The Human Microbiome: at the interface of health and disease

Characterizing the microbiome

Animals have had residential microbes performing metabolic functions for at least 500 million years, at a conservative estimate⁶,7. Extensive congruent phylogenies of animal hosts and their microbiota, involving both individual organisms and whole microbial populations¹,8,9, suggest the existence of specific selection based on co-adaptation. Cooperative interactions between microbes and their hosts typically involve microbial participation in host functions such as defence, metabolism, and reproduction¹⁰. For example, comparing germ-free and conventional mice indicates that microbiota are responsible for most of the metabolites that are detected in plasma¹¹.

Functional variation of host microbiota can be mediated by the introduction or extinction of particular microbial groups or by a change in population structure¹²–14. Such alterations can in turn be induced by selection by environmental factors¹⁰,15 such as dietary changes or exposure to antibiotics¹⁰,15. Below we describe the efforts to categorize the composition and complex dynamics of the microbiota.

Taxonomic and functional variation

The composition of the microbiome varies by anatomical site (Figure 1). The primary determinant of community composition is anatomical location: interpersonal variation is substantial²³,24 and is higher than the temporal variability seen at most sites in a single individual²⁵. Such relative temporal stability suggests that individuals can be grouped according to the major enterotypes present in the colon⁵ or the vagina⁴. However, dietary changes can rapidly cause substantial intestinal metagenomic changes, and enterotypes are known to cluster based on dietary abundance of animal protein vs carbohydrate²⁶. Similarly, nasopharyngeal microbiota in young children varies seasonally²⁴, and vaginal microbiota varies with menses⁴. In the absence of marked perturbations, the aggregate microbiota of an individual appears to vary relatively narrowly within host-specific boundaries; the basis of such boundaries have not been established, but may represent Nash equilibria¹³. Because minor microbial populations have the potential to bloom, the temporal variation observed in a host may be mirrored by the interindividual variation at a single time²⁷; that the system is dynamic suggests that there are greater interpersonal similarities than a snap-shot view indicates. However, large perturbations such as antibiotic exposure²⁸ or enteric infections (LA David, personal communication), can lead to transient disequilibrium²⁹, or to the development of a new stable state.

Figure 1

Compositional differences in the microbiome by anatomic site

High-throughput sequencing has revealed substantial intra-individual microbiome variation at different anatomical sites, and inter-individually for the same anatomical sites ⁴,5,25,52,89,93. However, higher level (e.g. phylum) taxonomic features display temporal (longitudinal) stability in individuals at specific anatomical sites. Such site-specific differences as well as observed conservation between human hosts provide an important framework to determine the biological and pathological significance of a particular microbiome composition. The figure indicates percentages of sequences at the taxonomic phylum level from selected references. Certain features, such as the presence or absence of Helicobacter pylori, can lead to permanent and marked perturbations in community composition⁹³.

Among all mammals, the microbiota is extensively conserved at high taxonomic levels⁷, but variation increases at progressively lower taxonomic levels³⁰. Consequently, 85% of the sequences obtained from the distal mouse gut represent genera that are not detected in humans³¹. Furthermore, intra-species variability of the microbiota within the human population is also substantial⁵, This degree of variation was unanticipated a priori. However, in retrospect, extensive taxonomic variation is unsurprising: a human harbours a climax population of about 10¹⁴bacterial cells, can host 10 ^{– 10 bacterial generations per human generation, is omniverous, and has accumulated genetic and epigenetic diversity as a host species for >1 million years. Indicator organisms such as Helicobacter pylori32} and Streptococcus mutans³³ highlight some differences across the microbiota³⁴ and metagenome³⁵ among human ethnic groups; however, the extent of ethnic variation in overall metagenomic composition is unknown. The microbiomes of monozygotic twins are more closely related to one another than those of unrelated individuals³⁶,37 but not strikingly so, indicating important post-natal influences on composition.

The extensive lower-level taxonomic variation and large compositional differences observed even among highly related host organisms (e.g. mice and humans) is counterbalanced by the substantial conservation of metagenomic core functions (Figure 2). This reflects the conservation of core bacterial properties involved in nucleic acid and protein synthesis, and in metabolic and structural requirements. Of the >50 known phyla, most of the human microbiota is composed of <10 (and mostly 6) phyla. Bacteria from other phyla, usually of plant origin, that may be present in skin, nasopharyngeal, or gut samples²⁴,25,38; are generally infrequent (<0.01% of the sequences) and probably represent transient carriage from food- and air-borne exposures. Why did the particular restriction of diversity within a few phyla evolve, not only in humans, but in perhaps all other vertebrates? One possibility is that within the relatively conserved boundaries for the microbiome permitted by the human genome, there exists a large array of contingency organisms and contingency genes. According to this hypothesis, the genes may only be active at some moment in the host’s lifespan, or perhaps at a frequency of less than once per lifespan.

Figure 2

Conservation of bacterial genes despite taxonomic variation

A) Turnbaugh et al. studied the distal gut microbiome in lean and obese twins and their mothers³⁷. There were substantial and significant taxonomic variations amongst the individuals, although Firmicutes and Bacteroidetes still constituted the majority of the taxa. B) Through metagenomic analyses, the functional characteristics of the microbiomeas identified by COG pathways are largely conserved, despite the taxonomic variation³⁷. COG pathways are denoted by: S – Unknown; R – General function; L – DNA; G – Carbohydrates; E – Amino acids; M – Envelope; K – Transcription; J – Translation; C – Energy; T – Signal transduction; P – Inorganic; V – Defense; H – Coenzymes; O – Protein turnover; F – Nucleotides; U – Secretion; I – Lipids; D – Cell cycle; B – Chromatin; Q – Second metabolites; N – Cell motility; W – Extracellular; Z – Cytoskeleton; A – RNA. Reproduced with permission from Turnbaugh et al37, Nature and the authors © Macmillan Publishers Ltd

The parallel needs of individual bacteria lead to both competition for key substrates and to functional redundancy in the microbiota. Nevertheless, the enormous bacterial biomass also provides many unique or minimally redundant bacterial genes¹⁹.

Tools for studying the metagenome

The taxonomic diversity inherent in complex environmental communities and the task of identifying specific associations with host traits create unique challenges. One approach to metagenome analysis involves assigning unassembled sequences generated by shotgun high-throughput sequencing (HTS)¹⁶ to the NCBI non-redundant Clusters of Orthologous Groups (COG) or the Kyoto Encyclopedia of Genes and Genomes (KEGG) databases¹⁷. This method facilitates the assessment of interactions that occur within the microbiome, and potentially between microbiome and host¹⁸. However, because a substantial fraction of the metagenome (~ 33%) is not well-represented by reference genomes, this strategy provides only a limited understanding of the functional potential of the microbiota. An alternative approach is to use catalogues of known genes to identify functional clusters within a sample; such clusters could correspond to the proposed taxonomic enterotypes⁵. A catalogue of the microbial genes present in the human gut, for example, is being generated using several approaches including sequencing, assembling and characterizing non-redundant microbial genes from fecal samples¹⁹, and whole genome sequencing of reference microbial species²⁰.

As sequencing and bioinformatics technologies continue to evolve (see ref ²¹ for a review of the state-of-the-art technologies), scientific priorities will include elucidating the ‘core’ metagenome that occupies a specific human niche, and discerning the differences between normal and diseased hosts.. As an example of the latter, Greenblum et al. applied new tools to understand interhost metagenomic variation in relation to phenotypes such as obesity and inflammatory bowel disease (IBD)²². By categorizing metagenomic sequences based on gene function, they constructed community-level metabolic networks varying in gene abundance, and examined the topological features of these networks in relation to host phenotype. Their analysis identified specific network topologies related to obesity and IBD; skewed topologies chiefly differ in genes related to host interactivity, particularly in relationship to metabolism. Such topological tools can be applied to explore differences in other host disease states.

Taxonomic and functional variation

The composition of the microbiome varies by anatomical site (Figure 1). The primary determinant of community composition is anatomical location: interpersonal variation is substantial²³,24 and is higher than the temporal variability seen at most sites in a single individual²⁵. Such relative temporal stability suggests that individuals can be grouped according to the major enterotypes present in the colon⁵ or the vagina⁴. However, dietary changes can rapidly cause substantial intestinal metagenomic changes, and enterotypes are known to cluster based on dietary abundance of animal protein vs carbohydrate²⁶. Similarly, nasopharyngeal microbiota in young children varies seasonally²⁴, and vaginal microbiota varies with menses⁴. In the absence of marked perturbations, the aggregate microbiota of an individual appears to vary relatively narrowly within host-specific boundaries; the basis of such boundaries have not been established, but may represent Nash equilibria¹³. Because minor microbial populations have the potential to bloom, the temporal variation observed in a host may be mirrored by the interindividual variation at a single time²⁷; that the system is dynamic suggests that there are greater interpersonal similarities than a snap-shot view indicates. However, large perturbations such as antibiotic exposure²⁸ or enteric infections (LA David, personal communication), can lead to transient disequilibrium²⁹, or to the development of a new stable state.

Figure 1

Compositional differences in the microbiome by anatomic site

High-throughput sequencing has revealed substantial intra-individual microbiome variation at different anatomical sites, and inter-individually for the same anatomical sites ⁴,5,25,52,89,93. However, higher level (e.g. phylum) taxonomic features display temporal (longitudinal) stability in individuals at specific anatomical sites. Such site-specific differences as well as observed conservation between human hosts provide an important framework to determine the biological and pathological significance of a particular microbiome composition. The figure indicates percentages of sequences at the taxonomic phylum level from selected references. Certain features, such as the presence or absence of Helicobacter pylori, can lead to permanent and marked perturbations in community composition⁹³.

Among all mammals, the microbiota is extensively conserved at high taxonomic levels⁷, but variation increases at progressively lower taxonomic levels³⁰. Consequently, 85% of the sequences obtained from the distal mouse gut represent genera that are not detected in humans³¹. Furthermore, intra-species variability of the microbiota within the human population is also substantial⁵, This degree of variation was unanticipated a priori. However, in retrospect, extensive taxonomic variation is unsurprising: a human harbours a climax population of about 10¹⁴bacterial cells, can host 10 ^{– 10 bacterial generations per human generation, is omniverous, and has accumulated genetic and epigenetic diversity as a host species for >1 million years. Indicator organisms such as Helicobacter pylori32} and Streptococcus mutans³³ highlight some differences across the microbiota³⁴ and metagenome³⁵ among human ethnic groups; however, the extent of ethnic variation in overall metagenomic composition is unknown. The microbiomes of monozygotic twins are more closely related to one another than those of unrelated individuals³⁶,37 but not strikingly so, indicating important post-natal influences on composition.

The extensive lower-level taxonomic variation and large compositional differences observed even among highly related host organisms (e.g. mice and humans) is counterbalanced by the substantial conservation of metagenomic core functions (Figure 2). This reflects the conservation of core bacterial properties involved in nucleic acid and protein synthesis, and in metabolic and structural requirements. Of the >50 known phyla, most of the human microbiota is composed of <10 (and mostly 6) phyla. Bacteria from other phyla, usually of plant origin, that may be present in skin, nasopharyngeal, or gut samples²⁴,25,38; are generally infrequent (<0.01% of the sequences) and probably represent transient carriage from food- and air-borne exposures. Why did the particular restriction of diversity within a few phyla evolve, not only in humans, but in perhaps all other vertebrates? One possibility is that within the relatively conserved boundaries for the microbiome permitted by the human genome, there exists a large array of contingency organisms and contingency genes. According to this hypothesis, the genes may only be active at some moment in the host’s lifespan, or perhaps at a frequency of less than once per lifespan.

Figure 2

Conservation of bacterial genes despite taxonomic variation

A) Turnbaugh et al. studied the distal gut microbiome in lean and obese twins and their mothers³⁷. There were substantial and significant taxonomic variations amongst the individuals, although Firmicutes and Bacteroidetes still constituted the majority of the taxa. B) Through metagenomic analyses, the functional characteristics of the microbiomeas identified by COG pathways are largely conserved, despite the taxonomic variation³⁷. COG pathways are denoted by: S – Unknown; R – General function; L – DNA; G – Carbohydrates; E – Amino acids; M – Envelope; K – Transcription; J – Translation; C – Energy; T – Signal transduction; P – Inorganic; V – Defense; H – Coenzymes; O – Protein turnover; F – Nucleotides; U – Secretion; I – Lipids; D – Cell cycle; B – Chromatin; Q – Second metabolites; N – Cell motility; W – Extracellular; Z – Cytoskeleton; A – RNA. Reproduced with permission from Turnbaugh et al37, Nature and the authors © Macmillan Publishers Ltd

The parallel needs of individual bacteria lead to both competition for key substrates and to functional redundancy in the microbiota. Nevertheless, the enormous bacterial biomass also provides many unique or minimally redundant bacterial genes¹⁹.

Resilience and community disturbance

Resilience, the ability to withstand disturbance, is a central concept in ecology. The resilience of the human colonic microbiome is beautifully illustrated by recent studies of twins examined before, during, and after 7-weeks of ingesting a fermented milk product containing a sample composed of ~10 ^{Bifidobacterium, Lactobacillus, Lactococcus, and Streptococccus species36}. Despite the daily oral inoculations, the composition of the microbiota at the 16S rDNA level and the metagenome were essentially unaffected. While the microbiome of human adults appears highly resilient, no comparable studies have been performed in children, but because population structures appear more dynamic and developmental³⁹, resilience may be lower. An important natural experiment has been occurring over the past 70 years in which most of the world’s population has been exposed to pharmacological doses of antimicrobial agents. Such usage has been based on the implicit belief that the human microbiome is completely resilient, and returns to the status quo ante after antibiotic-induced perturbation. However, study of indicator organisms such as H. pylori, indicates that extirpations of a bacterial species can occur in individual hosts⁴⁰. There also may be medium- and long-term selection of resistant organisms, and destabilization of the microbiome with new species compositions, in the absence of further antibiotic exposure²⁸,41. Thus, despite the extensive resilience inherent in a complex eco-system, there may be loss of recovery from continued perturbations²⁹, with important implications for human health⁴².

Medical scientists are familiar with Koch’s postulates, which are used as criteria to determine whether a microbe causes disease⁴³. However, pathogenetic considerations about the microbiome may better focus on community characteristics, which are largely governed by richness, composition, and interactions among the constituent members⁷,16,44. Substantial perturbation (community disturbance⁴⁵) tests the resilience of the community, e.g., it ability to resist invasion by exogenous microbes; stable diverse communities resist pathogens⁴⁶. At present, 16S rRNA analyses focus on taxonomic differences at or above the species level. However, examinations below the subspecies level, relating to strain, or even allele, ultimately may be more significant, although the technology, importantly the informatics tools, are not yet developed enough.

Extinctions

The human microbiome represents one or more complete ecosystems. The trophic organization of species-rich communities is similar to other complex network topologies, in that it shows extreme heterogeneity, with a relatively small number of highly connected nodes dominating⁴⁷. Such communities may resist random perturbations, but if keystone species⁴⁸ are lost, effects may cascade, with secondary extinctions; high biodiversity diminishes this risk¹². The substantial non-linear interactions in complex co-evolved systems ensure that ecological networks are robust against random removals⁴⁹. However, with repeated perturbations, the effects of gene loss can be amplified by downstream effects on co-colonizing microbes (secondary extinctions) and on the host. Because of allelopathy, the effects of extinctions may magnify⁵⁰. In the short-term, functional redundancy may mask extinction effects, but in the longer-run, extinctions lead to loss of contingency responses that can cause ecological crashes⁴⁹. Considering the importance of guilds of bacteria exploiting parallel and sequential metabolic pathways, these concepts are germane to the human metagenome. With modernity, horizontal microbial transmission has been diminishing, and there has been unprecedented selection against existing, long-present microbes⁴⁰. As has occurred in the human stomach with the loss of a dominant species⁵¹,52, body sites then can harbor alternative stable states.

In summary, as with other complex ecosystems, the microbiomes that populate specific human anatomical niches are species-rich, but possess particular overall community characteristics at higher organizational levels. All are subject to perturbation in the course of normal development and aging, and especially with disease. As our knowledge of the fundamental characteristics and biology of the human microbiome grows, so will our ability to understand disease-related variation.

Influences on the microbiota during the life cycle

As described in the previous section, differences in microbiota composition exist across body sites and between individuals. However, changes are also evident across the human lifespan. Important questions in this field involve determining if such temporal changes are life-stage specific, and whether they are predetermined by host genetic characteristics, or by environmental factor.

Inheritance of microbiota

The congruent phylogenies of mammals and their microbiota⁸ provide strong evidence for the inheritance of the microbiota⁷ Although inheritance of the microbiota from the father is presently little studied, increasing evidence supports inheritance from the mother³⁴,53. Until the amniotic sac ruptures, a fetus is considered to be sterile, or essentially sterile. Immediately after vaginal delivery, founding microbial populations in the baby closely resemble that of their mother’s vagina⁵⁴, with lactobacilli predominating. Since lactic acid-producing bacteria dominate in both vagina and in mother’s milk, the initial gastro-intestinal tract bloom of lactobacilli in the baby cannot be considered accidental. Lactobacilli represent the pioneer community in mice⁵⁵ as well as humans³⁹, preparing the gastrointestinal tract for subsequent microbial successions, until microbial maturity is reached.

The multiple opportunities for the microbiota to be transferred from a mother to her baby may be disrupted by modern lifestyles. Caesarian section instead of vaginal delivery is an obvious example of the potential impact of medical practice on microbiota composition, with substantial differences in founding population⁵⁴, that may persist for months⁵⁶ (Figure 3). In many host species, paternal contributions to offspring traits have been well documented⁵⁷,58; these observations have been extended to the microbiome, where paternal contributions to offspring H. pylori allele composition have been shown⁵⁹, In any event, there is evidence for extensive horizontal gene transfer (HGT) within human populations, functional classes, and ecological niches⁶⁰, indicating the site-specificity and dynamism of selection on the human microbiome. Even so, microbial inheritance can provide important confirmation of human ancestry⁶¹.

Figure 3

Acquisition of the microbiome in early life by vertical transmission and factors modifying mother-to-child microbial transmission

Through live-birth, mammals have important opportunities for mother → child microbial transmission, via direct-surface contact. However, many modern practices can reduce the organism and gene flow; several examples are illustrated. After initial introductions, there is strong selection by hosts for microbes with specific phenotypes, consistent with the extensive conservation shown in Figure 1. Acquisition is modified by offspring genetic and epigenetic differences (with respect to both maternal and paternal genes) that inform the competition for host resources by the vertically transmitted and environmentally acquired microbes. Ancestral organisms that have particular tissue- and niche-specific adaptations facilitate tissue tropisms and are selected, explaining the conserved niche-specificity compositions.

In Drosophila melanogaster, microbial influences have an effect on mating preference for >30 generations⁶²; can microbiome composition affect mating in humans? Odor is one means to affect mating preference, since human axillary and oral odors are largely influenced by microbial products, especially mercaptans⁶³. In general, the greater the force of mating preference, the more likely those populations become sexually isolated⁶⁴,65; this could affect tribal differentiation and other ethnic differences in humans. We speculate that metagenome composition may have affected mating preference in humans, representing another phenotype under strong selective pressure.

Inheritance of microbiota

The congruent phylogenies of mammals and their microbiota⁸ provide strong evidence for the inheritance of the microbiota⁷ Although inheritance of the microbiota from the father is presently little studied, increasing evidence supports inheritance from the mother³⁴,53. Until the amniotic sac ruptures, a fetus is considered to be sterile, or essentially sterile. Immediately after vaginal delivery, founding microbial populations in the baby closely resemble that of their mother’s vagina⁵⁴, with lactobacilli predominating. Since lactic acid-producing bacteria dominate in both vagina and in mother’s milk, the initial gastro-intestinal tract bloom of lactobacilli in the baby cannot be considered accidental. Lactobacilli represent the pioneer community in mice⁵⁵ as well as humans³⁹, preparing the gastrointestinal tract for subsequent microbial successions, until microbial maturity is reached.

The multiple opportunities for the microbiota to be transferred from a mother to her baby may be disrupted by modern lifestyles. Caesarian section instead of vaginal delivery is an obvious example of the potential impact of medical practice on microbiota composition, with substantial differences in founding population⁵⁴, that may persist for months⁵⁶ (Figure 3). In many host species, paternal contributions to offspring traits have been well documented⁵⁷,58; these observations have been extended to the microbiome, where paternal contributions to offspring H. pylori allele composition have been shown⁵⁹, In any event, there is evidence for extensive horizontal gene transfer (HGT) within human populations, functional classes, and ecological niches⁶⁰, indicating the site-specificity and dynamism of selection on the human microbiome. Even so, microbial inheritance can provide important confirmation of human ancestry⁶¹.

Figure 3

Acquisition of the microbiome in early life by vertical transmission and factors modifying mother-to-child microbial transmission

Through live-birth, mammals have important opportunities for mother → child microbial transmission, via direct-surface contact. However, many modern practices can reduce the organism and gene flow; several examples are illustrated. After initial introductions, there is strong selection by hosts for microbes with specific phenotypes, consistent with the extensive conservation shown in Figure 1. Acquisition is modified by offspring genetic and epigenetic differences (with respect to both maternal and paternal genes) that inform the competition for host resources by the vertically transmitted and environmentally acquired microbes. Ancestral organisms that have particular tissue- and niche-specific adaptations facilitate tissue tropisms and are selected, explaining the conserved niche-specificity compositions.

In Drosophila melanogaster, microbial influences have an effect on mating preference for >30 generations⁶²; can microbiome composition affect mating in humans? Odor is one means to affect mating preference, since human axillary and oral odors are largely influenced by microbial products, especially mercaptans⁶³. In general, the greater the force of mating preference, the more likely those populations become sexually isolated⁶⁴,65; this could affect tribal differentiation and other ethnic differences in humans. We speculate that metagenome composition may have affected mating preference in humans, representing another phenotype under strong selective pressure.

Postnatal influences on the microbiota

Over a lifetime, each human develops a densely populated microbiome that is recapitulated in every individual and in every generation. The eruption of teeth is responsible for major successions in the oral microbiota⁶⁶,67, suggesting that succession may be a general property of microbiome dynamics in humans. In mice, this clearly occurs in the GI tract⁶⁸. Exposure (or not) to environmental microbes is another important but highly variable reservoir for the resident microbiota. Early life antibiotic use produces major shifts in both microbiota characteristics and in host developmental phenotypes, both on the farm⁶⁹ and in experimental animals⁷⁰,71. Whether such precedents are applicable to human children is unknown, but seems likely. If so, then both the timing of microbiome succession and the specific organisms present may affect development. The concept of time-dependent compositional variation affecting host immunological, metabolic, cognitive, and reproductive development is a potentially important and testable hypothesis. We further speculate that nature orchestrates microbiome development so to optimize fecundity, reaching a climax state at or near parturition to maximize success for the next generation. The noted heterozygote advantage for fecundity⁷² may be an analogue to a genetically diverse microbiota.

Microbiome dynamics in adults

Our knowledge of microbiome dynamics, especially age-related changes, during human adulthood is limited. The older literature (predating the use of HTS), clearly shows that the post-menopause vaginal microbiota differs substantially from that during the reproductive period⁷³,74. Similarly, in the stomach,, the age-related progressive development of gastric atrophy (enhanced by H. pylori presence⁷⁵,76) selects for gastric microbiota that are substantially different from the norm ⁷⁷. Analogous changes may be occurring in other body sites as senescence advances. In the gut, the ratio of Bacteroidetes to Firmicutes changes with age⁷⁸.

These concepts are particularly relevant to oncogenesis, which is generally age-related. In the multi-step Nordling hypothesis of oncogenesis⁷⁹, 4–6 somatic cell mutations are needed for cancer development. We propose that age-related microbiota shifts contribute to this multi-step process. Residential microbes can contribute to somatic mutagenesis by causing genotoxicity secondary to inflammation, increased cell proliferation, and production of pro-mutagenic metabolites (e.g. butyrate)⁸⁰. Genes may have alternative effects at different life stages, illustrating the idea of antagonistic pleiotropy⁸¹. We hypothesize that specific human microbiota and their genes that are beneficial early in life may be harmful later in life. The dominant gastric bacterium, H. pylori provides an example:early in life, inflammatory responses to the organism improve control of infection⁸²,83 and allergy⁸⁴, but later in life promote atrophy and oncogenesis⁸⁵. A related hypothesis is that co-evolved microbiota are adaptive for the human species both by supporting early-in-life host functions and by leading to later-in-life host demise ⁸⁶.

Disease links and health implications

How, then, does the microbiome affect human health? Current studies focus on describing the variant microbe populations that appear in specific disease states or the temporal microbial changes over the course of a disease. For many conditions, the challenge is to discover whether there is a causal link between microbiome variation and significant pathology. Limitations in the definitions and stratification of clinical syndromes, including irritable bowel syndrome (IBS) and non-ulcer dyspepsia (NUD), unfortunately reduce the potential of microbiome studies. The diseases listed below review some current recent investigations (Table 1). Investigations are preliminary, but some observations are promising.

Cutaneous microbiome

The cutaneous microbiome is an obvious target in specific diseases, such as psoriasis, a chronic, idiopathic inflammatory dermatologic condition⁸⁷. In studies predating HTS, the use of PCR and cloning led to observations that Firmicutes were significantly over-represented and Actinobacteria were significantly under-represented in psoriatic lesions compared to both unaffected skin in psoriasis patients and in normal controls⁸⁸. Studies to explore these findings using HTS are currently underway⁸⁹. Atopic dermatitis, another chronic inflammatory condition, has increased in incidence approximately three-fold over the last 30 years in industrialized countries, suggesting a potential role for microbiome alterations. Classic atopic dermatitis occurs in areas, such as the antecubital and popliteal fossae, with similar microbial populations⁸⁹, suggesting a microbiome role. Similarly, Propionibacterium acnes has been implicated in the common dermatologic condition, acne. Although P. acnes thrives in the cutaneous pilosebaceous units and secretes enzymes that cause local injury and inflammation, and is widely accepted to have a function in acne development⁹⁰, continuing investigations also are examining other microbes in the development of acne. Chronic skin ulcers, often secondary to venous stasis or diabetes, lead to substantial morbidity. Cutaneous microbiome shifts, such as an increased abundance of Pseudomonadaceae in patients with chronic ulcers treated with antibiotics and an increased abundance of Streptococcaceae in diabetic ulcers have been noted⁹¹. Such shifts may interact with aberrantly expressed host cutaneous defense response genes⁹², thereby increasing disease risk.

Gastric microbiome

The discovery that H. pylori was adapted to survive in the acidic gastric environment overturned the dogma that the stomach is sterile. In H. pylori-negative individuals, gastric microbiota diversity is high; most of the prominent gastric phylotypes (Streptococcus, Actinomyces, Prevotella, Gemella) also are abundant in the oropharynx of these individuals⁹³; indicating that either many constituents are swallowed from more proximal sites, or that close relatives of the oral microbiota colonize more distally. In contrast, among H. pylori-positive persons, H. pylori usually accounts for >90% of sequence reads from the gastric microbiota⁹³, dramatically reducing the overall diversity of this microbiota. The ability of H. pylori to dominate indicates an evolved fitness for that specialized niche. H. pylori is a classical amphibiont; the presence (or absence) of an H. pylori dominated gastric microbiota is strongly associated with particular diseases with important age-related differences⁸⁵. Its presence increases risk for developing peptic ulcer disease, gastric mucosa-associated lymphoid tissue (MALT) tumors, and gastric adenocarcinoma⁹⁴ but also is associated with decreased reflux esophagitis⁹⁵ and childhood-onset asthma⁹⁶; demonstrating the complex biological interactions with our microbiota.

The colonic microbiota and colorectal cancer

The colonic microbiota has been suspected for a long time to be involved in the development of colorectal cancers⁹⁷, possibly by synthesizing short-chain fatty acids (SCFAs) and other metabolites. SCFAs, in particular butyrate, may induce apoptosis, cell cycle arrest, and differentiation, via Wnt signaling⁹⁸. Microbes may also be genotoxic to colonic epithelial cells, as demonstrated by the induction of aneuploidy and tetraploidy by Enterococcus faecalis⁹⁹. The colonic microbiota also might promote colorectal cancer by eliciting host responses, for example, by stimulating exaggerated immune responses, potentially via Th17 cells⁹⁹.

Further evidence of a link between colonic microbiota and colorectal cancer is suggested by the ability of antibiotic administration to alter the composition of the colonic microbiota and to affect the expression of host genes involved in cell cycle regulation, reducing epithelial proliferation¹⁰⁰. Early studies evaluating specific microbes were limited to identifying culture-dependent species, such as Streptococcus bovis, but could not adequately assess anaerobic constituents. However, members of the anaerobic genus Fusobacterium have recently been associated with colorectal cancer; whole genome sequences of Fusobacterium species were compared between cases and matched controls using both quantitative PCR analysis and HTS¹⁰¹,102. Fusobacterium nucleatum is a mucosally adherent, pro-inflammatory microbe that was first identified in the mouth¹⁰³. In colorectal cancer samples, F. nucleatum sequences were significantly enriched compared to samples obtained from control subjects, while both Bacteroidetes and Firmicutes were relatively depleted in those with Fusobacterium-rich malignancies¹⁰². The enrichment of Fusobacterium species (not limited to F. nucleatum) was confirmed when evaluating the mucosal microbiome of colorectal cancers compared to adjacent normal tissues from 99 subjects¹⁰¹. However, the causal direction of the association has not yet been ascertained.

The colon microbiota and inflammatory bowel disease

The microbiome is essential for the activation of host immune responses ¹⁰⁴. For example, Th17 cell differentiation in the mouse lamina propria requires the presence of segmented filamentous bacteria (SFB)¹⁰⁵, and polysaccharide A produced by Bacteroides fragilis mediates conversion of CD4+ T cells into regulatory T cells¹⁰⁶. The inflammatory bowel diseases have long been considered to reflect interactions between microbes and the host. IBD susceptibility is associated with host polymorphisms in bacterial sensor genes such as nucleotide-binding oligomerization domain-containing protein 2 (NOD2) also known as caspase recruitment domain-containing protein 15 (CARD15)¹⁰⁷,108 and toll-like receptor 4 (TLR4)¹⁰⁹, and IBD patients sometimes improve following antibiotic treatment¹¹⁰. Early childhood antibiotic exposure has been associated with significantly increased risk for Crohn’s disease¹¹¹, suggesting that gut microbiome perturbations may be critical for disease risk. Microbial diversity is significantly diminished in Crohn’s disease¹¹², suggesting decreased gut microbiome resilience that could affect immune interactions. Gut microbiome population structures of patients with ulcerative colitis or Crohn’s disease¹⁹ depart from normality, but remain clustered by disease within their characteristic deviated patterns. Specific bacteria of the Enterobacteriaceae genus may act together with a disordered microbiome to increase the risk of ulcerative colitis¹¹³. Among twins discordant for ulcerative colitis, those affected had significantly reduced bacterial diversity, but increased proportions of Actinobacteria and Proteobacteria¹¹⁴. Patients with Crohn’s disease have over-representation of E. faecium and of several Proteobacteria compared to controls¹¹⁵. The microbial patterns observed for the conditions described above are preliminary and their specificity and causal direction have not been established.

The gut microbiota and diseases of the liver

The gut microbiota may be involved in hepatologic conditions, including non-alcoholic fatty liver disease (NAFLD)¹¹⁶, alcoholic steatosis, and hepatocellular carcinoma. The liver is the first solid organ exposed to the metabolic products generated by the gut microbiome, including acetaldehyde, ammonia, and phenols. Compared to germ-free mice, the presence of a microbiome in conventional mice leads to suppression of intestinal epithelium angiopoietin-related protein 4, which inhibits lipoprotein lipase, increasing downstream triglyceride accumulation in the hepatic parenchyma and adipocytes¹¹⁷. Chronic ethanol exposure disturbs the gut microbiome¹¹⁸,119, but roles for the microbiome in steatosis are unresolved. Particular murine colonic commensals (e.g. H. hepaticus) promote the development of hepatocellular carcinoma¹²⁰. Patients with cirrhosis have a substantially altered microbiome, including community-wide changes at multiple taxonomic levels, with enrichment of Proteobacteria and Fusobacteria (phyla), and Enterobacteriaceae, Veillonellaceae, and Streptococacceae (family) ¹²¹. Although many observations suggest links between microbiome composition and liver disease, definitive associations in humans are lacking.

The gut microbiota and pbesity

Genetically obese (ob/ob) mice have decreased Bacteroidetes/Firmicutes ratios compared with lean (ob/+ and +/+ wild-type) siblings³¹. Transplantation of gut microbiota from the obese (ob/ob) to germ-free mice conferred an obese phenotype, demonstrating transmissibility of metabolic phenotypes¹⁷; the transferred microbiomes had increased capacity for energy harvest. In humans, relative Bacteroidetes proportions increase with weight loss¹²². In mono- and dizygotic twins, obesity was associated with decreased Bacteroidetes and diminished bacterial diversity, with enrichment of genes related to lipid and carbohydrate metabolism. Despite substantial taxonomic variation, functional metagenomic differences were minor³⁷. Changing exposures/selection pressures on the early-life microbiome may contribute to development of obesity. Antibiotic use in human infancy, before age 6 months, was significantly associated with obesity development¹²³. In contrast, perinatal administration of a Lactobacillus rhamnosus GG-based probiotic decreased excessive weight gain during childhood¹²⁴. These early studies provide support for the concept that microbiota alterations could lead to childhood-onset obesity, which could be modifiable. Alterations in the gut microbiome also occur with interventions used to treat obesity. Roux-en-Y surgery significantly increases levels of Proteobacteria, while altering specific metabolic markers, such as the production of urinary amines and cresols¹²⁵.

The gut microbiota and rheumatoid arthritis

Disregulation of host responses secondary to dysbiosis within the gut lumen could affect distant anatomical sites through activation of host immune responses. This may be a mechanism in rheumatoid arthritis, another chronic idiopathic inflammatory condition. In mice, the presence of SFBs in the gut microbiome causes local expansion of Th17 cells¹²⁶ that then migrate to peripheral immune compartments and activate B cells into antibody-producing plasma cells. Antibody production leads to immune-mediated destruction of the joints that mirrors rheumatoid arthritis¹²⁷.

Cause or effect?

Microbiome analysis in humans has been largely based on observation, with associations of disease phenotypes with particular microbiota constituents. But which is causal: does factor A cause factor B, does factor B cause factor A, or does factor C cause both A and B? Hill developed criteria to address the questions: “In what circumstances can we pass from this observed association to a verdict of causation? Upon what basis should we proceed to do so?”¹²⁸ The criteria include the strength of association, its consistency, specificity, temporality, and biological plausibility, and whether biological gradients are present, experimental support exists, and support can be extrapolated from known causal relationships. Although these criteria were advanced largely to unravel epidemiological relationships, they are applicable to genetics as well and to metagenomics, in particular. Sometimes successful treatment trials with amelioration or cure of a particular condition provide the critical evidence for a causal relationship. The changed natural history of peptic ulcer disease following elimination of H. pylori¹²⁹ demonstrated its pathogenic role.

For understanding causation and pathogenesis, model organisms provide an important approach. Animal models approximate some human diseases (e.g. asthma, atherosclerosis), but many diseases are not well reproduced (e.g. psoriasis). For diseases that can be studied in model organisms, microbiota roles can be explored within the constraints of particular model systems (Table 2). Standard models of inbred mice are limited by their uncontrolled microbiome diversity. Certain disease states are well-studied in these models, such as the effects of SFBs on Th17 development or the susceptibility to type I diabetes in non-obese diabetic (NOD) mice. The use of gnotobiotic mice eliminates the above-mentioned microbiome variability, but the animals are expensive and require specialized facilities and expertise limiting their widespread use. The recent availability of gnotobiotic animals from commercial sources permits conventionalization of animals with experimental or control microbiota without needing xenobiotic facilities; such approaches allow for the direct observation of microbiota effects on the host. The extension of this concept to humanized model organisms¹³⁰ allows better approximation of the effects of the human microbiome on disease processes in tractable animal models.

Perspectives

Inherent complexities in the composition of the microbiome may preclude investigations of microbe-associated diseases using classical approaches such as Koch’s postulates. Instead of single organisms associated with disease, community characteristics (composition and metagenomic functionality) may be more relevant. The principles of host interaction with pathogens and commensals contain many parallel features, which can help tutor the new field, but the nature of the selection for commensalism is more complex and highly dynamic. The scale of the interface suggests that microbiome–host interactions have important bearings on disease susceptibility, and the microbial effects on the balance of host metabolism and immunity¹³¹ provides an excellent model for the broader phenomenon of disease susceptibility. Modifying disease risk by altering metabolic, immunological, or developmental pathways are obvious strategies.

Given the ongoing extinction of our ancient commensal organisms, the future of a healthy human microbiome may include restoration of our ancestral microbial ecology. There are two types of restoration. The first involves restoring ancient organisms (or pathways) in healthy hosts lacking them, as prophylaxis against future imbalances. The second type of restoration could be therapeutic, when the etiologic extinctions and imbalances are recognized. This scientific frontier will require understanding the biology of re-introductions, as well as developing microbial breeding programs. In addition to the technical problems associated with restoring particular organisms to specific hard-to-reach niches, such as the distal ileum, there also will be substantial biological problems related to understanding how reintroductions affect the population structure of the extant organisms, and host interactions.

To better understand the implications of microbiota and metagenome variation in human health and disease, the field needs improved informatics tools, including new approaches for understanding the complexity of the metadata¹³². The multidimensionality of the human and microbial phenotypes and the dynamic, non-linear interactions challenge deterministic solutions. For example, in analyses of 16S-defined operational taxonomic units (OTU) populations in mice receiving traditional or Western-type diets, Reshef et al., examined top-scoring non-linear abundance relationships¹³³. These often involved “noncoexistence”, sometimes related to known factors (e.g. diet, host gender), but were often unexplained. Although incomplete, such work leads to new approaches to understand the underlying complexity.

We also will need new tools to implement the health implications presented by metagenomic analyses. Although the known principles of evolution and genetics apply, study of the microbiome provides new applications, and will lead to new understandings of complex traits. Important questions to pursue are listed in Box 2. As such, this is a frontier for human preventive medicine, and for medical management of chronic diseases.

Acknowledgements