The gut microbiome beyond the gut
The gut microbiome doesn't act in isolation. The trillions of microbes living in the digestive tract interact with the food we eat and produce compounds that can affect the rest of the body. Some of these microbial products, such as short-chain fatty acids, can interact with cells and tissues beyond the gut [1].
The gut microbiome also communicates with the immune system and helps influence the gut barrier, which controls what passes from the intestine into the body. Through these pathways, changes in the gut microbiome can be associated with processes throughout the body, including immune activity, metabolism, and signaling between the gut and other organs [2].
Some organs have hard-wired connections to the gut. The brain is one example: it links to the gut through the vagus nerve, a direct line that carries signals in both directions between the brain and gut. Gut microbes can influence the signals traveling along this nerve, which helps explain how gut health is tied to mood, stress, and cognitive function [3].
This is why researchers study the gut microbiome in connection with conditions that don't necessarily start in the gut. Research has explored associations between gut microbiome patterns and health conditions involving systems such as the brain, skin, heart, bones, and immune system.
‍
Importantly, finding a microbiome connection does not mean that the microbiome causes the condition. A microbiome pattern may be one part of a much more complex relationship involving genetics, diet, medications, lifestyle, and other factors. Understanding what the research actually shows—and what it doesn't—is an important part of how we built Gut Body Connections.
But how do you determine which connections the research supports?
That's where our research process comes in.
First, we defined what belongs in Gut Body Connections
Before looking for individual microbiome associations, we first researched the field to identify health conditions that could be considered for the new Gut Body Connections.
Our clinical team provided input on conditions to consider. This gave us a broader starting point and helped bring together perspectives from both clinical and emerging microbiome research.
From there, we established criteria for the types of evidence we would consider.
We focused on human research comparing people with a health condition to healthy people, in addition to research designed to investigate potential causal relationships, such as Mendelian randomization studies. Mendelian randomization uses genetic differences between people to investigate whether an exposure may causally relate to an outcome.
We also paid attention to who was actually included in each study.
For example, if we were researching a condition for an adult report, we prioritized evidence from adult populations and excluded studies conducted in children. The same principle applied in reverse. This matters because the microbiome changes throughout life, so findings from one age group aren't automatically applicable to another [14].
That population-level relevance was an important part of deciding which evidence belonged in our research.
We used AI to search the literature at scale
The scientific literature is enormous. Searching it thoroughly can take substantial time, particularly when you're looking across many different conditions and types of microbiome measurements.
We used AI to accelerate this initial stage, screening up to 500,000 research articles per condition.
AI helped us identify potentially relevant studies and findings for closer scientific review. It wasn't used to make the final decision about whether an association should appear in a report.
This distinction matters.
AI can help researchers process large amounts of information quickly, but it can also make mistakes [15]. A paper can be misinterpreted, an important detail missed, or a finding can appear more meaningful than it is. That's why our process doesn't end with the AI search.
More data doesn't automatically mean better evidence. If the studies we find are limited, inconsistent, or don't match the population we're studying, AI can amplify those limitations across many findings. A microbiome association can also be statistically significant without being meaningful enough to include in a consumer report.
That's why we don't collect every significant association we discover. AI helps us find the evidence, but scientific review is still needed to understand what those findings actually tell us. We focus on results that are supported by the broader body of evidence.
We looked at several types of microbiome findings
For each condition, we searched for differences in specific gut microbiome characteristics.
Specific microbes
We looked for individual microbial groups reported at higher or lower levels in people with a condition than healthy groups.
These findings can help describe how microbiome composition differs between groups.Â
Microbial diversity
We also looked at measures of alpha diversity, which describe the variety of microbes within an individual sample.
Different diversity metrics capture different aspects of the microbial community, so we considered the specific measure used in each study rather than treating all diversity findings as interchangeable.
Microbial functions
The microbiome isn't only about which microbes are present. Microbes perform many biological functions, including producing and transforming compounds that interact with the body.
We therefore looked for functional differences, including measures related to:
- LPS production
- Short-chain fatty acid production
- Fiber breakdown
These functional measurements can provide another way to understand how microbial communities differ between groups.
We focused on the types of studies that could answer our question
Our research primarily focused on case-control studies, Mendelian randomization analyses, and relevant meta-analyses.
Case-control studies compare people with a particular condition with a comparison group without that condition [16]. These studies can identify associations between the microbiome and health conditions.
Mendelian randomization studies use genetic information to investigate potential causal relationships. They can provide a different type of evidence from observational research.
Meta-analyses combine results from multiple studies, helping researchers evaluate evidence across a larger body of research [17].
We also deliberately excluded intervention studies, such as studies asking whether probiotics changed a microbiome difference or improved a health outcome. That wasn't because intervention studies aren't valuable. They simply answer a different question.
Our research goal was to understand which microbiome patterns have been associated with each condition, rather than determining whether a particular treatment changes those patterns.
Not all studies carry the same weight
Finding an association in a published paper doesn't automatically mean the finding is reliable, consistent, or relevant to everyone.
We therefore evaluated the strength and consistency of the evidence.
Among other factors, we considered:
- Study design: How was the research conducted?
- Sample size: How many people were included?
- Potential confounding: Could other differences between groups help explain the finding?
- Replication: Has a similar association been reported in other studies?
- Consistency: Do different studies point in the same direction, or are their findings conflicting?
- Population relevance: Does the study population match the population for which we're reporting the association?
This step is especially important in microbiome research. Different studies can report different results, and an association found in one population doesn't necessarily apply universally.
Rather than simply collecting every reported microbiome difference, we considered the broader evidence surrounding each finding.
We built a second AI workflow to audit the first
Because AI can make mistakes, we didn't rely on a single AI-powered research workflow.
We developed an internal AI workflow to independently audit the initial research.
This second layer was designed to help identify potential errors, missing studies, or associations that deserved another look. In other words, we used AI not only to make the initial literature review more efficient, but also to challenge and check the results of that first pass.
This creates an important safeguard: the first round of AI research isn't treated as the final answer.
Human microbiome researchers reviewed the findings
After the AI workflows identified relevant evidence, human microbiome researchers with years of experience in the field independently reviewed the selected findings.
They went back to the underlying research to verify the scientific claims. They checked for mistakes, looked for missing or contradictory associations, and assessed whether the findings were represented appropriately.
This human review is the most critical part of the process.
The researchers didn’t simply check whether a paper existed. They evaluated what the study actually found and whether that finding belonged in your Gut Body Connections section.
That distinction helps prevent a common problem with large-scale literature reviews: taking an isolated result out of context and presenting it as established fact.
We combined published research with our own data
Published research gives us an important starting point, but it isn't the whole picture. We also have large internal datasets of microbiome samples and health information that we can analyze ourselves.
Our data is carefully collected and quality-controlled, allowing our data science team to look for patterns across thousands of samples. We use these data to see which findings from published studies hold up in our own population, as well as discover novel associations that have not previously been described. Sometimes our results support the literature. Other times, they challenge or add context to what has been reported. And we don’t just look for statistical significance, but also effect size, or how big the differences actually are. We can then focus on the strongest, most actionable biological signals.
We also look beyond which microbes are present to understand what they may be doing. Taxa are commonly reported in microbiome studies, but different microbes can perform similar functions, so looking at bacteria alone doesn't always tell the full story. Analyzing functional features of the microbiome gives us another way to understand these connections, highlighting not just who is there but what they can do.
Together, published research and our own data give us a stronger, more definitive way to evaluate microbiome connections and build reports based on more than individual bacterial associations.
Why this approach matters
Microbiome research is developing quickly, and there's a lot to learn. That makes rigorous evidence review especially important.
We use technology where it can genuinely help researchers: searching large bodies of literature, organizing information, and identifying areas that deserve closer attention. But technology doesn't replace human scientific judgment.
Our approach combines large-scale research, defined evidence criteria, AI-assisted screening, independent AI auditing, review by microbiome researchers, and internal validation.
Most importantly, we try to be clear about what the evidence actually shows.
A microbiome association can be interesting and potentially useful without being a prediction or a diagnosis. Keeping that distinction intact is part of how we build research that is informative, scientifically grounded, and useful in the real world.
If you're curious what your own gut microbiome may look like, an Adult Gut Health Test can give you a personalized view of your microbial community and help you better understand the patterns we can measure today.
