Abstract

Introduction

The completion of the Human Genome Project in the early 2000s strongly suggested that, even if the genomic content of an organism is relevant to know some of its phenotypic features, it is by no means the “holy grail” that would allow uncovering all the determinants of the human phenotype—including the determinants of human behaviour—as some had incorrectly supposed. As a matter of fact, instead of revealing the genetic determinants of the human phenotype, the Human Genome Project revealed that the phenotype is a polyhedral product of multiple and heterogeneous factors, including both nature (genetic determinants) and nurture (developmental and environmental cues). This influenced the emergence of the so-called “post-genomic” approaches to the study of life, in which the number and kind of molecular elements affecting the phenotype of the organism expands over and above its genome (Griffiths & Stotz, 2013). This movement apparently puts organisms and their causal complexity at the centre of the debate, partially involving a movement from the century of the gene—as Fox Keller (2000) baptized the twentieth century—to the century of the organism (Baedke, 2019; Montévil et al., 2018). Post-genomic approaches to the study of life include the study of epigenetic factors, the interaction between the genotype and the environment (e.g., niche construction theory), a closer look to the patterns of gene expression (functional genomics, proteomics), or the study of the microbiome, among others. The beginning of the postgenomic era suggested that some of the topics and controversial philosophical assumptions embedded in the Human Genome Project could be overcome, including its tendency towards essentialising the human nature, its lineal conception of causality (from genes, to proteins, and from proteins to phenotypes), its genocentrism and genoreductionism, and its biases towards genetic determinism (Gannett, 2019).¹

However, in recent years, many have emphasised that the postgenomic era is ultimately grounded in the same (or some of the same) underlying philosophical assumptions that grounded the Human Genome Project. For example, in a rich and fundamental work on the topic, Meloni & Testa (2014, p. 439, original emphasis) show that a paradox in postgenomic research “rest(s) on [its] bivalent [and, I’d say, contradictory] understanding of its relationship with genomics: on the one hand as a missing link that can succeed where genomics purportedly failed, on the other as a quantum leap enabled by the very success of genomics”. Note that the first reading entails that postgenomic research inherits all the problematic philosophical assumptions embedded in the Human Genome Project, as the very conceptual or ontological foundations would remain unaltered. In a similar vein, Waggoner & Uller (2015) argue, based on the type of language used in at least three areas of postgenomic research (“genetic” control, programming, and inheritance), that postgenomic approaches restate the determinism but in a form of environmental determinism. To quote: “[Postgenomic language] may simultaneously promote a novel form of determinism, one that highlights the “influential” role of the environment and behaviour in determining individual characteristics and even the expressed genetic code of future generations” (Waggoner & Uller, 2015. p. 178, emphasis added). Similar claims can be found in more recent works, such as Richardson & Stevens (2015), Dubois et al. (2018), Santaló (2018) or Dupras et al. (2019). An area of particular interest in this context is microbiome research, which also put forward its own version of the Human Genome Project by launching the Human Microbiome Project in 2008 (Turnbaugh et al., 2007; see https://www.hmpdacc.org/ for all the information about the Human Microbiome Project between 2008 and 2016; web last accessed on 28/12/2023). My goal in this paper is to examine how microbiome research implicitly endorses deterministic conceptions on its own, what these conceptions specifically consist in, what they are grounded on, and their limitations given our contemporary knowledge of the microbiome and its relationship to host biology.

Before proceeding, I would like to clarify that this paper is not about the potential causal role of the microbiome, which has been abundantly discussed in the literature so far (Hanage, 2014; Cani, 2018; Bourrat, 2018; Walter et al., 2020; Lynch et al., 2019; O’Malley & Parke, 2020), but rather about a specific dimension of this causal role. While the topic of determinism and the topic of causality may be linked, they are conceptually different (Hoefer, 2016). In biology, genetic—and postgenomic, as I will show—determinism concerns whether the genetic fixes the traits that an organism expresses. In this sense, genetic determinism presupposes that causation works bottom-up (from genes to phenotypic traits) and that the genes play the most important causal role in fixing the traits. In contrast with this, the debate about causality admits different degrees or forms in which causal claims can be made, and includes different types of causal influences or interactions, including both top-down and bottom-up forms. To put it differently, the issue of determinism concerns whether the strength of the nomic relationship between a specific cause and an effect is enough for the cause to be determining that the effect is primarily produced by such cause. Thus, even if my conclusions will suggest that deterministic claims about the microbiome—in the two forms I will distinguish later—are misguided, the reader should not presuppose that I am simultaneously arguing the microbiome does not play any type of causal role for its host.²

My agenda will be as follows. In Sect. 2, I introduce the concept of genetic determinism as it is used in today’s biological research. In Sect. 3 I motivate the application of the notion to microbiome science by distinguishing between “host-microbiome determinism” and “microbiome-phenotype determinism”. In the second part of this section, I offer an analysis of the most common methodological practices used and the main question raised in today’s microbiome research to illustrate the specific character of these two types of determinism. In Sect. 4, by drawing on our current knowledge of host-microbiome systems, I show that both types of determinism are ungrounded. Finally, in Sect. 5, I present my conclusions.

Genetic determinism

Roughly speaking, genetic or biological determinism is the belief that the traits expressed by an organism during its life cycle are a product of its genetic composition. These traits range from basic phenotypic traits such as eye or skin colour, the capacity for tongue rolling, or boldness, to more complex traits such as cancer, schizophrenia, diabetes, or Alzheimer’s disease, and also the so-called social or personality traits like openness, extraversion or sexuality phenotypes such as being gay. Genetic determinism is usually motivated by the correct observation that the expression of some traits is strongly connected to the presence or absence of specific genes or alleles (Sarkar, 1998, p. 10). For example, if a person inherits two mutations (one maternal, one paternal) in the beta globin gene, she will likely develop sickle cell anaemia; and if a person inherits a trisomy in the chromosome number 21, she will likely be born with Down syndrome. A key potential consequence of this basic and under elaborated view of genetic determinism is thus that environmental interventions are useless when it comes to modify the expression of certain traits, because their expression depends on the genetic composition of the organism expressing the traits. In other words, if a trait is genetically determined, then the organism bearing the genes for the trait will carry the trait, no matter how many efforts one makes in altering the environment to avoid its instantiation (Rosoff & Rosenberg, 2006, p. 123).

When it comes to genetic determinism, however, it is convenient to distinguish different meanings, as the expression is polysemic, and it is not always clear what concept of genetic determinism is being used across different contexts (Gayon, 2009; Kaplan, 2000). Part of the problem stands from the complexity of the concept of determinism itself, and its relationship to debates about human agency, causality of the future, and prediction (Hoefer, 2016; Müller & Placek, 2018). In fact, some of the arguments against genetic determinism emphasise the complexity and the lack of predictability of some biological facts (Robertoux & Carlier, 2011), whereas others put their emphasis on the plurality of (causal) factors beyond genetics affecting human nature (Dupré, 2003). These two extreme visions of what genetic determinism is, and the responses that need to be articulated to resist genetic deterministic claims, reflect the existence of at least two different concepts potentially meant by the expression. On the one hand, what Kaplan (2000) calls the “complete information” strand, and Gayon (2009) names “Laplacian (scientific) determinism”. This is an epistemological interpretation of genetic determinism, according to which the genetics of an organism provide complete information about its phenotype, to the point that if the details of the genotype were fully known, it would be possible to predict the complete phenotype that the organism would develop during its ontogeny. Kaplan (2000, p. 11) argues that this form of determinism is considered trivially false, although as a matter of fact some researchers defend its falsehood on the basis of the complex mapping that exists between the genotype of an organism and its phenotype (Roubertoux & Carlier, 2011).

On the other hand, what Kaplan (2000) names the “intervention is useless” strand, which can roughly be equated to Gayon’s (2009) “Laplacian (metaphysical/ontological) determinism”. The main idea underlying this interpretation is that the genes fix the phenotypic traits to the point that the possession of one gene determines the expression of the trait, regardless of the potential interventions in the environment. To give an example, if sickle cell anaemia is fixed by having two mutations in the beta globin gene, then it does not matter how we modify the environment of a person who carries two copies of the gene, because she will bear the trait. Note that this conception of genetic determinism is not strictly the same as the type of determinism used in physics or in other disciplines. The point here is not that the genes strictly fix the phenotype, but rather that they constrain its expression so narrowly that human intervention in the environment is unwarranted. “Intervention is useless” genetic determinism is also almost impossible to maintain, both in its application to simple traits—with phenylketonuria as a classic example of why this conception of genetic determinism does not hold—but particularly in its application to complex diseases as well as social or personality traits, where the potential for intervention is especially broad, and biologists are generally conscious of this (Ellison & De Wet, 2018).

Given that these two conceptions of genetic determinism are inadequate to capture what biologists mean when they use the expression, as well as to capture the research founded on deterministic hypotheses that is carried out in today’s biology, Kaplan urges to consider a third meaning that he characterizes as the conjunction of two theses, one methodological and the other ontological: “(a) the genetic is the natural place to look when attempting to explain, predict, and control traits with even partial genetic etiologies, and (b) traits with partial genetic etiologies are best understood as being primarily genetic, and it is only through directed intervention that the expression of genes for traits with partial genetic etiologies can be avoided or controlled” (Kaplan, 2000, p. 12). In this context, genetic means something similar to “vertically inherited”, i.e., passed from parent to offspring during zygote formation (cf. Merlin & Roboli-Sasco, 2021; Veigl et al., 2022). Under this conception, genetic determinism is a driving hypothesis of biological research which encourages researchers to look for the hereditary basis of the trait (methodological primacy of the genetic) as it is assumed that its expression can be manipulated by controlling such basis (ontological control by the genetic). In a motto, this concept of genetic determinism can be summarized in the hope of “controlling the trait by controlling the genetic”. Kaplan argues that this third form of genetic determinism is present in most of today’s biological research, and will usually explicitly appear due to the emphasis of the power of the genetic in allowing both predictions and control.

It must be noted at this point how the issues of genetic determinism and biological causality differ. Claims about biological causality may emphasise that a specific factor plays a causal role in producing a trait, for example, if someone appeals to the joint role of developmental constraints and selective effects in producing a specific form of a trait, like the obstetric conundrum (Grunstra et al., 2019). In these cases, while biologists clearly make causal claims, some of which involve the genetic level, it makes little sense to speak of genetic determinism, because the genetic is at the same level as other factors, and the evolution of the trait is considered complex, instable, and highly contingent. Thus, genetic determinism differs from genetic causality in that the former, but not the latter, prioritizes the genetic over any other factor. Secondly, biological claims about causality may be both top-down or downward and bottom-up or upward (see Sect. 4). For instance, research on natural selection may appeal to how selection on traits expressed at the higher-level causes the distribution of genetic forms in the lower-level to differ from what would be expected if natural selection was a cause of change by acting exclusively on the lower-level (Lloyd & Wade 2019; Suárez & Triviño, 2019; Suárez & Lloyd, 2023). When causation is predicated downwardly, it makes little sense to talk of genetic determinism, as one would be presupposing that the genetic is neither methodologically nor ontologically privileged, but just an effect of what happens in non-genetic levels (organism-environment interaction). However, this would be a case of biological causality. Therefore, biological determinism, as it is used in today’s biology, concerns both the strength and the direction of the causation, whereas the range of potential causal claims being made in biology is wider. Grounded on this conception, in the next section, I show how microbiome research introduces deterministic considerations on its own, and how they relate to Kaplan’s concept of genetic determinism.

Two types of microbiome determinism: host-microbiome determinism and microbiome-phenotype determinism

Microbiome research is a growing field in contemporary biology which studies the microbial composition of multicellular organisms, specially animals and plants, as well as their influence on animal phenotype, including behaviour, physiology, reproduction and health. The growth of microbiome research started in early 2000, with the development of high-throughput sequencing technologies. These technologies allowed divorcing bacterial identification from bacterial culture, and permitted discovering that animal and plants harboured millions of microbes in their bodies. This led some authors to interpret that this discovery should have profound implications for our understanding of biological individuality (Gilbert et al., 2012; McFall-Ngai et al., 2013; Stencel & Proszewska, 2017; Suárez, 2018). For example, Dupré (2010) argued that organisms should now be conceived as polygenomic (de-essentialized) entities, whose expressed traits were highly indetermistic as they depended on the unfolding of several genomes (see also Dupré, 2021). However, contrary to the expectations, it led some to consider that the microbiome constitutes “a second genome”, serving as a source of genetic variation that could even hypothetically account for phenomena like the missing heritability problem (Sandoval-Motta et al., 2017). This led to the introduction of deterministic conceptions in microbiome research, conceived in terms Kaplan’s third meaning of “determinism” (Sect. 2). While he introduced this third meaning for studies carried out under the hopes raised by the Human Genome Project, I think the conception of genetic determinism he offers can serve as a basis to understand the very specific forms of determinism that are implicit in today’s research on the microbiome, even if the latter supposes an expansion of the former as I will show later. Particularly, in most microbiome research it is also usually assumed that the genetic determines the expression of certain traits: (a) by being a methodologically privileged place to investigate the source of these traits and (b) by playing a primary causal role in their expression, provided these traits have partial genetic aetiologies.

The analogy between genetic and microbiome determinism is grounded on the observation that the microbiome affects the expression of several host traits. Well-described cases include: the influence of Vibrio fischeri on the bioluminiscense of the Hawaiian bobtail squid; the role of different Treponema species in favouring the production of short-chain fatty acids from wood fibres in Nasutitermes takasagoensis; the role of cellular microbes of the exoskeleton of the leaf-cutting ant Acromyrmex echinatior in nestmate recognition; or the role of Bacteroides plebeius in the digestion of seaweed in humans (see Lynch & Hsiao, 2019 for a review of these and other examples). While this evidence opens up the possibility of thinking that host traits can eventually be primarily modified by manipulating the microbiome, the situation is not as simple, for the microbiome needs to be assembled de novo each generation (Chiu & Gilbert, 2015). Furthermore, most of the microbes composing the microbiome of a host are acquired horizontally, during the life cycle of the organism. While traces of vertical transmission can be found for some microbial species and some animal taxa, they constitute a minority. The host has thus to assemble its microbiome “afresh” each generation. Therefore, there is no a priori guarantee that the microbial components being acquired belong to the same lineage, or even to the same taxa, as the components borne by their progenitors. The key question that microbiome researchers need to ask before considering whether the microbiome can play a primary causal role in determining the host phenotype is whether there is any host determinant conditioning microbiome assembly, or rather the process that depends on other factors (e.g., ecological or behavioural). In most microbiome research, it is further assumed that the primary host determinant of microbiome assembly must be the host genotype, and that the components to be acquired correspond to the genetic components of the microbiome. It is thus inferred that if the microbiome were reassembled by factors that do not depend on the genotype of the host, this would hamper the potential for “controlling the trait by controlling the genetic”, where genetic refers to the genetic composition of the microbiome. This assumption grounds that many deterministic practices from the Human Genome Projects have been extended to microbiome research, and this extension came hand in hand with an extension and refinement of the older deterministic views.

Henceforth, an important part of microbiome research is aimed at showing: On the one hand, that microbiome assembly is not primarily guided by ecological or behavioural processes, but it is rather strongly determined by the host genotype. This type of research gives rise to what I call “host-microbiome determinism”; on the other hand, that the microbiome influences on part of the host phenotype in a substantial way, having causal primacy over other factors for the production of certain traits. This gives rise to what I call “microbiome-phenotype determinism”.³ Both forms of determinism are specifications of Kaplan’s third meaning of genetic determinism, as these are applied in microbiome research. This occurs because both forms of microbiome determinism entail an excessive reliance on the genetic and the molecular as the primary causal agents of some traits, and in doing so they entail that other causal factors (some of which may ultimately play a more prominent causal role than the genetic in the production of the trait) are ignored. This a crucial feature of the denounce of microbiome determinism that I am making. Table ​Table11 introduces a comparison of the two types of determinism isolated in this work.⁴

I contend that these two lines of research in contemporary microbiome science constitute two ways in which the old non-trivial deterministic inclinations embedded in the Human Genome Project and diagnosed by Kaplan re-appear in the context of microbiome science. In other words, I claim that the hopes of “controlling the trait by controlling the genetic” and, in doing so, masking the study of non-genetic factors influencing the expression of the phenotypic traits, are still fully present in some parts of microbiome science, and they manifest specifically in “host-microbiome determinism” and “microbiome-phenotype determinism”. In what follows, I explain in detail what these two form of determinism consist in, show how they manifest in today’s microbiome research and the dangers they entail.

Problems of microbiome determinism

Understanding the limitations, perils and false hopes of the two types of determinism in microbiome science analysed in this work requires first understanding the nature of host-microbiome associations. Particularly, it requires understanding the complexities of the interactions, how these are built and maintained, how these can change during host development and evolution and, importantly, how these are stabilized and can be destabilized. This is because the whole question surrounding determinism is whether an attitude that privileges a methodology and an ontology grounded on the primacy of some parts over the rest of parts of the whole in explaining/predicting and controlling trait expression is coherent with our scientific knowledge of the relationship between the privileged parts and the other parts of the whole. In other words, do the parts deemed responsible for the primacy over the control really have that primacy? And, if they do so, do they acquire that primacy regardless of the context, or is their primacy just a mere causal influence when the context is favourable for that type of action? In the specific case of microbiome science, these two questions may be asked with regards to the host-microbiome determinism and microbiome-phenotype determinism.

A possible way of replying to these questions for host-microbiome determinism would consist in showing that it is empirically false that the host genetically controls microbiome assembly (ontological thesis). Rather, the microbiome is assembled de novo in each generation and the evidence suggests that this assembly depends on the host ecological opportunities, rather than on the host genetics, as many authors have shown (e.g. Moran & Sloan, 2015; Douglas & Werren, 2016; Skilling, 2016; Hurst , 2017; Bourrat & Griffiths, 2018; Stencel & Wloch-Salamon, 2018; Koskella & Bergelson, 2020; Roughgarden, 2023). Hence, the host genetics is not be best place to primarily study microbiome assembly (methodological thesis). Note that this does not deny that the host genotype may play some causal role in determining the microbiome assembly. It simply states that it does not play a primary role, and thus any of the methods I described in Sect. 3 as fostering host-microbiome determinism should be complemented with others studying other potential factors determining microbiome assembly. For example, studies on the ecology of the microbiome, or studies on how specific environmental factors may drive the acquisition of certain microbiome components, or the acquisition of specific microbiome functions (Boem & Suárez 2024).

However, while this argument could work for the specific case of host-microbiome determinism, I do not think it would be a fully convincing strategy to oppose microbiome-phenotype determinism. For while the critics would have a powerful empirical claim against the host genotype having a primary causal influence on microbiome assembly, this argument does not touch on the conceptual and empirical grounds of microbiome-phenotype determinism, though. There are two reasons for this. Firstly, conceptually speaking, rejecting that the host genotype plays a primary causal role in assembling the microbiome is consistent with the thesis that, if it did so, then the microbiome would be a primary causal factor in producing the host phenotype. To put it differently, it is consistent with stating that if the assembly of the whole microbiome, or the assembly of some microbes in the microbiome (e.g. the mitochondria in eukaryotic cells), were controlled by the host genotype (e.g., by controlling its mode of transmission), then the microbiome could control the expression of certain traits of its host (ontological thesis), and/or should be the primary source for investigating these traits (methodological thesis). Secondly, denying that the host genotype is the primary causal factor of the microbiome assembly leaves untouched the empirical question about the causal relationship between the microbiome and the host phenotype. This is because it may happen that the microbiome plays a primary role in determining the host phenotype even if the microbiome were environmentally acquired. All that is required is that the microbiome bears some genes that play a primary causal role in the production of some host traits (ontological thesis) and thus the microbiome would be the best source to predict or explain why the host bear those traits (methodological thesis). In short, the mode of acquisition of the microbiome is, to a certain extent, empirically irrelevant to explore the validity of microbiome-phenotype determinism. Therefore, I claim this not to be a good conceptual move against microbiome-phenotype determinism.

A more promising avenue to contest microbiome-phenotype determinism consists in pointing out the nuances of phenotype expression, as well as the specific role that the microbiome may play in that phenomenon. Concretely, I refer to three well-described biological phenomena that question microbiome-phenotype determinism: (1) epistasis plays a key, foundational role, in the type of synergies driving host-microbiome associations; (2) not every part of the microbiome can play a causal role for phenotypic expression, but only discrete traits which are frequently redundant; (3) host-microbiome systems exhibit a network structure, in which network properties play a more determinant role on what happens to the parts than the other way around. While these three phenomena are in the end interconnected, they can be analytically separated to facilitate the study of how each of them independently questions the validity of microbiome determinism in any of the two forms isolated in Sect. 3. Particularly, I argue that: (1) epistasis questions methodological microbiome-phenotype determinism; (2) microbiome redundance suggests the untenability of methodological and ontological host-microbiome determinism; (3) the network structure of host-microbiome systems challenges the possibility of ontological microbiome-phenotype determinism. Let us deal with these points in detail.

Epistasis. Epistasis refers to the non-additive interactions between genes—alleles—in different loci that result in an expressed phenotype which is different from the phenotype that would be expressed if each of the interacting genes would have acted separately. Evidence supporting the role of epistasis in determining several phenotypic effects is abundant, and so are the long term effects of epistatic interactions between genes (Phillips, 2008). Epistasis between the host and either the taxa or the functional traits of its microbiome is speculated to be a primary driver of host-microbiome interactions, and several theoretical and empirical studies have already proven this to be so (Bourrat, 2019; Lloyd & Wade, 2019). As a matter of fact, epistasis is hypothesised to underlie the main biological processes leading to host-microbiome evolution, including the potential for the evolution of mutualism and vertical transmission.

Epistasis poses a serious obstacle for the possibility of methodological microbiome-phenotype determinism due to the non-linearity of the interactions among genes at different loci, and the fact that this non-linearity eventually leads to emergent phenotypes, as consistently documented in the literature on emergence (Bedau, 1997; Suárez & Triviño, 2019, 2020; Wilson, 2016).⁷ Microbiome-phenotype methodological determinism works by assuming that the microbiome is the most adequate place to investigate, predict or explain host traits with partial microbiome aetiologies. Thus, it leads to research methodologically comparing the microbiome of hosts expressing the trait of interest x, and those not expressing it or expressing a different value of the trait—if the trait is quantitative. This hypothesis would be feasible if the interactions between the genes in the microbiome, or between the microbiome and the host, were merely additive. In that case, it would be possible to predict host states from microbiome states, and the likelihood that the microbiome primarily controls the expression of some traits would increase. In contrast with this, epistatic interactions are non-linear, both in the host-microbiome relationship and in the interrelationships between the different components of the microbiome. Therefore, the specific phenotypic state that the host expresses turns out to be unpredictable. This is important because it suggests that instead of interpreting the work on mGWAS or on gnotobiotic research as if it suggested that some differences in the microbiome may allow the prediction of differences in the expressed traits, it should be read the other way around: as if some differences and/or analogies in the expressed traits may allow the prediction of differences in the microbiome. That is, as if the differences or similarities in the microbiome were methodologically irrelevant to account for the differences in the expressed phenotype at the higher-level, because epistasis precisely suggests that the higher-level is a better predictor of what happens at the lower-level. Therefore, epistasis questions the possibility of any form of methodological microbiome-phenotype determinism insofar as it entails admitting that that the expression of the traits will depend on their global context, and thus the microbiome cannot be the most adequate place to investigate, predict or explain host traits with partial microbiome aetiologies.

Discrete and (mostly) redundant traits. Most host-microbiome interactions do not affect the microbial components in the same way that they affect the host species, generating a host-centric unit based on the non-symmetry of the relationships between the components (Schneider, 2021; Stencel, 2022; Suárez & Stencel, 2020). This non-symmetry manifests in the interaction between the host species and a pool of variable bacterial species that bear or express similar traits. This leads to a functional redundancy of the microbiome, in which the traits contributing towards the global functionality of the host-microbiome system will tend towards stabilization, regardless of the species bearing them.

While the fact that discrete traits play the causal role in host microbiome systems does not necessarily hamper the possibility of developing a coherent and empirically grounded form of microbiome-phenotype determinism, it poses a serious challenge to the promise of host-microbiome determinism, both in its ontological and methodological variants. Let me sketch the reasons why this is so. If the main association is between the host and a set of discrete traits borne by the bacterial taxa, then it seems implausible that the host genetics may have any primary role in determining microbiome assembly. On the one hand, it seems evolutionarily implausible that the host genetics has an evolutionary primacy in determining which bacterial taxa the host associates with, especially given that the biologically relevant host interaction is not with the taxa but with some of the traits borne by these taxa. At most, microbiome assembly may be associated to host genetics by a common cause, but this is a far cry from the tenets of host-microbiome determinism which requires the host genetics to be a primary cause of microbiome assembly. On the other hand, even if one may correctly assume that host-microbiome determinism could be reformulated such that the deterministic primacy runs from the host genetics to the functions or traits expressed by the microbiome (Greslehner, 2020), the paradox would now be to explain how some traits, mostly known to be involved in abundant horizontal gene transfer and able to be borne by several bacterial taxa may guide microbiome assembly due to their association with the host genetics. Rather, what seems to be happening is that selection within the host-microbiome system after assembly and during the lifespan of the host drives the microbiome towards an optimal state where the necessary traits become dominant and redundant (Suárez, 2020). In fact, the empirical evidence suggests that colonization depends both on the context of the host as well as on bacteria-to-bacteria interactions, rather than on host control over bacterial assembly (Jones et al., 2022). This is important because it should oblige biologists to investigate other potential determinants of microbiome assembly, including host behaviour and/or host environment, as well as the role of the immunological system of the host and the very ecology of the microbiome in determining microbiome composition. But, as discussed in Sect. 3, microbiome-phenotype determinism, unfortunately, often precludes these investigations.

Network structure of the host-microbiome system. A system is said to possess a network structure when the association between its parts generate an emergent regime of causation due to the appearance of constraints resulting from the interaction between the parts. A constraint is a material structure that results from the interaction between the parts and harnesses their interactions, so that the range of possible states that each of the parts occupies is smaller than the potential states that it would occupy if no constraint existed (Emmeche et al., 2000; Umerez & Mossio, 2013). Green (2018), drawing on Emmeche et al. (2000), grounds the importance of the network structure on the existence of boundary conditions, “conditions under which a given mathematical model or equation hold (e.g., by specifying a value interval for the possible solution)” (Green, 2018, p. 1001; see also Green, 2020). Boundary conditions ontologically define how the higher-level constrains the behaviour of the lower-level parts, as their range of behaviours is always restricted to those delimited by the mathematical model defining the boundary conditions—which, by definition, must be narrower than the possibilities offered by the laws governing the behaviour of the lower-level parts. Simultaneously, a system with a network structure generates the potential for the appearance of new structures and functions that would be unlikely if the system lacked such structure (Moreno & Suárez, 2020).⁸

Two important features of network structures are their recursion and their resistance to perturbations or robustness. By the latter I mean that a system with a network structure is often capable of remaining in the same state—i.e., realizing the same dynamics—despite a range of potential threads that alter or affect their component parts (Green, 2022; Moreno & Suárez, 2020). In short, what happens when a network is perturbed is that it gets driven out of equilibrium, but the dynamics of a network are such that a new stable equilibrium will be reached after a certain time (Wagner 2005, for an analysis in terms of the concept of “distributional robustness”). This feature concerns the specific form of physical realization in network systems, being generally taken to entail a form of ontological emergence (Wilson, 2016), manifested by the strong degree of causal autonomy of the higher-level (in the case of the host-microbiome association, the host-microbiome system as a whole).

Different authors have already emphasised the network structure of host-microbiome systems (Doolittle & Booth, 2017; Doolittle & Inkpen, 2018; Huitzil et al., 2018; Bapteste & Papale, 2021). It is important to note, although in passing, that the realization of a network structure—regardless of the type, which varies across systems—affects all dimensions of these systems, including their ecology, evolution, physiology, etc. (Suárez, 2020). An important feature of most systems realizing a network structure is that it will be both robust—it will maintain its global, network properties despite perturbations or alterations at the lower-level—and resilient—capable of recovering quickly after a perturbation that the system is not vulnerable to (see also Green & Batterman, 2017). These two features definitely rule out the possibility of ontological microbiome-phenotype determinism. In a network system, what happens to the parts is conditioned upon the type of constraints that emerge from their interactions. In this sense, no part can be privileged over the rest, isolated, or attributed a function over and above the properties of the whole—the network. This is because the network exhibits a characteristic dynamic and every element that enters the network will be harnessed so that the dynamic is maintained. The network dynamics results from the very interactions between the elements, and even if the host assumes the role of a central node within the network, which is something known to happen in host-microbiome systems (Suárez & Stencel, 2020), its behaviour cannot be reduced to the behaviour of the host. This also affects the microbiome and its potential for affecting the phenotypic expression of the host. It is true the microbiome may alter that expression, but only if it is able to bring the whole host-microbiome network towards a new dynamic. For this to happen, the changes in the microbiome need to be structural, such that the action of the host, or the action of other components of the microbiome, do not hamper the possible changes brought about by the former. In this context, the possibility that the microbiome primarily controls any host trait is ruled out. At most, alterations in the microbiome can bring about new dynamics to the host-microbiome system, provided the former does not resit these alterations. Importantly, these new dynamics will likely be temporal, as the host-microbiome network will tend towards a stable equilibrium. Even when these alterations are not temporal, the causality is not linearly produced, from the microbiome to the host. It is rather a network dynamic. Therefore, ontological microbiome-phenotype determinism in the form defined in this paper is untenable.

Conclusion

This paper argued that, despite the promises that microbiome science could definitely debunk some deterministic hypotheses originally embedded in the Human Genome Project, an important part of contemporary research in microbiome science is grounded on deterministic theses. Concretely, I argued that some of the gene determinism underlying the Human Genome Project was expanded and became applicable to study the causal influences between symbiotic organisms. In microbiome science, determinism primarily permeated through the methods being used—mainly molecular and focusing on the microbiome species composition, rather than on other biological characteristics—and the approaches being neglected—for example, ecological approaches. Grounded on this, I distinguished two forms of determinism in contemporary microbiome science: host-microbiome determinism and microbiome-phenotype determinism. I showed that both forms of determinism presuppose the ideas of “genetic control” and “methodological primacy of the genetic” that permeated genomic determinism. Due to the nuances of microbiome science, I also showed that these approaches permeated in a distinctive way: namely, by reducing microbiome research to molecular methods, mainly gene-centred based methodologies, and ignoring other types of methodologies currently available, such as those based on ecological methods. Secondly, by relying on recent research on host-microbiome systems, I argued that these forms of determinism score poorly with our best ways of characterizing the relationships between microbiomes and their hosts. Concretely, I first argued that our current evidence suggests that the host genotype is not the main factor driving microbiome assembly, thus questioning host-microbiome determinism; second, I emphasized that the role of epistasis in shaping host-microbiome interactions, the fact that host-microbiome interactions do not affect the whole microbiome, or even the species composition of the microbiome, but rather microbial traits, and the network structure of the microbiome, jointly make questionable the main assumptions underlying microbiome-phenotype determinism and host-microbiome determinism.

I conclude by encouraging others to investigate other possible forms of microbiome determinisms I may have missed. In addition, I suggest, that more research needs to be conducted on how the microbiome may still be an important causal factor contributing to variation in the host phenotype, as well as how causality generally acts in the microbiome, beyond the constraints imposed by any (gene) deterministic programme. Finally, I encourage researchers to address other factor than transcend the genetic level when uncovering the primary causes of phenotypic traits in order to avoid that microbiome research repeats the same ontological and epistemological mistakes that were on the basis of the Human Genome Project.

Acknowledgements

Funding

Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature.

Footnotes

References