Emerging antimicrobial peptides in gastrointestinal disorders: Dual role in immunity and therapy

Abstract

Antimicrobial peptides (AMPs) are small, cationic effectors of innate immunity that are central to gastrointestinal homeostasis and increasingly attractive as therapeutics for infection, inflammation, and malignancy. Produced by intestinal epithelial and immune cells, AMPs such as defensins and cathelicidins act as frontline regulators, responding to microbial cues through pattern-recognition receptors, shaping microbiota composition, reinforcing the mucosal barrier, and tuning the balance between pro-inflammatory and anti-inflammatory signalling. This review frames AMP biology around a central duality: The same peptides that defend the gut can, when dysregulated in concentration, location, or context, contribute to pathology, from the α-defensin deficiency that initiates ileal Crohn’s disease to the concentration-dependent switch by which LL-37 turns from a wound-healing mediator into a tumour-promoting one. This duality dictates therapeutic strategy, favouring restoration where disease arises from deficiency and restraint where it arises from aberrant signalling. We examine the mechanistic basis of AMP action (membrane disruption, intracellular targeting, and receptor-mediated immunomodulation) and evaluate preclinical and clinical evidence across inflammatory bowel disease, enteric infection, and gastrointestinal cancers, drawing the recurring lesson that microbiological success rarely guarantees clinical benefit. Persistent barriers of proteolytic instability, narrow therapeutic windows, and manufacturing cost are increasingly addressed through convergent advances in artificial intelligence–guided peptide design, sequence stabilisation, gut-targeted and stimulus-responsive delivery, and engineered live biotherapeutics that produce peptides in situ. Together, these developments are transforming AMPs from endogenous immune mediators into next-generation precision therapeutics for gut disease—agents designed to be not merely potent, but deliverable, selective, and stable where needed most.

Key Words: Antimicrobial peptide; Gut microbiota; Antimicrobial resistance; Immunomodulation; Intestinal barrier integrity; Inflammatory bowel disease; Nanotechnology-based delivery; Artificial intelligence in peptide design; Microbiome-targeted therapeutics

Core Tip: Antimicrobial peptides represent promising next-generation therapeutics that counter antimicrobial resistance through membrane disruption, immune modulation, and microbiome regulation. Their dual role in host defence and tissue homeostasis, particularly within gastrointestinal environments, underscores their broader physiological significance. Advances in artificial intelligence–driven peptide design and nanotechnology-based delivery systems are progressively addressing pharmacokinetic limitations and enhancing clinical viability. However, challenges remain, including cytotoxicity, proteolytic instability, and high production costs. Integrating bioengineering, precision medicine, and microbiome-guided strategies holds considerable potential to establish antimicrobial peptides as personalised, clinically translatable anti-infective and immunomodulatory therapeutics.

INTRODUCTION

Antimicrobial resistance is now one of the greatest public health challenges of the twenty-first century and is a continuous threat to the effectiveness of traditional antibiotics. With an increasing threat from multidrug-resistant organisms and a dwindling pipeline of antibiotics, alternative anti-infective approaches are necessary. Among the most promising candidates are antimicrobial peptides (AMPs)[1,2].

This phenomenon is particularly prominent in the gastrointestinal (GI) tract. Diseases of the gut, such as inflammatory bowel disease (IBD), Crohn’s colitis, ulcerative colitis (UC), irritable bowel syndrome, recurrent enteric infections (e.g., Clostridioides difficile colitis), and colorectal cancer, impose a huge and growing global burden, but they have a common theme: A breakdown in the communication between the host epithelium and its microbiota[3]. The current strategy is quite aggressive, with a strong reliance on broad-spectrum antibiotics and systemic immunosuppression, which indiscriminately disrupt the commensal microbiome, become ineffective with the emergence of resistant strains, often fail to produce sustained remissions, and have significant adverse effects. Thus, there is a strong need for more targeted therapies that restore the gut microbiota, rather than simply kill bacteria, while also restoring barrier function and microbial balance. This is an attractive approach because endogenous regulators function at the host–microbe interface in the gut, where they fine-tune the antimicrobial and immune tone in response to microbial signals. As ‘rheostats of intestinal homeostasis’, they provide a mechanistically different pathway for the treatment of GI disease[3].

AMPs are small, evolutionarily conserved, cationic, amphiphilic molecules that play a fundamental role in the innate immune response in bacteria, plants, insects, and mammals. They exhibit antimicrobial activity against bacteria, fungi, viruses, and even cancer cells by disrupting cell membranes, targeting the intracellular environment, and modulating the immune system[2,4,5]. The mode of action of AMPs is different from that of conventional antibiotics, which target specific molecules; AMPs usually disrupt microbial membranes within seconds, significantly reducing the chance of resistance arising[6].

The therapeutic possibilities of AMPs have broadened thanks to advances in molecular biology, peptide chemistry, and computation. Cyclisation, residue substitution, and conjugation increase stability, selectivity, and pharmacokinetics, whereas nanocarrier delivery overcomes the limitations of proteolytic degradation and toxicity and facilitates clinical application[5,7]. Most importantly, artificial intelligence (AI) and machine learning are now being used for the discovery, optimisation, and development of AMPs with increased potency and decreased toxicity and, when integrated with high-throughput screening and omics, are transforming the therapeutic paradigm[8].

Beyond direct killing, AMPs are multifunctional, with immunomodulatory, wound-healing, and anti-biofilm properties, broadening their biomedical reach[4,6]. Yet this same multifunctionality is a double-edged sword. The central theme of this review is the functional duality of AMPs: The very molecules that defend intestinal homeostasis can, when dysregulated, contribute to pathology. Persistent obstacles such as toxicity, instability, manufacturing cost, and regulatory hurdles must be resolved through multidisciplinary bioengineering before this duality can be harnessed clinically[7,9]. We trace this theme from AMP biology and gut homeostasis, through disease-associated dysregulation and therapeutic application, to the engineering, AI-guided design, and delivery strategies now converting endogenous immune mediators into precision therapeutics for gut disease.

CLASSIFICATION AND SOURCES OF AMP

Classifying AMPs by organismal origin, secondary structure, net charge, and amino acid composition is not merely taxonomic (Table 1). It directly governs the mechanism of action, spectrum of activity, and therapeutic tractability, explaining why certain AMPs have advanced clinically while others remain preclinical[4,5].

Table 1 Key antimicrobial peptides: Structural classes, mechanisms, molecular targets, and gastrointestinal relevance.

AMP	Structural class	Primary mechanism(s)	Molecular target(s)	GI relevance	Ref.
HD-5, HD-6	α-Defensin	Membrane disruption; microbiome shaping via crypt gradients	Lipid II (HD-5); self-assembly nets (HD-6)	Paneth cell-derived; reduced in ileal Crohn’s disease	[4,14]
hBD-2	β-Defensin (inducible)	Membrane disruption; CCR6-mediated chemotaxis	Lipid II; CCR6	NF-κB-induced in IBD epithelium; inducible barrier	[4,5]
LL-37	Cathelicidin (α-helix)	Membrane disruption; DNA binding; receptor-mediated signalling	FPR2/ALX (approximately 200 nM), EGFR, TLR9 via DNA	Wound healing; context-dependent pro-/anti-tumour in CRC	[17,21]
Buforin II	α-Helix (proline hinge)	Non-lytic entry; DNA/RNA binding	dsDNA minor groove; rRNA	Model for resistance-refractory intracellular targeting	[5,9]
Nisin	Lantibiotic (cyclic)	Lipid II sequestration and pore formation	Lipid II pyrophosphate (approximately 1 nM)	Narrow-spectrum; resistance-resistant dual-mechanism paradigm	[2,20]
Microcin J25	Lasso peptide	RNA polymerase inhibition; membrane-independent	RNA polymerase β-subunit	Efficacious in DSS-induced colitis; selective microbiome modulator	[22]
CRAMP	Cathelicidin (α-helix)	Membrane disruption; FPR2/ALX immunomodulation	Microbial membranes; FPR2/ALX	Anti-inflammatory in murine colitis; restores diversity	[23]

AMP: Antimicrobial peptide; CCR6: C-C chemokine receptor type 6; CRAMP: Cathelicidin-related antimicrobial peptide; CRC: Colorectal cancer; DSS: Dextran sodium sulfate; dsDNA: Double-stranded DNA; EGFR: Epidermal growth factor receptor; FPR2/ALX: Formyl peptide receptor 2/Lipoxin A4 receptor; GI: Gastrointestinal; IBD: Inflammatory bowel disease; NF-κB: Nuclear factor kappa B; RNA: Ribonucleic acid; rRNA: Ribosomal RNA; TLR9: Toll-like receptor 9.

Mammalian AMPs

In mammalian cells, neutrophils, macrophages, and Paneth cells at epithelial surfaces, including the skin, respiratory mucosa, and intestinal wall, constitutively synthesise AMPs upon activation of pattern recognition receptors[4]. The 2 dominant human families are defensins and cathelicidins.

Defensins (3–5 kDa) are characterised by 6 conserved cysteine residues that form 3 disulfide bonds, stabilising a β-sheet scaffold[4,5]. The α-defensins HD-5 and HD-6 are secreted constitutively by Paneth cells into the lumen of intestinal crypts, where they establish antimicrobial gradients that govern colonisation resistance and shape microbiota composition[10,11]. By contrast, the β-defensins hBD-1 through hBD-4 are produced inducibly by epithelial cells through nuclear factor-κB (NF-κB) and mitogen-activated protein kinase (MAPK) signalling downstream of Toll-like receptor (TLR)2/4[12]. hBD-1 is notable for its redox-dependent activation: Reduction of its disulfide bonds by thioredoxin markedly enhances its potency, a mechanism not seen in α-defensins and one of direct relevance to the redox-dysregulated mucosa of IBD[13]. θ-Defensins, by contrast, are absent in humans because of a premature stop codon, although synthetic retrocyclins remain of therapeutic interest[4]. One inconsistency should be noted: While α-defensin deficiency in ileal Crohn’s disease (CD) is well established, changes in defensin expression in colorectal cancer vary by stage and anatomical location, with reports of both upregulation and downregulation, preventing a single designation of defensins as either protective or promotive in malignancy[14,15].

The sole human cathelicidin, LL-37, is proteolytically released from its pro-peptide hCAP-18 by neutrophil serine proteases, yielding a 37-residue amphipathic α-helix[16,17]. Its structural plasticity—disordered monomers that transition into ordered helical oligomers upon membrane contact—allows it to engage microbial membranes, host receptors, and nucleic acids[9]. Acting through receptor-mediated signalling, LL-37 promotes proliferation, migration, and cytokine production. The same signalling pathways account for both its wound-healing functions and, at the chronically elevated concentrations found in the tumour microenvironment, its paradoxical tumour-promoting activity in colorectal cancer. This duality is elaborated on in the sections that follow.

Non-mammalian AMPs

Bacteriocins from the gut microbiota (microcins ≤ 10 kDa; lantibiotics) are ribosomally synthesised and have a narrow spectrum of activity, acting via lipid II sequestration, cell-wall inhibition, or intracellular nuclease activity[2,18]. They enable precise suppression of pathogens without the broad microbiome disruption caused by conventional antibiotics, a property central to dysbiosis correction.

Insect AMPs (cecropins, attacins) and amphibian peptides (magainins, dermaseptins) disrupt membranes through amphipathic helical mechanisms and have served as templates for synthetic optimisation[1,9]. The translational failure of magainin analogues, despite decades of work, foreshadows the clinical challenges further discussed in the review[19]. Plant-derived AMPs offer protease-resistant scaffolds: Cyclotides such as kalata B1 combine backbone cyclisation with a cystine-knot motif, a framework that has directly informed cyclic peptide drug design[2]. A synthesis-level gap remains: The field lacks systematic comparative data on which structural classes are most amenable to GI-specific development, the relative contributions of membrane versus intracellular targeting across classes, and how structural class correlates with microbiome selectivity. Addressing these is a prerequisite to evidence-based AMP design.

MECHANISMS OF ACTION OF AMPs

Current evidence supports a multi-target model in which a single peptide may simultaneously destabilise membranes, engage intracellular targets, and modulate host signalling, with the dominant mechanism depending on concentration, organism, and microenvironment[1,5,6]. These mechanisms are neither mutually exclusive nor fixed, and conflating them risks both misinterpreting the data and misdirecting peptide design.

Membrane disruption: Mechanisms, selectivity, and in vivo limitations

Initial engagement results from electrostatic attraction between the peptide’s net positive charge (+2 to +9) and the anionic components of the bacterial cell membrane—phosphatidylglycerol, cardiolipin, and lipopolysaccharides in Gram-negatives, and lipoteichoic acid in Gram-positives[4,6]. Eukaryotic membranes, by contrast, resist AMP action. Their outer leaflet is dominated by zwitterionic phosphatidylcholine and sphingomyelin and is further stabilised by cholesterol[4,9]. From this initial contact, 3 structural mechanisms may follow. In the barrel-stave model, helical AMPs insert perpendicularly into the bilayer to line a stable aqueous channel, as exemplified by alamethicin, which forms electrophysiologically confirmed voltage-gated channels[6,9]. In the toroidal-pore model, peptides such as magainin 2 and melittin induce curvature strain that produces transient peptide-lipid pores, a process confirmed by solid-state nuclear magnetic resonance and neutron scattering[6]. In the carpet model, peptides such as the dermaseptins accumulate on the membrane surface until, at a critical coverage, they trigger detergent-like micellisation[9]. A single peptide can shift between these modes depending on the peptide-to-lipid ratio and membrane composition, which complicates the interpretation of whole-cell data[5,6]. Beyond pore formation, AMPs cause membrane thinning, lipid phase separation, and curvature stress that impair membrane–protein function at sub-lytic concentrations. They also disrupt electrostatic interactions within the biofilm matrix to prevent adhesion at levels below the minimum inhibitory concentration[6,7]. A critical translational limitation is that membrane activity is substantially attenuated by physiological NaCl, low luminal pH, and serum proteins—conditions endemic to the GI tract. This physicochemical sensitivity, not merely protease susceptibility, explains why peptides with potent in vitro activity routinely underperform in GI models. These limitations must be addressed through formulation rather than minimised in efficacy reporting.

Intracellular targets and immunomodulation: Molecular specificity

AMPs access intracellular targets via pore-dependent leakage, direct translocation, or endocytosis—the dominant route being peptide-specific[5,6]. Buforin II crosses Escherichia coli membranes with minimal lysis owing to a proline hinge at position 11, then binds double-stranded DNA and ribosomal RNA with nanomolar affinity to inhibit replication and transcription[5,9]. By contrast, indolicidin has DNA-binding capacity but kills primarily by membrane disruption—a distinction relevant to resistance prediction because membrane-based resistance cannot counter intracellular-targeting peptides[6]. Tachyplesin I’s rigid β-sheet fits the double-stranded DNA minor groove while disrupting Gram-negative membranes[5,9]. At the protein level, pyrrhocoricin and oncocin inhibit the chaperone DnaK[6], whereas apidaecin-1b depends on the transporter SbmA for uptake. SbmA-deficient strains are intrinsically resistant, a mechanism predictable from genomic screening[5,6]. Nisin exemplifies dual action: It sequesters lipid II (Kd approximately 1 nM), preventing cell-wall synthesis, then uses the complex to scaffold membrane pores[2,20]. This mechanistic redundancy underlies its potency and slow development of resistance and serves as a key design principle for next-generation AMPs.

Immunomodulation demands precision beyond the generic ‘TLR/NF-κB activation’ language that pervades the literature. LL-37 offers the best-characterised network and serves as the mechanistic anchor for the duality developed throughout this review. It binds formyl peptide receptor 2/Lipoxin A4 receptor (FPR2/ALX) (Kd approximately 200 nM) on macrophages and epithelial cells, recruiting β-arrestin and activating extracellular signal-regulated kinase 1/2 (ERK1/2) and p38 to drive CXCL1/CXCL8 expression and neutrophil chemotaxis[17]. It also activates epidermal growth factor receptor (EGFR) directly via phosphoinositide 3-kinase/protein kinase B to promote epithelial survival and migration—the basis of its wound-healing activity and the same mechanism that accounts for tumour promotion under sustained high concentrations[17,21]. In addition, it complexes with self-DNA to engage endosomal TLR9 in plasmacytoid dendritic cells, amplifying interferon-α with implications for sterile intestinal inflammation[17]. The anti-inflammatory dimension operates differently. LL-37 and hBD-2 neutralise lipopolysaccharides directly (Kd approximately 3 μmol/L), blocking TLR4-mediated NF-κB activation independently of any killing effect—a property particularly relevant at endotoxin-driven sites in IBD[17]. α-Defensins contribute further by promoting dendritic cell maturation and Th1/Th2 polarisation, supporting commensal tolerance[5,17].

Whether direct killing or immunomodulation dominates the therapeutic efficacy of peptides such as cathelicidin-related AMP in colitis remains unresolved. Deconvolution using receptor-knockout models and mechanism-uncoupled mutants that separate FPR2/ALX binding from membrane activity is a methodological priority before the rational selection of immunomodulatory versus bactericidal AMPs for GI use. These molecular mechanisms set the stage for the physiological roles AMPs play in the intact gut.

PHYSIOLOGICAL ROLES OF AMPs IN GI HOMEOSTASIS

Having established how AMPs act at the molecular level, we now turn to how these actions sustain the gut. The gut maintains a remarkable truce: Some 100 trillion microorganisms coexist with the host while being kept separate from sterile systemic tissues. AMPs are central to maintaining that truce, continuously tuning the antimicrobial and immune tone in response to microbial signals. As Akoh-Arrey and Brooks[3] describe them, they act as ‘rheostats of intestinal homeostasis’.

Regulation of gut microbiota

AMP-mediated regulation of the gut microbiota is well supported experimentally. Peptides such as microcin J25 and cathelicidin-related AMP have each been shown to reshape microbial community composition and restore eubiosis in murine models of colitis[22,23]. Paneth cell-derived AMPs—chiefly the α-defensins HD-5 and HD-6 in humans and the lectin RegIIIγ in mice—are secreted into the lumen following microbial stimulation through MyD88-dependent TLR signalling, establishing steep gradients that keep luminal bacteria separated from the epithelial surface[10,24,25]. Crucially, AMPs do more than kill. At sub-lethal concentrations, they suppress virulence genes and interfere with quorum sensing, shifting the competitive balance against pathobionts without necessarily reducing bacterial counts[18]. HD-6 illustrates a separate, non-lytic strategy. On contact with microbial surface proteins, it polymerises into ‘nanonets’ that entrap pathogens and block invasion, defending the mucosa without releasing inflammatory lysis products[4].

Maintenance of intestinal barrier integrity

Barrier integrity rests on the combined action of physical (tight junctions and adherens junctions), biochemical (mucus and secretory IgA), and chemical (AMP) defences. Acting through the vitamin D receptor, vitamin D upregulates intestinal tight-junction proteins, including claudin-1, occludin, and zonula occludens-1[26]. In parallel, it serves as a direct transcriptional inducer of cathelicidin via a vitamin D response element in the CAMP promoter[27]. Cathelicidin (LL-37) can itself reinforce epithelial tight junctions, although the characterised mechanisms, namely Rac1/atypical protein kinase C/glycogen synthase kinase 3/phosphoinositide 3-kinase, demonstrated in keratinocytes, are independent of the vitamin D receptor[28]. This vitamin D receptor–AMP link is clinically important. Vitamin D deficiency, common in IBD, simultaneously weakens cathelicidin expression and tight-junction maintenance, producing a self-reinforcing cycle of barrier dysfunction that may explain why low vitamin D tracks with IBD severity[29,30]. AMPs also strengthen the mucus layer, with LL-37 inducing MUC5AC through EGFR transactivation[31].

Immunomodulatory functions

AMPs link innate and adaptive immunity in the mucosa. FPR2/ALX binds LL-37, and C-C chemokine receptor type 6 (CCR6) binds β-defensins. These interactions activate innate immune cells and prime adaptive CCR6⁺ cells, respectively, leading to the recruitment of neutrophils, monocytes, immature dendritic cells, and CCR6⁺ memory T cells[17]. AMPs also help maintain tolerance to commensal organisms. Their constitutive expression is kept low enough that the resulting TLR stimulation remains below the threshold required to trigger adaptive immune activation. When that balance is lost through Paneth-cell dysfunction, inappropriate activation against commensals may drive the autoimmune-like pathology of IBD[9,11].

Role in epithelial repair and regeneration

AMPs also promote mucosal healing. Cathelicidins drive epithelial migration and proliferation through EGFR/ERK and peroxisome proliferator-activated receptor-γ signalling, with LL-37 showing pro-healing activity at nanomolar concentrations[32-34]. These homeostatic activities, however, are not unconditionally beneficial. Their net effect on the tissue depends on context, the subject of the next section.

THE FUNCTIONAL DUALITY OF AMP IN GI DISEASE

Reframing AMPs as context-dependent mediators

AMPs have historically been cast as a uniformly beneficial arm of innate immunity[17,24], yet this framing poorly predicts their behaviour in human disease. They are more accurately understood as conditional rather than intrinsically protective molecules, with their net tissue effect determined by which peptide is expressed, at what concentration, in which microenvironment, at which disease stage, and against which microbial community[3]. This context-dependence reflects functional pleiotropy: Beyond microbicidal action, host-defence peptides recruit and polarise leukocytes, shape cytokine output, influence angiogenesis, and bridge innate and adaptive immunity—functions that are dissociable from killing[10]. A peptide’s contribution to disease therefore cannot be predicted from antimicrobial potency alone, a distinction with direct consequences for whether AMP therapeutics should supplement, suppress, or modulate peptide activity. The following sections trace this duality across the gut: Where AMP dysregulation initiates and sustains disease, where AMPs limit injury, and the variables governing the switch.

AMP deficiency as a driver of disease initiation

The clearest evidence that absent AMP activity initiates disease comes from ileal CD, where Paneth-cell α-defensin deficiency precedes rather than follows inflammation. Reduced HD-5 and HD-6 are detectable in non-inflamed ileal tissue and track with genotype rather than local inflammatory activity[14], arguing against a reactive loss; the deficiency occurs even in NOD2/CARD15 wild-type patients and is further accentuated in carriers of the SNP13 mutation[14]. Mouse genetics support causality: Animals expressing human DEFA5 show defensin-dependent microbiota remodelling, with selective loss of segmented filamentous bacteria and reduced interleukin (IL)-17-producing T cells, while defective processing produces the opposite shift[11]. α-Defensins thus actively sculpt community structure and its T-cell compartment, rather than simply restraining bacterial load.

What converts correlation to mechanism is the convergence of Crohn’s risk genes on the Paneth-cell secretory pathway. The autophagy gene ATG16 L1 is required for normal granule formation and antimicrobial exocytosis; hypomorphic mice and risk-allele-homozygous patients show identical granule pathology[35]. The link resolves to a single residue: The T300A variant increases susceptibility to caspase-3 cleavage, blunting autophagy and amplifying cytokine release[36]. With NOD2, which couples bacterial sensing to defensin induction, these define a cascade in which a genetic lesion impairs Paneth-cell function, lowers luminal AMP delivery, and permits microbial encroachment, notably the expansion of adherent-invasive E. coli, recovered at high frequency from Crohn’s ileal mucosa, which invades the epithelium and replicates within macrophages with sustained tumour necrosis factor-α output[37]. Defective secretion, pathobiont expansion, and granulomatous Th1/Th17 inflammation thus form a causal chain rather than co-occurring abnormalities.

This logic extends, more tentatively, to functional disorders. In irritable bowel syndrome, where overt inflammation is absent, neuroimmune regulation of AMP expression offers a mechanistic entry point for the gut–brain axis: Chronic social-defeat stress reduces Paneth-cell α-defensin secretion and induces dysbiosis and metabolite disturbance, both reversed by exogenous α-defensin[34], establishing defensin reduction as a necessary intermediate rather than an epiphenomenon.

Aberrant AMP activity as a driver of disease exacerbation

The colon illustrates the opposite failure mode, in which excessive or misdirected AMP activity amplifies pathology. LL-37 is paradigmatic, behaving as a concentration- and receptor-dependent molecule rather than an inherently harmful one. Signalling through FPR2/ALX on endothelial cells, it drives angiogenesis across endothelial assays, the chick chorioallantoic membrane, and ischaemic hindlimb models, with cathelicidin-deficient mice showing impaired revascularisation[25]. In wound healing, this is reparative; in the tumour microenvironment, the identical receptor signalling becomes pro-oncogenic via MAPK/ERK and EGFR transactivation[21]. At low concentrations, receptor signalling predominates and favours proliferative, angiogenic responses; only at substantially higher concentrations does membrane-disruptive cytotoxicity emerge, underlying anti-tumour potential[21]. Precise thresholds vary, and the proliferative–cytotoxic transition is best treated as a qualitative regime change rather than a fixed boundary.

AMPs as protective and barrier-maintaining mediators

Against these pathogenic roles, AMPs remain indispensable for mucosal order. The C-type lectin regenerating islet-derived protein 3 gamma (REG3γ) sustains an approximately 50-µm exclusion zone separating microbiota from the small intestinal epithelium; in its absence, bacteria encroach and adaptive activation rises sharply despite little change in total bacterial number[38]. This reframes AMP protection in spatial rather than purely bactericidal terms. Thus, its homeostatic value lies as much in organising where microbes reside as in how many survive. The same families that drive pathology when dysregulated maintain order when appropriately expressed, orchestrating chemotaxis, tempering pro-inflammatory signalling, supporting restitution, and contributing to resolution[10,39]. In experimental colitis, cathelicidin- and defensin-derived peptides attenuate inflammation, restore barrier integrity, and favour reparative macrophage phenotypes—LL-37’s angiogenic activity operating in a reparative rather than neoplastic context[25,40].

Determinants of the protective-to-pathogenic switch

The duality is governed by a small set of intersecting variables. Peptide identity is first: Defensins and REG3 lectins act predominantly to exclude and segregate, whereas cathelicidin’s heavier signalling load renders it more prone to pathological repurposing. Concentration is most decisive for LL-37: Low-level receptor engagement and high-level membrane disruption are functionally distinct modes with opposite consequences[21,25]. Microenvironment supplies the third: Receptor availability (FPR2/ALX, EGFR), inflammatory tone, and microbiota composition dictate whether a given concentration reads as a repair or proliferative signal[3,21,25]. Disease stage and host genotype overlay these because the defensin deficiency initiating Crohn’s operates in a host whose NOD2 and ATG16 L1 background has already constrained the secretory response[14,35,36]. Microbial interaction is a final, often-overlooked determinant: Helicobacter pylori selectively downregulates constitutive hBD-1 in a type IV secretion-dependent manner while inflammation drives inducible hBD-2 upward, producing net remodelling rather than simple suppression[41], favouring persistence and, through sustained inflammation, the carcinogenic trajectory for which H. pylori is prototypic.

The inflammation–infection–cancer continuum as integrated duality

In plain terms, long-standing gut inflammation creates conditions that let infection take hold and persist; that same combination of prolonged irritation and infection, sustained over years, is what raises the risk of cancer developing in the gut. The sections below trace exactly how this happens at the molecular level and why the same AMPs that normally prevent this progression can, in the wrong context, contribute to it instead. These determinants converge in the progression from chronic inflammation through persistent infection to malignancy. AMP deficiency in Crohn’s sustains pathobiont colonisation and a carcinogenic field of reactive oxygen species, DNA damage, and pro-oncogenic cytokines[14,37]; H. pylori-mediated remodelling secures colonisation and erodes immune surveillance[41]; and C. difficile links epidemiologically to colorectal cancer, recurrent positivity carrying an adjusted hazard ratio of approximately 2 whereas single positivity does not—a dose-dependent relationship coherent if the relevant exposure is cumulative dysbiosis rather than a single event[42]. The malignant compartment recapitulates the duality cellularly: Defensins suppress tumour growth through mitochondrial apoptosis and caspase activation, with reduced expression linked to adverse features, whereas LL-37 promotes proliferation and angiogenesis at tumour-microenvironment concentrations[15,21,25].

A context-dependent model implies that no single therapeutic strategy is correct across disorders; the duality must dictate the intervention. Where disease arises from deficiency (ileal Crohn’s) or stress-associated dysfunction, restoration or supplementation is rational, as DEFA5 and α-defensin rescue models suggest[11,34]. In conditions driven by aberrant signalling, such as chronic colonic inflammation, where LL-37 has pro-tumour activity, restraint or receptor-level antagonism is the more appropriate strategy because supplementation risks worsening the underlying pathology[21]. The LL-37 concentration switch makes this concrete: A therapeutic window exists only where cytotoxic concentrations are achievable locally without traversing the proliferative regime, favouring direct or targeted over systemic delivery[7].

THERAPEUTIC APPLICATIONS OF AMP IN GI DISEASES

The therapeutic logic follows directly from this duality. In conditions where AMP deficiency drives disease, such as CD and several enteric infections, supplementing exogenous AMPs or inducing endogenous expression is the rational approach. Where dysregulated signalling contributes to pathology, such as LL-37 in UC and the colorectal tumour microenvironment, selective modulation rather than indiscriminate augmentation is needed. This section evaluates the evidence, separating approaches with preclinical proof-of-concept from those supported by clinical or ex vivo human data, with particular attention to the delivery and stability barriers that have kept promising in vitro activity from the clinic.

AMPs in IBD

The rationale is strongest where endogenous deficiency is established, most notably in ileal CD, where Paneth-cell α-defensin expression is reduced[10,14]. Two strategies have been pursued: Direct administration of exogenous peptides and restoration of endogenous expression. Preclinical evidence for exogenous peptide administration is accumulating across dextran sodium sulfate (DSS)-induced colitis models (Table 2). DP7 significantly reduced DSS-induced colitis in mice, reducing the disease activity index, weight loss, and colon shortening, while preserving epithelial integrity. This effect was linked to enrichment of Muribaculaceae, and faecal microbiota transplantation from DP7-treated donors reproduced the therapeutic benefit in recipients, confirming a microbiota-dependent mechanism[43]. The bee-derived peptide abaecin similarly alleviated mucosal barrier damage and lowered pro-inflammatory cytokines (IL-1β, IL-6, tumour necrosis factor-α, and interferon-γ) by suppressing NF-κB/MAPK signalling, alongside improved microbial composition[44]. Chensinin-1b, a Rana chensinensis-derived peptide, reprogrammed macrophages from the M1 to the M2 phenotype through the same pathway in vitro, and its intraperitoneal administration reproduced this shift in vivo, attenuating weight loss, disease activity, colon shortening, and splenomegaly[45].

Table 2 Preclinical studies of antimicrobial peptide activity in gastrointestinal and related disease models.

No.	Ref.	Peptide	Model / route	Dose	Key findings	Sig.
1	Moghaddam et al[49]	CM11; CM15 (cecropin-melittin hybrids)	In vitro: P. aeruginosa, S. aureus, V. cholerae, A. baumannii, E. coli (ATCC and clinical)	MIC 4 mg/L; MBC 16 mg/L	Complete kill at 30-40 minutes; CM11 MIC/MBC stable at 48 hours; CM15 MIC doubled at 48 hours	P < 0.05
2	Shang et al[22]	MccJ25 (lasso peptide, 21-aa)	DSS-induced UC; C57BL/6J/oral gavage	5 mg/kg and 10 mg/kg	↓ DAI, ↑ colon length; ↓ TNF-α, IFN-γ, IL-1β, IL-6; ↑ TJ proteins; ↑ Lactobacillus, ↓ Bacteroides and Akkermansia; co-housing confirmed microbiota as key mechanism	P < 0.05
3	Jiang et al[23]	CRAMP (cathelicidin-related; murine LL-37 homologue)	3% DSS-induced acute UC; C57BL/6J/IP	4 mg/kg/day	↑ body weight, ↑ colon length, ↓ DAI, ↓ MPO; ↑ TJ proteins; ↓ IL-6, TNF-α, MCP-1, CRP; ↑ GSH-PX, ↓ MDA; ↑ Verrucomicrobiota	P < 0.05-0.001
4	Liu et al[44]	Abaecin (proline-rich, bee-derived; non-lytic)	2.5% DSS-induced acute UC; C57BL/6J/rectal	5 mg/kg	↑ colon length and body weight, ↓ DAI; ↓ LPS, D-LA, DAO; ↓ IL-1β, IL-6, TNF-α, IFN-γ (IL-10 unchanged); ↑ ZO-1, occludin, claudin-1; ↓ NF-κB/MAPK; ↓ Bacteroides, Barnesiella, Escherichia, ↑ Lactobacillus	P < 0.05-0.01
5	Sun et al[45]	Chensinin-1b (from R. chensinensis)	RAW264.7 + LPS (in vitro) and 4% DSS-induced UC; BALB/c/IP	10-40 μmol/L; 1.5 and 3 mg/kg	↓ M1 markers (TNF-α, IL-6, NO, CD86), ↑ M2 markers (IL-10, TGF-β1, Arg-1, CD206); ↓ NF-κB/MAPK; ↑ body weight and colon length, ↓ DAI; high dose superior to cyclosporine A	P < 0.05-0.01
6	Zhao et al[43]	DP7 (12-aa; machine-learning designed)	4% DSS-induced UC (multi-arm: FMT, antibiotic, pseudo-germ-free); C57BL/6/IV	0.5 mg/kg q2d	↓ weight loss, ↓ DAI, ↑ colon length; ↑ ZO-1, claudin-1, occludin; ↓ IL-1β, IL-6, ↑ IL-10; ↑ Muribaculaceae; DP7-FMT > DSS-FMT, immune effects microbiota-dependent	P < 0.05-0.0001
7	Sun et al[46]	R7I (IRPI × 7; anti-proteolytic; trypsin/chymotrypsin/pepsin-resistant)	E. coli ATCC 25922 enteritis; C57BL/6/oral gavage	20 mg/kg, 30 mg/kg, 40 mg/kg	↓ intestinal and hepatic inflammation, restored barrier; modulated microbiota (↓ Clostridia, ↑ Odoribacteraceae); normalised gut metabolites	P < 0.05

Arg-1: Arginase-1; ATCC: American Type Culture Collection; CD206: Cluster of differentiation 206; CRAMP: Cathelicidin-related antimicrobial peptide; CRP: C-reactive protein; DAI: Disease activity index; DAO: Diamine oxidase; DP7: Designed peptide 7; DSS: Dextran sodium sulfate; FMT: Faecal microbiota transplantation; GSH-PX: Glutathione peroxidase; IFN-γ: Interferon gamma; IL: Interleukin; IP: Intraperitoneal; IRPI: Isoleucine-arginine-proline-isoleucine motif; IV: Intravenous; LPS: Lipopolysaccharide; MAPK: Mitogen-activated protein kinase; MBC: Minimum bactericidal concentration; MccJ25: Microcin J25; MCP-1: Monocyte chemoattractant protein-1; MDA: Malondialdehyde; MIC: Minimum inhibitory concentration; MPO: Myeloperoxidase; NF-κB: Nuclear factor kappa B; NO: Nitric oxide; q2d: Every 2 days; R7I: Arginine-isoleucine repeat peptide (IRPI × 7); RAW264.7: Murine macrophage cell line; TGF-β1: Transforming growth factor beta 1; TJ: Tight junction; UC: Ulcerative colitis; ZO-1: Zonula occludens-1.

AMPs link innate and adaptive immunity in the mucosa. FPR2/ALX binds LL-37, and CCR6 binds β-defensins. These interactions activate innate immune cells and prime adaptive CCR6⁺ cells, leading to the recruitment of neutrophils, monocytes, immature dendritic cells, and CCR6⁺ memory T cells[17]. AMPs also sustain commensal tolerance: Low constitutive expression keeps TLR stimulation below the threshold for adaptive activation. When that balance is lost through Paneth-cell dysfunction, inappropriate activation against commensals may drive the autoimmune-like pathology of IBD[9,11,33].

A single limitation unites these candidates: The inflamed lumen is protease-rich, and most native AMPs are degraded before reaching their target (LL-37’s circulating half-life is roughly 1 hour). Two engineering responses are emerging. The first is sequence stabilisation: The anti-proteolytic peptide R7I retained oral efficacy in an E. coli enteritis model, restoring barrier integrity and modulating the microbiota, though it was tested against infection rather than sterile inflammation; a more disease-matched example is the stabilised LL-37 fragment KR-12, which, upon systemic administration, reduced macroscopic, microscopic, and ulcer scores across acute, semi-chronic, and chronic 2,4,6-trinitrobenzene sulfonic acid–induced colitis[40,46]. The second is targeted delivery, in which nanocarriers shield the peptide through gastric transit and release it at the mucosa[7,47].

Strategies that restore endogenous AMP tone provide the clearest human evidence. High doses of vitamin D were shown to significantly increase cathelicidin (hCAP/LL-37) expression compared with placebo in a randomised controlled trial of patients with UC[48]. Commensal signals offer a parallel strategy: Short-chain fatty acids induce expression of β-defensin in colonic epithelial cells[24,39]. By boosting physiological production at the mucosa rather than delivering peptide systemically, these approaches largely bypass the stability problem.

AMPs in GI infections

Activity against enteric pathogens is well established in vitro. The cecropin–melittin hybrids CM11 and CM15 show rapid, broad-spectrum killing, but translation is limited by pathogen-specific resistance and difficulty sustaining mucosal concentrations[49]. Helicobacter pylori shows how targeted delivery converts in vitro promise into in vivo efficacy: A bioresponsive, genetically encoded antimicrobial crystal stays stable through gastric transit and releases its LL-37-based payload only where H. pylori urease raises local pH; given orally, it eradicates infection and restores the microbiome[50]. This stimulus-responsive design exemplifies the delivery-first logic now shaping the field. Clostridioides difficile poses a distinct problem: Its disease is toxin-mediated and arises on antibiotic-disrupted microbiota, so AMPs and bacteriocins are explored both as selective bactericidal agents and within microbiome-sparing regimens that, unlike broad-spectrum antibiotics, spare protective commensals and may also limit toxin production[51]. AMPs also serve as adjuvants, permeabilising the outer membrane to grant co-administered antibiotics intracellular access to them, though the experience with membrane-targeting peptidomimetics warns that a narrow therapeutic index can derail agents with strong in vitro activity[17].

AMPs in GI cancers

Because AMP activity in GI cancers is concentration- and context-dependent, therapy must reach tumouricidal local concentrations while avoiding exposures that drive proliferation—a constraint distinguishing oncological from infectious indications. Defensins capture this tension: Depending on the peptide and setting, they suppress tumour initiation or associate with tumour-promoting processes, and their lower expression in colorectal tissue than in normal tissue supports locally restorative strategies while warning against indiscriminate augmentation[15]. LL-37 must be viewed through its cytotoxic-versus-proliferative switch[21,34]; rational strategies aim to uncouple cytotoxic membrane activity from growth-promoting signalling through structural modification or pathway inhibitors rather than maximising delivery[21]. More broadly, the immunomodulatory recruitment of cytotoxic effectors makes host-defence peptides candidate immunotherapy partners, though this remains an early hypothesis needing GI-specific evidence[17].

Microbiome modulation and gut-targeted delivery

A defining advantage of AMPs over broad-spectrum antibiotics is selectivity: They preferentially target pathobionts, alter bacterial behaviour at sub-lethal concentrations, and reshape community structure, acting as ecosystem regulators rather than blunt antimicrobials with post-antibiotic microbiome restoration, including recovery from C. difficile-associated dysbiosis, a promising application[24,39]. Engineered live biotherapeutics extend this by producing peptide in situ: Lactococcus lactis engineered to deliver hCAP-18/LL-37 alleviated 2,4-dinitrobenzene sulfonic acid–induced colitis and raised IL-10 and IL-17A, showing that a food-grade commensal can deliver an AMP locally and continuously[52]. This directly addresses the proteolysis and short half-life that limit systemic peptides, and inflammation-responsive, biocontained probiotic and biomaterial platforms are now being developed for IBD[53]. From the formulation side, GI-adapted nanocarriers pursue the same goal[47]. Together, these advances indicate that delivery, rather than the peptides themselves, is increasingly the rate-limiting step in translating AMP therapy[7].

Lessons from human trials: Why AMPs have underperformed in the clinic

Across 10 trials (Table 3), the agents were consistently safe, yet efficacy was inconsistent, and several well-designed studies missed their primary endpoint. Critically, none targeted a primary GI disease; the indications were oral mucositis, skin and nasal colonisation, catheter infection, candidiasis, and a respiratory virus. The clinical evidence base for AMPs in GI disease thus remains essentially preclinical, and the failures in adjacent indications expose the barriers GI programmes must clear.

Table 3 Clinical studies of antimicrobial peptide-based interventions.

No.	AMP	Indication	Phase	Design /groups	Key findings	Safety	Ref.
1	Pexiganan (MSI-78)	Infected diabetic foot ulcers	3	2 double-blind RCTs; pexiganan 1% cream vs oral ofloxacin vs placebo (n = 835)	Cream equivalent to ofloxacin (85%-90% improvement; 42%-47% eradication); no pexiganan resistance (ofloxacin resistance emerged); study 304 + combined met equivalence, study 303 failed	Well tolerated; no systemic toxicity	[19]
2	Iseganan (IB-367; protegrin-1 analogue)	Oral mucositis (stomatotoxic chemotherapy)	3	Double-blind RCT; iseganan 9 mg oral rinse 6 ×/day vs placebo (n = 323; 163 vs 160)	Primary endpoint (UOM prevention by day 21) not met: 43% vs 33% UOM-free (P = 0.067); significant reductions in peak mouth pain (P = 0.041), peak throat pain (P = 0.048), and NCI CTC stomatitis (P = 0.013)	Well tolerated; no systemic absorption	[60]
3	Iseganan (IB-367)	Oral mucositis (radiotherapy, H&N cancer)	3	Double-blind 3-arm RCT; iseganan + SOC vs placebo + SOC vs SOC (n = 545)	OM prevention not met (9% vs 9% OM-free, P = 0.998); both intervention arms > SOC alone; benefit attributed to oral hygiene/vehicle, not AMP	Nausea higher with iseganan (51%); no systemic absorption	[59]
4	hLF1-11 (lactoferrin 1-11)	Infection prevention, autologous HSCT	1	Open-label single 5-mg IV dose in autologous HSCT recipients (n = 8; part of a 3-study first-in-human programme, total n = 56)	Well tolerated; no immunogenicity (no anti-hLF1-11 IgG/IgE); IL-6/TNF-α attenuation trend on LPS stimulation (NS); safety/PD only; PK not determinable (peptide unquantifiable in plasma), no efficacy data	No serious drug-related AEs; reversible transaminase rise	[54]
5	LTX-109 (Lytixar; peptidomimetic)	Nasal MRSA/MSSA carriage	1/2a	Dose-escalating vehicle-controlled; 1%, 2%, 5% nasal gel TID × 3 days	Significant decolonisation below detection limit at 2% and 5% doses from day 2 (P = 0.0008 and P = 0.0012 vs vehicle, respectively) and sustained through day 4 (P = 0.0180 and P = 0.0105); 1% dose showed reduction from day 1 but did not reach significance vs vehicle; effect not durable; recolonisation occurred in all but one subject by approximately 5 days post-treatment, with no significant difference vs vehicle from baseline to week 9 (P = 0.2754); low resistance propensity supported by preclinical/mechanistic data, not demonstrated in this trial; minimal systemic absorption (Cmax 3.72–11.7 ng/mL in the 5% group; undetectable by 1 week)	No systemic issues; minor reversible local lesions	[55]
6	Dusquetide (SGX942; IDR pentapeptide)	Severe oral mucositis (H&N CRT)	2 (Ph 3 failed)	Double-blind dose-escalating RCT; 0.5 mg/kg, 1.0 mg/kg, 1.5 mg/kg IV twice weekly vs placebo (n = 111)	In overall population: 50% ↓ severe OM duration, 18 to 9 days (67% ↓ in high-risk cisplatin subgroup, 30 → 10 days, P = 0.04); 39% ↓ AUC (WHO Grade-time score); 71% ↓ SOM rate at 1-month follow-up; ↓ infection rate; 7% relative ↓ in SOM incidence: 74% → 69%	No dose-limiting toxicity; AEs consistent with CRT	[64]
7	Omiganan (CLS001)	Atopic dermatitis (mild-moderate)	2	Double-blind vehicle-controlled; 1%, 1.75%, 2.5% gel BID × 28 days (n = 80)	93.5% S. aureus reduction at 2.5% (P = 0.02); dysbiosis recovered; clinical EASI/SCORAD not met; microbiome normalisation insufficient for symptom relief	No systemic AEs; good local tolerability	[56]
8	Omiganan (CLS001)	Facial seborrheic dermatitis	2	3-arm RCT; omiganan 1.75% vs ketoconazole 2% vs placebo BID × 4 weeks	Omiganan showed no improvement vs placebo (SDASI P = 0.143, IGA P = 0.097, %BSA P = 0.522) and did not significantly reduce Malassezia (-3.7%, P = 0.563), explaining the failure; staphylococcus declined in all arms including placebo with no omiganan-specific effect (not a primary driver); ketoconazole met all clinical endpoints (SDASI P = 0.025, IGA P = 0.005, %BSA P = 0.005) with reduced Malassezia and restored barrier function	Mild application-site reactions; no systemic AEs	[57]
9	Omiganan (CLS001)	HPV anogenital warts and vulvar HSIL	2	Two concurrent vehicle-controlled RCTs; 2.5% gel QD × 12 weeks (n = 36)	First demonstration of topical AMP antiviral activity in HPV-induced disease: Significant reduction in HPV viral load in AGW patients (-96.6%; 95%CI: -99.9 to -7.4%; P = 0.045), but no significant difference in lesion clearance or wart count/size vs placebo (near-significant trend in wart height reduction, -30.3%, P = 0.054). In the vulvar HSIL arm, no significant difference in viral load or lesion response vs placebo. Reduction in viral load did not translate into clinically meaningful lesion regression in either indication	No serious AEs; excellent local tolerability	[58]
10	Oral LL-37 (CAS001; via GMO L. lactis)	COVID-19 (SARS-CoV-2 Omicron BA.5.1.3)	2	Open-label, randomized, placebo-controlled, single-center; Oral LL-37 vs L. lactis placebo (n = 238; 129 vs 109); early (≤ 6 days) vs late (≥ 7 days) initiation	Significantly shortened nucleic-acid negative conversion time when started early (9.80 days vs 14.04 days, P = 0.0044; early vs placebo HR 2.427, P = 0.0097); early > late initiation; LL-37 acts via viral envelope disruption (and ACE2 blockade)	No serious AEs; no systemic toxicity	[63]

ACE2: Angiotensin-converting enzyme 2; AE: Adverse event; AGW: Anogenital warts; AMP: Antimicrobial peptide; BID: Twice daily; COVID-19: Coronavirus disease 2019; CRT: Chemoradiotherapy; EASI: Eczema Area and Severity Index; GMO: Genetically modified organism; H&N: Head and neck; hLF1-11: Human lactoferrin peptide 1-11; HR: Hazard ratio; HSCT: Haematopoietic stem cell transplantation; HSIL: High-grade squamous intraepithelial lesion; HPV: Human papillomavirus; IDR: Innate defence regulator; IGA: Investigator’s Global Assessment; LPS: Lipopolysaccharide; MRSA: Methicillin-resistant Staphylococcus aureus; MSSA: Methicillin-sensitive Staphylococcus aureus; NCI CTC: National Cancer Institute Common Toxicity Criteria; NS: Not significant; OM: Oral mucositis; Ph: Phase; PK: Pharmacokinetics; QD: Once daily; RCT: Randomized controlled trial; SARS-CoV-2: Severe acute respiratory syndrome coronavirus 2; SCORAD: Scoring Atopic Dermatitis; SDASI: Seborrheic Dermatitis Area and Severity Index; SOC: Standard of care; TID: Three times daily; UOM: Ulcerative oral mucositis.

The most important lesson is that microbiological success does not guarantee clinical benefit. Topical omiganan reduced cutaneous Staphylococcus aureus by over 90% without improving disease scores in atopic dermatitis[54], left the true driver Malassezia untouched in seborrhoeic dermatitis[55], and, in anogenital warts specifically, reduced human papillomavirus viral load without a corresponding reduction in lesions—an effect not replicated in the vulvar high-grade squamous intraepithelial lesion arm of the same programme[56]. An AMP succeeds only when the organism it kills is genuinely causal, so in IBD, where dysbiosis is intertwined with immune dysregulation rather than caused by one pathogen, reducing bacterial load is necessary but unlikely sufficient, and immunomodulatory actions may matter more than bactericidal potency.

A second lesson concerns trial design and vehicle effects. In the iseganan radiotherapy trial, drug and placebo gave an identical 9% mucositis-free rate, and both beat standard care, indicating that intensified oral hygiene and the rinse vehicle drove benefit[57]; the companion chemotherapy trial narrowly missed significance (43% vs 33% mucositis-free, P = 0.067) amid a dispensing error affecting 102 of 323 patients (32%)[58]. A topical peptide competes against a vehicle that may itself be therapeutic, and operational rigour is as decisive as pharmacology.

A third pattern is anatomical: Every clear success has occurred at a surface or lumen, where the peptide could be applied directly and at high local concentration—decolonising the nares[59], reducing colonisation at catheter sites, treating candidiasis as a rinse, or clearing virus through orally delivered LL-37[60-62]. By contrast, AMPs struggle to reach deep or systemic compartments, where their short half-life and protease sensitivity prevent therapeutic exposure—a single intravenous dose of hLF1-11 in autologous haematopoietic stem cell transplantation recipients was well tolerated but unquantifiable in plasma, yielding no efficacy data[63]. This bias is encouraging for gastroenterology: The gut lumen is exactly the accessible, high-concentration compartment in which AMPs succeed, underscoring why luminal delivery is central.

Across these trials, safety was rarely the failure point; efficacy and pharmacokinetics were. Pexiganan, initially equivalent to oral antibiotics, was abandoned when phase 3 showed no benefit and excess infection—a reminder that favourable early tolerability cannot be assumed to persist at efficacious exposures[19]. Encouragingly, immunomodulatory agents fared better in principle: Dusquetide shortened severe mucositis with a clinically meaningful, if non-significant, phase 3 effect, and engineered-probiotic LL-37 met its endpoint via in situ delivery[62,64]. Because no AMP has yet reached controlled clinical trials for a primary gastrointestinal indication, the translational lessons drawn here are necessarily extrapolated from AMP trials in other accessible mucosal and cutaneous compartments—nasal, dermatological, oropharyngeal, and anogenital—rather than from GI-specific clinical data. These trials were selected not for disease-specific relevance but because each illustrates a translational principle—anatomical accessibility, the dissociation of microbiological and clinical endpoints, or the fragility of trial execution—directly applicable to the design of future GI-focused AMP studies. The human data suggest the future of GI AMP therapy lies less in direct bactericidal use than in harnessing immunomodulatory and barrier-restoring functions through locally delivered, stabilised, or engineered systems.

CHALLENGES IN AMP-BASED THERAPEUTICS

The foregoing sections make clear that AMPs exhibit mechanistic versatility and have been shown to be effective in a range of preclinical models of GI infection, inflammation, and malignancy. But the irony is that this promise has not yet been converted into approved systemically acting therapeutics.

The translational gap: Lessons from clinical failure

Pexiganan (MSI-78), among the most clinically advanced AMPs, is instructive: In randomised trials for mildly infected diabetic foot ulcers, it achieved clinical and microbiological outcomes statistically equivalent (in pooled analysis) to oral ofloxacin[19,65], yet in subsequent phase III testing it failed to demonstrate superiority over a topical placebo (vehicle plus standardised wound care) and was not approved. It failed not on safety or intrinsic activity but on trial design and the absence of a differentiated niche. The lesson for GI applications is that demonstrating that an AMP works is insufficient; sponsors must define an indication—such as protease-stable, mucosally restricted activity against a resistant enteric pathogen—where peptides offer an advantage that conventional antibiotics cannot. Much of the preclinical GI literature has not been designed with this comparative endpoint in mind.

Stability, bioavailability, and pharmacokinetics as a single liability

Proteolytic instability and unfavourable pharmacokinetics are 2 expressions of the same vulnerability: The cationic, amphipathic architecture that drives membrane activity also makes AMPs substrates for luminal proteases and rapid renal clearance, producing short half-lives and erratic exposure. The underappreciated consequence is that this directly undermines the clinical-differentiation argument—a peptide degraded within minutes cannot sustain the concentration-time profile needed to outperform a stable small molecule. Stability is therefore not a formulation inconvenience but the rate-limiting determinant of whether an AMP can occupy a defensible niche.

Manufacturing economics as a translational filter

Cost is the most quantitatively neglected barrier, yet it silently shapes which indications are attempted. Solid-phase peptide synthesis scales poorly, with a process-mass intensity roughly 2 orders of magnitude greater than that of small-molecule drugs and reagent costs reaching tens of thousands of dollars per gram[66]. Recombinant expression is a credible alternative: Unit costs fell from approximately 253 €/mg to approximately 42 €/mg as batch size rose from 100 mg to 1000 mg, approaching competitiveness with chemical synthesis[67]. Molecular-farming platforms expressing AMPs in plants promise further reductions while easing endotoxin and scalability constraints[66]. Production economics thus act as a covert filter: The dominance of topical, low-dose indications reflects not only formulation logic but affordability, and systemic GI indications are economically penalised before efficacy is ever tested.

Cytotoxicity and the optimisation problem

The narrow therapeutic window arises because the amphipathicity that permeabilises microbial membranes also engages mammalian bilayers at higher concentrations, producing dose-limiting cytotoxicity and haemolysis. Widening it through empirical residue substitution has been slow. Critically, however, selectivity, stability, and synthetic tractability are all functions of primary sequence, reframing toxicity from an intractable caveat into a defined, multi-objective design problem amenable to computation.

Viewed critically, these barriers form a single causal chain: Instability erodes pharmacokinetics, which negates the clinical-differentiation argument, while cost dictates which indications are viable. Progress depends less on discovering more peptides than on integrating computational design, scalable low-cost manufacturing, and GI-targeted delivery into a coherent strategy and on designing trials around endpoints where peptides can demonstrably outperform existing agents. The principal gaps are evidentiary: Head-to-head comparative efficacy, long-term safety in chronic GI models, and transparent cost-of-goods analyses at clinical scale. The next 2 sections address the first and third elements of this chain—computational design and targeted delivery.

AI AND COMPUTATIONAL DESIGN OF NEXT-GENERATION AMPs

The rationale for computational discovery

The barriers above, such as narrow therapeutic windows, proteolytic instability, and prohibitive cost, are ultimately problems of sequence selection. A peptide of length n occupies a space of 20ⁿ possibilities, far exceeding empirical screening. Computational approaches reframe AMP development as a navigable optimisation problem, allowing potency, selectivity, and stability to be co-engineered before synthesis, which is particularly pertinent to the gut, where peptides must additionally survive a protease-rich lumen.

Discriminative learning and microbiome-scale mining

The most striking demonstration of scale comes from discriminative machine learning applied to the global microbiome: Mining 63410 metagenomes and 87920 prokaryotic genomes, a deep-learning pipeline predicted close to 1 million candidate AMPs, of which 79 of 100 synthesised peptides were active in vitro, and several matched a clinical antibiotic in murine models[68]. The relevance is direct—the intestinal microbiome is itself a vast, untapped reservoir of host-defence peptides, and machine learning is the only tractable route to surveying it systematically.

Generative and de novo design

Beyond mining, generative models create novel peptides absent from nature. Architectures have progressed from generative adversarial and variational networks to diffusion models coupled with protein language models: One such framework couples a pretrained protein language model with a diffusion model operating in a shared latent embedding space to generate thousands of structurally diverse candidate sequences within hours. In this pipeline, 44 of 45 synthesised peptides showed activity against Gram-positive or Gram-negative bacteria, and the lead candidate combined potency with low haemolysis and demonstrated in vivo efficacy against a drug-resistant E. coli strain in a model of acute peritonitis[69]. Generative design has also yielded bifunctional peptides—a generative adversarial network generator paired with an activity regressor produced AMPs with combined antibacterial and antiviral action, with 16 validated in vitro and in animals[70]. This multifunctionality is well suited to the mixed infectious-inflammatory pathologies of gut disease.

Predictive modelling of toxicity, stability, and proteolysis

Computation’s decisive advantage is filtering for liabilities at the design stage. Models predict cytotoxicity, haemolysis, and, critically for the gut, proteolytic susceptibility: A random-forest cleavage-site predictor applied across human proteomes revealed that de-extinct archaic peptides resist enzymatic degradation better than modern counterparts, explicitly linking design to the stability barrier limiting oral AMPs[71]. Equally important is interpretability: Feature-attribution methods such as SHapley Additive exPlanations (SHAP) quantify each residue’s contribution, transforming opaque predictions into design rules and yielding validated sequences[72]—pre-empting the concern that AI-designed therapeutics are unaccountable ‘black boxes’.

From extinct proteomes to preclinical leads

Computational promise is meaningful only if it produces tractable molecules. The de-extinction paradigm provides proof of concept: The Antibiotic Peptide de-Extinction (APEX) deep-learning model mined extinct-organism proteomes and identified preclinical candidates, including mammuthusin and elephasin, with demonstrable in vivo efficacy[73]. With microbiome mining, this establishes a reproducible candidate-to-preclinical pipeline, the translational evidence distinguishing a forward-looking review from a speculative one[68].

Toward GI-targeted in silico design

A critical gap remains: Almost none of these platforms have been optimised for the GI context. Current models train predominantly on systemic antibacterial activity, with little attention to luminal proteases, mucus penetration, microbiota selectivity, or commensal sparing. The logical next step is to embed GI-specific constraints—regional pH, host and bacterial protease repertoires, and commensal-sparing selectivity—directly into generative objectives, coupling in silico design with the gut-targeted delivery described in the next section and validating these approaches prospectively in intestinal infection and colitis models. Computation has expanded the search space by orders of magnitude; the priority now is to focus it on the unique demands of the gut, so the next generation of peptides is designed to be not merely potent but deliverable, selective, and stable where needed. Figure 1 brings together the full AMP development pipeline, from computational discovery through molecular optimisation, formulation engineering, and clinical translation. Together, these stages show how machine learning, nanocarrier systems, and engineered live biotherapeutics are turning AMPs from endogenous immune mediators into precision therapies for GI disease.

Figure 1 Antimicrobial peptide development pipeline: From computational discovery through molecular optimisation, formulation engineering, and clinical translation. AI: Artificial intelligence; AMP: Antimicrobial peptide; GI: Gastrointestinal.

ENGINEERING, FORMULATION, AND GUT-TARGETED DELIVERY STRATEGIES

If computational design solves the sequence half of the problem, delivery solves the rest. A potent peptide that is degraded in the lumen, cleared before reaching its target, or toxic at systemically required doses fails regardless of its intrinsic design—so the strategies below are best read as engineering responses to the stability, pharmacokinetic, and toxicity barriers.

Nanoparticle, lipid, and stimulus-responsive carriers

Polymeric, lipid, inorganic, and hybrid nanoparticles transfer the burden of survival from the peptide to a robust carrier, shielding it from proteases while enabling controlled release. The most compelling GI example couples encapsulation with disease-specific triggering: An oxidation-responsive nanoparticle carrying the annexin A1-mimetic peptide Ac2-26 shields its cargo from the lumen and releases it selectively at inflamed colonic sites in response to elevated reactive oxygen species, reducing inflammation, accelerating healing, and reshaping the microbiota toward short-chain fatty acid production[74]. Lipid carriers are the most clinically mature platform; surface engineering extends them to the oral route, where a tetraether-lipid–stabilised liposome decorated with cyclic cell-penetrating peptides withstood the gastric milieu while acting as a permeation enhancer, significantly increasing the oral bioavailability of a vancomycin-derived antibiotic with retained efficacy[75].

Hydrogels, self-assembly, and biomimetic systems

Hydrogels decouple therapeutic duration from peptide half-life: An alginate hydrogel delivering Chol-37 (F34-R) sustained activity and accelerated healing in a murine Pseudomonas aeruginosa infection[76]. An alternative is to engineer the peptide itself into a protective structure—proteolytically resistant self-assembling nanofibres cross-link into nanonets that trap and kill bacteria while resisting trypsin, chymotrypsin, and pepsin, uniting stability and antibacterial function in a single biomimetic design suited to the protease-dense gut[77]. Multifunctional oral nanomedicines integrate several disease axes at once: A silk-fibroin nanoparticle co-delivering epigallocatechin-3-gallate and surface-functionalised with cathelicidin-BF, administered within a chitosan/alginate hydrogel, restored barrier integrity, drove anti-inflammatory macrophage polarisation, and enriched beneficial taxa in experimental UC[78].

Live biotherapeutics and emerging frontiers

Rather than delivering a finished peptide, engineered probiotics can produce it in situ in the gut, in principle sidestepping both bioavailability and cost barriers—exemplified by L. lactis engineered to deliver human cathelicidin (hCAP-18), where bacterial delivery of hCAP-18-encoding DNA for expression by host epithelial cells alleviated experimental colitis more effectively than direct bacterial secretion of the peptide, and by genetically engineered Bacillus subtilis displaying cathelicidin-BF within secreted extracellular vesicles that resist digestive fluid and alleviate DSS-induced colitis while restoring the barrier and microbiota[52]. Looking further ahead, several synthetic-biology and engineering approaches could, in principle, be adapted to gut-targeted AMP delivery: Chassis organisms governed by inflammation-responsive promoters that would confine peptide production to active disease, CRISPR-based strategies aimed at pathogen-selective killing, AI-guided formulation, 3-dimensionally printed personalised dosage forms, and theranostic systems that would couple delivery with disease monitoring. These directions remain speculative for AMPs specifically and should be regarded as conceptual possibilities rather than demonstrated approaches.

CONCLUSION

AMP biology is defined by a delicate balance between protection and pathology. The same cationic, amphipathic molecules that establish colonisation resistance, segregate microbes from the epithelium, restore the barrier, and resolve inflammation can, when their concentration, location, timing, or microbial context shifts, drive angiogenesis, sustain pathobionts, and promote malignancy. This duality is not a curiosity at the margins of AMP biology but its organising principle, and it dictates therapeutic strategy: Deficiency-driven disease calls for restoration, whereas signalling-driven disease calls for restraint. The LL-37 concentration switch makes the stakes concrete—the same peptide that heals at low local levels can promote tumour growth if delivered indiscriminately.

Translating this understanding has been slow because the obstacles are interdependent: Instability undermines pharmacokinetics, pharmacokinetics undermine clinical differentiation, and manufacturing cost constrains which indications are even attempted. Yet the convergence now underway is precisely what these coupled barriers require. Peptide engineering and AI-guided design widen the therapeutic window and confer stability at the sequence level; targeted, stimulus-responsive delivery confines activity to diseased mucosa; and synthetic biology produces peptides in situ, dissolving the bioavailability and cost barriers together. The gut lumen, which is accessible, tolerant of high local concentrations, and the compartment most amenable to topical AMP delivery, is an unusually favourable arena for this convergence. The path forward is therefore integrative rather than incremental. Advances in peptide engineering, targeted delivery, synthetic biology, and AI are collectively transforming AMPs from endogenous immune mediators into next-generation precision therapeutics for gut disease, agents designed not merely to be potent but to be deliverable, selective, stable, and deployed in the disease context where their duality favours protection over harm. Realising this will require GI-relevant computational training data, head-to-head comparative efficacy studies, long-term safety evaluation in chronic disease models, and transparent cost analyses at clinical scale. The biology is no longer the limiting factor; the discipline of matching the right peptide, in the right form, to the right disease context now is.