Battling obesity in industrialized nations and mitigating malnutrition in impoverished ones; delivering diagnoses to patients afflicted with idiopathic infections; administering antibiotics to treat disease in a targeted manner; providing personalized probiotics to promote health and prevent disease — these are but a few of the hopes for applications of human microbiome research.
Unraveling the metagenomics of the human microbiome in an effort to understand its roles in health and disease is a large undertaking. Understanding the compositional and functional dynamics of the human microbiome to the point of clinical application — to promote health as well as to treat, and perhaps even prevent, disease — will be an even larger feat. It's a goal that a growing group of researchers and clinicians is taking great interest in.
And for good reason, according to Dusko Ehrlich at the Institut National de la Recherche Agronomique in Jouy-en-Josas, France. "Our other genome — that is, the collective genome of the microbiome — appears to be much more variable than our own genome," he says. While the tremendous diversity of the microbiome across body sites, individuals, and populations has left researchers largely perplexed as to its functional interpretation, the human microbiome has shown promise for clinical applications. The complexity of the human microbiome, Ehrlich says, makes it "seem to be a much more interesting target to develop powerful new diagnostic and prognostic tools" against than the human genome.
With an eye toward clinical applications, government and nonprofit research funding organizations have invested heavily in human microbiome projects undertaken by increasingly large international consortia. The majority of these collaborations have characterized the composition of the human microbiome at various sites. In March 2010, a team led by investigators at BGI in Shenzhen, China, published the largest human metagenomic sequencing study to date in Nature. Working in collaboration with members of the MetaHIT Consortium — the €11.4 million initiative Ehrlich directs that aims to assess the metagenomics of the human intestinal tract — the BGI-led team generated a catalog of 3.3 million human microbial genes derived from fecal samples from 124 European individuals. That gene set, the team reported, is about 150 times larger than of human gene complement.
Recently, however, there has been a shift in the human microbiome research agenda. Beyond "who's there" in terms of microbial community composition, researchers are now interested in "what do they do?" and, increasingly, how they might best be used to improve patient care.
First form, now function
In their early phases, initiatives like the National Institutes of Health's five-year, $157 million Human Microbiome Project, or HMP, put fleets of researchers — supported by growing stores of sequencing capabilities — to the task of generating a reference set of genome sequences from the scores of bacteria and other microbes that inhabit the GI tract, the urogenital tracts, the oral cavity, the nasopharyngeal tract, and the skin, among other sites.
But despite having invested nearly four years scouring the human microbiome, "people are still puzzled because most studies, including ours, in the beginning are very descriptive in terms of who's there, but they don't really answer the question as to what they do," says the University of Maryland's Jacques Ravel. "The big drive should be on function."
Ravel, who is funded as part of the HMP to evaluate the microbial ecology and metagenomics of bacterial vaginosis, adds that in order to properly diagnose and effectively treat a disease, it is imperative to understand its etiology. In general, he says, "people study bacterial vagin-osis by looking at a woman who has just experienced it." However, just like trying to determine the cause of a plane crash by examining only the wreckage, trying to understand the origins of the disease by sampling the vaginal microbiome only once it has shifted into a vaginosis state is inadequate. After a crash, investigators generally go straight to the plane's black box, Ravel says, as it contains a record of what happened just before the crash occurred. For his team's bacterial vaginosis project, "our prospective, longitudinal design is really built to capture every single daily event. We get the sample, but we also get the metadata … anything that's happening in the daily lives of those subjects. Every single day we get this information — up to the event, through the event, and after the event," he says. "I think that is the only way we're going to be able to really understand the role of the microbiota in health and disease."
Because the human microbiome is constantly changing, cross-sectional sampling — used in nearly all of the early characterization projects — cannot adequately power functional analyses to disentangle cause from effect. Sampling at a single time point "is really limiting because what a subject is today might be very different from what it's going to be tomorrow and what it was yesterday," Ravel says. For the bacterial vaginosis study's demonstration phase, Ravel and his team sampled 33 reproductive-age women, from whom they also collected relevant metadata, twice weekly for 16 weeks. Throughout this pilot phase, the team observed rapid changes in vaginal microbiota communities that correlated with the elevation of estrogen levels during certain stages of the menstrual cycle. Ravel is quick to point out, though, that with only 33 subjects, any associations his team found in its preliminary investigation are not statistically significant. In a forthcoming trial, Ravel's team will sample 200 women daily over the course of 10 weeks. "Our essential hypothesis is that the composition, the structure, the function, and the dynamics of the microbiota are really what leads to a higher risk of bacterial vaginosis or even the acquisition of different STIs," he says, referring to sexually transmitted infections. "We're now looking at gene function in the context of the community and the host."
Longitudinal sampling could be especially informative for episodic conditions like bacterial vaginosis, for which researchers have established the functions of key associated bacteria — in this case, Lactobacillus acidophilus. Having a clear picture of when Lactobacilli dominate the microbiome could help researchers identify what causes the shift, and it could also inform diagnostic tests to detect vaginosis-associated shifts.
For this and other reasons, the University of British Columbia's Deborah Money says that the path to the clinic may be clearer for vaginal microbiome researchers than for their colleagues who study other body sites.
"Our advantage is that the vaginal microbiome is probably a bit simpler than the microbiome in other areas, like the gut, so we may have a little bit of a head start in understanding the complexity there and perhaps will be able to get into genomics-based diagnostics sooner," Money says. Compared with other sites, the microbial communities that inhabit the vagina are "a little less diverse — the depth of organism range and functionality, we think, is a bit simpler," she adds.
In fact, Maryland's Ravel says that, so far as he is aware, "there's no other body site where people are actually able to group subjects based on their microbiota composition; it's something that's quite unique to the vaginal microbiota." Based on the relative abundance of certain species, "we can actually assign women now, very commonly, to what we call a community state type — a state of the community [composition] which is similar to others'," he adds.
As the vagina is dominated by a small number of genera, particularly by Lactobacilli, Money and her colleagues are using the CPN60 gene as the primary target of their metagenomic sequencing studies, rather than the 16S rRNA marker that is commonly used for gut-derived samples, as it offers more sensitive and specific detection of predominant vaginal microbiota. "We really need species and sub-species differentiation capabilities," says Money, who directs the Canadian Institutes of Health Research and Genome British Columbia's five year, C$2.3 million Vaginal Microbiome Project.
Still, "the biggest issue with coming up with diagnostics is that everybody's different," says Patrick Gillevet at George Mason University. Like their colleagues investigating the vaginal microbiome, researchers studying the GI tract have made significant strides in detecting diseases linked to distinct community shifts. For episodic diseases, like Crohn's, Gillevet says that because researchers have already identified key species and have garnered "a fairly good idea of their functionality, because we've done experiments on them," diagnostics are likely not far off.
Using length-heterogeneity fingerprinting and multi-tag pyrosequencing — or barcoding based on 16S rRNA and other markers — Gillevet and his collaborator Ali Keshavarzian at Rush University Medical Center have been comparing samples from inflammatory bowel disease patients with those from healthy controls for nearly 10 years. "We have accumulated a fair bit of data and we think we actually have [identified] patterns in the microbiome that can distinguish, using bioinformatics tools, inflammatory bowel disease from normal," Gillevet says. He adds that he is currently writing an SBIR Phase II grant application on behalf of Metabiomics, the firm he co-founded in 2000, in the hope that the company can "get some more funds to actually do enough samples to validate these computational tools."
But not every group has had the advantage of having functional data. For the Bascom Palmer Eye Institute's Valery Shestopalov, investigating the metagenomics of the ocular surface in an effort to better diagnose microbial eye infections like keratitis is beginning to feel like a game of catch-up.
"We have a little bit of a different situation in the eye compared to what we have in the gut," Shestopalov says. The ocular surface microbiome "is underappreciated hugely. ... It's one of those that's been overlooked by far."
Perhaps one reason for this site-specific disparity is that, when compared with sampling the vagina or the GI tract, collecting enough sample to sequence from the surface of the eye is a delicate task. In order to keep pace with his fellow HMP researchers, Shestopalov has amassed a cohort of 100 healthy controls to sequence in order to build a reference catalog of ocular surface microbes, and has also designed a variety of plans for prospective, longitudinal pre-clinical trials to assess the microbial ecology of contact lenses and their storage cases. According to Shestopalov, up to 60 percent of eye infections remain undiagnosed as many candidate microbes are unculturable using standard media, and because PCR-based tests are only useful to identify bacteria that are well-characterized. "Our first goal is to characterize what we know at the community level, and the second — most importantly — is we need to have some understanding of what's going on on the dark side of the moon," Shestopalov says. Much like UBC's Money, Shestopalov envisions metagenomics-based diagnostic tests on the most immediate horizon. In the future, he also hopes to see the strategic use of antibiotics to treat microbial eye infections.
No matter at what stage they may be, there are a handful of universal roadblocks hindering researchers' attempts to harness the human microbiome for clinical applications. The University of Colorado's Rob Knight says the main challenge is that researchers still "don't really know what 'normal' looks like," which makes it particularly difficult to design clinical trials as they hinge upon carefully selected groups of controls. One underlying issue, which "even the largest microbiome studies" are not immune to, is sample size, he adds. "The HMP has spent $170 million, roughly, looking at 300 people. While it's useful as a start, we may find that we need 10,000 people to really get somewhere," Knight says.
The extent to which there is a "core" human microbiome — a complement of microbes shared by all — remains up for debate. "It's semantic game-playing," Knight says. "Depending on what you're trying to play up, either you're going to say, 'It's amazing how much diversity there is' — because anything that's common in one person, you'll be about to find some other person that lacks it — or you play up the essential similarities, 'There are a lot of species that are found in many people.'"
In the end, it all comes down to effect sizes. "We don't know the relative sizes of the effects of your geographical location, or your environmental exposure, or whether you were delivered by C-section or vaginally," he says. To determine these, and what shared percentage of core microbes constitutes 'normal,' exponentially more data will be required.
Working in collaboration with Harvard University's George Church, Knight's group is adding a microbiome component to the Personal Genome Project. His team plans to sequence the microbiomes of the first 1,000 PGP participants who provide all of their clinical information. While the HMP is assessing the -microbes of healthy people, with no clinical symptoms, Knight says the PGP presents a fairly unique opportunity. "In the PGP, we're taking everybody," he says, regardless of their health status. "The great thing about that it we'll have all sorts of different clinical phenotypes and we'll be able to go trolling for the really strong associations."
Going forward, while researchers interrogate the microbiomes of increasingly diverse populations, "one thing that's going to be really fascinating is compiling enough data over the next, say, three to 10 years [so] that we can start fitting, in a good way, the effect size of different phenomena that affect the microbiome," he says. "Being able to figure out what the effect sizes are is what's going to let us realize when our studies are adequately powered, when we've recruited enough people."
In the meantime, Knight says animal model studies are essential. Washington University in St. Louis' Jeffrey Gordon, Knight's long-time collaborator, published a PNAS paper in March in which he and his colleagues showed they could transplant microbes derived from human stool samples into humanized, germ-free mice such that the animal's gut flora composition recapitulated much of the donor's. Using high-throughput sequencing technologies on these personalized culture collections in humanized mice, Gordon, Knight, and their collaborators hope to "be able to culture, say 1,000 different organisms out of a person, sequence all their genomes, and then just put them into mice into different combinations — maybe leaving one out every time — and just asking, definitively, what is the function of each," Knight says.
While it can be "really hard to get rid of one bug from a human subject, if you have a thousand of them in pure culture, it's really easy to put 999 into a mouse and leave out the one that you don't want and then just see what the effect of each microbe is in the context of that consortium," he adds.
It's that level of precision that Curtis Huttenhower at the Harvard School of Public Health is applying at the molecular level to deduce the signaling interactions that occur between the microbiome and the host genome. In parsing HMP data and applying computational tools to characterize biomolecular functions from meta-genomic sequences, Huttenhower says that microbial function seems to be more constant from person to person than microbial community membership is. "There are bugs performing the same processes in different communities," he says. "But those processes are fairly consistent, particularly from person to person, even though the specific microbial members vary tremendously. So, there's consistency at the metabolic level that's not present at the organismal level, which gives hope for things like drug-targeting."
By applying computational tools to functional and metagenomic data sets, Huttenhower hopes to ascertain the extent to which the microbiome and its host communicate with one another using small molecules as well as whether direct, physical protein-protein interactions occur between them. "There are definitely individual proteins in microbes in the microbiome that might be potential drug targets," he says. "And there may well be drug targets on the host side."
While much computational work remains to be done, Huttenhower says it will be well worth the effort. "A lot of the areas that have been highlights for potential clinical applications revolve around probiotics and antibiotics," he says. The problem with both is that they impose systematic effects. "It's perhaps the most complicated part of the biology and where we have the least understanding at this point. Both of those kinds of pharmaceutical interventions ... perturb an entire community of hundreds of different types of organisms," he says. "Understanding what's going on in the initial perturbation and in the recovery involves understanding what's going on among hundreds of different organisms."
Rather than focusing on probiotics, antibiotics, or prebiotics, Huttenhower says researchers should instead set their sights on "limiting the range of potential complications from hundreds of different organisms to single points of intervention" by determining "the biomolecular functions, either on the microbe's or the host's side, that might be potential points for characterization or drugability."
In the nearer future, Huttenhower envisions the development of biomarker-based diagnostics. "That's something I think we're actually fairly close to," he says. "Microbial community membership will be a good biomarker in the short term. ... Basically, any condition that potentially touches the microbiome tends to have a strong impact on membership and community structure that can pretty straightforwardly be used as a diagnostic tool."
But there are still gaps in the research on the basic end of the spectrum, not least, Maryland's Ravel says, is that investigators still know little as to how the microbiome develops. To that end, he adds, researchers require access to samples across the human lifespan, so as to adequately characterize the formation of the microbiome. "Once we can understand where it starts, where it ends, and how it ends, then I think we can start better controlling the ... microbiota," Ravel says.
Patients show support
While microbiome-based diagnostics and therapeutics are not yet ready for prime time, researchers have already begun to assess patients' perceptions of them. Equipped with HMP funding, a team led by the Cleveland Clinic's Richard Sharp has held, to date, 22 focus group sessions with 136 patients with GI tract disorders. During each two-hour session, the researchers "frame a number of different possibilities" in which metagenomics might be applied to improve clinical care, in particular, those pertaining to genetically modified probiotics and metagenomics-based diagnostics. Overall, Sharp says his group aims to ascertain whether "patients will be excited about these [potential clinical applications] or express some of the concerns that, in the past, have been associated with new types of genetic testing."
Thus far, in general, "we've found that patients are very supportive of this type of research," Sharp says. For many patients, "it's been a difficult road for them to find a diagnosis, so anything that might help to shorten that path and make it easier ... they're very excited about," he adds. "They're very optimistic about this and, especially given the limited range of therapeutic interventions that are available, they recognize the need for more studies. They're willing to enroll."
Harvard's Huttenhower says there is a similar sense of camaraderie among microbiome researchers. "I've been really struck by the amount that we're learning quickly and by the degree of capacity-building and community-building that's been going on," he says. "It's a very exciting project and there's a tremendous amount more work to be done and ... I'm glad the field is still growing. I think it's a great opportunity."