Editing the Microbiome in Place: From Engineered Probiotics to Programmable Ecological Therapies

Key Takeaways
In situ microbiome editing shifts the therapeutic model from administering engineered probiotics to programming bacteria already living in the patient.
Current approaches include transient gene regulation, CRISPR-mediated bacterial depletion, base editing, and pathway-scale DNA insertion.
Engineered bacterial donors and phage-derived systems offer different advantages in host range, payload capacity, persistence, and biological containment.
CMC strategies must connect the manufactured vector with editing efficiency, functional output, ecological effects, and the persistence of modified bacteria.
CRISPR-armed phage therapies have entered human testing, but durable genomic rewriting and functional installation in native human microbiomes remain preclinical.

From Administering Microbes to Programming Microbial Communities

The first generation of engineered bacterial medicines has largely followed a familiar product model: select a microbial strain, modify it outside the body, manufacture it under controlled conditions, and administer it to the patient. The organism may be designed to sense a disease-associated signal, consume a harmful metabolite, produce a therapeutic molecule, or interact with the immune system, but the engineered strain itself remains the manufactured therapeutic agent. This model has opened a path toward living medicines while also creating persistent challenges involving strain selection, engraftment, biological containment, potency, stability, and consistent manufacturing.¹

A newer set of technologies establishes a different strategy to achieve those ends. Rather than relying on an introduced strain to establish itself and perform the desired function, these systems deliver genetic instructions to bacteria that already occupy the patient’s microbiome. The administered product may be an engineered bacterial donor, a bacteriophage, or a nonreplicative phage-derived particle. Its purpose is not necessarily to persist as the therapeutic organism. Instead, it transfers molecular machinery that changes the abundance, behavior, or genome of a selected resident population.

Early demonstrations of this concept used orally delivered bacteriophages to alter gene expression within gut bacteria. Subsequent systems employed engineered probiotics or phages to deliver clustered regularly interspaced short palindromic repeats (CRISPR) machinery that selectively killed targeted bacterial strains. More recent studies have expanded the available interventions to include nucleotide-level base editing and the insertion of multikilobase DNA cargoes into commensal bacteria living in the gut.^2–6

These advances place several distinct therapeutic mechanisms under the broad label of microbiome editing. A delivered construct might transiently suppress a bacterial gene without permanently changing the genome. A CRISPR antimicrobial might remove a selected strain while sparing closely related organisms that lack the targeted sequence. A base editor might introduce a precise nucleotide substitution that remains after the delivery vehicle disappears. A CRISPR-associated transposase might install an entire metabolic pathway or other functional program in a resident organism. Each approach modifies the microbiome, but each creates a different relationship among the administered product, the bacterial target, and the resulting biological effect.

That distinction has practical consequences for drug development. With a conventional engineered live bacterial therapeutic, developers can generally define the product around the identity, viability, purity, genetic stability, and functional activity of the organism in the administered dose. In an in situ editing system, the manufactured vector is only the first component of a multistep process. It must survive administration, reach the relevant gastrointestinal environment, encounter a susceptible recipient, transfer its payload, execute the intended molecular change, and generate an altered bacterial population capable of producing the desired effect. The abundance and persistence of that population may then depend on its fitness within a complex and variable microbial community.

The resulting product concept is less straightforward than it is for an administered therapeutic strain. The delivery organism or particle can be manufactured, tested, released, and dosed, but the modified organism may arise only after administration. In some architectures, the vector may disappear while the genomic change and its functional consequences remain. In others, continued maintenance of the delivered construct may be necessary to sustain the effect. Selective depletion adds another variation: the relevant outcome may be the removal of a bacterial population rather than the creation of a new therapeutic organism.

This emerging field therefore extends beyond the idea of an improved probiotic. It introduces a family of programmable interventions that can regulate microbial genes, eliminate selected strains, rewrite bacterial sequences, or install new functions within an established community. The possibilities are significant, but editing organisms in place does not remove the development challenges associated with living medicines. It relocates many of them from the manufactured strain to the interaction among the delivery system, the intended recipient, and the surrounding ecosystem.

The most advanced examples also remain unevenly distributed across the development spectrum. CRISPR-armed phage products designed to reduce targeted gastrointestinal bacteria have entered human clinical evaluation, but pathway-scale insertion and durable genomic rewriting of resident gut organisms have so far been demonstrated in animal models rather than patients.^5–7 The near-term opportunity is therefore grounded in credible technical progress, while the broader vision of programmable ecological therapy still depends on solving fundamental questions of delivery, specificity, persistence, containment, potency, and control.

Why Edit Resident Bacteria?

Resident bacteria offer a compelling target because they have already succeeded in the ecological task that often challenges an administered therapeutic strain: occupying a niche within the host. They have adapted to local nutrient availability, neighboring organisms, immune pressure, and the physical conditions of their intestinal habitat. Modifying an established organism could therefore allow a therapeutic program to operate from within an existing microbial community rather than requiring a newly introduced strain to compete for durable residence.

This approach may also expand the range of organisms that can be used therapeutically. Some commensals may possess valuable metabolic or immunological functions but remain difficult to isolate, engineer, manufacture, formulate, or administer as conventional live biotherapeutic products (LBPs). In situ delivery could, in principle, reach those organisms without turning each target strain into a separately manufactured drug substance. A platform that delivers programmable cargo to selected bacteria might also be adapted more readily across targets than a development strategy that requires a new therapeutic strain for every indication.

The potential advantage is not limited to engraftment. A transient delivery vector could produce a longer-lived change if the edit becomes integrated into the target genome and the altered organism remains competitive. That separation between vector persistence and functional persistence is central to the appeal of in situ editing. It could reduce the need for continuous administration while retaining an effect within the microbial population, although the duration and predictability of that effect will depend on the target organism, the edit, and the surrounding ecology.^5,6

However, working with resident bacteria introduces a different form of variability. The target may be abundant in one patient, scarce in another, or absent altogether. Strains classified as the same species may differ in their susceptibility to a phage, conjugative system, guide sequence, or editing payload. Antibiotic exposure, diet, inflammation, gastrointestinal transit, and baseline microbiome composition may further alter access to the intended recipient. A dose that delivers adequate editing in one microbial environment may perform differently in another.

The relevant delivery problem therefore extends beyond getting a formulation through the gastrointestinal tract. The vector must enter the correct anatomical region, encounter the target organism, overcome bacterial defense systems, transfer the payload, and execute the intended change. The edited bacterium must then remain viable and sufficiently fit to persist or expand. Failures at any of these stages could produce a weak or inconsistent response, even if the manufactured product meets all conventional release specifications.

This distinction between administration and intracellular delivery will shape development programs. Measuring how much vector reaches the intestine cannot establish how much editing occurred, and detecting an edit does not prove that the intended function reached a therapeutically meaningful level. Developers will need to connect vector exposure, recipient access, molecular modification, functional output, and ecological persistence through a chain of assays and biomarkers.

A Spectrum of Programmable Interventions

Microbiome editing encompasses several mechanisms with different therapeutic objectives. The least permanent approach is transient regulation of bacterial gene expression. An orally delivered phage system has been used to carry a nuclease-deactivated CRISPR construct that repressed a selected bacterial gene in the gut without necessarily creating a permanent genomic change. This type of intervention could be useful when reversibility is desirable or when temporary suppression is sufficient, although its duration may depend on continued maintenance or redelivery of the construct.²

Selective depletion takes a more aggressive approach. Engineered bacterial donors and bacteriophages can deliver CRISPR machinery programmed to recognize specific DNA sequences in a target population. Cleavage of those sequences can kill the recipient bacterium while leaving organisms without the target sequence relatively unaffected. This level of discrimination could allow treatment of pathogenic, antibiotic-resistant, or otherwise harmful strains without the broad collateral effects associated with conventional antibiotics.^3,4

Sequence-directed killing also creates its own development pressures. The target sequence must be sufficiently conserved across clinically relevant strains, yet sufficiently distinct to avoid unwanted activity in related commensals. A bacterial population may escape through mutation, altered phage receptors, loss of the targeted element, or acquisition of protective mechanisms. Phage cocktails and complementary host ranges can reduce some escape routes, but they also increase manufacturing and characterization complexity.⁴

Base editing changes the target bacterium without necessarily killing it. A phage-derived particle has delivered a base editor to gut-colonizing bacteria in mice, producing a specific nucleotide change in a substantial fraction of the target population. Edited bacteria remained detectable after the nonreplicative delivery construct was no longer intended to persist, illustrating how a short-lived vector can leave a longer-lived genomic effect.⁵

Pathway-scale insertion expands the concept further. CRISPR-associated transposases can integrate larger DNA cargoes than nucleotide editors, creating the possibility of installing multigene pathways, biosensors, control circuits, or therapeutic production functions. In mice, a conjugative system inserted multikilobase payloads into native commensal bacteria, including a pathway that enabled utilization of a defined dietary substrate.⁶

These mechanisms should not be grouped under a single development template. Transient repression may require repeated dosing but offer greater reversibility. Depletion can be evaluated through reduction of a target population, whereas base editing requires accurate measurement of allele conversion. Pathway insertion must establish not only integration efficiency and sequence accuracy but also expression, functional activity, genetic stability, and any fitness cost imposed on the recipient. To be clear: the term microbiome editing describes a technological family, not a uniform product class.

Choosing a Delivery Architecture

The two leading delivery architectures use engineered bacteria or bacteriophages as genetic couriers. Each offers distinct advantages, constraints, and safety considerations.

Conjugative donors exploit natural mechanisms of bacterial DNA transfer. An engineered donor carries a mobile construct containing the editing machinery and transfers it directly to recipient bacteria. This approach can potentially reach organisms that lack a suitable characterized phage and can support delivery of relatively large genetic payloads. Engineered probiotics have already been used as conjugative vehicles for CRISPR-mediated bacterial killing, while later systems extended the concept to genomic integration of larger functional cargoes.^3,6

A living donor must itself be controlled. Its identity, genetic stability, viability, antibiotic susceptibility, and transfer capacity become critical quality attributes. Developers must determine how long the donor persists, whether it transfers DNA to unintended recipients, and whether its containment features remain effective. Auxotrophy can restrict survival outside defined conditions, and kill switches may provide additional control, but each safeguard must function reliably during manufacture, administration, gastrointestinal transit, and environmental exposure.

Phage-based systems offer a different form of specificity. Their ability to infect bacteria depends heavily on host receptors and other strain-level features, allowing narrow targeting but also limiting coverage across diverse clinical isolates. Tail-fiber engineering, receptor-binding modifications, and phage cocktails can broaden host range, while nonreplicative phage-derived particles can deliver editing cargo without establishing a self-propagating vector population.^4,5

Host range is both a design feature and a source of variability. A phage may perform well against one strain but fail against another within the same species. Bacteria can alter or mask receptors, deploy restriction systems, or acquire resistance. A clinically useful phage product may therefore require a panel of complementary components or an adaptable platform capable of matching vectors to the patient’s target strain. Either strategy increases the analytical burden associated with identity, potency, and comparability.

Payload size may further influence platform selection. A compact repression or killing construct may fit within a phage-derived delivery system, whereas multigene pathway installation may favor conjugative transfer or another architecture with greater cargo capacity. The desired duration also matters. A self-limiting particle may be preferable when transient exposure is important, while durable genomic integration may require efficient transfer into a stable resident population.

Neither platform is inherently safer or more controllable in all circumstances. A bacterial donor can replicate and transfer DNA, while a phage may propagate, recombine, or encounter a narrow host range unless specifically designed otherwise. Platform choice must follow the biological target, payload, mechanism, and intended duration rather than a generalized preference for bacteria or phages.

Redefining Dose, Exposure, and Persistence

For conventional drugs, the administered dose often provides a useful starting point for understanding exposure. In situ microbiome editors create several linked but non-equivalent doses. A bacterial product may be quantified in colony-forming units, while a phage preparation may be described by particle number or infectious titer. Those values indicate how much vector was administered but not how much of the target population received or expressed the payload.

Biological exposure depends on the abundance and accessibility of the recipient bacteria, the frequency of transfer or transduction, and the fraction of successful editing events. A large administered dose could have little effect if the target is absent or resistant. A smaller dose might produce substantial editing if the vector encounters a susceptible and abundant population. The relationship between administered vector and edited cells may therefore vary across patients even when product quality and dosing remain constant.

Three timelines must be considered separately. The first is vector exposure, including how long the donor organism or phage remains detectable. The second is the period during which transfer and editing occur. The third is the persistence of the altered population and its functional effect. A vector could disappear quickly while the edited bacteria remain, or the vector could persist without producing sufficient editing.

This separation complicates pharmacokinetic and pharmacodynamic analysis. Stool recovery may help measure gastrointestinal passage or shedding, but it may not represent vector concentrations at the relevant mucosal site. Sequencing can quantify editing frequency in sampled bacteria, but the measured population may not reflect microbial niches elsewhere in the intestine. Functional biomarkers may therefore become essential for connecting the genomic intervention to the intended biological effect.

Persistence will also be shaped by microbial fitness. An edit that imposes a substantial metabolic burden may decline even when integration is technically successful. Conversely, a function that improves access to a nutrient could allow edited organisms to expand under selected dietary conditions. In one pathway-insertion experiment, providing a corresponding dietary substrate favored bacteria carrying the installed function, and withdrawing that substrate reduced the selective advantage.⁶

This result suggests that diet or another external input could potentially become part of a control strategy. Such an approach might support expansion of edited cells, limit persistence, or help modulate therapeutic intensity. It remains a preclinical concept, however, and human microbial ecosystems may respond less predictably. Diet could also introduce additional sources of variability, adherence challenges, and interactions with other community members.

The dose for these products may ultimately need to combine a conventional administered quantity with a biological target range. Developers may need to specify an acceptable level of vector exposure, editing frequency, functional output, and persistence rather than rely on a single numerical dose. Establishing which of those measures best predicts clinical benefit will be a central challenge in translational development.

Containing an Intervention Designed to Spread Genetic Information

Some in situ platforms rely on horizontal gene transfer to deliver their therapeutic cargo. The same property that enables treatment can create a safety risk if the payload reaches an unintended organism, persists longer than planned, or enters the environment.

Off-target activity can occur at several levels. The editor may modify an unintended sequence within the target bacterium. The delivery construct may enter a nontarget strain or species. The donor, phage, or mobile element may persist and continue to disseminate. Even a precise edit in the intended recipient may alter competition, metabolite exchange, or community structure in ways that were not predicted from isolated cultures.

Containment must therefore address both the vector and the resulting edit. Nonreplicative particles can limit maintenance of the delivery construct, while auxotrophic donors can be designed to depend on nutrients unavailable outside controlled conditions. Kill switches, restricted host ranges, nonmobilizable payloads, removal of selectable markers, and retained susceptibility to rescue antibiotics offer additional control options.^1,5,6

No single safeguard is likely to be sufficient. Mutations can disable genetic containment systems, environmental conditions can differ from laboratory assumptions, and mobile elements may behave differently across recipient strains. Layered containment may therefore combine biological restrictions, product design, clinical monitoring, and environmental surveillance.

Evolutionary escape adds another dimension. Phage receptors can change, target sequences can mutate, and edited organisms may be outcompeted by unedited neighbors. A therapy intended to reduce a harmful strain could select for variants that evade the vector. A pathway insertion could disappear if the new function carries a fitness cost. Resistance monitoring will need to consider both molecular escape from the editor and ecological replacement of the intended target.

Microbiome composition alone may not capture these outcomes. Broad taxonomic profiles can remain stable while strain-level abundance, gene content, or metabolic activity changes. Appropriate monitoring may require targeted culture, quantitative polymerase chain reaction, metagenomic sequencing, long-read analysis, vector-specific assays, metabolomics, and tests for transferable genetic material.

Shedding studies will also need to distinguish among the administered vector, the editing construct, and edited recipient bacteria. A donor strain may no longer be detectable even though the payload has moved into another organism. A phage may appear in stool without evidence of systemic distribution. An edited commensal may persist after vector clearance and could potentially be transferred between individuals. Existing guidance for bacteria- and virus-based gene therapy products provides a starting point for evaluating shedding and transmission, but in situ microbiome editing may require broader, platform-specific assays.^8,9

From Animal Models to Human Translation

The field has advanced through a sequence of increasingly capable preclinical systems. Early studies showed that orally administered phages could alter bacterial gene expression in the gut. Engineered probiotics then delivered CRISPR systems through conjugation to selectively deplete target bacteria. Phage cocktails extended precision killing across a broader set of strains, while nonreplicative particles enabled nucleotide editing without requiring continued maintenance of the payload. CRISPR-associated transposases have now expanded the potential cargo from small sequence changes to multikilobase functional programs.^2–6

Human translation has begun with the depletion branch of the field. An orally administered cocktail of four CRISPR-Cas–armed bacteriophages targeting gastrointestinal Escherichia coli has completed a randomized, double-blind, dose-escalation phase I study in healthy adults. The trial evaluated safety, recovery, and pharmacodynamics rather than clinical efficacy, but it demonstrates that a programmable, phage-based microbiome product can enter formal human testing.^7,10

This milestone should not be confused with clinical validation of native bacteria editing more broadly. The human product is designed to reduce a target bacterial population, not install a new pathway or create a durable edited commensal. Base editing and pathway insertion remain preclinical, and no verified source has demonstrated therapeutic benefit from genomic rewriting of resident human gut bacteria.

Translation will require confronting forms of heterogeneity that controlled mouse models cannot fully reproduce. Human target strains vary genetically, phage susceptibility differs among isolates, and the relevant organism may not be present in every prospective patient. Diet, medications, prior antibiotic exposure, inflammation, and intestinal anatomy can alter vector access and microbial behavior. Patient selection may therefore require diagnostic testing to confirm the presence and susceptibility of the target population.

Sampling presents another challenge. Stool is accessible and useful, but it does not provide a complete picture of microbial populations associated with mucosal surfaces or different intestinal regions. A product could affect a local niche without producing a strong fecal signal, or fecal recovery could overstate exposure at the site responsible for disease. Clinical programs may need to combine stool assays with functional biomarkers, circulating metabolites, imaging, or tissue sampling where feasible.

The first successful indications may be those with clearly defined bacterial targets and measurable pharmacodynamic outcomes. Selective depletion of a pathogenic or drug-resistant strain offers a relatively direct relationship among target presence, vector activity, and change in abundance. Installing a metabolic or immunomodulatory function creates a longer causal chain and may require more extensive evidence to connect editing with clinical benefit.

Defining and Controlling the Manufactured Product

Chemistry, manufacturing, and controls (CMC) become especially difficult when the manufactured vector is not identical to the biologically active entity that ultimately produces the therapeutic effect. The administered donor or phage can be characterized before release, but the edited bacterium is generated inside the patient and may differ according to the target strain and microbial environment.

Product identity must begin with the manufactured component. For a bacterial donor, this includes strain identity, complete genetic sequence, editing cassette, transfer machinery, containment modifications, and antibiotic susceptibility. For a phage or phage-derived particle, identity may include capsid- and receptor-binding components, production host, guide sequence, editor, payload, and replication characteristics. A cocktail requires confirmation of each component and its relative contribution.

Purity considerations will depend on the platform and manufacturing process. Bacterial products may contain unwanted microorganisms, adventitious phages, residual media components, or unintended mobile elements. Phage preparations may require control of residual production-host DNA and proteins, endotoxin, inactive particles, contaminating phages, and replication-competent material where the intended product is nonreplicative. The possible presence of antibiotic-resistance genes or other transferable sequences deserves particular scrutiny.

Potency must connect the released lot to the intended mechanism. Viability or particle titer alone may not predict transfer, transduction, or editing efficiency. A potency assay for a conjugative donor may need to measure transfer into a representative recipient and successful completion of the intended edit. A phage product may require both infectious or delivery activity and confirmation of CRISPR-mediated killing or modification. A pathway-insertion product may also need to demonstrate expression and functional output from the installed genes.

Representative recipient panels could become important. A product intended to target a genetically diverse species should be tested across clinically relevant isolates rather than a single laboratory strain. Host-range testing, editing efficiency, off-target recipient analysis, and resistance emergence could help define whether a lot retains the expected biological profile. These assays may be more complex than conventional release methods, but they address functions that are central to clinical performance.

Stability must include genetic and functional properties in addition to conventional measures. A donor can remain viable while losing transfer efficiency. A phage preparation can retain particle count while experiencing a reduction in infectivity or host range. An editing cassette can acquire mutations that alter guide activity, payload integrity, or containment. Storage and shipping studies must therefore evaluate the attributes that enable delivery and editing, not merely survival of the vector.

Comparability will present similar challenges. Changes to the production host, fermentation process, purification conditions, formulation, or scale could affect host range, transfer frequency, particle composition, or payload stability without producing an obvious change in standard quality tests. Orthogonal functional assays will be necessary to show that process changes preserve the complete chain of activity from vector administration to target modification.

Existing U.S. Food and Drug Administration (FDA) guidance for LBPs addresses identity, purity, potency, manufacturing controls, and stability, but it excludes products intended to serve as gene therapy vectors. Separate guidance for microbial vectors covers bacteria and other microorganisms used to deliver genetic material.^9,11 An engineered bacterial courier may sit near the boundary between these frameworks, reinforcing the need for early regulatory discussion.

A Regulatory Framework Divided Among Existing Categories

In situ microbiome editors do not fit neatly within a single established regulatory category. A live bacterial donor resembles an LBP in its manufacture and administration, but its primary purpose may be to transfer genetic material rather than act directly as the therapeutic organism. A phage product may function as an antibacterial agent, a gene-delivery vector, or both. A nonreplicative particle may disappear rapidly, while the resulting genomic edit persists in a resident organism.

Existing guidance provides relevant principles without resolving all of these questions. The FDA’s LBP guidance establishes CMC expectations for live microorganisms intended to prevent, treat, or cure disease but excludes LBPs intended as gene therapy vectors. Guidance for microbial vectors addresses products that use bacteria or other microorganisms to transfer genetic material and includes recommendations related to characterization, biological activity, biodistribution, persistence, antibiotic susceptibility, and environmental considerations.^9,11

This division raises questions about which component should anchor regulatory control. The manufactured vector is the material administered to the patient, but the edited recipient may be the durable source of activity. Regulators may need to evaluate both the consistency of the manufactured lot and the range of biological outcomes it can generate across variable microbial targets.

Permanent editing also introduces questions associated with reversibility and follow-up. Stopping administration may eliminate further vector exposure but may not remove the edit. If an altered commensal persists, developers will need to understand its long-term fitness, shedding, transmissibility, and functional stability. The appropriate duration of monitoring may differ from that for a transient phage antimicrobial or an LBP requiring continued dosing.

Environmental assessment could become relevant when recombinant bacteria, phages, or edited organisms may be released from treated patients. The risk will depend on persistence, host range, mobility of the genetic payload, and the ability of containment systems to restrict spread. Shedding assays may need to distinguish viable vector, vector DNA, edited recipient organisms, and independently mobile payloads.^8,9

No verified guidance currently defines a therapeutic ecological state as the regulated product. Changes in community composition or function may serve as pharmacodynamic endpoints, but the release specification must still apply to material that can be manufactured and tested before administration. A product intended to create a predictable ecological outcome will therefore need a validated relationship between its preadministration attributes and its in vivo effects.

Early engagement with regulators will be essential because different programs may require different combinations of LBP, bacteriophage, microbial vector, and genome-editing principles. The regulatory strategy for a self-limiting CRISPR phage designed to deplete a strain may differ substantially from that for a conjugative donor intended to install a durable metabolic pathway.

Toward Programmable Ecological Therapies

The most important advance in this field is the growing ability to deliver genetic instructions to selected members of an existing microbial community. This capability could support interventions that remove harmful strains, suppress disease-associated pathways, alter resistance determinants, install metabolic functions, or produce therapeutic molecules at a local site. More elaborate systems could eventually combine sensing, computation, and output functions within resident bacteria.

The appeal of such therapies lies partly in their potential durability and precision. A transient vector might establish a long-lived function in an organism already adapted to the patient. Sequence-directed targeting could distinguish among strains that conventional antibiotics or broad microbial interventions affect indiscriminately. Environmental inputs, including diet, might provide a way to tune the fitness or activity of edited populations.

Those possibilities remain contingent on control. A clinically viable product must reach a variable target population, execute the correct change at a reproducible level, avoid unintended recipients, and maintain an acceptable safety profile after the manufactured vector has disappeared. Developers must be able to measure not only the administered product but also the molecular, functional, and ecological consequences that follow.

The first human trial of a CRISPR-armed phage product shows that precision microbiome targeting can progress beyond animal models. The next stages will test whether the field can move from depletion toward durable rewriting and functional installation without losing control of specificity, persistence, and spread. Success will depend on close integration of synthetic biology, phage engineering, microbial ecology, translational pharmacology, analytics, manufacturing, and regulation.

In situ editing may ultimately change the product logic of microbiome therapeutics. Rather than manufacturing every therapeutic function within a dedicated strain, developers could manufacture a programmable delivery system that installs or modifies functions within bacteria already present in the patient. That model would create a new class of medicines whose activity emerges through controlled interaction with a living ecosystem. Its promise is substantial, but its development will require standards capable of controlling both the product that enters the patient and the biological state it leaves behind.