In terms of species and population, bacterial communities occur in dynamic equilibria related to environmental factors that drive interspecies competition and coexistence. Different species may facilitate their survival in a particular environment (e.g., gut and biofilms) by sharing information on existing sources of food or threats using small molecular messengers (Mukherjee and Bassler, 2019). Apart from aiding each other, bacteria most often compete using toxic molecules. In a close contact intercellular war, they may inject lethal molecules (peptidoglycan hydrolases, phospholipases, pore forming proteins, DNases, RNases, NAD(P)⁺ hydrolases, and ADP-ribosyltransferase among others), to suppress antagonistic species using the type VI secretion system (Coulthurst, 2019; Ross et al., 2019). Another type of antagonism is the secretion of toxic molecules destined for competitors (Richards et al., 2017). If the toxic effector molecules are proteinaceous they are called bacteriocins (Cotter et al., 2013). The term describes a broad and heterogenous category of molecules that can inhibit bacterial growth while being present in a defined contained environment (chemosphere) (Baquero et al., 2019). They are roughly divided in two categories: one of transmembrane proteins [colicins (Kleanthous, 2010)] and another of peptides (class I bacteriocins) (Riley and Wertz, 2002). Microcins are class I bacteriocins with pluripotent inhibitory actions that may act in synergy with the larger colicins for even broader inhibition effects (Baquero et al., 2019). Bacteriocins have a narrow target range and are designed to benefit the species that produces them over its closely related competitors.

Antimicrobial peptides (AMPs), a term first used for antibacterial peptides of the innate immune response of eukaryotes, are found in all kingdoms of life (Hancock, 2001). Eukaryotic AMPs mostly target and disrupt bacterial membranes at micromolar concentrations, without interacting with receptors (Kumar et al., 2018). They take advantage of the specific lipid constitution of bacterial membranes [very polar parts of the exposed lipid heads (Kumar et al., 2018)] and the curvature of the bacterial membrane (Matsuzaki et al., 1998; Drin and Antonny, 2010) to act as detergents lysing the bacterial cell. AMP-like detergents may constitute part of bacterial defenses against other bacterial species [e.g., class I bacteriocin nisin from Lactococcus lactis (Breukink et al., 1999), mersacidin from Bacillus sp. (Chatterjee et al., 1992), bacteriocins MccV, MccE492, and MccL]. The bacteria generating AMPs are not harmed due to specific immunity mechanisms that include specific AMP pumps (Beis and Rebuffat, 2019), or dedicated immunity proteins embedded within the cellular membrane, that specifically bind to the cognate AMPs on the extracellular side preventing its entrance-damage to the cell (Hassan et al., 2012; Beis and Rebuffat, 2019).

Antimicrobial peptides maintain a balanced microbiome and establish compatibility between bacterial populations in niches such as the human gut (Kerr et al., 2002; Kirkup and Riley, 2004; Majeed et al., 2011). Apart from their role in microbial biodiversity, AMPs have been employed as antimicrobials. Purified AMPs have been used to extend food preservation time, in the treatment of infectious diseases (e.g., skin infections and wounds) (Pfalzgraff et al., 2018) and cancer (Rodrigues et al., 2019). The therapeutic potential of AMPs as a replacement drug candidate for antibiotics has immense potential, in particular for the treatment of pathogens resistant to antibiotics (Yang et al., 2014).

Antimicrobial peptides that form pores are generally considered α-helical amphipathic molecules that may act as detergents on bacterial membranes (Blondelle et al., 1999). The amphipathicity and α-helical secondary structure characteristics have been exploited by a number of prediction tools in order to predict AMPs in a given protein sequence (Gabere and Noble, 2017). Detailed studies on mutations affecting the toxicity of eukaryotic bovine peptide bactenecin 2A (Hilpert et al., 2005) have been used for the formulation of AMPA, a predictor of antimicrobial peptides in a protein sequence (Torrent et al., 2009, 2012). Specifically, AMPA uses a sophisticated algorithm that gives an antimicrobial propensity value for a selected peptide stretch (Torrent et al., 2009). Each amino acid of the stretch is assigned an “antimicrobial index” value derived from high-throughput screening results (antimicrobial IC₅₀ values) concerning amino acid replacements of the AMP bactenecin 2A (Hilpert et al., 2005). AMPA has been thoroughly validated in silico exhibiting accurate prediction of 80–90% of the assessed antimicrobial proteins including human ribonucleases, lysozymes and bacterial bacteriocins (Torrent et al., 2009).

We wondered whether bacterial proteomes might hide antibacterial stretches that could be used for interbacterial competition among different genera. Assuming that sequence similarities of AMPs from the proteomes of different bacteria could form a basis for interspecies competition, a computational method termed Bacterial Wars (BW) was developed and employed herein to explore the relationship between putative AMPs and bacterial predominance. The BW method (i) analyzed and compared putative AMPs hidden in the proteomes of different bacterial taxa and (ii) used these AMPs to predict interbacterial antagonism based on their sequence similarities. Finally, we put the hypothesis to the test by comparing the outcome of BW, to the abundance of bacterial genera in eleven publicly available 16S rRNA derived datasets as well as the species from a high-resolution full shotgun metagenomics dataset. The overall outcome was that AMPs embedded within bacterial proteomes may affect prevalence of individual taxa in the gut. To differentiate this form of competition from previously known specialized mechanisms, the novel concept is coined as “putative AMP defense.”

The concept of peptides of bacterial origin as factors of interbacterial antagonism is not new (Baquero and Moreno, 1984). Examples date back to microcin E492 which is secreted by Klebsiella pneumoniae RYC492 and is active against various Enterobacteriaceae (de Lorenzo, 1984; Thomas et al., 2004; Gillor et al., 2008). Some of these peptides [e.g., mersacidin, nisin, subtilin, cinnamycin, duramycin, actagardine, epidermin, gallidermin, lanthiopeptin (Chatterjee et al., 1992)], may require post translational modification [addition of lanthionine (Knerr and van der Donk, 2012)] to stabilize the peptide and enhance the killing effect. Synthetic improvements of hairpin AMPs have also been used (Wu and Hancock, 1999; Liu et al., 2013). Despite the wide diversity of bacterial species, the isolated and characterized AMPs from bacteria are a fraction compared to those from eukaryotes (200 bacterial AMPs compared to 2159 from animals) (Kumar et al., 2018). The herein predicted 3000 putative AMPs for E. coli only (Figure 1B), suggests that the number of bacterial (and human) AMPs may be underestimated. The high number of bacterial AMPs is suggestive of a possible underlying function. Could it be that bacteria (and perhaps other organisms) employ AMPs encrypted in their proteomes to defend themselves against foe, especially in the relatively stable environment of a chemosphere (e.g., gut)? Could it be that a dying or dead bacterium might become a source of AMPs resulting from the proteolysis of its own proteins, to antagonize “enemy” species? This possibility would require the presence of proteases in a chemosphere. Bacterial proteases do exist in the colon (Ruiz-Perez and Nataro, 2014; Sears et al., 2014; Guyton et al., 2019) as well as proteases from other microorganisms (Espinosa-Cantellano and Martínez-Palomo, 2000; Ajjampur and Tan, 2016) to account for such proteolytic activity. Most of the proteins identified as potential sources of AMPs were cationic with nucleic acid binding properties (Figure 1B). Such proteins contain many basic residues (e.g., lysines) and are thus amenable to proteolytic degradation. The concept of AMP homology among bacterial proteomes as being the main prerequisite for bacterial species compatibility cannot be envisaged without proteolysis occurring in a chemosphere that would allow for the continuous production of AMPs. A mechanism as such would relieve the cell from the need of the energy expenses to synthesize de novo AMPs by protein synthesis.

Under the supposition that a bacterial proteome is a potential source of AMPs that are not toxic to the host species and/or species that share AMP homology, a methodology was developed (“Bacterial Wars” or BW) that could predict the outcome of interbacterial competitions. BW (i) detected all putative AMPs embedded in curated proteins from UniProt-derived bacterial proteomes, (ii) clustered the detected AMPs by sequence similarity (iii) created a database of bacterial taxa based on AMP sequence similarities and finally (iv) predicted specific taxa that were more likely to “die” in a given microbiome. The final step was achieved using a novel network-based scoring scheme (termed “Die score”) that provided a measure of how likely a given microbial genus/species was to “die” in a given niche. It should be noted that the Die score is a network measure that may not be restricted to this work. It can be applied to any network that may follow similar assumptions to the bacterial networks generated from the BW method (e.g., co-occurrence networks). The methodology was used to provide evidence in favor of the putative AMP defense hypothesis using eleven 16S rRNA gene datasets as well as one full shotgun metagenomics dataset from the gut microbiome, where the outcomes of the BW method were compared to co-occurrence networks derived from these data. In almost all examined cases, bacterial predominance seemed to be affected by the similarity of encoded AMPs in bacterial proteins.

Co-occurrence networks are subject to change depending on the parameters used during the algorithm employed for their construction (i.e., the choice of correlation metric, the p-value cut-off etc.). This directly impinges on the correlation scores between Die scores for BW and MB networks and although we have tried to be as judicious as possible in the use of these parameters across the twelve datasets assessed in this study, there is still an aspect of subjectivity in the use of these parameters that could potentially affect results. Another concern is the exclusive use of AMPA as a detector of AMPs. It could well be that not all AMPs are detectable by AMPA and not all AMPs predicted by AMPA may have antibacterial properties. All these concerns must be addressed experimentally, which is not a trivial task. We can only state that under the current handling and analysis of proteomes, AMPs and bacterial abundances, putative AMPs seem like a significant factor in the interaction of different bacteria in a chemosphere (putative AMP defense hypothesis).

Numerous databases describe AMPs from prokaryotic and eukaryotic organisms (Seshadri Sundararajan et al., 2012; Gogoladze et al., 2014; Fan et al., 2016; Lin et al., 2016; Pirtskhalava et al., 2016; Waghu et al., 2016; Wang et al., 2016). They host experimentally validated, as well as computationally predicted, AMPs and are fully searchable allowing for queries of multiple search criteria. However, they do not provide any information on the relationship among the different AMPs or some background to explain their efficiency against specific organisms. The herein developed database combined with the BW methodology has the potential of detecting relationships between AMPs in the proteomes of different bacterial taxa and provide such information for further experimentation purposes. Predictions on the predominance and outlook of specific taxa in a given microbiome may lie within the potential of the method.

A first glance on bacterial genome size and AMP count suggests that the two are not related (Figure 1A). The aphid dwelling Buchnera aphidicola with a relatively small genome appears as having the highest number of putative AMPs followed by Wigglesworthia glossinidia, that lives in the gut of tsetse flies and is considered as having one of the smallest known genomes of any living organism. On the other hand, the predatory Bdellovibrio bacteriovorus, and Fusobacterium nucleatum of the oral cavity, seem as having less putative AMPs. E. coli lab strain K12 has a large number of putative AMPs (Figure 1B) but relatively small compared to its proteome (Figure 1A). Could it be that bacteria that live in relatively host-independent conditions (more open chemospheres) may need less AMPs? The relationship between, abiotic conditions of living, surrounding microbiomes and hosts are some of the issues that could be the object of a comparative study with the aim of explaining the different numbers of AMPs in different bacterial taxa.

A large proportion of the putative AMPs were unique as they shared low sequence similarity (<80%) (Figure 1C). This could be viewed as the extreme versatility of the combinatorial capabilities of microorganisms to create diverse AMPs to face an opponent species. Comparison of the Die scores in bacteria with close phylogenetic origin showed that putative AMPs were specific to the level of genera (Figure 6).

In summary, we propose that putative AMPs hidden within the proteome of bacteria may affect their symbiosis with other species in a given chemosphere. In support of the putative AMP defense hypothesis, we provide computational data corroborated by gut microbiome datasets on the abundance of bacterial species in the enterobiomes of different diseases. We believe that our findings may add another jigsaw in the mosaic of bacterial interactions and could perhaps provide leads for the construction of species-specific AMPs.

Publicly available datasets were analyzed in this study. This data can be found here: https://zenodo.org/record/569601#.XdPrYmgzaUk, https://www.medrxiv.org/content/10.1101/2021.02.07.21251098v1.

AO, MZ, and GMS designed the study. AO and MZ conceived and developed the computational tools. AO, CC, and MT analyzed the data and interpreted the results. GMS and AV-G interpreted the results. UI contributed to code for the computational tools. GPS contributed to the Parkinson’s metagenome dataset. AO and AV-G designed the experiments and wrote the manuscript. AV-G conceived the project. All authors edited the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.