Mining human microbiomes reveals an untapped source of peptide antibiotics

Abstract

Drug-resistant bacteria are outpacing traditional antibiotic discovery efforts. Here, we computationally screened 444,054 previously reported putative small protein families from 1,773 human metagenomes for antimicrobial properties, identifying 323 candidates encoded in small open reading frames (smORFs). To test our computational predictions, 78 peptides were synthesized and screened for antimicrobial activity in vitro, with 70.5% displaying antimicrobial activity. As these compounds were different compared with previously reported antimicrobial peptides, we termed them smORF-encoded peptides (SEPs). SEPs killed bacteria by targeting their membrane, synergizing with each other, and modulating gut commensals, indicating a potential role in reconfiguring microbiome communities in addition to counteracting pathogens. The lead candidates were anti-infective in both murine skin abscess and deep thigh infection models. Notably, prevotellin-2 from Prevotella copri presented activity comparable to the commonly used antibiotic polymyxin B. Our report supports the existence of hundreds of antimicrobials in the human microbiome amenable to clinical translation.

Keywords: SEPs; antibiotics; antimicrobial peptides; computational mining; human microbiome; microproteins; peptides; smORF-encoded peptides.

Figure 1.. Schematic of the computational-experimental platform for the discovery of SEPs from smORFs

Metagenomes from four distinct body sites were analyzed to identify open reading frames (ORFs) containing more than 15 base pairs, using the MetaProdigal tool (see also Table S1). Subsequently, small ORFs (≤ 150 bp) were filtered out, and the encoded proteins were grouped into families, resulting in a total of 444,054 families. To further narrow down the selection, representatives of each family underwent analysis with SmORFinder, and the results were ranked using AmPEP to identify peptides with antimicrobial potential. The sequences that were identified by both SmORFinder and ranked as antimicrobials by AmPEP were considered as high-confidence families (see also Data S1A and S1B). These families were then subjected to further filtering based on specific criteria, as outlined in the inclusion and exclusion criteria for selecting peptides for activity testing section. The selected high-confidence families were subsequently tested against a range of pathogen and commensal bacterial strains. Promising candidates were further investigated through systematic characterization, including conformational studies, mechanism of action elucidation, assessment of synergistic interactions, and evaluation in preclinical mouse models. The figure was created with BioRender.com.

Figure 2.. Sequence-related features of SEPs

(A) Amino acid frequency was calculated from candidate SEPs, from EPs predicted in the human proteome, and from AMPs in the DBAASP database. SEPs had overrepresentation of acidic residues (aspartic acid, D; and glutamic acid, E), polar uncharged residues (methionine, M, and asparagine, N), and lower content of leucine (L) residues. Synthesized and validated active SEPs showed similar amino acid content compared with all SEPs, and thus, they were considered as representatives of the total numbers of candidate SEPs. Among the most relevant physicochemical features that are known to influence biological activities of peptides (see also Figures S1 and S2), (B) SEPs have lower net positive charge and (C) normalized hydrophobicity than AMPs and EPs. Thus, SEPs are not amphipathic as other classes of AMPs, and instead, they are slightly less hydrophobic sequences with higher tendency to be disordered (see also Figure S2 and Table S2). See also Figures S1 and S2.

Figure 3.. Antimicrobial activity and structure analysis of SEPs

(A) Antimicrobial activity of the tested SEPs. Briefly, a 10⁶ bacterial cell load was exposed to serially diluted SEPs (1–128 μmol L⁻¹) in 96-well plates and incubated at 37°C. 20 h after the beginning of the experiment, each condition was analyzed in a microplate reader at 600 nm to check for inhibition of bacterial growth compared with the untreated controls. The results are presented as a heat map of antimicrobial activities (μmol L⁻¹) against pathogenic and gut commensal bacterial strains. Assays were performed in three independent replicates (see also Figure S2 and Data S1C). (B) ColabFold was used to generate structural predictions using default parameters. Three-dimensional ribbon structures of the resulting PDB files were generated using Mol* 3D Viewer. (C and D) (C) Circular dichroism spectra of the active SEPs in helical inducer medium to assess the tendency of SEPs to the most common structure adopted by antimicrobial peptides. Five of the SEPs present significant helical content (faecalibacticin-3, fusbacticin-2, keratinobacin-1, staphylococcin-2, and prevotellin-2) expressed in (D) helical fraction (f_H) values of active SEPs in a heat map, where the higher f_H values are presented in red and the lowest in blue. The activity did not correlate with the antimicrobial activity, once again reinforcing the independency of this class of peptides from amphipathic and balanced hydrophobic/cationic residues sequences (see also Figures S3A and S3B). See also Figures S2 and S3.

Figure 4.. Synergy, mechanism of action, and cytotoxicity of SEPs

(A) The synergistic interaction between pairs of SEPs from the same biogeography (tongue dorsum, supragingival plaque, and stool) was assessed by checkerboard assays with 2-fold serial dilutions starting at 2 × MIC to MIC/32. The histogram shows the fractional inhibitory indexes (FICIs) values obtained for each pair of SEPs, where dark red represents synergistic interactions, light red indicates additive interactions, and blue shows indifferent interactions. Most of the pairs of SEPs presented synergistic or additive interactions. To assess whether SEPs act on the bacterial membrane, all active SEPs against each of the pathogenic strains were tested in outer membrane permeabilization and cytoplasmic membrane depolarization assays. In general, SEPs presented low permeabilization of the outer membrane effect, as shown in (B) the relative fluorescence measurements of SEPs on A. baumannii cell membranes (see also Figure S4). SEPs showed high depolarization properties as shown in (C) the relative fluorescence measurements of SEPs on vancomycin-resistant E. faecium cytoplasmic membranes (see also Figure S5). The relative fluorescence was calculated with a non-linear fitting using as baseline the untreated control (buffer + bacteria + fluorescent dye) as described in the STAR Methods section. The correlation between cytotoxicity on (D) human colorectal adenocarcinoma cells (Caco-2) or (E) immortalized human keratinocytes (HaCaT) and antimicrobial activity is shown in a scatterplot where the cytotoxicity is represented by the CC₅₀ values (cytotoxic concentrations causing 50% cell death) and MIC (minimal inhibitory concentration for complete bacterial killing). CC₅₀ values have been predicted by interpolating the dose-response with a non-linear regression curve. The green area represents the therapeutic window where those peptides could be safely used with no toxic effect to eukaryotic cells (see also Figures S5R and S5S). See also Figures S4 and S5.

Figure 5.. Anti-infective activity of SEPs in preclinical animal models

(A) Schematic of the skin abscess and deep thigh infection mouse models used to assess the anti-infective activity of the smORF-encoded peptides (SEPs) against A. baumannii cells. (B) In the skin abscess infection model, mice were infected with a load of A. baumannii and treated 2 h after infection with one dose of the SEPs at their MIC. Mice were euthanized 2 and 4 days post-infection, and the tissue samples were homogenized and plated on agar plates in a 10-fold dilution gradient for colony-forming units (CFU) counts after an incubation of 24 h at 37°C. Each group (treated and untreated) consisted of three mice (n = 3), and the bacterial loads used for infection of each mouse came from a different inoculum. The experiment was done in three independent replicates. All peptides had similar bacteriostatic effect 2 days after infection, and after 4 days, all the SEPs tested were significantly different than the control, and the peptide prevotellin-2 presented activity comparable to the positive control (polymyxin B), reducing the infection by four orders of magnitude. (C) To rule out toxic effects of the peptides, mouse weight was monitored throughout the whole extent of the experiment. We considered 20% weight change as acceptable considering the duration of the experiment. (D) In the deep thigh infection mouse model, mice were first immunosuppressed by two rounds of treatment (24 and 72 h pre-infection) of immune system suppressor (cyclophosphamide). Subsequently, an intramuscular injection of A. baumannii was administered in the right thigh, followed by intraperitoneal administration of the peptides, to evaluate their systemic anti-infective activity 2 h after infection. 6 days after the start of the experiment, corresponding to 2 days post-infection, mice were euthanized. Each group, comprising treated and untreated mice, consisted of three individuals (n = 3), with distinct bacterial loads used for infecting each mouse, originating from different inocula. The experiment was done in three independent replicates. All peptides presented significant activity (one to two orders of magnitude reduction in bacterial counts), and the SEP prevotellin-2 had a similar effect than the antibiotics used as positive controls, polymyxin B and levofloxacin, reducing three to four orders of magnitude the bacterial counts 2 days post-infection (day 6 of the experiment). 4 days post-infection, levofloxacin was the only treatment that led to a more than 2 orders of magnitude decrease compared with the untreated control. (E) During the entire 8-day period of the deep thigh infection model, mouse weight was closely monitored to eliminate the possibility of any toxic effects caused by cyclophosphamide injections, bacterial load, and the peptides. To determine statistical significance in (B) and (D), one-way ANOVA followed by Dunnett’s test was employed, and the respective p values are presented for each group. All groups were compared with the untreated control, and the violin plots display the median and upper and lower quartiles. Data in (C) and (E) are the mean plus and minus the standard deviation. Figure created with BioRender.com.