Differences in the composition of the gut microbiome have been associated with a diverse range of diseases. This is exciting as manipulations of the gut microbiome (using methods such as pre- and probiotics, microbiome transplants, or dietary interventions) could represent a wave of new therapies that have been relatively unexplored. However, we don't know which diseases have the greatest association with the gut microbiome, and which aspects of the gut microbiome are most associated with health/disease. The TwinsUK cohort presented a unique opportunity to explore these questions.
The cohort has been running for over 25 years. Its members carry out regular clinical visits, where they provide samples and data spanning various aspects of health and lifestyle; including faecal samples that have been used to generate gut microbiota profiles for thousands of participants. This provides a uniform platform to compare gut microbiome associations between different health factors - such as common diseases and use of prescription medications as in our recent study published in Nature Communications.
One of the largest hurdles to overcome in this project was the substantial data preparation involved. Combining decades worth of data to generate a disease table was a mammoth task that required careful recoding and combining of questions asked over numerous questionnaires. Similarly, the self-reported medication data provided by participants contained many brand names, spelling mistakes, and non-drugs. This required extensive cleaning and categorisation into drug types and the expert knowledge of clinicians.
Overall, we investigated 38 different diseases and 51 medications in the study. We found, reassuringly, that the diseases with most associations have previously been studied in relation to the gut microbiome, or might intuitively be expected to have a large effect – such as food allergies, type 2 diabetes, and inflammatory bowel disease. The results also highlighted where research might be less fruitful, for example, respiratory allergies had many cases but far fewer microbiome associations. Of course, there is an inevitable limit to the value of correlative results, but they should guide further mechanistic research towards diseases where gut microbiome effects are most likely implicated.
We found many associations with medications known to affect the gut microbiome, such as antibiotics and proton pump inhibitors. However, to our surprise, we also observed associations with several widely used medications that have been relatively unexplored in this regard. These included inhaled anticholinergic medications, selective serotonin reuptake inhibitors, and paracetamol. This highlights the importance of considering non-obvious medications as confounders in human microbiome studies. It will also be interesting to follow-up these associations to uncover the underlying mechanisms and their influence on human health.
We carried out clustering of the diseases and microbiome features based on their associations, expecting to find groupings of similar diseases (for example allergies) that were defined by associations with particular microbiome traits. This was not the case and there was little clustering within diseases. We did, however, observe a trend within the microbiome traits: almost all were associated with more than one disease with consistent directions of effect across them. As such, we could classify the gut microbiome traits into two groups – those generally higher in disease and those generally lower in disease (or higher in health).
Finding bacteria that define a general healthy/unhealthy gut microbiome is a core goal for the development of diagnostics and therapeutics. Current efforts to identify such features have produced conflicting results: which gut microbes are healthy or unhealthy depends on who you ask and how you define the microbes. Much more work is required to find the level where common health effects are manifesting in the gut microbiome (should we be looking at individual genes, taxa, or community-level functions?) and how such features could be used to improve human health, but these results provide further support that such features might exist.