The microbiome has been big news in recent years, as scientists have sought to unravel the connection between changes in bacterial populations within our bodies and increases in conditions ranging from rheumatoid arthritis to food allergies to asthma to autism. In a study published in Cell, Host & Microbe earlier this year, the largest longitudinal study of the microbiome to date, researchers saw dramatic shifts in the gut bacteria of children who went on to develop type 1 diabetes. The study could have important implications for identifying the disease at a very early stage, delaying its onset or even preventing it altogether. Dr. Julia Greenstein, vice president of discovery research at JDRF, which helped fund the study, says that the work is very preliminary, but points toward strategies for prevention that haven’t yet been explored. “We are trying to understand the observations and think through the therapeutic implications. It’s going to take some time to see the treatment avenues. But we are looking at interventions to see if we can alter the microbiome to keep it in a healthy state, maybe vaccination strategies, nutrition strategies, or prebiotics.”
Earlier this year, I spoke with the lead author of the study, Aleksandar Kostic, a postdoctoral fellow at the Broad Institute of MIT and Harvard, who uses powerful computational approaches to study microbial populations.
How did you get started in this field?
I did my PhD at Harvard and also at the Broad Institute in cancer genomics, and I got interested in pathogens and the roles they play in cancer. This was in 2009-2010, when high throughput sequencing [the process of determining the order of nucleotides within DNA] was something very novel that the Broad had. It became possible for the first time to sequence large numbers of whole genomes. We were sequencing tumors, and comparing them to cancer-free tissues, trying to figure out mutations relevant to cancer. We saw an association between colon cancer and a bacteria called fusobacteria that was very highly abundant on colon tumors but that you wouldn’t see at all in healthy people. We followed this through and described a mechanism that links this bacteria with colon cancer.
When I was graduating and finishing my PhD I wanted to take a step back and systematically characterize the role of the microbiome not just in cancer but in other diseases. I wanted to try to understand the mechanisms of how the microbiome is linked to disease. That brought me to work with Ramnik Xavier, who is a gastroenterologist and longtime researcher in the realms of immunology and inflammatory bowel disease. Type 1 diabetes is related in the sense that it’s a disease that’s spurred on by the immune system just like IBD. The unique thing about it is that it’s manifested extra-intestinally. So the connection to many people at the outset isn’t clear—how does the gut microbiome link to diabetes, this disease of the pancreas? The thing that really convinced me this was worth looking at was the mouse model, the NOD or non-obese diabetic mouse, which is a model of T1D. There were mouse studies showing that changing the gut microbiome has a large impact on the progression of this disease. In these studies, if you take a mouse that is genetically predisposed to developing type 1 diabetes and delete a single gene, MyD88, that is very important in recognizing bacteria, suddenly that mouse is completely protected from diabetes—one hundred percent. So it means there’s some kind of interesting relationship going on between microbial sensing and this gene.
Do humans have an analogue for that gene?
Yes, the same exact gene.
But of course in humans you can’t just go in and delete a gene.
Yes, that’s right, and that’s one of the big issues. But the great thing is that with the microbiome you can go in and change things. It’s very easy in mice right now, and the field will move quickly towards therapies in humans as well, I think. In mice you can take the gut microbiota, which just means stool pellets, from these MyD88-deficient mice, and you can give it to a regular NOD mouse with its MyD88 intact. Normally this mouse develops severe diabetes, but in this case it’s completely protected, just simply by stool transfer. This means that there’s something about that microbial composition that prevents those mice from developing the disease. No mechanism has been worked out—how this works is not yet understood.
How did you come to do this study on a group of children in Finland and Estonia?
In the lab, we have a strong connection with a group in Finland called DIABIMMUNE, headed by a physician-scientist named Mikael Knip. They do phenomenal work collecting these giant cohorts of children from Finland as well as neighboring countries, so we can start to understand more about how the disease is seeded in these children. They don’t just study the microbiome; they collect information on many other aspects of lifestyle and genetics from these children. All the children who are in the DIABIMMUNE cohort are recruited because they have high risk alleles that predispose them to type 1 and other autoimmune diseases. They’re HLA-typed for being in the highest risk group, which really isn’t all that meaningful because I think that only something like five percent of these children go on to develop diabetes. So what that means is, in order to actually get a group of children who will eventually develop the disease, the birth cohort needs to be giant. So this was really hundreds of kids who were included, and all of them had to have systematic stool sampling and blood sampling and that sort of thing. They were followed over a span of three years. Over that time a number of them did seroconvert.
That’s when the antibodies show up?
Yes. And then a very small number of them actually got diagnosed with the disease during the course of the study.
A small number of those who seroconverted?
Yes. If you are considered seropositive, meaning you have at least two of the autoantibodies, then there’s an eighty-five percent chance you’ll develop T1D by the time you’re fifteen years old. So that’s a pretty good indicator. But when that happens is highly variable. The fact that some of the children develop T1D before they’re three years old is very rare, so it’s almost like a unique form of type 1 diabetes. In the end we only had four children who developed the full blown disease, out of 33 that we sequenced and the hundreds of children who we’ve collected samples from.
What differences did you see between the groups?
From the outset, the way the experiment was designed, we expected to see strong differences in the microbiomes of those children who seroconverted. But we didn’t see that. For the most part, there were no major changes in the microbiome that made the seroconverted children unique from the children who didn’t seroconvert. The strongest differences by far were in those children who were diagnosed with disease; and there we saw very drastic differences in their microbiomes that occurred about a year before they were diagnosed. The strongest difference was the diversity of their microbiome—it was much less diverse, meaning it consisted of many fewer types of bacteria. The reason why this might be important is that one of the things a low-diversity community allows for is the invasion of other species, because it’s less fortified. One of the things we saw in this community with the lower overall diversity was an expansion of certain species that we call pathobionts—pathogens such as Streptococcus and Rikenellaceae, which have been shown to induce inflammation in the gut. In other words, in this study we saw a large difference in the composition of the gut microbial community in the children who develop the disease prior to their diagnosis, and we saw other signs of intestinal inflammation that we believe are linked to a systemic immune effect that eventually impacts beta cells in the pancreas.
Do you have a sense of cause and effect? Are the microbiome changes causing the conversion to type 1 diabetes, or is the disease causing the changes in the microbiome?
It’s a very good question, and no we don’t. That’s a question I’ve struggled with since I started my PhD. Establishing causation in the microbiome is very difficult. By definition you can’t do it in patients without interventions, and interventions are not studies you can easily do.
Because you can’t introduce a bacteria then wait to see if it causes a problem?
Yes. In type 1 diabetes we suspect that there are some changes that are happening in the immune system that are really driving things and not only driving the disease but causing these changes in the microbiome. So the changes in the microbiome are only secondary. What we think it will take is not just mouse studies—that is what we’re planning and what our next steps are going to be—but more mechanistic details about what the microbes are producing chemically that might be affecting the immune system of the host and might be driving them towards disease. This question about causality is very valid, and it’s at the center of all the experiments we are planning.
How will you conduct the experiments?
A lot of it is trying to drill into the mechanisms. One of the things that’s wonderful and at the same time not so wonderful about the microbiome field as it stands right now is that it’s really an ecology field: we talk about communities of organisms in this abstract way and the shifts that happen to them and how they’re associated with disease. But I want to understand the mechanism and how it’s tied in to immunology, so what I’m doing is taking these bugs that we call pathobionts and I’m isolating them directly from the patients and growing them in the lab and making biochemical fractions from them, which means I’m extracting pools of chemicals from them that they produce. Then I test them in a systematic way with immune assays in the lab to try to find something that can link the chemicals to an immune effect that’s important in type 1 diabetes. So I’m taking a step back from the clinical samples themselves in trying to delve into the details of the mechanisms. What do the bugs produce that talk to the immune cells?
So you are trying to understand the autoimmune attack that may be connected to the microbiome?
Yes. I want to know exactly what the microbiome is producing that is causing some changes in the immune system. Right now that detail is very foggy. But we’re specifically looking for microbial compounds that will shift the immune system, so if we, let’s say, find a molecule produced by the Streptococcus bacteria that causes a shift in the innate immune system, then this is something that adds to that pile of evidence that points to the bacteria changing the immune system. I think it’s a matter of finding those mechanisms, and as they accumulate, understanding that link. I think that’s the key to getting at causality in disease.
I was interested in what you were saying about Streptococcus being a pathogen, since that’s something they test women for when they’re pregnant. Is there any sort of association between positivity in the mother and the child’s likelihood of developing type 1?
That’s definitely a testable hypothesis, but we don’t have the numbers sufficient to test it in our cohort because there are only four cases of type 1. But I think it’s a valid hypothesis, the idea being that Streptococcus can be a pathogen in many different circumstances: orally, gastrointestinally, vaginally. It’s completely conceivable that a mother colonizes her infant at birth with Streptococcus and might predispose it to a disease like type 1.
I was also wondering about antibiotics: was that one of the variables that you tested in the infants? Whether they had had antibiotics?
We definitely included the use of antibiotics in all of the statistics that we did. Anytime we saw a change in the microbiome associated with a shift in autoantibodies or with diagnosis of disease, we always used a statistical model that also included whether those children took any antibiotics around that time or not. Obviously antibiotics have a large effect on the composition of the microbiome, but that’s the extent of what we saw. To address what is the effect of antibiotic usage on type 1 incidence requires much larger numbers because there’s large variability in how people’s microbiome responds to an antibiotic. I don’t think anyone has done that study systematically yet, thoroughly looking at the microbiome. But I have seen older epidemiological studies that associate higher antibiotic usage with a slightly increased rate of type 1 diabetes.
It kind of makes sense. Because of the effect antibiotics have on the microbiome.
Yes, for one thing because antibiotic use lowers the diversity of the microbiome, so if we’re seeing that as potentially something that is causally associated with the disease, then that could be an important link.
In the paper, you talked about the difficulty of developing a cohort to study. What were the difficulties?
Part of the reason is, especially in the U.S., it’s not the way science is done, unfortunately. The way that science is funded is by these short cycles of grants, maybe 3-5 years, for relatively small amounts of money. For a study like this one, just collecting all the samples takes five years, and that’s not even counting the planning, the analysis, or the sequencing. I think that in Finland you can see with this DIABIMMUNE consortium that the funding mechanisms tend to be a little more tailored for studies with long-term cohorts like this one. You think about it in terms of the career lifespan of any trainee scientist such as myself: I could never be as involved in creating a cohort for something like this because it involves too many years. Eventually I have to publish papers and get a job. That’s why we really depend on clinicians who are funded for this purpose of creating cohorts and setting up other scientists to be able to analyze them. It’s unfortunate that it is so rare, but it really is.
Stay tuned for part 2.