Gut Bacteria Predict Melanoma Recurrence After Immunotherapy With Up to 94% Accuracy
Pre-treatment gut microbiome signatures predicted cancer recurrence in 674 melanoma patients receiving immune checkpoint blockade, with AUC up to 0.94.
Summary
A landmark study published in Cell analyzed stool samples from 674 high-risk melanoma patients enrolled in the CheckMate 915 phase 3 trial. Researchers found that specific gut bacteria present before treatment — including Eubacterium, Ruminococcus, and Clostridium — were strongly linked to whether patients would experience cancer recurrence after receiving immune checkpoint therapy. When patients in validation groups had gut microbiome profiles similar to the discovery group, prediction accuracy reached an impressive AUC of 0.78 to 0.94. Importantly, gut microbiome composition remained largely stable after treatment, suggesting these bacterial signatures reflect a durable biological state. The findings position gut microbiome profiling as a potential clinical tool to personalize immunotherapy decisions in melanoma patients.
Detailed Summary
Immune checkpoint blockade has transformed outcomes for high-risk melanoma, but 25–40% of patients still experience recurrence after surgery and adjuvant treatment. Identifying who will respond — and who will relapse — remains one of oncology's most pressing challenges. This study suggests the gut microbiome may hold a key part of the answer.
Researchers from NYU and UC San Diego analyzed pre-treatment stool samples from 674 patients enrolled in CheckMate 915, a phase 3 clinical trial comparing nivolumab plus ipilimumab versus nivolumab alone across five geographic regions. Using region-specific and cross-region meta-analyses, they identified gut bacterial taxa whose abundance before treatment was associated with recurrence-free survival.
Several bacterial groups emerged as significant predictors, including Eubacterium, Ruminococcus, Firmicutes, and Clostridium — taxa previously implicated in immune regulation and short-chain fatty acid production. Prediction performance was strongest when the validation cohort's microbiome profile closely resembled the discovery cohort's, with AUC values ranging from 0.78 to 0.94 among closely matched individuals (Jensen-Shannon divergence ≤ 0.11). This suggests microbiome-based prediction models may need to account for regional and population-level variation in gut ecology.
Notably, gut microbiome composition remained largely stable following immunotherapy treatment, reinforcing the idea that pre-treatment profiling captures a meaningful and durable biological signal rather than a transient state.
The clinical implications are significant. If validated prospectively, gut microbiome profiling before adjuvant immunotherapy could help oncologists stratify patients, intensify monitoring for high-risk individuals, or explore microbiome-modifying interventions — such as dietary changes, probiotics, or fecal microbiota transplantation — to improve outcomes. This study represents one of the largest microbiome-oncology datasets to date and adds compelling evidence that the gut-immune axis is a clinically actionable target in cancer care.
Key Findings
- Pre-treatment gut microbiome predicted melanoma recurrence with AUC 0.78–0.94 in matched patient cohorts.
- Eubacterium, Ruminococcus, Firmicutes, and Clostridium were key taxa linked to recurrence-free survival.
- Analysis covered 674 patients across five geographic regions in a phase 3 clinical trial.
- Gut microbiome composition remained stable after immune checkpoint blockade treatment.
- Prediction accuracy was highest when discovery and validation cohorts had similar microbiome profiles.
Methodology
This prospective study analyzed pre-treatment stool samples from 674 patients in CheckMate 915, a phase 3 RCT comparing nivolumab plus ipilimumab versus nivolumab monotherapy in resected high-risk melanoma. Region-specific and cross-region meta-analyses were performed, with microbiome similarity quantified using Jensen-Shannon divergence to assess prediction model transferability.
Study Limitations
This summary is based on the abstract only, as the full text is not open access. The predictive model's performance was highly dependent on microbiome similarity between cohorts, limiting generalizability across diverse populations. Prospective validation studies are needed before clinical implementation.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.