A New Path Toward Microbiome-Informed Precision Nutrition

Short-chain fatty acids (SCFAs) are beneficial molecules created by the bacteria residing in our gut that are closely tied to improved host metabolism, lower systemic inflammation, better cardiovascular health, lower cancer risk, and more. However, SCFA profiles can vary widely between individuals consuming the same exact diet and we currently lack tools for predicting this inter-individual variation.

ISB researchers have developed a novel way to simulate personalized, microbiome-mediated responses to diet. They use a microbial community-scale metabolic modeling (MCMM) approach to predict individual-specific SCFA production rates in response to different dietary, prebiotic, and probiotic inputs.

In other words, ISB scientists can build a “digital twin” of gut microbiome metabolism that can simulate personalized responses to diet, using gut microbiome sequencing data and information on dietary intake to constrain each individual-specific model. They detailed their results in a paper published in Nature Microbiology.

“To a first approximation, the gut microbiome is a bioreactor that converts dietary fibers into these SCFAs,” said Dr. Sean Gibbons, ISB associate professor and co-senior author. “Understanding how the ecology of the gut and dietary intake can be quantitatively mapped to SCFA outputs will represent a major advance in translating microbiome science into the clinic.”

Unlike black-box machine learning approaches to prediction, MCMMs are transparent and mechanistic, with tens of thousands of metabolites and enzymes across dozens of organisms providing a high degree of knowledge about the specific microbes, dietary components, and metabolic pathways that contribute to SCFA production. Despite this transparency, the complexity of these models makes them difficult to experimentally validate.

One approach is to measure SCFA production rates for an entire ecosystem, and then compare these ecosystem-scale measures to their cognate model predictions. However, measuring SCFAs in the wild is tricky because the body rapidly consumes them after they are created. In order to overcome this challenge, the authors measured SCFA production rates from in vitro (i.e., test tube) communities of random mixtures of human gut bacterial isolates and from ex vivo (i.e., outside the body) stool homogenates from different humans incubated in an anaerobic chamber with a variety of dietary fibers.

By isolating microbiota-driven SCFA production from host absorption, ISB scientists were able to show that MCMM predictions were significantly correlated with measured production rates across a range of fibers for both butyrate and propionate, two of the most abundant and physiologically potent SCFAs.

While in vivo (i.e., in the body) measurements of butyrate and propionate production were not feasible, the authors were able to use indirect associations between SCFA production rates and blood-based health markers to validate the physiological effects of inter-individual differences of production. First, they showed that MCMM predictions could differentiate between individuals from a high-fiber feeding study who showed divergent immune responses: most individuals showed a reduction in systemic markers of inflammation, but a subset of people showed an increase in inflammation on a high-fiber diet. Individuals in the high-inflammation response group showed a significantly reduced capacity for producing propionate, according to MCMM predictions. Next, the authors showed that butyrate predictions were significantly associated with blood markers of cardiometabolic and immune health in a population of over 2,000 individuals. Specifically, higher MCMM-predicted butyrate production was significantly associated with lower LDL cholesterol, lower triglycerides, improved insulin sensitivity, lower systemic inflammation, and lower blood pressure.

“The predictive accuracy of MCMMs in vitro, coupled with the significant associations between SCFA predictions and health markers in human cohorts, gives us confidence in the utility of these models for precision nutrition,” said lead author Dr. Nick Quinn-Bohmann, a University of Washington graduate student at ISB who recently defended his dissertation.

After validating MCMM predictions in a variety of ways, the authors then demonstrated the potential of this approach for designing personalized prebiotic, probiotic, and dietary interventions that optimize SCFA production profiles. They simulated butyrate production rates for two different diets – the standard Austrian diet (i.e., standard European diet) and a vegan high-fiber diet – across a cohort of over 2,000 individuals from the Pacific West of the US. They found that a small subset of individuals showed almost no increase in butyrate production when switched to the high-fiber diet (termed “non-responders”) and another subset actually saw a small drop in butyrate production on the high-fiber diet (termed “regressors”). Next, they simulated three simple co-interventions on both background diets to try and augment butyrate production in the non-responders and the regressors: adding the prebiotic fiber inulin, adding the prebiotic fiber pectin, or adding a butyrate-producing probiotic (Faecalibacterium). The results showed that no single combinatorial intervention was optimal across all individuals: some benefited most from adding a prebiotic fiber, while others appeared to require the addition of a butyrate-producing probiotic to their microbiota.

“Together, these results represent an important proof of concept for a novel path forward in microbiome-mediated precision nutrition,” said Dr. Christian Diener, co-senior author and assistant professor at the Medical University of Graz in Austria. “But, of course, there is more work to do to validate the predictive capacity of these models in prospective human trials before they can enter clinical practice.”

Video Transcript

Below is the video transcript of the conversation between Drs. Sean Gibbons and Nick Quinn-Bohmann:

Sean Gibbons:

Hello, I’m Sean Gibbons, an Associate Professor at the Institute for Systems Biology, and our lab studies the human microbiome and how it affects human health and disease.

Today, I’m here to talk about our latest paper, which is about to come out online in the Journal of Nature Microbiology. This work was driven by Dr. Nick Quinn-Bohmann, a PhD student in the lab, who recently graduated and got his PhD, and also with my co-senior author Christian Diener, who’s now Assistant Professor at the Medical University of Graz in Austria.

So to give a little background on this paper, which is titled “Microbial Community Scale Metabolic Modeling Predicts Personalized Short-Chain Fatty Acid Production Profiles in the Human Gut,” it’s a bit of a mouthful, but basically it’s an implementation of our metabolic modeling tool set that we’ve been building for the last several years, pioneered by Dr. Christian Diener, who I mentioned earlier. It’s a method for simulating the metabolic outputs of an entire microbial community given a particular dietary input to a human host.

And in this particular paper, Nick is trying to point this modeling framework at a really important question; you feed different people the same exact food, and you get different things coming out the other end. There’s a lot of heterogeneity in how people respond to diet, and he focused on a particular set of molecules.

So with all that as background, I thought I’d turn it over to Nick to give us an overview of the paper, why this particular problem was chosen and why short-chain fatty acids. Nick.

Nick Quinn-Bohmann:

Thanks, Sean, for that great introduction.

Yeah, as you alluded to for this project, we were really looking to determine whether or not we could use this modeling platform to predict functional metabolic outputs from the gut microbiome on an individual specific level.

In this case, we were specifically looking at short-chain fatty acids, or SCFAs, which are a set of molecules produced by the microbiome that have a number of really important functions in the host. They help to keep the lining of our gut healthy. They help to improve insulin resistance, preventing the onset of pre-diabetes and diabetes. They’re important for the gut-brain access, helping to keep our brain healthy and prevent neurodegeneration.

And finally, but very importantly, they’re also important in modulating immune response. So, through the modulation of different immune cells, they can actually tamp down an inflammatory immune response and bring the body back into a sort of less inflammatory state. This is important for preventing the onset of certain diseases like inflammatory bowel and Crohn’s disease, certain cancers and other cardiometabolic diseases.

In this case, we were really interested in determining whether, on an individual-specific level, our model was able to predict outcomes since, as Sean alluded to, given the exact same dietary input, the gut microbiome and differences in the gut microbiome between individuals, leads to very different outcomes in terms of the production of short-chain fatty acids as well as many other metabolites.

Sean Gibbons:

The short-chain fatty acids are kind of ephemeral in the body, right? They’re produced quickly, but they’re also consumed quickly. It makes them very hard to measure. So, at the end of the day, how did you go about validating that the model predictions were correct?

Nick Quinn-Bohmann:

That’s a great question, and it’s one of the primary challenges that we faced when constructing this project.

In the body, it’s very difficult to measure SCFAs, and that’s because as soon as they’re produced by the microbiome, they’re just as quickly consumed by the cells in our colon. So not very high levels of these SCFAs actually reaches circulation where they can be measured. So to validate the predictions of our model, we actually turned to communities outside the body. So, either in vitro cultures of lab strains put together, or cultures of fecal samples collected directly from study participants that were then cultured anaerobically.

These model communities allowed us two things. We could measure directly the composition of the microbiome, so which microbes are present in each community, as well as, without interference from the host, measure the production of short-chain fatty acids. So, without the cells in our colon quickly consuming these metabolites, we could measure how much of each metabolite was produced by each individual community.

Sean Gibbons:

We show in the paper that one can predict these short-chain fatty acids relatively accurately, not perfectly, but relatively accurately.

So then, once you have this capacity to make these personalized dietary output predictions based on the microbiome, what do you do with that? What kind of phenotypes is that affecting in the host, and how does one go about validating those effects? What did you do in the paper?

Nick Quinn-Bohmann:

We were very interested in exactly that. After we showed that our predictions were relatively accurate on an individual-specific level, we next turned to a cohort where we had thousands of different individuals.

What we saw after building models for each of these individuals and simulating their growth, we saw that predicted levels of butyrate specifically showed very strong negative correlations with several markers for poor cardiometabolic health. So, essentially, individuals that had poor cardiometabolic health were predicted to produce less butyrate. Some of these markers included C-reactive protein, which is a measure of systemic inflammation, LDL cholesterol, which is often referred to as bad cholesterol that can lead to the onset of cardiovascular disease, measures of insulin resistance that can lead to pre-diabetes and diabetes, as well as several others.

So, not only did we see a quantitative agreement between our predictions and measurements of SCFA production, in this case, we also saw a qualitative agreement between predicted levels of butyrate production and the physiological effect that we would expect butyrate to have in the body.

Sean Gibbons:

That’s great. There were limitations to the way we were able to implement these models. So I thought maybe we’d briefly brush on maybe one of the more important limitations, which is the diet.

What choices did you make in the current paper and how you model diet, and what might we do in the future if we wanted to make these models perhaps more predictive?

Nick Quinn-Bohmann:

Absolutely. One of the primary inputs into these models is the diet, which constrains the model on which dietary precursors that we can use for growth. The difficulty here is that we didn’t have specific data on what each individual in our study was eating. So in this case, we had to sort of assume a standard European diet for each individual and apply that broadly across all individuals, which isn’t necessarily relevant, since every person eats something slightly different.

Excitingly, other work out of our lab has been working toward fixing this problem. Essentially, it uses metagenomic data to map to a database of known foods, which we can then work backwards and use that to build an individual specific medium of what that person was eating and apply that to our metabolic models to help improve the predictions that come out of our modeling framework.

Sean Gibbons:

All right, so that’s one way we can make these models better.

So, assuming that the modeling is working pretty well, and we have this capacity to simulate dietary inputs to a given person’s model and see how they respond, in particular, how do you envision the implementation of these models and the translation of these models into the real world? Where are we going with all of this?

Nick Quinn-Bohmann:

I think that’s the most exciting part of this work is that it’s ongoing, and it has a lot of promise.

One of the things that we see moving forward, one of the benefits of this modeling framework is that it allows us to test all sorts of different interventions, whether they be complete dietary switches or the addition of prebiotic fibers or probiotics, putatively healthy bacteria that we can add to an individual microbiome. It allows us to test all of these individually as well as in combination. What we showed in the paper is that we can use this framework to find an optimal intervention aimed at increasing SCFA production for an individual person given their unique compositional context.

Moving forward, I think this could serve as a powerful tool in precision nutrition, building optimal interventions aimed at improving metabolic outcomes from the gut microbiome for each individual person.

Sean Gibbons:

That’s great. Thanks Nick.

I think this is an exciting time in the microbiome field where we can start to map this ecological composition to these functional outputs. And you mentioned precision nutrition or personalized nutrition. So, obviously, this model has applications there, but we might actually even say precision medicine in the sense that butyrate as a molecule is as potent as many drugs that are on the market in terms of improving your cardiometabolic health and reducing inflammation. So, hopefully, in this next few years, we’re proving Hippocrates right, that food is medicine. It’s just not everyone needs the same medicine, that different foods apply to different people to get the same drug output, and that’s where we’re going with all this.

All right, well yeah, I’m excited for this paper to come out, and great work and congrats.

Nick Quinn-Bohmann:

Thank you so much, Sean. I’m very excited.