Role of nasal microbiota and host response in infants with respiratory syncytial virus infection: Causal questions about respiratory outcomes

Respiratory syncytial virus (RSV) is a major cause of acute respiratory infection (ARI), such as bronchiolitis, in infants.¹ As with other viral induced ARIs, the severity of RSV illness ranges from a minor nuisance to fatal and the pathobiology remains poorly understood. To date, for infants with RSV infection no effective treatment strategies have consistently proven helpful other than supportive care. In addition to the substantial acute morbidity, the literature has also demonstrated that RSV infection during infancy contributes to the development of chronic morbidities (e.g., recurrent wheeze, childhood asthma). Indeed, given the high prevalence of RSV infection, it is the risk factor with the largest population attributable fraction in incident asthma.² However, the underlying mechanism linking the acute infection with these chronic wheezing conditions remains unclear. Importantly, developing treatments for RSV infection and prevention strategies for asthma both require a comprehensive understanding of the underlying pathobiology which includes the airway microbiome and its crosstalk with the host immune response in infancy—a crucial period of airway development.¹

Several decades of RSV research hold important lessons for clinicians and scientists, if only by reminding us of familiar conundrums. Most notably, is clinically-significant RSV infection a causal factor for incident asthma, or is it simply a marker of infant who is predisposed to develop or already has asthma? As a marker, RSV infection is akin to a “positive stress test” that identifies infants as at-risk for asthma. These two models are not mutually exclusive—i.e., RSV infection may not only increase the risk for subsequently developing asthma (i.e., causal) but also identify infants prone to asthma (i.e., marker). The causal question has been debated for decades due to, among other factors, the lack of appropriate high-throughput molecular testing (e.g., immune profiling, microbiome testing) and the difficulties in robust causal inference from observational research. Indeed, mainstream statistical science has provided clinical researchers with few approaches to explicitly articulate, let alone answer, causal questions.³ Consequently, several major journals have virtually prohibited causal vocabulary in observational research.³ However, the recent advent of high throughput technologies and counterfactual causal inference frameworks (Table) now enable researchers to ask and answer important causal questions (e.g., Does the interplay between the microbiota and host response in the airway contribute to both acute severity in and chronic morbidities of infant RSV infection?).

Table.

Glossary

α-diversity	Within-sample measures of the richness (the total number of different microbial species) and evenness (relative differences in the abundance of various species) of a community of microbes in an ecological niche, such as Shannon diversity index and inverse Simpson index.
β-diversity	Between-sample measures of similarity or dissimilarity, such as unweighted and weighted UniFrac, and Bray-Curtis, and Jaccard (presence vs. absence) distances.
Causal inference	The process of using data in a sample in order to infer a cause-and-effect relationship in the target population of interest.
Counterfactuals	Counterfactuals are how humans intuitively reason the causal effect, for which we compare two outcomes: 1) an outcome (e.g., wheezing) that would have been observed with a hypothetical exposure (e.g., a specific microbiome signature) versus 2) an outcome that would have been observed without such exposure. These two outcomes are referred to as counterfactual (or potential) outcomes as they represent world(s) that may not exist—that is, counter-to-the-fact outcomes.
Counterfactual causal inference	Causal inference methods that are based on the framework of counterfactuals to define and estimate causal effects. For a binary exposure case (e.g., exposure yes/no), this framework presupposes the existence of two outcome conditions (i.e., two counterfactual or potential outcomes) to which all individuals in the population could be exposed.
Dysbiosis	An imbalance in a community of the microbes in a particular niche secondary to various changes (e.g., the use of antibiotics, infection, inflammation).
Effect modification	The effect of one exposure (e.g., specific microbiome signature) on the outcome (e.g., wheezing) varies—qualitatively and/or quantitively—across strata of a third variable, the effect modifier (e.g., host immune profile).
Interaction	While effect modification is sometimes called an “interaction” in statistical science, a more precise definition of interaction is the joint effect of multiple interventions (or exposures) on the outcome.
Identifiability conditions	Three fundamental assumptions (i.e., exchangeability, positivity, and consistency) that are required to identify the average causal effect of interest from data. Exchangeability means that the exposed and unexposed patients are exchangeable in terms of their risk factors (i.e., no confounding). Positivity implies that the probability of being exposed conditional on a set of covariates is positive. Consistency means that the observed outcomes for exposed patients equal their counterfactual outcomes if they had received the exposure, which requires a well-defined exposure. When three identifiability conditions hold true, an observational study can be conceptualized as a conditionally randomized experiment.
Microbiome	The collection of symbiotic, pathogenic, and commensal microbes (e.g., archaea, bacteria, fungi, viruses) and their genomes and related products in a particular biological niche.
Microbiota	All microbes that exist in a particular biological niche.
Mediation analysis	Causal mediation analysis is an approach to tease apart the total effect, natural direct effect, and mediation (natural indirect) effect with the use of a counterfactual framework. The natural direct effect denotes how much the risk of an outcome (e.g., wheezing) would increase/decrease if a subject were exposed (e.g., specific microbiome signature) vs. not exposed, while the mediator (e.g., immune profile) was held constant at the level it would have taken in the unexposed subject. The natural indirect effect denotes how much the outcome risk would change if a subject were exposed, but the mediator value was changed from the unexposed level to the level it would take in the exposed subject.
Reverse causation	The situation in which the outcome of interest temporarily precedes and causes the exposure, instead of the other way around.
16S rRNA gene sequencing	Sequencing DNA within the hypervariable regions of the 16S ribosomal RNA gene that identifies the taxonomy of bacteria and archaea.

A study by de Steenhuijsen Piters and colleagues⁴ previously provided an example of how we can begin to disentangle this question. In a cohort study of 132 children with RSV infection, investigators examined the effect of nasopharyngeal microbiota clusters on host transcriptome profiles (in whole blood) and disease severity. Interrelations between RSV infection and Streptococcus-dominant microbiota were associated with unique immune profiles (e.g., upregulated Toll-like receptor signaling) and higher severity. In a recent multicenter prospective study of 221 infants hospitalized for RSV infection, our research group applied integrated omics (nasopharyngeal airway transcriptome, metatranscriptome, virome, and metabolome) and network approaches to investigate which infants with RSV infection were at highest risk of incident asthma.⁵ The study identified biologically-distinct endotypes of RSV infection (e.g., clinical^atopicmicrobiome^{S. pneumoniae/M. catarrhalis}inflammation^IFN-high endotype) that have differential effects on the development of asthma by age five years. These studies collectively suggest the integrated role of the virus, microbiome, and host immune response in the pathobiology of viral respiratory infection and its chronic morbidities in later childhood.

In the study by Rosas-Salazar and colleagues,⁶ published in this issue of the Journal of Allergy and Clinical Immunology, the authors built on their earlier work⁷ and provide another elegant example of how we can begin to disentangle the causal structure—i.e., the web of potential etiologic (causal) factors, confounders, mediators, effect modifiers, and outcomes. In a population-based cohort study—the Infant Susceptibility to Pulmonary Infections and Asthma Following RSV Exposure (INSPIRE) cohort, the investigators examined the nasal microbiota by 16S rRNA gene sequencing and 53 cytokines, chemokines, and growth factors by immunoassays in 357 infants with RSV infection with a large spectrum of acute disease severity. The study reports on the association of higher α-diversity indices of the microbiota with higher acute severity and increased frequency of wheezing episodes in the fourth year of life. In contrast, beta diversity (by the Jaccard index) was only associated with the level of acute care. In the causal mediation analysis, there was a non-significant and moderate mediation (indirect) effect of the nasal immune profile on the link between α-diversity and respiratory severity score, with effect modification of the exposure (i.e., microbial diversity) by the mediator (i.e., immune profile). It should be noted that previous studies have demonstrated different findings related to microbial diversity during ARI. For example, Teo et al. examined the nasopharyngeal microbiota in 244 Australian infants and reported dominance by certain bacteria and lower α-diversity were associated with higher ARI illness severity.⁸ The discrepancy between study results is likely multifactorial (e.g., differences in study design, target population, measurement, outcome definitions, or any combination of these factors).

Regardless, the present study supports the idea that the airway microbiota plays a major role in increased ARI severity. Moreover, the present results support the larger story implicating a complex interplay between respiratory viruses, microbiome, and host response as an important cause of recurrent wheeze and asthma in childhood.¹ On the other hand, and as acknowledged by the investigators, the cross-sectional examination of the microbiota, immune response, and acute severity might raise questions about reverse causation and confounding. Particularly, in the causal mediation analysis, the effects of interest may not be identified or consistently estimated when several important assumptions are not met—e.g., no residual exposure-mediator, exposure-outcome, and mediator-outcome confounding. Notwithstanding this uncertainty, the identification of a role of the upper airway microbiome-immune response interplay in the relationship between RSV infection, its acute severity, and subsequent wheezing illnesses is an important advance. However, in conjunction with other evidence, it seems likely that these findings will extend the link between RSV infection, more generally, and the future risk of developing asthma. Although causal inferences remain premature, focusing on the virus-microbiome-immune response interplay (as compared with conventional virus-centric approaches) has exciting implications for the development of more targeted interventions for RSV infections and asthma prevention. Promising investigational areas include examining the heterogeneity of RSV infection,⁵ longitudinal microbiome (both composition and function),⁹ multi-level omics,⁵ causal interactions (which are different from effect modification [Table]) of multiple exposures, and causal structural learning.³

Methodological challenges remain to successfully implement such research into clinical practice, —e.g., how to fully satisfy the standard assumptions of counterfactual causal inference (i.e., consistency, positivity, and exchangeability) and appropriately model the time-varying exposure-confounder feedback in a complex biological system. Additionally, the inferences derived from these observational studies are not confirmatory. Their promise lies in symbiosis with, not a replacement of, randomized controlled trials (e.g., palivizumab trial for wheezing in the first year of life¹⁰) and experimental studies when feasible. Regardless, the application of high-throughput testing and robust causal inference approaches to these clinical and epidemiological studies will help scientists develop and test novel and well-calibrated hypotheses. These scientists may focus now on exactly how the microbe and host interact within a niche and between niches (e.g., through extracellular vesicles, metabolites) and on the development of tailored interventions to prevent severe RSV infection or to mitigate its sequelae.

The study by Rosas-Salazar and colleagues provides valuable guidance on how to refine our analytic approaches to RSV infection and its acute and chronic morbidities. The integration of clinical and high-throughput molecular data, domain knowledge, and robust causal inference approaches is critical to understand how complex systems behave.³ Patients—whose lives generate the data, shape the knowledge, and promote the methodology—will benefit the most as this new scientific integration advances research and clinical practice.