Causal role of the gut microbiome in certain human diseases: a narrative review

Establishing the aetiologic contribution of microbiome constituents to clinical outcomes

Establishing a causal relationship between observational microbiome data and health phenotypes shares many of the challenges experienced with determining putative pathogens12 as the cause of disease conditions. For phenotypes that reliably occur in a pragmatic timeframe, approaches conceptually related to Koch’s postulates (either classical, molecular or host dependent) can be applied.13 This is the case with Helicobacter pylori and gastritis, for example, as well as for most established agents of acute bacterial gastroenteritis (eg, Campylobacter jejuni) for which the capacity to cause human disease was assessed against Koch’s postulates of causality in human volunteer studies.12 In this setting, studies to determine the mechanism of an association are considered the most robust form of evidence. This usually requires a suite of methodologies including molecular and animal studies.

The phenotype of interest may rely on the interactions of many organisms or yield an outcome that manifests many years after exposure, beyond a practical timeframe for challenge studies, such as Tropheryma whipplei and Whipple’s disease. In these settings, epidemiologic criteria known as Hill’s criteria14 have been developed to infer causality without direct mechanistic evidence. These are best suited to situations where direct observation from exposure to outcome is unfeasible or unethical. Other situations may be more complex if an outcome occurs long after exposure and only in a subset of hosts. For example, Mycobacterium leprae can cause multibacillary leprosy among people with genetic variants in the NOD2-mediated signalling pathway.15 In such cases, approaches such as Hill’s criteria of causality would be used in combination with studies of bacterial virulence/protective factors, as well as studies of host susceptibility. However, caution is required for the interpretation of epidemiologic association data prior to establishing rigorous evidence of causality.

Many microbiome-associated disease states presumably involve perturbations of many constituents at once, or of another aspect in the interactions of the altered ecosystem with the host rather than stemming from a single species with pathogenic effects. This raises specific challenges to establishing causality. In this review encompassing four representative conditions, we integrated relevant in vitro, in vivo, and observational data to assess the plausibility of the association between the microbiome and the targeted disease states. We then examined RCT data to assess whether microbiome therapeutics can directly impact human health outcomes.

C. difficile infection (CDI)

CDI is a bacterial toxin-mediated infection of the colon causing severe diarrhoea. It is the most common nosocomial infection, with an estimated 500 000 cases diagnosed in the USA yearly.17 Approximately 6% of incident cases result in death and 20% experience a recurrence.17 CDI significantly prolongs hospitalisation and results in a significant economic burden in excess of US$4.8 billion in the USA alone.18 More effective therapies are needed for CDI and the microbiome has been proposed as a therapeutic target in the face of disappointing outcomes with traditional antibiotic approaches. The pathogenesis of CDI is postulated to be ingestion of and subsequent colonisation with C. difficile in conjunction with perturbation in the gut microbiome. Antibiotic exposure is the best-recognised risk factor for C. difficile, through the reduction in microbial competition. Recent evidence also elucidates a role for quorum sensing as a regulator of C. difficile toxin production.19 This involves the production of a signalling molecule that, at a threshold concentration achieved with a relatively high organismal burden, initiates toxin production and mediates disease.19

The gut microbiome of patients with CDI varies over the course of the disease. A study collecting stool samples of non-CDI patients on hospital admission found that the relative abundance of Bacteroidetes and Firmicutes was independently associated with development of CDI.20 Compared with healthy controls, the gut microbiomes of patients with CDI are significantly enriched with Enterococcus and Klebsiella spp, whereas there is relatively less Faecalibacterium, Bacteroides and Bifidobacterium spp.21 22 Recurrent cases of CDI have increased Enterobacteriaceae, as well as increased Veillonella, Parabacteroides, Streptococci and Lachnospiraceae spp23 and lower microbiome diversity than index episodes.

Murine models support the role of the microbiome in CDI. Reeves et al demonstrated that only mice pretreated with antibiotics developed CDI following C. difficile inoculation.24 Another study of mice demonstrated that faecal microbiota transplantation (FMT) significantly reduced the risk of recurrent CDI and eliminated C. difficile colonisation following CDI treatment.25 Sterilisation of the faeces prior to FMT negated this effect.25

In total, 19 RCTs of microbiome therapies for CDI met inclusion criteria6 7 26–42 (figure 2). The first RCT data on the use of FMT in patients with recurrent CDI was published in 201336 (online supplemental table 1). In this study, vancomycin treatment followed by a duodenal infusion of heterologous FMT significantly improved cure without relapse (93.8% vs 30.8%, respectively, p<0.001) and increased stool microbial diversity compared with vancomycin alone.36 In fact, the trial was stopped early for demonstrating benefit.36 Several other trials have further supported and optimised FMT’s benefit in CDI.37–39 41 Oral FMT capsules are non-inferior to colonoscopic infusions in prolonging the freedom from CDI recurrence (96.2% vs 96.2%, respectively, p<0.001 for non-inferiority).39 Frozen faeces are non-inferior to fresh for prevention of recurrent CDI (75.0% vs 70.1%, respectively, p<0.001 for non-inferiority),38 although lyophilised faeces are inferior to fresh (100.0 vs 78.3%, p=0.02).30 Multiple doses of FMT may be superior to a single dose at improving CDI cure (100.0% vs 75.0%, p=0.01).31 Donor-derived faeces are superior to autologous specimens at improving CDI resolution without recurrence (91% vs 63%, respectively, p=0.04).37 Further, CDI clinical resolution was superior with FMT via nasojejunal tube or colonoscopy after a 4–10-day course of vancomycin versus fidaxomicin alone (91.7% vs 41.7%, respectively, p=0.0002).41 Based on such evidence, the Infectious Diseases Society of America recommends FMT in patients who have experienced≥3 episodes of CDI.43 However, caution must be exercised because cases of fatal bacteremia secondary to FMT have been documented.44

Request permission

In response to the safety and logistical challenges of FMT, other microbiome therapeutics are emerging tools for reducing the recurrence of CDI. Various formulations of bacterial spores have been shown to significantly reduce CDI recurrence when administered following a treated episode.6 7 40 42 Two of these therapies have recently obtained FDA licensure following large RCTs. The first agent, VOWST, is an orally administered microbiome agent containing spore-forming Firmicutes. In the ECOSPOR III trial, VOWST significantly reduced the risk of CDI recurrence compared with placebo (12.4% vs 39.8%, p<0.001).7 Stool analyses of VOWST recipients revealed successful engraftment of Firmicutes, higher concentrations of stool bile acids that are known to inhibit C. difficile7 and reduced abundance of antimicrobial resistance genes. The second agent, REBYOTA, is a consortium of bacteria administered via a single enema.45 In the PUNCH CD3 trial, REBYOTA decreased the absolute recurrence of CDI by 13.1% (29.4% vs 42.5%, probability of superiority 99.1%) compared with placebo.40 However, the benefits of these agents may be overestimated because not all patients received fidaxomicin for the index episode of CDI, nor did they receive a pulse and taper regimen.

Obesity

Obesity is prevalent and elicits significant disability and mortality. The pathogenesis of obesity is complex and involves an interaction of genetics, environmental and lifestyle factors. Recently, a role of the microbiome has been elucidated.

The relationship between composition of the microbiome and obesity is complex with many conflicting studies, although an elevated Firmicutes to Bacteroidetes ratio in obese persons appears to be the most robust correlation.46 Early life antibiotic exposure has an independent and dose-dependent association with obesity.47 However, this may be due to residual confounding, as this association was not significant in sensitivity analyses using twin and sibling-matched data.47 Further, a large cohort study demonstrated that decreased microbiome diversity and butyrate-producing bacteria are associated with insulin resistance and type 2 diabetes.48

When normal adult mice and germ-free mice (GFM) are fed a western diet, GFM gains significantly less weight.49 GFM colonised with the microbiota of normal mice develop increased adipose tissue and insulin resistance, without a corresponding change in weight, despite decreased chow consumption compared with normal mice.49 Ridaura et al colonised GFM with the stool of human twins discordant for obesity and found that colonisation with the obese twins’ stool led to significantly more weight gain than the lean twins’ stool.50 Cohousing a GFM colonised by a lean donor’s stool with its twin colonised by an obese donor’s stool was protective against development of an obese phenotype.50 Another study found that early exposure of normal mouse pups to penicillin resulted in significantly increased weight and adipose tissue.51 These gains were also transferable to GFM by FMT, suggesting that the effect was microbiome-mediated.51 Moreover, an association between an elevated Firmicutes to Bacteroidetes ratio and obesity has also been reported in mice.52

Despite encouraging in vivo results, RCT results of microbiome therapeutics for the treatment of obesity have been less promising. None of the seven RCTs on microbiome therapies for obesity identified by our search demonstrated any weight or BMI benefit.53–59 Further, three small RCTs found that FMT did not significantly improve insulin resistance compared with placebo.54 59 60 A larger RCT comparing lean donor FMT to placebo in obese adolescents did not find a significant difference in weight or other markers of metabolic syndrome57 (online supplemental table 1). Similarly, a trial comparing FMT vs placebo in patients undergoing bariatric surgery did not demonstrate any weight loss benefit.56 Another RCT using a live or pasteurised Akkermansia muciniphila supplement compared to placebo in people with obesity/overweight did not find any statistically significant change in anthropometric parameters, but did find increased insulin sensitivity in the pasteurised supplement recipients compared with placebo (p=0.002).53 Overall, these findings suggest that while the microbiome may be disrupted in obesity, our current knowledge of microbiome manipulation alone may be insufficient to modulate a person’s metabolic profile to result in sustained weight changes.

Cancer immunotherapy

Immunotherapy is a novel approach to cancer treatment that primes the immune system to eliminate malignant cells. The most common immunotherapy agents are checkpoint inhibitors, which either interfere with the PD-1 and PDL-1 interaction between tumour cells and T cells or CTLA-4 and B7 between antigen presenting cells and T cells.61 Immunotherapy has transformed the therapeutic landscape of various malignancies. However, a significant proportion of tumours are refractory to immunotherapy. This resistance to immunotherapy has generated significant interest in adjunctive treatments to augment response.

Recent evidence suggests an association between the gut microbiome and immunotherapy response. Several studies have reported poorer immunotherapy responses in recipients of antibiotics even when controlling for disease severity and comorbidities.62 63 Among renal cell carcinoma (RCC) and NSCLC patients, the abundance of A. muciniphila is associated with an improved immunotherapy response.63 64 Among a cohort of patients with metastatic melanoma treated with immunotherapy, response to treatment was positively correlated with the gut microbiome’s abundance of Bifidobacterium longum, Collinsella aerofaciens and Enterococcus faecium.65 Another study of metastatic melanoma patients found that faecal microbiome diversity and the abundance of Ruminococcaceae spp were elevated among immunotherapy responders.66 While these differences are hypothesis generating, no specific microbiome signature predicting immunotherapy has been recognised and these studies are likely to subject to significant residual confounding, despite adjustments.

The link between immunotherapy response and the microbiome is further supported by murine models. Sivan et al demonstrated that in a melanoma mouse model, FMT from one mouse population to another in conjunction with immunotherapy significantly slowed melanoma growth compared with immunotherapy alone.67 Mice receiving FMT also had evidence of increased immune activation and tumour-specific T cells, which correlated with the proportion of Bifidobacterium spp in the gut microbiome.67 Another study found that antibiotics diminished the effect of immunotherapy in sarcoma and melanoma mouse models.63 When the sarcoma mouse model was colonised with the stool of human sarcoma patients and treated with immunotherapy, response to therapy correlated with the stool donor’s response to immunotherapy.63 Several mechanisms have been proposed to explain these findings, including antigen mimicry, microbial metabolites and dendritic cell activation.

Two RCTs of immunotherapy with adjunctive microbiome therapy met our inclusion criteria.11 68 Dizman et al conducted an RCT of an adjunctive Bifidobacterium spp supplement (CBM588) in patients with metastatic RCC receiving ipilimumab and nivolumab11 (online supplemental table 1). Although there was no significant increase in the abundance of Bifidobacterium spp in the CBM588 group, they had improved progression-free survival compared with controls (HR 0.15, p=0.001). A second RCT by the same group boosting cabozantinib and nivolumab with CBM588 in metastatic RCC demonstrated an improved objective response rate among recipients of CBM588 compared with controls (p=0.01), without any significant change in microbiome composition.68 Interpretation of these compelling results is limited by the open-label design, lack of placebo control, single centre location, small study population (n=29–30) and no mechanism linking the bacteria to any surrogate markers of response. Further RCTs are underway to determine the efficacy of these agents and to further investigate their mechanisms of action (NCT04264975, NCT03934827, NCT0382911, NCT05038150, NCT5103345 and NCT05122546).

Ulcerative colitis (UC)

UC involves chronic inflammation of the colon; its prevalence in the USA is 0.3% and is rising.69 UC precipitates significant mortality and morbidity associated with frequent hospitalisations and surgeries. The pathogenesis of UC is postulated to implicate a complex interaction between host immunity, barrier dysfunction, genetics and dysbiosis.70

Multiple lines of evidence in humans support a role for the microbiome in UC. First, antibiotic exposure is associated with an increased risk of UC71 and, paradoxically, antibiotics have also shown some activity in the treatment of UC.72 Second, the microbiomes of UC patients are significantly less diverse73 and the composition fluctuates more over time compared with healthy controls.74 Third, one study found that the microbiome of UC patients contained less Bacteroidetes and Lachnospiraceae, but more Actinobacteria and Proteobacteria.75 Another study found that UC remission was significantly associated with gut faecal microbiome diversity and the abundance of A. muciniphila.76 However, data on which bacteria are enriched in UC are heterogeneous.77 Fourth, the composition of the gut microbiome of UC correlates with disease activity.78

Various inflammatory bowel disease (IBD) mouse models have been used to investigate the role of the gut microbiome in UC. Colonisation of an IBD mouse model with stool samples from human patients with IBD, but not healthy controls, induced histologic evidence of colitis.79 Colonisation with healthy murine stool resulted in clinical improvement and reduced inflammatory markers.80 The colonisation of an IBD mouse model with Faecalibacterium prausnitzii attenuated pathologic evidence of colitis and corrected gut dysbiosis compared with controls.81

Several effective therapies exist for the treatment of UC; however, many cases are refractory and require prolonged courses of corticosteroids. Microbiome agents are a potential steroid-sparing non-immunosuppressive strategy (online supplemental table 1). Our search revealed 18 RCTs of microbiome therapies in UC.82–99 One RCT found that FMT administered by enema increased stool microbial diversity and rates of UC remission (23.7% vs 5.4%, p=0.03) compared with a sham control.90 Similarly, FMT administered by colonoscopy significantly improved steroid-free clinical remission and endoscopic response compared with placebo (26.8% vs 7.5%, respectively, p=0.02).92 Further, donor anaerobically cultured FMT may be superior to autologous aerobically cultured FMT at inducing remission in UC.84 Conversely, another RCT found no significant difference in UC remission between anaerobically cultured allogenic FMT with strict donor selection criteria versus autologous FMT.83 A small RCT demonstrated that oral FMT is efficacious as well.86

Limitations

This narrative review is constrained by certain limitations. Because of the wide scope of the topic, a systematic review of the in vitro, in vivo and observational studies was not performed; therefore, it is possible that relevant articles were omitted. The additional limitations of the review are predominantly inherent to the studies discussed.

The observational evidence supporting the link between the microbiome and human health outcomes is subject to several limitations. First, there is significant normal variation in the human gut microbiome associated with location, diet, age, comorbidities and medications.2 These factors were frequently unaddressed and a rationale for the selection of control patients was seldom provided. Second, reverse causality is difficult to exclude as a possibility. Third, this review, and most of the evidence based on the microbiome, focuses on its bacterial composition, which likely underappreciates the role of non-bacterial organisms and their biospatial localization.

Mouse models provide controlled conditions where the user can easily manipulate the microbiome. However, there is significant variability in microbiome composition between different mouse models and vastly different microbiome profiles in humans compared with mice. These mouse models are also often imperfect proxies for human diseases, such as in certain IBD models, where the mice have only pathologic evidence of intestinal inflammation due to an external stimulus.100 Mouse behaviours, such as coprophagia, further confound their applicability as microbiome models.

While phase III RCTs provide the highest level of evidence between microbiome manipulation and human health outcomes, certain limitations persist. Mechanistic data is lacking on microbiome therapeutics. It remains unclear whether the efficacy of these therapeutics is due to a by-product of the bacteria, the influence of the agent on the composition of the microbiome, or some combination of factors. For example, CBM588 did not significantly alter the composition of the recipients’ microbiomes, but significantly improved progression-free survival.11 Nonetheless, the emergence of RCT data in separate patient populations across distinct geographic areas argues for a causal role of the microbiome in human health in certain conditions.

Future directions

There remains a plethora of unanswered questions on the gut microbiome that should be addressed by future studies. First, there is a lack of data on the optimal formulation (eg, lyophilised, fresh, frozen), route of administration (eg, oral, duodenal infusion, rectal) and dosing of microbiome therapeutics. Further, whether administering antibiotics prior to administration of a microbiome agent improves efficacy is unknown. Second, current RCTs typically administer a single course of the microbiome agent and have limited follow-up periods. Long-term data are necessary to follow microbial engraftment, to determine the duration of effect, and whether longer or additional courses of treatment are beneficial. Third, long-term safety data are also required, especially considering that these therapeutics can induce lasting alterations to the microbiome. Fourth, it is unclear whether diet, probiotics and other therapies interfere with the efficacy of microbiome therapeutics. Fifth, refractoriness to microbiome therapeutics is poorly understood.7 Sixth, direct comparisons between specified microbiome formulations versus FMT are lacking. Seventh, it is unknown whether all donor FMT are equivalent or if certain donor characteristics are favourable. Eighth, regulatory oversight of good manufacturing practices specific to microbiome therapeutics is needed to ensure the ongoing safety and reproducibility of these agents.