent antimicrobial activity and low toxicity. We also demonstrate the utility of predicting the structure of AMPs and report a significant association between peptides predicted to adopt a helical conformation and observed antimicrobial activity (p = 6.6 × 10−5). The potential of structural prediction to prioritize putative AMPs is an exciting avenue for discovery of new therapeutics in our fight against bacterial infections.

4. Materials and Methods 4.1. Peptide Discovery Using rAMPagePeptide sequences were discovered using the Rapid Antimicrobial Peptide Annotation and Gene Estimation (rAMPage) pipeline v1.0 [22], which is publicly available on Github (https://github.com/bcgsc/rAMPage, accessed on 14 February 2021). rAMPage is a homology-based pipeline that uses RNA-seq reads as input to generate putative AMPs for downstream in vitro testing. Briefly, rAMPage generates putative AMPs by first processing the input RNA-seq reads using fastp v0.20.0 [43] and assembles the reads into transcripts with RNA-bloom v1.3.1 [44]. Transcripts are subsequently translated by Transdecoder v5.5.0 [45] and a homology search is conducted with HMMER v3.3.1 [46]. Precursor sequences are cleaved with ProP v1.0c [47] and putative mature AMP sequences are prioritized by AMPlify v1.0.3, a deep-learning classifier [48].The putative AMPs were filtered to retain peptides with a minimum charge of +2 and maximum length of 30 amino acids. Peptides were further characterized with ENTAP v0.10.7 [49], Exonerate v2.4.0 [50] and SABLE v4.0 [51] before being clustered with CD-HIT v4.8.1 [52]. Ninety sequences were prioritized for synthesis from this list using three selection criteria:

“Species Count”—peptides identified in two or more species;

“Insect Peptide,”—insect-derived peptides chosen using a reduced AMPlify prediction score threshold; and

“AMPlify Score”—the top-scoring peptides with the highest net positive charge.

Peptides were purchased from GenScript and provided in lyophilized 0.08 milligram aliquots. Two peptides were unable to be synthesized, resulting in a final experimental set of 88 putative AMPs.

4.2. Bacterial Isolates

Bacteria were grown overnight at 37 °C in Mueller-Hinton Broth (MHB; Sigma-Aldrich, St. Louis. MO, USA) in a shaking incubator and aliquoted into cryovials with a 50% glycerol solution in a 1:1 ratio. No additives were added to the MHB during the growth of bacterial isolates. Glycerol stocks were stored at −80˚C.

4.3. Antimicrobial Susceptibility Testing (AST)The antimicrobial activity of the putative AMPs was determined by a broth microdilution assay as described by the Clinical and Laboratory Standards Institute [27], with the published adaptations for testing of cationic peptides [28]. Escherichia coli 25922 and Staphylococcus aureus 29213 isolates were purchased from the American Type Culture Collection (ATCC; Manassas, VA, USA). A human clinical isolate of Salmonella enterica serovar Enteritidis was provided by the BC Centre for Disease Control. Bacteria from stocks stored at −80 °C were streaked onto nonselective Columbia blood agar with 5% sheep blood (Oxoid) and incubated overnight at 37 °C. Next, 2–4 colonies were streaked onto a new agar plate to ensure uniform health of colonies used in the broth microdilution assay. Isolated colonies were suspended in MHB and the optical density was measured with a spectrophotometer to create an initial inoculum concentration of approximately 1 × 108 cfu/mL. The inoculum was diluted 1:250 to a final concentration of 2–8 × 105 cfu/mL, which was confirmed by performing a Total Viability Count (TVC). The TVC consisted of plating a 1:1000 dilution of the final inoculum on nonselective media and was also used to confirm the health and purity of the inoculum in each trial.

Lyophilized 80 µg peptide aliquots synthesized by GenScript (Piscataway, NJ, USA) were resuspended to 1.28 mg/mL with UltraPure water (Thermo Fisher Scientific, Waltham, MA, USA) and serially diluted in polypropylene 96-well microtiter plates (Greiner Bio-One #650261, Kremsmünster, Austria) from 128 down to 0.5 µg/mL. Two columns per plate were reserved for growth and sterility controls. Inoculum was added to the wells containing the peptide and the growth control. Plates were incubated for 20–24 h at 37 °C. The minimum inhibitory concentration (MIC) was determined as the lowest peptide concentration where there was no visible bacterial growth.

4.4. Structure Prediction and ClusteringWe predicted the 3D structures of the mature peptides using a local installation of ColabFold [21]. Using ColabFold’s server, five structures were generated for each peptide with sub-models of AlphaFold. Structural templates were used as input along with the peptide sequences for better accuracy. The five estimated structures were relaxed using the Amber force field, which helps remove stereochemical violations in predictions [20]. The five structures for each sequence were ranked by the per-residue estimate of AlphaFold’s confidence in its prediction (pLDDT score). The PDB file of the Amber relaxed rank 1 model for each peptide was used for further analysis.Similarity of the predicted structures of peptides was determined using mTM-align (version 20180725) [53], which was used to prepare a distance matrix calculated from the pairwise TM-scores of the PDB files produced for each peptide. This distance matrix was clustered using scipy.cluster.hierarchy [54] in python with complete linkage to produce a dendrogram. PDB files were fed into STRIDE [26] and the residue assignments were used to categorize peptides. Peptides were classified into linear extended peptides (only possessing turn secondary structure and/or coils), mainly extended (>50% of the residues were in the extended category), mainly helical (between 50–80% of the residues were helical), helical (>80% of the residues were assigned as participating in an α-, pi- or 310-helical structure), or β-strand containing. The fisher.test() function in base R was used to conduct a Fisher’s exact test to investigate the independence of the helical or linear extended structure and possession of activity. Only peptides classified as helical (≥80% of residues assigned helical, 45 peptides) or linear extended (no residues in helical or β-strand structures, 17 peptides) were included in this statistical test. Peptides were labeled as active if they visually inhibited at least one of the bacterial strains at one of the tested concentrations. 4.5. Hemolysis Assay

Peptides were evaluated for toxicity using three independent hemolysis experiments. Whole blood from healthy donor pigs, supplemented with Na Citrate, was purchased from Lampire Biological Laboratories (Pipersville, PA, USA). Red blood cells (RBCs) were washed and isolated by three centrifugation cycles using Roswell Park Memorial Institute medium (RPMI; Thermo-Fisher Scientific) to create a 1% (v/v) RBC solution. Lyophilized AMPs were suspended and serially diluted from 128 down to 1 μg/mL using RPMI in a 96-well plate, before being combined with 100 μL of the 1% RBC solution. Following a minimum 30 min incubation at 37 °C, plates were centrifuged and ½ volume from each supernatant was transferred to a new 96-well plate. The absorbance of these wells was measured at 415 nm. To quantify hemolytic activity and determine the AMP concentration that lyses 50% of the RBCs (HC50), absorbance reading from wells containing RBCs treated with 11 μL of a 2% TritonX-100 detergent solution (TX-100) or RPMI (AMP solvent-only) were used to define 100% and 0% hemolysis, respectively. All centrifugation steps were performed at 500× g for five minutes in an Allegra-6R centrifuge (Beckman Coulter, Brea, CA, USA).

4.6. CytotoxicityThe human embryonic kidney cell line HEK293 and the corresponding growth media and supplements were purchased from the ATCC. The cells were maintained in Eagle’s Minimum Essential Medium (EMEM) supplemented with 10% fetal bovine serum (FBS) and 1% Penicillin-Streptomycin solution and grown in a 5% CO2 incubator at 37 °C. Approximately 1 × 104 cells were distributed to each well of a 96-well flat-bottom cell culture plate (Corning #3595, Corning, NY, USA). Cells were incubated overnight to allow them to adhere. Lyophilized 80 µg peptide aliquots were resuspended to 1.28 mg/mL with UltraPure deionized water and serially diluted in polypropylene 96-well microtiter plates from 128-0.5 µg/mL. TX-100 was diluted to 2% in UltraPure deionized water and added as a positive control. Complete growth medium was added to the peptides and TX-100 containing wells. Spent growth medium of wells with adhered cells was replaced with the contents of wells containing serially diluted peptides or TX-100. Cells were incubated for four hours at 37 °C in 5% CO2 incubator. Growth medium containing peptides or TX-100 was replaced with fresh growth medium containing 10% (v/v) alamarBlue (Bio-Rad Laboratories, Hercules, CA, USA) and incubated for 20 h. Fluorescence was measured at excitation of 540 nm and emission of 590 nm in a Cytation 5 plate reader (Agilent BioTek, Winooski, VT, USA) and results recorded and analyzed with Gen5 software (Agilent BioTek, Winooski, VT, USA). The average fluorescence reads of TX-100 and growth media only wells were used to calculate 0% and 100% cell viability, respectively. The percentage of viable cells at each peptide concentration was determined by:

cell viability%=1−fluorescence100% viable− fluorescencepeptidefluorescence100% viable−fluorescence0% viable×100%

4.7. BLAST and Phylogenetic AnalysisAmino acid sequences for the mature peptides tested in vitro were used as the query sequences for BLASTp analysis with NCBI BLAST+ v2.13.0 [29]. The sequences were searched against the non-redundant protein database (downloaded on 01/19/2022) [29]. Max target sequences was set to 5. To identify protein family memberships, mature sequences were fed into InterProScan v5.56-89.0 [55] in FASTA format with default parameters. Multiple sequence alignment was performed with the 51 validated AMPs using the Bioconductor R package msa v1.28.0 [56] with the ClustalW option. A distance matrix was created using the multiple sequence alignment with the seqinr package v5.6-2 [57] and a phylogenetic tree using neighbour-joining tree estimation was created using the ggtree v3.4.2 [58,59,60,61] and ape v5.6-2 [62] R packages. AMP protein families, origin of AMPs and BLASTp results were annotated on the phylogenetic tree using ggtree [58,59,60,61].

留言 (0)

沒有登入
gif