Towards understanding the link between gut microbiota and heart failure in the heart–gut axis

This editorial refers to ‘Causal relationships between the gut microbiome, blood lipids, and heart failure: a Mendelian randomization analysis’, by H. Dai et al., https://doi.org/10.1093/eurjpc/zwad171. Heart failure is a multisystem disease that has been recognized and documented in ancient texts for at least several millenia.1 The interactions between the heart and the gut (heart–gut axis) and the pathophysiology of heart failure have long been considered to be primarily haemodynamic and circulatory—comprising the pump, gastrointestinal venous reservoir, and mucosal barrier2—until about a decade ago when microbial metabolism from dietary nutrients such as trimethylamine N-oxide (TMAO) was mechanistically linked to heart failure disease progression.3,4 These findings reinvigorated interests in the scientific community to conceptually connect nutrition, gut microbiota, and cardiovascular disease through circulating metabolites. Aside from TMAO that may alter intestinal cholesterol transport mechanisms,5 the interrelationship between the gut microbiota and lipoproteins/cholesterol remains poorly understood, particularly in cardiovascular disease and heart failure. As a major contributor to host metabolism, the gut microbiota are a major player of nutrient release6 and modulator of lipids and lipid species, including short-chain fatty acids.7 In this issue of The Journal, Dai et al. used bi-directional Mendelian randomization to explore and evaluate the causal relationships between heart failure and gut microbiome or blood lipids through the use of single nucleotide polymorphisms as genetic instrumental variables and identified intriguing associations between an abundant species of the Bacteroides genus in the gut, Bacteroides dorei, and heart failure via apolipoprotein B (ApoB).8B. dorei has been previously implicated in the protection against atherosclerosis in comparative analysis of human patients with and without coronary artery disease; through mouse studies, its was observed that gut microbial lipopolysaccharide production was reduced.9 On the other hand, B. dorei has also been demonstrated to be a contributing factor in obesity via branched-chain amino acids and brown adipose tissue metabolism,10 in gestational diabetes, and in type 1 diabetes.11 In the study of Dai et al., B. dorei was associated with increased risk for heart failure mediated via ApoB by an estimated 10.1% and appeared to be independent of coronary artery disease.8 While the pathophysiology of heart failure and coronary artery disease may overlap in ischaemic cardiomyopathy, the divergent effects of B. dorei in the two diseases and the underlying mechanisms remained unexplained.8,9 To accept the findings and implications of this in silico analysis, we should first review the methodology that has been popularized over the past decade with the availability of these large genomic datasets. Mendelian randomization has gained significant interest as a method to assess causal effects by using genetic variations to investigate the impact of a modifiable exposure on disease in observational studies. However, this approach relies on three key assumptions: (i) the genetic markers should be associated with the exposure and primarily affect the outcome through the exposure; (ii) the genetic markers should be independent of the outcome given the exposure and all confounding factors; and (iii) the genetic markers should not influence the outcome through pathways other than the exposure. The first assumption may perhaps be the least expected especially when leveraging carefully curated datasets—when the exposure and outcomes of interest may not be tightly associated with the genetic markers to begin with. The use of large-scale data sets by Dai et al.8 including gut metagenome from the Dutch Microbiome Project, blood metabolome from the UK Biobank, and a collection of over 110 000 heart failure cases from seven multi-ancestry consortia/biobanks can be informative and should have adequate power. However, complications arise when genetic associations extend beyond the human host, particularly when trying to identify human host genes that may impact microbial species levels (as opposed to metagenomic identification of microbial species). In fact, many loci identified through human genome-wide association studies (GWAS) of microbial traits over the years have been deemed false positives, even when showing significant associations in their initial reports. This could explain the lack of reproducibility in most findings in the literature, with the exception of lactase and ABO loci that have been reproduced through rigorous validation.12 Due to this lack of reproducibility, leveraging independent cohorts (i.e. validation datasets completely different from learning datasets) to demonstrate consistent interactions among genetically mediated human and microbial traits is therefore necessary and should no longer be optional to draw reliable conclusions. The second assumption is more obvious but often overlooked in bioinformatic analyses when many of the identified human loci may exhibit associations with multiple traits (i.e. pleiotropic effects)—many independent of microbial influence.13 GWAS analyses often do not provide information on whether these effects are concordant or discordant across traits. Also, the authors may not have considered how these identified loci have pleiotropic effects on biological pathways that may lead to heart failure risk independent of their associated microbial traits. Microbial ecosystems are highly complex, and their metabolic output may be dependent on interactions across various species above and beyond their abundance and their interactions with the human host—something that is difficult to account for and may not have strong genetic influences. The third assumption may also be challenged when the human host genetic influence on microbial composition and function is often modest compared to environmental factors such as age, diet, and comorbidities. There is ample evidence supporting the role of ApoB in cardiovascular risk prediction, partly due to its presence in atherogenic particles.14 Interestingly, even the authors’ own mediation analysis indicated a stronger causal effect of ApoB on heart failure than B. dorei. Even if the effects of ApoB on B. dorei-mediated heart failure risk were true, the true clinical significance of the findings regarding the impact of ApoB on microbial-mediated heart failure risk is largely unclear. In fact, reverse causality may still hold true despite statistical support of directionality. While it is very tempting to make claims on casual relationships of novel disease mechanisms with sophisticated genetic analytical techniques, it is equally important to bear in mind the assumptions of Mendelian randomization when using these powerful tools as well as critically evaluate the quality and scientific rigor before championing the results of that study.8 While Mendelian randomization provide a unique tool to understand causality, potential concerns regarding potential violation of the aforementioned assumptions when exploring genetically mediated interactions between microbes and human traits call for the need for greater scientific rigour. Independent cohort validation is therefore essential to confirm these hypothesis-generating findings and ensure reproducibility, which cannot be replaced by sensitivity analyses or different statistical models. Meanwhile, many reports emphasise the necessity of in vivo demonstrations of host–microbial interactions in further investigations; yet, numerous studies are published without such validation, contributing to the lack of confidence in the findings. The findings from Dai et al. are provocative, hypothesis generating, but warrant replication and in vivo and functional studies to validate as we have learned from the century-old Koch’s postulates15 that should be modified but still holds true in the genomic era.