The Human Microbiome: Our Second Genome^*

Amplicon-Based Surveys of Human Microbial Communities

High-throughput surveys of microbial communities have been greatly enabled by culture-independent identification methods coupled with advancement in genomic technologies, especially DNA sequencing technology (Figure 1). The earliest human microbial surveys were based upon fingerprinting techniques, which relied upon physical separation of 16S rRNA genes by denaturing gradient gel electrophoresis and terminal restriction fragment length polymorphism analyses (49, 56). Greater specificity was gained with Sanger sequencing of the amplified and cloned 16S rRNA gene. The longer read lengths produced by Sanger sequencing are still a benefit to studies in the microbial community discovery phase. However, the advent and commercialization of highly parallel DNA sequencing technology have revolutionized our ability to quickly and precisely characterize microbial populations at a much lower cost and a greater depth than Sanger sequencing can provide.

At present, many researchers in the field rely upon the Roche/454 pyrosequencing platform for 16S rRNA gene sequencing, which produces ~1 million ~400-nucleotide reads per run. Incorporating bar codes into the 5′ primer sequence multiplexes samples into a single sequencing run and enables researchers to produce thousands of 16S rRNA sequences per sample. If the goal of the study is to distinguish species within a genus (e.g., Staphylococcus epidermidis versus the pathogenic S. aureus), then longer read lengths (> 300 nucleotides) are necessary. As read lengths of the 16S rRNA gene decrease, so does taxonomic precision, and thus the ability to distinguish between strains, species, and even genera. However, for some studies researchers have preferred the greater sampling depth offered by Illumina platforms, and have explored paired-end sequencing to mitigate the shorter read length. Caporaso et al. (10) sequenced 1,967 microbiome samples across six lanes in a single run of the Illumina Genome Analyzer IIx. Balancing sequence length and sampling depth is one of the moving targets for the field, and the appropriate balance for a particular study is clearly shaped by the overall objectives. In general, the final decision of which sequencing platform to utilize depends upon the question being posed.

In addition to the choice of sequencing platform, investigators must consider many associated pitfalls when embarking on a bacterial survey study. Although the next-generation sequencing technologies eliminate cloning biases associated with Sanger sequencing, 16S rRNA gene diversity surveys still rely upon limited PCR amplification, which can potentially introduce bias. Known bias can derive from the design of the PCR primers or from PCR conditions that can cause amplification bias and chimera formation (102). Many of these study design issues and their analysis outcomes were recently reviewed by the Knight and Schloss groups (41, 79).

Microbial Genome Sequencing

Clearly, the use of next-generation sequencing has revolutionized the way that microbial diversity surveys are performed, but this technology has also had a tremendous impact on deciphering individual microbial genomes. In 1995, Haemophilus influenzae became the first bacterial genome sequenced with Sanger sequencing, introducing whole-genome random sequencing (85). At 1.8 Mb, the H. influenzae genome was 10 times larger than previously sequenced viral genomes. However, generating this genome assembly required the equivalent of three years of Sanger sequencing. (It was in fact completed by 14 machines over three months.) Then, with the advent of nextgeneration sequencing, generating this amount of data required only one instrument and one week. The ability to bypass cloning of DNA into bacteria for shotgun sequencing was also a major turning point in obtaining better coverage. Promoter regions of bacterial genes are to some degree uncloneable, and as a result these regions are typically underrepresented in initial data sets generated from cloning-based Sanger sequencing. As of February 2012, the Genomes Online Database (GOLD; http://www.genomesonline.org) provided nearly 3,000 finished bacterial genomes, with many thousands more underway.

Technological advances now make sequencing a single bacterial genome seem trivial and the potential to sequence thousands realistic. One of the initiatives of the National Institutes of Health (NIH) Common Fund’s Human Microbiome Project (http://commonfund.nih.gov/hmp) is to sequence 3,000 cultivated and uncultivated bacterial reference strains. This catalog of reference genomes is intended as a scaffold for the assembly of metagenomic sequences and as a reference for 16S rRNA gene sequences (69). In May 2010, 178 bacterial genomes were published, with the great majority isolated from the gastrointestinal (GI) tract (58). From this work and others, it is clear that only the tip of the iceberg of species diversity has been uncovered.

Targeted sequencing projects have focused on the pan-genome—the collection of genes found across all members of a species—as a useful framework for describing genomic diversity within a taxon. For example, Tettelin, Fraser-Liggett, and colleagues (92) characterized the pan-genome of Streptococcus agalactiae, an important pathogen for newborn infants, and found that ~20% of any given genome is made up of genes that are shared only partially with other strains. Furthermore, they found that sequencing additional genomes was predicted to increase the size of the pan-genome, indicating that S. agalactiae has an open genome. This type of analysis has been used to characterize a number of bacterial species as well as genera and has been used to estimate the pan-genome size of all bacteria.

Recent studies foreshadow a transition in the near future to sequencing technologies that can provide even faster turnaround for microbial genomics and microbiome studies. During the recent Escherichia coli outbreak in Germany, the Ion Torrent Personal Genome Machine proved its utility when a draft genome of the outbreak strain was produced in three days, furthering efforts to determine the evolutionary origins and pathogenic potential associated with the strain (76).

Whole-Genome Shotgun Metagenomic Sequencing

Highly parallelized DNA sequencing, such as that offered by the Illumina HiSeq, has paved the way for whole-genome shotgun (WGS) metagenomic analysis of microbial communities (Figure 1). Metagenomics provides the potential to analyze microbial communities from the viewpoint of functionality while circumventing both PCR and cloning bias. WGS metagenomics is a powerful alternative to sequencing 16S rRNA genes because it can in part answer not only the question “Who is there?” but also “What can they do?” The general strategy for a metagenomic study (Figure 1) is to directly extract DNA from a clinical sample (such as feces) and then construct a library of inserts that are multiplexed and sequenced. Tens of millions of Illumina reads are typically generated to articulate the full genetic potential of a sample. Unassembled reads that match signature genes (such as 16S rRNA) can be classified by searching against databases (such as greengenes or RDP-II) to determine the taxonomic makeup of the metagenomic data set. 16S rRNA read counts are used to calculate the frequency of microbial species, and differences between samples are a measure of relative microbial abundances.

However, individual reads are too short to generate the sequence of a complete gene and thus predict a protein-coding function. Reads are therefore assembled into contiguous DNA fragments to provide maximum information. However, there are often multiple highly similar genomes present in a WGS metagenomic data set. The sheer volume of sequence data generated combined with incomplete catalogs of sequenced reference genomes (bacterial, viral, archaeal, eukaryotic) makes WGS metagenomics an extremely computationally intensive undertaking. Sequence reads can be assembled as much as possible using tools such as SOAPdenovo (47), as longer reads provide more accurate gene annotation and phylogeny prediction. To identify putative protein-coding genes, a BLASTX search can be performed against either a nonredundant database or a database of microbial genome sequences specific to the environment/site of interest (e.g., the gut).

Functional data can then be inferred from the predicted proteins by searching against the KEGG (Kyoto Encyclopedia of Genes and Genomes) (38) and COG (Clusters of Orthologous Groups) (91) pathway databases. MG-RAST (Metagenomics Rapid Annotation Using Subsystem Technology), a Web-based metagenomics analysis server, has implemented many of these functions into an open-source metagenome analysis pipeline that combines normalization, alignment, functional and taxonomic assignment, and subsystem reconstruction into an automated workflow (53). MEGAN (Metagenome Analyzer) is a similar tool that performs taxonomic classification, functional analysis using KEGG or SEED classification, and computational comparison of metagenomes, all in an interactive analysis and visualization platform (55) (Table 1). It should be noted that reference databases (e.g., phylogenetic, pathway, and functional databases) are highly biased toward those organisms that are readily cultivable and those genes with known functions.

The Gastrointestinal Tract: Readily Accessible with Abundant Microbiota

The GI tract contains the bulk of microbiota associated with the human body and has been one of its most thoroughly examined ecosystems. Fecal samples are commonly used for these analyses because they are easy to obtain and contain a large amount of biomass. The gut microbiome is quite low in diversity at higher phylogenetic levels, comprising primarily the bacterial phyla Firmicutes and Bacteroidetes (21), but contains great diversity at lower phylogenetic levels (species, strains) (45). One study suggested that more than 5,000 bacterial taxa may reside in the gut (19).

Initial microbial colonization of the gut in infants appears to be dependent on delivery mode; vaginally delivered babies acquire microbiota similar to those of their mother’s vagina (i.e., dominated by Lactobacillus and Prevotella), and babies delivered via Caesarian section acquire microbiota similar to those typically associated with the skin (i.e., Propionibacterium, Staphylococcus, and Corynebacterium) (20). Furthermore, there were significant differences between the gut microbiota of formula-fed and breast-fed infants, with some of those bacteria common in formula-fed babies being associated with a higher prevalence of antibiotic use, hospitalization, and prematurity (64). The gut microbiota eventually converge toward an adultlike profile during the first year of life (61).

Even more abundant than bacteria are the viruses that infect them—the bacteriophages. In their lysogenic phase, bacteriophages can integrate into the bacterial genome and provide additional genetic diversity. Virome analysis of the gut revealed that the major type of variation was interindividual, but significant changes were observed when the host was placed on a defined diet (54). The diet-induced changes in the gut virome covaried with changes in the gut bacterial community, and gut virus populations converged in individuals placed on similar diets. This is in sharp contrast to another study that found the gut virome to be stable over time (72). It is unclear what accounts for the differences observed between these two studies, but this example highlights the need for standardization in study design and result reporting.

The GI tract was one of the first human ecosystems to be examined by WGS metagenomic analysis. Early WGS metagenomic sequencing and analysis of the fecal matter of two individuals identified the metabolic potential of the gut microbiota by assigning function to Sanger sequence reads (31). The MetaHIT (Metagenomics of the Human Intestinal Tract) Consortium, funded by the European Commission and comprising 13 academic and industry partners, has been a key leader in the gut microbiome and metagenomics arena. Qin et al. (70) described a WGS metagenomic analysis of 124 Europeans that used Illumina Genome Analyzer technology and generated over 576 Gb of metagenomic sequence reads. Using this data set, they described the core gut metagenome, which consists of genes essential for host-microbe interactions such as those that degrade complex polysaccharides and those that synthesize short-chain fatty acids, vitamins, and amino acids. Even so, it should be noted that only ~12% of the genes derived from this sequence data set mapped to reference genomes. As more reference genomes are sequenced—for example, as part of the MetaHIT Consortium and NIH Human Microbiome Project initiatives—these numbers should improve. WGS metagenomic analysis of this data set and others subsequently demonstrated the existence of three distinct enterotypes—i.e., groups of individuals defined by the composition of their gut microbiota (3). The enterotypes were neither nation nor continent specific and could not be explained by body mass index, age, or gender. This study provided the first indication that the composition of the human gut microbiota is stratified and not continuous.

Because a fecal sample is theoretically a composite of bacteria collected throughout the length of the GI tract, one challenge lies in the analysis of spatial microbial diversity in the GI tract, for which there is very little data. One of the first genomic analyses of gut microbiota utilized 16S rRNA gene sequencing to begin to define the intrapersonal (spatial) and interpersonal variation of the adherent mucosal and fecal microbiota (21). Much of the sequence diversity along theGI tract was novel and revealed spatial differences not captured by a fecal sample. This is to be expected, as the host tissue varies greatly throughout the length of the GI tract, dictating distinct microenvironments depending on dominant cell types and functions.

This leads to the question of what determines the microbiota that colonize the gut. In addition to host- and tissue-specific factors, there are certainly environmental factors that must play some role in selecting for those microbes that constitute the gut microbiota. Wu and colleagues (108) demonstrated that gut enterotypes are strongly correlated with long-term dietary patterns, especially high-fat, high-protein diets as compared with high-carbohydrate diets. Gordon’s group (26) definitively demonstrated a relationship between both the total abundance and the relative levels of bacteria evoked by diet perturbations in mice colonized with defined bacterial communities. Genetic factors are also implicated in microbial gut colonization, with initial studies carried out in animal models. For example, studies in advanced intercross mice demonstrated that host genotype had a large influence on gut microbiota composition, independent of litter and cohort effects (5). In fact, a relative abundance of gut microbial taxa was associated by genome-wide linkage with 18 quantitative trait loci, which demonstrates the power of the host genotype in shaping microbial diversity.

Interesting features of the gut microbiome and its role in disease are beginning to emerge, especially its role in obesity, diabetes, and metabolic disease. Perhaps the best-known example is the contribution of gut microbiota to obesity. Although human geneticists continue to search for host variation tied to obesity, it is now appreciated that the gut microbiota of obese individuals are significantly altered and carry a greater capacity for energy harvest (95). Animal models have proven especially useful in demonstrating mechanistic and functional linkages of the gut microbiota to disease. An animal model deficient in Toll-like receptor 5 (Tlr5)— an important component of innate immunity and infection control in the gut—exhibits hallmarks of metabolic syndrome (adiposity, insulin resistance, hypertension, hyperlipidemia) and altered gut microbiota as compared with healthy control mice (99). Additionally, when transferred to germ-free mice, the gut microbiota of Tlr5-deficient mice were sufficient to recapitulate features of metabolic syndrome.

Interaction of the innate immune system with the gut microbiota was demonstrated to be a critical epigenetic factor governing the development of type 1 diabetes in nonobese diabetic (NOD) mice. Specific-pathogen-free NOD mice deficient in MyD88—an adaptor protein essential to multiple innate immune receptors and signaling pathways—do not develop diabetes. This effect appears to be dependent on the gut microbiota, as germ-free Myd88^−/− NOD mice develop robust diabetes. However, diabetes was attenuated in germ-free Myd88^−/− NOD mice colonized with defined microbiota from donor specific-pathogen-free Myd88^−/− NODmice (101). This is yet another example of how interactions between the microbiota and the host govern predisposition to disease.

Studies in Drosophila have provided insight into the complex interactions between host and microbe in the gut. In Drosophila, the homeobox gene Caudal represses the NF-κβ-dependent signaling of antimicrobial peptides, which in turn regulates host-microbe mutualism (77). RNA interference knockdown of Caudal results in altered gut microbiota and eventually gut cell apoptosis and death. In a separate study, the commensal bacterium Acetobacter pomorum proved to be a key link in the regulation of Drosophila development, body size, metabolism, and intestinal stem cell activity (84). A. pomorum–derived, pyrroloquinoline quinone–dependent alcohol dehydrogenase activity modulates Drosophila insulin/insulin-like growth factor signaling.

In humans, beyond aiding with digestion, bacteria were recently found to play a role in cancer. Colorectal carcinoma is associated with enriched abundance of Fusobacterium (11, 40), an invasive anaerobic bacterium that was previously associated with inflammatory bowel disease but is a rare constituent of the healthy gut microbiota. Though the exact role of Fusobacterium in colorectal carcinoma remains unclear, it is capable of eliciting a host inflammatory response, a known risk factor for colorectal cancer (52, 109). Additionally, Fusobacterium abundance was positively associated with lymph node metastasis (11). More work is needed to define the mechanism by which Fusobacterium is related to colorectal carcinoma, but a more immediate value may be in exploiting it as a marker for colorectal cancer presence, risk, or prognosis.

Finally, the role of the gut microbiota through their metabolic potential extends far beyond the intestine to the metabolism of systemic drugs and disease manifestation in other organ systems. Nicholson’s group (12) deciphered the first example of metabolomics intersecting with the microbiome, showing a clear link between an individual’s ability to metabolize acetaminophen (most commonly known in the United States under the brand name Tylenol) and bacterial metabolic state. Acetaminophen has been one of the most commonly used nonprescription medicines for decades, so one might think that its toxicology and metabolism would be well understood. However, Nicholson’s study showed a novel and striking association between an individual’s predose metabolite profile and his or her specific metabolism and excretion of acetaminophen. As these pathways described for acetaminophen impact the metabolism of many drugs, the gut bacterial metabolism might have a large influence on both drug-induced responses and disease development.

Microbiome profiling clearly has a role in personalized medicine that extends beyond the variation in drug-metabolizing enzymes. Hazen’s group (100) made the first link between diet, gut bacteria, liver metabolism, and atherosclerosis leading to cardiovascular disease. Their initial experiments screened for plasma metabolites in patients who had experienced a heart attack or stroke. While testing this metabolic cascade in animal models, they found dramatic differences in mice treated with broad-spectrum antibiotics. They teased apart this mechanism by examining atherosclerosis development in mouse models of human metabolic disorders that were placed on restricted diets and then treated with antibiotics. The hunt is now on to find a probiotic approach to selectively eliminate the gut microbes that process metabolites derived from high-fat foods before they are delivered to the liver and ultimately deposited in blood vessels. These studies clearly show the vast unexplored role of the gut microbiota in regulating human health beyond simply intestinal disorders.

The Oral Microbiome

In general, the dominant bacteria of the oral cavity are streptococcal species, with common representation also from Veillonella, Gamella, Rothia, Fusobacterium, and Neisseria (1, 6). Many diseases of the oral cavity, such as dental caries (cavities) and periodontitis (gum disease), have long been suspected to be caused at least in part by microbes. A metagenomic analysis of dental caries demonstrated that consortia of microbiota inhabit the caries and that these microbes differ functionally from those inhabiting the healthy oral cavity (4). Strikingly, those who never suffered from caries were colonized with microbiota enriched for genes encoding antimicrobial peptides and quorum-sensing molecules. In periodontitis, Porphyromonas gingivalis, an anaerobic bacterium, has historically been the suspected etiological agent. Mechanistic work in germ-free mice demonstrated that only minute quantities of P. gingivalis induce quantitative and qualitative changes in the oral microbiota, not directly but rather via exploitation of the complement cascade to cause periodontal bone loss (34).

Bacteriophages have been another focus of oral microbiome research and have provided the greater microbiome community with scientific insight. Pride et al. (67) recently analyzed the salivary viromes of five healthy individuals and found that the vast majority of human oral viruses are bacteriophages. Many of the viral genes encoded virulence factors, and thus the oral virome may be a reservoir for oral pathogenicity factors. In this same study, comparisons of the salivary virome with gut and respiratory viromes showed that habitat is an important selection factor for the virome.

Genome-encoded clustered regularly interspaced short palindromic repeats (CRISPRs) are a bacterial defense mechanism against mobile genetic elements such as bacteriophages and conjugative plasmids. CRISPRs are acquired from the invading element, forming a heritable, adaptive record of prior infection. Bacterial cells express CRISPR RNA to interfere with invading nucleic acids (51). An analysis of streptococcal CRISPR sequences in the oral cavity revealed great diversity within individuals, suggesting that each individual was exposed to unique viral populations (68). Although much remains to be learned about CRISPR elements, the historical perspective they provide about bacteria strains may be useful for ecological and epidemiological studies and eventually personalized medicine.

Epidemiological studies have suggested a correlation between periodontitis and diseases that affect seemingly unrelated organ systems, e.g., diabetes and atherosclerosis. A common link that could explain this apparent association is the human microbiota, their priming of the immune system, or their long-term effects on inflammation. Metagenomic sequence analysis of the oral cavity revealed enrichment of Streptococcus mitis bacteriophage SM1–derived genes encoding platelet-binding factors, which are key virulence factors in endocardium infection (104). Koren et al. (39) demonstrated not only that bacteria were present in the atherosclerotic plaques, but also that the types and abundance of those bacteria correlated with the abundance of those same bacteria in the oral cavity. Furthermore, the abundance of several bacterial taxa in the oral cavity and the gut showed correlation with plasma cholesterol levels. These studies also illustrate the potential of the microbiome as a disease marker.

The Reproductive Tract Microbiome

Even before genomics arrived on the scene, it was suspected that the vaginal microbiota play a key role in the prevention of multiple diseases, including bacterial vaginosis (BV), yeast infections, sexually transmitted diseases, urinary tract infections, and human immunodeficiency virus. In a study of 396 reproductive-aged women, 16S rRNA gene sequence analysis revealed that vaginal bacteria profiles generally fall into one of five clusters. Four of the clusters were dominated by Lactobacillus species, whereas the fifth group had a higher proportion of anaerobic species and overall greater bacterial diversity (71). This group was also associated with a higher vaginal pH and a higher Nugent score, the latter of which is a cellular morphological indicator of BV. Several independent analyses of women with BV compared with healthy women showed the same trend of greater vaginal microbiota diversity associated with BV (48, 60, 88). A longitudinal study of the vaginal microbiota indicated that levels of lactobacilli fluctuate with menses, with a notable increase in Gardnerella vaginalis during menstruation; the authors hypothesized that this is due to increased iron availability because of the presence of vaginal blood during menses (88). Longitudinal analysis of women with BV during a treatment time course demonstrated that BV-associated bacteria were eradicated by treatment with the antibiotic metronidazole but tended to reappear following treatment.

The male genitourinary tract microbiome does not get the same attention as the vagina’s, but could be a very important factor in sexually transmitted infections. In an analysis of the male urine microbiota, anaerobic, fastidious bacteria were associated with sexually transmitted infections (57). In an analysis of pre- and postcircumcision microbial diversity in 12 African men, circumcision was associated with a decrease in anaerobic bacteria and overall decreased bacterial diversity of the coronal sulci of the penis (66).

The Skin Microbiome

Genomic approaches characterizing skin bacteria have revealed a much greater diversity of organisms than was previously apparent from culture-based methods (14, 27, 29, 32, 33). Our group and others have demonstrated that the physiology of the skin site is a strong determinant of the dominant colonizing bacteria, with specific bacteria associated with moist, dry, and oily microenvironments (14, 33). In general, bacterial diversity appears to be lowest in oily sites (such as the back and face) and highest in dry, exposed sites (such as the arms and legs). In general, intrapersonal variation in microbial community membership and structure between symmetric skin sites is lower than interpersonal variation, as determined by 16S rRNA gene sequencing (14, 29, 33).

There is a particular need to develop better methods for typing fungal and other microeukaryotic species, as these organisms are known to thrive on the skin and in some cases are associated with skin disorders. Analysis of 18S rRNA gene sequences has shown that the vast majority of fungal organisms residing on the healthy skin resemble Malassezia species, closely mirroring culture-based data, but the identity of more rare species may still prove important to understanding human disorders such as toenail infections and athlete’s foot (28, 62, 63). Demodex mites are also considered part of the normal skin microflora, residing in the sebaceous glands and hair follicles of the facial skin and increasing in prevalence with age (25, 35). Molecular methods for typing Demodex do not exist, but there is evidence suggesting a role for Demodex in some skin disorders such as rosacea (30, 42, 46).

WGS metagenomic data for skin communities would provide a fuller articulation of gene content and function to address basic questions such as the potentially beneficial role of skin bacteria to the human host. But although skin holds tremendous advantages for ease of sampling, the clear disadvantage lies in the difficulty of obtaining the critical threshold of starting material required for WGS metagenomic sequencing. Swabbing a 1-cm² area of skin typically yields only tens of nanograms of DNA, which is at the very low end of the amount required to make a next-generation sequencing library. And although skin cells undergo a linear program of terminal differentiation that limits the amount of human DNA present at the surface, unbiased whole-genome amplification needs to be further developed before skin WGS metagenomic sequencing can be tractable on a large scale. It seems that beyond the diseases that affect an organ system, every microbial community comes with its own technical advantages and limitations for genomic analysis.