Precision Antimicrobial Therapy Against Fusobacterium nucleatum Using Bioengineered Probiotics Expressing Guided Antimicrobial Peptides (gAMPs).

Ankan Choudhury: conceptualization, investigation, writing – original draft, methodology, validation, visualization, writing – review and editing, formal analysis, data curation. Colin Scano: formal analysis, data curation, methodology. Allison Barton: methodology, formal analysis, data curation. Christopher M. Kearney: conceptualization, resources, supervision. K. Leigh Greathouse: conceptualization, investigation, funding acquisition, writing – original draft, writing – review and editing, methodology, formal analysis, project administration, supervision, resources.

The polymicrobial community was cultured from human faecal samples collected under the study published as ‘Randomized control trial of moderate dose vitamin D alters microbiota stability and metabolite networks in healthy adults’ ( https://doi.org/10.1128/spectrum.00083‐24 ) and was done with the consent of the participants. The study was approved by the Institutional Review Board (IRB) (1845028–2) at Baylor University and registered under the identifier NCT05387876 at ClinicalTrials.gov .

We first demonstrated the binding affinity of the Statherin‐derived guide peptide ligand by synthesising eGFP fused with the peptide ligand (stat‐eGFP) and quantifying the binding with flow cytometry post‐incubation with F. nucleatum (CRC Strain) and B. fragilis (non‐ Fusobacterium off‐target representative). Both the median fluorescence and the percent of fluorescent positive events were significantly higher ( p  < 0.01, ANOVA) for F. nucleatum incubated with stat‐eGFP than with either native eGFP‐treated or untreated F. nucleatum (Figure  2A,B ). The fluorescence values for native eGFP treatment were higher than the untreated sample, though not significant. In contrast, B. fragilis treated with either native or stat‐guided eGFP exhibited no significant difference between themselves ( p  = 0.58, ANOVA), although they were more fluorescent than the untreated samples ( p  < 0.05, ANOVA). Thus, we demonstrate that eGFP with the stat‐guide ligand peptide preferentially attaches to the F. nucleatum cells compared to native eGFP, unlike in the off‐target bacterium B. fragilis . These data indicate that the Statherin 6‐mer peptide is an effective ligand for F. nucleatum and may serve as a guide for other proteins or peptides. The Statherin‐derived guide peptide allows preferential attachment to F. nucleatum and exhibits targeted biofilm inhibition. (a) Median fluorescence intensity (MFI) of GFP in B. fragilis and F. nucleatum following treatment with eGFP, Stat‐eGFP and untreated controls. Statistically significant differences in GFP binding to F. nucleatum are indicated ( p  < 0.05). Error bars represent the standard deviation of three independent experiments. (b) Percent of events positive for GFP, as measured by flow cytometry, comparing Stat‐eGFP binding to F. nucleatum to B. fragilis . Statistical significance between treatments is noted ( p  < 0.05). (c) Biofilm inhibition of F. nucleatum and B. fragilis strains in the presence of increasing concentrations of synthetic AMPs (Ovispirin and CPSC5) and their guided versions (Stat‐Ovispirin and Stat‐CPSC5). The inhibition is plotted as a percentage of biofilm formation in comparison to the blank control, which is represented by the 100% line (grey‐dotted). We performed statistical comparison shown by pooling the biofilm inhibition from both non‐ Fusobacteria species against biofilm inhibition from both Fusobacteria strains for that AMP/gAMP at different dosages of the same (ANOVA, n  = 6). We omitted the 32 μM from the figure as almost all versions of AMP/gAMP were fully inhibitory against any of the bacteria used and hence did not have any statistical significance. Error bars represent the mean ± standard deviation of 6 replicates (2 technical replicates of 3 biological replicates). Statistical significance is indicated as p  < 0.05 (*), p  < 0.01 (**), p  < 0.001 (***) and p  < 0.0001 (****). Also see IC 50 values calculated from the dose–response curve (Table  1 ). Next, using synthetic versions of the antimicrobial peptides, Ovispirin and CPSC5, and their guided analogs (stat‐CPSC5 and stat‐Ovispirin), we demonstrated their effect on biofilm formation in vitro in a range of 0.5 μM to 128 μM. The synthetic peptides and their guided analogs were tested using two strains of F. nucleatum , one derived from CRC tumour samples ( F. nucleatum SB‐CTX3Tcol3) and another derived from an oral lesion ( F. nucleatum subsp. nucleatum Knorr—ATCC 25586), along with well‐described commensal strains of B. fragilis ATCC 43858 (anaerobic) and E. coli K12 ATCC 10798 (aerobic), as representative off‐target bacteria. Using crystal violet assays for detection of biofilm formation, we demonstrate that both the AMPs and their corresponding gAMPs have similar inhibitory effects against both strains of Fusobacterium (Figure  2C ) while the gAMPs showed a reduced biofilm inhibition towards the two non‐ Fusobacteria species. This differential biofilm inhibition was seen between 0.5 μM and 16 μM, as depicted in the figure. The two off‐target bacteria demonstrated two different responses, with B. fragilis less affected by AMP treatment while E. coli being more sensitive. However, both demonstrated a differential drop in sensitivity to gAMP compared to AMP. Fitting the biofilm inhibition percentage with the corresponding dose of AMPs/gAMPs in the drc package (R 4.3.1), we obtained a dose–response curve to mathematically deduce the IC 50 values (dose at which 50% of biofilm inhibition occurred) of each AMP/gAMP against each bacterium (Table  1 ). While there was no difference in IC 50 observed between each AMP and its corresponding gAMP against either F. nucleatum strain (0.71–2.15 μM for ovispirin, 0.91–2.3 μM for Stat‐ovispirin, p  = 0.83, 1.42–5.6 μM for CPSC5, 1.3–5.7 μM for Stat‐CPSC5, p  = 0.76, ANOVA), this differential increased drastically when using the gAMP against the off‐target bacteria, B. fragilis (23 μM for ovispirin, > 100 μM for Stat‐ovispirin, 18 μM for CPSC5, > 100 μM for Stat‐CPSC5, p   30 μM for Stat‐ovispirin, 1.3 μM for CPSC5, > 20 μM for Stat‐CPSC5, p  < 0.01, ANOVA), indicating a significant preferential inhibition caused by the presence of the guide peptide. Against these off‐target bacteria, the IC 50 of the gAMPs was 4–7 fold higher (less toxic) than the AMPs for E. coli and B. fragilis , causing the IC 50 values for the gAMP treatments to exceed the highest concentration used in our assays. The presence of the guide also lessened the biofilm virulence phenotype for E. coli , with gAMP‐treated off‐target E. coli demonstrating significantly more biofilm formation than AMP‐treated E. coli , while both AMP and gAMP treatment led to the same decrease in biofilm formation of the target bacterium F. nucleatum (Figure  2C ). Importantly, neither guided nor unguided AMPs showed significant cytotoxicity against human Caco2 or HT29 cells (Figure  S1 ). Combining this lack of cytotoxicity with the low IC 50 for gAMP treatment against F. nucleatum , we calculate a therapeutic window of approximately 8‐fold (2–16 μM). The fold calculations were made by dividing the IC 50 values of a gAMP against one bacterium versus its analog AMP. The 8‐fold therapeutic index was calculated by dividing the lowest concentration at which the gAMP showed at least 50% inhibition/toxicity towards mammalian cells (HT29 or Caco2) by the lowest concentration at which it showed at least 50% inhibition/toxicity towards the pathogen, that is, F. nucleatum . This gives the window of dosage in which the gAMP can be used to treat F. nucleatum infection where it will be toxic towards the bacteria but not toxic towards the host. IC 50 values of the synthetic AMPs—calculated by fitting biofilm inhibition assay data in a dose–response curve using ‘ drc ’ package (R). In summary, gAMP treatment shows a clear differential toxicity against F. nucleatum , sparing the off‐target E. coli and with no significant toxicity against human cells. Furthermore, B. fragilis was less susceptible to AMPs in any form and demonstrated even less susceptibility to the gAMPs compared to the AMPs tested. To better understand the effects of the gAMPs on the growth of F. nucleatum , the growth kinetics of planktonic cultures of two non‐ Fusobacterium species— E. coli and B. fragilis — and two Fusobacterium strains—Oral and CRC‐derived—were incubated with either AMPs or their gAMP analogs. This growth kinetics analysis showed a pattern of differential inhibition (Figure  3A–D ). For both the Fusobacterium strains, the unguided and guided versions of the AMPs reduced the growth, with inhibition of growth beginning between 2 and 8 μM among the 4 peptides, which continued for over 48 h and up to 72 h (Figure  3A,B ). In contrast, for the non‐ Fusobacterium species, only the unguided AMPs demonstrated any inhibitory effect below 8 μM, while the gAMPs either had no inhibition of growth or exhibited inhibition below 32 μM (Figure  3C,D ). For B. fragilis , the bacterium that was most resistant and was minimally affected by all the AMPs, the guided CPSC5 and Ovispirin did not subdue growth, even at 32 μM, while the unguided version was effective at 8 μM. For E. coli , the AMPs started inhibiting growth even at 8 μM, but the gAMPs had inhibitory effects beyond 16 μM, with more pronounced inhibition at 32 μM. This analysis demonstrates that both AMPs and gAMPs inhibit the growth of F. nucleatum strains at low concentrations (< 4 μM), making these peptides clinically viable. Moreover, gAMPs show minimal toxicity towards non‐ Fusobacterium species compared to unguided AMPs, highlighting their selective toxicity and preferential action against F. nucleatum over the commensals. Selective inhibition of bacterial growth by guided antimicrobial peptides (gAMPs). Growth kinetics of planktonic cultures for two Fusobacterium strains—Oral/ATCC 25586 (a) and CRC/Strain 14 (b)—and two non‐ Fusobacterium species, non‐enterotoxigenic B. fragilis (c) and E. coli (d), were measured in the presence of both unguided (red) and guided (blue) antimicrobial peptides (AMPs—CPSC5 and Ovispirin) across a concentration gradient (2 μM to 32 μM). Optical density (OD600) measurements were recorded over 60 h (Oral strain) and 40 h (CRC strain). Error bars represent standard deviation from 6 replicates (2 technical replicates of three biological replicates) for each concentration at each timepoint. Once the gAMPs demonstrated sufficient targeting abilities, probiotic strains expressing gAMPs and control AMPs were constructed next. The open reading frames (ORFs) of the antimicrobial peptides (AMPs), CPSC5 and Ovispirin, along with their guided analogs (gAMPs), were cloned into the pMSP3545 plasmid, as described in the Methods. This plasmid includes the nisin‐inducible PNisA promoter and the usp45 signal peptide, which ensures the secretion of the peptides into the extracellular environment (Figure  4A ). The engineered plasmids were introduced into competent Lactococcus lactis MG1363 cells via electroporation. Successful cloning was confirmed through colony PCR using ORF‐specific primers, followed by Sanger sequencing to verify the accuracy of the insertion. Based on these results, only the bioengineered probiotics expressing CPSC5 and Stat‐CPSC5 were selected for further experiments, as the Ovispirin and stat‐Ovispirin constructs failed the Sanger sequence screening. Engineered Lactococcus lactis with inducible plasmids demonstrates targeted inhibition of F. nucleatum compared to non‐ Fusobacterium species. (a) Schematic of the plasmids used to engineer L. lactis probiotic. The pMSP3545‐derived plasmids contain the nisin‐inducible PNisA promoter and the usp45 signal peptide for extracellular secretion of the peptides. The AMP/gAMP plasmids include ORFs for Ovispirin (left) or CPSC5 (right), with or without a Statherin‐derived guide peptide. The control probiotic plasmid was constructed using the same backbone, replacing the AMP/gAMP ORFs with the superfolding GFP (sfGFP) ORF. Only CPSC5 and Stat‐CPSC5 were successfully cloned into L. lactis . (b) Co‐culture experiments with F. nucleatum (CRC strain), Bacteroides fragilis and E. coli in the presence of control (sfGFP), AMP‐expressing (CPSC5), or gAMP‐expressing (Stat‐CPSC5) L. lactis probiotics. The relative number of bacterial genomes remaining in the co‐culture was quantified by species‐specific qPCR after 24 h of probiotic treatment and plotted against the CFU of probiotic added. Error bars represent standard deviation from three biological replicates, and statistically significant differences (ANOVA, p  < 0.05, n  = 6) between guided and unguided AMPs are indicated by asterisks. To further explore the specificity and potential clinical application of the bioengineered L. lactis probiotics, we conducted an in vitro co‐culture experiment to assess the targeted inhibition of F. nucleatum compared to non‐ Fusobacterium commensal species. The bioengineered probiotics expressing CPSC5 (AMP) and Stat‐CPSC5 (gAMP) demonstrated selective toxicity towards F. nucleatum when co‐cultured with non‐ Fusobacterium commensal species. Either the AMP‐ or gAMP‐expressing probiotic was co‐cultured at a gradient of seven bacterial titers, ranging from 10 9 to 10 12  CFU/mL, against an initial seed inoculation of F. nucleatum (CRC strain), B. fragilis , or E. coli . After 24 h of incubation, initially induced with 100 ng/mL of nisin in a 96‐well plate format, the co‐cultures were pelleted, and genomic DNA was extracted. Species‐specific qPCR was used to quantify the target bacteria and estimate the remaining cell counts after incubation with the engineered probiotic. The results showed that the AMP‐expressing probiotics inhibited the growth of all three bacterial species in a dose‐dependent manner, with the most significant reduction observed in F. nucleatum (Figure  4B ). In contrast, the gAMP‐expressing probiotics selectively inhibited F. nucleatum as the probiotic dose increased, while the inhibitory effects on B. fragilis and E. coli were significantly less. Again, B. fragilis was relatively resistant to AMP toxicity, making comparisons between gAMP and AMP treatment much less pronounced than that seen for E. coli . The toxic effect of the gAMPs against F. nucleatum was 10 3 to 10 4 times greater than against B. fragilis and E. coli at higher probiotic titers (~10 11  CFU/mL), a statistically significant difference ( p  < 0.05, ANOVA). Interestingly, an inhibitory effect was observed across all three bacterial species, even in the control probiotic expressing super‐folding GFP (sfGFP). This suggests that L. lactis may have inherent antimicrobial properties when co‐cultured at high densities, possibly due to resource competition or innate toxicity, contributing to the observed inhibition. To determine whether the gAMP‐engineered probiotic could selectively target F. nucleatum while preserving the overall microbial diversity in a gut‐like environment, we conducted a series of experiments using an in vitro human gut microbiome‐derived polymicrobial community model. The efficacy of the bioengineered probiotics (AMP, gAMP and Control) was tested within a polymicrobial community spiked with F. nucleatum (CRC strain) 48 h culture (~6.6 × 10 9 CFU/mL) at 20% v/v in each well, with resultant F. nucleatum content of ~1.3 × 10 9 CFU/mL. This polymicrobial community was derived from anaerobic cultivation of rejuvenated faecal samples from three human subjects (denoted as FS04, FS16 and FS36), designed to mimic the composition of their gut microbiota. The communities were incubated with F. nucleatum and probiotic cultures (20% v/v with ~3 × 10 10 CFU/mL culture, resultant L. lactis content ~6 × 10 9 CFU/mL)—either AMP (CPSC5), gAMP (Stat‐CPSC5), or Control (sfGFP). We took samples at 24‐ and 48‐h intervals, lysed them, and extracted genomic DNA using the phenol‐chloroform method. The extracted genomic DNA was then analysed using both qPCR with F. nucleatum ‐specific primers and 16S rRNA (V4) sequencing (Illumina) to assess the overall microbial composition. The 16S rRNA sequencing revealed that, while a diverse array of bacteria remained in the polymicrobial communities following each treatment, the reduction of F. nucleatum titre was most prominent in the samples treated with the engineered probiotics, including the control probiotic, indicating that the probiotic itself had a great effect on the reduction of Fusobacterium levels. This reduction was in stark contrast to the untreated controls, where F. nucleatum abundance increased considerably from around 10% to over 25% in 48 h (Figure  5A,B ). Other prevalent genera, including Enterococcus , Peptoniphilus , Eubacterium , Escherichia , Bacteroides , Sellimonas , Negativicoccus and Paraclostridium showed little to no change in relative abundance compared to Fusobacterium , indicating that the gAMP exhibits antimicrobial activity towards FomA‐expressing F. nucleatum without significantly affecting other members of the microbiota. Probiotic only and media‐only controls were used as negative controls; however, they had minimal differences in microbial composition compared to their corresponding time 0 (data not shown). Overall, these results suggest that FomA‐expressing Fusobacterium strains, which demonstrate higher pathogenicity, are more susceptible to our engineered probiotic targeting FomA‐expressing F. nucleatum . The gAMP probiotic maintained the diversity of the microbiota in the in vitro polmicrobial community. Heatmaps of the genus‐level (a) and family‐level (b) relative abundance. Polymicrobial community rejuvenated from human faecal microbiome sample spiked with F. nucleatum culture and then treated with bioengineered probiotics secreting CPSC5 (AMP), Stat‐CPSC5 (gAMP) and sfGFP (Control) at a ratio of 3:1:1 respectively ( n  = 9 at each timepoint, 3 biological replicates for each of the 3 faecal samples used). Three alpha diversity indices—Shannon (c), Faith's Phylogenetic diversity (d) and species richness (e), showing the ability of gAMP probiotic treatment to maintain the biodiversity of the polymicrobial community post spiking with F. nucleatum , compared to AMP and control probiotics. No significant changes either between any consecutive timepoints or the overall timeframe. ANOVA was used for testing for significance between treatments with p  < 0.05 considered significant. Measuring the alpha diversity of the samples using three different indices revealed that all probiotic treatments preserved microbial community diversity after exposure to F. nucleatum . Among the treatments, the gAMP probiotic exhibited the smallest shift in microbial diversity from 0 to 48 h. Using the Shannon index as a measure of diversity (Figure  5C ) indicated a decrease in diversity at 24 h in all probiotic‐treated samples, likely due to the proliferation of F. nucleatum . However, diversity began to recover in the following 24 h. This decline at 24 h was statistically significant in the control and AMP‐treated samples but not in the gAMP‐treated samples, with no significant differences in the recovery phase across all treatments. These results suggest that the gAMP probiotics caused the least disruption to microbial diversity, showing minimal and non‐significant changes. In the samples that were spiked with F. nucleatum without probiotic treatment, diversity significantly declined at both 24 and 48 h ( p  = 0.0019 and 0.004 for Shannon Index, p  = 0.0078 and 0.0005 for Faith PD, p  = 0.031 & 0.004 for species richness, ANOVA). Interestingly, this recovery pattern was not observed when analyzing the Faith Phylogenetic Diversity and Species Richness (Figure  5D,E ). Both of these indices showed a steady decline in diversity across all three probiotic treatments and in the F. nucleatum ‐only controls, with the gAMP‐treated samples exhibiting the least drastic and non‐significant changes in microbial diversity. To assess whether the bioengineered gAMP probiotic could effectively reduce F. nucleatum in a complex gut microbiome‐derived polymicrobial community without disrupting the overall microbial balance, we conducted a series of in vitro experiments. Using both 16S rRNA sequencing and qPCR, we found that the gAMP probiotic was highly effective in reducing F. nucleatum even within a complex polymicrobial community. Although all three probiotic variants demonstrated some antimicrobial activity, the gAMP probiotic produced the most significant reduction in F. nucleatum titers after 48 h of incubation. Specifically, the F. nucleatum abundance, measured by Centered‐Log Ratio (CLR) analysis, was lowest in the gAMP‐treated samples compared to the AMP‐treated ones, with a statistically significant reduction observed for the gAMP probiotic (Figure  6A–D ) ( p  = 0.031, ANOVA). In the probiotic‐treated samples, there was a slight increase in F. nucleatum CLR at the 24‐h mark, followed by a notable decline, which was significant only in the gAMP‐treated samples. In contrast, samples treated only with F. nucleatum exhibited a steady increase in CLR over time. Interestingly, these differential reductions in F. nucleatum abundance in gAMP vs. AMP/Control probiotic‐treated samples were not fully reflected in the qPCR results, where similar reductions were observed across all probiotic treatments. Only the untreated control samples showed a significant increase in F. nucleatum titre at both time points ( p  = 4.1e‐05, ANOVA) (Figure  6A–D ). This discrepancy between the two methods may be attributed to the nature of the measurements: qPCR focuses solely on the number of gene copies for the target species, whereas CLR analysis accounts for the relative abundance of F. nucleatum in relation to all other bacteria in the community, offering a more comprehensive view of its presence within a polymicrobial environment. Amplicon 16S rRNA sequencing produces compositional data, where abundances are relative and must sum to one; applying a CLR expresses F. nucleatum abundance in a natural log form relative to the geometric mean of all taxa in the sample. This means that even if F. nucleatum stays constant, its CLR value can shift if other microbes bloom or decline. Additional differences arise from 16S gene copy number variation, amplification bias, and the use of pseudocounts for low‐abundance taxa. Thus, CLR is a metric more appropriate for compositional data, while the number denoted by qPCR depends only on the raw abundance of F. nucleatum . Thus, the qPCR data shows us that regardless of the probiotic used, the raw number of F. nucleatum decreased in all samples with time, but CLR analysis shows us that this reduction, when seen relative to the abundance of all the microbes present in the samples, in F. nucleatum abundance is significant only in the gAMP probiotic treated sample, which is a more compositionally relevant result. The bioengineered probiotic effectively targets F. nucleatum in a gut microbiome‐derived polymicrobial community without disturbing the overall microbiota. (a) The effect of control probiotics, AMP, or gAMP probiotic on reduction in F. nucleatum titers after 48 h in an in vitro polymicrobial community ( n  = 9 at each timepoint). The abundance of F. nucleatum was measured by Centered‐Log Ratio (CLR), (ANOVA). (b) F. nucleatum titre measured by qPCR of the sample samples. The y‐axis gives the approximate number of F. nucleatum left in the sample at that timepoint ( n  = 9 at each timepoint) after treatment with the given probiotic. (ANOVA, p  < 0.05). (c) Compositionality Corrected by REnormalization and PErmutation (CCREPE) analysis of bacterial genera that were positively (red) or negatively (blue) correlated with F. nucleatum abundance. This analysis was used to create the Fusobacterium Dysbiosis Index. (d) The effect of probiotic, AMP, or gAMP probiotics on community structure, as indicated by the Fusobacterium Dysbiosis Index, which was calculated by comparing the abundance of F. nucleatum and its positively correlated genera against the negatively correlated genera. ( n  = 9 at each timepoint; Dysbiosis Index was calculated as described in Equation ( 1 ) with genera that fit the following criteria: |correlation| > 0.25, FDR‐adjusted p  < 0.15 by Benjamini‐Hochberg method). Next, we used Compositionality Corrected by REnormalization and PErmutation (CCREPE) analysis to identify bacterial species either positively or negatively correlated with F. nucleatum abundance (|correlation| > 0.25, FDR‐adjusted p  < 0.15) across the samples. This allowed us to determine which bacteria increased as F. nucleatum levels rose, and which bacteria decreased in response. In the gAMP probiotic‐treated samples, bacteria positively correlated with F. nucleatum showed a consistent reduction, while those negatively correlated were either maintained or increased in abundance at both time points (Figure  6C ). In contrast, the samples treated only with F. nucleatum exhibited the opposite trend. This information was used to develop the Fusobacterium Dysbiosis Index (Equation  1 ), which quantified the extent of dysbiosis caused by introducing F. nucleatum into the community. Using this index, the gAMP probiotics showed the most effective performance in vitro (Figure  6D ). The gAMP‐treated samples exhibited a significant reduction in dysbiosis, indicating that not only did the gAMP probiotics reduce F. nucleatum abundance, but they also preserved microbial diversity more effectively than AMP probiotics by selectively targeting the pathogen. Although all three probiotic treatments showed a similar pattern of a small increase in dysbiosis at 24 h followed by a reduction at 48 h, the dysbiosis index in the gAMP‐treated samples was significantly lower than the indices at both the 0‐h and 24‐h time points—an effect not seen with the AMP or control probiotics. These results confirm that the gAMP probiotic expressing Stat‐CPSC5 demonstrates preferential antimicrobial activity against F. nucleatum , consistent with findings from both monoculture and polymicrobial community experiments.

Our research presents critical insights into the specificity and efficacy of Statherin‐derived guide peptides as a first step in creating an engineered probiotic selectively targeting F. nucleatum , highlighting the potential for targeted antimicrobial therapies. By utilising a statherin‐derived guide peptide fused to an enhanced green fluorescent protein (eGFP), we observed preferential attachment to F. nucleatum over B. fragilis , a non‐Fusobacterium commensal anaerobe (Elsaghir and Reddivari  2024 ). Flow cytometry revealed significantly higher fluorescence in F. nucleatum treated with stat‐eGFP compared to both native eGFP and untreated controls, indicating a specific interaction (Getz et al.  2012 ). In contrast, B. fragilis exhibited no significant difference between native eGFP and stat‐eGFP treatments, confirming the specificity of the statherin‐derived peptide in guiding attached molecules to bind selectively to F. nucleatum . This specificity is mediated by the interaction between the Statherin‐derived peptide and the membrane protein FomA, as shown by previous studies in which F. nucleatum mutants lacking FomA lost binding specificity to the guide peptide ligand (Nakagaki et al.  2010 ). Recently, a study was published that also used the same Statherin‐derived peptide as a guiding ligand with AMPs to target the FomA protein in F. nucleatum in a mixed microbial population (Liu, Wang, et al.  2024 ). This specificity is crucial in designing targeted therapies that minimise the impact on non‐target microbiota, thereby reducing potential off‐target effects and preserving the delicate balance within the gut microbiome (Lozupone et al.  2012 ; Zhao et al.  2023 ). Biofilms provide a protective environment for bacteria, increasing their resistance to antimicrobial agents and immune responses (Shree et al.  2023 ). Our biofilm inhibition assays further demonstrated that statherin‐derived ligand‐guided AMPs (gAMPs), including stat‐CPSC5 and stat‐Ovispirin, maintained strong inhibitory effects against F. nucleatum biofilms despite the addition of the N‐terminal guide peptide fragment, while showing reduced efficacy against non‐target bacteria such as B. fragilis and E. coli . This selective inhibition is particularly important in the context of treating bacterial infections linked to biofilm formation. The ability of gAMPs to selectively inhibit F. nucleatum biofilms without affecting commensal bacteria suggests their potential as therapeutic agents for biofilm‐associated infections, particularly in the oral cavity and gastrointestinal tract (Pignatelli et al.  2023 ; Liu, Yu, et al.  2024 ). Growth kinetics data also supported the specificity of gAMPs, showing that both guided and unguided AMPs inhibited F. nucleatum at low clinically relevant concentrations. However, gAMPs exhibited significantly lower toxicity towards non‐ Fusobacterium bacteria. For example, in E. coli , gAMPs had a reduced inhibitory effect even at higher concentrations, whereas unguided AMPs were effective at much lower doses with a 6‐fold increase in IC 50 values. This selective growth inhibition emphasizes the therapeutic potential of gAMPs, where it is essential to target pathogenic bacteria without disrupting the broader microbial community (Lozupone et al.  2012 ; Zhao et al.  2023 ). Such precision could prevent the adverse effects commonly associated with broad‐spectrum antibiotics, such as dysbiosis and the emergence of antibiotic‐resistant strains. These results also confirm that the gAMP strategy is effective against both planktonic and biofilm forms of the target pathogen. The creation of bioengineered probiotics expressing gAMPs introduces an innovative approach to delivering targeted antimicrobial therapy within the gut. There has been recent research done using the approach of targeted antimicrobial peptide against F. nucleatum using the FomA protein (Liu, Wang, et al.  2024 ), but it did not offer a suitable delivery platform for the delivery of the peptides in situ of infection. Engineered Lactococcus lactis developed by us, expressing stat‐CPSC5, exhibited selective toxicity towards F. nucleatum when co‐cultured with non‐target bacteria. We observed a significant reduction in F. nucleatum populations, as indicated by quantitative PCR and species‐specific genome quantification, without comparable effects on B. fragilis and E. coli . This differential targeting is advantageous for maintaining a healthy gut microbiota while combating pathogens. Additionally, the inherent antimicrobial activity of L. lactis suggests it could serve as an effective vehicle for delivering therapeutic agents, enhancing its utility in microbial therapy. One of the most significant findings of this study is the ability of gAMP‐expressing probiotics to preserve the diversity of a polymicrobial community while effectively reducing F. nucleatum levels. 16S rRNA sequencing showed that gAMP probiotics maintained microbial diversity to a greater extent than AMP or control probiotics. Alpha diversity (Shannon Index) demonstrated that while all samples showed a decrease in diversity at 24 h, likely due to F. nucleatum proliferation, the gAMP probiotic‐treated samples exhibited a less significant decrease and a more complete recovery by 48 h. This suggests that gAMPs can selectively reduce pathogenic bacteria while minimizing collateral damage to the overall microbial community. Importantly, FomA expression is an important factor in F. nucleatum 's pathogenicity and facilitation through inflammatory and adhesive actions (Martin‐Gallausiaux et al.  2020 ; Zheng et al.  2024 ), which highlights the importance of this result. Moreover, compositional analysis using CCREPE (Schwager et al.  2024 ) revealed that gAMP probiotics were most effective in reducing the dysbiosis index, a measure of microbial imbalance caused by F. nucleatum overgrowth (Wu et al.  2024 ). This reduction in dysbiosis correlated with the selective decrease in bacteria positively associated with F. nucleatum and the preservation or increase of bacteria negatively associated with F. nucleatum . These findings are significant because maintaining microbial diversity and function is essential for gut health, and significant and/or prolonged disruptions can lead to conditions such as inflammatory bowel disease, cancer and other metabolic disorders (Hills et al.  2019 ). Our findings have broad implications for developing targeted antimicrobial therapies. The specificity of Statherin‐derived guide peptides for F. nucleatum opens up the possibility of using gAMPs in clinical settings to treat infections associated with this pathogen, particularly in the context of colorectal cancer, periodontal disease and even reproductive diseases such as pre‐term birth and endometriosis (Han et al.  2004 ; Vander Haar et al.  2018 ; Sun et al.  2019 ; Chen et al.  2022 ; Muraoka et al.  2023 ). Additionally, the use of bioengineered probiotics as a delivery system offers a promising avenue for administering these therapies in a controlled and sustained manner. Future research should focus on translating this therapeutic approach to in vivo models, where bioengineered L. lactis synthesizing and secreting gAMP could be administered orally for gut colonization, allowing the peptides to escape gastric proteolysis and exert antimicrobial effects in situ by preferentially binding to F. nucleatum 's FomA surface protein via the Statherin‐derived guide peptide. As a concept art of this translational research, we present a figure (Figure  7 ) that summarizes how this approach will behave in vivo. Model for in vivo application of bioengineered probiotics as a therapeutic. The probiotic schematic representation illustrates the engineered probiotic mechanism. The probiotic Lactococcus lactis is bioengineered with a Nisin‐inducible plasmid that includes the PNisA promoter and the usp45 secretion signal peptide. This vector enables the expression and secretion of antimicrobial peptides (AMPs) or guided antimicrobial peptides (gAMPs) in the gut upon oral administration. The statherin‐derived guide peptide on the gAMPs targets the FomA membrane protein expressed by F. nucleatum . Once secreted, the gAMPs preferentially bind to FomA on F. nucleatum , allowing targeted antimicrobial activity and thereby reducing F. nucleatum colonisation in the infected colon without affecting non‐target bacteria. Furthermore, the ability to maintain microbial diversity while selectively targeting pathogens could revolutionise the treatment of bacterial infections. As indicated, the current antibiotic therapies often disrupt the gut microbiota, leading to adverse effects and promoting the development of antibiotic resistance. The use of gAMPs, particularly when delivered via probiotics, could mitigate these issues by providing a more targeted approach that preserves the beneficial components of the microbiota in an easily delivered oral therapeutic. Despite the promising findings, this study has some important limitations. All experiments were performed in vitro, including biofilm inhibition, growth kinetics and polymicrobial community assays. While these models provided valuable insights into the selective activity of gAMPs against F. nucleatum , they do not fully replicate the complexity of the gastrointestinal environment, where host immune responses, nutrient availability and microbial interactions can alter therapeutic efficacy. In addition, although we demonstrated that both synthetic peptides and engineered L. lactis effectively inhibit F. nucleatum in controlled settings, we did not assess peptide stability under gastrointestinal conditions. Peptide degradation by host and microbial proteases remains a major barrier for oral delivery and may impact in vivo effectiveness. Furthermore, no pharmacokinetic, biodistribution, or toxicity data were collected for either the synthetic peptides or the probiotic strains. While our in vitro assays showed minimal cytotoxicity towards human epithelial cell lines, in vivo studies are essential to evaluate host safety, immune responses and potential off‐target effects. Explicitly recognizing these limitations provides transparency and highlights the need for future animal studies to evaluate gAMP stability, safety and efficacy before translation towards therapeutic application. In conclusion, this study demonstrates the effectiveness of Statherin‐derived guide peptides in selectively targeting F. nucleatum , inhibiting its biofilm formation and modulating its growth kinetics without significantly affecting non‐target commensal bacteria. The bioengineered probiotics expressing gAMPs further showed promise in reducing F. nucleatum within a complex polymicrobial community while maintaining microbial diversity. These findings lay a strong foundation for the development of targeted antimicrobial therapies that could offer a more precise and less disruptive alternative to traditional antibiotics.

The authors have nothing to report.

A detailed overview of the experimental workflow is visualized and described in Figure  1 . Experimental overview. The probiotic Lactococcus lactis was cloned with a Nisin‐inducible plasmid with the ORF of the AMPs/gAMPs downstream of it, to express and secrete the gAMP in vitro where the statherin‐derived guide peptide will allow the AMP to preferably attach to the FomA membrane protein on Fusobacterium nucleatum and exhibit targeted antimicrobial activity. The bioengineered probiotic was assayed by either co‐culturing with a single Fusobacterium and non‐ Fusobacterium species and analyzing by species‐specific qPCR (top) or co‐culturing with a fecal microbiome‐derived polymicrobial community spiked with F. nucleatum and analyzing through 16 s rRNA amplicon sequencing (bottom). The strains of target pathogen F. nucleatum used for the experiment were F. nucleatum SB‐CTX3Tcol3 derived from CRC tumour biopsy (gift from Dr. Suan Bullmann's group in MD Anderson Cancer Research Center, TX, USA) and F. nucleatum subsp. nucleatum Knorr (ATCC 25586) derived from Oral lesion (purchased from ATCC). Non‐target commensals used for the experiment were B. fragilis ATCC 43858 (Enterotoxigenic/ETBF), B. fragilis ATCC 25285 (Non‐Enterotoxigenic/NTBF) (both gifts from Dr. Cindy Sear's group at Johns Hopkins University, Baltimore, MD), and E. coli K12; sourced from ATCC (Manassas, VA, USA). Strains used for synthesis of eGFP and stat‐eGFP were E. coli 10β and E. coli BL21 (DE3) from New England Biolabs (Ipswich, MA, USA). Bioengineered probiotics were created using L. lactis subspecies cremoris MG1363 (LMBP 3019) acquired from the Belgian Coordinated Collections of Microorganisms (Louvain‐la‐Neuve, Belgium). Synthesis of eGFP and stat‐eGFP was done using pE‐SUMO (ampicillin) sourced from LifeSensors (Malvern, PA, USA). The cloning of AMP, gAMP and sfGFP into probiotic L. lactis was done using pMSP3545 sourced from Addgene (Watertown, MA, USA). We utilised the antimicrobial peptide database DBAASP (v3.0) to identify candidate AMPs active against F. nucleatum . Ovispirin (Yasin et al.  2003 ) and Cathelin‐related peptide SCF (CPSC5) (Ghosh et al.  2018 ) were selected as promising candidates. Through a comprehensive literature search, peptide ligand (YQPVPE) derived from the human salivary protein statherin (stat) was identified as a potential candidate for guiding ligand, which specifically binds the F. nucleatum membrane porin FomA (Nakagaki et al.  2010 ). The gAMPs were then constructed by linking the ligand to the AMP using a flexible triple glycine linker, resulting in Stat‐Ovispirin and Stat‐CPSC5 for downstream analysis (Table  S1 ). Synthetic peptides were employed in initial assays to confirm target‐specific antimicrobial activity. For the ORFs of eGFP and stat‐eGFP, we obtained them from IDT (Coralville, IA, USA) as gBlocks, with the stat‐eGFP containing the sequence YQPVPE (Nakagaki et al.  2010 ) at the N‐terminus of the eGFP protein. We amplified the ORFs using PCR, digested them with restriction enzymes, and inserted them into the pE‐SUMO vector between the MfeI and BamHI cut‐sites using T4 ligase (Thermo Fisher Scientific). We transformed the cloned plasmids into competent E. coli 10β using the heat‐shock technique to proliferate the plasmids. Transformed E. coli 10β were screened by plating on an ampicillin (50 μg/mL) agar plate and testing the colonies with colony PCR. Positive colonies were propagated in 15 mL Luria‐Bertani (LB) broth (Thermo Fisher Scientific) with 50 μg/mL ampicillin at 37°C overnight, pelleted, lysed and the extracted plasmid (using GeneJET Plasmid Miniprep Kit, Thermo Fisher Scientific) was used to transform E. coli BL21 (DE3) for protein synthesis. Positively transformed E. coli BL21 (DE3) colonies were grown in 2X Yeast‐Tryptone (YT) broth (Thermo Fisher Scientific) in a larger volume (~500 mL) at 37°C overnight and then induced with 0.5 mM isopropyl β‐d‐1‐thiogalactopyranoside (IPTG) (Thermo Fisher Scientific) for a subsequent overnight culture at 16°C. We pelleted the resultant culture at 7000× g and lysed it with 0.5 mg/mL lysozyme (Thermo Fisher Scientific) along with a freeze–thaw cycle. The thawed lysed pellet was sonicated and centrifuged at 50,000× g to obtain the supernatant containing the synthesised proteins, which were purified using Ni‐column ion affinity chromatography (Biorad), utilising the 6‐his tag found in the SUMO fusion partner of the pE‐SUMO vector. We cleaved the purified protein with SUMO protease, which cleaves the SUMO‐fusion partner and releases the synthesised protein. Flow cytometry was used to measure bacterial binding of the stat‐eGFP fusion protein against the targeted F. nucleatum (CRC strain) as well as against the off‐target bacterium, Bacteroides fragilis (ETBF). F. nucleatum and B. fragilis were both grown anaerobically (5% CO 2 , 5% H 2 , 90% N 2 ) in Columbia broth (Thermo Fisher Scientific) and brain‐heart infusion (BHI) broth (Thermo Fisher Scientific), respectively, for 48 h or until turbid, at 37°C. The bacteria were then standardised to an OD 600 of 1.0 and diluted 1:20 in 1× PBS supplemented with 125 μg/mL unguided eGFP or stat‐GFP and incubated for 30 min at 37°C, shaking at 180 rpm. We washed the cells and resuspended them in 1× PBS and flowed through a BD FACSVerse system (BD Biosciences). We excited the cells with a blue 488 nm laser using a 488/10 bandpass filter. For each sample, we measured fluorescence intensity for a total of ≥ 4000 events collected in triplicate. We analysed the data using FCS Express (De Novo Software, Glendale, CA, USA). Statistical analyses were performed for both the median fluorescence and percent of events positive for eGFP by using R 4.3.1, and all tests performed were one‐way ANOVA. Synthetic peptides (HPLC purified) for CPSC5, ovispirin and their Stat‐guided analogs (Table  S1 ) were purchased from ABI Scientific (Sterling, VA, USA). After reconstitution in sterile PBS buffer, the peptides were added to a 96‐well plate in a concentration gradient from 0.5 to 32 μM in appropriate media for the bacterium of choice (Columbia for F. nucleatum , BHI for B. fragilis , LB for E. coli ) up to a volume of 200 μL in each well. The bacteria, grown in appropriate medium either anaerobically or aerobically, were standardized to OD600 of 1.0 and 5 μL of it were added to each well, followed by incubation for 12 h aerobically for E. coli , 72 h anaerobically for F. nucleatum and 48 h anaerobically for B. fragilis . The incubation was done inside the Stratus Kinetic Microplate Reader (Cerillo, Charlottsville, VA, USA) at 600 nm to read the cell growth kinetics. The growth kinetics data were procured and analyzed using the Cerillo Labrador Software, with further statistical representation done in R. After incubation, the plate was assayed for biofilm formation using a modified protocol described by O'Toole ( 2011 ). Briefly, the medium was aspirated from the wells by pipetting, followed by fixing the biofilm with 200 μL of methanol added to each well, evaporation of the methanol by aeration/incubation at 50°C, addition of an equal volume of 1% (w/v) crystal violet solution, 3X washing with water or until the wash ran clear, dissolution of the crystal violet bound to the biofilm in 200 μL of 30% (v/v) acetic acid solution, and transfer of the crystal violet‐acetic acid solution to a fresh 96‐well plate for colorimetric assay at 600 nm. One row of the plate was used as a positive/vehicle control, and biofilm inhibition % was calculated for every other well from that control. Statistical analysis was performed using R 4.3.1, and all tests performed were one‐way ANOVA. The expression plasmid was designed by placing the open‐reading frames (ORFs) for both the guided and non‐guided AMPs into plasmid pMSP3545 (Bryan et al.  2000 ), which contained the PNisA promoter (Li et al.  2011 ) inducible by nisin—an endogenous peptide produced by L. lactis . This promoter facilitates controlled expression of the heterologous proteins and peptides. The ORFs were placed downstream of an L. lactis ‐specific secretion signal to create the bioengineered probiotic strains. Nisin‐inducible expression systems for heterologous proteins and peptides in both murine and human guts have been well established for over two decades (Steidler et al.  1998 , 2000 ; Chiabai et al.  2019 ; Arukha et al.  2021 ; Zahirović and Berlec  2022 ). The ORFs of the AMPs (Table  S1 ), codon‐optimised for L. lactis , were cloned into the pMSP3545 plasmid (Figure  4a ), which includes the PNisA Nisin‐inducible promoter and the usp45 signal peptide for extracellular secretion of the downstream peptide. The ORFs were amplified from a gBlock by PCR (1 min melting at 95°C; 35 cycles of 15 s melting at 95°C, 15 s annealing at melting temperature (T m ) + 3°C, 30 s extension at 72°C; and 5 min elongation at 72°C) using specific primers for Ovisipirin and CPSC5 and their stat‐analogs, respectively, and cloned into the pMSP3545 plasmid using restriction enzyme cut sites post agarose gel purification. The recombinant plasmid with AMPs/gAMPs was electroporated into electrocompetent L. lactis MG1363 (LMBP 3019) cells. Briefly, the procedure involved mixing thawed electrocompetent L. lactis with at least 0.25 μg of plasmid, resuspended in ice‐cold 0.5 M sucrose and 10% glycerol, electroporation in a cuvette (2000 V, 25 μF, 200 Ω) for a pulse of 5 ms, and grown out at 30°C in M17 media supplemented with 0.5% glucose (GM17). We screened for positive transformation by plating on GM17 agar plates with erythromycin (5 μg/mL). L. lactis clones engineered to express AMP or gAMP were propagated from glycerol stocks and grown in GM17 broth overnight at 30°C with erythromycin (5 μg/mL) with no shaking. We grew F. nucleatum in Columbia broth anaerobically at 37°C for 48–72 h. The L. lactis cultures were serially diluted in a 96‐well culture plate with GM17 broth to make up a volume of 100 μL with 100 ng/mL of nisin for induction of expression. To each well, 10 μL of the F. nucleatum culture was added, and each well volume was brought up to 200 μL with Columbia broth. We incubated the plate overnight in anaerobic conditions. After 24 h, the well contents were transferred to a 96‐well PCR plate. That PCR plate was sealed, heated for 15 min at 100°C, and then chilled at 4°C for 5 min. This step extracted the gDNA from the bacterial cultures by boiling. This plate was then centrifuged at 2000× g for 2 min, and the supernatant was used as the template for qPCR. We performed qPCR using primers for the FomA gene (forward: 5′‐ACTTTACCAGTTGCCCAGTT‐3′; reverse: 5′‐GGAGACCAAATGGTTCAGTAGAT‐3′, from IDTDNA, Coralville, IA, USA) to quantify F. nucleatum titre, and primers flanking the L. lactis acma gene (forward: 5′‐GGAGCTCGTGAAAGCTGACT‐3′; reverse: 5′‐GCCGGAACATTGACAACCAC‐3′, from IDTDNA, Coralville, IA, USA) were used to quantify L. lactis titre. The qPCR used SYBR Green as the amplification dye and ROX as the passive dye, and the thermal cycler had 2 min melting at 95°C followed by 40 cycles of 15 s melting at 95°C, and 1 min of annealing/extension at 60°C, ending with a melt curve. We constructed standard curves for F. nucleatum and L. lactis by determining C T values from the qPCR data for different dilutions of the gDNA extracted from the cultures of the respective bacteria (1/10, 1/100, 1/10 3 and 1/10 4 ) in the qPCR plates, and we calculated the approximate number of genomes represented in the gDNA by dividing the amount of gDNA by the average size of the genome of that species. We followed the same procedure with the off‐target bacteria where B. fragilis and E. coli were co‐cultured with serially diluted cultures of L. lactis for 24 h, and the titers of the off‐target bacteria were determined by qPCR using primers for species‐specific genes for either bacterium, DE3‐T7 polymerase for E. coli (forward: 5′‐GAAGCTTGCTTCTTTGCT‐3′; reverse: 5′‐GAGCCCGGGGATTTCACAR‐3′, from IDTDNA, Coralville, IA, USA) and Bft /Bf toxin for B. fragilis (forward: 5′‐GGTTTCAACCGTCAGGTACA‐3′; reverse: 5′‐GCGAACTCATCTCCCAGTATAAA‐3′, from IDTDNA, Coralville, IA, USA). The amount of L. lactis added to the co‐cultures of all three assays ranged from approximately 2.8 × 10 9 to 1.8 × 10 11 CFU/mL. L. lactis with sfGFP plasmid was used as a control for probiotic treatment, while the bacteria culture grown without any L. lactis added was used as a negative control. We performed statistical analysis using R 4.3.1, and all tests performed one‐way ANOVA, unless otherwise specified. We utilised three human faecal samples which were randomly selected from the placebo control arm of a separate project of our research group, a randomised control trial (approved by the Institutional Review Board (IRB) (1845028–2) at Baylor University and registered under the identifier NCT05387876 at ClinicalTrials.gov ), for which we collected daily faecal samples from 42 different subjects (mentioned in detail at the end of the article). These samples were treated individually for our experiments to capture the inter‐individual differences from exposure to our engineered probiotic. We created the polymicrobial community by reanimating faecal samples from human subjects kept frozen at −80°C in the complex media described by Li et al. ( 2019 ) for ex vivo microbiome culture in their MiPro model, which is a minimal culture media with bile salts and other essential amino acids and nutrients to simulate the environment inside the human large intestine, in anaerobic conditions at 37°C for 48 h. We added the polymicrobial culture in 96 well (2 mL, deep well) with F. nucleatum (48 h culture, ~6.6 × 10 9 CFU/mL) and probiotic (24 h culture, ~3 × 10 10 CFU/mL) of choice in the ratio 3:1:1. The media contained nisin (100 ng/mL) as an inducing agent. The probiotics used were those cloned with CPSC5 (AMP), stat‐CPSC5 (gAMP) and sfGFP (control). For samples that were F. nucleatum only negative control, the rest of the volume was made up by adding more of the complex media without nisin. The plates were incubated anaerobically at 37°C with separate plates that served as 0, 24 and 48 h timepoint samples. Each sample at each timepoint had 3 technical replicates, making the number of replicates for each treatment at each time point 9 total replicates. Post incubation, we extracted samples using Quick DNA Faecal/Soil Microbe 96 Kit (Zymo Research Corp.). We subjected the extracted gDNA to 16S rRNA Illumina sequencing and quantification of F. nucleatum content by qPCR using the FomA ‐specific primers as described in the previous section. We sequenced the extracted gDNA as per the protocol described in the Earth Microbiome Project (Caporaso et al. 2023 ). The 16S rRNA variable region V4 was amplified with PCR, using the following primers, 16S Forward Primer (515 F) with adapters: TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG GTGYCAGCMGCCGCGGTAA 3′ and 16S Reverse Primer (926 R) with adapters: 5′ GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG ACAG CCGYCAATTYMTTTRAGTTT 3′. The PCR mixture comprised 1 μL of each forward and reverse primer (12.5 μM), 5 μL of extracted DNA of approximately equal concentration from each sample, 12.5 μL of 2× Platinum Hot Smart Master Mix (Thermo Fisher Scientific), and water to make a final volume of 25 μL. The amplifications were performed under the following conditions: initial denaturation at 94°C for 5 min, followed by 35 cycles of denaturation at 94°C for 45 s, primer annealing at 50°C for 1 min, and extension at 72°C for 1 min 30 s, with a final elongation at 72°C for 5 min. We visualised the presence of PCR products by electrophoresis using a 1.5% agarose gel. We cleaned the products using Sera‐Mag select PCR clean up kit (Cytiva). We conducted a second PCR to attach the Illumina adapters and barcode index primers—I5 (forward) and I7 (reverse). After adapter and index attachment, we normalised the amplicons and pooled together in a DNA library at a concentration of 4 nM, measured by Quant‐iT dsDNA Assay Kit (Thermo Fisher Scientific). Using paired‐end sequencing on the pooled DNA library, we sequenced at 2 × 300 bp using a MiSeq Reagent Kit v3 on Illumina MiSeq platform (Illumina) at Baylor University. We evaluated, demultiplexed, and filtered the sequences using QIIME2 (version 2024.5) (Bolyen et al.  2019 ) plug‐ins on the Kodiak High‐Performance Computing (HPC) Cluster hosted by High Performance and Research Computing Services, Baylor University. Both paired reads were trimmed from the forward end, and the read length was at least 200 bp for further processing to generate the ASVs. The denoising and filtering of chimeric sequences was conducted using the DADA2 plug‐in of QIIME 2. The samples were rarefied at a depth of > 2500, which retained 97% of the samples. Taxonomic classification was performed utilizing the QIIME2‐compatible pre‐trained feature classifier based on the 16S rRNA gene database from Silva 16S rRNA 138 database. We calculated alpha diversity (Shannon, Faith Phylogenetic Diversity and Species Richness) and beta diversity (Bray‐Curtis, Aitchison's, Weighted and Unweighted Unifrac) analyses using QIIME2 plugins. The resultant α‐diversity, β‐diversity and ASV tables were downloaded and further analyzed in R 4.3.1 and RStudio platform 2023.12.1. The resultant taxonomic abundance data (at genus level) were analyzed using the CCREPE package in R (Schwager et al.  2024 ) to determine the correlation between the taxa according to relative changes in abundance upon the addition of F. nucleatum . The resultant coterie of genera (|correlation| > 0.25, p ‐value < 0.05, FDR‐adjusted p ‐value < 0.15, iterations = 1000) was then used to construct the equation for Dysbiosis Index Equation ( 1 ). Every other data visualization used default tools in the tidyverse and/or ggplot2 packages. Statistical analysis was performed using R 4.3.1 and all significance tests performed were one‐way ANOVA, unless otherwise specified. (1)

The oral pathogen, Fusobacterium nucleatum ( F. nucleatum ) , has been identified as a contributor to inflammation, inflammatory bowel diseases and cancer, particularly colorectal cancer (CRC) (Francescone et al.  2014 ). Several virulence factor proteins, including FomA, FadA, RadD and Fap2, expressed by F. nucleatum , have established connections with host cell invasion, aggregation, biofilm formation and the activation of multiple carcinogenesis (Han et al.  2000 ; Kaplan et al.  2010 ; Chew et al.  2012 ; Coppenhagen‐Glazer et al.  2015 ; Han  2015 ; Gholizadeh et al.  2017 ; Groeger et al.  2022 ). F. nucleatum has consistently been one of the most abundant bacteria found in colon adenomas and colon adenocarcinomas and is linked to colon cancer promotion (Sun et al.  2019 ; Abed et al.  2020 ; Tran et al.  2022 ). However, there is a lack of information regarding the effects of specifically eliminating F. nucleatum in the presence of a full consortia of commensal and pathogenic gastrointestinal microbiota on inflammation, tumorigenesis and metabolism. Therefore, as a first step, we pursued the development and testing of an engineered probiotic specifically targeting a virulence factor expressed by F. nucleatum . Seminal research demonstrates that biofilm formation is a biomarker for colorectal cancer (CRC) and oral squamous cell carcinoma, as well as a potential promoter of local inflammation. F. nucleatum is a key member of these biofilms. FomA, an outer membrane porin and virulence factor, serves as the anchoring binding site on F. nucleatum for attachment to the oral mucosa via the human salivary protein statherin. This attachment enables F. nucleatum to act as a bridging species for secondary colonisers, such as Porphyromonas gingivalis , thus promoting biofilm formation (Brennan and Garrett  2019 ). Targeting FomA offers the added benefit of reducing biofilm formation at mucosal sites, leading to a decreased risk of inflammation and disease. Furthermore, this targeting likely applies additional selection pressure to reduce FomA expression in these strains. However, methods that target specific pathogen species or eliminate virulence factors face significant challenges, including off‐target effects. One method to overcome this challenge is to utilise antimicrobial peptides (AMPs) attached to virulence‐factor targeting guides (gAMPs). Unlike broad‐spectrum antibiotics, guided AMPs can be designed to selectively bind and target specific microbial membrane proteins, minimising harm to beneficial bacteria that do not express the target protein (Xu et al.  2024 ). This specificity not only enhances efficacy but also has the potential to mitigate the rise of resistance (Mba and Nweze  2022 ). Additionally, AMPs have the advantage of rapid product development compared to the slower pace of traditional antibiotics. The extensive databases describing hundreds of AMPs, along with the ease of genetically modifying them for increased affinity, such as through peptide tags, make them an ideal alternative to conventional antibiotics (Pirtskhalava et al.  2021 ). Synthesising and optimising gAMPs to be effective in vitro, however, does not often translate to clinical achievement as degradation of peptide drugs through the oral route is a major hurdle to overcome. One innovative strategy involves using engineered probiotics to express and release guided AMPs (gAMPs) in situ, avoiding peptide degradation and reducing production costs for recombinant protein production. Using this approach with selective gAMPs may aid in the recovery of species richness; unlike antibiotic treatments; and reduce overall microbial dysbiosis (Hemarajata and Versalovic  2013 ). Lactococcus lactis ( L. lactis thereafter) MG1363, a widely used probiotic, serves as an effective delivery platform and is the first patented engineered probiotic to deliver therapeutic molecules to human patients (Steidler et al.  2003 ). Previous research has demonstrated the versatility of L. lactis as a delivery system for various antimicrobial peptides, including camel milk lactoferrin derivatives (lactoferrampin and lactoferricin) (Tanhaeian et al.  2020 ) and gram‐negative specific amphibian‐derived AMPs (Volzing et al.  2013 ; Forkus et al.  2017 ). Beyond antimicrobial applications, L. lactis has been utilised to deliver systemic therapeutics for conditions such as pulmonary inflammation (Bermúdez‐Humarán et al.  2003 ; Yumoto et al.  2020 ), inflammatory bowel disease (Chiabai et al.  2019 ; Arukha et al.  2021 ; Noguès et al.  2022 ; Zahirović and Berlec  2022 ), and more recently, for cancer immunotherapy (Zhu et al.  2022 ). Together, these factors make it an ideal probiotic in which to express and deliver gAMPs. The approach used in this study was inspired by our previous study that developed precision therapy against Helicobacter pylori in mice (Choudhury et al.  2023 ). Building on this foundation, our study focused on engineering L. lactis MG1363 to synthesise and deliver gAMPs specifically targeting FomA on F. nucleatum . Our results demonstrate that the Statherin‐derived guide peptide enhances the binding affinity of a GFP marker to F. nucleatum , significantly increasing the preferential attachment compared to control peptides. In vitro assays revealed that both unguided and guided AMPs effectively inhibited biofilm formation in F. nucleatum , with gAMPs showing reduced toxicity against non‐target bacteria ( Bacteroides fragilis and Escherichia coli ). The gAMPs were also more effective in modulating growth kinetics, exhibiting selective toxicity towards F. nucleatum at lower concentrations. Co‐culture experiments in a simulated human gut microbiome also demonstrated that the gAMP probiotic maintained microbial diversity while effectively reducing F. nucleatum abundance. Quantitative PCR and 16S rRNA sequencing confirmed that gAMP treatment preserved the richness of the microbiota, contrasting with significant dysbiosis observed in both the positive and negative control samples. Overall, this study advances the application of engineered L. lactis as a targeted delivery system for gAMPs, offering a promising strategy for combating pathogenic bacteria like F. nucleatum while maintaining commensal microbial communities. By specifically targeting biofilm‐forming pathogens associated with colorectal cancer and oral squamous cell carcinoma, our approach not only mitigates infection and inflammation but also preserves the integrity of the host microbiome, presenting a significant step forward in precision antimicrobial therapy.

The authors declare no conflicts of interest.

Table S1: AMPs used in the study—ovisiprin and cathelin‐derived peptide SC5. The statherin derived guide peptide (red) is attached to the N‐terminus of the AMPs, separated by a—GGG—linker, to create the corresponding guided AMPs (gAMPs). Figure S1: MTT assay of ovispirin and CPSC5 and their stat‐guided analogs against Caco2 and HT29 cells. The data shown is the mean with error bars representing standard error.