Actifensin Evolution in the Human Oral Cavity over the Past 100,000 Years Authors: Rosa Herbst, Anan Ibrahim, Alexander Hübner, Uwe Knüpfer, Lars Regestein, Christoph Wiedemann, Ute A. Hellmich, Christina Warinner, Pierre Stallforth Published: 2025 ABSTRACT Bacterially produced antimicrobial peptides (AMPs), or bacteriocins, play key roles in shaping microbial communities via interspecies competition. Unlike the more temporally dynamic gut microbiome, the oral microbiome exhibits long-term stability and is preserved into deep time in dental calculus, enabling evolutionary analysis across time. Here, we combine metagenomics, structural modeling, and experimental validation to investigate AMP diversity in ancient and modern dental biofilms from humans, Neanderthals, and nonhuman primates spanning 100,000 years. Using our newly developed platform, AMPcombi, we uncover evolutionary trajectories of bacteriocins and elucidate their ecological functions. Among these, we identify a conserved family of Actinomyces-derived defensin-like peptides, termed actifensins, present across all time periods. Phylogenetic, structural, and functional analyses revealed shared ancestry and adaptive diversification between ancient (paleo-) and modern actifensins, with evidence of positive selection and maintained antimicrobial activity. Our findings position the oral microbiome as a valuable reservoir for natural product discovery. In the face of rising antimicrobial resistance, evolutionary insights into AMP function open a door to next-generation therapeutics. AMPcombi streamlines this process, linking ancient biomolecules with biotechnology. Introduction Microbial natural products constitute the basis for the majority of all commercially available antibiotics, as well as a large number of anticancer drugs, immunomodulators, and cholesterol-lowering drugs. , Recent advances in the field of archaeogenetics and paleobiotechnology have enabled the reconstruction of biosynthetic enzymes from the past, thus opening the door to microbial natural product discovery spanning more than 100,000 years. Structural and functional diversification of these natural products requires extensive reprogramming of their complex biosynthetic assembly lines, resulting in an evolutionary rate substantially slower than that of peptides and proteins. In contrast, ribosomally produced antimicrobial peptides (AMPs) can evolve more rapidly due to the direct translation of changes in the DNA sequence into structural and functional peptide changes. AMPs are vital components of the defense systems in virtually all forms of life, displaying antimicrobial activities that range from narrow to broad-spectrum. In higher organisms, AMPs are integral to the innate immune system, acting as a first-line defense barrier against harmful microorganisms. Host and bacterially produced AMPs also play a major role in microbiomes, where they strongly influence the balance and dynamics of microbial communities. Bacterially produced AMPs, known as bacteriocins, provide a competitive advantage to the producer strain by eliminating competing bacteria in their ecological niche. The human oral microbiome is a highly complex microbial community containing more than 500–700 prevalent species. , Oral microbes are highly structured and spatially organized within different oral niches, and they exhibit remarkable stability over long periods. Within dental biofilms, oral microbial structure reflects complex interactions of adhesion, syntrophy, and secretion of signaling molecules and bacteriocins that influence bacterial colonization, persistence, and immune escape, rendering dental biofilms an excellent model for investigating the AMP evolution. Moreover, dental biofilms naturally calcify during life, forming dental calculus, which is preserved after death and therefore is nearly ubiquitous in the archeological record. This dental calculus bioarchive of the oral microbiome has been shown to exhibit excellent biomolecular preservation, enabling the direct study of human oral microbiome evolution over tens of thousands of years. These studies revealed a high degree of conservation within primate oral microbiota over at least 40 million years, with humans, great apes, and howler monkeys sharing a core dental biofilm microbiome of at least ten major genera. Here, we present a new software tool, AMPcombi, to explore AMP evolutionary changes within human and Neanderthal dental biofilms over a period of 100,000 years, as well as within dental biofilms of nonhuman primates, allowing us to retrace the structural and functional changes of bacteriocins over key periods of recent human evolution. In an era of escalating antimicrobial resistance, an evolutionary approach to discovering AMPs holds great promise as an alternative treatment strategy. By understanding the evolutionary adaptations and functions of AMPs, we can develop novel antimicrobial therapies to combat resistant pathogens. Results De Novo Assembly of Oral Metagenomes from Human and Nonhuman Primate Dental Calculus We compiled a data set of 119 published dental biofilm metagenomes derived from the dental calculus of present-day humans (n = 18), archeological humans (n = 43) and Neanderthals (n = 12), as well as nonhuman primates consisting of chimpanzees (n = 20), gorillas (n = 29), and howler monkeys (n = 5) , (Figure A; Supplementary Data 1). The Neanderthal samples date to the Middle Paleolithic (ca. 44–102 ka BP), while the archeological human populations originate from Europe and Africa and span multiple time periods, including the Upper Paleolithic (n = 12, ca. 14–31 ka BP), Mesolithic (n = 7, ca. 9–10 ka BP), Neolithic (n = 6, ca. 4–8 ka BP), and historic (n = 18, ca. 1850–550 BP) periods. We then de novo assembled these metagenomes using metaSPAdes for present-day human samples and MEGAHIT for nonhuman primates and archeological humans and Neanderthals, producing 3,274,184 contigs >500 bp in length for AMP discovery using AMPcombi (Supplementary Figure 1, Supplementary Data 2). Antimicrobial peptide sequence prediction from dental calculus using AMPcombi. (a) Geographical map showing the distribution of the dental calculus metagenomes (DCMs) used in this study indicating the species, time period, and origin they represent. (b) A graphical summary illustrating the workflow used to generate the input for AMPcombi, the intermediary steps and the downstream processing invoked to recover putative AMP classes from the DCM. (c) Bubble chart demonstrating the proportion of successfully predicted AMPs derived from the different DCM samples (small bubbles with sample codes) grouped by age period (large bubbles) using AMPcombi and data set specific filtering steps. The color code of the different age periods is shown in (a). (d) The bar plots display the number of putative AMPs recovered per one million nucleotides within the DCMs, while the stacked bar plot illustrates the proportion of the retained AMPs across the DCMs based on the prediction efficacy of each AMP prediction tool included in the workflow. The color code of the different age periods is shown in (a). Six AMP prediction tools were used in the workflow. HMMsearch was included as an additional sequence similarity–based method to capture known AMP motifs and to standardize tool performance, but is not considered a prediction tool. Logo artwork by Anan Ibrahim, used with permission. Enhanced AMP Discovery with AMPcombi Recent years have seen a surge in the development of bioinformatic tools applying various algorithms to identify genes encoding antimicrobial peptides (AMP). Each tool has specific strengths and weaknesses, and AMP discovery is best achieved by combining the results of multiple tools. However, integrating the results of such tools is cumbersome and error-prone because of different output formats and the large amount of generated data with only minimal overlap. To overcome this challenge, we developed AMPcombi, a Python-based software tool that integrates the results of six state-of-the-art computational tools to narrow down AMP hits in a single workflow, and provides a GUI interface to dynamically inspect the results. Applying AMPcombi to our data set of dental calculus metagenomes (DCMs), we identify a total of 4,628,215 potential AMP hits. However, this large number is likely an overestimate resulting from AMP misassignment, which is a longstanding problem in the identification of AMPs. Cross-tool comparison is an essential step for reducing false positives, but is challenging to manually perform on large data sets. AMPcombi (Figure b) automates this process by consolidating the outputs into a streamlined, tab-separated table, suitable for interactive exploration via a web interface. Given the variability in AMP prediction accuracy of the different AMP prediction tools, user-adjustable parameters are included to allow for standardization and refinement of results, ensuring that predictions from large-scale metagenomic data sets are manageable, and to provide experimentally relevant hits (Figure , Supplementary Figures 2 and 3 and Supplementary Section 1). To further increase reliability, AMPcombi also includes an integrated option using HMMsearch (HMMER v.3.4), which can be customized by specific Hidden Markov Models (HMMs). This standardization by HMMsearch is particularly well-suited for finding the most suitable tool for the respective data set (Supplementary Figure 4a). Further validation of the results is performed by the alignment of the amino acid sequences of the putative AMPs to reference databases, such as InterProScan, and either APD, DRAMP or UniRef100. Large numbers of predicted AMPs found in metagenomic data sets can be easily narrowed in AMPcombi to those with the strongest support by applying user-specified filters to the results table (Figure b). AMP Prediction Tool Discrepancies Across Metagenomes We applied 65 HMM models representing different AMP classes in HMMsearch to the initial 4.6 million putative AMP gene hits identified by AMPcombi, of which 48 were identified in the dental calculus metagenomes (Supplementary Figure 4a). Cross-validation revealed that each tool has a distinct prediction tendency, with NeuBI, Ampir and AmpGram out-performing all other tools using the thresholds we set for this study (Figure a). Among the six prediction tools, Neubl was the only tool to identify nearly all HMM models aligned by HMMsearch (Supplementary Figure 4a). Finally, each tool exhibited performance differences in predicting distinct AMP families and AMPs with specific physicochemical properties and secondary structure (Supplementary Figure 4b–g). Recovery of ancient and oral-derived antimicrobial peptides from dental calculus. (a) The UpSet plot displays the number of AMPs predicted by six AMP prediction tools and HMMsearch (a sequence similarity–based method) after removing peptides longer than 120 amino acids and with AMP probability scores <0.8. The numbers above each vertical bar indicate the count of unique or mutually predicted AMPs by the tools, while the horizontal bars indicate the total number of AMPs predicted solely by each tool in the data set. (b) Heatmap illustrating the final 78 clusters representing different putative AMP classes recovered from DCMs using AMPcombi and the subsequent filtering steps. The AMP clusters are ordered by their total abundance among archeological human samples and are classified based on annotations obtained from UniRef100. The heatmap color scale represents the number of AMPs recovered per cluster in each time period, with yellow indicating few AMPs and dark blue indicating more than 70 AMPs, as shown in the scale bar. AMP Curation and Refinement After filtering predicted AMPs for only peptides shorter than 120 amino acids and with an AMP probability above 0.8 predicted by at least one tool, 254,718 AMPs were retained. Most of these putative AMPs were detected solely by one tool, while those identified by multiple tools were mostly derived from the three high-performing prediction tools: NeuBI, Ampir and AmpGram (Figure a). In an additional step, 5% of these AMPs were eliminated by removing peptides annotated as ribosomal proteins by InterProScan (Supplementary Figure 3). AMPs from DCMs were further trimmed using a set of biologically relevant selection criteria. Given that some AMPs from DCMs were found on taxonomically unclassified contigs (Supplementary Figures 5 and 9), only AMPs found on contigs belonging to a genus known to include oral species (Supplementary Data 3) were retained (Supplementary Figure 6). As a result, AMP-bearing contigs in the DCM were predominantly assigned to Actinomyces, Streptococcus, Neisseria, and Capnocytophaga, with each AMP prediction tool exhibiting varying tendencies to identify AMPs from the different taxonomic groups (Supplementary Figures 5 and 6). Additionally, because predictable forms of DNA damage are known to accumulate over time, we removed archeological DCMs lacking these signals (Supplementary Figure 7). After these filtering steps, 140,069 putative AMPs remained. Although we observed the highest number of putative AMPs in samples from present-day humans (Figure c), when normalized by the total contig lengths, more AMPs were recovered collectively from archeological humans (Figure c,d). Temporal coverage analysis shows that the most abundant taxa, e.g. Actinomyces and Arachnia, were found in similar proportions across the different time periods (Supplementary Figure 8c), whereas AMPs from Capnocytophaga, Rothia, and Treponema, were less prevalent in human samples older than 200 years. We next dereplicated the AMPs by clustering them according to amino acid sequence homology and removed any clusters with fewer than 10 AMPs, as well as any cluster without at least one archeological human. We then further removed clusters that lacked either a predicted signal peptide or validation from at least three of the seven AMP prediction tools (Supplementary Figure 10a). The final data set consisted of 78 clusters comprising 1669 putative AMPs (see Figure b and Supplementary Section 2.2 for a summary of all filtering steps). Comparative Analysis of Clustered AMP Variants The 1669 putative AMPs retained for analysis span the complete temporal range of the DCMs (Figure b). Approximately 50% originate from present-day humans, about 10% from recent periods (18th–20th centuries), 13% from more ancient periods (Paleolithic, Mesolithic, and Neolithic) and 27% from nonhuman primates. Of the total AMPs recovered from the historical samples, 50% were detected in the DCM of GRF001.A-D (Figure c), which originates from King Richard III of England (AD 1452–1485), and is likely due to the very deep sequencing of the dental calculus of this individual. Only ten clusters contained AMPs derived from human samples spanning all time periods (Figure b, Supplementary Figure 12 and Supplementary Table 1). In contrast, 17 clusters contained no AMP from any older archeological periods (Paleolithic, Mesolithic, Neolithic), and only two clusters contained no AMP hits from any present-day human samples (Figure b). The predicted secondary structures of the AMPs indicate that most clusters include peptides with either exclusively helical or mixed helical and β-sheet structures (Supplementary Figure 10c). Annotation of the AMPs was possible in 53% of the clusters by identifying protein homologies using an HMM model and databases such as DRAMP, APD, and UniRef100 (Figure b and Supplementary Figure 12), of which 22% were identified as bacteriocins. Although AMP-coding genes were mainly located on shorter contigs across all time periods (Supplementary Figure 8a,b), many clusters still contained enough longer contigs to examine neighboring genes. These frequently included genes related to peptide maturation, transport, and enzymatic modification (Supplementary Figure 13a,b). However, some clusters also produced false positives, with peptides unrelated to AMPs, such as conjugal transfer proteins, found alongside with respective associated elements (Supplementary Figure 13c). Evolution of Actinomyces Defensin-like Peptides Across the Past 100,000 Years Due to its well-preserved metagenomes and conserved microbial composition, the distinct microbial community of dental calculus is ideally suited for analyzing the structure–function relationships of AMPs in an evolutionary context. Among the final AMPs cluster set we selected cluster 59324 for further functional and evolutionary analysis. The 78 AMPs in this cluster represented the largest number of peptides among all identified clusters and were annotated as defensin-like peptides and identified across all archeological periods, as well as in present-day humans and nonhuman primates (Figure b and Supplementary Table 1). Taxonomic classification of the contigs revealed that the defensin-like AMPs all originate from oral-derived Actinomyces species (Figure and Supplementary Figure 14). Actinomyces is a genus that has been conserved within the primate lineage for over 40 million years, and it is a core member of the human oral microbiome. Actinomyces play a major role in dental biofilm initiation, and its interactions with other oral species structure the formation of dental plaque. Because AMPs play key roles in shaping microbiota, characterizing the evolutionary history of Actinomyces AMPs is an important and necessary step in understanding the long-term evolution of human dental biofilms. Overall diversity of oral derived defensin-like peptides originating from Actinomyces. (a) The phylogenetic tree represents the diversity of mature defensin peptide sequences found in Actinomyces-derived contigs from the dental calculus data set, as well as in Actinomyces genomes retrieved from NCBI that correspond to oral-derived strains. The tree was rooted using the mature peptide sequence from micasin-A. Sample age and type are indicated by symbols, and rings indicate isolation source (inner), host (middle), and species (outer). There is a wide distribution of defensin classes independent of host origin, species, time, or location. (b) SplitsTree neighbor network based on the consensus hybridization network of both amino acid- and nucleotide-based alignments (Supplementary Figure 18) computed using the PhyloFusion algorithm. Colors represent different oral-derived actifensin phylogroups across the split network. The network suggests a common evolutionary origin among the seven actifensin phylogroups, showing varying levels of reticulate evolution and divergence. Among the two further investigated actifensin pairs, which fall in Clade 1 (PAFI/MAF I) and Clade 6 (PAF/MAF II), PAF is consistently located basal to MAF. Phylogenetic analysis with different classes of defensin-like AMPs confirmed that the DCM-derived peptides are mostly related to bacterial homologues (Supplementary Figure 15). These peptides, recently found in Actinomyces, have been named actifensins, a nomenclature we also adopted here. To distinguish between ancient and more recent actifensins, we refer to actifensins originating from archeological DCMs >200 years old as paleo-actifensins (PAF), and actifensins from more recent DCMs as modern-actifensins (MAF). Most homologues of PAFs and MAFs found in public databases clustered with Actinomyces species of oral origin (Figure a and Supplementary Figure 16). Although most contigs contained only a single gene copy encoding a defensin-like peptide, in rare cases, two or more isoforms were found encoded adjacent to each other. All actifensins were encoded as larger precursor molecules with an N-terminal signal peptide, which is typically cleaved off during secretion at approximately three residues N-terminal to the first cysteine to release the mature peptide. For the signal peptide regions of both PAFs and MAFs, amino acid sequence similarities range from 55% to 100% (mean = 77.4%). In contrast, the mature regions have a greater variation, ranging from 33% to 100%, and lower average similarity with a mean of 58.7%, excluding self-comparisons. Aligning the amino acid sequences of the mature peptides, together with 127 homologous sequences of oral cavity-associated Actinomyces strains, resulted in a phylogenetic tree with seven distinct clades (Figure ). Across all clades, nonhuman primate actifensins, when present, appear to exhibit ancestral characteristics. However, no distinct ancestral pattern was observed among human PAFs across the clades. In some cases, such as clades 2 and 3, PAFs (including those represented by Neanderthals dating to 100,000 years ago) shared complete amino acid sequence identity with their MAF counterparts. Notably, only two clades (1 and 6) showed PAF as an ancestral lineage (Figures b and , Supplementary Figure 17) with clade 6 also including a nonhuman actifensin basal to the ancestral human PAFs (Figures and a). Genetic and structural adaptation of paleo and modern oral actifensins. (a) The phylogenetic tree represents all mature peptide sequences from actifensins recovered from DCMs and from the dereplicated sequences found to be homologues from all oral Actinomyces genomes (Supplementary Figure 16). Four mature peptides, labeled as PAF/MAF I and PAF/MAF II, were selected for further structural and functional analysis. (b) Sequence alignment illustrates sequence identity along the lineage, residues are colored to display the respective physicochemical properties. The bar plot at the bottom shows the frequency of each amino acid residue at every alignment position. The bar height reflects the distribution frequency of residues, while the color encodes selection pressure (light gray indicates no detected selection, hatching indicates conserved residues, and black denotes positions under positive or diversifying selection) based on pervasive and episodic selection pressure analyses. The estimated positions of the three main features of the secondary structureone α-helix and two β-strandsare highlighted. (c,d) Tertiary protein structures represent selected members from the seven actifensin phylogroups. Protein surface is shown from two perspectives, with the second view rotated 180° relative to the first. While (c) shows the electrostatic surface potential of the peptides, with blue regions indicating positive charge, red regions indicating negative charge, and white regions representing neutral areas, providing insight into potential electrostatic interactions, (d) displays surface hydrophobicity, where yellow regions represent hydrophobic areas and turquoise regions indicate hydrophilic areas, highlighting the peptides’ potential interactions with aqueous or lipid environments. Construction of a consensus hybridization network using amino acid and nucleotide alignments (Supplementary Figure 18) showed a common evolutionary origin for the seven actifensin phylogroups, with varying levels of reticulation and divergence (Figure b). Regardless of whether the alignment was based on structural features, amino acids or nucleotide sequences, the resulting lineages showed minimal variation (Supplementary Figure 18). Although each branch in the split network contains a different degree of splitting, all diverging branches are connected at the base in a single line, suggesting a shared evolutionary history and the presence of a common ancestor for this family of proteins. The polygonal connections between the splits, more prominent in clade 6, indicate conflicting signals signifying possible “ongoing” recombination events, horizontal gene transfer, or purifying selection pressures acting on some sites on the sequences on those branches (Figure b). Indeed, analyses of different models indicating pervasive (constant selection pressure) and episodic diversifications along the mature peptide sequence of all seven phylogroups showed nine conserved amino acid residues across oral-derived actifensins (Figure b and Supplementary Figure 19). Seven of these amino acid residues under positive selection pressure are located within distinct α-helical or β-strand secondary structure elements (Figure b). Secondary Structure and Antimicrobial Activity of Modern and Ancient Actifensins From the sequence-based predicted secondary structures of representative peptides from each clade, we found that all peptides exhibited a conserved secondary structure consisting of one α-helix and two β-strands, possibly connected via three disulfide bonds (Supplementary Figure 20). Most peptides showed amphipathic character with more extensive positively charged regions and negatively charged areas are localized in specific patches. While some structures are predicted to have more compact 3D structures, the majority exhibit extended surface conformations (Figure c,d). To confirm the theoretically predicted secondary structures of the peptides, we recombinantly expressed two pairs (I and II) of ancestral PAFs and their related MAFs in Escherichia coli as thioredoxin fusion proteins. The mature peptides were purified using a combination of affinity and chromatographic separation methods following removal of the fusion protein by enterokinase cleavage. Analysis of the secondary structure using circular dichroism spectroscopy revealed that all peptides display the expected secondary structure content and that the composition of α- and β-helical content derived from experimental data is in good agreement with the respective theoretically predicted secondary structures (Supplementary Figure 20a). Molecular dynamic (MD) simulations of the peptides in water during 100 ns revealed an overall molecular compactness supported by consistent radius of gyration (Rg) values and structural stability during the entire simulation with almost no alterations in root-mean-square deviation (RMSD). Root mean square fluctuations (RMSF) were only observed in the loop region between the two antiparallel β-strands (Supplementary Figure 20b). The structural stability was experimentally confirmed by CD spectroscopy using thermal denaturation of the recombinant peptides; all exert high melting temperatures (T m) of >70 °C in all cases and can refold after cooling (Supplementary Figure 21). We tested the antimicrobial activities of the peptides against three Gram-positive strains, which were previously shown to be susceptible to actifensins, , representing diverse branches of two bacterial phyla: Bacillota, Bacillus megaterium and Pediococcus acidilactici; and Actinomycetota, Lysinibacillus fusiformis. Based on these assays, peptides of group II were more potent compared to the peptides of group I. All MAFs exhibited lower MIC values than their ancient counterparts, except for MAF I, which showed a slightly higher MIC against P. acidilactici than its PAF counterpart. None of the peptides were active against Gram-negative E. coli K12 in the Pseudomonadota phylum (Figure ). Functional adaptation of paleo- and modern actifensins (PAF/MAF). (a) The pruned phylogenetic tree illustrates the clades in which PAF/MAFI and PAF/MAFII are located (see Figure ). In both clades, the paleo-representative is positioned basal to its modern counterpart, suggesting potential structural and functional divergence over an evolutionary period. (b) The secondary structures predicted for each member of the PAF/MAF I and PAF/MAF II groups. Superposition of the peptide structures was performed in PyMOL using the fit command. For PAF/MAF I, the calculated root-mean-square deviation (RMSD) was 0.941 Å, while PAF/MAF II yielded an RMSD of 1.280 Å, indicating a high level of structural similarity among the compared peptides. (c) The minimum inhibitory concentration (MIC) values are expressed in μM and represent the antimicrobial activity of each peptide against four different bacteria, revealing functional differences among the peptide variants. No antimicrobial activity was observed for E. coli, i.e., its MIC could not determined (n.d.). Discussion The dramatic increase of antimicrobial resistance over the past century has substantially reduced the efficacy of treatment options for previously curable diseases. Antimicrobial resistance is now a leading cause of mortality worldwide, thus posing a major threat to the world’s healthcare systems. Addressing this urgent challenge requires innovative bioprospecting strategies to discover novel antimicrobial peptides that are intrinsically less prone to the development of bacterial resistance. AMPs more effectively augment conventional antibiotics due to their ease of modification, e.g., alteration of their amino acid sequence, to circumvent resistance. However, they remain understudied due to outdated models of their evolution and activity. Recently, there has been a growing interest in AMPs as potential therapeutic agents, not only with renewed efforts for novel AMPs identification , and the expansion of specialized AMP databases, , but also in the development of publicly available mining tools. Moreover, the increasing availability of large metagenomic data sets from previously inaccessible environments , offers unprecedented opportunities for AMP discovery. However, the current lack of workflows to systematically compare, filter, and cross-validate results of different AMP tools currently limits their potential. To address this issue, we developed AMPcombi, a workflow that simplifies the comparison of outputs from different AMP prediction tools by implementing customizable data filtering based on peptide features that generate an extensive, high quality collection of potentially bioactive AMPs ready to be validated experimentally. In addition to therapeutic applications, it is now also possible to explore the evolutionary history of AMPs due to the growing availability of homologous AMP sequences from diverse species and environments. For instance, analysis of peptide sequences, such as the prepropeptides of dermaseptin from frogs, suggests an origin in a common ancestral species approximately 200 million years ago, coinciding with tectonic events that separated the Hylidae and Ranidae families. Other studies have identified and characterized encrypted AMPs in Neanderthals and Denisovans defensins, from extinct birds and a prehistoric mammal, and cathelicidins from extinct marsupials. Defensins and defensin-like peptides represent a very diverse AMP class that has been widely examined in the context of evolutionary biology. Homologues have been identified in a broad range of organisms across the eukaryotic kingdom and, more recently, in bacteria. , In humans, defensins are a key part of the innate immune system. In the oral cavity, periodontal inflammation leads to the release of defensins and other antimicrobial compounds to restrict the growth of dental plaque. In fact, human defensins are among the most prevalent and abundant proteins identified in dental calculus proteomes. Some bacterial members of dental plaque, notably Actinomyces, are also known to produce defensins (actifensins), and the secretion of such defensins by this basal member of the dental biofilm may influence subsequent oral microbial colonization and biofilm growth. Although defensins in bacteria are most commonly found in actinomycetes and myxobacteria, defensins from the latter have previously been shown to share a common evolutionary history with invertebrate defensins. It has been proposed that defensins comprise two evolutionarily independent superfamilies, namely cis- and trans-defensins, with extensive structural and functional similarities arising through convergent evolution. Accordingly, MAFs and PAFs are classified as members of the cis-defensins superfamily. Using DCMs, we expand the current insights into structural and functional adaptations across bacterial AMPs within the primate lineage representing a period of approximately 40 million years, and through ancient DNA we provide direct and time-resolved evidence of AMP evolution within Homo over the past 102,000 years. For this purpose, dental calculus offers a magnitude of advantages due to its preservation of microbial content over time, with minimal contamination from exogenous sources. Moreover, the presence of similar bacterial genomes in DCMs from nonhuman primates and present-day humans supports the existence of a core set of highly conserved oral microbes that are suitable for long-term AMP evolutionary studies. This is also reflected by the recovered AMP cluster sets in which members derived from DCMs of different ages and types share high sequence similarity (Figure and Supplementary Figure 12). Most of the putative AMP hits in all DCM samples originate from Actinomyces (Supplementary Figures 8 and 9), which is a prominent core member of the human oral microbiome and a key early colonizer involved in dental biofilm initiation. However, in addition to its high relative abundance within DCMs, the large number of Actinomyces AMPs also likely reflects these microbes’ prolific antimicrobial capacity to produce secondary metabolites to sustain their survival in the oral cavity. Although oral actifensins appear to have a shared common ancestor, nucleotide-level differences show that they have diverged into at least seven distinct phylogroups that are independent of host type or sample age (Figure ), suggesting evolutionary adaptation. We observe distinct differences in polarity and in hydrophobicity of surface properties in these peptides (Figure ), and comprehensive evolutionary and selection pressure analyses identified distinct amino acid residues under positive or diversifying selection (Figure , Supplementary Figures 17 and 19). These sites are located near the N-termini, within the loop between the antiparallel β-sheets, and in the α-helix region. The latter is of particular interest because sites in the α-helix may significantly influence peptide specificity. In addition to variable positions within the peptide sequences, there are also conserved residues-most notably, a characteristic pattern of six cysteine residues-that classify these peptides as defensin-like (Figure ). Furthermore, a GXC motif, known as the γ-core motif, is present in all actifensin sequences. This motif is proposed to be an archetypal element of antimicrobial peptides, found in all disulfide-bridged antimicrobial peptides throughout the phylogenetic kingdoms, thereby reflecting an ancient evolutionary relationship. It is likely that specific configurations of these structural mosaics optimize the function of antimicrobial peptides against their targets in distinct environmental or physiological contexts. While the antimicrobial assays were conducted with strains previously shown to be highly susceptible to actifensins, , the objective was to assess whether sequence divergence between ancient and modern peptides results in measurable functional differences, independent of specific ecological interactions. The observed diversification in actifensin surface properties and the presence of positively selected sites suggest adaptive fine-tuning of antimicrobial activity, potentially reflecting long-term interactions between Actinomyces-derived peptides and co-occurring members of the oral microbiome. This interpretation aligns with the view that the evolution of defensin-like peptides involves balancing conservation of core structural motifs with diversification of residues that influence target specificity. Although it is not possible to directly assess the activity of paleo-actifensins against ancient bacterial strains, it is possible that these peptides exhibited higher activity in their original microbial context. Antimicrobial peptides represent a rich and underexplored source of antimicrobials for translational applications. In addition to being attractive targets for the development of clinical therapeutics, AMPs play essential roles in the regulation of symbiotic communities and thus are of key interest in understanding the formation and maintenance of healthy host-associated microbiota. Before undertaking translational applications, however, it has been argued that it is essential to understand AMP biology in natural contexts to maximize effectiveness and to avoid repeating mistakes that have led to widespread bacterial resistance. Recently, it has become possible to reconstruct oral microbiota metagenomes through the sequencing and de novo assembly of ancient DNA from dental calculus, a calcified form of dental plaque that can preserve biomolecules for millions of years. This enables the direct investigation of AMP evolution within vertebrate microbiota and presents a new source of potentially therapeutic AMPs. However, although resources for AMP discovery have dramatically expanded in recent years, AMP studies continue to be limited by the absence of a universally accepted framework for AMP identification and a lack of workflows to systematically compare, filter, and cross-validate results of different AMP tools. Conclusions In conclusion, we streamline the discovery of antimicrobial peptides (AMPs) by developing AMPcombi, a versatile workflow for the identification, structural analysis, and functional characterization of predicted AMPs from large metagenomic data sets. Applying AMPcombi to ancient and modern dental calculus metagenomes, we demonstrate that integrating evolutionary insights with structure–function analyses, exemplified by PAFs and MAFs, offers a deeper understanding of AMP evolution and paves the way for future efforts to optimize these molecules for clinical applications. In an era of rising antibiotic resistance, the integration of bioinformatics, evolutionary biology, and experimental validation is essential to fully harness the therapeutic potential of AMPs. Supplementary Material