Colorectal cancer (CRC) is the third most common cancer in the world and the second leading cause of cancer-related death. While surgery is the main treatment for early and middle-stage disease, more than 30% of patients who have surgery still see their cancer return within a few years.
Postoperative recurrence is particularly problematic because about 80% of colon cancer recurrences happen within the first three years after surgery. Currently, doctors use tumor staging systems based on how far the cancer has spread to nearby tissue and lymph nodes, but these systems often fail to accurately predict who will relapse.
Researchers have long suspected that the gut microbiota - the community of trillions of bacteria living in the digestive tract - plays a role in how colorectal cancer develops and progresses. This study set out to identify specific bacterial and chemical markers in the tumor tissue that could predict which patients are most likely to have their cancer return after surgery.
The gut microbiota is not just a passive passenger in the digestive system - it actively interacts with cancer cells and the immune system. Certain bacteria can damage DNA, trigger inflammation, and activate cancer-promoting signaling pathways that help tumors grow.
Fusobacterium nucleatum is one of the most studied bacteria in CRC. Multiple studies have linked it to cancer initiation, faster progression, chemotherapy resistance, and worse patient outcomes. Once F. nucleatum colonizes a primary tumor, it can even travel with cancer cells to distant metastatic sites.
Beyond bacteria, the intestinal metabolome - the chemical products created by bacteria and tumor cells - also shapes cancer behavior. Some metabolites like butyrate can actually help prevent cancer, while others promote tumor growth. This study analyzed both the bacteria and the metabolites present in tumor tissue to build a more complete picture of recurrence risk.
The study enrolled 126 colorectal cancer patients who had surgical removal of their tumors. During surgery, researchers collected small tissue samples from the tumor surface, froze them immediately, and stored them for analysis. After surgery, all patients were followed for over 33 months on average to track whether the cancer came back.
16S rRNA gene sequencing was used to identify which bacteria were present in 125 tissue samples. This technique reads a specific section of bacterial DNA that acts like a barcode, allowing researchers to identify hundreds of different bacterial species from a single sample. Separately, liquid chromatography-mass spectrometry (LC-MS) was used to measure metabolite levels in 120 samples.
Machine learning algorithms were then applied to find the most informative bacterial and metabolite features for predicting recurrence. Specifically, the researchers used LASSO regression (a statistical method that selects the most useful predictors from a large pool of candidates) combined with Random Forest classifiers to build predictive models. This approach handles the complexity of biological data better than traditional statistical methods.
To ensure the results were reliable, the data was split into training sets (used to build the model) and independent test sets (used to evaluate performance without bias). The final models were also tested for their ability to predict not just recurrence status but actual recurrence-free survival time.
27 out of 126 patients (21.4%) developed cancer recurrence during follow-up. The researchers found that the bacteria living in the tumors of patients who later had recurrence looked quite different from those who remained cancer-free - and these differences were actually more pronounced than differences between early and late-stage cancers.
Microbial diversity - a measure of how many different types of bacteria are present - was significantly reduced in patients who later had recurrence. Lower microbial diversity is often a sign of an unhealthy gut ecosystem. Recurrent patients showed an increase in potentially harmful bacteria, particularly Fusobacterium (3-fold higher) and Peptostreptococcus (2.3-fold higher), compared to non-recurrent patients.
These two bacterial groups together made up more than 30% of the total mucosal microbial abundance in recurrent patients. In contrast, beneficial bacteria like Bacteroides were depleted in recurrent patients. This pattern - enrichment of harmful bacteria and loss of protective ones - creates a microenvironment that seems to favor cancer recurrence.
Metabolomics analysis identified 721 metabolites present in the tumor tissue. Among these, several showed significantly different levels between patients who had recurrence and those who did not. Interestingly, the metabolite differences between recurrent and non-recurrent patients were much larger than those between different cancer stages - mirroring the pattern seen with bacteria.
Patients who later had recurrence showed elevated levels of polyamines - a class of molecules including putrescine and cadaverine - in their tumor tissue. These polyamines are known to promote cell growth and may help bacteria form dense protective communities. Meanwhile, arginine, an amino acid with potential cancer-suppressing properties, was significantly depleted in the recurrence group.
Pathway analysis revealed that arginine and proline metabolism was one of the most disrupted pathways in recurrent patients. This metabolic shift is significant because arginine can be converted into putrescine through a chain of biochemical reactions - creating a direct chemical link between altered amino acid metabolism and the bacteria associated with recurrence.
Machine learning analysis identified five bacterial genera and five metabolites as the most powerful predictors of recurrence. The five bacteria were Peptostreptococcus, Fusobacterium, Bacteroides, Porphyromonas, and Prevotella; the five metabolites were alanylglutamic acid, putrescine, arginine, histidine, and sebacic acid.
When bacteria alone were used for prediction, the model achieved an AUC of 0.81 (where 1.0 would be perfect). Adding the metabolite data significantly boosted performance to an AUC of 0.89 - a 14% improvement in accuracy and 31% improvement in specificity. This demonstrates that combining bacterial and metabolic information provides a richer and more reliable picture of recurrence risk than either alone.
A risk score derived from these 10 biomarkers successfully separated patients into high-risk and low-risk groups. High-risk patients had significantly shorter recurrence-free survival, and this risk score remained an independent predictor of outcomes even after accounting for tumor stage (adjusted hazard ratio = 1.59, P less than 0.0001). This means the microbiome-metabolome signature provides information beyond what conventional staging already tells us.
Co-aggregation experiments revealed that F. nucleatum and P. anaerobius have a strong physical affinity for each other. When the two bacteria were mixed together in laboratory conditions, 58% of bacterial cells formed clumps within 2 hours - rising to 79% after 5 hours. In contrast, neither bacterium co-aggregated significantly with a neutral control bacterium (E. coli).
This physical partnership is mediated by a surface protein on F. nucleatum called RadD, an arginine-sensitive adhesin. When researchers added arginine to the bacterial mixture, co-aggregation dropped dramatically. This is a key finding - it connects the metabolic finding (low arginine in recurrent patients) to the bacterial finding (high co-aggregation of these two species), suggesting that depleted arginine in the tumor tissue removes a natural check on bacterial co-aggregation.
Together, F. nucleatum and P. anaerobius form dense dual-species biofilms - structured communities of bacteria encased in a protective matrix. Biofilms are much harder for the immune system to attack than free-floating bacteria, and they allow the bacteria to maintain a persistent presence in the tumor tissue. Critically, P. anaerobius was found to significantly enhance how strongly F. nucleatum attaches to tumor cells, creating a cooperative invasion strategy.
Putrescine is a polyamine produced when bacteria metabolize arginine. The study found that putrescine was elevated in the tumors of patients who later had recurrence. To understand why this matters, researchers tested whether putrescine affects bacterial behavior directly.
Adding putrescine to bacterial cultures significantly stimulated growth of the F. nucleatum and P. anaerobius co-culture at concentrations of 100-500 micromolar. At lower concentrations (10-50 micromolar), putrescine still strongly promoted biofilm formation in a dose-dependent manner - the more putrescine present, the bigger the biofilm.
This creates a potentially self-reinforcing cycle: bacteria convert arginine to putrescine, putrescine encourages more bacterial growth and biofilm formation, and the resulting bacterial community further depletes arginine in the local environment. This metabolic feedback loop between bacteria and their chemical environment may be a key driver of the microbial community changes seen in recurrent patients.
This study introduces a new way of thinking about recurrence risk in colorectal cancer. Rather than relying solely on tumor stage and appearance under the microscope, doctors might one day use the microbial and metabolic profile of a patient's tumor tissue to more accurately predict who needs more intensive surveillance or additional treatment after surgery.
The finding that RadD-mediated interactions between bacteria may drive colonization suggests a potential therapeutic target. Disrupting the physical interaction between F. nucleatum and P. anaerobius - for example with antibiotics targeting specific adhesion proteins, or by restoring normal arginine levels - could theoretically reduce the ability of these bacteria to colonize tumor tissue.
However, the researchers acknowledge important limitations: the study was conducted at a single institution with a relatively small cohort of 126 patients, and the predictive model has not yet been validated in larger independent groups of patients. Detailed information about each patient's chemotherapy regimen was also not available. Larger prospective studies are needed before this approach could be used in clinical practice.
This study demonstrates that the microbial and metabolic environment of colorectal tumors carries important information about recurrence risk that goes beyond traditional staging. Using machine learning to integrate both types of data produced a predictive model with strong performance (AUC 0.89).
The discovery of a biological partnership between F. nucleatum and P. anaerobius provides a mechanistic explanation for why these two bacteria are so strongly associated with recurrence. Their ability to co-aggregate, form biofilms, and collectively enhance tumor adhesion - fueled by the polyamine metabolite putrescine - suggests a coordinated biological process that promotes cancer recurrence.
Future research should focus on validating these findings in diverse patient populations, developing clinically practical ways to measure the relevant bacteria and metabolites, and testing whether interventions targeting the arginine-putrescine metabolic axis or the bacterial adhesion proteins can reduce recurrence rates in high-risk patients.