What "good" and "bad" gut bacteria actually means

What it is

The large intestine carries a dense community of bacteria, and studies repeatedly find that some species are more abundant in people who are well and others more abundant in people who are ill. Species commonly labelled beneficial include Faecalibacterium prausnitzii, Akkermansia muciniphila and Bifidobacterium; species commonly labelled harmful include Clostridioides difficile, some Escherichia coli strains and Fusobacterium nucleatum. The labels come mostly from observational comparisons of abundance between groups, not from experiments that added or removed the organism in people. Several of the organisms called harmful are also carried by people with no symptoms at all.

What the research says

The honest summary is that "good" and "bad" are shorthand for statistical associations plus a small number of mechanistic experiments, and that the same species can sit on either side of the line depending on strain, site and host. Faecalibacterium prausnitzii is consistently lower in inflammatory bowel disease, but that is an association. A three-month placebo-controlled pilot trial of Akkermansia muciniphila improved insulin sensitivity in 32 overweight adults while leaving body weight essentially unchanged. On the other side, Escherichia coli strains carrying the pks island make a genotoxin whose mutational fingerprint appears in human colorectal cancers, yet the century-old probiotic E. coli Nissle 1917 carries that same island, and probiotic Lactobacillus strains have been sequenced out of the bloodstream of intensive-care patients who were given them.

Evidence grade: Limited evidence.

What goes wrong

Escherichia coli carrying the pks island produces colibactin, and repeated exposure of human intestinal organoids to pks+ E. coli left a mutational signature that is also present in human colorectal cancer genomes. (Source 1)

• Lab study in cells, Moderate certainty.
• Size: organoid experiments plus a subset of 5,876 human cancer genomes from two cohorts.
• Who: human intestinal organoids; human cancer genome cohorts.
• How long: repeated luminal injection over five months.
• Result: a distinct mutational signature absent from organoids injected with isogenic pks-mutant bacteria, detected predominantly in colorectal cancer genomes.
• Funding: not stated in the abstract.

The same mutational signature was detected in a subset of 5,876 human cancer genomes from two independent cohorts, predominantly in colorectal cancer.

The probiotic strain Escherichia coli Nissle 1917 carries the same colibactin-producing pks island, which illustrates how thin the line between a "good" and a "bad" strain can be. (Source 2)

• Lab study in cells, Low certainty.
• Size: bacterial genetics plus a mouse salmonellosis model.
• Who: E. coli Nissle 1917 and mutants; mice.
• How long: not applicable.
• Result: a single amino acid substitution in ClbP inactivated genotoxic activity while keeping the antagonistic activity.
• Funding: Agence Nationale de la Recherche grants (independent of a probiotic manufacturer)

This co-evolution observed in EcN illustrates the fine margin between pathogenicity and probiotic activity, and the need to address both the effectiveness and safety of probiotics.

Fusobacterium nucleatum was more abundant in colorectal tumour tissue than in paired normal tissue taken from the same patients. (Source 3)

• Survey study, Low certainty.
• Size: 112 colorectal cancer tumours, of which 108 had paired normal tissue sequenced.
• Who: a cohort of individuals with colorectal cancer; tumour tissue compared with paired normal tissue from the same person.
• How long: cross-sectional, tissue sampled at one time point.
• Result: F. nucleatum present in 86.6% of tumours (97/112), relative abundance 0.01% to 40% of total microbial reads; significant enrichment in tumour versus paired normal tissue, p < 0.0001.
• Funding: not stated in the sections of the article we could reach.

Analysis of the 108 tumor/normal pairs revealed significant enrichment of F. nucleatum in the tumor microbiome compared to paired normal samples (p < 0.0001)

A review of the Lactobacillus bacteraemia literature reports that these bloodstream infections are uncommon, and that they are seen more often in people taking probiotics than in people who are not; the review puts no numbers behind that comparison. (Source 4)

• Expert review, not systematic, Low certainty.
• Size: published reports of Lactobacillus spp. bacteraemia identified by the authors' literature search; no pooled denominator given in the abstract.
• Who: patients with Lactobacillus bloodstream infection, including intensive care patients and people with central venous catheters.
• How long: literature reviewed up to 2023.
• Result: prevalence described as infrequent but more common in patients taking probiotics; three probiotic species (Lacticaseibacillus rhamnosus GG, Lactiplantibacillus plantarum and Lacticaseibacillus paracasei) were directly linked with patients' blood isolates by molecular identification assays. No risk ratio or absolute rate is given in the abstract.
• Funding: partly industry-funded; the paper's funding statement reads "Partial funding was provided by Bio-K+, a Kerry Company." Bio-K+ is a probiotic manufacturer, and the paper also discloses that several authors sit on its advisory board or consult for it.

Limit of this finding: The review says Lactobacillus bloodstream infection is more common in people taking probiotics, but it gives no numerator and no denominator: it never says how many probiotic users out of how many developed one. So this cannot be read as a rate, a percentage, or a risk of taking probiotics, only as a direction the published reports point in. It is also a narrative review of case reports and series, not a study that followed probiotic users and non-users side by side, and it was part-funded by a probiotic manufacturer.

The prevalence of Lactobacillus bacteremia is infrequent but is more common in patients taking probiotics compared to those not taking probiotics.

The same review reports that Lactobacillus bloodstream infection is infrequent but carries a higher risk of death, and it lists severe underlying disease, a suppressed immune system, admission to intensive care and use of a central venous catheter as the things that make someone more likely to develop it. (Source 4)

• Expert review, not systematic, Low certainty.
• Size: published reports of Lactobacillus spp. bacteraemia identified by the authors' literature search.
• Who: hospitalised patients with Lactobacillus bacteraemia.
• How long: literature reviewed up to 2023.
• Result: the review states a higher risk of mortality without giving a pooled figure in the abstract; risk factors listed are severe underlying disease, immune suppression, intensive care admission and central venous catheters.
• Funding: partly industry-funded; the paper's funding statement reads "Partial funding was provided by Bio-K+, a Kerry Company." Bio-K+ is a probiotic manufacturer, and the paper also discloses that several authors sit on its advisory board or consult for it.

Limit of this finding: The sentence quoted here is loosely built in the source itself, and is reproduced as the review prints it. Read literally it suggests that severe underlying disease, immune suppression, intensive care and central venous catheters are risks caused by the infection. The review means the opposite: these are risk factors for getting it in the first place. The review also gives no figure for the higher risk of death and names no comparison group, so the size of that risk cannot be taken from this sentence. It was part-funded by a probiotic manufacturer.

Lactobacillus bacteremia is infrequent but has a higher risk of mortality and risk factors, including severe underlying diseases, immune system suppression, admission to intensive care units, and use of central venous catheters.

What the evidence supports

Faecalibacterium prausnitzii is found at lower abundance in people with inflammatory bowel disease than in healthy controls. (Source 5)

• Meta-analysis, Low certainty.
• Size: 11 studies, 1,180 subjects, from 48 abstracts reviewed.
• Who: patients with Crohn's disease or ulcerative colitis versus healthy controls.
• How long: cross-sectional comparisons.
• Result: F. prausnitzii 6.7888 +/- 1.8875 log10 CFU/g feces in IBD versus 7.5791 +/- 1.5812 in healthy controls, P < 0.0001; standardised mean difference -0.94 (95% CI -1.07 to -0.80); Crohn's disease -1.13 (95% CI -1.32 to -0.94), ulcerative colitis -0.78 (95% CI -0.97 to -0.60)
• Funding: not stated.

The abundance of F. prausnitzii was decreased in IBD patients compared with healthy controls.

The observational literature links these labels to disease states rather than to demonstrated cause. (Source 5)

• Meta-analysis, Very low certainty.
• Size: 11 included studies, 1,180 subjects.
• Who: IBD patients and controls.
• How long: not applicable.
• Result: no experimental manipulation of the organism was performed in the included studies.
• Funding: not stated.

Numerous observational studies have suspected dysbiosis, an imbalance between protective and harmful bacteria to be relevant to the etiology and pathogenesis of IBD.

In a three-month placebo-controlled pilot trial, pasteurised Akkermansia muciniphila improved insulin sensitivity in overweight, insulin-resistant volunteers. (Source 6)

• Randomized trial, Low certainty.
• Size: 40 enrolled, 32 completed.
• Who: overweight/obese insulin-resistant adult volunteers at a single centre.
• How long: three months.
• Result: insulin sensitivity +28.62 +/- 7.02%, P = 0.002; insulinemia -34.08 +/- 7.12%, P = 0.006; plasma total cholesterol -8.68 +/- 2.38%, P = 0.02; dose 10^10 bacteria daily for three months.
• Funding: authors include inventors on patent applications for A. muciniphila and cofounders of A-Mansia Biotech S.A.

Compared to placebo, pasteurized A. muciniphila improved insulin sensitivity (+28.62 +/- 7.02%, P = 0.002), and reduced insulinemia (-34.08 +/- 7.12%, P = 0.006) and plasma total cholesterol (-8.68 +/- 2.38%, P = 0.02).

What the evidence does not support

In that same trial the supposedly beneficial organism did not significantly change body weight or fat mass. (Source 6)

• Randomized trial, Low certainty.
• Size: 32 completers.
• Who: overweight/obese insulin-resistant adults.
• How long: three months.
• Result: body weight -2.27 ± 0.92 kg, P = 0.091; fat mass -1.37 ± 0.82 kg, P = 0.092; hip circumference -2.63 ± 1.14 cm, P = 0.091 (all versus placebo or baseline, none reaching P < 0.05)
• Funding: authors include patent holders and company cofounders.

Pasteurized A. muciniphila supplementation slightly decreased body weight (-2.27 +/- 0.92 kg, P = 0.091) compared to the placebo group, and fat mass (-1.37 +/- 0.82 kg, P = 0.092) and hip circumference (-2.63 +/- 1.14 cm, P = 0.091) compared to baseline.

Raising the abundance of organisms usually called beneficial did not produce better metabolic outcomes in a randomised prebiotic sub-study; several inflammatory and lipid markers went the wrong way. (Source 7)

• Randomized trial, Low certainty.
• Size: 21 supplemented with 20 g/d inulin-type fructans versus 22 controls, within a trial of 100 randomised (82 completers)
• Who: adults aged 18-65 with BMI 27-40 kg/m2 on an ad libitum plant-based diet.
• How long: 10 weeks.
• Result: Bifidobacterium and Faecalibacterium selectively increased (q < 0.05) while Shannon diversity fell; LDL:HDL ratio, IL-10, MCP-1 and TNF-alpha significantly elevated in the prebiotic subgroup.
• Funding: EU Horizon 2020 grant 818318 and King Saud bin Abdulaziz University for Health Sciences.

Addition of ITF to the plant-based diet reduced microbial diversity (Shannon index) and selectively increased Bifidobacterium and Faecalibacterium (q < 0.05). The change in the latter was significantly associated with higher values of insulin and HOMA-IR and lower HDL cholesterol. In addition, the LDL:HDL ratio, IL-10, MCP-1 and TNFa levels were significantly elevated in the ITF-subgroup.

There is no agreed scientific threshold that defines a healthy microbiome, so calling a person's bacteria "good" or "bad" on a test report is not currently supportable. (Source 8)

• Official position, Certainty not rated.
• Size: expert workshop of 21 experts from 8 countries.
• Who: not applicable.
• How long: not applicable.
• Result: no thresholds exist for calling a microbiome healthy, diverse or optimal.
• Funding: part of the EU-funded International Human Microbiome Coordination and Support Action consortium.

without a scientific consensus on thresholds, it is impossible to say whether a microbiome is healthy

Where the evidence is mixed

Clostridioides difficile, the textbook "bad" organism, is carried without symptoms by a substantial and highly variable share of people. (Source 9)

• Systematic review, Low certainty.
• Size: 51 studies, 39,447 patients.
• Who: healthy individuals, healthcare workers, hospitalised patients, cancer and cystic fibrosis patients, outbreak settings.
• How long: cross-sectional colonisation surveys.
• Result: toxigenic C. difficile asymptomatic colonisation prevalence ranged from 0.5% to 51.5%, highest in cystic fibrosis, outbreak settings and cancer patients, lowest in healthy individuals and healthcare workers.
• Funding: not stated on the page fetched.

The tCDAC prevalence ranged from 0.5 to 51.5%.

After broad-spectrum antibiotics, organisms usually described as harmful bloomed while Bifidobacterium and butyrate producers were depleted, and some species had not returned six months later. (Source 10)

• Cohort study, Low certainty.
• Size: 12 healthy men.
• Who: healthy young adult men given meropenem, gentamicin and vancomycin for 4 days.
• How long: 6 months of follow-up.
• Result: near-baseline composition within 1.5 months, but 9 common species present in all subjects beforehand remained undetectable in most subjects at 180 days.
• Funding: not stated in the abstract.

Initial changes included blooms of enterobacteria and other pathobionts, such as Enterococcus faecalis and Fusobacterium nucleatum, and the depletion of Bifidobacterium species and butyrate producers.

Where the research disagrees

Whether a laboratory report can call a person's gut bacteria good or bad

• European expert panel convened by the Human Microbiome Action consortium (Rodriguez et al., 2024), expert workshop, 21 experts from 8 countries: without a scientific consensus on thresholds, it is impossible to say whether a microbiome is healthy (Source 8)
• Consumer testing companies, as observed by NIST investigators (Servetas et al., 2026), analytical evaluation of seven services against a standard faecal material: Additionally, some companies recommended that customers start taking costly supplements (e.g., probiotics) that are sold by the same company and for which there is very little clinical evidence for efficacy. (Source 11)

A common belief, and what the research shows

The belief: Gut bacteria divide neatly into good ones that should be increased and bad ones that should be eliminated.

What the research shows: The same organism can be on either side depending on strain and situation. The probiotic E. coli Nissle 1917 carries the colibactin island: the authors write that this "illustrates the fine margin between pathogenicity and probiotic activity, and the need to address both the effectiveness and safety of probiotics". Clostridioides difficile is carried symptom-free by up to half of some populations, and probiotic Lactobacillus has been sequenced from the blood of ICU patients who were given it.

Questions and answers

"Good" and "bad" are labels the research literature attaches to gut bacteria mainly by comparing which species are more or less abundant in people who are ill and people who are well. A 2025 review of inflammatory bowel disease describes that pattern directly: in people with the disease the make-up of the gut microbiota is altered, with less of the species the review calls beneficial and more of the ones it calls harmful. The labels therefore summarise an association with a disease state in a group, not a demonstrated effect of the organism in any one person, and the same species can sit on either side of the line depending on strain and setting. (Source 12)

Organisms called beneficial are mostly studied for what they make or displace: butyrate producers and Akkermansia muciniphila have been linked to metabolic measures, and in one small placebo-controlled trial pasteurised A. muciniphila improved insulin sensitivity over three months. Organisms called harmful are studied for toxins and invasion, such as colibactin from pks-positive Escherichia coli. Most of the rest of the literature is association, not demonstrated action in people. (Source 6)

It depends on strain and setting, and the literature says so directly. The probiotic Escherichia coli Nissle 1917 carries the same colibactin-producing pks island that has been tied to mutations in colorectal cancer, and Clostridioides difficile is carried without symptoms by between 0.5% and 51.5% of people depending on the group studied. No agreed threshold exists for calling any person's microbiome good or bad. (Source 13)

Trials that change diet change these organisms. A 17-week randomised trial found a fermented-food diet steadily increased microbiota diversity and decreased inflammatory markers, while a high-fibre diet left community diversity stable. A 10-week randomised sub-study found 20 g/day of inulin-type fructans selectively increased Bifidobacterium and Faecalibacterium. These are descriptions of what trials did, not a recommendation. (Source 14)

Antibiotics do remove organisms, including the ones considered harmful, but they are indiscriminate: a four-day course of meropenem, gentamicin and vancomycin in healthy men wiped out Bifidobacterium and butyrate producers while enterobacteria and Fusobacterium nucleatum bloomed. Nine species present in everyone beforehand were still undetectable in most men after 180 days. The literature we found does not describe any way to selectively remove one unwanted species from a healthy gut. (Source 10)

Birth mode, antibiotics in the mother, and not being breastfed all shape which organisms establish. A study of 596 UK babies found caesarean birth disrupted transmission of maternal Bacteroides strains and led to high-level colonisation by hospital-associated opportunists. In adults, antibiotic courses deplete Bifidobacterium and butyrate producers, and some species do not return within six months. (Source 15)

Fermented foods and fibre are the two food classes tested head-to-head in a randomised trial: the fermented-food arm increased diversity and lowered inflammatory markers, the high-fibre arm increased glycan-degrading enzymes in the microbiome without changing diversity. Chicory-type inulin fed as a 20 g/day supplement raised Bifidobacterium and Faecalibacterium. No whole food supplies Akkermansia muciniphila or Faecalibacterium prausnitzii directly; the foods feed organisms already present. (Source 14)

Low Faecalibacterium prausnitzii travels with inflammatory bowel disease across eleven studies, with a pooled standardised mean difference of -0.94 and a larger gap in Crohn's disease than in ulcerative colitis. That is an association measured at one point in time, so it does not establish that the loss caused the disease rather than resulting from it. No study we found shows what happens to a healthy person who simply lacks one of these species. (Source 5)

Gut bacteria are measured by sequencing bacterial DNA from a stool sample, and the reliability of that measurement has been tested head to head. Researchers sent one standardised human faecal material to seven direct-to-consumer gut microbiome testing services and found major discrepancies both within and across the providers. The spread between providers was on the same scale as the biological difference between different donors, which the authors put down to methodological variability and a lack of quality control. On that evidence a consumer report's verdict on whether your bacteria are good or bad is not a reliable measurement. (Source 16)

Which organisms does the literature associate with benefit, and which with harm?

Benefit is most consistently associated with Faecalibacterium prausnitzii (lower in inflammatory bowel disease across 11 studies, pooled SMD -0.94), Akkermansia muciniphila (inversely correlated with obesity and untreated type 2 diabetes, and improving insulin sensitivity in one 32-person pilot trial) and Bifidobacterium (depleted by antibiotics, increased by inulin). Harm is associated with pks-positive Escherichia coli, whose colibactin leaves a mutational signature found in colorectal cancer genomes, with Fusobacterium nucleatum, over-represented in tumour tissue in 99 patients, and with Clostridioides difficile. Almost all of these are abundance comparisons rather than experiments.

Do the "good" and "bad" labels hold up, or do they depend on context and strain?

They depend on context and on strain. Escherichia coli Nissle 1917 has been used as a probiotic for a century and carries the colibactin gene cluster implicated in cancer-associated mutations; researchers had to mutate a single amino acid to separate the two activities. Toxigenic C. difficile is carried silently by anywhere from 0.5% to 51.5% of people depending on the population. Faecalibacterium prausnitzii itself contains at least two phylogroups distributed differently between healthy people and patients.

If I raise my "good" bacteria, do I get healthier?

Not automatically. In a 10-week randomised sub-study, adding 20 g/day of inulin-type fructans to a plant-based diet selectively increased Bifidobacterium and Faecalibacterium, yet diversity fell and LDL:HDL ratio, IL-10, MCP-1 and TNF-alpha rose compared with controls, attenuating some of the diet's benefits. In the Akkermansia pilot trial, insulin sensitivity improved but body weight did not change significantly. Higher counts of a labelled organism are not themselves an outcome.

References

1. Nature. Mutational signature in colorectal cancer caused by genotoxic pks+ E. coli. 2020. PMID 32106218, DOI 10.1038/s41586-020-2080-8. Read the source
2. PLOS Pathogens. Deciphering the interplay between the genotoxic and probiotic activities of Escherichia coli Nissle 1917. 2019. PMID 31545853, DOI 10.1371/journal.ppat.1008029. Read the source
3. npj Biofilms and Microbiomes. Fusobacterium nucleatum is enriched in invasive biofilms in colorectal cancer. 2025. DOI 10.1038/s41522-025-00717-7. Read the source
4. Microorganisms. Lactobacillus Bacteremia and Probiotics: A Review. 2023. DOI 10.3390/microorganisms11040896. Read the source
5. Gastroenterology Research and Practice. Association between Faecalibacterium prausnitzii Reduction and Inflammatory Bowel Disease: A Meta-Analysis and Systematic Review of the Literature. 2014. PMID 24799893, DOI 10.1155/2014/872725. Read the source
6. Nature Medicine. Supplementation with Akkermansia muciniphila in overweight and obese human volunteers: a proof-of-concept exploratory study. 2019. PMID 31263284, DOI 10.1038/s41591-019-0495-2. Read the source
7. Frontiers in Nutrition. Supplementation with inulin-type fructans affects gut microbiota and attenuates some of the cardiometabolic benefits of a plant-based diet in individuals with overweight or obesity. 2023. DOI 10.3389/fnut.2023.1108088. Read the source
8. Microbiome. Microbiome testing in Europe: navigating analytical, ethical and regulatory challenges [body text, thresholds]. 2024. PMID 39695869, DOI 10.1186/s40168-024-01991-x. Read the source
9. Gut Pathogens. Heterogeneity of Clostridioides difficile asymptomatic colonization prevalence: a systematic review and meta-analysis. 2025. PMID 39871276, DOI 10.1186/s13099-024-00674-0. Read the source
10. Nature Microbiology. Recovery of gut microbiota of healthy adults following antibiotic exposure. 2018. PMID 30349083, DOI 10.1038/s41564-018-0257-9. Read the source
11. Communications Biology. Evaluating the analytical performance of direct-to-consumer gut microbiome testing services [Discussion]. 2026. DOI 10.1038/s42003-025-09301-3. Read the source
12. Frontiers in Cellular and Infection Microbiology. Gut microbiota dysbiosis in inflammatory bowel disease: interaction with intestinal barriers and microbiota-targeted treatment options. 2025. DOI 10.3389/fcimb.2025.1608025. Read the source
13. PLOS Pathogens. Deciphering the interplay between the genotoxic and probiotic activities of Escherichia coli Nissle 1917 [Author summary]. 2019. PMID 31545853, DOI 10.1371/journal.ppat.1008029. Read the source
14. Cell (text read from the University of California eScholarship repository copy). Gut-microbiota-targeted diets modulate human immune status. 2021. PMID 34256014, DOI 10.1016/j.cell.2021.06.019. Read the source
15. Nature. Stunted microbiota and opportunistic pathogen colonization in caesarean-section birth. 2019. PMID 31534227, DOI 10.1038/s41586-019-1560-1. Read the source
16. Communications Biology. Evaluating the analytical performance of direct-to-consumer gut microbiome testing services. 2026. DOI 10.1038/s42003-025-09301-3. Read the source