Investigational

Three Enterococcus species inside lung tumors switch on a hidden immune sensor

Most microbiome-and-cancer research says bacteria matter without saying how. A Johns Hopkins team has now traced one complete chain inside human lung tumors, and it turns on a vitamin B2 breakdown product and a little-known molecule called MR1. The effect is startlingly species-specific, and it is entirely laboratory work.

Derrick Hayes

Archives · August 20, 2026 · 9 min read

If you have been following the microbiome and cancer story, you already know the headline: the bacteria you carry seem to influence whether immunotherapy works. What has been almost entirely missing is the machinery — a step-by-step chain from one named bacterium, through one named molecule, to one named immune cell doing something measurable. A Johns Hopkins team has now traced exactly that chain inside human lung tumors, and it runs through a breakdown product of vitamin B2. The result is genuinely specific, which is what makes it interesting, and it is also entirely laboratory work, which is what makes it something to watch rather than something to act on.

Key takeaways

Three species of Enterococcus found inside lung tumors made antigen-presenting immune cells display far more of a sensor molecule called MR1, which in turn switched on MAIT cells — but only when a vitamin B2 breakdown product was also present.
A fourth close relative, E. avium, did nothing. Neither did E. coli or Bifidobacterium longum. This is a narrow, species-level effect, not a general vote of confidence in tumor bacteria.
The mechanism is unusual: the bacteria do not make the immune cells produce more MR1, they jam the cell's internal recycling system so existing MR1 piles up on the surface.
This is cell-culture and tissue-sequencing work with no animal model, no patient outcomes, and no probiotic implication. It changes nothing about anyone's treatment today.

Riboflavin-auxotrophic Enterococcus drives MR1 to the cell surface

Three Enterococcus species that cannot make their own vitamin B2 — E. hirae, E. faecalis and E. faecium — sharply increased how much MR1 sat on the surface of antigen-presenting cells, and thereby how strongly MAIT cells were activated, according to work from Pakhi Birla, Franck Housseau and colleagues at the Johns Hopkins Kimmel Cancer Center published in PNAS on 18 August 2026. A freely readable preprint version carries the full figures.

Two pieces of jargon, defined once. MR1 (MHC class I-related protein 1) is a display case on the outside of immune cells; unlike its famous cousins, which show fragments of protein, MR1 was found in 2012 to show fragments of bacterial vitamin chemistry. MAIT cells (mucosal-associated invariant T cells) are an innate-like T-cell population that reads those display cases — a rapid-response force stationed at the gut, lung and liver that reacts to bacterial vitamin B2 manufacturing as a sign that microbes are somewhere they should not be.

The crucial twist is that these Enterococcus species are riboflavin auxotrophs: they cannot build vitamin B2 at all, and so cannot produce the activating molecule themselves. On their own they did nothing. Only when researchers added 5-OP-RU — the unstable molecule formed downstream of bacterial riboflavin synthesis — did the combination light up MAIT receptors far beyond what 5-OP-RU produced alone. The bacterium is not the fuel. It is the amplifier.

Why this matters for lung cancer patients on checkpoint immunotherapy

The population directly in frame is people with non-small-cell lung cancer receiving PD-1 blockade, the checkpoint inhibitor class that includes drugs like nivolumab and pembrolizumab. Roughly speaking, these drugs release the brakes on T cells; they work spectacularly in a minority of patients and not at all in many others, and nobody can reliably predict which. The microbiome has been a leading suspect for years, but mostly as a statistical association between stool composition and response rates.

This work proposes something more mechanical: that the bacteria physically living inside the tumor, not in the gut, set how visible the tumor's antigen-presenting cells are to a whole class of innate immune cells. If that holds up, the composition of a tumor's own resident microbiome becomes a variable worth measuring, and possibly worth adjusting. That would matter to anyone facing a lung cancer treatment decision, and by extension to people with other tumors known to harbor internal bacteria.

MR1-knockout controls and a cloned patient T-cell receptor

The methodological move that gives this study its teeth is that the immune receptor tested came from an actual patient's tumor and was then validated against knockout cells. The team re-analyzed a published Johns Hopkins single-cell dataset of tumor-infiltrating T cells from 16 patients with non-small-cell lung cancer treated with PD-1 blockade before surgery. MAIT cells made up 6.3% of all T cells — plentiful, but present at similar levels in responders (8.6%) and non-responders (9.6%), so raw abundance predicts nothing.

Narrowing to conventional MAIT cells with the canonical receptor left just 7.5% of that cluster. Two of their receptors were cloned into reporter cells. One responded to MR1 and 5-OP-RU; the other did not, a useful reminder that looking like a MAIT cell in sequencing data does not guarantee behaving like one. The working receptor was then tested against human cell lines with and without MR1 deleted, and its response was blocked stepwise by a competing MR1 ligand — three independent ways of confirming the signal genuinely runs through MR1.

Separately, 16S sequencing of archived tumor tissue from 4 patients showed that on average 11% of the bacteria inside those tumors carried all three genes needed to make the riboflavin precursor. Notably, that predicted gene capacity did not correlate with MAIT cell abundance — the authors' own data undercutting the simplest version of their story, which is a good sign about the paper.

What this does and does not mean for you

Nothing here justifies taking a probiotic, and that point deserves emphasis rather than a footnote. Enterococcus faecalis and E. faecium are not friendly supplement organisms; they are among the more common causes of hospital-acquired bloodstream infections and include drug-resistant strains. Nobody should be seeking them out, and no commercial product delivers this effect.

Vitamin B2 supplements are equally beside the point. The active molecule was 5-OP-RU, a fragile intermediate of bacterial riboflavin manufacture that your own cells do not make and that no pill contains. Swallowing riboflavin does not produce it.

If you or someone close to you is being treated for lung cancer, the honest translation is: this is a mechanism paper, and mechanism papers precede treatments by years. The reasonable thing to ask an oncologist is not about bacteria but about whether any tumor-microbiome trials are open — and to expect the answer to be that this specific idea is not yet at that stage. Antibiotic timing around checkpoint therapy is a separate, better-established conversation already worth having.

The finding rests heavily on one patient's expanded T-cell clone

The largest limitation is arithmetic. Conventional MAIT cells were far more abundant in responders than non-responders — 16.7% versus 0.6% of MAIT cells — but the authors state plainly that this gap was driven mainly by the dramatic expansion of a single clonotype in a single patient, MD01-005. That same patient supplied the receptor used in nearly every downstream experiment. It is a well-chosen clone, not a representative sample.

Second, the readout throughout is a reporter cell line lighting up, not MAIT cells killing cancer cells. Increased MR1 on the surface plausibly means better immune surveillance; it has not been shown here to mean smaller tumors. And a curious wrinkle cuts against a simple interpretation: the synergy appeared only with added 5-OP-RU, not with genuinely riboflavin-producing bacteria such as E. coli, hinting that how the molecule gets into the cell matters as much as whether it is there.

What comes next: animal models and species-level tumor profiling

The obvious next experiment is the one this paper does not contain — putting these bacteria and the metabolite into a tumor-bearing animal with an intact immune system and measuring whether tumors actually shrink. Until that exists, the causal chain stops at "more MR1, stronger receptor signal."

Two other threads are open. Because E. avium failed while its three close relatives succeeded, something in the Enterococcus genome at species level is responsible, and identifying it is a tractable genetics problem. And because the mechanism turned out to involve bacteria disrupting the cell's endosomal recycling — MR1 messenger RNA did not change at all, so the cells are not making more, merely failing to dispose of what they have — the door opens to drugs that mimic the effect without any bacteria present. That would be a far more practical therapeutic than introducing Enterococcus into a patient.

Common questions

Does this mean bacteria in tumors are good? No. Three specific Enterococcus species amplified an immune signal under laboratory conditions when a bacterial metabolite was supplied alongside them. A fourth Enterococcus species, E. coli, and Bifidobacterium longum all failed to do so. Tumor bacteria have also been linked to worse outcomes in other cancers.

Should I take a probiotic or a vitamin B2 supplement? No. The activating molecule is 5-OP-RU, a short-lived by-product of bacterial riboflavin synthesis that supplements do not contain and your body does not make. The Enterococcus species involved are opportunistic pathogens, not probiotics.

Was this study done in people? Partly. The tumor sequencing and the T-cell receptor came from real lung cancer patients, but every experiment testing the bacteria was done in cell culture using human cell lines and dendritic cells. No patients received anything, and no treatment outcomes were measured.

Is it peer reviewed? Yes. It appeared in the Proceedings of the National Academy of Sciences on 18 August 2026, after circulating as a bioRxiv preprint from November 2025. Peer review confirms the work was competently done; it does not make a cell-culture finding clinically applicable.

Evidence Rating

INVESTIGATIONAL — 78/100

Caveats

Entirely laboratory work: human tumor sequencing re-analysis plus cell-line and dendritic-cell experiments. No animal model, no clinical trial, and no patient outcome data of any kind.
The immune ‘activation’ measured is a Jurkat reporter cell line carrying a patient’s MAIT receptor lighting up, not primary MAIT cells killing tumor cells. Nobody showed a tumor shrinking.
Sample sizes are small: 16 patients in the sequencing re-analysis, only 4 patients for tumor bacterial sequencing, and the key T-cell receptor came from one expanded clone in one patient.
Highly species- and metabolite-specific. E. hirae, E. faecalis and E. faecium worked; E. avium, E. coli and Bifidobacterium longum did not. This is not a general ‘tumor bacteria are good’ finding.
The effect required adding the metabolite 5-OP-RU from outside. The Enterococcus species tested cannot make riboflavin themselves, so the bacterium alone does nothing.
No probiotic implication whatsoever. Enterococcus faecalis and E. faecium are also common causes of hospital-acquired bloodstream infections and are not benign supplements.
Specific figures cited here are drawn from the authors’ freely readable bioRxiv preprint; the peer-reviewed PNAS version added four authors and may report additional or revised data.
The preprint declares no competing interests and lists foundation funding; the PNAS competing-interest statement could not be independently retrieved.

Peer-reviewed mechanistic study published in PNAS (123(34):e2617943123, 18 Aug 2026), with a freely readable earlier preprint on bioRxiv (10.1101/2025.11.25.690539). Combines a re-analysis of published single-cell RNA and TCR sequencing from 16 non-small-cell lung cancer patients treated with neoadjuvant PD-1 blockade, 16S rRNA sequencing of archived tumor tissue from 4 patients, and in vitro reporter assays using one patient-derived MAIT T-cell receptor against human cell lines and dendritic cells. Causal chain is well controlled (MR1-knockout and B2M-knockout comparator lines, MR1 antagonist dose-response, MyD88-knockout cells, pharmacological trafficking inhibitors). No animal model, no patient outcomes, no primary MAIT cell killing assay. Much of the human signal traces to a single expanded clonotype in a single patient. The preprint declares no competing interests; funding is from the Commonwealth and Willowcroft foundations.

References

Pakhi Birla, Lansaol Yang, Wanting Shan, et al. (2026). Select intratumoral riboflavin-auxotrophic Enterococcus species enhance cell surface MR1 expression and MAIT TCR activation in lung cancer. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2617943123
Synergistic role of riboflavin-auxotrophic Enterococcus for MR1 expression and intra-tumoral MAIT cell activation (bioRxiv preprint). bioRxiv. https://www.biorxiv.org/content/10.1101/2025.11.25.690539v1.full
Justina X. Caushi, Jiajia Zhang, Zhicheng Ji, et al. (2021). Author Correction: Transcriptional programs of neoantigen-specific TIL in anti-PD-1-treated lung cancers. Nature. https://doi.org/10.1038/s41586-021-03893-6