R2.1. Techniques and mouse models

Other commentaries focus more closely on specific neuroscientific techniques that could potentially contribute to MGB insights. Aarts & El Aidy suggest that brain imaging in the form of functional magnetic resonance imaging (fMRI) may increase the depth of neuroscientific knowledge in microbiome-related investigations. They cite a handful of studies that have done just that. Even though there are limitations to what fMRIs reveal (Klein Reference Klein2010; Logothetis Reference Logothetis2008), we agree that brain-imaging methods have the potential to provide more detail about the activation of brain structures in relationship to microbiome conditions. Other visualization techniques (not just fMRIs) might also reveal associations between visualized brain states and microbiome composition; again, they may not (e.g., Carlson et al. Reference Carlson, Xia, Azcarate-Peril, Goldman, Ahn, Styner, Thompson, Geng, Gilmore and Knickmeyer2018). Although everyone has his or her favourite technique, it is not always the case when a field is new and still developing that “more is better” all the time. Moreover, it is not at all clear that additional types of data are what are needed to push microbiome-brain findings over the causal threshold.

Alberts et al. emphasize the developmental importance of parent-offspring interactions for gut microbiome colonization and make a plea for more careful behavioural approaches. They are disappointed that we believe there are limits to mouse models of human disorders, which they think is not justified because of the “evolutionary conservation of core mechanisms” (para. 7). The problem we see is that there is little evidence that these claims are true for many aspects of behaviour. Assuming shared evolved mechanisms for human and rodent behaviour is part of the methodological and translational problem indicated by our target article. Alberts et al. go on to argue that “sophisticated experimental” use of germ-free animals will advance the field and allow mechanisms to be postulated and therapies to be discovered. They suggest the answers lie in “methods that are sensitive, objective, and sufficiently nuanced to capture social behaviour” (para. 3).

In contrast, we believe that even optimal methods for measuring social behaviour in mice are just as likely to provide evidence for species-specific behavioural patterns, rather than conserved core behaviours relevant to all mammals. These species-specific behaviours will probably be much harder to compare to human behaviour in general, let alone mental disorders, than the study of one or two parameters in rodents that are supposed to reflect one symptom of a human condition. In addressing this issue of human-mice comparability, Kelsey & Grossmann argue that we should have paid more attention to “humanized mouse models” (para. 4), in which human microbiota are transferred into mice. Possibly, but there are also well-known problems in transplanting human microbiota into mice (see Arrieta et al. Reference Arrieta, Walter and Finlay2016; Nguyen et al. Reference Nguyen, Vieira-Silva, Liston and Raes2015). These problems greatly exacerbate the basic translational obstacles we noted in our target article; Birk also points to these challenges. Mouse-human translation becomes even more troublesome when treatments derived from mice studies are applied to humans. A recent study (also noted by Korte & Korte) demonstrates how probiotics often fail to colonize mice intestines, whereas humans may have far more individualized colonization responses and hence different modes of potential host impact (Zmora et al. Reference Zmora, Zilberman-Schapira, Suez, Mor, Dori-Bachash, Bashiardes, Kotler, Zur, Regev-Lehavi, Brik, Federici, Cohen, Linevsky, Rothschild, Moor, Ben-Moshe, Harmelin, Itzkovitz, Maharshak, Shibolet, Shapiro, Pevsner-Fischer, Sharon, Halpern, Segal and Elinav2018).

R2.2. Therapies

The greatest treatment hopes of MGB and some broader microbiome research are probiotics and dietary interventions. Kelsey & Grossmann think the probiotic aspect of MGB research is “somewhat overemphasized” (para. 5). We concur, on the grounds that these are not straightforward therapies, as our target article discussed. Korte & Korte agree and go on to urge more careful hypotheses about how probiotics might affect host physiology. They encourage attention to new research that reveals a “personalized” response to probiotic therapy (Zmora et al. Reference Zmora, Zilberman-Schapira, Suez, Mor, Dori-Bachash, Bashiardes, Kotler, Zur, Regev-Lehavi, Brik, Federici, Cohen, Linevsky, Rothschild, Moor, Ben-Moshe, Harmelin, Itzkovitz, Maharshak, Shibolet, Shapiro, Pevsner-Fischer, Sharon, Halpern, Segal and Elinav2018). Therapeutically, this means that different strains of probiotic bacteria (i.e., below the species level) may have to be used on very specific populations of people. A personalized approach also means that different probiotic organisms may help different conditions, and that detailed diagnosis and treatment plans would have to accompany any therapy. Doing so would undermine expectations of generic probiotics able to provide a fast universal fix of conditions ranging from anxiety and Alzheimer's to schizophrenia and Tourette's syndrome. The scientific detail required for personalized probiotics undermines the DIY appeal of such therapies. More personalized probiotic treatments might also not be so attractive to commercialization (see the Supplementary Material that accompanies our target article) because of decreased consumer bases.

Expanding the treatment theme further, Dalile, Van Oudenhove, Verbeke, & Vervliet (Dalile et al.) note that our target article ignored research on prebiotics, which are foods believed to favour “good” microbes in the gut; prebiotics may also refer to any dietary fibres that are primarily broken down by microorganisms. It is true we do not even mention prebiotics in our target article, but that is simply because the 25 most cited MGB articles paid them no attention. Dalile et al. argue that prebiotics are therapeutically better than probiotics, and to substantiate their claim, they discuss a systematic review of prebiotic-microbiome studies. Although it is always useful to draw on comprehensive analyses of a topic, we are concerned that many of the problems present in probiotic research would manifest themselves in prebiotic and other food-based studies. Dalile et al. suggest that prebiotics are good therapy because they increase the numbers of certain bacteria (or their fermentation products) that correlate with neurobehavioral benefits. However, it is not obvious to us how universally “good” any strain of bacteria can be for mental health; plus, dietary confounding is a recognized problem in microbiome research. In a nutshell, we do not think there is sufficient evidence of sufficient quality to warrant Dalile et al.’s conclusion that “prebiotics have a higher potential than probiotics [for] causal claims” (para. 6) about microbiomes, brains, and behaviours.

To add to the complications of microbiome-oriented therapies, Borghi, Vignoli, & D'Agostino (Borghi et al.) note that poor nutritional and feeding habits are common in patients with mental illness. Their commentary then provides more data showing correlations between diet and mental health disorders. We think they may be implying that altered diet can be a common cause that leads both to host disorders and “reduced diversity microbiota.” This is by no means the same as saying the microbiota is the causal agent; instead, it could be merely a causally unrelated correlate of disease. We address the issue of causality in more detail in Theme 2 but agree that however microbiome causation of disease is understood, it must be investigated in the context of diet. Most human microbiome research does this already but does not always carefully analyse the causal structure of the relationships between diet, microbiome composition, and health (Lynch et al. Reference Lynch, Parke and O'Malley2019). Other commentators (Valles-Colomer, Falony, Vieira-Silva, & Raes [Valles-Colomer et al.]; Aarts & El Aidy) also point to additional confounders in microbiome studies such as intestinal transit time, inflammation, and medication. More attention to causal structure (i.e., common causes, confounders, causal order, and causal interactions) would certainly help with the evaluation of causal hypotheses in MGB and other microbiome research.

R3.1. Causal explanation

As we emphasized in our target article, a major issue in all microbiome research, and particularly MGB research, is how to detect and evaluate causal relationships. When seeking ways forward, it can be useful to turn to philosophy of science and see what it has to say. Korte & Korte suggest that “the field of MGB research would greatly improve when more attempts are made to falsify the hypothesis (Popper Reference Popper and Popper1963)” (para. 1). A flood of philosophical ink has been spilled over the last few decades on examining Popper's rather appealing characterization of good science. The general conclusion of this massive literature is that Popper's abstract methodology of falsification is an ideal of good science that simply does not really reflect how science works; even Popper eventually reached this conclusion (Thornton Reference Thornton and Zalta.2018). This is especially the case in today's life sciences, which value positive findings and may not always be driven by hypotheses at every stage of inquiry. Nevertheless, without worrying too much about the truth or falsity of “falsificationism,” it is good general practice to try to formulate and test hypotheses about causal processes. For us, the question is whether broad reference to “microbiomes” (compositional or functional) will really allow causal hypotheses to be tested and clear conclusions drawn.

This key topic emerges in a subset of commentaries that examine whether microbiomes as a whole are the causal agents of brain-behavioural disease and health states, or whether specific components of the microbiome are what should be understood causally. In tackling this topic, Lynch addresses whether whole microbiome explanations will succeed, or whether explanations involving particular microbes (or small groups of different microbes) are more likely to be successful in accounting for host phenomena. Using the criterion of “proportionality” (found in explanations in which causes are proportionate to the effect), she suggests smaller-scale causality is the way forward for MGB. However, Lynch also notes how microbiome research (or its media uptake) is particularly attached to much broader and vaguer “microbiome” explanations. This is rather odd in light of a historical view of genomics, she observes, which has always sought proportionate gene-based explanations rather than disproportionate whole-genome ones. Some sociology of science may eventually illuminate this curious trend in microbiome research.

Clark largely echoes Lynch’s advice, pointing specifically to neurotropic enterovirus research as “demonstrating causal relationships between gut-brain-axis status and infected host health, development, behavior, and mind” (para. 2). Microbiome researchers used once to focus exclusively on the bacterial components of gut microbiota, but these days, they are increasingly concerned with the viral components (or “virome”) of gut microbial communities. A “virome” focus seeks to understand not only all of the viral groups in the gut, but also their interactions with prokaryotic and eukaryotic microbes in relation to host health (Pfeiffer & Virgin Reference Pfeiffer and Virgin2016). These interactions are barely understood at present, largely because experimental exploration of the virome is limited. Although work on neurotropic enteroviruses precedes microbiome research by many years and is limited to a very specific range of viruses, it is an example of how insights into causal relationships are facilitated by having a more specific experimental focus, be that organismal, viral, or molecular.

Fetissov too argues along these lines when he proposes that microbiota effects on behaviour are potentially related to the generation of antibodies against gut microbial components, which then recognize mammalian peptide signalling molecules. Although any particular claims about causal entities and mechanistic pathways might be questioned, homing in on specific relationships such as a recognized group of organisms, molecules, or viruses is likely to be more experimentally tractable and generate more precise findings. Conversely, after proposing neurotropic viruses as causal agents of disease, Clark dwells on “cosmopolitan virus-microbe relations” (para. 2), which are again broad and unclear. We suggest that as soon as the research focus moves to this more general and unspecified level, problems with attributing causality are likely. Alberts et al. think researchers can have it both ways: They can consider “the microbial community in its entirety” even while looking for “missing or over-represented” microorganisms that just might be implicated in host phenotypes (para. 5). This is reasonable advice, but being able to do the latter already relies on decomposing the “entire community” and knowing what each individual microbe does to the host. MGB and other microbiome research is not able to do this yet except for a small selection of microorganisms.

In a similar vein, Johnstone & Cohen Kadosh believe that developmental cognitive neuroscience (DCN) is well equipped for “establishing causal relationships between dysbiosis and mental health problems” (para. 5). They see DCN as multilevel and multifactorial in its explanations, and argue that this allows the specification of mechanisms and the identification of causal relationships from correlational ones. Although strategies to improve the scientific understanding of complex phenomena are essential, DCN is probably not alone in having the potential to provide a helpful perspective. Psychoneuroimmunology, for example, is another field that may be operating in a similar spirit (e.g., Konsman et al. Reference Konsman, Parnet and Dantzer2002). Although Johnstone & Cohen Kadosh's short commentary is not detailed enough to lay out how such causal explanations have been achieved in DCN, we doubt that any of its putative explanatory success depends on positing large-scale complex entities without decomposing them and experimenting with individual components one by one. After that, of course, these components can be recomposed into larger networks of causal influence, but the initial breakdown is crucial for detecting experimental effects. We fully acknowledge Borghi et al.’s emphasis on the complexity of interactions between the gut, brain, and microbiota but think that a more traditional emphasis on the experimental targeting of proportionate and specific causes will make steps in the right direction.

R4.1. Bibliometric issues

Ostinelli, Gambini, & D'Agostino (Ostinelli et al.) find that our bibliometric methods are deficient, particularly because at least one highly cited MGB paper was missed in our original analysis. To them, this indicates “cherry-picking.” Moreover, they argue, high citations are not good sources of evidence because the former can occur for bad reasons (provocation, etc.). They believe we did not distinguish adequately between article types or citation sources. They also suggest we should have included “systematic reviews” and not excluded the large numbers of MGB reviews our literature analysis found.

Borghi et al. also say that a focus on top-cited papers is bad methodology. Moreover, recent MGB literature has better microbiome analyses. They say generally that MGB is a new field and that it is just a matter of time before the research situation improves. Wissel & Smith echo this position when they say that looking at top-cited papers is biased in favour of outdated ones. Clark also notes that our perspective is based on “25 narrowly chosen literature examples” (para. 2), and he prefers to pay attention to some less cited material that is able to demonstrate causal relationships (see Theme 2).

These comments require us to reiterate the “snapshot” methodology of our target article. As we stated in our target article, we selected these 25 papers because of their centrality to the established experimental core of MGB. Their “most cited” status indicates these papers have influenced the field, in terms of their findings, methodologies, and conceptual machinery. By focusing on them, we hoped to delve more deeply into an influential corpus of work than we could have if we had conducted only a shallow survey of the entire body of literature. It did not escape our attention that a focus on “top-cited” papers would mean leaving aside more recent and potentially improved work in the field. We do not believe that citations are the apotheosis of bibliometric analysis, but citations do mean researchers are paying attention and often gaining inspiration and structure from these influential papers.

We also made very clear that we were not carrying out a systematic review, and even that we thought it was probably too early to do this for MGB, which is not only fairly youthful in its career, but also quite diverse. It will be difficult to compare effectively its different methods, data, and hypotheses not only now, but also for the field as a whole in the future. However, we take this opportunity to remind readers of systematic reviews and meta-analyses of microbiome and obesity correlations and experiments, which showed that accumulated findings were inconsistent, whether concerned with bioinformatic patterns or experimental effects (Duvallet et al. Reference Duvallet, Gibbons, Gurry, Irizarry and Alm2017; Sze & Schloss Reference Sze and Schloss2016). MGB research is concerned with many more host phenomena than obesity, but we suspect that sub-categories of research (e.g., depression-related MGB) might find themselves in the same position as obesity when systematic reviews and meta-analyses are eventually conducted.

We further note that the highly cited paper Ostinelli et al. say our methodology missed (Berer et al. Reference Berer, Mues, Koutrolos, Rasbi, Boziki, Johner, Wekerle and Krishnamoorthy2011) did not fit our specifications. We searched for highly cited articles linking microbiota/microbiomes and brain-behaviour phenomena. The “missed” article is about connections between the microbiota and multiple sclerosis, which is a neurological disorder rather than a behavioural or mental state. From our perspective, it is therefore not fully relevant to the body of work we loosely refer to as MGB. In other words, although we do not claim that our bibliometric exploration of the field is perfect, this single example of an important but omitted article is not compelling. However, perhaps this means that Ostinelli et al. have a different conception of what counts as MGB and think it should include basic neurological disorders. We agree we did not make this exclusion as clear as it should have been in the target article.

Clark makes a general point about whether top-cited articles can accurately depict trends in a field. His solution is to turn to meta-analyses, particularly to capture less cited but more robust findings. As we said in our target article and again in this response, there are major issues of when it is timely in the emergence of a field to do such studies. One consequence of thinking it is too early for systematic reviews and meta-analyses might be that MGB research is not yet a fully formed field. There are no universal definitions of what constitutes fields of research, but it is probably unobjectionable to describe them as loose collective research endeavours that use different tools to illuminate shared but fairly general research questions. Most clearly, perhaps, a research field tends not to be described as an established discipline and hence is free to draw on many different disciplinary tools and perspectives. This proclivity is both a strength and (sometimes) a weakness of MGB research.

R4.2. Translation

Translational issues abound in MGB research, where they range from mouse-human translation problems to how scientific findings are translated across disciplines and then into media reports, everyday talk, and health regimes (all discussed in our target article). Issues of multidisciplinarity lie on the “inward-facing” end of translation, which is concerned with overcoming internal barriers to scientific success. Put another way, translation needs to occur between disciplines for MGB to work. Andreoletti & Rescigno raise the possibility that the issues discussed in our target article are “typical of an emerging multidisciplinary field” (para. 1). They recommend seeking advice from experts to improve matters. To us, this recommendation falls a bit short with regard to the practical demands and everyday difficulties of multidisciplinary research. Aarts & El Aidy have narrower but more concrete suggestions that revolve around trial registration and preregistration. They believe that having formal requirements for clear hypotheses in advance of inquiry will help the field develop more rigorously as disciplines continue to interact and make potential MGB findings.

On the outward-facing (or publicly oriented) end of the translation spectrum, Gligorov speculates as to whether or not commonsense conceptions of health are revised by scientific discoveries. She distinguishes between easily revised common sense (beliefs that share similarities with scientific findings) and more intransigent common sense (beliefs that are in opposition to scientific findings). We are sure it will be worth doing some solid research on this topic but want to go back to the single mention of common sense in our target article. There, we discussed the advantage sometimes taken by scientists of commonsense beliefs.

Numerous MGB articles, reviews, and popular press pieces (as well as one of our commentaries) play on the phrase “gut feelings.” Some elaborate why using this phrase is an appropriate vindication of ordinary thinking. For example, one MGB review suggests that “in daily life, we use common phrases such as ‘butterflies in my stomach’ or ‘I have a gut feeling’ to articulate our intuitions, the first warning signs when something is wrong. This concept that the gut can reveal, or even predict, our thoughts and feelings is one that is ingrained in our psyche and is now gaining the attention it deserves in the scientific literature” (Cowan et al. Reference Cowan, Hoban, Ventura-Silva, Dinan, Clarke and Cryan2018, p. 1). They go on to say that such intuitions have something to do with the gut-brain-axis, which now includes the microbiota (hence MGB).

We fully understand how tempting catchphrases can be and why connections between “folk” uses of language might be made in scientific research areas. But there is seldom any straightforward relationship between folk use and scientific terminology, as detailed investigations of folk and scientific concepts such as “species,” “innateness,” “free will,” and “causation” have shown (e.g., Griffiths Reference Griffiths2002; Hunn Reference Hunn1976; Nichols Reference Nichols2004; Norton Reference Norton2003). Even without thinking too hard, it should be obvious that whatever microorganisms are doing in our gut, these activities simply will not translate into “thoughts and feelings” in any comprehensible sense. Our main reason for urging caution with such appeals to commonsense terminology is because of the tendency to “oversell” microbiome research in general (Eisen Reference Eisen2017) and MGB in particular (as outlined in our target article). One of the temptations to oversell can come from appeals to commonsense notions, when these verbal vehicles may already come with strong and potentially incompatible perspectives and may simply not capture at all the phenomena the science is investigating.

For example, Wissel & Smith mention findings about “the microbiome of romantic couples” (para. 3) when they cite a recently published study (Dill-McFarland et al. Reference Dill-McFarland, Tang, Kemis, Kerby, Chen, Palloni, Sorenson, Rey and Herd2019). It finds that married couples reporting “closeness” have more similar microbiomes than less close spouses, non-spouses, and even siblings. This article explains these correlations as attributable to sustained marital intimacies that presumably transfer microbes from one spouse to the other. Caregiving relationships that involve personal hygiene probably achieve similar effects. But by reporting this study with the label “romantic” (never mentioned in the cited paper), and then immediately invoking a microbiome-neurotransmitter correlation from another paper, Wissel & Smith might be thought to offer up these findings to the very sensationalism they criticize. We expect to read headlines any day now saying, “Microbiomes responsible for romance!”Footnote ¹ One of the points in our target article was that scientists too have to take responsibility for their language and interpretations, and talking about “romance” rather than close cohabitation effects on microbiome composition is not going to help the field's relationship with media.

Clark partly defends the use of some catchy phrases saying that “many playful, flashy, even personifying interpretations and labels, now common to modern-day multimedia presentation styles, enrich the public's imagery of difficult-to-learn science concepts” (para. 3). Although we think this can be justified in appropriate contexts, there is also a very fine line between finding good metaphors to communicate science to non-specialists and over-simplification accompanied by false promises. Borghi et al. notice “pressure towards public engagement for professionals without specific training in science communication” (para. 1). There is a whole academic discipline of science communication out there – in most universities, in fact – and working with this discipline might be valuable for MGB research, scientific groups in general, and the future of science.

R4.3. Terminology

Being careful with language is not just about wording, but also conceptualization and operationalization. Birk urges MGB and other researchers to be more precise about the use of the term stress. He locates several definitions in our sample of highly cited MGB papers and finds “no coherent theoretical paradigm and … very different methods of inducing ‘stress’” (para. 2). Our target article noted efforts to avoid anthropomorphizing of anxiety and depression, yet this does not happen with “stress,” observes Birk. He gives very good examples of problems with the unqualified use of this term, which occurs well beyond the MGB area of research. We greatly appreciated his concluding salvo: that rather than grand discussions of microbiomes and their impact on the human self (as is done in Rees et al. Reference Rees, Bosch and Douglas2018), the field needs to deflate itself to deal with basic operationalization of terms.

Birk also shows that “social” is another inconsistently deployed concept in rodent behavioural tests. He goes on to argue more generally that all scientific terms and concepts need careful operationalization. We agree wholeheartedly. Our pet peeve in MGB research and beyond is the term “dysbiosis.” Johnstone & Cohen Kadosh, for example, take the term to be part of established microbiome explanations. For them, “dysbiosis has been shown to be related to symptoms of anxiety and depression,” and that there are “consequences of dysbiosis [that are] particularly critical during development” (para. 3). They suggest that improved MGB research can be achieved simply by taking into account “the timing of dysbiosis” (para. 3). Andreoletti & Rescigno likewise believe that dysbiosis contributes to disease, and that it is a generic phenomenon untroubled by the inter-individual variability of microbiomes.

As we outlined in our target article and have addressed in earlier work (Hooks & O'Malley Reference Hooks and O'Malley2017), the term dysbiosis suffers from loose and messily overlapping definitions that often tie disease diagnosis and explanation together in a vicious explanatory circle. By throwing the term around so liberally (in the broader microbiome field, as well as MGB), causality is further obscured. Rather than talking about mere “changes” or “reduced diversity” in microbiomes as drivers of disease (e.g., Johnstone & Cohen Kadosh), researchers would be considerably better off specifying which fine-grained taxa are increasing or decreasing, their potential or actual roles in the causal pathways that lead to disease outcomes, and whether many of the microbes might simply be along for the causal ride when the host is diseased. All of these aspects of host-microbiome relationships are beginning to be addressed in broader microbiome research but have some way to go.

Alberts et al. try to avoid the conceptual issues of dysbiosis by referring to microbiomes as “dysregulated” instead of “dysbiotic.” Borghi et al. refer coyly to “unbalanced ratios” of microbiota composition. Do these rewordings avoid the problem? We do not think so. First, there is no evidence that the microbiome should be treated by the host as a unitary entity that can be regulated to achieve a definitive desirable state rather than less desirable ones. Microbes come and go and compete furiously among themselves and with the host (Foster et al. Reference Foster, Schluter, Coyte and Rakoff-Nahoum2017). Discussing stable states as “homeostasis” (the supposed opposite of dysbiosis) misleads about the dynamics producing such states. Although there may be microbiome states that have specific pathogens that affect host health, it is simpler and more accurate to refer to those taxa. If the microbiota is influenced by diet, and the diet seems to have effects on both host and microbiome composition, it is more straightforward to say just that. As for balance, this is the revival of an antiquated notion of thinking about health, and there is neither evidence nor theory to warrant claims about microbiome imbalance as a mechanism for disease (Olesen & Alm Reference Olesen and Alm2016).

When referring to “dysbiotic microbiota,” Fetissov argues for a focus on “specific bacteria in modulating brain function and behaviour” (para. 1). This, we suggest, identifies very nicely where the true problems of the term dysbiosis are located along with their potential solutions. Although attempts have been made to identify “dysbiotic” signatures of whole microbiomes, ultimately, very particular groups of microbes are probably producing the harmful effects on the host (as Lynch notes, this is a more “proportionate” causal explanation than the whole microbiome). In such a situation, there is no need to mention “dysbiosis.” Instead, it will be sufficient to specify which organisms are doing what and how these relationships hold under distinct (experimental) conditions. Ultimately, the issues to do with dysbiosis come back to the causal uncertainty of microbiome research, which has come up repeatedly in these three themes and the commentaries grouped under them.