OUP user menu

Colonization resistance and microbial ecophysiology: using gnotobiotic mouse models and single-cell technology to explore the intestinal jungle

Bärbel Stecher, David Berry, Alexander Loy
DOI: http://dx.doi.org/10.1111/1574-6976.12024 793-829 First published online: 1 September 2013


The highly diverse intestinal microbiota forms a structured community engaged in constant communication with itself and its host and is characterized by extensive ecological interactions. A key benefit that the microbiota affords its host is its ability to protect against infections in a process termed colonization resistance (CR), which remains insufficiently understood. In this review, we connect basic concepts of CR with new insights from recent years and highlight key technological advances in the field of microbial ecology. We present a selection of statistical and bioinformatics tools used to generate hypotheses about synergistic and antagonistic interactions in microbial ecosystems from metagenomic datasets. We emphasize the importance of experimentally testing these hypotheses and discuss the value of gnotobiotic mouse models for investigating specific aspects related to microbiota–host–pathogen interactions in a well-defined experimental system. We further introduce new developments in the area of single-cell analysis using fluorescence in situ hybridization in combination with metabolic stable isotope labeling technologies for studying the in vivo activities of complex community members. These approaches promise to yield novel insights into the mechanisms of CR and intestinal ecophysiology in general, and give researchers the means to experimentally test hypotheses in vivo at varying levels of biological and ecological complexity.

  • stable isotope labeling
  • ecophysiology
  • NanoSIMS
  • Raman microspectroscopy
  • pathogen
  • infection
  • single cell


The mammalian gastrointestinal tract microbiota comprises hundreds of different species of bacteria and archaea and an estimated 1014 cells that together form a highly dense and diverse microbial ecosystem. Microbial cells with an array of physiological capacities and ecological strategies are embedded in a complex and layered network of cooperation and competition. The extent of this intricate interplay is only partially understood because of the difficulty in studying it; simplified experimental culture conditions are in many cases inadequate to recapitulate in vivo activities and interactions and to accurately determine their contributions to the overall microbiota functionality. Within the last 10 years, technological progress in the field of next-generation sequencing technology has tremendously advanced our understanding of the wide variety of physiological and pathological processes that are influenced by the commensal microbiota. A clearer picture is emerging of the composition of the human microbiota in healthy individuals and the diverse factors affecting its variability (Yatsunenko et al., 2012). A robust definition of what is ‘normal’ allows the identification of ‘microbiome signatures’ associated with disease or disease susceptibility. However, at this stage, it is often unclear whether these signatures are causally involved or epiphenomena of health or disease. An important focus of current research is thus finding causative links in correlations between the microbiota and various human disease conditions, such as inflammatory bowel diseases, type 2 diabetes, obesity, and colorectal cancer (de Vos & de Vos, 2012).

Commensal microbes provide a variety of beneficial services to their host such as harvest of otherwise unavailable nutrient sources, training of the immune system, and vitamin production (Hooper et al., 2012; Nicholson et al., 2012). Among the longest known is the ability of an uncompromised microbiota to prevent pathogen infections, a phenomenon termed colonization resistance (CR). In humans, antibiotic-mediated microbiota perturbations can result in loss of CR and drastically increase susceptibility to infections, in particular with nosocomial pathogens such as Clostridium difficile, Enterococcus spp., Enterobacteriaceae, and Gram-negative nonfermenters (Stecher & Hardt, 2010; Britton & Young, 2012; Ubeda & Pamer, 2012). In particular, multimorbid patients undergoing several rounds of antibiotic therapy are at high risk of acquiring severe systemic infections by gut-dwelling multiple-antibiotic-resistant bacteria (e.g. vancomycin-resistant Enterococcus spp., multiresistant unidentified Gram-negative rods; Taur et al., 2012). To date, the mechanisms underlying CR are insufficiently understood. However, it is evident that CR results from a mutualistic, complex and multilayered interaction between the host, its mucosal immune system, and the commensal microbiota. Furthermore, environmental factors such as dietary habits, hygiene conditions, and other lifestyle aspects as well as genetic factors may play a role. In order to restabilize the microbiota after a perturbation and thus restore CR in affected patients, it is of key importance to identify the parts of the microbiota involved and to elucidate the molecular basis that mediates interference with pathogen colonization and replication.

Moving forward, it will be vital to further unravel the structure of the intestinal ecosystem and more precisely delineate how individual members of this ecosystem feed, interact, and survive. New statistical tools as well as ecological models can be used to identify putative structure–function relationships in microbial ecosystems (e.g. from metagenomic datasets) and to propose hypotheses to guide experimental study. In addition, valuable information can be obtained by molecular or metabolic pathway analysis of isolated strains based on genome analysis and in vitro experiments. However, care must be taken in order to translate the gained information to the actual in vivo activities of the strains. To better disentangle the complex microbial community and selectively analyze the role of individual species or their products, germfree or gnotobiotic mouse models have emerged as attractive experimental systems to test hypotheses generated by the numerous correlative studies and experimental in vitro data. Furthermore, highly resolved analyses of the spatial structure and activity of microbial communities using single-cell tools such as fluorescence in situ hybridization (FISH) in combination with new imaging techniques can give new insights into structure–function relationships in the intestinal ecosystem (Berry et al., 2013). Finally, methods involving metabolic labeling with stable isotopes open new paths to assess bacterial metabolism in vivo and to directly test the metabolic activity of individual members of complex microbial communities in their undisturbed ecological context.

A significant number of articles have recently reviewed and critically discussed the role of the gut microbiota in health and disease (Chow et al., 2011; Blumberg & Powrie, 2012; Cho & Blaser, 2012). In this review, we will present a ‘microbial ecologists view’ on this topic and highlight new findings about CR. We will discuss novel experimental approaches that can be used to refine our understanding of the mechanisms of the interplay of pathogens, the indigenous microbiota, and the mammalian host.

Gut microbial community structure: how to read signatures in the complex gut microbiome?

Bacterial interactions and microbiota community structure: general concepts

The symbiotic intestinal microbiota and its host have been described as a superorganism in which the microbiota acts as a virtual organ that provides manifold services to the body (Lederberg, 2000; O'Hara & Shanahan, 2006; Blumberg & Powrie, 2012; Hooper et al., 2012). This view has been supported by metagenomic evidence that the intestinal microbiota is not only numerically abundant, but also harbors a vast genetic potential estimated to encode roughly 100 times the number of genes present in the human genome and has therefore been coined ‘our second genome’ (Qin et al., 2010; Weinstock, 2012). While this analogy highlights the importance of this ecosystem to health and nutrition, framing the intestinal microbiota as a virtual organ minimizes important differences between organismal biology and microbial ecosystems. Unlike the component cells of animal organs, each microbial population in the gut must compete for limited resources to expand its population in order to survive over evolutionary time. Various strategies for achieving this end such as cooperation, competition, niche specialization, and metabolic flexibility can play a role in shaping the composition and ecosystem function of the intestine. Whereas each cell type in an organ plays a specific role in the functioning of the whole, the intestinal microbiota may be partially described as a collection of coexisting species without strong or specific interactions due in part to functional overlap of different coexisting or interchangeable populations.

Competition between microbial populations or species, also called interspecific competition, can take many forms. Competition for resources such as particular food sources, trace metals, or spatial habitats is considered exploitative competition because species compete to utilize one or more limited resources (Little et al., 2008). Resources subject to competition include nutrient sources such as host- or dietary-derived glycans (Bertin et al., 2012; Koropatkin et al., 2012), trace metals such as zinc (Gielda & DiRita, 2012) and iron (Dostal et al., 2012), and very likely spatial niches such as crypts or mucosal surfaces (Swidsinski et al., 2007b; Belzer & de Vos, 2012; Pedron et al., 2012; Van den Abbeele et al., 2012). Even compounds that are believed to give certain pathogens a competitive advantage such as ethanolamine (Thiennimitr et al., 2011; Bertin et al., 2012) and electron acceptors such as fumarate (Guccione et al., 2010; Jones et al., 2011), nitrate (Pittman et al., 2007; Jones et al., 2011; Winter et al., 2013), and tetrathionate (Barrett & Clark, 1987; Winter et al., 2010) have been shown to be utilized by more than one species, and it is likely that hitherto unknown commensals using these compounds will be identified in the future.

In contrast to exploitative competition, species can directly antagonize one another by producing compounds that negatively affect the other's growth, a process known as interference competition (Little et al., 2008). A well-studied example of interference competition that is believed to be widespread and important in the gut is bacteriocin production. Bacteriocins are bacterially produced antimicrobials that have many forms and can have diverse activities. Some, such as thuricin CD, which is produced by Bacillus thuringiensis and targets C. difficile, has a very narrow activity spectrum (Rea et al., 2010). Others, such as the Lactococcus lactis-produced lantibiotic nisin, target a wide group of organisms (Mota-Meira et al., 2000). Bacteriocins might play a role in CR and are discussed below. Further, bacteriocins are of medical interest because they offer the possibility to specifically target and eliminate a certain pathogen without causing collateral damage of the commensal microbiota, but they are likely also key in shaping the structure of the commensal microbiota. Because they can be so narrow in their activity spectrum-specific bacteriocins offer a potentially powerful research tool to study the importance of individual players in the gut by specifically targeting and eradicating them.

The fundamental ecological niche of an organism encompasses all possible life strategies that it can exploit, its ecological or metabolic potential as determined by genomic or pure-culture studies. The realized niche, in contrast, is the actual metabolism or strategy utilized by the organism in an ecosystem, which can be considerably more restricted depending on the ecological context (Fig. 1). In the face of asymmetric competition, when one species is a superior competitor and is able to inhibit another species without being much affected, the outcompeted species would either be forced to switch activities to exploit an alternate niche or go extinct from the ecosystem. For example, Lactobacillus reuteri and Lactobacillus johnsonii can partition the resources glucose and maltose when both species are present, allowing both to coexist though both organisms are capable of utilizing both compounds in pure culture (Tannock et al., 2012). The realized niche can also be modified by cooperation between organisms. For example, the presence of methanogen Methanobrevibacter smithii can enhance the ability of Bacteroides thetaiotaomicron to degrade polyfructose-containing glycans (Samuel & Gordon, 2006), Ruminococcus bromii can stimulate the utilization of resistant starches by a number of other species (Ze et al., 2012), and cocultivation of Bacteroides vulgatus with Akkermansia muciniphila facilitates its growth in a medium with mucin as the sole substrate (Png et al., 2010). Hydrogenogenic fermenters facilitate the activity of hydrogenotrophic organisms, such as methanogenic archaea, reductive acetogens, and sulfate-reducing bacteria, which in turn benefits hydrogenogenic fermenters by lowering the partial pressure of hydrogen (Carbonero et al., 2012). While there is abundant evidence that metabolic flexibility and shifting activities dependent upon the ecological context are hallmarks of the functioning of the intestinal microbiota, the role of functional redundancy and the relative importance of specific nutrient niches vs. metabolic flexibility are poorly understood. Simplified cultivation-based experiments and in silico predictions for ecological interactions such as competition (Freilich et al., 2010, 2011), while providing provocative hypotheses, are insufficient to demonstrate activity and interactions in the intestinal ecosystem and in vivo studies of complex microbiota are needed to experimentally test these hypotheses.

Figure 1

Representative microbial activities and ecological interactions in different experimental systems. Pure cultures or mono-associated animals (left) can be used to study the activities and fundamental niche of individual microorganisms and their interactions with the host. Set-ups employing multiple microorganisms in gnotobiotic animals or complex (undefined) microbiota (right) are useful for defining the realized niche of microbiota members and allow the study of diverse physiological and ecological processes such as cooperative substrate degradation and trophic interactions, interference and exploitative competition, and metabolic flexibility or niche switching. Microorganisms can be commensals, pathogens, or a combination of commensals and pathogens. Black arrows indicate substrate/metabolite transformations. Light gray arrows indicate substrate utilization, and the dashed arrow indicates that an organism was outcompeted for a substrate. Light gray lines with bars indicate antagonistic interactions. Examples of factors important under all experimental conditions are indicated in the box at the bottom of the figure.

Detecting structure in complex microbial communities using 16S rRNA gene-targeted analysis

Microbial communities are often characterized by PCR amplification and sequencing of a portion of the 16S rRNA gene (or its transcript) using primers targeting almost all bacteria. The recovered environmental 16S rRNA gene sequences are then clustered into phylotypes (or operational taxonomic units) based on sequence similarity. While a powerful approach for characterizing diversity, it does come with limitations. Biases that can affect the composition of sequence libraries have been identified at different steps of this process, including template inhibition in end-point PCR (Acinas et al., 2005), variability due to barcoded pyrosequencing adaptors (Berry et al., 2011), and pyrosequencing errors (Kunin et al., 2010). Therefore, results from sequence libraries, whether they produced by cloning and Sanger sequencing or next-generation sequencing technologies, should be considered semi-quantitative at best. In selected cases, phylogenetic information within the 16S rRNA gene can be insufficient to readily differentiate bacteria on a species- [e.g. Escherichia coli and Shigella spp. (Fukushima et al., 2002)] and strain level (i.e. E. coli pathotypes), which can result in underestimation of actual functional diversity (discussed further below; Wirth et al., 2006). Microbial communities are often described by the diversity within a single sample, termed alpha diversity. Because the community is almost never exhaustively sampled the real richness and diversity of a complex sample must typically be estimated. Several estimators have therefore been developed and applied to quantify species (or gene) richness and alpha diversity, including for example the Chao1 and ACE estimators for richness and the Simpson and Shannon indices for diversity (Hill et al., 2003; Lozupone & Knight, 2008). These estimators are all sensitive to undersampling of the microbial community, a problem that is particularly acute for highly rich or uneven communities (Hill et al., 2003; Curtis et al., 2006). The obstacle of undersampling has mostly been overcome by the advent of next-generation sequencing, which has increased the feasible sampling intensity by orders of magnitude. Comparisons of the diversity between several samples can be carried out by comparing differences in alpha diversity metrics or by applying beta diversity metrics to compare the membership of two communities (e.g. Jaccard and Sorenson similarity indices) or the differences in community structure, which includes both the presence and the relative abundance of each community member (e.g. Bray–Curtis similarity and the weighted UniFrac distance metric; Lozupone & Knight, 2008). Beta diversity can also be associated with spatial location or environmental parameters such as pH or other factors such as levels of immune effectors using statistical tools such as the Mantel test or Procrustes analysis (Ramette, 2007). Phylotypes statistically associated with a certain parameter (i.e. either detected exclusively or enriched) such as health/disease state, a genetic risk locus, or susceptibility of the host to a disease, can be identified by methods such as indicator species analysis (De Caceres & Legendre, 2009). Studies of community change over time, which are valuable for evaluating the resistance and resilience of the microbiota to perturbations such as those caused by antibiotics, infection, or inflammation, have mostly been observational or have used general beta diversity tools (Caporaso et al., 2011; Koenig et al., 2011), but some specialized tools for time-series analysis have been developed and applied (Trosvik et al., 2010; Gerber et al., 2012). Standardization and benchmarking of community stability measures will be important to ensure that suitable descriptions of disturbance and resilience are used in order to facilitate cross-study comparisons.

Association, or correlation, networks are a popular approach for microbiota analysis using ‘Omics’ datasets. Correlation networks can identify structures in ecological datasets that involve multiple nonrandomly associated variables such as phylotypes and are therefore useful for a basic ecological understanding as well as for moving beyond the ‘one-disease one-pathogen’ paradigm in order to detect possible multipartner structures predictive for medically important outcomes such as disease susceptibility or CR. Correlation networks are constructed by making pairwise correlations between each variable unit, whether the unit is a ‘species’ or 16S rRNA gene defined phylotype, gene or gene transcript (or functional category), protein, metabolite, environmental parameter or combination of different types of variables (Freilich et al., 2010; Qin et al., 2010; Arumugam et al., 2011; Bapteste et al., 2012; Greenblum et al., 2012; Xiong et al., 2012). Correlations describe the relationship between two variables across multiple samples based on the covariance in their abundances (typically relative abundances). Correlations can be quantified using the Pearson correlation coefficient, which detects linear associations, or rank correlations such the Spearman coefficient, which are less sensitive to requirements of linearity. Other approaches to detect associations include the maximal information content (Reshef et al., 2011), generalized boosted linear models (Faust et al., 2012) and ensembles of multiple approaches to increase robustness (Faust et al., 2012). The use of compositional data (i.e. relative abundance) presents a special problem because the unit of measurement intrinsically relates all variables with each other, making no population truly independent from any other. From a statistical perspective, it would be preferable to use total abundance so that measured abundance changes of each population would be independent of other populations. Measurement of absolute abundance, however, is not commonly practiced and can be problematic because the measurement (e.g. gene copies) must be normalized to either sample volume, which will be highly variable depending on water content, or mass, which is not strictly an absolute (i.e. independent of microbiota) measurement because intestinal contents are composed so predominately of bacteria, with estimates of up to half of feces being microbial (De Cruz et al., 2012). An approach that has recently been proposed uses the concept of ‘subcompositional coherence’ to overcome the limitations of relative abundance data by calculating pairwise correlations between species using only the abundances of the two species relative to each other rather than to the whole community (Friedman & Alm, 2012). Application of this approach should help to minimize the number of artifactual correlations in networks.

Care must be taken that correlation networks are not named or considered to be ‘interaction networks’, as interaction is an assumption that is easily violated. Co-occurrence can result from a shared preference for the same environment or same sensitivity to an environmental stressor (or opposite preferences in the case of negative correlations), from an indirect interaction via a shared predator (e.g. a protozoa such as Entamoeba, or sensitivity to a phage), or from the activity of an intermediate species that exerts an effect on two noninteracting species (e.g. shared sensitivity to a bacteriocin). This can lead to apparent competition and spurious correlation. Cooperative or competitive interactions must therefore be validated by other approaches or at least circumstantial evidence such as the inferred genetic potential of related reference genomes must be posited in order to establish a hypothesis of interaction. This is particularly important because it is known that the stability and the relative importance of individuals in ecological networks is determined not only by network features such as connectivity or centrality but also by specific activity or behaviors of members, especially important being trophic interactions (Allesina & Pascual, 2009; Gross et al., 2009). Experimental confirmation using stable isotope probing (SIP) and gnotobiotic mouse models, as discussed below, is therefore vital to supporting putative interactions suggested by network analysis.

Phylogenetic niche conservatism: the importance of phylogenetically resolved analysis

Understanding the relationship between microbial diversity and intestinal ecosystem function is a fundamental challenge. An appreciation of the ecologically equivalence or functional redundancy of closely related organisms is important for a basic understanding of community assembly and function (Hubell, 2005) as well as identification of health-state biomarkers (Berry, in press) such as microbiota signatures indicating CR. Phylogenetic niche conservatism, or phylogenetic signal, is an intuitive concept that states that the activity, traits, and occurrence of populations or species are conserved over evolutionary time and therefore related to organismal phylogeny. One might reasonably expect that closely related organisms would be more similar than more distantly related organisms due to their shared evolutionary history and similar genomes. Phylogenetic signal can be attenuated or modified by gene mutation, gene loss, lateral gene transfer (gene gain), and convergent evolution (Cavender-Bares et al., 2009). Phylogenetic attraction, also termed phylogenetic underdispersion or clustering, can be caused by a number of mechanisms such as environmental filtering, which is when environmental factors select for related species in the same community (Webb & Pitman, 2002; Helmus et al., 2007). Alternatively, phylogenetic repulsion (or overdispersion) can be caused by mechanisms such as competitive exclusion, where competing related species exclude each other from a community, or convergent evolution, where less related species converge on the same niche (Webb & Pitman, 2002; Helmus et al., 2007). A number of quantitative tools have been developed to measure phylogenetic signal, including Pagel's lambda (Pagel, 1999), Purvis and Fritz's D (Fritz & Purvis, 2010) and consentTRAIT (Martiny et al., 2012). Phylogenetic signal can be of variable strength in different ecosystems as well as in different clades (Losos, 2008). In the intestine, different bacterial lineages carry a different strength of phylogenetic signal, as determined based on their co-occurrence patterns (Berry et al., 2012; Koeppel & Wu, 2012). 16S rRNA gene-based studies have revealed a strong phylogenetic signal for the Enterobacteriaceae in the mouse intestine (Stecher et al., 2010; Berry et al., 2012), but many other families have less strong signal, particularly large phylotype-rich and genetically diverse groups such as the Lachnospiraceae (Berry et al., 2012). Therefore, microbiota analysis at higher taxonomic levels can obscure meaningful patterns in phylotype abundance, particularly for phylotype-rich taxa.

There is no clear single threshold of genetic relatedness at which ecological forces act in microorganisms, a problem that has called to question whether a biologically meaningful species definition can be proposed for microorganisms (Achtman & Wagner, 2008). It is known that very closely related organisms can have dramatically different ecological niches and consequences for health, for example, the different pathovars and commensal strains of E. coli (Wirth et al., 2006). Microdiversity is not only a feature of the evolution of pathogenesis, but also for basic metabolism. A recent analysis suggests that the phylogenetic signal of basic aspects of central metabolism, the utilization of different small carbon compounds, is attenuated at proximal phylogenetic distances (i.e. between closely related organisms) such that clustering of phylotypes at 97% similarity of a fragment of the 16S rRNA gene, as is commonly performed, can cluster groups of organisms with unlike metabolic potential together into a single group (Martiny et al., 2012). Strain-resolved metagenomic analysis of the low-diversity infant intestinal microbiota has revealed that Citrobacter strains occupy different ecological niches (Morowitz et al., 2011) and that the abundances of three Staphylococcus epidermidis strains and their phages vary dramatically over time, which may be due to either ecological/physiological differences or phage activity (Sharon et al., 2013). Methods unable to distinguish organisms to a higher phylogenetic resolution, such as amplicon and shotgun sequencing of fragments of single genes using 454, Illumina, or Ion Torrent technologies, can be insufficient to distinguish metabolically distinct groups, a caveat that should be kept in mind when these data are analyzed and functions are inferred. Use of ‘metagenomic species’ or metagenomic linkage groups, which are sequences assumed to be physically linked (i.e. on the same chromosome) because they highly covary over multiple samples (Qin et al., 2012), can increase the resolution of analysis by increasing the amount of the genetic content analyzed. However, this also introduces the problem of how to translate metagenomic linkage groups into real microbial populations because a single organism may be composed of multiple linkage groups. Phylogenetic resolution can also be improved without sacrificing a microbial population-based approach by employing methods such as multilocus sequence analysis (MLSA) or whole genome sequencing (WGS) of isolates or separated single cells (Achtman & Wagner, 2008). MLSA and WGS are relatively straightforward for cultivable organisms. Single-cell separation and genomic amplification technologies have been developed extensively recently (Stepanauskas, 2012), and though the methods remain challenging and low-throughput they hold the potential to greatly improve our understanding of intestinal microbiota structure and function.

Pathogen physiology and nutrition in the intestine

General determinants of intestinal colonization and pathogenicity

Only by employing a successful colonization strategy can bacteria, including both commensals and pathogens, be upgraded from transient to permanent colonizers and ensure continuous establishment in the gut. A strategy can be considered successful if it enables long-term niche occupation within the intestinal microbial ecosystem. In order to trigger disease, pathogens have evolved to occupy a specific niche and ensure efficient colonization of the intestinal tract. Microbial density in the gastrointestinal tract increases from the oropharynx to the colon (Berg, 1996). Bacterial titers in the stomach, duodenum, and jejunum are kept in check by low pH (i.e. stomach), bile acids and antimicrobial peptides produced by intestinal epithelial and specialized Paneth cells. This ensures that dietary nutrients which are directly accessible to the host (e.g. simple sugars, amino acids, ions, vitamins) can be absorbed in the upper small intestine in the absence of high numbers of bacterial competitors. Despite the higher number of bacterial competitors, the lower intestinal tract including the ileum, cecum, colon, and rectum are preferential infection sites for a number of enteropathogens (Heesemann et al., 1993; Hapfelmeier & Hardt, 2005). Pathogenesis can be mediated by toxin production in the intestinal lumen, such as is the case for enterohemorrhagic or enterotoxigenic E. coli (EHEC; ETEC), C. difficile, and Vibrio cholerae. Other infection strategies involve either the intimate attachment to epithelial cells such as performed by enteropathogenic (EPEC) or enteroaggregative E. coli (EaggEC) and Citrobacter spp. or mucosal tissue invasion as in the case of Salmonella spp., Shigella spp., Campylobacter spp., and Yersinia spp. In order to establish infection, a requirement common to all pathogens is the ability to replicate and grow to a certain threshold density in the respective region of the intestine. To this end, sufficient nutrient supply must be ensured.

E. coli: the most popular commensal and its pathogenic alter ego

Accounting for up to about 1% of total bacteria, E. coli is the most abundant facultative anaerobic commensal in mammalian gut ecosystems. The species also comprises a large number of ‘pathotypes’ such as EHEC and EPEC (Tenaillon et al., 2010). Until now, E. coli is the most well-studied bacterium in terms of intestinal lifestyle and metabolic requirements, mainly owing to seminal work by the laboratories of Paul Cohen and Tyrell Conway who elucidated nutrient sources and the main metabolic pathways of E. coli in the mammalian gut (Conway, 2007). In general, little is known about energy metabolism of enteropathogens during infection. The majority of human pathogens are facultative anaerobes. Their preferred form of energy metabolism is aerobic respiration, but in the absence of oxygen, they switch to anaerobic respiration and fermentation. Oxygen is the most energetically favorable electron acceptor, yet oxygen tension is very low in the intestine (He et al., 1999). Oxygen originates from capillary blood flow and is hence would only be expected to be available in close proximity to the intestinal epithelium because it would be rapidly consumed by bacterial respiration. Jones et al. (2011) demonstrated that E. coli lacking the high-affinity terminal oxidase (cytochrome bd type) exhibited reduced intestinal colonization, while the mutant with a low-affinity oxidase (cytochrome bo3 type) was not attenuated, suggesting that microaerophilic conditions prevail in the gut of streptomycin-treated mice. The ability to use nitrate and fumarate as terminal electron acceptors was also of importance for colonization of E. coli, though mutants deficient in respiring nitrite, dimethyl sulfoxide, or trimethylamine N-oxide were not attenuated in colonization (Jones et al., 2011). Recently, it was shown that the ability to respire host-derived nitrate boosts E. coli overgrowth in the inflamed intestine (Winter et al., 2013).

The mucus layer paradox: barrier function and nutrient source

What is known about pathogen nutrition in the intestine? Essentially, pathogens could utilize nutrients from three different sources: they may be primary consumers and feed (1) on components of the host's diet or (2) on host-derived compounds, including epithelial debris and mucosal secretions (i.e. mucins). Alternatively, (3) they may be secondary consumers and cross-feed on metabolic byproducts by other commensal bacteria. There are two types of mucins which may serve as nutrient source for pathogens: transmembrane mucins covering the epithelial surface (i.e. glycocalyx) and the secreted, gel-forming mucins which form the intestinal mucus layer. In addition, epithelial cell debris, which is shed from the mucosa into the gut lumen, can be used as source of nutrients (i.e. membrane compounds phosphatidylcholine and ethanolamine). The secreted colonic mucus layer in the small and large intestine of mice consists of the main mucin Muc2, forming of an inner layer of about 50 μm in thickness, which is firmly attached to the colonic epithelium and limits bacterial access to this compartment. This inner layer transits into an outer, loose layer, which measures 4–5 volumes of the inner layer, where the tight network of polymeric mucin glycoproteins is loosened up by cleavage via endogenous proteases (Johansson et al., 2011; Hansson, 2012). The mucus gel provides a matrix for the retention of antimicrobial molecules (e.g. defensins, cathelicidins, lysozymes, lectins; McGuckin et al., 2011; Vaishnava et al., 2011). The outer, loose mucus layer of the colon constitutes an excellent bacterial habitat enriched in glycans and amino acids (Johansson et al., 2011). The Muc2 protein backbone is rich in proline, serine, and threonine and highly decorated by O-glycosylation. The oligosaccharides contain five major sugars, N-actetyl-d-galactosamine, N-acetyl-d-glucosamine, N-acetylneuraminic acid (sialic acid), l-fucose, and d-galactose (Allen et al., 1984). Sulfation and sialylation of terminal sugar residues alters mucin viscoelastic properties and confers stability against degradation (McGuckin et al., 2011).

Spatial localization or enrichment of bacteria within the mucus layer may already hint at characteristics of their potential metabolic niche (e.g. mucus-specific adhesins, metabolization of mucin components). It has been demonstrated using FISH that growth of both commensal and pathogenic E. coli strains mainly takes place in the mucus layer of the large intestine (Poulsen et al., 1994; Moller et al., 2003). Other enteropathogens, such as Salmonella enterica serovar Typhimurium (S. Typhimurium), Citrobacter rodentium, and V. cholerae were shown to grow in the mucus layer (Freter et al., 1981; Stecher et al., 2004, 2008; Bergstrom et al., 2010). Escherichia coli genes and pathways induced upon growth in cecal mucus include catabolism of the major mucin-derived sugars N-acetylglucosamine (NAG), N-acetylneuraminic (sialic) acid, glucosamine, fucose and ribose, tryptophan, threonine, serine, aspartate, and phosphatidylethanolamine (Chang et al., 2004). Whether mucin-derived sugars are also metabolized in vivo, that is, upon growth in streptomycin-treated mice remains to be demonstrated.

Direct evidence may be obtained by an experimental approach making use of SIP in combination with Raman microspectroscopy or NanoSIMS (see below for further details). The relevance of mucin-derived compounds for E. coli colonization was assessed by creating mutant strains deficient in several pathways and then analyzing their colonization ability in competitive infection with the wild-type E. coli strain. Interestingly, these studies revealed that some sugars were required for the initiation of colonization at early time points post infection (day 1 p.i.), while other components were more relevant for maintenance of long-term colonization (day 9 p.i.; Conway, 2007). Strikingly, the relative importance of these carbon sources varied between different E. coli pathovars (i.e. pathogenic EHEC strain EDL933 and commensal K-12 strain MG1655) in vivo, while they showed the same order of nutrient preference in vitro. Once more, this evidences that the realized metabolic niche occupied in vivo can significantly differ from the fundamental ecological niche of pathogens and commensals. Gene expression profiles of E. coli grown on mucin as sole carbon source revealed that several sugars can be utilized simultaneously (Chang et al., 2004). This finding may in fact explain the mechanism how different strains of the same species can co-occur in the same intestinal ecosystem: if each strain specializes on a different nutrient source (i.e. sugar), several strains can occupy independent niches in the same ecosystem: they exploit two different realized niches.

So far it remains unknown if E. coli can use bistable gene expression (i.e. phenotypic noise) to split in subpopulations each specializing on using a different carbon compound in vivo. This scenario would allow the same E. coli population to exploit different niches in the intestine at the same time. In support of this notion, promoters of genes involved in energy metabolism of carbon sources (e.g. glycolysis, the pentose phosphate shunt, fermentation, aerobic respiration) were shown to be among those with the highest level of noise (Silander et al., 2012). In the case of S. Typhimurium, a transporter required for d-galactose uptake (mglB) is only expressed by a subpopulation of cells in the mouse intestine upon infection (Stecher et al., 2008).

How can enteric pathogens utilize mucin components for growth? Mucins are highly complex and stable glycoproteins, and their degradation requires the concerted action of a variety of bacterial enzymes. Therefore, breakdown of mucin is most efficiently accomplished by a community of cooperating commensals. Pathogens can adhere to and penetrate the mucus layer and access the underlying epithelium. The attaching and effacing (AE) pathogen C. rodentium was shown to replicate underneath the inner mucus layer in close proximity to the epithelium (Bergstrom et al., 2010; Muller et al., 2012). To facilitate mucus penetration, some pathogens express enzymes for mucus degradation that can lead to loosening of the firm mucus layer (McGuckin et al., 2011). Shigella spp. and enteroaggregative E. coli (EaggEC) encode pic, encoding a mucin serine protease that recognizes O-glycosylated serine (Navarro-Garcia et al., 2010). Enzymes that target a complex mucin scaffold have also been reported for EHEC (StcE; Metalloprotease), Campylobacter pyloridis (mucin protease), V. cholerae (hap, tagA mucinase) and Yersinia (mucin-depolymerizing enzyme; Table 1). Yet, evidence for enzymes that liberate monosaccharides or amino acids from mucus in order to eventually be metabolized has not been provided for Enterobacteriaceae. Thus, we assume that some pathogens (such as E. coli) largely depend on cross-feeding by commensals to utilize mucin for growth (Fig. 1). Yet, the direct experimental evidence for this scenario taking place in vivo is still lacking. Recently, in vitro experiments demonstrated that fucose liberated by B. thetaiotaomicron from pig gastric mucin can be sensed by EHEC in vitro (Pacheco et al., 2012).

View this table:
Table 1

Mucolytic commensal and pathogenic bacteria and their enzymatic capacities

MicroorganismSubstrateCharacterized enzymesActivityReferences
Human fecal microbiota (c. 1%)PGMExtracellular glycosidases (β-d-galactosidase, β-N-acetylglucosaminidase, and sialidase)Hoskins & Boulding (1981), Miller & Hoskins (1981)
Akkermansia muciniphila PGM, human MUC2Sulfatase and glycosidase activityDerrien et al. (2004), Png et al. (2010)
Bacillus indicus and Bacillus firmusPGMManzo et al. (2011)
Bacteroides fragilis PGM, pig colon mucinα-galactosidase, β-galactosidase, α-fucosidase, α-N-acetylgalactosaminidase, β-N-acetylglucosaminidase, glucose-6-sulfatase (mucin-desulfating sulfatase), β-d-galactose-3-sulfatase, β-d-galactose-6-sulfatase, N-acetylneuraminidase, sialidaseRoberton & Stanley (1982), Wright et al. (2000b)
Bfragilis and other BacteroidesBovine submaxillary gland mucinsgu gene locusSeveral glycosyl hydrolase activitiesNakayama-Imaohji et al. (2012)
Bacteroides vulgatus PGMPng et al. (2010)
Bacteroides thetaiotaomicron α-fucosidase, β-galactosidase, α-N-acetylgalactosaminidase, β-N-acetylglucosaminidase, and neuraminidase. Also, a novel glycosulphatase was identified using glucose-6-sulphate as substrateTsai et al. (1991)
Anaerobic sulfatase-maturing enzymeBenjdia et al. (2011)
Bifidobacterium bifidum PGM, human MUC2AfcA (1,2-alpha-l-fucosidase)Png et al. (2010)
Endo-alpha-N-acetylgalactosaminidaseKatayama et al. (2005)
α-galactosidase, α-N-acetylgalactosaminidaseHydrolyzed α1,3-linked Gal in branched blood group B antigen [Galα1-3(Fucα1-2)Galβ1-R], but not in linear xenotransplantation antigen (Galα1-3Galβ1-R)Kiyohara et al. (2012), Wakinaka et al. (2013)
1,3-1,4-alpha-l-fucosidaseThe enzyme specifically released alpha1,3- and alpha1,4-linked fucosyl residues from 3-fucosyllactose, various Lewis blood group substances (a, b, x, and y types), and lacto-N-fucopentaose II and III. However, the enzyme did not act on glycoconjugates containing alpha1,2-fucosyl residue or on synthetic alpha-fucoside (p-nitrophenyl-alpha-l-fucoside)Ashida et al. (2009)
Comparative genome analysis of various Bifidobacterium bifidum strainsTurroni et al. (2011)
Enzymes involved in the degradation of major core 1 and 2 O-glycansTurroni et al. (2010)
Bifidobacterium infantis strain VIII-240PGMBlood group H-degrading alpha-glycosidase activities, sialidase, and the requisite beta-glycosidases: extracellularHoskins et al. (1985)
Bifidobacterium longum Human intestinal mucinRuiz et al. (2011)
Prevotella strain RS2PGMα-galactosidase, β-galactosidase, α-fucosidase, α-N-acetylgalactosaminidase, β-N-acetylgalactosaminidase, glucose-6-sulfatase (mucin-desulfating sulfatase), β-d-galactose-3-sulfatase, β-d-galactose-6-sulfatase, N-acetylneuraminidaseWright et al. (2000b)
PGMSulfoglycosidase Sgl (AY158021)Cleaves terminal 2-acetamido-2-deoxy-β-d-glucopyranoside 6-sulfate (6-SO3-GlcNAc) residues from sulfomucin and from the model substrate 4-nitrophenyl 2-acetamido-2-deoxy-β-d-glucopyranoside 6-sodium sulfateRho et al. (2005)
MdsAN-acetylglucosamine-6-sulfataseWright et al. (2000b)
Ruminococcus AB strain VJ-268PGMBlood group B-degrading alpha-galactosidase activity, but this strain lacked beta-N-acetylhexosaminidases to complete degradation of B antigenic chains. Extracellular activityHoskins et al. (1985)
Ruminococcus gnavus PGM, human MUC2Png et al. (2010)
Ruminococcus torques PGM, human MUC2Salyers et al. (1977), Png et al. (2010)
Ruminococcus torques strains IX-70 and VIII-239PGMBlood group A- and H-degrading alpha-glycosidase activities, sialidase, and the requisite beta-glycosidases. Extracellular activity.Hoskins et al. (1985)
Campylobacter pyloridis Gastric mucinMucin proteaseSlomiany et al. (1987)
Escherichia coli, Shigella Bovine submaxillary mucin; Mouse cecal mucusPic: Mucin serine proteasePromotes growth in cecal mucusHarrington et al. (2009), Navarro-Garcia et al. (2010)
EHECMucin 7; glycoprotein 340StcE; metalloproteaseGrys et al. (2005), Yu et al. (2012)
Vibrio cholerae HapA and TagA mucin proteasesHapA might aid mucin penetration. TagA might modify host cell surface molecules during V. cholerae infectionSilva et al. (2003), Szabady et al. (2011)
  • PGM, pig gastric mucin. Mucolytic bacteria are listed along with the mucin substrates tested, the enzymatic activities measured, and enzymes (or relevant genes/gene products) that have been characterized. Results are from pure-culture studies.

Do commensals feed intestinal pathogens?

Besides being used as nutrient source, mucin components can be strong chemoattractants for pathogens. Salmonella enterica serovar Typhimurium, E. coli, C. rodentium and V. cholerae use flagella-driven chemotaxis to penetrate the mucus layer (Miranda et al., 2004; Osorio et al., 2005; Bergstrom et al., 2010). Nonchemotactic mutants are often attenuated in intestinal pathogenesis (Stecher et al., 2004; Butler & Camilli, 2005). Escherichia coli typically harbors 5 methyl-accepting chemotaxis proteins, Campylobacter jejuni 7 (Lertsethtakarn et al., 2011), S. Typhimurium 11 and V. cholerae encodes more than 40 different methyl-accepting chemotaxis proteins (Butler & Camilli, 2005). Yet, chemoreceptors of these pathogens can only sense monosaccharides (ribose, glucose, galactose), amino acids (serine, aspartate) or dipeptides. For this reason, the liberation of chemotactic-active components by mucus-associated commensal bacteria may be a prerequisite for directed pathogen movement in the intestinal lumen. This hypothesis could be challenged by appropriate experiments in germfree mice or gnotobiotic mice lacking any mucus-degrading commensals. In addition, all the named pathogens are also able to move toward higher oxygen concentration (positive aerotaxis), which should also guide the pathogens to the mucosa in the absence of free mucin sugars (Taylor et al., 1999). However, it is unclear how far the oxygen gradient would extend into the gut lumen in the presence of an oxygen-consuming competitive microbiota. Thus, oxygen may be sensed by pathogens only within the mucus layer and guide final orientation toward the epithelial lining.

Mining of intestinal microbiome datasets of human and animal origin as well as full genome sequences of commensal bacteria reveals a large diversity of bacterial enzymes that target different parts of the mucin structure (Table 1). These include sulfatases, sialidases, peptidases, and glycosyl hydrolases. Many commensal bacteria including A. muciniphila, Prevotella spp., and Bacteroides spp. have been shown to encode for or produce sulfatases (Benjdia et al., 2011; van Passel et al., 2011). Bacteroidetes, the most abundant Gram-negative bacteria in the anaerobic communities of the rumen and the human large intestine, are known to degrade a wide range of complex plant-derived polysaccharides (Salyers et al., 1977; Hooper et al., 2002; Chassard et al., 2005). Bacteroides thetaiotaomicron encodes 88 polysaccharide utilization loci (PUL, i.e. glycoside hydrolases) to forage both on dietary and host-derived glycans (Martens et al., 2008). A B. thetaiotaomicron mutant deficient in the production of sulfatases exhibits reduced capacity to upregulate PUL genes upon growth on mucin (Benjdia et al., 2011). When low amounts of polysaccharides are available through the diet, B. thetaiotaomicron was shown to switch toward mucin consumption (Sonnenburg et al., 2005). Via positive feedback regulation, degradation of mucin-derived l-fucose can in turn stimulate production of fucosylated glycans by the host (Hooper et al., 1999). Conversely, germfree mice lacking any microbiota in their gut exhibit a hypotrophic mucin layer (Smith et al., 2006). Expression of B. thetaiotaomicron PULs is also modulated in the presence of other members of the gut microbiota. Co-colonization with Eubacterium rectale, Bifidobacterium longum, and M. smithii can alter B. thetaiotaomicron metabolism and extend its nutrient range (Samuel & Gordon, 2006; Sonnenburg et al., 2006). A variety of other commensal bacteria commonly present in mammalian gut ecosystems are also involved in mucin degradation, and their contribution to the complex overall metabolic network of the microbiota is being unraveled. Addition of mucin stimulated sulfate reduction by sulfate-respiring bacteria in fecal slurries, suggesting that mucin is a potential electron donor and/or acceptor source for these bacteria (Gibson et al., 1988). Akkermansia muciniphila, a member of the Verrucomicrobia, is a common mucin degrader and present in mammalian gut ecosystems (Belzer & de Vos, 2012). The type strain has been isolated from human feces in media with mucin as only carbon source (Derrien et al., 2004). Its genome sequence revealed the presence of a large number of glycosyl hydrolases, proteases, sulfatases, and sialidases. Recently, it was shown using stable isotope labeling in combination with NanoSIMS-FISH that Akkermansia spp. also forages on mucin proteins in vivo in the mouse intestine (see below; Berry et al., 2013). In addition, members of the Ruminococcus spp., Bifidobacterium spp. and Lactobacillus spp. genera were demonstrated to produce glycoside hydrolases (Hoskins et al., 1985; Table 1). To conclude, although direct experimental evidence is lacking, commensals may play an important function in regulating pathogen nutrition, chemotaxis and infection by modulating the nutritional milieu of the intestine. While the detailed mechanisms are not yet fully resolved, there is abundant evidence that the commensal microbiota principally plays a beneficial role by preventing enteric infections by inducing CR.

Mechanisms of direct microbiota-mediated CR

The origin of the concept of CR dates back more than 40 years to studies of René Dubos and Rolf Freter (Mushin & Dubos, 1965; Freter & Abrams, 1972). Even earlier work by Bonhoff & Miller and Freter had already demonstrated that antibiotic treatment increases the susceptibility of mice to oral infection with S. Typhimurium and V. cholerae (Bohnhoff et al., 1954; Freter, 1955), which is also the case for a variety of other enteric and nosocomial pathogens (Wilson & Freter, 1986; Taur et al., 2012). In the same way, germfree mice are highly susceptible to infections with E. coli, C. difficile, V. cholerae or C. rodentium (Collins & Carter, 1978; Butterton et al., 1996; Stecher et al., 2005; Kamada et al., 2012; Reeves et al., 2012). Disrupted CR (in germfree, gnotobiotic or antibiotic-treated mice) can be reversed by fecal microbiota transplantation (FMT) from a donor animal harboring a complex microbiota, confirming that CR is mediated by the microbiota (Barman et al., 2008; Lawley et al., 2008; Endt et al., 2010; Stecher et al., 2010). Yet, due to the vast complexity of the microbiota and its intricate cross-talk with the host, the exact mechanisms underlying CR are still poorly understood.

Freter's nutrient-niche hypothesis: exploitative competition by the microbiota

In 1983, Rolf Freter formulated his ‘nutrient-niche hypothesis’, stating: ‘Our current hypothesis (…) holds therefore that the populations of most indigenous intestinal bacteria are controlled by substrate competition, that is, that each species is more efficient than the rest in utilizing one or a few particular substrates and that the population level of that species is controlled by the concentration of these few limiting substrates’ (Freter et al., 1983a, b) (Fig. 2a). Thus, if all possible nutrient niches are blocked by the indigenous microbiota, pathogens cannot colonize the intestinal environment. In this case, CR would increase commensurate with the complexity of the commensal microbiota's competitive metabolic potential. Indeed, several studies already suggested that CR is not mediated by a single species or its products but rather by a complex bacterial consortium (Freter et al., 1983a, b; Koopman et al., 1984). Re-association of germfree mice with a complex microbiota derived from intestinal content of conventional mice gradually increased CR against E. coli (Freter & Abrams, 1972). Gnotobiotic mice, colonized with a low-complexity microbial community (LCM) harboring only four members of the altered Schaedler flora (ASF; Dewhirst et al., 1999), are highly susceptible to intestinal colonization by S. Typhimurium (Stecher & Hardt, 2010). Yet, transplantation of feces from conventional mice significantly increased microbiota diversity and restored CR. When LCM mice were first colonized by S. Typhimurium and thereafter transplanted with conventional microbiota, a gradual rather than a sudden decrease in fecal S. Typhimurium colonization levels ensued, initially starting at 109 cfu g−1, to < 105 cfu g−1 within 3 weeks (Endt et al., 2010). These observations is in line with a scenario in which there is a sequence of events in which newcomers fill niches previously occupied by S. Typhimurium and thereby successively outcompete the pathogen by niche exploitation. One could imagine that competition for electron acceptors as well as carbon sources and micronutrients (i.e. iron) might play a role. A similar scenario may prevail in the course of an antibiotic therapy: due to disruption of gut ecosystem ecology and eradication of competitors, many niches suddenly become vacant and thereby CR is alleviated (Lozupone et al., 2012).

Figure 2

Freter's nutrient-niche hypothesis and application of gnotobiotic mouse models for CR research. (a) The nutrient-niche hypothesis raised by Freter et al. (1983a, b) states that the populations of most indigenous intestinal bacteria are controlled by substrate competition, that is, that each species is more efficient than the rest in utilizing one or a few particular substrates and that the population level of that species is controlled by the concentration of these few limiting substrates (Freter et al., 1983a, b). Thus, if all possible nutrient niches are blocked by the microbiota, CR is maximal, and pathogens cannot invade the gut ecosystem. In the pebble analogy, each pebble in the flask represents the realized niche space of a species and the space between pebbles the available niche space. In species-rich ecosystems, the available niche space is minimized. (b) Members of the gut microbiota are in constant interaction with itself involving competitive (i.e. inhibitory) or stimulatory (i.e. cross-feeding) processes and with the intestinal mucosa and its associated immune system (inhibition/stimulation of the immune system). The high complexity in the natural system (left side) precludes any mechanistic analysis of the contribution of individual bacterial strains or their products to CR. Gnotobiotic mouse models in combination with a defined set of commensal bacteria that can be selectively added to the system (right side) offer an attractive experimental tool to dissect the differential contribution of isolated microbial strains, their interaction with each other, and with the host to protection against pathogens.

The principle of niche exclusion has been used to mechanistically explain the resolution of C. difficile infection by the therapeutic application of nontoxicogenic strains. Clostridium difficile, a Gram-positive, anaerobic, spore-forming bacterium, is a paradigm opportunistic pathogen causing antibiotic-associated nosocomial infections in humans (Britton & Young, 2012). A severely disturbed intestinal microbiota plays a crucial role in the pathogenesis of this infection. In most healthy individuals, the microbiota controls C. difficile colonization of the intestinal tract. Antibiotic therapy (i.e. clindamycin, cephalosporins) disrupts the protective gut microbiota, whereupon ingested or existent antibiotic-resistant C. difficile spores germinate, colonize the gastrointestinal tract and produce toxins (toxins A and B; Bartlett et al., 1977). This can lead to antibiotic-associated diarrhea (AAD). Symptoms can range from mild diarrhea to the severe forms of pseudomembranous colitis, toxic megacolon, and multiple-organ dysfunction syndrome. Precolonization of hamsters with nontoxinogenic C. difficile spores in the course of clindamycin treatment prevented disease in the majority of hamsters that were challenged with toxigenic strains (Sambol et al., 2002). Niche exclusion may be mediated by the nontoxinogenic strain outcompeting the virulent strain by more efficiently utilizing limiting nutrient sources. AAD can be treated and cured by antibiotic therapy (e.g. metronidazole, vancomycin), but this often leads to recurrent infections. This situation is becoming an increasingly alarming healthcare problem. Use of alternate narrow-spectrum antibiotics that spare the microbiota and more selectively target C. difficile (i.e. fidaxomicin) has proven more efficient in preventing recurrent infections (Cornely et al., 2012).

Further evidence that CR could be largely directly mediated by the microbiota stems from in vitro gut fermentation models in continuous flow (CF) cultures. CF cultures can be inoculated with human or animal fecal material and, under controlled pH, temperature, and oxygen conditions, different regions of the intestinal tract can be modeled (Freter et al., 1983b; Payne et al., 2012). These systems are employed to assess the impact of probiotics, food additives, dietary components or drugs on microbiota composition or metabolism. In addition, CF cultures are used to study the interaction of the microbiota and pathogens. A main feature of these systems over in vivo studies is that functions of the microbiota are completely uncoupled from the host: the microbiota, for example, can be studied in the absence of host structures, enzymes, cells, and immune defenses (i.e. defensins, mucin, and cellular immunity) as well as neuroendocrine responses. Furthermore, control of environmental conditions, addition of other species or nutrients and controlled perturbation (i.e. by antibiotics, drugs) of the microbial ecosystem are more readily achieved in CF models than in animal models or human studies. CR is provided in anaerobic CF cultures, as demonstrated by repression of E. coli growth (Freter et al., 1983b). Further disturbance of the microbiota in CF cultures alleviates CR and leads to increased titers of S. Typhimurium, demonstrating that a complex fecal microbial community can keep pathogen overgrowth in check independently of the host (Le Blay et al., 2009).

Production of inhibitors: pathogen interference competition mediated by the microbiota

Early studies have provided evidence that CR could, at least in part, be mediated by the production of metabolites or inhibitors by the microbiota. In addition, changes in the oxidation–reduction potential or in pH induced by microbial metabolic activity might contribute to inhibition of pathogen replication. General candidate inhibitors include short-chain fatty acids (SCFA), bile salts, H2S as well as bacteriocins (Savage, 1977; Freter et al., 1983b). Decreased CR has frequently been correlated with reduced intestinal SCFA concentrations, that is, in antibiotic-treated or germfree mice (Hoverstad et al., 1986; Que et al., 1986). Further, in mice, a dysbiotic C. difficile ‘supershedder’ state is characterized by a reduction in butyrate and acetate and an increase in succinate levels (Lawley et al., 2012). The pathogen inhibitory effect of SCFAs may be indirectly caused by the associated reduction in pH, which can lead to reduced growth rates as has been shown for S. Typhimurium (Durant et al., 2000). In addition, SCFAs can impact virulence gene expression: acetate and propionate were shown to lead to induction of S. Typhimurium invasion genes (Durant et al., 2000), while butyrate had an inhibitory effect (Gantois et al., 2006). On the other side, SCFAs can also indirectly mediate protection against pathogens. Butyrate is a major energy source for colonic epithelial cells, leading to increased mucosal barrier function, that is, by promoting the formation of tight junctions (Ploger et al., 2012). Recently, it has been shown that the production of acetate by certain Bifidobacterium spp. strengthens mucosal barrier function and thereby protects from fatal shiga toxin producing E. coli infection (Fukuda et al., 2011).

Bile acids that in general are inhibitory to bacteria are synthesized by the liver and thereupon released into the gut lumen. Initially, the primary bile acids cholate and chenodeoxycholate are produced and are conjugated to the amino acids glycine or taurine to enhance solubility. In the gut, commensals dehydroxylate cholate and chenodeoxycholate into deoxycholate and lithocholate, respectively. Further, the microbiota can transform primary bile acids into secondary, unconjugated bile acids by hydrolysis (Midtvedt, 1974). Generally, unconjugated bile acids exhibit a stronger inhibitory effect on various tested bacteria than their conjugated forms (Floch et al., 1972). However, some pathogens including S. Typhimurium have developed resistance mechanisms against bile acids (Hernandez et al., 2012). An example of a mechanism by which transformation of bile acids by the microbiota may impact C. difficile infection has very recently been outlined in detail (Britton & Young, 2012). Taurocholate and glycine were shown to be potent germinants for C. difficile spores. In contrast, chenodeoxycholate has a strong inhibitory effect on spore germination. This suggests that the ratio of cholate derivatives to chenodeoxycholate may determine whether or not spores will germinate.

Another important group of inhibitors are bacteria-derived proteinaceous toxins (i.e. bacteriocins, lantibiotics colicins). In general, bacteriocins tend to have a narrow spectrum of activity and only act against closely related competitors. Colicins are produced by E. coli and relatives and are active against members of the Gram-negative Enterobacteriaceae including many pathogens (Cascales et al., 2007). A single E. coli strain was shown to have antagonistic activity by colicin production against a variety of enteric pathogens in vitro and in vivo in germfree mice (Cursino et al., 2006). Multiple types of colicins have been described which differ both by the mode of action as well as by their target range. Moreover, the distribution of colicin-producing strains in mammalian gut ecosystems can be highly variable (Gordon et al., 1998). Lantibiotics are bacteriocins produced by intestinal Gram-positive lactic acid bacteria such as Lactobacillus spp. and Enterococcus spp. Lantibiotics are a group of ribosomally synthesized, post-translationally modified peptides, containing unusual amino acids (i.e. lanthionine). They show strong antimicrobial activity toward a wide range of other Gram-positive bacteria. Yet, a recent study suggests that bacteriocin-producing strains are relatively infrequent among the human microbiota (Lakshminarayanan et al., 2012). Thus, bacteriocins might in selected cases contribute to protection against bacterial infections but may not account for CR in general.

Bacteriocins do offer, however, a great chance for treatment of enteric infections. Thuricidin, a bacteriocin produced by B. thuringiensis was shown to have a narrow-spectrum activity against C. difficile (Rea et al., 2010). In contrast to the antibiotics vancomycin and metronidazole as well as broad-spectrum bacteriocin lacticin 3147, which kill a wide range of anaerobic commensal bacteria of the microbiota, thuricidin selectively eliminated C. difficile from human distal colon cultures (Rea et al., 2011). Narrow-spectrum antimicrobials are considered very useful to develop novel therapeutic approaches for curing enteric infections as they do not interfere with CR provided by the normal microbiota. Along these lines, designer-bacteriocins offer a promising therapeutic strategy to target specific groups of detrimental bacteria while sparing the beneficial ones (Molloy et al., 2012).

Candidate microorganisms providing resistance against infections: search for agents of bacteriotherapy

Given the variety of possible CR mechanisms discussed above, it still remains to be answered which bacteria of a ‘normal’ mammalian gut microbiota are causally involved in CR. Ground-breaking work by Rolf Freter in the 1970s initiated the identification of the strains that mediate the ‘normalization’ (= CR) of germfree mice (Freter & Abrams, 1972). Germfree mice display a wide spectrum of abnormalities, of which the most striking is the vastly enlarged cecum (Bleich & Hansen, 2012). Inoculation of such mice with fresh or cultivated fecal material from normal mice ‘normalizes’ cecal size and other parameters such as villus/crypt ratio in the jejunum and ileum, mucus and IgA-production as well as cecal SCFA and bile acid levels (Koopman et al., 1984). Additionally, CR is re-established. Yet, reduction of cecal size does not always correlate with CR, for example, against E. coli (Itoh & Mitsuoka, 1980). Germfree mice inoculated with ethanol-treated feces (only ethanol-resistant, e.g. spore-forming, bacteria survive) exhibit reduction of cecal size but not E. coli levels compared with germfree mice (only Clostridia and fusiform bacteria were detectable; Itoh & Mitsuoka, 1980). In contrast, when fecal material was treated with chloroform or heat, which in addition preserved Bifidobacteria and Lactobacilli, both cecal size and E. coli levels declined (Itoh & Mitsuoka, 1980). Thus, chloroform and heat treatment seems to be a less efficient way to eliminate nonspore-forming intestinal bacteria. A consortium of 95 individual anaerobic strains isolated from the murine gut (Arank et al., 1969) was shown to promote CR against E. coli (Freter & Abrams, 1972). Yet, since methodology available at that time did not allow further characterization of the isolates (and most of the isolates have been lost in the meantime), the identity of the bacteria remains unknown. A number of other studies performed in the 1980s by Itoh and colleagues showed that a combination of Bacteroides, Lactobacilli and Clostridia was required to induce CR against Pseudomonas aeruginosa (Itoh et al., 1986). In contrast Bacteroides, Lactobacilli, or Clostridia on their own were unable to reduce C. difficile levels in germfree mice. This suggests that a combination of phylogenetically different species may be essential for providing complete CR (Itoh et al., 1987). Interestingly, a combination of Clostridium spp. and Lactobacillus spp. isolates alone could restore CR against E. coli (Itoh & Freter, 1989) and those strains have been preserved and further characterized recently (Momose et al., 2009). Reeves et al. (2012) demonstrated that a murine Lachnospiraceae isolate could partially restore CR against C. difficile in germfree mice.

In summary, these early studies suggest that obligate anaerobic bacteria (e.g. Clostridia) could play a major role in the induction of CR but that full protection is only achieved by a more diverse bacterial consortium. Clostridia belong to the low G+C Gram-positive Firmicutes which were formerly classified in several clostridial clusters (Collins et al., 1994), a system that is now outdated. The order Clostridiales contains many paraphyletic families, and their systematic classification is complex and under constant revision. A recent update based on curated databases of processed small-subunit rRNA primary structures (http://www.arb-silva.de; Ludwig et al., 2004) is published in Bergey's Manual, Vol 3, sec. edition 2009. The majority of characterized members of the Clostridiales are saccharolytic and produce a variety of SCFAs such as acetate and butyrate (Duncan et al., 2007), yet the exact mechanism of how they may contribute to CR remains unknown to date.

Recently, a milestone study by Lawley et al. (2012) identified a simple mixture of phylogenetically diverse intestinal bacteria that mediated clearance of C. difficile infection (‘supershedder’ state) in a mouse model. A collection of isolates from murine fecal material was used to screen for combinations of strains that mediate resolution of the ‘supershedder’ state. One successful mixture of only six strains could be identified and contained members of the Bacteroidetes, Firmicutes, and Actinobacteria phyla harboring three previously described species (Staphylococcus warneri, Enterococcus hirae, L. reuteri), and three novel species (Anaerostipes sp. nov., Bacteroidetes sp. nov., and Enterorhabdus sp. nov.). However, each of the strains failed to clear C. difficile when administered alone or in combination with other strains. Thus, displacement of C. difficile may require competition from a phylogenetically diverse and physiologically distinct collection of living bacteria. Further characterization of the strains and their genomes may, for the first time, allow insights to be gained into the mechanisms of CR provided by a defined bacterial consortium.

Studies in human patients have shown that CR against C. difficile can be successfully re-established by human fecal FMT (Landy et al., 2011; Palmer, 2011; van Nood et al., 2013). However, FMT is not used as standard therapy due to general patient aversion and due to the fact that it bears potential risks of transmission of viral, bacterial, or parasitic pathogens. To improve safety of this effective therapeutic approach, it would be favorable to transplant a known microbial mixture to patients that could mediate C. difficile clearance from stool and eventually block relapse. To this end, approaches such as the one described above (Lawley et al., 2012) could be highly useful to identify parts of the human microbiota with the potential to cure recurrent infections. A pioneering study showed clearance of human C. difficile infection by rectal application of a mixture of 10 strains (Tvede & Rask-Madsen, 1989). A more recent study demonstrated that a fecal microbiota culture that originated from a healthy donor and had been regularly recultivated under strict anaerobic conditions for more than 10 years was able to resolve relapsing C. difficile infection in 69% of treated human patients (Jorup-Ronstrom et al., 2012).

How can pathogens overcome CR?

Adaptive radiation of clonal populations to exploit different niches

To overcome CR imposed by the microbiota, pathogens have, in turn, evolved elaborate strategies. It has recently become clear that E. coli accumulates ‘adaptive mutations’ in the course of colonization of its host and that these mutations are central for efficient intestinal colonization (Giraud et al., 2001). Genetically diverse mutant subpopulations emerge from an initially clonal E. coli population upon colonization of germfree or streptomycin-treated mice (Leatham-Jensen et al., 2012). These mutant populations stably coexist as they exploit different niches in the intestine, an evolutionary process termed ‘adaptive radiation’. Initially, Leatham et al. (2005) had shown that nonmotile variants of the E. coli K-12 strain MG1655 with deletions in the flagellar master regulator flhDC were reproducibly selected in streptomycin-treated infected mice. Those mutants exhibit enhanced growth rates on a variety of carbon sources. Similarly, Giraud et al. (2008) reported that mutants with decreased motility exhibiting specific mutations in the envZ/ompR locus were systematically detected in 90% of bacteria harvested from E. coli MG1655 mono-colonized mice. The two-component regulatory system EnvZ/OmpR is involved in the regulation of E. coli membrane permeability but also controls flhDC and thereby motility. EnvZ/ompR mutants also have increased resistance to bile salts, which was considered as advantageous in the gut. De Paepe et al. (2011) further elaborated that in addition to the previously identified mutations in envZ/ompR, flhDC and malT mutants are reproducibly selected and all three mutants coexisted after 1 month of colonization. Thus, selective forces prevailing in the germfree mouse intestine generate a trade-off between stress resistance (i.e. bile acids) and nutritional competence and thus promote ‘adaptive radiation’ of E. coli. Similar albeit not identical findings were obtained in the study by Leatham-Jensen et al. (2012), which may be due to the fact that the streptomycin mouse model was used instead of germfree mice.

This led to the formulation of the ‘restaurant hypothesis’, which is an extension of the ‘nutrient-niche’ hypothesis by Freter to consider spatially separated niches, which are analogous to a wide variety of menu choices in a restaurant (Leatham-Jensen et al., 2012). Freter's hypothesis assumes that in the intestine, all nutrients are perfectly mixed, yet this is likely not the case. A large diversity of spatially separated niches is created due to spatial structures offering adhesion sites (i.e. mucus, epithelium, food particles), differential distribution of commensals providing nutrients to pathogens, for example, by enzymatic cleavage of polysaccharides as well as varying concentrations of antibacterials (bacteriocins, defensins). Thus, sympatric diversification allows E. coli to efficiently exploit all those niches and thus to keep up with competing microbes. Future work has to show whether sympatric diversification is a trait specific to E. coli or also occurs with other pathogens or commensal bacteria.

Timing and regulation of metabolic and virulence gene expression

Colonization of the epithelial surface, which is for the most part devoid of commensals, is a well-established infection strategy of AE pathogens (e.g. EPEC, EHEC, C. rodentium) and Shigella spp. (Miranda et al., 2004; Marteyn et al., 2010). Recently, it has been shown that this property is in fact crucial for C. rodentium to outcompete the microbiota (Kamada et al., 2012). At late infection stages, C. rodentium down-regulates the genes required for the formation of AE lesions, including the locus of enterocyte effacement (LEE) and the translocated intimin receptor (Tir). This leads to redistribution of the pathogen to the gut lumen where it is readily outcompeted by commensal E. coli exhibiting the same nutrient preferences. This scenario substantiates Freter's hypothesis according to which two strains cannot co exist in the intestine when one competes less well than the other for the same nutrient(s), unless the metabolically less efficient one adheres to the intestinal wall (Freter et al., 1983a, b).

In addition, timing of metabolic and virulence gene expression allows human pathogenic EHEC 0157:H7 to outcompete the microbiota (Pacheco et al., 2012). LEE expression in EHEC is positively regulated by two quorum sensing systems, which can sense the host-derived neurotransmitters adrenalin and noradrenalin in close proximity to the intestinal wall (Sperandio et al., 2003). On the other hand, LEE is repressed by the two-component regulatory system FusKR, which senses extracellular free fucose which is available at high concentrations in the intestinal mucus layer. In addition, FusKR negatively controls the genes involved in fucose uptake and metabolism. In the mucus layer, where LEE is repressed, EHEC efficiently grows on diverse carbon sources (e.g. galactose and hexuronates) and leaves fucose consumption to commensal E. coli strains (Fabich et al., 2008). Thus, EHEC can compete efficiently against commensal E. coli in the mucus layer by shutting down expression of the metabolically costly LEE virulence factors and utilizing its preferred carbon sources for growth. Yet, in close proximity to the intestinal wall, LEE expression is turned on by quorum sensing and infection can proceed in the absence of commensal competitors (Pacheco et al., 2012). This strategy of EHEC serves as an excellent example of how the tight location-mediated regulation of metabolism and virulence genes allows pathogens to successfully compete against a complex microbiota upon infection.

Induction of inflammation to generate new niches for pathogen replication in the intestine

Inflammatory changes of the intestinal mucosa are accompanied by alterations in the physicochemical gut environment and microbial ecology (Packey & Sartor, 2009; Li et al., 2012b). Typical changes involve decrease of microbial richness and diversity, reduction of obligate anaerobic bacteria, and outgrowth of facultative anaerobic bacteria such as Enterobacteriaceae, Enterococci, and Lactobacilli. Enteric pathogens such as S. Typhimurium and C. rodentium can subvert the inflammatory response they induce to outcompete the otherwise inhibitory microbiota (Lupp et al., 2007; Stecher & Hardt, 2008). Various pathogenic as well as certain commensal E. coli, which are generally minor members of mammalian gut ecosystems, have likewise adapted to taking advantage of inflammatory conditions (Barnich & Darfeuille-Michaud, 2007; Stecher et al., 2012, 2013). Conditions prevailing in the inflamed intestine are hostile to some obligate anaerobic commensals, but can be exploited by pathogens. Antimicrobial mediators such as the RegIIIβ and neutrophil elastase produced at elevated levels by the inflamed gut epithelium or infiltrating phagocytes were shown to selectively inhibit commensal but not pathogenic bacteria (Stelter et al., 2011; Gill et al., 2012). In the mouse colitis model, synthesis and uptake of the siderophore salmochelin confers resistance to the antimicrobial peptide lipocalin-2 to and boosts luminal growth of S. Typhimurium (Raffatellu et al., 2009). Similarly, resistance to the zinc-sequestering neutrophil protein calprotectin is mediated by expression of the high-affinity zinc transporter ZnuABC (Liu et al., 2012). Furthermore, the concentration of bile acids is elevated in murine S. Typhimurium colitis, and bile resistance is a critical trait enabling pathogen overgrowth under these conditions (Crawford et al., 2012).

Besides the elaborate resistance mechanisms against innate immune defenses, S. Typhimurium can also profit from inflammatory conditions by exploiting electron acceptors which are exclusively available under these conditions. In the presence of reactive oxygen species generated during inflammation tetrathionate is formed from thiosulfate as an epithelial detoxification product of microbial H2S (Winter et al., 2010). Tetrathionate is believed to be an electron acceptor exclusively used by S. Typhimurium, thus giving it a competitive advantage relative to the commensal microbial community. Several other enteric pathogens such as C. rodentium and Yersinia enterocolitica also harbor genes for tetrathionate reduction and possibly exploit similar mechanisms for enhanced fitness in the inflamed intestinal environment (Barrett & Clark, 1987; Winter & Baumler, 2011). Besides using tetrathionate, S. Typhimurium can manipulate the host by employing the type three secretion effector-protein SopE to induce generation of nitrate, which can in turn be used as terminal electron acceptor for anaerobic respiration (Lopez et al., 2012). Respiratory metabolism further enables S. Typhimurium to use highly abundant host-derived ethanolamine as electron donor and carbon source, which is not possible upon fermentative metabolism (Price-Carter et al., 2001; Thiennimitr et al., 2011).

Gnotobiotic mouse models: indispensable systems for addressing the mechanisms of microbiota–host–pathogen interactions

CR is the result of a highly complex interplay of the commensal microbiota, comprising more than 150 individual bacterial species, the intestinal mucosa, and the mucosal immune system (Qin et al., 2010). In addition to directly interfering with pathogen replication in the gut lumen, the microbiota also acts on the host and induces specific alterations of mucosal gene expression and immune defense (Hooper et al., 2012). A recent study showed that the microbiota of mice obtained from ‘Taconic Farms’ harbors a deep-branching clostridial species referred to as ‘segmented filamentous bacteria’ (SFB), which is absent in the ‘Jackson Laboratory’ colony (Ivanov et al., 2009). The presence of SFB led to dramatic differences in the Th17 cell levels and in turn influenced susceptibility to C. rodentium infection. On the other hand, a defined consortium of Clostridiales derived from mouse feces stimulated an increase in intestinal Foxp3 (+) T(reg) cell numbers, which can also drastically influence the outcome of infections by immunosuppression (Atarashi et al., 2011).

While certain components of the microbiota can modulate the immune system, the host can also affect gut microbial composition, adding yet another layer of complexity (Fig. 2b): mice exhibiting altered innate immune effector functions (i.e. expression of antimicrobial peptides, defensins, IgA, or mucin) can reveal drastic differences in their microbiota. Mice expressing human defensin 5 or lacking functional sIgA exhibit higher abundance of SFB (Suzuki et al., 2004; Salzman et al., 2010). Further, the different mouse strains 129 and C57BL/6 express a completely different pattern of antimicrobial peptides in the gut and display differential microbiota composition after being colonized with the same microbiota from a germfree state (Gulati et al., 2012). Similar effects have also been described to various other genetically modified mice defective in the innate and adaptive arms of the immune system (Elinav et al., 2011; Kawamoto et al., 2012). Further, mice with disrupted mucosal barrier and chronic gut inflammation show pronounced differences in their microbiota when compared with healthy littermate controls (Bibiloni et al., 2005; Lupp et al., 2007).

Pathogen infection is affected by both the intrinsic microbiota and the host's immune defense. As the host genetic background can also shape microbiota composition, it is very challenging to experimentally dissect the differential contribution of the microbiota and the immune defect to the outcome of pathogen infection in mice colonized with a complex, undefined microbiota. Gnotobiotic mice, for example, harboring a defined and well-characterized ‘model’ microbiota (Greek: gnotos = ‘known’) represent a ‘reductionist’ approach to dissect host–microbiota–pathogen interactions. While offering a versatile experimental model to address the effect of single species on the host, its immune system and resistance against infections, it has to be kept in mind that a defined ‘minimal’ microbiota with only selected members may lack important (but yet unknown) components of the microbiota (and their cellular and metabolic functions). Thus, experiments in gnotobiotic mouse models are essential for elucidating specific aspects of host–microbiota–pathogen interaction at a molecular basis, but findings should, if possible, be confirmed for their general validity in gut ecosystems with increased complexity.

In the past, gnotobiotic mice have been instrumental to address the role of the microbiota in nutrient degradation, stimulation of host tissue differentiation, and immune system development (Falk et al., 1998; Sonnenburg et al., 2005; Smith et al., 2007; Ivanov & Littman, 2010; Macpherson et al., 2012). Combination of gnotobiology with genetically engineered mice offers the opportunity to analyze reciprocal microbiota–host interactions at a molecular basis. Thus, causative links between targeted microbiota alterations and diseases including obesity, multiple sclerosis, inflammatory bowel disease, and others have been uncovered (Balish & Warner, 2002; Turnbaugh et al., 2006; Lee et al., 2010; Berer et al., 2011).

Which gnotobiotic model microbiota could be used in experimental preclinical (i.e. for preclinical drug or probiotics testing) mouse models for CR research? In the past, the ASF, a mixture of eight bacterial strains of murine intestinal origin, has been used as a ‘starter’ inoculum in order to generate specific pathogen-free mice after germfree rederivation for experimental research (Schaedler et al., 1965; Orcutt et al., 1987). Regrettably, the original ASF strains are currently not available through public strain collections. In addition, the ASF contains strains from merely three of the seven abundant phyla of the murine microbiota, the Bacteroidetes, Firmicutes, and Deferribacteres (Dewhirst et al., 1999), while lacking members of the Actinobacteria, Proteobacteria, Verrucomicrobia, and Tenericutes. Despite the reduced richness and phylogenetic diversity, the ASF has been shown to partially restore the mucosal immune system of germfree mice (Geuking et al., 2011). However, it does not fully reconstitute cardinal functions of a complex murine microbiota such as CR (Stecher et al., 2010), isolated lymphoid follicle formation (Bouskra et al., 2008), and reduction of cecal size (Bleich & Hansen, 2012).

For establishing mouse models with standardized murine microbiota for a wide spectrum of applications in biomedical research (i.e. including research on the CR mechanisms), a well-characterized murine microbiota is indispensable, but comparably, little energy is put into the analysis and isolation of murine microbiota. In contrast, tremendous efforts are put into isolation and detailed characterization (i.e. genome sequencing) of bacteria from the human intestinal tract due to the potentially higher clinical relevance (Turnbaugh et al., 2007). From these human strain collections, a number of ‘humanized’, gnotobiotic mouse models have been established recently (Faith et al., 2010; Becker et al., 2011; Wos-Oxley et al., 2012). However, it deserves consideration that the microbiota and its host share a long-standing co-evolution that has naturally resulted in bacterial adaptation to its host (Ley et al., 2008; Oh et al., 2010). Recent work has provided evidence that the intestinal microbiota is highly host specific and would exert its ‘normal’ functions only in the respective ‘original’ host from which it has been derived (Chung et al., 2012; Laycock et al., 2012). These data are intriguing and suggest that we have underestimated the importance of microbiota–host specificity for complete microbiota ‘functionality’. So far, cause-and-effect studies in gnotobiotic mice have frequently employed human bacterial isolates (Samuel & Gordon, 2006; Derrien et al., 2008; Faith et al., 2010; Becker et al., 2011). Humanized mouse models also can generate conditions that are more favorable for the colonization and replication of human enteropathogens, as has been shown for C. jejuni recently (Bereswill et al., 2011), which allows for the study of human pathogens which otherwise cannot be studied in animal models. Studies employing mice with human-derived microbiota have beyond any doubt yielded fundamental insights into this emerging research field. However, the availability of gnotobiotic mouse models based on murine microbiota will nonetheless be needed to generate novel insights into the role of bacteria–host adaptation and the mechanisms underlying host specificity.

The grand challenge: deciphering single-cell physiology in the intestinal wilderness

Microbial metabolism and energy gain in the intestinal environment are centered on the breakdown of host-derived and dietary compounds. Physiological processes and activities in the intestine are inherently complex given the wide variety of substrates (including but not limited to various glycans, proteins, and lipids), and the many different archaeal, bacterial, and host cells that are involved (Koropatkin et al., 2012). As discussed above, individual cells of the intestinal microbiota either compete for substrates due to functional overlap or cooperate in syntrophic substrate degradation, resulting in a dynamic network of synergistic and antagonistic physiological interactions that are reciprocally influenced by the temporarily and spatially fluctuating physical and chemical environment of the gut. An attractive experimental option to study the complex nature of the intestinal ecosystem is through reductionism. In vitro experiments (Ze et al., 2012) and in vivo studies in gnotobiotic mice (Samuel & Gordon, 2006; Hooper et al., 2012) have been instrumental in revealing the physiological capabilities of cultivated intestinal microorganisms – either alone or in mixed cultures of defined composition. However, given the physiological and physical interdependencies of the diverse intestinal microorganisms and host cells, the metabolic functions of individual microorganisms might be considerably different or of different magnitudes in a complex microbial background. Most of what we know of the phylogenetic and metabolic diversity of the gut microbiota comes from application of metagenomic and postgenomics (metatranscriptomics, proteomics, metabolomics) methods that capitalize on the high sample–throughput capacity of modern nucleic acids sequencing and mass spectrometry instruments. ‘Omics’ methods are of great importance for detecting the physiological properties and activities of uncultivated gut microbiota members (Turnbaugh et al., 2006, 2009; Qin et al., 2010; Arumugam et al., 2011; Consortium, 2012; Yatsunenko et al., 2012), but they also have clear limitations. For example, neither the detection of a homolog to a specific metabolic gene nor its transcription and translation definitely proves a specific physiological activity (Imachi et al., 2006; Mussmann et al., 2011). Furthermore, novel genes involved in known pathways or entirely novel pathways and metabolic functions cannot be unambiguously identified by sequencing-based approaches alone and thus an absence of homologs to a known functional gene does not prove the absence of a certain metabolic capability.

A valuable complement to ‘Omics’ methods are substrate-mediated stable isotope labeling techniques because they allow the study of microbial physiology without requiring a pure culture or an a priori knowledge of metabolic genes or pathways (Wagner & Horn, 2006; Neufeld et al., 2007; Orphan, 2009). The diverse methods now available are all based on the same principle, exposure of a microbial community to a stable isotope-labeled substrate, but the type of isotope (mostly 13C, 15N, but also 18O or 2H), the choice of isotope-labeled substrate, and the conditions under which the labeled substrate is introduced to an undisturbed microbial community can be varied depending on the physiological function that is being investigated. Specialized analytical instrumentation is employed to trace the fate of the isotope label into cellular components of a microbial cell and to identify the corresponding microorganisms by molecular methods. Existing substrate-mediated stable isotope labeling methods can be divided into two broad categories: (1) exploratory community-screening approaches such as DNA- and RNA-SIP and (2) directed approaches utilizing nucleic acids probes (mostly targeting rRNA) for identification of defined phylogenetic groups of microorganisms (Loy, 2010). The latter category includes two rather recently invented methods for investigation of individual cells – FISH combined with either Raman microspectroscopy (Raman-FISH) or multi-isotope imaging mass spectrometry with nanoscale resolution (NanoSIMS-FISH; Wagner, 2009; Musat et al., 2012; Pett-Ridge & Weber, 2012; Fig. 3). Though isotope-labeling methods for structure–function analysis of uncultivated members of complex environmental microbiota are well established in microbial ecology research, there has been to date surprisingly little use of these methods for functional analysis of the intestinal microbiota (Barclay et al., 2008; de Graaf & Venema, 2008; Venema, 2010), with a few noteworthy exceptions. Pioneering 16S rRNA gene-SIP studies of an in vitro model of the human intestine revealed the identity of microorganisms that were actively involved in degradation of 13C-glucose (Egert et al., 2007), 13C-starch (Kovatcheva-Datchary et al., 2009), and 13C-galacto-oligosaccharides (Maathuis et al., 2012). During DNA/RNA-SIP metabolically active populations are identified based on incorporation of the substrate-derived isotope label into nucleic acids, and subsequent molecular marker analysis of a continuum of ‘light’ (unlabeled) to ‘heavy’ (labeled) nucleic acids fractions that were separated by density ultracentrifugation. DNA/RNA-SIP is a discovery tool to screen a microbiota for not yet identified organisms that carry out a specific function, though it cannot be used as a quantitative measure of their metabolic activity. Furthermore, there is an increasing awareness that the role of single or subpopulations of cells is hidden in community measurements of physiological activities, which are inherently averages (Dethlefsen et al., 2007; De Souza, 2010, 2012). The individual cell is in immediate contact with the environment and can be regarded as the ultimate biological unit on which ecological and evolutionary forces can act upon.

Figure 3

In vitro/in vivo stable isotope labeling and combination of nanoscale multi-isotope secondary ion mass spectrometry imaging (NanoSIMS) or Raman microspectroscopy with FISH for physiological characterization and identification of individual cells in the complex intestinal microbiota. These single-cell approaches are best accompanied by complementary bulk and compound-specific isotope analyses in tissue (e.g. mucosa), blood, lumen content, exhalation air, and intestinal gas by isotope ratio mass spectrometry.

It is now possible to study the substrate utilization behavior of single cells and physiological differences between cells and quantitatively measure isotope label incorporation using Raman-FISH or NanoSIMS-FISH (Fig. 3). Raman-FISH requires a specialized Raman microspectroscope with an extra unit for fluorescence microscopy to identify a fluorescent probe-labeled cell and acquire its Raman spectrum (Huang et al., 2007). Raman microspectroscopy exploits the process of Raman scattering in which molecular bonds are excited by a laser, thereby producing a characteristic, inelastically scattered spectra of light. Due to the laser beam width and the numerical aperture of the microspectroscope objective, the lateral resolution is about 1 μm, which enables Raman spectra of individual microbial cells to be acquired. Naturally, the spectrum of a single cell is highly complex and consists of peaks that are assigned to several abundant compound classes such as nucleic acids, proteins, lipids, and carbohydrates (Huang et al., 2004). While essentially a composite of the individual Raman spectra of the many different cellular compounds, abundant individual molecules with the capacity for significant Raman scattering or resonance Raman scattering, such as phenylalanine, polyhydroxybutyrate (De Gelder et al., 2007) cytochrome C, and elemental sulfur (Freeman et al., 2001), can nevertheless be identified by characteristic peaks within the complex Raman spectrum of a microbial cell. Overall, the Raman peak pattern provides a useful fingerprint of the cellular biochemical composition (Wagner, 2009; Haider et al., 2010). In addition, incorporation of heavy isotopes, as shown for 13C and 15N, leads to characteristic changes in the cellular Raman spectrum (so-called peak shifts) that can be exploited to detect and quantify isotope-labeled substrate utilization (Huang et al., 2004). To date, the prominent phenylalanine peak has been used to diagnose and quantify isotope label incorporation because the position and height of this peak changes dependent on the relative amount of 13C enrichment (Huang et al., 2007). For the current state of the methodology, a cell must be enriched to minimally about 10 atom% 13C to detect the phenylalanine peak shift (for comparison, this is equivalent to or superior to the detection limit of RNA/DNA-SIP). While the technology will continue to improve, Raman microspectroscopy is best applied to analysis of the primary consumers of an isotope-labeled substrate, as secondary consumers of substrates produced by the primary consumers will be less enriched in isotope label. Raman microspectroscopy is a nondestructive technology and thus also enables sorting of individual living 13C-labeled cells for subsequent cultivation or whole genome amplification and sequencing (Huang et al., 2009; Li et al., 2012a, b). These technological innovations provide a remarkable opportunity for genomics of individual cells with specific molecular components or defined metabolic capabilities. While one study used near-infrared Raman spectroscopy to detect Helicobacter pylori infection and intestinal metaplasia in human gastric tissue (Teh et al., 2010), the potential of Raman microspectroscopy for intestinal microbiota analysis is yet to be exploited.

Owing to another technological innovation, the NanoSIMS, the isotopic composition of a biological specimen can even be imaged with nanoscale lateral resolution (down to c. 50 nm), but this is not the only outstanding feature of this secondary ion mass spectrometer (Lechene et al., 2006). Up to seven ion masses, can be measured in parallel, and distinguished with very high mass resolving power and sensitivity. The NanoSIMS can detect incorporation of an isotope label in a cell if the stable isotope ratio of an element (e.g. 13C/12C or 15N/14N) just slightly exceeds natural abundance levels (natural abundance of 13C and 15N is c. 1.1% and 0.4%, respectively). Several options exist for simultaneous, oligonucleotide probe-based visualization and identification of the microbial cells. Probes can be labeled with either a halogen (such as iodine, bromine, and fluorine) that has a low background level in the sample, a stable isotope that is different from the one used for measuring physiological activity, or a standard fluorescent dye. The label is introduced into the cell either directly attached to the probe (Li et al., 2008) or by employing horseradish peroxidase–tagged probes for subsequent deposition of radicalized tyramides conjugated with halogen-containing fluorophores. The latter, reporter deposition approach dramatically improves detection sensitivity through enzymatic signal amplification and is particularly useful if numbers of probe target molecules (i.e. number of ribosomes for rRNA-targeted probes) per cell are low (Behrens et al., 2008; Musat et al., 2008). Cells stained with standard fluorophores without halogens cannot be directly visualized with the NanoSIMS. Instead, FISH signals and isotope data are linked through an indirect spatial correlation approach that involves initially imaging probe-stained cells by confocal or standard fluorescence microscopy and subsequently imaging the same field of view with NanoSIMS (Berry et al., 2013). Because of its high sensitivity in accurately measuring even small differences in isotope ratios, NanoSIMS is particularly useful for analysis of the flow of the isotope label between trophically interacting organisms that have been exposed to a pulse of isotope-labeled substrate and sampled at different time points. An intriguing early demonstration of the power of the NanoSIMS for biological applications revealed that the bacterial symbiont Teredinibacter turnerae, which thrives in gills of wood-eating marine bivalves (shipworms), fixes 15N2 in situ, and provides nitrogen compounds to its host (Lechene et al., 2007). More recently, NanoSIMS has been used to study the unusual mutualistic symbiosis between an uncultured cyanobacterium (UCYN-A) and a unicellular marine alga, which is driven by reciprocal exchange of fixed carbon and nitrogen (Thompson et al., 2012). These and other NanoSIMS applications in microbiology (Wagner, 2009; Musat et al., 2012; Pett-Ridge & Weber, 2012) illustrate the great opportunities for studying metabolic interactions of the intestinal microbiota–host symbiosis and addressing fundamental physiological questions. This potential is exemplified by the findings from a NanoSIMS-based investigation of microbial foraging of host-derived substrates in the intestine. The gut ecosystem offers the resident microorganisms two basic nutrient options: direct utilization of dietary compounds or consumption of host-derived substrates. As aforementioned, a prominent example for the latter is utilization of complex mucin glycoproteins, major structural components of the secreted mucus overlying the intestinal epithelium (McGuckin et al., 2011; Koropatkin et al., 2012). Distinguishing which of the two major nutrient sources is used by the diverse gut microorganisms in vivo, that is, directly in the intestine of their hosts, is difficult and has so far been restricted to gnotobiotic mouse models with simplified microbiota (Sonnenburg et al., 2005). A new NanoSIMS-FISH-based approach now enables investigations of in vivo host compound foraging of individual cells within the naturally complex intestinal microbiota (Berry et al., 2013). 13C/15N-threonine was injected intravenously to mice as a building block for cellular protein biomass (e.g. threonine-rich mucin), and appearance of maximum label enrichment was measured over time in blood, mucus, and lumen content to identify the ideal postpulse time point for NanoSIMS-FISH analysis. Using genus- and species-specific probes for several microorganisms in the murine intestine, Bacteroides acidifaciens and A. muciniphila were identified as important foragers of host-derived protein compounds. Although A. muciniphila is a dedicated mucin forager in pure culture (Derrien et al., 2004), this proved for the first time that this bacterium also displays this phenotype in vivo. Interestingly, experiments with gnotobiotic mice showed that the host-foraging phenotype of these bacteria is quantitatively dependent on the complexity of the intestinal microbiota. Decreased utilization of host proteins by B. acidifaciens and A. muciniphila in a low-complexity microbiota is either due to artificial realization of niches (e.g. availability of dietary compounds) through absence of otherwise competing microorganisms, lack of syntrophic partners for efficient degradation of complex host compounds such as mucin or general changes in mucus production and turnover. It will be interesting to see if such foraging pattern still hold true if other mucus-building blocks (such as NAG and N-acetylgalactosamine, constituents of the mucin O-glycosylated side chains) or epithelial cell precursors [such as ethanolamine i.e. synthesized to the predominant membrane phospholipid of released enterocytes, phosphatidylethanolamine (Kawai et al., 1974; Gibellini & Smith, 2010)], are used intravenously for stable isotope labeling. This study demonstrated that the combination of FISH and NanoSIMS is readily applicable in the gut environment, but its potential for detailed spatial analysis is yet to be fully exploited. For example, tissue structure is better preserved by embedding in an epoxy resin, but this is not compatible with FISH, thus acrylic resin (LR White) is used for NanoSIMS-FISH of microorganisms in tissue sections. Resin embedding also introduces additional carbon that can interfere with carbon, but not nitrogen, isotope measurements by NanoSIMS (Berry et al., 2013). Depending on the type of sample preparation, hybridization, or signal amplification, exogenous carbon and/or nitrogen might be introduced into the specimen, which would reduce detection sensitivity and hamper quantification of the isotope label. Despite these obstacles, further development of the combination of FISH with Raman microspectroscopy or NanoSIMS, but also methodological or technical improvements of each individual technique will certainly advance analysis of specific metabolic activities of individual cells.

Equipped with such tools, the physiological roles of individual microorganisms and their relative contribution to the concerted activities of the intestinal microbiota, whether synergistic or antagonistic, in the degradation of complex substrates and recycling of degradation intermediates such as SCFA can now be determined. Hypotheses on the function of specific members of the intestinal microbiota, as inferred from use of ‘Omics’ technologies or put forward to explain CR (e.g. the nutrient niche and the restaurant hypotheses, as outlined above), can now be proven or disproven in vivo under natural conditions. In addition, pulse labeling and time-course studies potentially allow tracing the fate of label through the different levels of the tightly interwoven trophic network in the intestinal tract. Time-course analyses will enable measuring average rates of substrate incorporation into a population of cells that is defined by distinct morphology or a diagnostic probe and thus reveal how the fluxes of specific nutrients are quantitatively influenced by the various intestinal microorganisms.

Higher resolution and accuracy on all frontiers: introducing advanced FISH methods for a better understanding of intestinal stereobiology

Ribosomal RNA-targeted oligonucleotide probes are commonly applied in FISH to detect and quantify bacterial, archaeal, or fungal community members in the intestinal tract of animals and humans (Swidsinski et al., 2005; Scupham et al., 2006; Bergstrom et al., 2010). In comparisons with the microbiota in environments with lower nutrient supply such as sediments or soils, the intestinal microbiota is well nourished, metabolically more active with a higher cellular ribosome content, and thus well accessible to detection by FISH with standard mono-labeled rRNA-targeted probes; additional signal amplification is generally not required. In contrast to PCR amplicon-based sequencing approaches that are intrinsically biased (Acinas et al., 2005; Kunin et al., 2010; Berry et al., 2011). FISH also allows true quantitative measures of probe-defined cell populations in the gut. Cells in a defined volume of lumen content or feces are fixed and suspended, hybridized, and counted by flow cytometry (fluorescence-activated cell sorting) to reveal absolute cell numbers (Lay et al., 2007; Fallani et al., 2011). Alternatively, images of FISH-stained organisms in unmixed fecal/lumen samples or embedded tissue sections are recorded by fluorescence microscopy. Subsequently, digital image analysis is used to quantify relative abundances of specific probe-stained populations (Berry et al., 2012). To investigate the spatial localization of target microbes, it is essential that fixation and embedding procedures maintain the structural integrity of the sample. A well-established tissue sample preparation for FISH that preserves the mucus layer is fixation with Carnoy solution and embedding in paraffin wax (Swidsinski et al., 2005, 2007a, b; Canny et al., 2006; Johansson & Hansson, 2012). Such spatial FISH analyses provided valuable insights into the mucosal infiltration and localization of intestinal microorganisms in healthy human individuals and patients with different intestinal disorders (Swidsinski et al., 2007a, b; Pedron et al., 2012).

While these studies clearly exemplify the usefulness of FISH for gut microbiota research and diagnosis, many advanced FISH methods are now available (Amann & Fuchs, 2008; Moraru & Amann, 2012; Wagner & Haider, 2012) that have not yet been exploited for analyses of the intestinal microbiota, such as more phylogenetically resolved identification, multiplexed detection, and/or quantitative spatial localization of multiple target organisms. Before the introduction of some of these advanced options, however, it is valuable to emphasize the very fundament of FISH – the accurate identification of phylogenetically defined populations of target cells by specific rRNA-targeted probes (DeLong et al., 1989) – as this strongly influences biological interpretation of results. The online database probeBase offers a large collection of published rRNA-targeted probes that target microorganisms at different phylogenetic or taxonomic levels (Loy et al., 2007). However, these probes need to be carefully selected and periodically evaluated for their in silico coverage and specificity (Table 2; Loy et al., 2008) against the rapidly increasing number of sequences in public rRNA repositories (ARB-Silva, Greengenes, RDP II), because the initially expected specificity and coverage might not be given anymore or might be outdated. FISH with such ‘outdated’ probes is unfortunately a common problem, also in gut microbiota research, and leads to erroneous results and conclusions. Furthermore, for most FISH probes in probeBase, including probes for microbial taxa found in the intestinal tract, the optimal hybridization conditions have either not or not adequately been evaluated empirically by melting curve analysis of representative target and nontarget cells.

View this table:
Table 2

Up-to-date in silico specificity and coverage of selected, group-specific 16S rRNA gene-targeted probes for FISH of intestinal bacteria

Probe nameProbe sequence (5′ -> 3′)RDP II probe match*References
Total hitsMajor target taxa (coverage%, total hits in taxon)Total nontarget hitsMajor nontarget taxa (coverage%, total hits in taxon)
Phylum Firmicutes
Lab158GGT ATT AGC AYC TGT TTC CA45 351 Order Lactobacillales (27.8%, 40 908) Family Enterococcaceae (96.1%, 9718) Family Lactobacillaceae (93.6%, 27 762) Family Leuconostocaceae (59.4%, 3003) 4443 Order Clostridiales (1.5%, 3322) Family Lachnospiraceae (1.8%, 1920) Family Ruminococcaceae (1.2%, 677) Hermie et al. (1999)
Strc493GTT AGC CGT CCC TTT CTG G64 277 Order Lactobacillales (46.8%, 64 077) Family Streptococcaceae (96.6%, 63 938) 339Franks et al. (1998)
Erec482GCT TCT TAG TCA RGT ACC G88 687 Order Clostridiales (37.7%, 88 160) Family Lachnospiraceae (75.8%, 86 075) 2612Franks et al. (1998)
Clep866GGT GGA TWA CTT ATT G22 344 Order Clostridiales (11.1%, 19 508) Family Ruminococcaceae (44.1% 19 346) 2998Lay et al. (2005)
Fprau645CCT CTG CAC TAC TCA AGA AAA AC14 206 Order Clostridiales (5.8%, 12 689) Family Ruminococcaceae (23.2%, 12 656) Genus Faecalibacterium (87.5%, 12 620) 1586Suau et al. (2001)
Phasco741TCA GCG TCA GAC ACA GTC2097 Order Selenomonadales (9.6%, 2072) Family Acidaminococcaceae (90.9%, 1916) 181Harmsen et al. (2002)
Phylum Bacteroidetes
CF319aTGG TCC GTG TCT CAG TAC12 1462 Phylum Bacteroidetes (43.6%, 119 230) Order Bacteroidales (30.5%, 47 232) Family Bacteroidaceae (2.7%, 1483) Family Marinilabiaceae (80.1%, 803) Family Porphyromonadaceae (78.7%, 34 321) Family Prevotellaceae (10.4%, 4509) Family Rikenellaceae (92.5%, 3718) Order Flavobacteriales (86.6%, 37 773) Family Cryomorphaceae (95.9%, 2620) Family Flavobacteriaceae (85.7%, 33 250) Order Sphingobacteriales (15.7%, 5503) Family Cytophagaceae (9.4%, 561) Family Flammeovirgaceae (18.8%, 359) Family Sphingobacteriaceae (89.1%, 4020) 2232Manz et al. (1996)
Bac303CCA ATG TGG GGG ACC TT96 867 Order Bacteroidales (62.2%, 96 059) Family Bacteroidaceae (95.9%, 52 153) Family Porphyromonadaceae (19.6%, 8544) Family Prevotellaceae (77.5%, 33 409) 808Manz et al. (1996)
Bac1080GCA CTT AAG CCG ACA CCT79 867 Order Bacteroidales (88.9%, 78 353) Family Bacteroidaceae (97.1%, 36 481) Family Porphyromonadaceae (91.0%, 18 928) Family Prevotellaceae (93.7%, 19 766) 1514Kong et al. (2010)
CFB1082TGG CAC TTA AGC CGA CAC79 920 Order Bacteroidales (89.2%, 78 644) Family Bacteroidaceae (97.1%, 36 492) Family Porphyromonadaceae (90.8%, 18 891) Family Prevotellaceae (93.7%, 19 769) 1276Weller et al. (2000)
Phylum Actinobacteria
Bif164CAT CCG GCA TTA CCA CCC2580 Order Bifidobacteriales (85.8%, 2568) Family Bifidobacteriaceae (85.9%, 2568) Genus Bifidobacterium (91.0%, 2541) Genus Parascardovia (100%, 15) 12Langendijk et al. (1995)
Phylum Deferribacteres
Mcs487GCC GGG GCT GCT TAT ACA GGT838Genus Mucispirillum (96.8%, 837)1Berry et al. (2012)
Mcs547CAG TCA CTC CGA ACA ACG CT883Genus Mucispirillum (96.9%, 822)61Berry et al. (2012)
Phylum Verrucomicrobia
Akk1437CCT TGC GGT TGG CTT CAG AT285Genus Akkermansia (88.2%, 284)1Derrien et al. (2008)
  • RDP II probe match was performed with database release 10, Update 31 (7 December 2012) containing 2 639 157 bacterial and archaeal 16S rRNA gene sequences. The search for each probe was restricted to sequences of good quality with data in the respective probe-binding region. Coverage is the percentage of sequences within the RDP II target taxon that shows a full match to the probe sequence. The number of nontarget hits indicates the total number of sequences outside the respective RDP II target taxa that show a full match to the probe sequence.

  • See Lay et al. (2005) for competitor probes.

Differentiating between different species using rRNA-targeted probes is sometimes difficult or even impossible due to the high sequence conservation of the rRNA molecules (Fox et al., 1992; Fukushima et al., 2002). An alternative target for whole cell hybridization in microorganisms is precursor rRNA (e.g. transcripts of the 16S and 23S rRNA gene intergenic spacer region) that consists of sequence regions with higher variability than 16S rRNA gene and thus is well suited to even distinguish different strains (Schmid et al., 2001; Wagner, 2004). Further oligonucleotide probe targets that could offer higher phylogenetic resolution than rRNA are the actual genes on the chromosome or their mRNA transcripts. Current FISH methods for specific detection of single-copy genes or mRNA in microorganisms are still far from routine use as they are laborious to establish, difficult to apply in an environmental context, do not detect all cells within a target population, and often depend on polynucleotide probes that have better sensitivity than oligonucleotide probes but cannot distinguish between similar sequence variants of the same gene (Pernthaler & Amann, 2004; Hoshino & Schramm, 2010; Moraru et al., 2010). However, if this effort is made and FISH is performed simultaneously with rRNA-targeted probes, the presence of a known or unknown gene can be directly linked with the identity of a cell in an environmental sample. A further integration with single-cell stable isotope labeling would allow associating some of the many unknown genes in the intestinal microbiome (Qin et al., 2010) to defined physiological properties, an exciting new approach that has yet to be explored.

Digital analysis of fluorescence microscopy images is an essential step in FISH analysis that, if done right, provides objective and quantitative data on the abundances, morphologies, signal intensities, and spatial localization (2D and 3D) of cells in complex samples in a semi-automated way and should not be considered merely a technical gimmick. Though in the intestine, there are homogenizing or disturbing factors such as peristalsis, mucus secretion, and mucosal shedding, the spatial arrangement of microbial cells relative to each other and to the different epithelial host cells is not just random, especially not for mucus-associated cells, and thus can hold important clues about the quality of intercellular interactions. Smaller and greater distances between cells could simply be indicative for beneficial and antagonistic interactions, respectively – at least for one of the partners. Subjective visual inspection is usually inadequate in discerning nonrandom spatial arrangements of microorganisms, but quantitative digital image analysis offers sensitive detection of patterns at small spatial scales (Daims et al., 2006). Using specialized software such as DAIME (Daims et al., 2006), quantitative colocalization analysis highlighted ecological niche differentiation of uncultured members of the nitrite-oxidizing genus Nitrospira in nitrifying biofilm and activated sludge (Maixner et al., 2006) and disclosed significant proximity of Fusobacterium nucleatum/periodonticum and Tannerella forsythia cells in an in vivo grown subgingival biofilm (Schillinger et al., 2012).

FISH combined with microscopy is a low-throughput method with the restriction that only few target populations can be distinguished in parallel using different fluorophore-labeled probes. FLUOS, Cy3, and Cy5 are the three most commonly applied fluorescence dyes for FISH because most standard confocal laser scanning or epifluorescence microscopes are readily equipped with appropriate light sources, emission and excitation filters, and additional optics for their detection. Thus, maximally up to seven populations can be detected at the same time by the classical FISH approach, that is, simultaneous use of probes mono-labeled with these three different dyes (Amann et al., 1996). Novel approaches now exploit combinations of multiple fluorescence labels per probe variant (Valm et al., 2011) or probe molecule (Behnam et al., 2012) to overcome this limitation. Tens to potentially hundreds of microbial populations can be detected in parallel in a single field of view of the microscope by the new combinatorial labeling and spectral imaging FISH (CLASI-FISH) approach (Valm et al., 2011, 2012). Fifteen microbial taxa were imaged simultaneously in human dental plaque biofilm in a proof-of-concept application of CLASI-FISH. Colocalization analysis revealed that cells of the genera Veillonella, Prevotella, and Actinomyces displayed the most interspecies cell–cell contacts, indicating a pivotal role in establishing and maintaining plaque biofilm architecture and function (Valm et al., 2011). In addition to the benefit of increased data output, CLASI-FISH will also prove to be very useful in combination with single-cell SIP and allow unprecedented insights into the cellular structural organization and metabolic complexity of the intestinal mucosal microbiota and other microbial habitats of the human and animal body.

Considering in vivo stable isotope labeling for structure–function analyses of the intestinal microbiota

The technical and methodological advances for isotope labeling and molecular identification of single cells open up new opportunities for studying the role of intestinal microorganisms at a hitherto unmatched resolution in space and activity. For initial evaluation of hypotheses, stable isotope labeling of the complex intestinal microbiota can be performed in straightforward ex vivo setups such as through anaerobic incubations of fecal material/lumen content/tissue biopsies or sophisticated in vitro bioreactor models of the human gastrointestinal tract (de Graaf & Venema, 2008; Kovatcheva-Datchary et al., 2009). However, this is just an intermediate step toward in vivo applications of stable isotope-labeled substrates for functional analysis of the complex animal and human microbiota but also of host cells (Hamer et al., 2012). Substrates can be administered to the gastrointestinal tract via the oral route (i.e. added to food or enclosed in enteric-coated capsules that release the substrates in defined regions), directly to a certain intestinal location via endoscopy, or indirectly by intravenous injection of (precursors of) substrates that are metabolically processed by the host and translocated into the gut. The choice of administration route is a nontrivial problem and will depend upon the research question as well as logistic factors such as, for example, the body size of the host of interest. Certainly, many approaches for in vivo stable isotope labeling and analysis of the intestinal microbiota will initially be developed in animal models (Berry et al., 2013) before they will be applied to humans. However, stable isotope tracer application to humans in vivo is not fiction but already for some applications a standard tool in biomedicine for revealing host physiology and quantification of metabolic fluxes (Kelleher, 2004; Dolnikowski et al., 2005; de Graaf & Venema, 2008). For example, whole-body turnover of SCFA in animals and humans, which depends on endogenous turnover and exogenous production by intestinal microbiota, has been determined after intravenous infusion of 13C-labeled acetate, propionate, and/or butyrate (Pouteau et al., 2003). Stable isotope-labeled substrates are also routinely administered to patients for diagnosis of gastrointestinal tract disorders (Szarka & Camilleri, 2012). For example, oral ingestion of 13C-urea followed by a breath analysis is a standard test for H. pylori infections (Israeli et al., 2003). A similar breath test, after rectal administration of 13C-butyrate, can be used to evaluate colonic mucosal function in patients with inflammatory bowel disease (Kato et al., 2007). Gastroenterologists routinely retrieve samples (i.e. feces or lumen content/tissue biopsies via endoscopy) from the gastrointestinal tract environment of such patients for diagnostic purposes. Analysis of individual in vivo labeled cells by NanoSIMS-/Raman-FISH is thus literally within reach and promises novel insights into the various physiological states of dysbiosis among the commensal microbiota and the host during infectious (e.g. S. Typhimurium, H. pylori, C. difficile), chronic (e.g. obesity, metabolic syndrome, diabetes, inflammatory bowel disease) or malignant (e.g. colorectal cancer) intestinal diseases.

Concluding remarks

The current view of the intestinal microbiota is as a highly complex and structured ecosystem with many participants exercising diverse metabolisms and harboring varying metabolic potentials and with a major impact on host health and nutrition and susceptibility to infection. The composition of the microbiota is shaped by ecological and evolutionary forces, and the activities of members of the ecosystem are tightly interlinked via various mechanisms of competition and cooperation that affect both commensal microbiota structure and CR against pathogens. Microbial niches in the intestine are spatially structured, and there is an increasing appreciation that spatially resolved analysis will be necessary to fully describe commensal and pathogen processes and interactions. Sequencing-based methods and pure-culture characterization can describe the diverse interdependencies in this system only to a limited extent, and different tools are needed to plunge further into the intestinal jungle. For this task, gnotobiotic animals are key models to deeply characterize how known assemblages of commensal (and pathogenic) microbes interact with each other and with their host. Single-cell tools such as FISH are vital for exploring spatial structuring of the intestinal microbiota. Additionally, single-cell SIP analysis using Raman microspectroscopy or NanoSIMS offers an unparalleled opportunity to experimentally test hypotheses about the activity and physiology of microbiota members in complex ecosystems. These are promising approaches that are valuable to test hypotheses generated in simple systems or by ‘Omics’ methods and to reveal new insights into commensal and pathogen ecology in the intestinal wilderness.

Author's Contribution

B.S. and D.B. contributed equally to this work.


We thank Markus Schmid and Christoph Böhm for providing images for Fig. 3 and Julia Ramesmayer for help with the probe table. This work has been funded by the German Research Foundation (DFG) and the Bundesministerium für Bildung und Forschung (BMBF Infektionsgenomik), the Austrian Federal Ministry of Science and Research (GEN-AU III InflammoBiota), and the Vienna Science and Technology Fund (WWTF) through project LS12-001.


  • These authors contributed equally.

  • Editor: Sebastian Suerbaum