Current Capabilities of Gut Microbiome–Based Diagnostics and the Promise of Clinical Application

Abstract

There is increasing evidence for the importance of the gut microbiome in human health and disease. Traditional and modern technologies - from cell culture to next generation sequencing - have facilitated these advances in knowledge. Each of the tools employed in measuring the microbiome exhibits unique capabilities that may be leveraged for clinical diagnostics. However, much still needs to be done to standardize the language and metrics by which a microbiome is characterized. Here we review the capabilities of gut microbiome-based diagnostics, review selected examples, and discuss the outlook towards clinical application.

Evidence for the role of the gut microbiome in human disease is accelerating. Next-generation technologies are producing orders of magnitude more data than were available to the scientific community <2 decades ago. As understanding of complicated interactions between microscopic ecosystems and human health grows, so will pressure for translational microbiome-based diagnostics to detect or predict disease. Despite major efforts to characterize microbiome profiles from apparently healthy individuals, criteria that might be applied to diagnose, much less quantify, abnormal configurations or to reassure patients and providers remain limited to a small number of diseases. Toward understanding the bounds of this challenge, we describe the current landscape of validated microbiome measures with diagnostic relevance, outline key modalities used in microbiome research, and illustrate the capabilities of these modalities through selected examples.

MEASURES OF THE MICROBIOME

Here we refer to the human microbiome as the totality of a microbial ecosystem encompassing organisms, genomes, and microbial products (eg, proteins and metabolites) [1]. Occasionally this term is used to describe more specifically distributions of genomes or marker genes to characterize a microbial ecosystem [2, 3]. Investigation of microorganisms within a microbiome—that is, the microbiota—has emphasized prokaryotes in the gastrointestinal tract in addition to the lungs, skin, and female genitourinary tract, among other sites. Fungal (mycobiome) and viral (virome) components are also of increasing interest with improved amplicon sequencing and extraction methods but less well described. Numerous methods for characterizing the microbiome exist, but they generally lack standardization. It is therefore helpful to delineate measures of the microbiome that may be used in clinical diagnostic applications.

Detection of specific taxa, ideally at the species or strain/pathotype level, is the most straightforward measure evaluated by microbiome-based diagnostics. This is also the most analogous result to currently approved diagnostics such as multiplex polymerase chain reaction (PCR) panels. Quantitative PCR (qPCR) or flow cytometry has also been used to estimate microbial load, or the agnostic quantification of microbiota in a sample.

Sequencing taxonomic results—facilitated by next-generation sequencing (NGS) methods including 16S ribosomal RNA (rRNA) amplicon and shotgun metagenomic sequencing (MGS)—are generally reported as relative abundance, the proportion of the total sequencing effort that match a reference. This may be counterintuitive for clinicians used to quantitative or semiquantitative results reported by clinical microbiology laboratories, because a major limitation of relative abundance as a diagnostic measure is its failure to reflect true changes among microbiota when absolute quantities change but proportions remain unchanged [4]. Changes in overall quantities of bacteria are captured in measures of absolute abundance, expressing the quantity of each species in units relative to an external measure and in theory the product of relative abundance and total microbial load. Equality of abundance between species of a microbiome may be described in terms of evenness. While in the research realm this does not inform analysis, it may be an important aspect if quantitation of a bacterial family is important for a diagnostic phenotype, such as for diagnosis of inflammatory bowel disease [5].

Investigations of microbiome diversity draw on concepts in biodiversity long used by ecologists and can be described in terms of an individual microbiome (alpha diversity [6]) or how individual microbiomes compare to the diversity of a larger community (beta diversity). In this context, various metrics common to biodiversity are found in the microbiome literature. Richness is an aspect of diversity that refers to the total number of species present in a microbiome [7]. At first glance, it may be tempting to assume that more diversity is always better, but purely compositional taxonomic measures generally fail to account for measures of microbiota function in situ.

The resistome refers to the subset of the microbiome that confers or exhibits resistance to antimicrobial agents. Resistome measurements currently lack standardized thresholds for defining presence or absence [8], yet are increasingly reported and meaningful metrics in microbiome research with important clinical implications. Importantly, validated resistome measures may support improved public health surveillance for taxonomic-independent transmission of antimicrobial resistance determinants. Finally, dysbiosis is a term intended to describe pathologic deviation of the microbiome from normal but is imprecise and is usually expressed in terms of alterations in presence or absence of taxa and resistance patterns, changes in microbial load, abundance, richness, or some combination of these. The following reviews major methods specifically in gut microbiome–based diagnostics and describes capabilities of each (Table 1) with respect to these measures.

Table 1.

Capabilities of Methods for Measuring the Microbiome

Method	Bacteria	Fungi	Viruses	Relative Abundance	Absolute Abundance	Species Richness	Resistome
Culture	Culturable organisms	Culturable organisms	Culturable organisms	Culturable organisms	Semiquantitative, culturable organisms	Culturable organisms	In vitro phenotypic susceptibility
qPCR	Known targets	Known targets	Known targets	Limited	Limited	Limited	If known resistance sequences exist
16S rRNA sequencing	All bacterial targets	Analogous methods (ITS and 28s)	No	Yes	No	Yes	Possibly inferred from taxonomic information
MGS	Yes	Yes	Yes	Yes	No	Yes	Identifies known resistance sequences
QMP	Yes	Depends on sequencing and quantitation technique	Depends on sequencing and quantitation technique	Yes	Yes	Yes	Depends on sequencing technique (16S or MGS)
Metabolomics	Reflects impact of microbiome composition	Reflects impact of microbiome composition	No	No	No	No	Undefined

Abbreviations: ITS, internal transcribed spacer; MGS, metagenomic sequencing; QMP, quantitative microbiome profiling; qPCR, quantitative polymerase chain reaction; rRNA, ribosomal RNA.

GUT MICROBIOME–BASED DIAGNOSTIC METHODS

Culture-Based Methods

Culture-based methods have been the workhorse of clinical microbiology for almost a century. These methods are ubiquitous in clinical settings but too laborious for broad characterization of a microbiome. Furthermore, anaerobes and low-abundance microbes are notoriously difficult to culture with standard techniques. As such, culture-based microbiome diagnostics remain limited to narrow hypothesis-driven identification of specific taxa or antibiotic resistance (AR) patterns and, although capable of describing results semiquantitatively (eg, colony counts), are not amenable to broad measurements of abundance and richness. The best examples of culture-based microbiome diagnostics in clinical prediction generally involve infection control surveillance of antibiotic-resistant pathogens harbored in the microbiome of hospitalized patients, such as active screening for vancomycin-resistant Enterococcus spp. [9]. Culture-based methods may be better harnessed if an appropriate surrogate could be established to represent an indicator of “abnormality,” such as drug-resistant organisms that should not be present [10] or overgrowth of yeast or Pseudomonas on a routine stool culture [11].

qPCR Testing

The standard of care for pathogen detection in many infectious diseases, qPCR platforms are available in most clinical settings and generally increase the sensitivity and limit of detection of a diagnostic test compared to culture, if that is preferable for the clinical syndrome. qPCR can detect taxonomy-specific and AR-determining targets but is limited in that targets for amplification must generally be prespecified. In one example, qPCR detection of the Fusobacterium nucleatum butyryl-CoA dehydrogenase gene and the rpoB gene (encoding RNA polymerase subunit β) from Parvimonas micra in the gut microbiome was investigated as a screening diagnostic test for colorectal cancer [12]. These targets were first derived from MGS data in multiple cohorts and then validated for screening in a prospective cohort. In a second example, PCR targets specific for Bacteroidetes have been used as enrollment criteria for clinical studies of microbiome therapies, though validation for this approach is needed [13]. The speed, sensitivity, specificity, and quantitative component of qPCR data will likely retain an important role in clinical detection of biomarkers for which significance has been characterized with untargeted discovery methods.

16S rRNA Sequencing

The RNA sequence encoding the 16S rRNA subunit in bacteria has been a key component of taxonomic study for decades owing to highly conserved regions interspersed with hypervariable regions across species [14]. NGS technology allows for relatively rapid sequencing of these targets, and clinical 16S sequencing has found a niche for pathogen identification in culture-negative infections, especially endocarditis [15]. 16S sequencing can be used to target bacteria-specific nucleic acid, which reduces the eukaryotic genetic signal (including the host’s own genetic material) and is even the reference standard for diagnosis of some difficult-to-culture organisms [16], but it does not capture any additional genomic data. Therefore, this technique is primarily useful for species- and genus-level identification, and not identification of other genetic elements, such as AR genes. Similar targets for fungal organisms (internal transcribed spacers and the 28S subunit) are amenable to an analogous approach [14], but viral genomes do not exhibit such a target. In contrast with culture and qPCR methods, NGS instruments necessary for 16S sequencing are most common in referral laboratories. These can be limited as to whether fresh samples can be sequenced, or whether only isolates can be sequenced in these laboratories.

Research applications for 16S sequencing have focused on identifying gut microbiome features that predict pathogen colonization and infection, including with antibiotic-resistant organisms. For example, gut microbiome profiling in a prospective cohort of patients admitted to an intensive care unit showed that high abundance of “protective” bacterial taxa (eg, Prevotella and Morganella spp.) was associated with lower odds of subsequent carbapenem-resistant Pseudomonas aeruginosa colonization [11]. In addition to allowing for detection of infrequent commensal species that confer protection from subsequent pathogen colonization or infection, 16S sequencing allows for determination of relative abundance of pathobionts, which may be important in determining risk for subsequent infection [17].

A relative abundance of Klebsiella pneumoniae carbapenemase–producing K. pneumoniae ≥22% among 506 patients in long-term acute care predicted subsequent K. pneumoniae carbapenemase–producing K. pneumoniae bacteremia [18], and similar associations between intestinal “domination” (≥30% relative abundance) and subsequent bacteremia have been shown with vancomycin-resistant Enterococcus [19] and several gram-negative species (including Escherichia and Enterobacter spp.) in hematopoietic stem cell transplant recipients [20]. Use of 16S sequencing to detect specific microbiome features that either protect from or confer risk for subsequent infections has not yet found clinical use but may allow for antibiotic prophylaxis in selected high-risk patients or targeted antibiotic therapy in patients with suspected AR. This approach can be used as a crude surrogate that can then be used to do deeper MGS, or to screen for putative bacteria to cultivate or use in microbiome therapeutics.

MGS Applications

MGS uses unbiased deep sequencing of all genetic material in a sample. This is facilitated by both NGS technology and modern computational tools that perform reconstruction of sequence fragments into complete genomes, separate host from microbial genetic material, and identify sequence patterns of interest—a method that has proved helpful in the diagnosis of culture-negative infections [21, 22]. An advantage with respect to the microbiome of the unbiased and agnostic MGS approach is that bacterial, viral, fungal, and other eukaryotic genetic sequences are included. Furthermore, functional aspects of the gut microbiota can be derived though metabolic modeling, and AR can be identified based on sequences.

A relatively practical diagnostic application of microbiome MGS is the identification of taxa or metabolic patterns as biomarkers. A meta-analysis combining data from 4 geographically distinct cohorts identified 4 clusters of 29 bacterial species that were associated with colorectal cancer stage and tumor location [23]. Performance as a screening method was validated in 2 separate cohorts and had greater accuracy than similar methods based on 16S sequencing [23]. Three genes for bacterial virulence factors—information that cannot be derived from 16S sequencing—were also associated with colorectal cancer in the validation cohorts [23]. In another example, a machine learning model based on gut microbiome MGS in patients with nonalcoholic fatty liver disease identified 37 bacterial species that, along with age, body mass index, and alpha diversity, could be used to diagnose advanced liver fibrosis [24].

Antibiotic exposure has been associated with reduced survival in patients with cancer undergoing anti–programmed cell death protein 1 (PD-1) immunotherapy, a phenomenon thought to be linked to the microbiome [25]. Three studies examining the relationship of the microbiome with efficacy of PD-1 immunotherapy were simultaneously published in 2018 and included robust MGS microbiome analysis [26]. The presence of multiple specific bacterial species was associated with response to anti–PD-1 therapy in a cohort of patients with non–small cell lung cancer and renal cell carcinoma [25].

A second study combined 16S sequencing and MGS of the gut microbiome in patients with metastatic melanoma to identify 43 unique operational taxonomic units differentially expressed between immunotherapy responders and nonresponders [27]. A ratio of beneficial to nonbeneficial operational taxonomic units >1.5 was associated with clinical response [27]. A third study, also in patients with melanoma, combined taxonomic information from 16S sequencing with metabolic models from MGS data to identify an anabolism-predominant group with a greater response to anti–PD-1 therapy than a catabolism-predominant group [28]. Although these studies each concluded that the gut microbiome impacts human immunity with implications for immunotherapy—and that antibiotic exposure is a highly important influence—they each used MGS in distinct ways, thereby highlighting the method’s multiple uses.

By virtue of “deep sequencing” the microbiome, MGS data contain sequence information from which AR may be inferred through comparison to a reference library. In one example, the gut resistomes of a group of returning travelers demonstrated acquisition of extended-spectrum β-lactamase, AmpC, and quinolone resistance genes that persisted as long as 6 months after returning [29]. A second example used MGS to identify broad changes to the gut microbiome associated with use of various medications including alterations in species abundance, metabolic pathways, and AR [30]. The effect of pharmaceuticals on the resistome was quantified by determining the number of AR genes present in each microbiome—proton pump inhibitors and metformin were associated with an increase in a general cohort, while opiates and tricyclic antidepressants were associated with increased number of AR genes in a cohort with inflammatory bowel disease [30].

MGS can be used to analyze the incredible complexity of the gut microbiome and as a diagnostic serves as a platform from which hypotheses of interventions can be generated. Its use as an actionable clinical diagnostic is currently hampered by the turnaround time and lack of standardization of the analysis pipeline. However, given the progress in these aspects, it is possible that phenotypes of a complex composition could be interpreted and stratified, as in the risk for disease outcome or for acquisition of another chronic disease.

Quantitative Microbiome Profiling

NGS, including both 16S sequencing and MGS, is compositional rather than quantitative. Absolute quantitation of abundance may be important to fully capture how an intervention affects the microbiome when relative abundance remains unchanged but absolute quantities of microbiota are affected. Methods for quantitative microbiome profiling (QMP) usually describe coupling NGS with flow cytometry or qPCR for bacterial enumeration.

Differences in relative versus QMP were demonstrated in a cohort of patients with Crohn disease, compared with controls, by determining taxonomic information with 16S sequencing and microbial load by flow cytometry [31]. Results showed that relative differences in Bacteroides spp. between the groups did not retain significance when examined by QMP, while differences in Prevotella spp. between the groups were seen in QMP but not relative profiling [31]. The authors point out that this could contribute to misclassification bias. An alternative to flow cytometry, qPCR has also been used to estimate microbial load with “universal” bacterial primers [32]. Total microbial load was multiplied by relative abundance (from 16S sequencing) to determine absolute abundance and validated by comparison with separate qPCR quantification of 4 unique taxa (each a different rank: phylum, family, genus, and species) [32].

QMP is not without challenges. In fecal samples gathered from 16 healthy volunteers, 3 methods in conjunction with 16S sequencing were compared: qPCR, flow cytometry, and flow cytometry followed by treatment of the specimen with propidium monoazide (a photo-activated dye that inhibits amplification of DNA not contained within viable organisms) before sequencing [33]. QMP with the methods using flow cytometry (with or without propidium monoazide) produced reasonable agreement, but stark differences emerged between flow cytometry and qPCR [33]. In addition to these discrepancies, it is not clear what method for sampling or basis for quantification most accurately reflects the microbiome in situ. For QMP to be considered as a clinical diagnostic, a means to standardize input and output would be critical, to enable consistent interpretation of QMP results.

Transcriptomics, Proteomics, and Metabolomics

This review focuses on methods commonly used for detecting and analyzing the microbiome based on analysis of 16S sequences or total DNA. Total RNA (transcriptome), gene products (proteome), metabolic products (metabolome), and other functional microbiota readouts also have major potential as diagnostic biomarkers. Transcriptomic measurements, for example, reveal the expression of genes rather than simply their presence and may reveal active organisms and their associated pathways while omitting bystanders [34]. Coupled metagenome and metatranscriptome analysis in healthy subjects demonstrated the presence of species whose transcriptional abundance was not completely reflected in metagenome analysis [35]. Meanwhile, a longitudinal study in patients with ulcerative colitis and Crohn disease compared with controls without inflammatory bowel disease identified disease-associated patterns attributable to several species, including Bacteroides spp., Faecalibacterium prausinitzii, and Alistipes putredinis, that were reflected in metatranscriptome measurements but not metagenomic measurements [36].

Meanwhile, at the metabolomic level, it has been shown that gut commensals differentially metabolize nutrients to fecal volatile organic compounds (VOCs) [37]; therefore, VOC composition may reflect microbiome composition. In a prospective study of preterm infants, fecal VOCs were able to predict late-onset sepsis, especially sepsis secondary to Escherichia coli and Staphylococcus aureus [38]. Microbial by-products (including VOCs) may be profiled more quickly than the microbiome itself (eg, with 16S or MGS), potentially allowing for prediction and diagnosis on a more clinically actionable time frame. In analogous approaches to translation of discoveries from analysis of NGS data, untargeted proteomic and metabolomic studies will likely dramatically expand the diagnostic utility of microbiome research. In a 2019 study, stool metabolomics was applied to identify metabolite profiles that accurately distinguished patients with Clostridioides difficile diarrhea from those who were colonized or not colonized with C. difficile [39]. Transcriptomic, proteomic and metabolomic measures would require standardization and calibration, but application as clinical diagnostics seem feasible.

OUTLOOK

We are only on the verge of leveraging the information contained within the gut microbiome to advance human health, and tools to reliably produce diagnostic information are essential. The example of qPCR for colorectal cancer detection demonstrates a process in which broad interrogation of the gut microbiome with advanced diagnostic tools can lead the way to more precise hypothesis-driven assays based on accessible methods. Meanwhile, the possibility remains that unbiased MGS coupled with machine learning may lead to predictive analytic tools not reducible to simple assays, such that NGS with advanced computing will be increasingly used in the clinical microbiology laboratory [40]. Ultimately, robust prospective data validating clinical performance of these biomarkers will be needed before they can be applied at scale. Finally, standardization and development of robust reference controls is needed, not only from a methodological perspective, but even when defining the metrics by which microbiomes are evaluated and compared. This is especially crucial in a clinical diagnostic context, where reproducibility is only one aspect to having an accurate and precise test [41]. Another unanswered aspect in diagnostics is what the next intervention may be for a patient with a diagnosis of an “abnormal” gut microbiome, and how this intervention will subsequently be measured.

CONCLUSIONS

The human gut microbiome is generally described in terms of the content and richness of its taxonomic composition and abundance of microorganisms. Tools for making these assessments include traditional culture, qPCR, 16S sequencing, and computationally heavy MGS, and each technique has limitations and benefits. Much work remains to define and standardize a “normal” or “healthy” microbiome. Nevertheless, data supporting the notion that deviation from a normal state has profound implications for both chronic and acute illnesses as well as response to therapeutics—a concept that seems especially important in the context of immune therapy and AR. Analysis of products of transcription, translation and metabolism are already accompanying the study of genomic material and will complement this understanding. From this, diagnostic and predictive tools can emerge in a spectrum of clinical contexts.

Notes

Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The funding agency had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Financial support. This work was supported by the National Institute for Allergy and Infectious Disease (grants UM1AI104681, K23AI144036 to M. H. W.).

Supplement sponsorship. This work is part of a supplement sponsored by the National Institute of Allergy and Infectious Diseases (NIAID), part of the National Institutes of Health (NIH) and the Centers for Disease Control and Prevention (CDC).

Potential conflicts of interest. G. L. D. reports consulting fees from Prenosis, in the last 36 months outside the submitted work; he also has 2 patents pending for biomedical diagnostic technologies. C. S. K. reports personal fees from Rebiotix, outside the submitted work, and she is part of a group holding a patent titled “Systems, Devices, and Methods for Specimen Preparation.” All other authors report no potential conflicts to. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.