CHAPTER 3: THE HUMAN MICROBIOME

The technologies that allowed us to decipher the human genome have revolutionized our ability to delineate the composition and functions of the microbial communities that colonize our bodies and make up our microbiota. Each body habitat, including the skin, nose, mouth, airways, gastrointestinal tract, and vagina, harbors a distinctive community of microbes. Efforts to understand our microbiota and its collection of microbial genes (our microbiome) are changing our views of “self” and deepening our understanding of many normal physiologic, metabolic, and immunologic features and their interpersonal and intrapersonal variations. In addition, this area of research is beginning to provide new insights into diseases not previously known to have microbial “contributors” and is suggesting new strategies for treatment and prevention. Key terms used in the discussion of the human microbiome are defined in Table 3-1.

We are holobionts—collections of human and microbial cells that function together in an elaborate symbiosis. The aggregate number of microbial cells in our microbiota exceeds the number of human cells in our adult bodies by up to 10-fold, and each healthy adult is estimated to harbor 10⁵–10⁶ microbial genes, in contrast to ~20,000 Homo sapiens genes. Members of our microbiota can function as mutualists (i.e., both host and microbe benefit from each other’s presence), as commensals (one partner benefits; the other is seemingly unaffected), and as potential or overt pathogens (one partner benefits; the other is harmed). Many clinicians view pathogens as individual microbial species or strains that can elicit disease in susceptible hosts. An emerging, more ecologic view is that pathogens do not function in isolation; rather, their invasion, emergence, and effects on the host reflect interactions with other members of a microbiota. An even more expansive view is that multiple organisms in a community conspire to produce pathogenic effects in certain host and environmental contexts (a pathologic community).

The ability to characterize microbial communities without culturing their component members has spawned the field of metagenomics (Table 3-1). Metagenomics reflects a confluence of experimental and computational advances in the genome sciences as well as a more ecologic understanding of medical microbiology, according to which the functions of a given microbe and its impact on human biology depend on the context of other microbes in the same community. Traditional microbiology relies on culturing individual microbes, but metagenomics skips this step, instead sequencing DNA isolated directly from a given microbial community. The resulting datasets facilitate follow-up functional studies, such as the profiling of RNA and protein products expressed from the microbiome or the characterization of a microbial community’s metabolic activities.

Metagenomics provides insight into how microbial communities vary in several situations critical to human health. One such situation is how microbial communities are assembled following birth and how they operate over time, including responses of established communities to various perturbations. Another is how microbial communities normally vary between different anatomic sites within an individual and between different groups of people representing different ages, physiologic states, lifestyles, geographies, and gender. Yet another is how microbial communities vary in disease; whether such variations are consistent among individuals grouped according to current criteria for a disease or its subtypes; whether the microbiota or microbiome provides new ways of classifying disease states; and, importantly, whether the structural and functional configurations of microbial communities are a cause or a consequence of disease.

Analysis of our microbiomes also addresses one of the most fundamental questions in genetics: How does environment select our genes and directly influence their function? Each human encounters a unique environment during the course of his or her lifetime. Part of this personally experienced environment is incorporated into the genes and capabilities of our microbial communities. The microbiome therefore expands our conceptualization of “human” genetic potential from a single set of genes “fixed” at birth to a microbiome with additional genes and capabilities acquired via a process influenced by our family and life experiences, including modifiable lifestyle choices such as diet. This view recognizes a previously underappreciated dimension of human evolution that occurs at the level of our microbiomes and inspires us to determine how—and how fast—this microbial evolution effects changes in our human biology. For example, Westernization is associated with loss of bacterial species diversity (richness) in the microbiota, and this loss may be associated with the suite of Western diseases. The study of our microbiomes also raises important questions about personal identity, how we define the origins of health disparities, and privacy. Further, it offers the possibility of entirely new approaches to disease prevention and treatment, including regenerative medicine, which involves administration of microbial species (probiotics) to individuals harboring communities that have not developed into a mature, fully functional state or that have been perturbed in ways that can be restored by the addition of species that fill unoccupied “jobs” (niches).

This chapter provides a general overview of how human microbial communities are analyzed; reviews ecologic principles that guide our understanding of microbial communities in health and disease; summarizes recent studies that establish correlations and, in some cases, causal relationships between our microbiota/microbiomes and various diseases; and discusses challenges faced in the translation of these findings to new therapeutic interventions.

Life on Earth has been classified into three domains: Bacteria, Archaea, and Eukarya. The habitats of the surface-exposed human body harbor members of each domain plus their viruses. In large part, microbial diversity has not been characterized by culture-based approaches, partly because we do not know how to re-create the metabolic milieu fashioned by these communities in their native habitats and partly because a few organisms tend to outgrow the others. Culture-independent methods readily identify which organisms are present in a microbiota and their relative abundance. The gene widely used to identify microbes and their evolutionary relationships encodes the major RNA component of the small subunit (SSU) of ribosomes. Within each domain of life, the SSU gene is highly conserved, allowing the SSU gene sequences present in different organisms in that domain to be accurately aligned and regions of nucleotide sequence variation to be identified. Pairwise comparisons of SSU ribosomal RNA (rRNA) genes from different microbes allow construction of a phylogenetic tree that represents an evolutionary map on which previously unknown organisms can be assigned a position. This approach, known as molecular phylogenetics, permits characterization of each organism on the basis of its evolutionary distance from other organisms. Different phylogenetic types (phylotypes) can be viewed as comprising branches on an evolutionary tree.

Characterization of bacteria

Because members of the Bacteria dominate our microbiota, most studies defining our various body habitat–associated microbial communities have sequenced the bacterial SSU gene that encodes 16S rRNA. This gene has a mosaic structure, with highly conserved domains flanking more variable regions. The most straightforward way to identify bacterial taxonomic groups (taxa) in a given community is to sequence polymerase chain reaction (PCR) products (amplicons) generated from the 16S rRNA genes present in that community. PCR primers directed at the conserved regions of the gene yield PCR amplicons encompassing one or more of that gene’s nine variable regions. PCR primer design is critical: differential annealing with primer pairs designed to amplify different variable regions can lead to over- or underrepresentation of specific taxa, and different regions within the 16S rRNA gene can have different patterns of evolution. Therefore, caution must be exercised in comparisons of the relative abundance of taxa in samples characterized in different studies, as methodologic differences can lead to larger perceived differences in the inferred taxonomy than actually exist.

A key innovation is multiplex sequencing. Amplicons from each microbial-community DNA sample are tagged by incorporation of a unique oligonucleotide barcode into the PCR primer. Amplicons harboring these sample-specific barcodes can then be pooled together so that multiple samples representing multiple communities can be sequenced simultaneously (Fig. 3-1). One important choice is the tradeoff between the number of samples that can be processed simultaneously and the number of sequences generated per sample. Interpersonal differences in the bacterial components of the microbiota are typically large, as are differences between communities occupying different body habitats in the same individual (see below); thus fewer than 1000 16S rRNA reads are characteristically required to discriminate community type. However, the identification of systematic differences in microbiota composition that correlate with physiologic status or disease state is confounded by the substantial interpersonal variation that occurs normally.

Sequencing of bacterial 16S rRNA genes creates a challenge for medical microbiology: how to define the taxonomic groups present in a community in a systematic and informative manner, so that one community can be compared with and contrasted to another. Within each domain of life, microbes are classified in a hierarchy beginning with phylum (the broadest group) followed by class, order, family, genus, and species. To determine taxonomy, 16S rRNA sequences are aligned on the basis of their sequence similarity—a process known as picking operational taxonomic units (OTUs). Grouping of 16S rRNA sequences from a given variable region into “bins” that share ≥97% nucleotide sequence identity (97%ID OTUs) is a commonly accepted, albeit arbitrary, way to define a species.

Looking beyond the 16S rRNA gene, we find that different isolates (strains) of a given bacterial species have overlapping but not identical sets of genes in their genomes. The aggregate set of genes identified in all isolates (strains) of a given species-level phylotype represents its pan-genome. Most species are represented by multiple strains, sometimes with markedly different functions (for example, enteropathogenic versus commensal Escherichia coli). Differences in genome content among strains of a given species reflect differences in community membership as well as differences in the selective pressures these strains experience within and between habitats. Horizontal gene transfer among members of a microbiota—mediated by phage, plasmids, and other mechanisms—is a major contributor to this strain-level variation.

Strain-level diversity can be important in any consideration of how microbial communities differ between individuals and how these communities accommodate perturbations. For example, the great bacterial strain-level diversity that exists in the gut is thought to be one of the features that allows this microbiota, which occupies a constantly perfused ecosystem exposed to the complex and varying set of substances we ingest, to adapt to changing circumstances rather than depending on one strain to occupy a given niche important for proper community functioning. In ecologic studies of different environments, such as grasslands, forests, and reefs, increased diversity within a community increases its capacity to respond to disturbances and to restore itself (i.e., its resilience); the same is likely true of microbial ecosystems. When characterizing the mechanisms by which a given species produces an effect or effects on humans, it is important to consider the strain being tested; strain-level diversity has an impact on discovery and development efforts aimed at identifying next-generation probiotics that can be used therapeutically to promote health or treat disease.

Identification of archaeal and eukaryotic members

Surveys based on SSU rRNA gene sequencing have largely focused on Bacteria, yet the census of “who’s there” in human body habitat–associated communities must also include the other two domains of life: Archaea and Eukarya. Differences in the sequences of archaeal and bacterial 16S rRNA genes, first recognized by Carl Woese in 1977, allowed these two domains of life to be distinguished. The representation of Archaea in human microbial communities is less well defined than that of Bacteria, in part due to the difficulty in optimizing the design of PCR primers that specifically target conserved regions of archaeal (versus bacterial) 16S rRNA genes. Identifying archaeal members is important to our understanding of the functional properties of the microbiota. For example, a major challenge faced by microbial communities when breaking down polysaccharides (the most abundant biologic polymers on Earth) is the maintenance of redox balance in the setting of maximal energy production. Many microbial species have branched fermentation pathways that allow them to dispose of reducing equivalents (e.g., by the production of H₂, which is energetically efficient). However, there is a caveat: the hydrogen must be removed or it will inhibit reoxidation of pyridine nucleotides. Therefore, hydrogen-consuming (hydrogenotrophic) species are key to maximizing the energy-extracting capacity of primary fermenters.

In the human gut, hydrogenotrophs include a phylogenetically diverse group of bacterial acetogens, a more limited group of sulfate-reducing bacteria that generate hydrogen sulfide, and methane-producing archaeal organisms (methanogens) that can represent up to 10% of the anaerobes present in the feces of some humans. However, the degree of archaeal diversity in the gut microbiota of healthy individuals appears to be low.

Culture-independent surveys of eukaryotic diversity are also confounded by challenges related to the design of PCR primers that target the eukaryotic SSU gene (18S rRNA) as well as the internal transcribed spacer regions of rRNA operons. Metagenomic studies of healthy human adults living in countries with distinct cultural traditions and disparate geographic features and locations have revealed that the degree of eukaryotic diversity is lower than that of bacterial diversity. In the gut, which contains far more microbes than any other body habitat, the representation of fungi is significantly lower in individuals living in Westernized societies than in those living in non-Western societies. The most abundant fungal sequences belong to the phylum-level taxa Ascomycota and Microsporidia. The phyla Ascomycota and Basidiomycota appear to be mutually exclusive, and the presence of Candida in particular correlates with recent consumption of carbohydrates.

Elucidation of viral dynamics

Viruses are the most abundant biologic entity on Earth. Viral particles outnumber microbial cells by 10:1 in most environments. Humans are no exception in terms of viral colonization; our feces alone contain 10⁸–10⁹ viral particles per gram. Despite this abundance, many eukaryotic viral communities remain incompletely characterized, in part because the identification of viruses within metagenomic sequencing datasets is itself very challenging. Characterizing viral diversity requires different approaches: because no single gene is found in all viruses, no universal phylogenetic “barcode of life” equivalent to the SSU rRNA gene exists. One approach has been to selectively purify virus-like particles from community biospecimens, amplify the small amounts of DNA that are recovered, and randomly fragment the DNA and sequence the fragments (shotgun sequencing). The resulting sequences can be assembled into larger contigs whose function can be computationally predicted from homology to known genes, and the information obtained can be used to populate/expand nonredundant viral databases. These annotated nonredundant databases can then be used for more targeted mining of the rapidly expanding number of shotgun sequencing datasets generated from total-community DNA for known or putative DNA viruses.

Given the dominance of bacteria in the gut microbiota, it is not surprising that phages (viruses that infect bacteria) dominate the identifiable components of the gut’s DNA virome. Prophages are a manifestation of a so-called temperate viral–bacterial host dynamic, in which a phage is integrated into its host bacterium’s genome. This temperate dynamic provides a way to constantly refashion the genomes of bacterial species through horizontal gene transfer. Genes encoded by a prophage genome may expand the niche and fitness of their bacterial host, for example, by enabling the metabolism of previously inaccessible nutrient sources. Prophage integration can also protect the host strain from superinfection, “immunizing” the strain against infection by closely related phages. A temperate prophage life cycle allows the virus to expand in a 1:1 ratio with its bacterial host. If the integrated virus conveys increased fitness, the prevalence of the bacterial host and its phage will increase in the microbiota. Induction of a lytic cycle, where the prophage replicates and kills the host, may follow. Lytic cycles can cause high bacterial turnover. Lysis debris (e.g., components of capsules) can be used as nutrient sources by surviving bacteria; this change in the energy dynamic in a community is referred to as a phage shunt. A subpopulation of bacteria that undergoes lytic induction may sweep away other sensitive species present in the community, thus increasing the niche space available for survivors (i.e., those bacteria that already have an integrated prophage). Periodic induction of prophages leads to a “constant diversity dynamic” that helps maintain community structure and function.

Interest in viral communities has expanded in recent years, especially given a potentially therapeutic role for phages as an alternative or adjunct to antibiotics. Virome members have evolved elegant survival mechanisms that allow them to evade host defenses, diversify, and establish elaborate and mutually beneficial symbioses with their hosts. A number of recent studies have tried to adapt these mechanisms for therapeutic purposes (e.g., the use of synthetic phages to treat Pseudomonas aeruginosa infections in burn patients or in other settings). Phage therapy is not a new idea: Félix d’Herelle, co-discoverer of phages, recognized their potential medical applications nearly a century ago. However, only recently have our technologic capabilities and our knowledge of the human microbiota made phage therapy realistically attainable within our lifetimes.

At many levels, different people are very much alike: our genomes are >99% identical, and we have similar collections of human cells. However, our microbial communities differ drastically, both between people and between habitats within a single human body. The greatest variation (beta diversity, described below) is between body sites. For example, the difference between the microbial communities residing in a person’s mouth versus the same person’s gut is comparable to the difference in communities residing in soil versus seawater. Even within a body site, the differences among people are not subtle: gut, skin, and oral communities can all differ by 80–90%, even from the broad, bacterial species–level view. The English poet John Donne said that “no man is an island”; however, from a microbial perspective, each of us consists of not just one isolated island but rather a whole archipelago of distinct habitats that exchange microbes with one another and with the outside environment at some as yet undetermined level. Before we can discuss these differences and understand their relevance to human disease, it is important to understand some basic terms and ecologic principles.

Alpha diversity

Alpha diversity is defined as the effective number of species present in a given sample. Communities that are compositionally more diverse (i.e., have more OTUs) or that are phylogenetically more diverse are defined as having greater alpha diversity. Alpha diversity can be measured by plotting the number of different types of SSU rRNA sequences identified at a given phylogenetic level (species, genera, etc.) in a sample as a function of the number of SSU rRNA gene reads collected. The most commonly used metrics of alpha diversity are S_obs (the number of species observed in a given number of sequences), Chao1 (a measure based on the number of species observed only once), the Shannon index (a measure of the number of bits of information gained by revealing the identity of a randomly chosen member of the community), and phylogenetic diversity (a measure of the total branch length of a phylogenetic tree encompassing a sample). Diversity estimators are particularly sensitive to errors introduced during PCR and sequencing.

Beta diversity

Beta diversity refers to the differences between communities and can be defined with phylogenetic or nonphylogenetic distance measurements. UniFrac is a commonly used phylogenetic metric that compares the evolutionary history of different microbial communities, noting the degree to which any two communities share branch length on a tree of microbial life: the more similar communities are to each other, the more branch length they share (Fig. 3-1). UniFrac-based measurements of distances between communities can be visually represented with principal coordinates analysis or other geometric techniques that project a high-dimensional dataset down onto a small number of dimensions for a more approachable analysis (Fig. 3-1). Principal coordinates analysis can also be applied to nonphylogenetic methods for comparing communities, such as Euclidean distance, Jensen-Shannon divergence, or Bray-Curtis dissimilarity, which operate independent of evolutionary tree data but can make biologic patterns more difficult to identify. The taxonomic data or distance matrices can also be used as input into a range of machine-learning algorithms (such as Random Forests) that employ supervised classification to identify differences between labeled groups of samples. Supervised classification is useful for identifying differences between cases and controls but can obscure important patterns intrinsic to the data, including confounding variables such as different sequencing runs or patient populations.

As noted above, the greatest beta diversity is that among body sites. This fact underscores the need to specify body habitat in microbiota analyses of any type, including microbial surveillance studies examining the flow of normal and pathogenic organisms into and out of different body sites in patients and their health care providers. Several other key points have emerged from beta diversity studies of human-associated microbial communities—notably, that (1) there is a high level of interpersonal variability in every body habitat studied to date, (2) intrapersonal variation in a given body habitat is less pronounced, and (3) family members have more similar communities than unrelated individuals living in separate households. Thus, a person is his/her own best control, and examination of an individual over time as a function of disease state or treatment intervention is desirable. Similarly, family members serve as logical reference controls, although age is a major covariate that affects microbiota structure.

Studies of fecal samples obtained from twins over time have shown that the overall degree of phylogenetic similarity of bacterial communities does not differ significantly between monozygotic and dizygotic twin pairs, although monozygotic twin pairs may be more similar in some populations at earlier ages. These results, together with intervention studies in mice and epidemiologic observations in humans, emphasize that early environmental exposures are a very important determinant of adult-gut microbial ecology. In humans, the initial exposures depend on delivery mode: babies sampled within 20 min of birth have relatively undifferentiated microbial communities in the mouth, the skin, and the gut. For vaginally delivered babies, these communities resemble the specific microbial communities found in the mother’s vagina. For babies delivered by cesarean section, the communities resemble skin communities. Although studies of older children and of adults stratified by delivery mode are still rare in the literature, these differences have been shown to persist until at least 4 months of age and perhaps until age 7 years. The infant gut microbiota changes to resemble the adult gut community over the first 3 years of life; comparable studies have not been done in other body habitats to date.

Exposures to environmental microbial reservoirs can continue to influence community structure. For example, unrelated cohabiting adults have more similar microbiotas in all of their body habitats than do non-cohabiting adults, and humans resemble the dogs they live with, at least in terms of skin microbiota. Gender and sexual maturation may also affect the microbiota structure, although efforts to isolate these variables are complicated by many confounding factors; any gender effect must be small compared with the effects of other variables such as diet (except in the case of the female urinary tract, which is influenced by the vaginal microbiota).

The vaginal microbiota illustrates another intriguing aspect of the contributions made by various factors to interpersonal differences in microbial community structure within a given body habitat. Bacterial 16S rRNA–based studies of the midvaginal microbiota in sexually active women have documented significant differences in community configurations between four self-reported ethnic groups: Caucasian, black, Hispanic, and Asian. Unlike most other body habitats that have been surveyed, this ecosystem is dominated by a single genus, Lactobacillus. Four species of this genus together account for more than half of the bacteria in most vaginal communities. Five community categories have been defined: four are dominated by L. iners, L. crispatus, L. gasseri, and L. jensenii, respectively, and the fifth has proportionally fewer lactobacilli and more anaerobes. The representation of these community categories is distinct within each of the four ethnic groups and correlates with vaginal pH and Nugent score (the latter being a biomarker for bacterial vaginosis). Longitudinal studies of individuals are being conducted to identify factors that determine the assembly of these distinct communities—both within and among ethnic groups—as well as their resistance to or resilience after various physiologic and pathologic disturbances. For example, the menstrual cycle and pregnancy turn out to be surprisingly significant factors (cause larger changes) compared with sexual activity.

Yet another factor affecting beta diversity is spatial location within a habitat. Several surveys show that the skin harbors bacterial communities with predictable, albeit complex, biogeographic features. To determine whether these differences are due to differences in local environmental factors, to the history of a given site’s exposure to microbes, or to a combination of the two, reciprocal microbiota transplantation has been performed. Microbial communities from one region of the skin were depleted by treatment with germicidal agents, and the region (plot) was inoculated with a “foreign” microbiota harvested from different regions of the skin or from different body habitats from the same or another individual. Community assembly at the site of transplantation was then tracked over time. Remarkably, assembly proceeded differently at different sites: forearm plots receiving a tongue microbiota remained more similar to tongue communities than to native forearm communities in terms of their composition and diversity, while forehead plots inoculated with tongue bacteria changed to become more similar to native forehead communities. Thus, in addition to the history of exposure to tongue bacteria, environmental factors operating at the forehead plot likely shape community assembly. Intriguingly, the factors that shape fungal skin communities appear to be entirely different from those that shape bacterial skin communities. The palm and forearm have high bacterial and low fungal diversity, whereas the feet have the opposite diversity pattern. Moreover, fungal communities are generally shaped by location (foot, torso, head), whereas bacterial communities are generally shaped by moisture phenotype (dry, moist, or sebaceous).

Co-occurrence analysis

Co-occurrence analysis seeks to identify which phylotypes are co-distributed across individuals in a given body habitat and/or between habitats and to determine the factors that explain the observed patterns of co-distribution. Positive correlations tend to reflect shared preferences for certain environmental features, while negative correlations typically reflect divergent preferences or a competitive relationship. Syntrophic (cross-feeding) relationships reflect interdependent interactions based on nutrient-sharing strategies. For example, in food webs, the products of one organism’s metabolism can be used by the other for its own unique metabolic capabilities (e.g., the interactions between fermentative organisms and methanogens).

Enterotype analysis

Enterotype analysis seeks to classify individuals into discrete groups based on the configuration of their microbiotas, essentially drawing boundaries on a map defined by principal coordinates analysis or other ordination techniques. The first enterotype analysis used supervised clustering to define three major types of human-gut microbial configurations across three distinct human studies and provided a view that presupposed the existence of three clusters. Subsequent work has shown that the range of variability in the gut microbiota of children and of non-Western populations greatly exceeds the variability captured in the populations used to define the original enterotypes; in addition, even in Western populations, the variability follows more of a continuum dominated by a gradient in the abundance of the genera Bacteroides and Prevotella. Another consideration in enterotype analysis is whether location on a map defined by healthy human variation is relevant to predisposition to disease or whether instead rare species with particular functions are more important discriminants.

Functional redundancy

Functional redundancy arises when functions are performed by many bacterial taxa. Thus interpersonal differences in bacterial diversity (i.e., which bacteria are present) are not necessarily accompanied by comparable degrees of difference in functional diversity (i.e., what these bacteria can do). Characterization of a microbiome by shotgun sequencing is important because, unlike SSU rRNA analyses, shotgun sequencing provides a direct readout of the genes (and, via comparative genomics, their functions) in a given community. One fundamental question is the degree to which variations in the species occupying a given body habitat correlate with variations in a community’s functional capabilities. For example, the neutral theory of community assembly developed by macroecologists suggests that species are added to the community without respect to function, automatically endowing the community with functional redundancy. If applicable to the microbial world, neutral community assembly would predict a high level of variation in the types of microbial lineages that occupy a given body habitat in different individuals, although the broad functions encoded in the microbiomes of these communities could be quite similar.

Shotgun sequencing of the fecal microbiome has revealed that different microbial communities converge on the same functional state: in other words, there is a group of microbial genes represented in the guts of unrelated as well as related individuals. The same principle holds true at other body sites (Fig. 3-2). The “core” gut microbiome is enriched in functions related to microbial survival (e.g., translation; metabolism of nucleotides, carbohydrates, and amino acids) and in functions that benefit the host (nutrient and energy partitioning from the diet to microbes and host). The latter functions encompass the food webs mentioned above, in which products of one type of microbe become the substrates for other microbes. These webs, which can be incredibly elaborate, change as microbes adjust their patterns of gene expression and metabolism in response to alterations in nutrient availability. Thus the sum of all the activities of the members of a microbial community can be viewed as an emergent rather than a fixed property.

It is important to note that pairwise comparisons have shown that family members have functionally more similar gut microbiomes than do unrelated individuals. Thus, intrafamilial transmission of a gut microbiome within a given generation and across multiple generations could shape the biologic features of humans belonging to a kinship and modulate/mediate risks for a variety of diseases.

Stability

Like other ecosystems, human body habitat–associated microbial communities vary over time, and an understanding of this variation is essential for a functional understanding of our microbiota. Few high-resolution time series of individual healthy adults have been published to date, but one available daily time series suggests that individuals tend to resemble themselves microbially day to day over a span of 6–15 months, retaining their separate identities during cohabitation. The development of low-error amplicon sequencing methods has provided a much more reliable way for defining stability at the strain level than was available in the past. Application of these methods to the guts of healthy individuals sampled over time has disclosed that a healthy adult gut harbors a persistent collection of ~100 bacterial species and several hundred strains. The stability of the bacterial components follows a power law: bacterial strains acquired early in life can persist in the gut for decades, although their proportional representation changes as a function of numerous factors, including diet. Whole-genome sequencing of culturable components of the microbiota of study participants has confirmed that strains are retained in individuals for prolonged periods and are shared among family members.

Resilience

The ability of a microbiota or microbiome to rebound from a short-term perturbation, such as antibiotic administration or an infection, is defined as its resilience. This capacity can be visualized as a ball rolling over a landscape of local minima; essentially, the community moves into a new state and, to recover, must move through another, unstable state. In some cases, recovery will lead to the original stable state; in others, it will lead to a new stable state, which may be either healthy or unhealthy. Changes in, for example, diet or host physiologic status may introduce alterations into the landscape itself, making it easier to move from the initial state to any one of a number of other states, potentially with different health consequences. Microbial communities in our body habitats differ widely in resilience. For example, hand washing leads to profound changes in the microbial community, greatly increasing diversity (presumably because of the preferential removal of high-abundance, dominant phylotypes such as Propionibacterium). Within 6 h, the hand microbiota rebounds to resemble the original hand communities. The effects of repeated hand washing still need to be defined; for example, the surface microbiota on the skin (as measured by scrape biopsies) consists of ~50,000 microbial cells/cm², whereas the subsurface microbiota (as measured by punch biopsies) consists of ~1,000,000 microbial cells/cm².

In a study of three healthy adult volunteers given a short course of ciprofloxacin (500 mg by mouth twice a day for 5 days—a regimen commonly used against uncomplicated urinary tract infections), overall gut-community configuration came to resemble baseline within 6 months after treatment cessation, although some taxa failed to recover. However, the effects of the antibiotic perturbation were highly individualized. Administration of a second course of treatment months later led to altered-community states, relative to baseline, in all three volunteers; again, the extent of the alteration differed with the individual. Crucially, as shown in this and other studies, a given bacterial taxon can respond differently to the same antibiotic in different individuals; this observation suggests that the rest of the microbial community plays an important role in determining the effects of antibiotics on a per-individual basis.

In any body habitat, the microbial-community state after disturbance may be degraded. However, this degraded state may itself be resilient, and it may therefore be difficult to restore a more functional state. For example, Clostridium difficile infection can persist for years. The development and resilience of a degraded state may be driven by positive feedback loops, such as reactive oxygen species cascades involving host macrophages that promote the further growth of proinflammatory Proteobacteria, as well as negative-feedback loops such as depletion of the butyrate needed for promotion of a healthy gut epithelial barrier and further establishment of beneficial members of the microbiota. Consequently, microbiota-based therapies may require either (1) the elimination of a feedback loop that prevents establishment of a new community or (2) identification of a direction for change and a stimulus of sufficient magnitude (e.g., invasion and establishment of microbes from a fecal transplant or from a defined consortium of cultured, sequenced members of the human gut microbiota; see below) to overcome the resilience mechanisms inherent in the degraded state. A critical unresolved question that especially affects infants, whose microbiota is changing rapidly, is whether intervention during periods of rapid change or during periods of relative stability is generally more effective.

Gnotobiotic animals are raised in germ-free environments—with no exposure to microbes—and then colonized at specific stages of life with specified microbial communities. Gnotobiotic mice provide an excellent system for controlling host genotype, microbial community composition, diet, and housing conditions. Microbial communities harvested from donor mice with defined genotypes and phenotypes can be used to determine how the donors’ microbial communities affect the properties of formerly germ-free recipients. The recipients may also affect the transplanted microbiota and its microbiome. Thus gnotobiotic mice afford investigators an opportunity to marry comparative studies of donor communities to functional assays of community properties and to determine how (and for how long) these functions influence host biology.

The cardiovascular system

The gut microbiota affects the elaborate microvasculature underlying the small-intestinal epithelium: capillary network density is markedly reduced in adult germ-free animals but can be restored to normal levels within 2 weeks after gut microbiota transplantation. Mechanistic studies have shown that the microbiota promotes vascular remodeling in the gut through effects on a novel extravascular tissue factor–protease-activated receptor (PAR1) signaling pathway. Heart weight measured echocardiographically or as wet mass and normalized to tibial length or lean body weight is significantly reduced in germ-free mice; this difference is eliminated within 2 weeks after colonization with a gut microbiota. During fasting, a gut microbiota–dependent increase in hepatic ketogenesis (regulated by peroxisome proliferator–activated receptor α) occurs, and myocardial metabolism is directed to ketone body utilization. Analyses of isolated, perfused working hearts from germ-free and colonized animals, together with in vivo assessments, have shown that myocardial performance in germ-free mice is maintained by increasing glucose utilization. However, heart weight is significantly reduced in both fasted and fed mice; this heart-mass phenotype is completely reversed in germ-free mice fed a ketogenic diet. These findings illustrate how the gut microbiota benefits the host during periods of nutrient deprivation and represent one link between gut microbes and cardiovascular metabolism and health.

Conventionally raised apoE-deficient mice develop a less severe form of atherosclerosis than their germ-free counterparts when fed a high-fiber diet. This protective effect of the microbiota is obviated when animals are fed a diet low in fiber and high in simple sugars and fat. A number of the beneficial effects attributed to diets with high proportional representation of whole grains, fruits, and vegetables are thought to be mediated by end products of microbial metabolism of dietary compounds, including short-chain fatty acids and metabolites derived from flavonoids. Conversely, microbes can convert otherwise harmless dietary compounds into metabolites that increase risk for cardiovascular disease. Studies of mice and human volunteers have revealed that gut microbiota metabolism of dietary L-carnitine, which is present in large amounts in red meat, yields trimethylamine-N-oxide, which can accelerate atherosclerosis in mice by suppressing reverse cholesterol transport.

Yet another facet of microbial influence on cardiovascular physiology was revealed in a study of mice deficient in Olfr78 (a G protein–coupled receptor expressed in the juxtaglomerular apparatus, where it regulates renin secretion in response to short-chain fatty acids) or Gpr41 (another short-chain fatty acid receptor that, together with Olfr78, is expressed in the smooth muscle cells present in small resistance vessels). This study demonstrated that the microbiota can modulate host blood pressure via short-chain fatty acids produced by microbial fermentation.

Bone

Adult germ-free mice have greater bone mass than their conventionally raised counterparts. This increase in bone mass is associated with reduced numbers of osteoclasts per unit bone surface area, reduced numbers of CD11b+/GR1 osteoclast precursors in bone marrow, decreased numbers of CD4+ T cells, and reduced levels of expression of the osteolytic cytokine tumor necrosis factor α. Colonization with a normal gut microbiota resolves these observed differences between germ-free and conventionally raised animals.

Brain

Adult germ-free and conventionally raised mice differ significantly in levels of 38 out of 196 identified cerebral metabolites, 10 of which have known roles in brain function; included in the latter group are N-acetylaspartic acid (a marker of neuronal health and attenuation), pipecolic acid (a presynaptic modulator of γ-aminobutyric acid levels), and serine (an obligatory co-agonist at the glycine site of the N-methyl-d-aspartate receptor). Propionate, a short-chain fatty acid product of gut microbial-community metabolism of dietary fiber, affects expression of genes involved in intestinal gluconeogenesis via a gut–brain neural circuit involving free fatty-acid receptor 3; this effect provides a mechanistic explanation for the documented beneficial impact of dietary fiber in enhancing insulin sensitivity and reducing body mass and adiposity.

Studies of a mouse model (maternal immune activation) with stereotyped/repetitive and anxiety-like behaviors indicate that treatment with a member of the human gut microbiota, Bacteroides fragilis, corrects gut barrier (permeability) defects; reduces elevated levels of 4-ethylphenylsulfate, a metabolite seen in the maternal immune activation model that has been causally associated with the animals’ behavioral phenotypes; and ameliorates some behavioral effects. These observations highlight the importance of further exploration of potentially co-evolved relationships between the microbiota and host behavior.

Immune function

Many foundational studies have shown that the gut microbiota plays a key role in the maturation of the innate as well as the adaptive components of the immune system. The intestinal epithelium, which is composed of four principal cell lineages (enterocytes plus goblet, Paneth, and enteroendocrine cells), acts as a physical and functional barrier to microbial penetration. Goblet cells produce mucus that overlies the epithelium, where it forms two layers: an outer (luminal-facing) looser layer that harbors microbes and a denser lower layer that normally excludes microbes. Members of the Paneth cell lineage reside at the base of crypts of Lieberkühn and secrete antimicrobial peptides. Studies in mice have demonstrated that Paneth cells directly sense the presence of a microbiota through expression of the signaling adaptor protein MyD88, which helps transduce signals to host cells upon recognition of microbial products through Toll-like receptors (TLRs). This recognition drives expression of antibacterial products (e.g., the lectin RegIIIγ) that act to prevent microbial translocation across the gut mucosal barrier.

The intestine is enriched for B cells that produce IgA, which is secreted into the lumen; there it functions to exclude microbes from crossing the mucosal barrier and to restrict dissemination of food antigens. The microbiota plays a key role in development of an IgA response: germ-free mice display a marked reduction in IgA+ B cells. The absence of a normal IgA response can lead to a massive increase in bacterial load. B cell–derived IgA that targets specific members of the gut microbiota plays an important role in preventing activation of microbiota-specific T cells.

Gut bacterial species elicit development of protective T_H17 and T_H1 responses that help ward off pathogen attack. Members of the microbiota also promote the development of a specialized population of CD4+ T cells that prevent unwarranted inflammatory responses. These regulatory T cells (T_regs) are characterized by expression of the transcription factor forkhead box P3 (FOXP3) and by expression of other cell-surface markers. There is a paucity of T_regs in the colonic lamina propria of germ-free mice. Specific members of the microbiota—including a consortium of Clostridium strains isolated from the mouse and human gut as well as several human-gut Bacteroides species —expand the T_reg compartment and enhance immunosuppressive functions.

The microbiota is a key trigger in the development of inflammatory bowel disease (IBD) in mice that harbor mutations in genes associated with IBD risk in humans. Moreover, components of the gut microbiota can modify the activity of the immune system to ameliorate or prevent IBD. Mice containing a mutant ATG16L1 allele linked to Crohn’s disease are particularly susceptible to IBD. Upon infection with mouse norovirus and treatment with dextran sodium sulfate, expression of a hypomorphic ATG16L1 allele leads to defects in small-intestinal Paneth cells and renders mice significantly more susceptible to ileitis than are wild-type control animals. This process is dependent on the gut microbiota and highlights how the intersection of host genetics, infectious agents, and the microbiota can lead to severe immune pathology; i.e., the pathogenic potential of a microbiota may be context-dependent, requiring a confluence of factors. An important observation is that members of the gut microbiota, including B. fragilis or members of Clostridium, prevent the severe inflammation that develops in mouse models mimicking various aspects of human IBD.

The gut microbiota has been implicated in promoting immunopathology outside of the intestine. Multiple sclerosis develops in conventionally raised mice whose CD4+ T cell compartment is reactive to myelin oligodendrocyte protein; their germ-free counterparts are completely protected from development of multiple sclerosis–like symptoms. This protection is reversed by colonization with a gut microbiota from conventionally raised animals.

Inflammasomes are cytoplasmic multiprotein complexes that sense stress and damage-associated patterns. Mice deficient in NLRP6, a component of the inflammasome, are more susceptible to colitis induced by administration of dextran sodium sulfate. This enhanced susceptibility is associated with alterations in the gut microbiota of these animals relative to that of wild-type controls. Mice are coprophagic, and co-housing of NLRP6-deficient mice with wild-type mice is sufficient to transfer the enhanced susceptibility to colitis induced by dextran sodium sulfate. Similar findings have been reported for mice deficient in the inflammasome adaptor ASC (apoptosis-associated speck-like protein containing a caspase recruitment domain). ASC-deficient mice are more susceptible to the development of a model of nonalcoholic steatohepatitis. This susceptibility is associated with alterations in gut microbiota structure and can be transferred to wild-type animals by co-housing.

Obesity and diabetes

Germ-free mice are resistant to diet-induced obesity. Genetically obese ob/ob mice have gut microbial-community structures that are profoundly altered from those in their lean wild-type (+/+) and heterozygous +/ob littermates. Transplantation of the ob/ob mouse microbiota into wild-type germ-free animals transmits an increased-adiposity phenotype not seen in mice receiving microbiota transplants from +/+ and +/ob littermates. These differences are not attributable to differences in food consumption but rather are associated with differences in microbial community metabolism. Roux-en-Y gastric bypass produces pronounced decreases in weight and adiposity as well as improved glucose metabolism—changes that are not ascribable simply to decreased caloric intake or reduced nutrient absorption. 16S rRNA analyses have documented that changes in the gut microbiota after this surgery are conserved among mice, rats, and humans; animal studies have demonstrated these changes along the length of the gut but most prominently downstream of the site of surgical manipulation of the bowel. Notably, transplantation of the gut microbiota from mice that have undergone Roux-en-Y gastric bypass to germ-free mice that have not had this surgery produces reductions in weight and adiposity not seen in recipients of microbiotas from mice that underwent sham surgery.

The gut microbiota confers protection against the development of type 1 diabetes mellitus in the non-obese diabetic (NOD) mouse model. Disease incidence is significantly lower in conventionally raised male NOD mice than in their female counterparts, while germ-free males are as susceptible as their female counterparts. Castration of males increases disease incidence, while androgen treatment of females provides protection. Transfer of the gut microbiota from adult male NOD mice to female NOD weanlings is sufficient to reduce the severity of disease relative to that among females receiving a microbiota from an adult female or an unmanipulated female. The blocking of protection by treatment with flutamide highlights a functional role for testosterone signaling in this microbiota-mediated protection against type 1 diabetes.

NOD mice deficient in MyD88, a key component of the TLR signaling pathway, do not develop diabetes and exhibit increased relative abundance of members of the family-level taxon Lactobacillaceae. Consistent with these findings, investigators have documented lower levels of representation of members of the genus Lactobacillus in children with type 1 diabetes than in healthy controls. Components of lactobacilli have been shown to promote gut barrier integrity. Studies in various animal models indicate that translocation of bacterial components, including bacterial lipopolysaccharides, across a leaky gut barrier triggers low-grade inflammation, which contributes to insulin resistance. Mice deficient in TLR5 exhibit alterations in the gut microbiota and hyperphagia, and they develop features of metabolic syndrome, including hypertension, hyperlipidemia, insulin resistance, and increased adiposity.

The gut microbiota regulates biosynthesis as well as metabolism of host-derived products; these products can signal through host receptors to shape host physiology. An example of this symbiosis is provided by bile acids, which direct metabolic effects that are largely mediated through the farnesoid X receptor (FXR, also known as NR1H4). In leptin-deficient mice, FXR deficiency protects against obesity and improves insulin sensitivity. In mice with diet-induced obesity that are subjected to vertical sleeve gastrectomy, the surgical procedure results in elevated levels of circulating bile acids, changes in the gut microbiota, weight loss, and improved glucose homeostasis. However, weight reduction and improved insulin sensitivity are mitigated in animals with engineered FXR-deficiency.

Xenobiotic metabolism

Evidence is accumulating that pharmacogenomic studies need to consider the gene repertoire present in our H. sapiens genome as well as that in our microbiomes. For example, digoxin is inactivated by the human gut bacterium Eggerthella lenta, but only by strains with a cytochrome-containing operon. Expression of this operon is induced by digoxin and inhibited by arginine. Studies in gnotobiotic mice established that dietary protein affects (reduces) microbial metabolism of digoxin, with corresponding alterations in levels of the drug in both serum and urine. These findings reinforce the need to consider strain-level diversity in the gut microbiota when examining interpersonal variations in the metabolism of orally administered drugs.

Characterizing the effects of the human microbiota on host biology in mice and humans

Questions about the relationship between human microbial communities and health status can be posed in the following general format: Is there a consistent configuration of the microbiota definable in the study population that is associated with a given disease state? How is the configuration affected by remission/relapse or by treatment? If a reconfiguration does occur with treatment, is it durable? How is host biology related to the configuration or reconfiguration? What is the effect size? Are correlations robust to individuals from different families and communities representing different ages, geographic locales, and lifestyles?

As in all studies involving human microbial ecology, the issue of what constitutes a suitable reference control is extremely important. Should we choose the person himself or herself, family members, or age- or gender-matched individuals living in the same locale and representing similar cultural traditions? Critically, are the relationships observed between microbial community structure and expressed functions a response to disease state (i.e., side effects of other processes), or are they a contributing cause? In this sense, we are challenged to evolve a set of Koch’s postulates that can be applied to whole microbial communities or components of communities rather than just to a single purified organism. As in other circumstances in which experiments to determine causality of human disease are difficult or unethical, Hill’s criteria, which examine the strength, consistency, and biologic plausibility of epidemiologic data, can be useful.

Sets of mono- and dizygotic twins and their family members represent a valuable resource for initially teasing out relationships between environmental exposures, genotypes, and our own microbial ecology. Similarly, monozygotic twins discordant for various disease states enhance the ability to determine whether various diseases can be linked to a person’s microbiota and microbiome. A twin-pair sampling design rather than a conventional unrelated case–control design has advantages owing to the pronounced between-family variability in microbiota/microbiome composition and the potential for multiple states of a community associated with disease. Transplantation of a microbiota from suitable human donor controls representing different disease states and communities (e.g., twins discordant for a disease) to germ-free mice is helpful in establishing a causal role for the community in pathogenesis and for providing insights relevant to underlying mechanisms. In addition, transplantation provides a preclinical platform for identifying next-generation probiotics, prebiotics, or combinations of the two (synbiotics). Obesity and obesity-associated metabolic dysfunction illustrate these points.

The gut microbiotas (and microbiomes) of obese individuals are significantly less diverse than those of lean individuals; the implication is that there may be unfilled niches (unexpressed functions) that contribute to obesity and its associated metabolic abnormalities. Le Chatelier and colleagues observed a bimodal distribution of gene abundance in their analysis of 292 fecal microbiomes: low-gene-count (LGC) individuals averaged 380,000 microbial genes per gut microbiome, while high-gene-count (HGC) individuals averaged 640,000 genes. LGC individuals had an increased risk for type 2 diabetes and other metabolic abnormalities, whereas the HGC group was metabolically healthy. When gene content was used to identify taxa that discriminated HGC and LGC individuals, the results revealed associations between anti-inflammatory bacterial species such as Faecalibacterium prausnitzii and the HGC group and between proinflammatory species such as Ruminococcus gnavus and the LGC group. LGC microbiomes had significantly greater representation of genes assigned to tricarboxylic acid cycle modules, peroxidases, and catalases—an observation suggesting a greater capacity to handle oxygen exposure and oxidative stress; HGC microbiomes were enriched in genes involved in the production of organic acids, including lactate, propionate, and butyrate—a result suggesting increased fermentative capacity.

Transplantation of an uncultured fecal microbiota from twins stably discordant for obesity or of bacterial culture collections generated from their microbiota transmits their discordant adiposity phenotypes as well as obesity-associated metabolic abnormalities to recipient germ-free mice. Co-housing of the recipient coprophagic gnotobiotic mice results in invasion of specific bacterial species from the transplanted lean twin’s culture collection into the guts of cage mates harboring the obese twin’s culture collection (but not vice versa), thereby preventing the latter animals from developing obesity and its associated metabolic abnormalities. It is noteworthy that invasion and prevention of obesity and metabolic phenotypes are dependent on the type of human diets fed to animals: prevention is associated with a diet low in saturated fats and high in fruit and vegetable content, but not with a diet high in saturated fats and low in fruit and vegetable content.

This approach provides evidence for a causal role for the microbiota in obesity and its attendant metabolic abnormalities. It also provides a method for defining unoccupied niches in disease-associated microbial communities, the role of dietary components in determining how these niches can be filled by human gut–derived bacterial taxa, and the effects of such occupancy on microbial and host metabolism. It also provides a way to identify health-promoting diets and next-generation probiotics representing naturally occurring members of our indigenous microbial communities that are well adapted to persist in a given body habitat.

A key to this approach is the ability to harvest a microbial community from a donor representing a physiology, disease state, lifestyle, or geography of interest; to preserve the donor’s community by freezing it; and then to resurrect and replicate it in multiple recipient gnotobiotic animals that can be reared under conditions where environmental and host variables can be controlled and manipulated to a degree not achievable in clinical studies. Since these mice can be followed as a function of time prior to and after transplantation, in essence, a snapshot of a donor’s community can be converted into a movie. Transplantation of intact uncultured human (fecal) microbiota samples from multiple donors representing the phenotype of interest, with administration of the donors’ diets (or derivatives of those diets) to different groups of mice, is one way to assess whether transmissible responses are shared features of the microbiota or are highly donor specific. A second step is to determine whether the culturable component of a representative microbiota sample can transmit the phenotype(s) observed with the intact uncultured sample. Possession of a collection of cultured organisms that have co-evolved in a given donor’s body habitat sets the stage for the selection of subsets of the collection for testing in gnotobiotic mice, the determination of which members are responsible for effecting the phenotype, and the elucidation of the mechanisms underlying these effects. The models used may inform the design and interpretation of clinical studies of the very individuals and populations whose microbiota are selected for creating these models.

Human-to-human fecal microbiota transplantation (FMT) is currently the most direct way to establish proof-of-concept for a causal role for the microbiota in disease pathogenesis. A human donor’s feces are provided to a recipient via nasogastric tube or another technique. Numerous small trials have documented the effects of FMT from healthy donors to recipients with diseases ranging from C. difficile infection to Crohn’s disease, ulcerative colitis, and type 2 diabetes. Only a few of these studies have used a double-blind, placebo-controlled design.

In a double-blind, controlled trial involving men 21–65 years old with a body mass index of >30 kg/m² and documented insulin resistance, FMT was performed using a microbiota from metabolically healthy lean donors or from the study participants themselves. A microbiota from lean donors significantly improved peripheral insulin sensitivity over that in controls. This change was associated with an increase in the relative abundance of the butyrate-producing bacteria related to Roseburia intestinalis (in the feces) and Eubacterium hallii (in the small intestine).

The efficacy of FMT for the treatment of recurrent C. difficile infection has been assessed in a number of small trials. One unblinded, placebo-controlled trial assessed the use of FMT in 42 patients with recurrent C. difficile infection (defined as at least one relapse after treatment with vancomycin or metronidazole for ≥10 d). Patients were pretreated with oral vancomycin. The experimental group then received FMT via nasoduodenal tube from healthy volunteer donors (<60 years of age) selected from the community. Controls underwent sterile lavage or received oral vancomycin alone. In 10 weeks of follow-up, infection was cured (with cure defined as three negative fecal tests for C. difficile toxin) in 81% of patients in the FMT group (13 of 16) but in only 23% (3 of 13) in the bowel-lavage control arm and 31% (4 of 13) in the vancomycin-only group. Metagenomic analysis of microbiota samples collected before and after treatment revealed an increased representation of Bacteroidetes and Clostridium clusters IV and XIVa, along with a 100-fold decrease in the relative abundance of Proteobacteria, in the FMT group.

A meta-analysis of FMT in C. difficile infection examined 20 case-series publications, 15 case reports, and the one unblinded study described above. All but one of these studies used fresh (not frozen) fecal samples. Donor selection varied, although most donors were family members or relatives and most studies excluded donors who had recently received antibiotics. It is noteworthy that the concentrations of infused donor feces varied widely (i.e., from 5 g to 200 g, resuspended in 10–500 mL); these fecal suspensions were introduced at different sites along the gastrointestinal tract, including the stomach and points throughout the small intestine and colon. Resolution of infection, which was frequently assessed on the basis of symptom resolution (with C. difficile toxin testing rarely performed), was documented in 87% (467) of 536 treated patients. The most common adverse events reported were diarrhea (94% of cases) and abdominal cramps (31%) on the day of infusion. The meta-analysis was limited to clinical outcomes and did not specifically address the role of the microbiota in disease resolution (e.g., the extent of invasion of donor taxa; their persistence; or the long-term effects of transplantation on various facets of host biology, which generally have not been evaluated).

Sober and thoughtful consideration needs to be applied to the therapeutic use of FMT, which represents an early and rudimentary approach to microbiota manipulation that very likely will be replaced by administration of defined collections of sequenced, cultured members of the human microbiota (probiotic consortia). A number of published reports on FMT have garnered significant public attention. This attention, coupled with an increasing public appreciation of the beneficial nature of our interactions with microbes, demands that the precautionary principle be honored and that risks versus benefits of such interventions be carefully evaluated.

To date, most FMT trials have failed to define (or have differed in) significant confounders, including (1) the criteria used for donor sample selection; (2) the methods used for donor sample preparation and characterization as well as the decision about whether or not to create a repository for donor and recipient samples that will permit retrospective analyses (and meta-analyses for given disease states); (3) the development of minimal standards for assessing the invasion of recipient gut communities by taxa from donor microbiota (using microbial source-tracking methods) as well as the timing, duration, nature, and breadth of sampling of the recipient as a function of transplantation; (4) the adoption of minimal standards for collection of patients’ clinical data (e.g., age, diet, antibiotic use) and the establishment of databases for entering these data (including use of a defined vocabulary for annotating the clinical data); and (5) the development of standards for informed consent in lieu of knowledge of the long-term effects of the procedure. The regulatory landscape is evolving. The U. S. Food and Drug Administration recently issued an enforcement policy specifically addressing the use of FMT for the treatment of recurrent C. difficile infection; this policy indicates that the agency intends to “exercise enforcement discretion regarding the investigational new drug (IND) requirements for the use of FMT to treat C. difficile infection not responding to standard therapies,” but it does not waive IND requirements for other FMT studies.

The design of human microbiome studies is rapidly evolving, in part because the data are highly multivariate, are compositional, and do not meet distributional assumptions of standard statistical tests such as analysis of variance. Consequently, the proper number of subjects to enroll and the proper populations to target remain to be established. One useful approach is to review published studies and ask whether the reported conclusion could be obtained with fewer subjects (sample rarefaction) and/or fewer sequencing reads from SSU rRNA genes, whole-community DNA (microbiomes), or expressed community mRNA (metatranscriptomes) per subject (sequence rarefaction). A common yet critical problem to avoid is under-sampling of the types of objects under study. For example, if the goal is to compare factors applying to individuals (e.g., individual diet), then dozens of individuals in each clinical category may be needed. If the goal is to compare factors applying to populations (e.g., demographic properties), then many populations may be needed.

Another key issue is whether the effect size to be studied, especially in meta-analysis, is greater than or less than technical effects. As noted above, different PCR primers will lead to different readouts of the taxonomy of a microbial community; these differences are, for example, greater than the differences between lean and obese subjects’ fecal microbiota but less than the difference between fecal communities in newborns and adults.

A central challenge in human microbiome research is establishing the extent to which diagnostic tests and therapeutic approaches are generalizable. This challenge is illustrated by studies of the capacities of gut microbiomes to metabolize orally administered drugs. The results could be very informative for the pharmaceutical industry as it seeks new and more accurate ways to predict bioavailability and toxicity. However, these studies should prompt consideration of the fact that many clinical trials are outsourced to countries where trial participants have diets and microbial community structures that differ from those of the intended initial recipients of the (marketed) drug. Capture and preservation of the wide range of microbial diversity present in different human populations—and thus of the capacity of our microbial communities to catalyze elaborate and in many respects uncharacterized biotransformations—represent potentially fertile ground for the discovery of new drugs (and new industrial processes of societal value). The chemical entities that our microbial communities have evolved to synthesize in order to support their mutually beneficial relationships and the human genes that these chemotypes influence may become new classes of drugs and new targets for drug discovery, respectively. Therefore, characterization of groups of individuals living in countries that are undergoing rapid transformations in cultural traditions and socioeconomic conditions and are witnessing the emergence of a variety of diseases associated with increasingly Western lifestyles (globalization) is a timely challenge. Birth cohort studies (including studies of twins) initiated every 10 years in these countries may be able to capture the impact of globalization, including changing diets, on human microbial ecology.

Although microbiome-associated diagnostics and therapeutics provide new and exciting dimensions for personalized medicine, attention must be paid to the potentially broad societal impact of this work. For example, studies of the human gut microbiome are likely to have a disruptive effect on current views of human nutrition, enhancing appreciation of how food and the metabolic output of interactions of dietary components with the microbiota are intimately connected to myriad features of human biology. Underlying the efforts to elucidate the relations among food, the microbiome, and human nutrition is a need to proactively develop materials for educational outreach with a narrative and vocabulary that is understandable to broad and varied consumer populations representing different cultural traditions and widely ranging degrees of scientific literacy. The results have the potential to catalyze efforts to integrate agricultural policies and practice, food production, and nutritional recommendations for consumer populations representing different ages, geographic locales, and states of health.

Defining our metagenome (the genes embedded in our H. sapiens genome plus those in our microbiome) will likely lead to an entirely new level of refinement in our description of self, our genetic evolution, our postnatal development, the microbial legacy of our connection to family, and the consequences of personal lifestyle choices. While this information can help us understand the origins of certain yet unexplained health disparities, care must be taken to avoid stigmatization of individuals or groups of individuals having different cultural norms, belief systems, or behaviors. In partnership with human microbiome researchers, anthropologists need to examine the impact of studies of the human microbiome on the participants, assessing how this field and participants’ cultural traditions interact to affect these individuals’ perceptions about the natural world, the forces that affect their lives, and their connections to one another within the context of family and community.

Studies of human microbial ecology are an important manifestation of progress in the genome sciences, represent a timely step in our quest to achieve a better understanding of our place in the natural world, and reflect the evolving focus of twenty-first-century medicine on disease prevention, new definitions of health, new ways to determine the origins of individual biologic differences, and new approaches to evaluating the impact of changes in our lifestyles and biosphere on our biology. As microbiome-directed diagnostics and therapeutics emerge, we must be sensitive to the societal impact of this work.