Microbiome Engineering

Established The claim under test is that the microbial community living in and on a person is a control surface — something that can be read, diagnosed against a reference state, and rewritten to produce a chosen physiological outcome. Established The measured record supports a much narrower statement. One indication has a regulatory approval, an early-stopped randomised trial and two licensed products behind it: recurrent Clostridioides difficile infection. Frontier Everything else — obesity, depression, autism, inflammatory bowel disease, metabolic syndrome, the whole gut–brain literature — rests on associations whose measured magnitudes, where anyone has gone back and measured them properly, are small enough to be indistinguishable from methodological noise.

Established This is the subject where the gap between what is known and what is said is widest, so this brief is organised around the gap. Established The single most-repeated fact about the human microbiome — ten bacterial cells for every human cell — is wrong by a factor of ten; the ratio is approximately one to one. Established The single most-cited disease signature in the field, the type 2 diabetes gut signature, was substantially a signature of metformin. Established An entire claimed organ microbiome, the placental one, dissolved into reagent contamination and delivery-room acquisition when somebody ran the controls at scale. Established These are not fringe corrections; each comes from a peer-reviewed primary and each overturned something that had been repeated for years.

Speculative The exotic end of this subject is genuinely interesting and is treated seriously here: designed consortia assembled to specification, whole-community artificial selection, engineered strains carrying therapeutic circuits, orthogonal nutrient niches that would make colonisation deterministic rather than probabilistic. Handwave What none of them has yet done is install a specified microbial community in an arbitrary human and keep it there. Established The one intervention that reliably changes a human gut community does so by replacing it wholesale with somebody else’s, and nobody can say which part does the work.

Established Faecal microbiota transplantation for recurrent C. difficile infection is a real result and should be stated as one. The founding randomised controlled trial was published in the New England Journal of Medicine in January 2013 and was stopped early because continuing to randomise patients away from the treatment arm was judged unethical. Established Its three arms: donor faeces by nasoduodenal tube resolved 13 of 16 patients after a single infusion, with two of the remaining three resolving after a second, for 15 of 16 overall; vancomycin alone resolved 4 of 13; bowel lavage alone resolved 3 of 13. Established P was below 0.001 and adverse events in the treatment arm were limited to mild abdominal cramping. A later review of 317 patients reported resolution in 92% of persistent and recurrent cases, the 2017 IDSA guidelines recommend the procedure after a second recurrent episode, and open-label case-series efficacy in antibiotic-refractory disease is quoted at 85–90%.

Established Keep the denominators. The 81% that founded a field is 13 divided by 16. Established That does not make it wrong — the effect is enormous, it replicated in clinical practice, and it drove a guideline change and two drug approvals — but it is a point estimate with very wide confidence bounds, and it is routinely quoted as though it were a population statistic. Frontier The honest summary is that FMT in refractory recurrent CDI produces one of the largest treatment effects in modern gastroenterology, established on a small trial and a large uncontrolled experience, for reasons nobody has isolated.

Established The approved products show a 13.1-point effect over placebo, not a 90% cure rate, and the difference matters. Rebyota (faecal microbiota, live, as a rectal suspension) was approved by the FDA in November 2022; Vowst (faecal microbiota spores, live, as an oral capsule) in April 2023; Biomictra in Australia in November 2022. These are the first approved microbiome-derived therapeutics anywhere. Established Rebyota’s pivotal PUNCH CD3 phase III trial randomised 289 patients and treated 267 — 180 on RBX2660 and 87 on placebo. Established Its primary endpoint, absence of C. difficile diarrhoea within eight weeks, gave a model-estimated success rate of 70.6% for the product against 57.5% for placebo: a 13.1 percentage-point difference, with a posterior probability of superiority of 0.991, and over 90% of responders in both arms sustained their response to six months. Established The pivotal ECOSPOR III trial behind Vowst reported superiority to placebo for recurrence at eight weeks with comparable safety between arms; only its record and abstract were retrieved for this brief, so no effect size from it is quoted here. Established Putting “85–90% effective” next to “FDA approved” splices two incompatible evidence bases: the first is an open-label response rate in refractory patients, the second is a blinded absolute success rate against a placebo arm of patients who had just finished standard antibiotics and who did rather well on their own.

Established Safety is a donor-screening problem and the screening has failed. In 2019 one patient died and another was seriously infected by drug-resistant organisms present in donor stool, prompting an FDA warning about the potentially life-threatening consequences of transplanting material from improperly screened donors. Other documented adverse events include bacteraemia, fever and inflammatory response, gastrointestinal disturbance, and exacerbation of inflammatory bowel disease in susceptible patients. Frontier Because the active ingredient is unidentified, the product cannot be purified or characterised in the way a drug can; screening the donor is the entire safety model, which is why the defined-spore route that Vowst represents is the more interesting regulatory object even though its effect size is smaller than the case-series numbers.

Established Outside CDI, the clinical record is close to empty. FMT has been tried in ulcerative colitis (typically requiring repeated infusions), irritable bowel syndrome, fibromyalgia, multiple sclerosis and Parkinson’s disease, with varying and unconvincing results and no approvals anywhere. Established The probiotic record is worse. Established EFSA has rejected every single commercial petition for a probiotic health claim in Europe on grounds of insufficient evidence, and the European Commission banned the word “probiotic” on packaging on the grounds that it misleads consumers into inferring a health benefit; in the United States probiotics are generally recognised as safe, which speaks to safety and establishes nothing whatever about efficacy. Established Cochrane’s findings by indication: a protective effect for certain probiotics in antibiotic-associated diarrhoea in children; no clear evidence of improved remission or reduced adverse events in Crohn’s disease; low-certainty evidence of increased remission probability in ulcerative colitis; and in pregnancy, no benefit for gestational diabetes together with evidence of increased pre-eclampsia risk.

Established The measurement base is thinner than the literature built on it. The Human Microbiome Project spent $170 million of NIH Common Fund money between 2007 and 2016 and produced, at its 2012 milestone, over 5,000 samples from 242 healthy US volunteers across 15 to 18 body sites, an estimate of more than 10,000 microbial species in the human ecosystem, and over 650 papers cited more than 70,000 times by 2017. That is a valuable reference set and it is not a definition of health. Established The field’s workhorse instrument, 16S rRNA amplicon sequencing, generally cannot resolve to species level; it recovers phylogenetic relationship, not function, and a great deal of the literature makes species-level and functional claims from data that cannot carry them. Established Functional annotation is largely absent: 40–70% of the annotated genes in fully sequenced microbial genomes have no known or predicted function, and at the time of the cited analysis 85 of the then-established 118 bacterial phyla had not had a single species described. Established Relic DNA from dead cells can comprise up to 40% of sequenced soil DNA and up to 80% in some samples, though reassuringly it has minimal effect on diversity estimates. Even the vocabulary was unstable until 2020, when a consensus definition separated microbiota (the living organisms) from microbiome (the organisms plus their theatre of activity — metabolites, structural elements, conditions).

Established Causal inference is where the field is weakest, and three results establish that in different ways. First, medication confounding: a 2015 Nature paper from the MetaHIT consortium showed that a prominent reported gut-microbiome signature of type 2 diabetes was substantially a signature of metformin treatment rather than of the disease. Established That paper’s record and its 2017 corrigendum were verified for this brief but its full text was not read, so no effect size from it appears here; the qualitative finding is the point, and every microbiome–disease claim should be read against it. Established Second, effect size: a 2016 re-analysis of ten obesity datasets found limited support for the widely accepted hypothesis that microbiome changes are associated with obesity. Established Pooled random-effects analysis did detect significant relationships with Shannon diversity and OTU counts, but the actual difference between non-obese and obese individuals was 2.07%; only one of the ten datasets was adequately powered; and random-forest classifiers trained on one dataset and tested on the other nine achieved 33.01% to 64.77% accuracy, a range that spans chance. Third, contamination: a 2019 Nature study of 578 placentas across two cohorts concluded that the human placenta does not have a microbiome, with almost all prior signal traceable either to bacteria acquired during labour and delivery or to contamination of laboratory reagents with bacterial DNA. The one genuine exception was group B Streptococcus in about 5% of pre-labour samples. Established This is not a controversy any more; it is a settled negative result, and it is the cleanest demonstration available that a low-biomass microbiome literature can exist for years and be entirely artefact.

Established The causal toolkit itself is compromised. Germ-free mice, the field’s main instrument for demonstrating causation, have documented defects in immune function and energy uptake precisely because they lack a microbiome, so the comparison baseline is pathological rather than normal; and gut microbiota vary between research facilities, which is a confounder and a documented cause of non-reproducibility. Frontier For the gut–brain axis the founding result is Sudo and Chida’s 2004 demonstration of an exaggerated HPA-axis stress response in germ-free mice. Established The literature’s own summary as of 2016 was that most of the work had been done in animals, that human studies are small and cannot be generalised, and that whether microbiota changes are cause, consequence or both remains unclear — the associations are described in that literature as correlational and not causal.

Frontier The two flagship results of engineered and designed probiotics, stated as they were reported. SYNB1618, an engineered E. coli Nissle for phenylketonuria, ran a first-in-human phase 1/2a study in 70 participants — 56 healthy adult volunteers and 14 patients with PKU at blood phenylalanine of 600 µmol/l or above. Established What it reported was dose-responsive increases in strain-specific phenylalanine metabolites: trans-cinnamic acid in plasma and hippuric acid in urine. It did not report a reduction in blood phenylalanine, which is the clinical variable that PKU treatment exists to lower, and the authors correctly describe their result as proof of mechanism. Established Separately, the widely-cited human Akkermansia muciniphila trial enrolled 40 overweight and obese volunteers of whom 32 completed, dosed 10¹⁰ bacteria daily for three months, and reported insulin sensitivity up 28.62% (P = 0.002), insulinaemia down 34.08% (P = 0.006) and total cholesterol down 8.68% (P = 0.02) — for pasteurised, which is to say dead, bacteria. Body weight fell by 2.27 kg at P = 0.091 and fat mass by 1.37 kg at P = 0.092, neither significant, and the paper labels itself exploratory and proof-of-concept. Frontier A non-viable heat-killed preparation with a bioactive surface protein is a genuinely interesting pharmacology and it is not a probiotic in any mechanistic sense, which changes what technology is being demonstrated.

Frontier Can “dysbiosis” be operationalised at all? The concept anchors most clinical microbiome rhetoric and there is no consensus definition of a healthy microbiome, no reference state, and no agreed metric of deviation from one. Established Phylosymbiosis — the assumption that host phylogeny predicts microbiome composition — is the exception rather than the rule in vertebrates, with fish and bird studies showing much weaker correlations than mammals, so there is no universal principle to anchor a reference state to. Speculative A term that cannot be operationalised cannot be tested, and proposing a quantitative, prospectively-validated reference state against which deviation predicts a held-out outcome is a fundable, tractable and largely unattempted project.

Frontier Can FMT be reduced to a defined consortium? The procedure works at roughly 90% in refractory CDI and nobody can name the organisms or metabolites doing the work. Established Vowst, a defined spore preparation, is a partial answer and is the reason a manufacturable product could be approved at all. Speculative A full reduction — a minimal defined community that reproduces the CDI effect — would be this field’s equivalent of isolating penicillin from mould broth, and it has not happened. Frontier The co-cultivation approach, in which a consortium is produced by growing organisms together rather than assembling it from separately-grown monocultures, is the most interesting current method and its principal published claim is robustness rather than efficacy; the source was record-verified for this brief but not read, so nothing quantitative is attributed to it.

Frontier Is colonisation resistance a wall or a parameter? An adult gut is a dense, saturated, continuously-immigrated ecosystem in which essentially every metabolic niche is occupied, and an introduced strain must displace an incumbent to persist. Speculative No published result demonstrates dose-controlled, durable engraftment of a defined strain in an unmanipulated adult human gut. Speculative The most promising escape is niche creation: pair an engineered strain with an exclusive catabolic pathway for a substrate no resident and no environmental organism can use, and colonisation becomes deterministic and dose-controllable rather than probabilistic. Handwave That converts an ecology problem into a carbohydrate-chemistry problem — find a polysaccharide safe to administer chronically that nothing else in the gut can eat — which is a real research programme and, as far as this brief could establish, an under-attempted one.

Frontier Can causality be recovered by instrument rather than by intervention? Mendelian randomisation is being imported into microbiome epidemiology at speed, using host genetic variants as instruments for microbiome composition. Speculative The method rests on weak instruments — host genotype explains very little microbiome variance — and it is being applied considerably faster than it is being validated. Frontier The honest reading is that microbiome MR is currently generating hypotheses that look like conclusions.

Frontier Is precision decolonisation a better target than precision addition? Narrow-spectrum antibacterials drawn from the microbiome itself — microcins that selectively inhibit enteric organisms and reduce carriage of carbapenem-resistant Klebsiella in animals — attack the community by subtraction rather than addition. Speculative Subtraction has the advantage of not requiring engraftment, which is the step nobody has solved. Frontier The source here was record-verified but not read; the direction is what matters and no number is attributed.

Frontier Which claimed microbiomes are real? The placental one is settled as artefact. Frontier The blood microbiome remains genuinely open and is exactly the same kind of claim, at the same biomass, with the same contamination exposure. Speculative A reasonable prior, given that the placenta went the way it did at n = 578, is that low-biomass tissue microbiomes should be treated as contamination until an adequately-controlled study of comparable size says otherwise.

Frontier Are enterotypes a thing? The proposal that human gut communities fall into a small number of discrete compositional classes has been contested in the primary literature for over a decade — a 2014 Cell Host & Microbe paper is titled “Rethinking ‘Enterotypes’” — while enterotype language remains common in secondary coverage and in consumer microbiome testing. Frontier That paper’s abstract was not retrieved for this brief, so it is cited for the existence of the dispute and no conclusion is attributed to it.

Speculative The binding constraint for engineering is deterministic colonisation, and the binding constraint for the science is a testable definition of the target state. Everything else on this list is downstream of one or the other.

Frontier Colonisation resistance. Nothing in the published record shows that a chosen strain can be installed in an unmanipulated adult gut at a chosen abundance and held there. Established Every intervention that reliably changes a human gut community at scale — FMT, antibiotics, sustained dietary change — works by bulk replacement or bulk perturbation, not by precision addition. Speculative Until engraftment is dose-controllable, an engineered strain is a transient dosage form rather than a resident device, and the whole vocabulary of “installing” a microbiome is aspirational.

Speculative The containment paradox. A live engineered organism must colonise to be effective and must not persist to be safe. Frontier Biocontainment resolves the tension by engineering in a kill switch, which reduces colonisation, which reduces efficacy. Speculative The reductio is worth stating plainly: if a transiently-dosed engineered bacterium is only an enzyme-delivery vehicle, it is competing against enzyme replacement therapy, which is a mature, better-characterised and more easily regulated modality. Handwave The engineered-probiotic field has not answered this and mostly does not pose it.

Established Measurement resolution. Species- and function-level claims are being made from 16S data that cannot resolve species and does not report function, against reference genomes in which 40–70% of genes have no predicted function. Frontier The fix — shotgun or long-read strain-resolved sequencing as the default for any causal claim — is a cost and convention problem rather than a scientific one.

Established Confounder discipline. Medication, diet and animal-facility covariates are the difference between a disease signature and a drug signature, and the metformin case shows that the difference can be most of the effect. Frontier No journal in this field requires the covariate handling that the 2015 result implies is necessary.

Established No reference state. There is no consensus definition of a healthy microbiome and no validated metric of deviation from one, which means the field has no target to engineer toward outside the single case where the target is the absence of one named pathogen. Speculative A discipline that can measure a system in enormous detail and cannot say what a good value looks like is not yet an engineering discipline; it is a survey.

Established Manufacturing and characterisation. A product whose active ingredient is unidentified cannot be assayed for potency in the ordinary sense; donor screening substitutes for characterisation, and in 2019 that substitution killed a patient. Frontier The defined-spore route solves the assay problem and costs effect size, which is the trade the whole field will have to make repeatedly: everything that makes a microbiome product manufacturable makes it less like the uncharacterised thing that worked.

Frontier This subject waits on results it does not generate, and the largest of them belongs to genetic engineering. Every engineered live biotherapeutic assumes a construction step — a stable, non-mobilisable circuit in a gut-competent chassis, with a containment mechanism that does not cost fitness — that is Genetic Engineering’s problem rather than this one’s. Established The typed dependency in section 13 records exactly that and nothing else.

Established Strain-resolved sequencing at population scale. The causal claims this field wants to make are about strains and gene content, and the instrument used for two decades reports neither. Frontier Long-read and deep shotgun methods exist; what does not exist is a cohort large enough, longitudinal enough and interventional enough to use them on.

Frontier Gnotobiotic models with a defensible baseline. The germ-free mouse is the field’s causal instrument and its baseline is pathological by construction. Speculative What is needed is a defined-community animal whose baseline is a deliberately specified normal rather than an absence — which is itself a designed-consortium problem, so the causal tool waits on the engineering capability it is supposed to validate.

Frontier Metabolomics that closes the loop from community to host. The organisms are not the mechanism; the metabolites are. Speculative Until routine, quantitative, host-side metabolite measurement accompanies every compositional survey, compositional differences will keep being reported as if they were physiological ones. Established The SYNB1618 trial is the instructive case in both directions: metabolite measurement is exactly what let it demonstrate that its bacterium was alive and working, and exactly what let a metabolite stand in for a clinical endpoint.

Speculative Carbohydrate chemistry, of all things. The one route to deterministic colonisation that does not require solving gut ecology requires a substrate that is safe to eat every day and that nothing else in the gut or the environment can metabolise. Handwave That is a synthesis and safety-pharmacology programme rather than a microbiology one, and it is not obvious that anybody in this field has the right chemists.

Speculative The workback plan, in dependency order, from what exists to a designed community that produces a specified physiological outcome in an arbitrary human.

Frontier 1. Reduce FMT to a defined consortium. Identify the minimal defined community that reproduces the CDI cure rate. The spore preparation behind Vowst is a partial answer; a full reduction has not been done, and it is a tractable scientific unknown rather than a hard one.

Speculative 2. Operationalise dysbiosis. Publish a quantitative reference-state definition, pre-register it, and show that deviation from it predicts a clinical outcome in a held-out cohort. Until this exists, most of the field’s claims are not merely unproven but untestable. This link binds the science.

Established 3. Retire 16S for species- and function-level claims. Require strain-resolved sequencing wherever a causal claim is made. Institutional and cost, not scientific.

Established 4. Mandate confounder handling. Medication, diet and facility covariates at the standard the metformin result implies, in every association study. Institutional.

Speculative 5. Run an adversarial replication programme. A pre-registered replication of the twenty most-cited gut-microbiome–disease associations, in independent cohorts, with covariates handled properly. The cost is modest, the expected result is that a large fraction fail, and nobody will fund it because no participant benefits from the answer. This is the cheapest high-value link in the chain and it is blocked institutionally rather than scientifically.

Speculative 6. Demonstrate deterministic colonisation. Dose-controlled, durable engraftment of a defined strain in an unmanipulated adult human gut — most plausibly via an orthogonal nutrient niche. This has never been demonstrated. This link binds the engineering.

Speculative 7. Containment without a fitness cost. A kill switch or auxotrophy that does not reduce engraftment. Scientific unknown, and currently the reason engineered strains are dosed rather than installed.

Frontier 8. Win once outside displacement. One randomised, adequately-powered, replicated positive trial in an indication that is not single-pathogen displacement. None exists. This is the honest scoreboard for the whole field.

Speculative 9. Try whole-community selection on a therapeutic phenotype. Serially propagate hundreds of replicate communities, select on a system-level readout rather than a composition, and see whether the therapeutic property is heritable across transfers. Established The method was demonstrated on soil ecosystems in 2000 and the heritability objection was answered then; what is missing is a cheap enough assay and anyone willing to run it on a clinical phenotype.

Established Manufacturing strict anaerobes to pharmaceutical standard is the unglamorous capability the approvals actually rest on. The organisms that matter in a gut community mostly die on contact with oxygen, which makes cultivation, formulation, fill-finish, stability testing and cold chain harder than for any conventional biologic. Established The spore route sidesteps most of it, which is a large part of why an oral spore capsule became a licensed product.

Established Donor screening is an industrial process and not a laboratory one. A stool-bank supply chain has to screen for pathogens that have not yet been named as risks; the 2019 deaths came from drug-resistant organisms that were not on the screening panel at the time. Frontier Every expansion of the panel raises the donor rejection rate, and a screening regime strict enough to be safe may be strict enough to make the supply uneconomic — which is an argument for defined products and against donor material, on engineering grounds alone.

Speculative An engineered resident strain needs four things this field cannot yet specify. A chassis that colonises the adult gut; a circuit stable over thousands of generations under real selection; a containment mechanism that does not cost fitness; and a dosing regime that reaches a target abundance rather than a target dose. Handwave No published programme has all four, and the trials that exist have essentially none of them — they dose a strain, measure something while it is passing through, and stop.

Speculative The orthogonal-nutrient route implies a small chemical-manufacturing programme — a food-grade substrate produced at scale, chronically administered, and demonstrably inedible to everything else in the gut and in the environment. Handwave That is a tractable chemistry problem attached to an untested ecological premise, which is roughly the best position an exotic proposal in this subject can occupy. Speculative It also has a clean failure mode, which is unusual for an exotic proposal: if a resident organism turns out to be able to eat the substrate after all, the experiment says so within weeks.

Frontier Agricultural microbiome engineering is further along commercially and has the same endpoint problem. An engineered Klebsiella variicola strain that colonises maize roots and retains nitrogenase activity in nitrogen-replete soil — wild diazotrophs shut fixation off when fertiliser is present — reported field gains across the US Corn Belt of 0.35 t/ha (5.2 bushels/acre) with reduced within-field yield variance. Established The endpoint there is yield in already-fertilised fields, not a measured quantity of nitrogen substituted for fertiliser; a gain on top of full fertiliser is a real agronomic result and it is not the Haber–Bosch displacement claim it is often reported as. Established The naturally-occurring version — a Mexican maize landrace whose aerial-root mucilage hosts a diazotrophic microbiota — establishes that a cereal can obtain fixed nitrogen from bacteria without nodulation, and is discussed further in Future Agriculture.

Frontier Whole-community artificial selection is the environmental route and it is older than most people assume. A 2000 PNAS experiment subjected entire soil microbial ecosystems to artificial selection and found that laboratory-initiated ecosystems exhibit phenotypic variation with a heritable proportion, despite comprising thousands of species and millions of individuals. Speculative Community-level heritability is therefore not a speculation; it was demonstrated twenty-six years ago, and almost nothing in therapeutic microbiome engineering uses it. Speculative Selecting communities on a system-level phenotype, rather than designing them organism by organism, is the most under-explored idea in this brief.

Frontier Closed ecological systems are the same problem at a different scale. Synthetic Ecosystems asks whether a community can be assembled, kept stable and kept invasible-resistant for years; so does this subject, with a human wrapped around it. Speculative A serious cross-reading of closed-ecosystem ecology and gut microbial ecology does not exist and should. Frontier Synthetic Biology and Designer Organisms supply the circuits; Precision Medicine supplies the stratification this field would need to find the subpopulations in which its effects might actually be large.

Established Two regulators looked at the same evidence base and reached opposite operational conclusions, which is itself the most informative institutional fact in this subject. The FDA approved two faecal-microbiota products for recurrent CDI in 2022 and 2023 on randomised placebo-controlled trials. Established EFSA has rejected every commercial petition for a probiotic health claim ever submitted to it, and the European Commission banned the word from packaging. Established These are not in conflict: one is a licensed medicine with a trial behind it, the other is a food-supplement claim with none. Frontier The public reads them as the same category, and the industry has not worked hard to correct that.

Frontier The category problem for live organisms is half-solved. Rebyota and Vowst established that a living, incompletely-characterised biological product can be licensed as a drug, which is a genuine regulatory achievement. Speculative No comparable framework exists for a genetically engineered living organism intended to persist in a patient, where the questions are shedding, horizontal gene transfer, environmental release and reversibility rather than potency and purity. Speculative The nearest analogues are in agricultural biosafety, not in medicines regulation, and the two communities barely talk.

Speculative Nobody owns replication. The single highest-value experiment available in this field is an adversarial pre-registered replication of its most-cited associations, and there is no funder whose mandate covers it: agencies fund discovery, companies fund products, and journals reward novelty. Handwave The absence is structural rather than accidental, and it is the same institutional pathology that leaves benchmark work undone in every field where the incentive to confirm exceeds the incentive to check.

Frontier Stool banking is an infrastructure with no settled legal status. Donor material is regulated variously as a drug, a tissue, or nothing at all depending on jurisdiction, and the do-it-yourself practice that grew up around the 85–90% figure sits entirely outside all of it. Speculative Approving manufactured products has a second-order effect that is rarely discussed: it gives regulators a licensed alternative, and therefore the standing to restrict the unlicensed practice that got there first and that remains cheaper.

Speculative The publication conventions are the cheapest available intervention and nobody has made it. Requiring strain-resolved data for a species-level claim, and requiring medication and diet covariates for an association claim, would cost a journal nothing and would retire a large share of the literature at the point of submission rather than a decade later. Handwave That no editorial board has done it, a decade after the metformin result, is the clearest evidence that the field’s problem is incentives rather than methods.

Established The consent objects here are unusual and the frameworks were built for other things. A stool donation is simultaneously a tissue donation, a genetic sample, a microbial community with its own future, and a manufacturing input; donors are typically compensated modestly and screened intensively, and no jurisdiction has settled whether they retain any interest in what is made from the material.

Frontier Overclaiming is the field’s main ethical failure, not its exotic risks. Consumer microbiome tests are sold on enterotype language contested in the primary literature since 2014 and on disease associations whose measured magnitude is around two percent of a diversity index. Established Probiotic marketing survives an outright regulatory finding of no established benefit in Europe. Speculative The harm is not usually direct; it is the substitution of a purchased intervention for an effective one, and the erosion of the credibility that the one real result — CDI — needs in order to be believed.

Established The 2019 death is the concrete safety case and it should not be softened. A patient died from drug-resistant organisms in screened donor stool. Frontier “It is only stool” is a category error: it is an uncharacterised living transplant, and it carries transplant-grade risk.

Speculative Engineered organisms intended to persist raise a consent question that has no medical precedent. A resident engineered strain can be shed, can transfer genes, and cannot be recalled; the exposed population includes people who never consented to anything. Handwave Reversibility — a demonstrated, reliable way to remove an installed community — is a prerequisite that nobody has built and that few programmes list. Speculative The containment paradox has an ethical face as well as an engineering one: the version of the technology that is easiest to justify releasing is the version least likely to work, and a field under commercial pressure will be tempted to resolve that in the direction that produces a product.

Frontier There is also a quieter equity question. Donor stool comes disproportionately from young, healthy, screenable people, and defined products will be priced as medicines while the do-it-yourself practice they replace was free. Speculative An intervention that was, uniquely in modern medicine, available without industry is being converted into one that is not, and the conversion is well justified on safety grounds and is still a real loss.

Frontier The most valuable thing a microbiome community could plausibly deliver at civilizational scale is colonisation resistance as public-health infrastructure. The one demonstrated mechanism in this field — competitive exclusion of a pathogen by a resident community — is precisely the mechanism antibiotic use destroys, and it addresses the resistance problem from the ecological side rather than the chemical one. Speculative A reliable way to restore or fortify colonisation resistance after antibiotic exposure would be worth more than every metabolic claim in the field combined, and it is the one extrapolation the CDI result actually licenses.

Speculative Closed habitats make this a load-bearing engineering discipline rather than an optional therapeutic one. In a sealed environment the human microbiome, the crop microbiome, the waste-processing community and the surface flora are one coupled system with no external reservoir to buffer it, and every result in Synthetic Ecosystems about drift and guild loss applies. Handwave Nobody has specified what microbial community a multi-year closed habitat should be launched with, and on the evidence in this brief nobody can.

Frontier The agricultural version is where the near-term tonnage is. Root-associated communities that fix nitrogen or suppress pathogens act on gigatonnes of soil, and the endpoints are measurable in yield rather than in diversity indices — which is why the agricultural literature, for all its own overclaiming, is currently better disciplined than the clinical one. Speculative A field trial that reports bushels has an argument-ending property that no diversity index has, and the clinical wing of this subject would be healthier if it could find its own equivalent.

Speculative The largest civilizational risk in this subject is not a runaway organism; it is credibility. A field with one excellent result and a great many exaggerated ones is one bad decade away from having the excellent result disbelieved along with the rest. Established That has already begun in Europe, where a regulator’s blanket rejection of commercial probiotic claims and a ban on the word itself are the institutional record of a scientific community having spent its credit. Handwave Protecting the C. difficile result from the reputation of everything sold alongside it is, unglamorously, the most valuable thing this field could do in the next ten years.

These horizons track two clocks that run at very different speeds: the clinical record, which has produced one indication in twenty years, and the measurement base, which improves steadily and keeps deflating the claims made on the previous generation of data.

• 10 yr: Frontier Defined-consortium products displace donor material for CDI, because manufacturing and screening economics favour them and the precedent already exists. Frontier At least one further indication reaches a well-powered randomised trial; the honest prior, given that none has succeeded in twenty years, is that it fails. Speculative Strain-resolved sequencing becomes the default for causal claims and a visible fraction of the 2010s association literature quietly stops being cited. Speculative An engineered live biotherapeutic reports a clinical endpoint rather than a strain-metabolite endpoint, or the modality loses its remaining credibility.
• 25 yr: Speculative Deterministic colonisation is demonstrated or abandoned, and that single result decides whether this subject is an engineering discipline or a branch of infectious-disease medicine. Speculative A prospectively-validated reference-state definition exists, or “dysbiosis” is retired as a technical term the way “autointoxication” was. Speculative Community-level selection — breeding whole consortia on a system phenotype rather than designing them organism by organism — is either the standard method for producing therapeutic communities or has been shown not to scale beyond soil microcosms.
• 50 yr: Handwave If colonisation is solved, a specified community becomes a durable installed device and the interesting questions become maintenance, interoperability and removal — the vocabulary of implants rather than of drugs. Handwave If it is not, this subject in 2076 looks like this subject now: one excellent treatment for one infection, a very large descriptive literature, and a consumer industry regulators periodically discipline.
• 100 / 250+ yr: Handwave Designed microbial communities as standard equipment for closed habitats and engineered biospheres is the coherent long-horizon endpoint, and it requires community-level stability guarantees that no branch of ecology can currently provide for any system, anywhere. Handwave The deepest open question on this horizon is whether a microbial community is the kind of object that can be specified at all — whether there exists a description short enough to be a design and complete enough to determine an outcome — and nothing in the present literature answers it either way.

• Depends on The one typed dependency here runs to Genetic Engineering, and it covers everything on the engineered side of this subject and nothing on the side that works. The FMT result required no editing at all; the engineered live biotherapeutic requires a stable non-mobilisable circuit in a gut-competent chassis, a containment mechanism that does not cost fitness, and an editing toolkit that this brief assumes and does not produce. That is the only typed dependency here, and stating it narrowly is the point: the field’s single approved technology is upstream of nothing in genetics, while its entire speculative wing is downstream of results genetics has not yet delivered. Everything else this subject waits on is either an instrument, a convention or a decision about money rather than a frontier result — strain-resolved sequencing at cohort scale, mandatory medication and diet covariates, and a funder willing to pay for replication — and none of those belongs on a map of hypotheses. It is worth being explicit that the binding engineering constraint, deterministic colonisation of an unmanipulated adult gut, is a scientific unknown that no other brief on this map is working on either. This subject has to produce it itself, or wait.
• Enables Synthetic Ecosystems in the specific sense that a closed habitat needs a specified microbial community and currently has no way to obtain one; supplies Future Agriculture with the root-associated engineering that the nitrogen programmes assume; and gives Precision Medicine a candidate stratification variable that, on present evidence, it should treat with considerable suspicion. It also supplies Biodiversity Restoration with the only demonstrated method of changing a whole community deliberately — bulk replacement — and with the reason to distrust the precision version of the same idea.
• Adjacent to Synthetic Biology and Designer Organisms, which own the chassis and the circuit; to Biological Energy Systems, which shares the problem of engineering a mixed microbial community toward a specified output and has the same habit of reporting the flattering metric; to Longevity Therapies, which is the largest consumer of this field’s weakest associations; and to Xenobiology, which is where the orthogonal-nutrient idea would have to come from if it is ever built, since a substrate no resident organism can metabolise is precisely a xenobiological object. Two further adjacencies are worth naming because they are usually missed. Antimicrobial stewardship is adjacent in the strong sense that colonisation resistance is the ecological half of the resistance problem and this is the only field working on it. And consumer genomics is adjacent in the weak but commercially decisive sense that it supplied the business model — a mail-in sample, a report, and a claim the underlying data cannot support — which microbiome testing then copied wholesale.

Established “There are ten bacterial cells in your body for every human cell.” Wrong by a factor of ten, and corrected in 2016: roughly 38 trillion bacterial cells against roughly 30 trillion human cells, approximately one to one, and about 1:350 by mass — around 0.2 kg in a 70 kg human. Established The 10:1 figure traced back to a 1972 estimate never intended as a rigorous calculation, and the ratio near 100:1 that is real applies to genes, not cells: roughly two million bacterial protein-coding genes against about twenty thousand human ones. Established A factor-of-ten error circulated in the peer-reviewed literature and popular science for four decades, which is the cleanest available illustration of how little friction a satisfying number meets in this field.

Established “FMT is about 90% effective, and the FDA approved it on that basis.” Two evidence bases spliced together. Established The 85–90% figures are open-label case-series response rates in refractory recurrent CDI; the pivotal randomised trial behind Rebyota reported 70.6% success against 57.5% on placebo — 13.1 percentage points, in 267 treated patients; and the founding 2013 randomised trial had sixteen patients in the FMT arm. Established FMT works. It does not work by as much as the round numbers imply, and the placebo arms in this indication are not small.

Established “Every tissue has a microbiome” — and the placenta is the counter-example that settles it. A study of 578 placentas in Nature concluded flatly that the human placenta does not have a microbiome: almost every signal traced either to bacteria acquired during labour and delivery or to reagent contamination, the single genuine exception being group B Streptococcus in about 5% of pre-labour samples. Frontier That is not a live controversy but a resolved negative, and the blood microbiome should be read in its light: when the contamination controls much of this literature never ran were finally run at scale, an entire claimed organ microbiome disappeared.

Established “The gut microbiome differs in obesity.” On re-analysis of ten datasets the difference between obese and non-obese individuals in diversity indices was 2.07%; only one of the ten was adequately powered; and classifiers trained on one cohort predicted obesity in the others at 33.01% to 64.77% accuracy, a range that includes chance. Established The association is not zero; it is small, inconsistent between cohorts, and swamped by inter-individual variation — which is a description of a finding that will not support a therapy.

Established “The gut microbiome differs in type 2 diabetes.” Much of the reported signature was a signature of metformin rather than of diabetes, per the 2015 MetaHIT re-analysis in Nature — which means any disease association drawn from a cohort where the cases are medicated and the controls are not is uninterpretable until the drugs are modelled.

Frontier “Gut bacteria cause depression, obesity or autism.” The gut–brain literature’s own summary is that most work has been done in animals, that human studies are small and cannot be generalised, and that whether microbiota changes are cause, consequence or both remains unclear; the associations are described there as correlational and not causal. Established The germ-free mouse that anchors the causal claims has immune and energy-uptake defects because it is germ-free, and facility-to-facility variation is a documented cause of non-reproducibility. Speculative None of this proves the causal story false — effects could be large in subpopulations and averaged away, or strain-level and invisible to 16S — and what would settle it is strain-resolved longitudinal cohorts with interventional arms, which are not funded at anything like the scale of the observational literature.

Established “Engineered probiotics have been shown to work in humans.” The flagship first-in-human trial — SYNB1618 in phenylketonuria, 70 participants — reported dose-responsive increases in the engineered strain’s own metabolic by-products, trans-cinnamic acid in plasma and hippurate in urine, which the authors correctly call proof of mechanism. Established It did not report a fall in blood phenylalanine, so what stands is a surrogate for a surrogate: evidence that the bacterium was alive and metabolically active in the gut, not evidence that the patient’s disease burden moved. Frontier When someone says engineered probiotics have worked in humans, this is the trial they mean, and this is what it measured.

Established “Akkermansia supplementation causes weight loss.” In the Nature Medicine proof-of-concept trial the weight change was 2.27 kg at P = 0.091 — not significant — in 32 completers, and the metabolic effects that were significant were produced by pasteurised, non-viable bacteria — an interesting pharmacological object, and not a probiotic, since it works through a surface protein rather than through anything that colonises.

Established “Probiotics are proven to work” and “16S sequencing tells you which species are present.” EFSA has rejected every commercial probiotic health claim submitted to it and the European Commission banned the term from packaging as misleading; US GRAS status is a safety designation and nothing more. And 16S generally cannot resolve to species level, reports phylogeny rather than function, and is read against genomes in which 40–70% of genes have no predicted function.

Speculative The strongest argument that most of this will not replicate, stated in full. The weaknesses are cumulative and each is documented above: 16S cannot resolve species, yet species claims are routine; most genes have no predicted function, yet functional inference is routine; low-biomass sites are contamination-dominated, and one claimed organ microbiome has already dissolved; germ-free animals are a pathological baseline; drug confounding can generate an entire spurious disease signature; and the European regulator has rejected every commercial claim put to it. Speculative For the mainstream causal story to hold anyway, effect sizes would have to be large enough to survive all of that, demonstrated in humans, in randomised designs, with pre-registration — a combination that exists for exactly one indication. Established And CDI is the case where the mechanism is simple, ecological, and requires none of the inferential machinery above: this field’s one unambiguous success is the case where you do not need to understand the microbiome at all.

Speculative The minority reading that best fits the clinical record: FMT works by displacement, not by restoring balance. On this account the mechanism in CDI is competitive exclusion of a single pathogen plus bile-acid metabolism — no ecological subtlety, no restoration of a reference state — which predicts that the procedure will work well for single-pathogen displacement and poorly for everything else. Established That is exactly the clinical record: approvals for CDI only, mixed or unconvincing results in ulcerative colitis, IBS, multiple sclerosis and Parkinson’s, and no non-displacement win after two decades and $170 million of Human Microbiome Project funding alone. Handwave The furthest defensible version of the sceptical case is that precision microbiome engineering is the wrong abstraction for the substrate — that adding one designed organism to a continuously-immigrated 10¹⁴-organism ecosystem is like adding one engineered plant to a rainforest and expecting a specified outcome. Speculative The honest counter, and the reason this brief does not end on the sceptical note, is that nobody has yet tried the one manoeuvre that would make precision addition deterministic, which is to build the niche first.