What Long COVID Microbiome Research Has Found (And What Patients Should Make of It)
A note before we start
This is a research summary, not medical advice. We are writing it because the gut microbiome is a genuinely active area of long COVID research, and we think patients deserve a careful read of what has actually been found, rather than another round of overconfident claims.
If you are reading this as a long COVID patient, you have probably already been let down by overconfident science. Treatments have been announced and walked back. Mechanisms have been proposed and revised. We are not going to tell you that the microbiome is the answer. We are going to lead with what has been observed, be plain about what is hypothesized, and end with what remains uncertain. The honest position today is that researchers have found reproducible signatures in some cohorts, the mechanisms are not settled, and no microbiome-based intervention has proven itself in a properly powered randomized trial.
We work on microbiome research tooling at Graphomics. We are not a clinical group, and nothing here should be read as a recommendation.
What’s been observed
The strongest claim that can be made today is descriptive: across multiple cohorts, the gut microbiomes of long COVID patients look different from controls, and from acute COVID patients who recovered fully.
The Hong Kong group led by Siew Ng and Yun Kit Yeoh has produced the most-cited body of work here. In Yeoh et al. 2021 (Gut), the team profiled 100 patients with COVID-19 and 78 controls and reported that gut microbiota composition was altered during acute disease and that those alterations correlated with disease severity and inflammatory markers. The follow-up by Liu et al. 2022 (Gut) extended this into post-acute sequelae, profiling 106 patients at six months and reporting that those with persistent symptoms (post-acute COVID-19 syndrome, or PACS) had distinct gut microbiome composition compared to those who recovered. They reported lower abundance of Faecalibacterium prausnitzii and several butyrate-producing taxa in PACS patients, and elevated abundance of opportunistic species. The cohort was modest, around 106 patients followed longitudinally, and recruitment was from a single regional health system, which matters for generalization.
Independent groups have reported broadly compatible findings. Zuo et al. 2021 (Gastroenterology) found altered fecal microbiomes during acute infection. Vestad et al. 2022 (Journal of Internal Medicine) reported that respiratory dysfunction three months after severe COVID-19 was associated with gut microbiota alterations including shifts in Veillonella and Flavonifractor and reductions in selected SCFA producers, though the specific taxa flagged differ between studies. Stanford’s Bhatt group has separately led work on gut viral persistence in long COVID — Natarajan et al. (Bhatt lab, 2022) reported SARS-CoV-2 RNA in stool of patients up to seven months after mild infection, framing long COVID as potentially driven in part by hidden gut reservoirs.
A separate but related thread is gut viral persistence. Gaebler et al. 2021 (Nature), from the Nussenzweig group at Rockefeller, reported SARS-CoV-2 RNA and antigen in intestinal biopsies of patients months after infection, even when nasopharyngeal swabs had been negative for some time. Subsequent work from multiple groups has reproduced gut-mucosal viral RNA detection in subsets of post-acute patients, though the proportion of patients with detectable persistence and the relationship to symptoms varies across studies.
What replicates well: long COVID cohorts, on average, look dysbiotic relative to controls, and depletion of butyrate producers comes up across studies. What replicates less cleanly: which specific taxa are up or down, by how much, and whether the signature is specific to long COVID or shared with other post-infectious and chronic inflammatory conditions. Cohort sizes are typically in the dozens to low hundreds. Cross-cohort meta-analyses are starting to appear but are still limited by methodological heterogeneity, particularly around sequencing approach, sample handling, and how “long COVID” is defined.
What dysbiosis signatures might mean
A signature is not a mechanism. The literature has converged on three non-exclusive hypotheses, none of which is settled.
The first is gut viral persistence. If SARS-CoV-2 RNA or antigen sits in intestinal tissue for months, as Gaebler et al. and follow-up groups have reported, that could plausibly drive ongoing local immune activation, alter the mucosal environment, and reshape which microbes thrive there. This is mechanistically coherent but the causal arrow has not been established. We do not know whether persistence drives dysbiosis, whether dysbiosis sustains conditions favorable to persistence, or whether both are downstream of something else.
The second is immune dysregulation. Long COVID patients show a range of immune abnormalities in some studies: altered cytokine profiles, persistent T cell activation, autoantibodies in subsets of patients. The gut is the body’s largest immune interface, so any chronic systemic immune perturbation should be expected to leave a footprint there. Under this view, the microbiome signature is a readout of immune state rather than a driver of symptoms.
The third is gut-brain signaling. Many long COVID symptoms — fatigue, brain fog, mood changes, dysautonomia — overlap with the symptom set that gut-brain-axis research has historically been interested in. Microbial metabolites, particularly short-chain fatty acids and tryptophan-derived compounds, are known to influence neuroinflammation and vagal signaling in animal models. Whether they meaningfully drive long COVID neurocognitive symptoms in humans is, at this point, a hypothesis with circumstantial support, not an established mechanism.
We want to be plain. The signatures are real. The mechanisms are unclear. Any treatment derived from this is, today, speculative.
The interventions that have been tried
A small number of interventions have been tried, and the data is preliminary in every case.
Fecal microbiota transplantation (FMT) has been explored in case series and small pilots. The Hong Kong group has reported preliminary FMT case-series data in discharged COVID-19 patients (Liu et al. 2021), and a more recent open-label interventional study from the same network reported FMT-associated improvements in post-acute COVID-19 sleep disturbance (Clinical Gastroenterology and Hepatology, 2024), but the studies have been small, often open-label, and without proper placebo controls. FMT is also not a routine intervention. It carries known risks including infection transmission, and outside of C. difficile it is investigational. The signal in long COVID, where it exists, is preliminary.
Probiotics have been tested in several small randomized studies. The most-cited is the SIM01 RECOVERY trial from CUHK (Lau et al. 2024, Lancet Infectious Diseases), a 463-patient randomized double-blind placebo-controlled trial of a synbiotic preparation in PACS that reported significant improvement in fatigue, memory, concentration, and gastrointestinal symptoms at six months. The trial is the largest in the space and is genuinely better-powered than the case series, though it is single-network and has not yet been independently replicated. Outside SIM01, sample sizes have been small, primary endpoints have varied, and effect sizes have not been large. There is no probiotic regimen that has cleared the bar of properly powered, adequately blinded, replicated randomized trials for long COVID.
Dietary interventions — Mediterranean-style diets, fiber-forward eating, low-FODMAP for GI-predominant patients — have biological rationales rooted in microbiome modulation, and they are reasonable things to discuss with a clinician. Whether they specifically address long COVID, as opposed to general health, is not established in trial data.
We are not going to recommend any of these. The honest read is: small samples, mixed outcomes, no replicated RCT. Patients trying these on their own, sometimes at significant cost, are not unreasonable for trying — but the evidence does not yet support endorsement.
What we would flag for any patient considering an intervention is the same standard of skepticism that applies to any underpowered field. Ask whether the trial was randomized. Ask the sample size. Ask whether the endpoint was patient-reported or biomarker-based, and which was pre-registered. Ask whether the result has replicated. These are not unreasonable questions, and clinicians and researchers worth talking to will not be defensive about them.
What research is doing next
The next phase of this work is methodological more than it is therapeutic.
Several large standardized cohorts are now in place or being assembled. The NIH RECOVER initiative in the United States has enrolled thousands of patients with longitudinal sampling, including stool collection in subcohorts. International long COVID consortia have been working toward harmonized phenotyping so that cross-cohort comparison becomes meaningful rather than confounded. The shift from dozens-to-hundreds-of-patients single-site studies to thousands-of-patients multi-site cohorts is the prerequisite for any of the existing signatures to either solidify or fall apart.
Multi-omic integration is the second methodological push. A microbiome composition snapshot tells you what is present. To get at mechanism, researchers are pairing 16S or shotgun sequencing with metabolomics, host transcriptomics, immune phenotyping, and in some cases viral persistence assays on the same patients. This is expensive and slow, but it is the way the field will move from “the microbiome looks different” to “here is what the microbiome is doing, in patients with this immune phenotype, and here is the candidate intervention.”
Trial design is also tightening. A handful of randomized, placebo-controlled trials of microbiome-targeted interventions are registered or running, with pre-registered primary endpoints and adequate blinding. Results from these will take time. We would expect read-outs over the next two to four years rather than the next several months. Anyone telling you a microbiome-based long COVID treatment is imminent is getting ahead of the data.
Realistically, the near-future contribution of microbiome research to long COVID is more likely to be diagnostic stratification — identifying which patients have a microbiome-and-immune phenotype that might respond to which class of intervention — than a single broadly applicable treatment.
Closing
We are aware that “more research is needed” reads as dismissive when you are the person living with the symptoms. It is not meant that way. The research community is taking long COVID seriously. It is also moving slowly, partly because the condition is heterogeneous, partly because rigorous trials take years, and partly because the field is being careful — appropriately, in our view — not to repeat the early-pandemic pattern of announcing treatments before the evidence justifies them.
If you are a patient, the honest summary is that microbiome science has produced real observations, plausible hypotheses, and no proven treatments yet. That is a frustrating place for the field to be, and we don’t want to dress it up.
This article does not recommend treatments. If you are managing long COVID, work with your clinical team.