Microbiome and Metabolome Data Integration Provides Insight Into Health and Disease

Abstract

For much of our history, the most basic information about the microbial world has evaded characterization. Next-generation sequencing has led to a rapid increase in understanding of the structure and function of host-associated microbial communities in diverse diseases ranging from obesity to autism. Through experimental systems such as gnotobiotic mice only colonized with known microbes, a causal relationship between microbial communities and disease phenotypes has been supported. Now microbiome research must move beyond correlations and general demonstration of causality to develop mechanistic understandings of microbial influence, including through their metabolic activities. Similar to the microbiome field, advances in technologies for cataloguing small molecules have broadened our understanding of the metabolites that populate our bodies. Integration of microbial and metabolomics data paired with experimental validation has promise for identifying microbial influence on host physiology through production, modification or degradation of bioactive metabolites. Realization of microbial metabolic activities that affect health is hampered by gaps in our understanding of 1) biological properties of microbes and metabolites, 2) which microbial enzymes/pathways produce which metabolites and 3) the effects of metabolites on hosts. Capitalizing upon known mechanistic relationships and filling gaps in our understanding has the potential to enable translational microbiome research across disease contexts.

The Microbiome

An important concept for understanding the state and challenges of microbiome research was articulated by Donald Rumsfeld in 2002, when he said, “There are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns – the ones we don’t know we don’t know…. it is the latter category that tend to be the difficult ones¹.”

The modern microbiome field began with a great discovery of “unknown unknowns”. Starting in the mid-1980s when researchers began using targeted sequencing of ribosomal RNA (rRNA) genes to survey microbial populations², studies began to reveal deep branches of the tree of life previously unknown either because the organisms were invisible to the naked eye or uncultivable in the laboratory³. Considering how many “unknown unknowns” there were about the microbes that live in, on and around us just 30 years ago, the progress made in the field of microbiome research is remarkable.

Several key aspects of microbiome research have contributed to this rapid progress. First, the early adoption of the highly conserved small subunit rRNA gene as a phylogenetic marker allowed for relating all cellular life on earth onto one phylogenetic tree^3,4. Examining novel environmental 16S rRNA gene sequences in an evolutionary context allowed for some properties of uncultivated organisms to be predicted. Even when such functional predictions were not possible (e.g. because the 16S rRNA gene sequence was extremely different from any previously characterized organism), simply moving our understanding of microbial biology from one of “unknown unknowns” to “known unknowns,” allowed for the identification of key gaps in understanding. For instance, the 16S rRNA sequences from an uncultivated group of bacteria in the α-proteobacterial lineage called SAR11 was found in nearly every pelagic marine bacterioplankton community, accounting for 26% of all rRNA genes in sea water. Targeted efforts led to the isolation and biological characterization of SAR11 isolates⁵. Using other –omics techniques coupled with experimental validation, the diverse and complex functions of the SAR11⁶ lineage in the world’s oceans continue to be elucidated⁷.

A second factor that has contributed to rapid progress in microbiome research is next-generation sequencing. These technologies were used for 16S rRNA targeted characterization of microbial communities through use of the 454 sequencing platform in 2008^6,8,9. Starting in 2011, the 454 platform was largely replaced with the more high-throughput Illumina platforms¹⁰. Coupled with development of bioinformatics pipelines for 16S rRNA targeted microbiome analysis such as QIIME¹¹ and of taxonomic databases for linking 16S rRNA gene sequences to named organisms/lineages (e.g. Silva¹² and greengenes¹³), deep sequencing of microbial populations in many different environments, including many human body sites in disease contexts, has occurred. This has led to a plethora of papers describing associations of human microbial communities with diseases ranging from obesity^14,15, depression¹⁶, inflammatory bowel disease (IBD)^17,18 and heart disease^19–21 to autism^22–24.

Next generation sequencing has also enabled the sequencing of shotgun metagenomes and transcriptomes. Shotgun metagenomes survey the “functional potential” of a microbial community by mapping the genes present. The metatranscriptome adds an extra layer of information of the genes actively expressed. Effective functional analysis of microbial communities using shotgun metagenomic and metatranscriptomic data is highly dependent on our underlying knowledge of how gene sequences relate to enzymatic or other functions, which is summarized in databases such as KEGG²⁵ and MetaCyc²⁶. However, as for 16S rRNA databases where many known sequences represent organisms of unknown characteristics, the genes detected in shotgun metagenomes and metatranscriptomes contain many genes of unknown function. As an example, only a median of 18.9% percent of the genes present in bacterial genomes can be mapped to a KEGG Orthologue (KO) of known function based on homology. In some cases, only general function is known for a given gene. For instance, Bacteroides thetaiotaomicron has 280 genes predicted to be glycoside hydrolases in the Carbohydrate Active EnZymes database (CAZY)²⁰ because they have sequence domains related to families of structurally-related catalytic and carbohydrate-binding modules, but only 44 of them are in KEGG.

As described for the SAR11 cluster and 16S rRNA, identifying “known genes of unknown function” that may be of particular relevance due to their ubiquity or a strong correlation with disease, can focus targeted efforts to elucidate their functional roles. In one study, a family of nonribosomal peptide synthetase gene clusters in gut bacteria was targeted for functional analysis because they were present in 88% of the National Institutes of Health Human Microbiome Project stool samples and transcribed under conditions of host colonization²⁵. By expressing a subset of these clusters in Escherichia coli or Bacillus subtilis, it was determined that their gene products lead to the production of peptide aldehydes with protease inhibitory activity that selectively targets a subset of cathepsins in human cell proteomes²⁷. The ubiquity of these enzymes indicates that they have the potential to be an important mediator of gut microbiome/host interactions.

While there have been many advances in microbiome sequencing workflows, the key challenge for the microbiome field, and essential for translational research, is to move beyond associations to an understanding of mechanism. One challenge for generating testable hypotheses regarding mechanistic links between microbial communities and disease is a lack of understanding of microbial function in disease causality. A potential way to address this gap is through the application of metabolomics with microbiome analysis.

The Metabolome

An important tool for understanding microbial community function is metabolomics. Metabolomics is a growing field that comprises the measurement of hundreds to thousands of small molecules in a system²⁸. It encompasses metabolite profiling, lipidomics, and metabolic flux measurements^28,29 and can also include the targeted analyses of specific small molecules, including sex hormones, lipid mediators³⁰, and amino acids³¹.

While genomics, transcriptomics, and metadata provide valuable information regarding phenotypes and molecular mechanisms, metabolomics provides a real-time view of dynamic changes as they occur through the quantitation of small molecules. Small molecules play several roles in the cell including membrane structure (lipids), signaling (lipid mediators, neurotransmitters), and building blocks (amino acids). They are also involved in many processes affected by exogenous influences; these include response to oxidative stress (redox potential), inflammatory response, and energy metabolism, to name a few. Overall metabolomics can be thought of as a functional measurement of dozens to hundreds of molecules. This functional knowledge is essential for understanding the links between microbial communities and disease.

Metabolomics can be conducted using either mass spectrometry-based or nuclear magnetic resonance (NMR)-based platforms. NMR offers reproducible, quantitative measurement of approximately 125 small molecules, including carbohydrates, nucleic acids, and fatty acids^32,33. NMR is highly robust, can be performed in vivo, for instance to obtain biochemical information about the tissues of the human body in a non-invasive way (without the need for a biopsy), and is non-destructive. However, due to the relatively lower sensitivity of NMR, mass spectrometry-based platforms have become increasingly adopted by clinical and research laboratories wishing to conduct metabolomics.

One advantage of metabolomics is that this technical platform is agnostic to the nature of the sample type or treatment group; hence, it is a valuable tool for both hypothesis generation (untargeted approaches) and testing (targeted approaches). However, numerous challenges exist³⁴ including multiple compound name assignments to a single mass, lack of database search scoring algorithms, limited coverage of important pathways using a single technology, uncharacterized metabolic and degradation products, and lack of standardized nomenclature and analytical methods that compare datasets. High throughput samples and analysis as well as data interpretation pose additional challenges; traditionally, the interpretation step has been challenging for researchers when only long lists of potential biomarkers or compound masses (MS) are provided as the final results. Despite the availability of sophisticated pathway analysis programs^35,36, a thorough interpretation of the data has often been beyond the scope of many metabolomics studies; this is especially emphasized in the context of microbiome/host studies, where the origin of metabolites is unclear.

Metabolomics is still limited by many “known unknowns”. Due to their chemical diversity, it is impossible to quantitate every species or even class of small molecule using a profiling approach. Therefore, targeted small molecule analysis (based on robust mass spectrometry assays) are used to quantitatively pinpoint the roles of individual molecules. For example, inflammatory responses can be elucidated by measuring individual lipid mediators that are part of the arachidonic acid pathway³⁰. Examples of targeted small molecule analyses are most often found in newborn screening, where small molecules are used to determine the presence of many metabolic and genetic disorders³⁴. While a precise and accurate means of quantitating small molecules, targeted assays require knowing the molecule of interest and are often limited in the numbers of molecules that can be quantitated.

Additionally, many metabolites cannot be mapped to current pathways. Even knowing the chemical class or species of these molecules may not be enough to deduce their function. Overall, the challenges in both microbiome and metabolome data interpretation highlights the need for efforts to further develop our knowledge so that the results of multiple ‘omics studies can be used to understand mechanisms and phenotypes in human health and disease.

Microbiome Effects on the Host Metabolome

Metabolomics has been applied to a variety of human sample types where bacterial metabolites may be found. These include sites where the bacteria reside and are active including nasal, feces, and skin^37,38, as well as peripheral sites without resident bacterial communities such as blood^39–41, urine^40,41 and liver⁴². The latter are of interest because these metabolites have translocated through host barriers (e.g. skin or mucosal sites) and have potential for more systemic health effects. Metabolomics analyses of microbiome-influenced samples often have a primary aim to measure well-characterized metabolites such as short chain fatty acids (SCFAs), siderophores, and biotransformation products from host metabolites and xenobiotics^28,43–46. Untargeted analyses aimed at understanding microbiome effects on the metabolome of host-derived samples are complicated because the compounds detected represent a complex milieu of environmental compounds as well as those produced by microbes and/or the host.

Determining the origins of metabolites that cause or influence disease is important for developing targeted therapies. Microbes have highly enhanced metabolic capabilities compared to humans; the gut microbiome encodes an estimated ~ 10 million genes⁴⁷ compared to an estimated 19,000 in humans⁴⁸. Microbiomes have the genetic capacity to produce a wide variety of structurally diverse specialized metabolites including natural products, small molecule virulence factors, and signaling metabolites including quorum sensors and autoinducers^37,38,49. Most specialized metabolites, their biosynthetic origins, and their roles in microbiome related disease remain understudied³⁷.

A broad understanding of how the presence of a microbiome effects the host metabolome has come from comparing germ free (GF) mice, (i.e. mice with no associated microbial communities), to mice with colonizing microbes including 1) conventional mouse microbiota³⁹ 2) the microbiota of different humans (i.e. humanized mice)⁴⁰ or 3) specific individual bacterial species^39,40. Similarly, other studies have compared the metabolomes of mice treated with antibiotics to non-antibiotic treated controls^50,51. These studies have found that microbial colonization/composition has a profound influence on both the presence and relative abundances of many metabolites in various sites including the blood ^39,40,42, urine^40,42, feces⁴⁰, liver⁴², and ileum⁴². Differences observed in GF and/or antibiotic treated mice in the various studies include changes in amino acid metabolism pathways (e.g. a loss of transformation of tryptophan into tryptamine and indole derivatives^39–41,51), polysaccharide metabolism (e.g. increases in non-digestible plant-polysaccharides in feces⁴⁰), decreases in SCFAs, increases in primary bile acids and decreases in secondary bile acids⁵¹ and elevation of creatine and creatinine^40,52,53. Microbiota colonization also results in differences in many small compounds of unknown identity/function⁴⁰.

Another way to estimate which metabolites in host samples originate from host versus microbial metabolism is to use metabolic networks described in databases such as KEGG²⁵. These networks capture the relationship between metabolites, the enzymes that produce them, and the organisms (both host and microbial) that contain genes encoding those enzymes (Figure 1A). These networks can, in principle, provide a framework for relating the genes/enzymatic reactions present in the colonizing bacteria (genome/shotgun metagenome), the expression of those genes (meta-transcriptome), and presence of the encoded proteins (meta-proteome), and the metabolites (metabolome) in a sample (Figure 1B). They also provide a framework for investigation of microbiome/host co-metabolism (Figure 1C).

Figure 1.

(A) The central dogma of molecular biology is the information flow of genes to transcripts to proteins. In the –omics age, we can observe this information flow through the characterization of genomes, transcriptomes and proteomes. The metabolome, the collection of small compounds present in a biological sample, can be seen as a natural extension of the central dogma. These metabolites can be associated with disease and we can work back from metabolites, to proteins that catalyze the reactions that produce them, to the transcripts that are translated to the genes that encode those transcripts. (B) This model can be extended to include multiple organisms as would be present in a microbial community and the reactions and metabolites that are produced and consumed form connections between the community members. (C) The host can be added back into this network to find host-microbiome interactions.

Linking Microbiome/Metabolome with the Exposome

Another important level of complexity is how the metabolome and microbiome are linked with the exposome (e.g. exogenous factors such pollutants, hygiene products, and the diet). One of the most studied facets of the exposome is diet. Diet composition substantially alters the human microbiome^54,55 and diet induced changes can occur as rapidly as within a day^54,56. This change is accompanied by alteration in bacterial gene expression⁵⁴, suggesting that the microbiome is elastic and poised to quickly respond to varied diet exposures.

Studies in both humans^57–60 and animal models⁴⁰ have shown that consumption of different diets is associated with changes in microbially-produced metabolites. For instance, high fiber diets have been associated with the increases in both fecal^61,62 and circulating⁶³ levels of SCFAs. Another study of humans on a high protein diets saw an increase in levels of the carcinogenic N-nitroso compounds compared to a maintenance diet, thought to be driven by increased protein fermentation and decreased carbohydrate fermentation by colonic microbiota⁶². Broader effects of fiber consumption on the fecal and urine metabolome were observed in a study comparing mice on a standard low fat (LF) to a high fat (HF) diet⁶⁴ which suggests that lack of complex polysaccharides may be a large contributor to the HF-diet associated microbiome/metabolome.

Though a significant facet of the exposome, daily exposures extend beyond diet. A large area of exposure is the skin which is composed of a diverse and rich microbiome, varying by body site⁶⁵. Spatial mapping of the microbiome and metabolome across 400 sites of the human body revealed associations between microbes and the metabolites present⁶⁶. Most notably Propionibacterium was associated with sebaceous sites and highly correlated with 491 metabolome features, 73% of which were lipids. Further in vitro analysis revealed that Propionibacterium acnes can oxidize oleic acid, the breakdown products of which were found in sebaceous sites⁶⁶. These findings suggest that the skin microbiome influences the chemical landscape of the skin metabolome.

The host microbiome is also important in xenobiotic degradation. Gut microbes can metabolize a range of environmental chemicals in ways that can affect their toxicity to the host, including polycyclic aromatic hydrocarbons (PAHs), nitrotoluenes (which are important intermediates in the manufacture of dyes and plastics), and pesticides⁶⁷. PAH degrading microbes have also been observed on human skin⁶⁸. Additionally, gut microbes have also been associated with breakdown and reactivation of a variety of drugs including the cardiovascular drug, digoxin and the cancer drug, irinotecan⁶⁹.

Identifying Important Microbially-Produced Metabolites for Health and Disease

Studies aimed to explore an association between microbial metabolism and disease states often focus on well-characterized microbial metabolites such as SCFAs⁷⁰. However, approaches that can identify novel bio-active microbial metabolites as mediators of microbiome-host or microbe-microbe interactions are increasingly being applied⁷¹. These include: 1) Bioactivity-guided screening, in which a phenotypic bioassay is performed by testing a collection of microbes, chemical extracts, or cosmid libraries generated from metagenomics DNA against a panel of host pathogens or signaling pathways^72,73. 2) Candidate molecule workflows, which mine existing (meta)genomics data to identify Biosynthetic Gene Clusters (BGCs) that may produce metabolites of interest because they are prevalent or correlate with disease⁴⁹. 3) Integrative analysis of microbiomes and untargeted metabolomes in disease contexts^{20,50,74–76}. All of these approaches can identify molecules with potential roles in mediating host-microbe interactions that can be further validated in in vitro or in vivo systems.

Bioactivity-guided workflows have led to the identification of several metabolites that mediate interactions between microbes and the host or other microbes. For instance, commendamide, which activates the G-protein-coupled receptor G2A/GPR132 (G2A has been implicated in disease models of autoimmunity and atherosclerosis), was discovered initially in a screen of 3,000 megabases of metagenomic DNA for effectors that activate NF-κB, a transcription factor that plays a role in mediating responses to environmental stimuli⁷². Lugdunin, a novel antibiotic that prohibits colonization by methicillin-resistant Staphylococcus aureus (MRSA), was discovered by screening for the mechanism by which Staphylococcus lugdunensis, a microbe that occupies the same niche as MRSA, could inhibit the growth of MRSA in vitro⁷³. Bioactivity-guided workflows can also include random mutagenesis of the microbe of interest combined with high-throughput phenotypic screening for loss of function, and imaging mass spectrometry of competing microbes in vitro ^72,77,78.

Candidate molecule workflows identify BGCs that allow microbes to produce metabolites and capitalize on knowledge of the core enzymatic machinery responsible for their biosynthesis, which has been well-characterized for many classes of specialized metabolites⁴⁹. This knowledge has been used to develop and implement bioinformatics platforms, such as anti-SMASH and related platforms, to detect widely distributed and closely related gene cluster families, identify potential metabolite producers, and guide purification workflows^79–84. Candidate molecule methods have also led to the identification, structural elucidation and/or synthesis of metabolites that mediate interactions between microbes and the host and/or other microbes. These include colibactin, a secondary bacterial metabolite with genotoxic properties⁸⁵, lactocillin, a potent thiopeptide antibiotic made by a prominent member of the vaginal microbiota ^27,49, (dihydro)pyrazinones, which are prevalent in the human gut and that release a peptide aldehyde that inhibits cathepsins in human cell proteomes²¹, and humimycins, antibiotics with activity against MRSA⁸⁶. Candidate molecule workflows can also be combined with tandem mass spectrometry (MS/MS) data to link MS/MS spectra of unknown metabolites to their biosynthetic origins, simplifying compound isolation and characterization^87,88.

Integrative analysis of microbiomes and untargeted metabolomes in disease contexts have also allowed for the identification of microbially produced compounds found to mediate disease upon further functional characterization. For example, one study used a maternal immune activation (MIA) murine model that developed symptoms commonly associated with Autism Spectrum Disorders (ASD). These MIA mice also had altered gut microbiome and serum metabolome profiles compared to control mice. Oral treatment of the MIA mice with the human commensal Bacteroides fragilis ameliorated ASD-like behaviors. The metabolite 4-ethylphenylsulfate (4EPS) was focused on for functional study because its levels were increased in the MIA mice and restored to control levels by B. fragilis treatment. Interestingly, systemic administration of naïve wild-type mice with 4EPS potassium salt was sufficient to induce anxiety-like behavior similar to that observed in MIA offspring⁷⁴.

In another study that characterized the fecal microbiome and serum metabolome of non-diabetic Danish individuals, serum levels of branch-chain amino acids (BCAAs) were higher in insulin-resistant individuals and also correlated with fecal microbiomes with higher biosynthetic potential for BCAAs conferred by the species Prevotella copri and Bacteroides vulgatus. These predictions, made based on correlative analyses and functional predictions, were further validated by showing that P. copri could induce insulin resistance while increasing circulating BCAA levels when introduced into High Fat Diet (HFD)-fed mice⁸⁹.

In some cases, associations between microbially produced metabolites and disease have also been linked with the exposome, including diet. One study showed that a dietary fiber associated reduction in allergic lung inflammation was driven by the SCFA propionate, which exerted its health effects through G protein–coupled receptor 41 mediated signaling in bone marrow⁷². Other work demonstrated that risk of cardiovascular disease and atherosclerosis was linked with trimethylamine-N-oxide (TMAO), which is produced by host enzymes from trimethylamine (TMA). TMA is a metabolite of dietary phosphatidylcholine and L-carnitine that can be produced by intestinal bacteria but not host enzymes and whose production is associated with particular human gut microbiome compositions^20,76.

Metabolites from non-targeted metabolomics can be prioritized for functional exploration using various analytic techniques. As mentioned above, correlative analyses link the presence of a specific metabolite to genotypic and/or phenotypic data and may suggest that a metabolite is produced or induced by the presence of a specific microbe^90–94. Other methods include Self Organizing Map generation, which applies multivariate statistics to differentiate samples by identifying the largest changes in metabolite abundance between case and control^95–97, and dereplication, or the identification of known molecules. Dereplication by molecular networking uses MS/MS data to visualize the chemical space of samples to pinpoint an in vitro producer of a metabolite produced in vivo ^66,98–100. Molecular networking measures relatedness between MS/MS spectra as a proxy for molecular similarity. MS/MS metabolomics data derived from in vitro cultivation of members of the microbiota can be used to seed molecular networks in order to identify either known compounds produced by those members of the community or close analogs^99–101

Functional Validation

Once associations between microbiomes and metabolomes in disease have been identified, the roles of specific metabolites or metabolomics patterns must be validated. If a compound is characterized and commercially available, validation is most straightforward. For uncharacterized compounds, molecular biology workflows can be used; for instance, metagenomics-derived BGCs can be expressed in heterologous hosts as was recently used to characterize the (dihydro)pyrazinones from human gut microbiota²⁷. Additionally, if the microbe that produces the metabolite can be grown in pure culture and genetically manipulated, regulators of BGCs can be removed, replaced, or enhanced to increase in vitro metabolite production to levels necessary for isolation, as was recently demonstrated during the characterization of lugdunin^73,102.

Despite recent success characterizing a handful of individual metabolites, many specialized metabolites of interest are understudied, in part due to limitations of cultivation. To circumvent these limitations, conventional approaches for microbiota growth have been modified to include environment specific media^103–106, optimization of growth conditions based upon (meta)genomic information¹⁰⁶, inclusion of helper organisms^107,108 and higher throughput cultivation strategies^{105,109–112}. Efforts are also underway to cultivate both native and designer microbiome communities using in vitro and in vivo model systems¹⁰⁴ and new model systems continue to be developed^113,114.

Animal models are widely used to explore the functional attributes of microbes and their metabolites, including both conventionally raised and GF models of zebrafish, mice, rats, and pigs¹¹³. While GF models have been a powerful tool for experimental validation of microbial contributions to disease, use of these models has sometimes been criticized as these animals have alterations in essential physiological processes^113,115. Treatment of conventionally raised animals with broad-spectrum antibiotics cocktails followed by repopulation with chosen microbes is a commonly used alterative^{48,50,116,117} and may circumvent some issues with GF animal model systems, but have the limitation that antibiotics never completely remove indigenous microbes and also may have other effects on the host. Successful use of animal models to functionally validate the effects of microbially-produced metabolites in disease were described above for studies of atherosclerosis^20,76,118, ASDs⁷⁴, and diabetes⁶². Another elegant example of using molecular biology in conjunction with animal models to identify key metabolites is the elucidation of the role of microcins produced by probiotic E. coli strains in modulating Enterobacteriaceae in the inflamed gut¹¹⁹.

While animal models are most often chosen to explore functional activity of microbes and their metabolites, in vitro systems can provide an intriguing alternative¹¹³ due to their higher throughput, flexibility, scaling, and ability to simplify the system for high-resolution molecular analyses. These in vitro models include independent cultivation of the microbiota and interfacing the members of the microbiota with host cell lines using Transwell systems or, eventually, organ-on-a-chip microfluidic devices¹¹³. While a number of groups are attempting to systematically construct artificial communities for robust evaluation of microbiome modulation by both organisms and metabolites, such as probiotics and pharmaceuticals respectively, development of environment-specific media has led to some success recapitulating in vivo microbiome communities in vitro for similar purposes^{103–105,110,112,120}. For example, stable oral microbial biofilm communities representing 60–80% of the taxonomic diversity of the original sample has been accomplished¹²¹, while fecal communities were recapitulated at 65–95%¹²². Miniaturized bioreactors and chemostats may provide opportunities to maintain continuous culture of communities removing concerns about sample limitation from patient donors, particularly of microbiome niches that are not as easily accessible as stool^113,123. To incorporate host cells, co-culture systems have been developed and models for the context of gut microbiome investigation have been recently reviewed¹²⁴. These models include co-culture in tubes (HoxBan system), Transwell systems, and microfluidic devices. While microfluidic systems, such as the HuMiX model, are more compatible with multi-omics analyses, it suffers from poor chemical compatibility, challenges with parallelization, and lack of widespread availability^113,114,124.

Towards Microbiota-targeted Therapeutics

Microbiome/metabolome targeted therapies include both those that inhibit detrimental microbes/metabolites and those that promote beneficial microbes/metabolites (Figure 2). The microbiome/metabolome can also potentially inform personalized medicine, if the treatment of an individual is tailored based on the composition or metabolic capabilities of their microbiome. The complicated and diverse roles in disease of the microbiome are only beginning to be elucidated; thus, most promising research directions have yet to reach their full potential in the clinic.

Figure 2.

Microbiome-targeted therapies can target detrimental microbes and/or their metabolites (red) or promote beneficial microbes and the production of their metabolites (green). In principle, the presence of beneficial microbial metabolites in the host can be promoted by adding the metabolite itself (dietary intervention or supplementation), the microbe that produces the metabolite (probiotics), the substrate the microbe uses to produce the metabolite (prebiotics) or the beneficial metabolite itself (postbiotics). The presence of detrimental microbial metabolites can be reduced by targeting the microbe that produces it directly (antibiotics) or through competitive exclusion (probiotics/fecal microbiota transplant), by inhibiting the enzymes (shown in pink) that produce the metabolite, or by removing the substrates for that metabolite from the diet.

Targeting detrimental bacteria and metabolites

Although antibiotic treatment can be valuable for discouraging growth of bacteria that produce detrimental metabolites, broad effects on other (beneficial) microbes and rapid acquisition of antibiotic-resistance limits their utility. Thus, another promising approach is competitive exclusion by other “probiotic” bacteria. Perhaps the best evidence that this approach can be successful clinically is the 92% success rate of fecal microbiota transplantation (FMT) for the treatment of recurrent Clostridium difficile infection¹²⁵. However, understanding metabolic mechanisms of inter-species interaction can allow for better “precision microbiome reconstitution”¹²⁶ for treatment of C. difficile and potentially of diseases where FMT has had more variable success such as IBD^127,128. In the case of C. difficile, protection in mice has been conferred by introducing specific bacteria (e.g. Clostridium scindens) that can produce key metabolites (secondary bile acids) that affect C. difficile sporulation and growth and thus infection¹²⁶; however, efficacy and safety trials in human populations still need to be conducted.

Another mechanism for preventing detrimental microbial metabolic activity is to inhibit the metabolic pathway rather than the bacteria that contain it. This can be accomplished by limiting inputs to the pathway (e.g. by altering dietary intake) or by inhibiting bacterial enzymatic activity (Figure 2). One promising example is the treatment of cardiovascular disease by inhibiting microbial production of TMA using a structural analog of choline that can inhibit microbial TMA lyases¹¹⁸. In another example, the amino acid arginine can inhibit enzymatic conversion of the cardiac drug digoxin into an inactive form by the gut microbe Egerthella lenta, indicating that co-administering this drug with arginine may prevent its deactivation by microbial enzymes²¹. However, both of these enzyme-targeted treatment strategies still require validation in clinical cohorts.

Promoting beneficial bacteria and metabolites

Therapies that promote the prevalence/activity of beneficial microbes and their metabolites also have promise for clinical applications that has likely yet to be fully realized. Current probiotic treatments use a small number of intestinal microbes (e.g. various Lactobacillus or Bifidobacteria species, the yeast Saccharomyces boulardii, or E. Nissle) that have already been approved for use in humans and that only represent a small component of bacteria that can colonize the human gut. In some cases, specific probiotics have been shown to affect both microbiome and metabolome composition while promoting health; for instance, a trial in patients with liver cirrhosis found supplementation with Lactobacillus GG improved outcomes (decreased endotoxemia and TNF-α levels) with concomitant changes in microbiome composition and metabolite/microbiome correlations pertaining to amino acid, vitamin and secondary BA metabolism¹²⁹. As the literature on these probiotics is vast, consensus opinion regarding their utility in the clinic are valuable; the Triennial Yale/Harvard Workshop on Probiotic Recommendations¹³⁰ gives A–C evidence ratings for the use of specific probiotics in necrotizing enterocolitis, childhood diarrhea, IBD, irritable bowel syndrome, C. difficile diarrhea and liver disease.

There are several bacteria that have shown positive effects in mouse models that may represent a “next generation” of probiotics once safety and efficacy studies in humans can be conducted. For instance Bacteroides uniformis and Akkermansia muciniphila both reduce weight gain and metabolic disease pathology in HFD-fed mice^131,132; Bacteroides fragilis, Bacteroides cellulosilyticus, and cocktails of T regulatory cell inducing, spore-forming Clostridia have all been shown to protect against intestinal inflammation in mouse models^133–135. The mechanism by which probiotics may confer a health benefit are often not completely understood and include other functions besides the production of small molecules, such as the production of immune-modulatory capsular components¹³³.

Probiotic applications can be challenged by colonization resistance; This challenge can potentially be circumvented by promoting growth and metabolic activity of beneficial bacteria using prebiotics or synbiotics. Prebiotics are broadly defined as any compound that affects the composition or function of the microbiome to exert a beneficial effect upon the host following bacterial metabolism¹³⁶. Prebiotics are typically fibers and starches extracted from plants or post-fermentation processes with SCFAs being one desirable metabolic product of their digestion ^137–142. Synbiotics are formulations that contain both probiotics and prebiotics together. Understanding the metabolic environment necessary for the successful growth of microbes can lead to the intelligent design of prebiotics and synbiotics. For instance, galactooligosaccharide mixtures defined in laboratory culture experiments to have particularly strong prebiotic properties for Bifidobacteria, effectively increased the proportion of Bifidobacteria in the gut in a dose dependent manner when given to healthy human volunteers¹³⁸. Use of genomic and metabolic data to design prebiotic strategies that target specific microbes and metabolites will become more powerful as our knowledge bases of microbial enzymes and metabolites expand.

Beneficial microbial metabolites can also be directly introduced to the host to circumvent the need for microbial production from dietary substrates, a process referred to as “microbiome-based metabolite treatment” or “postbiotics”¹⁴³(Figure 2). For instance, the anti-inflammatory microbial fermentation product, butyrate, has been added directly to the colonic environment via enema in efforts to ameliorate inflammation in patients with ulcerative colitis 144,145, although only minor improvements were observed¹⁴⁶. The exploration of “postbiotic” therapies are in their infancy and potentially challenged by unintended effects of the compound on gut microbiome composition and activity, and difficulty in determining and delivering physiologically important levels of the metabolite to its active site in the body¹⁴³.

Personalized medicine

Finally, microbiome/metabolome information has the potential to inform personalized medicine. In one elegant example, gut microbiota composition was included in machine learning-algorithms that also considered blood parameters, dietary habits, anthropometrics, and physical activity, to predict personalized glycemic responses to meals. This model was used to perform personalized dietary intervention to lessen elevated postprandial blood glucose and its metabolic consequences¹⁴⁷.

Another promising direction of microbiome/metabolome informed personalized medicine is to consider the role of the microbiome on drug metabolism/pharmacokinetics for proper drug dosing. For instance, evaluation of the presence of digoxin-deactivating strains of the gut bacterium E. lenta has the potential to inform proper administration of this cardiac drug¹⁴⁸. Another study suggested that urinary levels of the microbial product p-cresol sulfate (as determined by (1)H NMR spectroscopy) could indicate the metabolic fate of acetaminophen¹⁴⁹.

Future Directions

It’s been almost nine years since we were all invited to the marriage of metagenomics and metabolomics¹⁵⁰ and progress in integrating these data types and elucidating the function of key small-molecules that are produced by the microbiome is being made. Understanding the metabolic capabilities of the microbes that inhabit all areas of the human body will be key to determining their effects on health and disease. To advance this knowledge, more work remains to be done. While 16S rRNA can be used to conveniently survey the composition of microbial communities, its ability to predict metabolic effects is limited both by the incomplete knowledge present in bacterial genomic databases and strain level variation. Metagenomic sequencing gives more knowledge of the genes present, but the majority of these genes still have unknown function. KEGG is still the most used database for orthologous gene group assignment and is the most comprehensive database for linking orthologous gene groups to reactions and annotated compounds²⁵. KEGG has moved to a subscription service model since 2011 and as such cannot be integrated into new open source tools. MetaCyc²⁶ is free to use and redistribute but is missing features from KEGG such as a universal gene orthology. As the gaps in these databases are filled through genome sequencing and both large-scale and targeted biochemistry and/or functional studies, our power to understand the functional potential of host associated microbial communities will increase.

Metabolomics has perhaps even greater challenges than the evaluation of microbial community structure and function using DNA and RNA sequencing. Extraction techniques are not standardized and subject to technical variability. Peak extraction and annotation of mass spectroscopy data is challenged by multiple metabolites matching to the same mass³⁴. Current best practices do not give confident results without further experimental validation and quantification which is not currently possible on the scale of detected metabolites in untargeted experiments. As the technology for untargeted mass spec metabolomics becomes more mature, these issues will hopefully become less prominent allowing understanding of metabolomes across studies. Paired with an increased biochemical understanding of the gene products encoded by microbiomes, the ability to develop a mechanistic understanding of how microbial communities influence the host via their metabolic activities will be enhanced. To rapidly advance microbiome and metabolome integration to understand roles for microbial metabolism in diverse disease contexts, advanced yet “user friendly” methods for multi-omic integration need to be developed.

Even with these formidable challenges, many studies have identified key microbes and their enzymes and metabolites that can be targeted therapeutically in the context of varied diseases. A key challenge in the field is to move many of these novel discoveries from the bench to the bedside. By understanding the metabolic interplay between microbiome and host, novel prebiotics and probiotics can be explored, as well as personalized treatment of disease that capitalize on knowledge of microbiomes and their interactions with the host.

Abbreviations

SCFA

Short chain fatty acids

BGC

Biosynthetic Gene Cluster

MS/MS

Tandem mass spectrometry

rRNA

Ribosomal RNA

GF

Germ Free

KO

KEGG Orthologue

LC/MS

liquid chromatography/mass spectroscopy