Analysing the oral microbiome could help diagnose colon and stomach tumours, according to a study

Overall, the work is of good quality. The topic is also clinically relevant, as being able to diagnose cancer from an analysis of a saliva sample using microbial markers would be a major breakthrough.

The concept is not entirely novel (particularly in the pancreas, where there are publications reporting microbial signatures associated with cancer). Perhaps the most novel aspect of this publication is the MF index [from ‘mouth to faeces’], derived from the intra-patient analysis of saliva and faecal samples. This index is based on the presence of bacterial sequences common to both types of sample (indicating potential transmissibility) rather than on overall relative abundance (as most studies do). The results regarding its diagnostic utility in colorectal cancer are promising (sensitivity is higher than that of the faecal occult blood test) but require independent validation, as the cohort in which the faecal occult blood test was examined is very small. The results for stomach cancer are less robust when validated in independent cohorts.

However, there are many limitations. The first and most important is that the analytical methodology (16S gene sequencing) does not allow for the precise classification of bacteria at the species and strain level. In other words, although the authors identify the same bacterial sequences in the two samples (saliva and faeces from the same patient) — and thereby generate the MF index — this does not necessarily indicate that the same bacterial strain has ‘travelled’ from the mouth to the gut. To determine this, more in-depth sequencing would be required. Secondly, the results do not allow for the establishment of causality (Do these bacteria that travel from the mouth to the gut have any effect once they are there? Do they help the tumour to grow?), and therefore do not allow for the evaluation of possible therapeutic interventions that might reduce the mouth-to-gut flow of these bacteria or promote that of other, more beneficial ones. And, clearly, a prospective clinical trial is needed to validate the MF index as a diagnostic test.

As with all studies of predictive models for microbiome signatures, it is very difficult to ascertain whether the validation methodology is reliable – that is, free from noise and bias. This type of analysis tends to suffer from subtle methodological ‘pitfalls’. For external validations, rather than applying a ‘locked’ predictive model with pre-specified cut-off points for each variable and a technically predefined score, the authors adapt the model in various ways (which are complex and difficult to reproduce or understand for non-bioinformaticians) to suit the external validation datasets.

In short, it has no impact on clinical practice and is unlikely to be validated in the future.

This is a methodologically sound and important study, particularly due to the high number of paired oral and faecal samples and because the samples from cancer patients were collected prior to any treatment, thereby avoiding any potential impact of therapy. The presence of oral bacteria in the gut had previously been reported as a factor associated with gastrointestinal cancer, and the development of the mouth–gut transmission index is of interest in this regard. The predictive results are particularly promising for colorectal cancer and support the idea that the overlap between oral and gut microbiota may provide additional diagnostic information.

Clinical conclusions should be drawn with caution. The study uses 16S sequencing; therefore, detecting identical signals in the mouth and faeces does not necessarily prove direct transmission at the strain level. The design is cross-sectional and lacks some important technical controls. The colorectal cancer samples include many cases of advanced cancer, meaning the disease may already have had a significant impact and does not reflect the aim of population-based screening, which seeks to detect earlier stages and even lesions that predate tumour development (high-risk lesions).

Predictive algorithms based on faecal material exist, including some developed by us at the IRB-BSC, which yield comparable results for colorectal cancer, calling into question the need for additional oral samples. Although validated in independent cohorts, the cohort used for each disease is limited. I therefore view this study as a very interesting proof of concept with diagnostic potential, but prospective validation—with higher metagenomic resolution and in real-world screening populations—would still be required before its clinical application can be considered.